diff options
Diffstat (limited to 'src/video_core')
269 files changed, 5911 insertions, 4260 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 6a6325e38..40e6d1ec4 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -1,7 +1,10 @@ +# SPDX-FileCopyrightText: 2018 yuzu Emulator Project +# SPDX-License-Identifier: GPL-2.0-or-later + add_subdirectory(host_shaders) if(LIBVA_FOUND) - set_source_files_properties(command_classes/codecs/codec.cpp + set_source_files_properties(host1x/codecs/codec.cpp PROPERTIES COMPILE_DEFINITIONS LIBVA_FOUND=1) list(APPEND FFmpeg_LIBRARIES ${LIBVA_LIBRARIES}) endif() @@ -12,26 +15,14 @@ add_library(video_core STATIC buffer_cache/buffer_cache.h cdma_pusher.cpp cdma_pusher.h - command_classes/codecs/codec.cpp - command_classes/codecs/codec.h - command_classes/codecs/h264.cpp - command_classes/codecs/h264.h - command_classes/codecs/vp8.cpp - command_classes/codecs/vp8.h - command_classes/codecs/vp9.cpp - command_classes/codecs/vp9.h - command_classes/codecs/vp9_types.h - command_classes/host1x.cpp - command_classes/host1x.h - command_classes/nvdec.cpp - command_classes/nvdec.h - command_classes/nvdec_common.h - command_classes/sync_manager.cpp - command_classes/sync_manager.h - command_classes/vic.cpp - command_classes/vic.h compatible_formats.cpp compatible_formats.h + control/channel_state.cpp + control/channel_state.h + control/channel_state_cache.cpp + control/channel_state_cache.h + control/scheduler.cpp + control/scheduler.h delayed_destruction_ring.h dirty_flags.cpp dirty_flags.h @@ -51,7 +42,31 @@ add_library(video_core STATIC engines/maxwell_3d.h engines/maxwell_dma.cpp engines/maxwell_dma.h + engines/puller.cpp + engines/puller.h framebuffer_config.h + host1x/codecs/codec.cpp + host1x/codecs/codec.h + host1x/codecs/h264.cpp + host1x/codecs/h264.h + host1x/codecs/vp8.cpp + host1x/codecs/vp8.h + host1x/codecs/vp9.cpp + host1x/codecs/vp9.h + host1x/codecs/vp9_types.h + host1x/control.cpp + host1x/control.h + host1x/host1x.cpp + host1x/host1x.h + host1x/nvdec.cpp + host1x/nvdec.h + host1x/nvdec_common.h + host1x/sync_manager.cpp + host1x/sync_manager.h + host1x/syncpoint_manager.cpp + host1x/syncpoint_manager.h + host1x/vic.cpp + host1x/vic.h macro/macro.cpp macro/macro.h macro/macro_hle.cpp @@ -192,6 +207,7 @@ add_library(video_core STATIC texture_cache/render_targets.h texture_cache/samples_helper.h texture_cache/slot_vector.h + texture_cache/texture_cache.cpp texture_cache/texture_cache.h texture_cache/texture_cache_base.h texture_cache/types.h @@ -258,10 +274,6 @@ if (MSVC) target_compile_options(video_core PRIVATE /we4242 # 'identifier': conversion from 'type1' to 'type2', possible loss of data /we4244 # 'conversion': conversion from 'type1' to 'type2', possible loss of data - /we4456 # Declaration of 'identifier' hides previous local declaration - /we4457 # Declaration of 'identifier' hides function parameter - /we4458 # Declaration of 'identifier' hides class member - /we4459 # Declaration of 'identifier' hides global declaration ) else() target_compile_options(video_core PRIVATE @@ -269,7 +281,6 @@ else() -Wno-error=sign-conversion -Werror=pessimizing-move -Werror=redundant-move - -Werror=shadow -Werror=type-limits $<$<CXX_COMPILER_ID:GNU>:-Werror=class-memaccess> @@ -277,3 +288,7 @@ else() $<$<CXX_COMPILER_ID:GNU>:-Werror=unused-but-set-variable> ) endif() + +if (ARCHITECTURE_x86_64) + target_link_libraries(video_core PRIVATE dynarmic) +endif() diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index be2113f5a..f9a6472cf 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -13,6 +12,7 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "common/div_ceil.h" +#include "common/settings.h" #include "core/memory.h" namespace VideoCommon { @@ -37,7 +37,7 @@ struct NullBufferParams {}; template <class RasterizerInterface> class BufferBase { static constexpr u64 PAGES_PER_WORD = 64; - static constexpr u64 BYTES_PER_PAGE = Core::Memory::PAGE_SIZE; + static constexpr u64 BYTES_PER_PAGE = Core::Memory::YUZU_PAGESIZE; static constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE; /// Vector tracking modified pages tightly packed with small vector optimization @@ -212,7 +212,7 @@ public: void FlushCachedWrites() noexcept { flags &= ~BufferFlagBits::CachedWrites; const u64 num_words = NumWords(); - const u64* const cached_words = Array<Type::CachedCPU>(); + u64* const cached_words = Array<Type::CachedCPU>(); u64* const untracked_words = Array<Type::Untracked>(); u64* const cpu_words = Array<Type::CPU>(); for (u64 word_index = 0; word_index < num_words; ++word_index) { @@ -220,6 +220,9 @@ public: NotifyRasterizer<false>(word_index, untracked_words[word_index], cached_bits); untracked_words[word_index] |= cached_bits; cpu_words[word_index] |= cached_bits; + if (!Settings::values.use_pessimistic_flushes) { + cached_words[word_index] = 0; + } } } diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp index ab32294c8..a16308b60 100644 --- a/src/video_core/buffer_cache/buffer_cache.cpp +++ b/src/video_core/buffer_cache/buffer_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/microprofile.h" diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index fa26eb8b0..8e26b3f95 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -1,17 +1,14 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <algorithm> #include <array> -#include <deque> #include <memory> #include <mutex> #include <numeric> #include <span> -#include <unordered_map> #include <vector> #include <boost/container/small_vector.hpp> @@ -22,10 +19,10 @@ #include "common/literals.h" #include "common/lru_cache.h" #include "common/microprofile.h" -#include "common/scope_exit.h" #include "common/settings.h" #include "core/memory.h" #include "video_core/buffer_cache/buffer_base.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" @@ -59,12 +56,12 @@ using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFE using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>; template <typename P> -class BufferCache { +class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { // Page size for caching purposes. // This is unrelated to the CPU page size and it can be changed as it seems optimal. - static constexpr u32 PAGE_BITS = 16; - static constexpr u64 PAGE_SIZE = u64{1} << PAGE_BITS; + static constexpr u32 YUZU_PAGEBITS = 16; + static constexpr u64 YUZU_PAGESIZE = u64{1} << YUZU_PAGEBITS; static constexpr bool IS_OPENGL = P::IS_OPENGL; static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = @@ -78,8 +75,9 @@ class BufferCache { static constexpr BufferId NULL_BUFFER_ID{0}; - static constexpr u64 EXPECTED_MEMORY = 512_MiB; - static constexpr u64 CRITICAL_MEMORY = 1_GiB; + static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB; + static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB; + static constexpr s64 TARGET_THRESHOLD = 4_GiB; using Maxwell = Tegra::Engines::Maxwell3D::Regs; @@ -118,10 +116,7 @@ public: static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB); explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - Runtime& runtime_); + Core::Memory::Memory& cpu_memory_, Runtime& runtime_); void TickFrame(); @@ -131,7 +126,7 @@ public: void DownloadMemory(VAddr cpu_addr, u64 size); - bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<u8> inlined_buffer); + bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer); void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size); @@ -218,8 +213,8 @@ private: template <typename Func> void ForEachBufferInRange(VAddr cpu_addr, u64 size, Func&& func) { - const u64 page_end = Common::DivCeil(cpu_addr + size, PAGE_SIZE); - for (u64 page = cpu_addr >> PAGE_BITS; page < page_end;) { + const u64 page_end = Common::DivCeil(cpu_addr + size, YUZU_PAGESIZE); + for (u64 page = cpu_addr >> YUZU_PAGEBITS; page < page_end;) { const BufferId buffer_id = page_table[page]; if (!buffer_id) { ++page; @@ -229,7 +224,7 @@ private: func(buffer_id, buffer); const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); - page = Common::DivCeil(end_addr, PAGE_SIZE); + page = Common::DivCeil(end_addr, YUZU_PAGESIZE); } } @@ -264,8 +259,8 @@ private: } static bool IsRangeGranular(VAddr cpu_addr, size_t size) { - return (cpu_addr & ~Core::Memory::PAGE_MASK) == - ((cpu_addr + size) & ~Core::Memory::PAGE_MASK); + return (cpu_addr & ~Core::Memory::YUZU_PAGEMASK) == + ((cpu_addr + size) & ~Core::Memory::YUZU_PAGEMASK); } void RunGarbageCollector(); @@ -355,7 +350,7 @@ private: void NotifyBufferDeletion(); - [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const; + [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr, bool is_written = false) const; [[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size, PixelFormat format); @@ -369,9 +364,6 @@ private: void ClearDownload(IntervalType subtract_interval); VideoCore::RasterizerInterface& rasterizer; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; Core::Memory::Memory& cpu_memory; SlotVector<Buffer> slot_buffers; @@ -438,26 +430,43 @@ private: Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache; u64 frame_tick = 0; u64 total_used_memory = 0; + u64 minimum_memory = 0; + u64 critical_memory = 0; - std::array<BufferId, ((1ULL << 39) >> PAGE_BITS)> page_table; + std::array<BufferId, ((1ULL << 39) >> YUZU_PAGEBITS)> page_table; }; template <class P> BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - Runtime& runtime_) - : runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_} { + Core::Memory::Memory& cpu_memory_, Runtime& runtime_) + : runtime{runtime_}, rasterizer{rasterizer_}, cpu_memory{cpu_memory_} { // Ensure the first slot is used for the null buffer void(slot_buffers.insert(runtime, NullBufferParams{})); common_ranges.clear(); + + if (!runtime.CanReportMemoryUsage()) { + minimum_memory = DEFAULT_EXPECTED_MEMORY; + critical_memory = DEFAULT_CRITICAL_MEMORY; + return; + } + + const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory()); + const s64 min_spacing_expected = device_memory - 1_GiB - 512_MiB; + const s64 min_spacing_critical = device_memory - 1_GiB; + const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD); + const s64 min_vacancy_expected = (6 * mem_threshold) / 10; + const s64 min_vacancy_critical = (3 * mem_threshold) / 10; + minimum_memory = static_cast<u64>( + std::max(std::min(device_memory - min_vacancy_expected, min_spacing_expected), + DEFAULT_EXPECTED_MEMORY)); + critical_memory = static_cast<u64>( + std::max(std::min(device_memory - min_vacancy_critical, min_spacing_critical), + DEFAULT_CRITICAL_MEMORY)); } template <class P> void BufferCache<P>::RunGarbageCollector() { - const bool aggressive_gc = total_used_memory >= CRITICAL_MEMORY; + const bool aggressive_gc = total_used_memory >= critical_memory; const u64 ticks_to_destroy = aggressive_gc ? 60 : 120; int num_iterations = aggressive_gc ? 64 : 32; const auto clean_up = [this, &num_iterations](BufferId buffer_id) { @@ -488,7 +497,11 @@ void BufferCache<P>::TickFrame() { const bool skip_preferred = hits * 256 < shots * 251; uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; - if (total_used_memory >= EXPECTED_MEMORY) { + // If we can obtain the memory info, use it instead of the estimate. + if (runtime.CanReportMemoryUsage()) { + total_used_memory = runtime.GetDeviceMemoryUsage(); + } + if (total_used_memory >= minimum_memory) { RunGarbageCollector(); } ++frame_tick; @@ -529,8 +542,8 @@ void BufferCache<P>::ClearDownload(IntervalType subtract_interval) { template <class P> bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { - const std::optional<VAddr> cpu_src_address = gpu_memory.GpuToCpuAddress(src_address); - const std::optional<VAddr> cpu_dest_address = gpu_memory.GpuToCpuAddress(dest_address); + const std::optional<VAddr> cpu_src_address = gpu_memory->GpuToCpuAddress(src_address); + const std::optional<VAddr> cpu_dest_address = gpu_memory->GpuToCpuAddress(dest_address); if (!cpu_src_address || !cpu_dest_address) { return false; } @@ -588,7 +601,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am template <class P> bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { - const std::optional<VAddr> cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address); + const std::optional<VAddr> cpu_dst_address = gpu_memory->GpuToCpuAddress(dst_address); if (!cpu_dst_address) { return false; } @@ -612,7 +625,7 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { template <class P> void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) { - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); const Binding binding{ .cpu_addr = *cpu_addr, .size = size, @@ -650,7 +663,7 @@ void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) { if (is_indexed) { BindHostIndexBuffer(); } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) { runtime.BindQuadArrayIndexBuffer(regs.vertex_buffer.first, regs.vertex_buffer.count); } @@ -710,9 +723,9 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, enabled_storage_buffers[stage] |= 1U << ssbo_index; written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index; - const auto& cbufs = maxwell3d.state.shader_stages[stage]; + const auto& cbufs = maxwell3d->state.shader_stages[stage]; const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset; - storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr); + storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr, is_written); } template <class P> @@ -747,12 +760,12 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, enabled_compute_storage_buffers |= 1U << ssbo_index; written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index; - const auto& launch_desc = kepler_compute.launch_description; + const auto& launch_desc = kepler_compute->launch_description; ASSERT(((launch_desc.const_buffer_enable_mask >> cbuf_index) & 1) != 0); const auto& cbufs = launch_desc.const_buffer_config; const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset; - compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr); + compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr, is_written); } template <class P> @@ -813,6 +826,19 @@ void BufferCache<P>::CommitAsyncFlushesHigh() { const bool is_accuracy_normal = Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::Normal; + auto it = committed_ranges.begin(); + while (it != committed_ranges.end()) { + auto& current_intervals = *it; + auto next_it = std::next(it); + while (next_it != committed_ranges.end()) { + for (auto& interval : *next_it) { + current_intervals.subtract(interval); + } + next_it++; + } + it++; + } + boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads; u64 total_size_bytes = 0; u64 largest_copy = 0; @@ -903,8 +929,8 @@ void BufferCache<P>::PopAsyncFlushes() {} template <class P> bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { - const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); - for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE); + for (u64 page = addr >> YUZU_PAGEBITS; page < page_end;) { const BufferId image_id = page_table[page]; if (!image_id) { ++page; @@ -915,7 +941,7 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { return true; } const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); - page = Common::DivCeil(end_addr, PAGE_SIZE); + page = Common::DivCeil(end_addr, YUZU_PAGESIZE); } return false; } @@ -923,8 +949,8 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { template <class P> bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) { const VAddr end_addr = addr + size; - const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE); - for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const u64 page_end = Common::DivCeil(end_addr, YUZU_PAGESIZE); + for (u64 page = addr >> YUZU_PAGEBITS; page < page_end;) { const BufferId buffer_id = page_table[page]; if (!buffer_id) { ++page; @@ -936,15 +962,15 @@ bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) { if (buf_start_addr < end_addr && addr < buf_end_addr) { return true; } - page = Common::DivCeil(end_addr, PAGE_SIZE); + page = Common::DivCeil(end_addr, YUZU_PAGESIZE); } return false; } template <class P> bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) { - const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); - for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE); + for (u64 page = addr >> YUZU_PAGEBITS; page < page_end;) { const BufferId image_id = page_table[page]; if (!image_id) { ++page; @@ -955,7 +981,7 @@ bool BufferCache<P>::IsRegionCpuModified(VAddr addr, size_t size) { return true; } const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); - page = Common::DivCeil(end_addr, PAGE_SIZE); + page = Common::DivCeil(end_addr, YUZU_PAGESIZE); } return false; } @@ -968,19 +994,19 @@ void BufferCache<P>::BindHostIndexBuffer() { const u32 size = index_buffer.size; SynchronizeBuffer(buffer, index_buffer.cpu_addr, size); if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { - const u32 new_offset = offset + maxwell3d.regs.index_array.first * - maxwell3d.regs.index_array.FormatSizeInBytes(); + const u32 new_offset = offset + maxwell3d->regs.index_array.first * + maxwell3d->regs.index_array.FormatSizeInBytes(); runtime.BindIndexBuffer(buffer, new_offset, size); } else { - runtime.BindIndexBuffer(maxwell3d.regs.draw.topology, maxwell3d.regs.index_array.format, - maxwell3d.regs.index_array.first, maxwell3d.regs.index_array.count, - buffer, offset, size); + runtime.BindIndexBuffer(maxwell3d->regs.draw.topology, maxwell3d->regs.index_array.format, + maxwell3d->regs.index_array.first, + maxwell3d->regs.index_array.count, buffer, offset, size); } } template <class P> void BufferCache<P>::BindHostVertexBuffers() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { const Binding& binding = vertex_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; @@ -991,7 +1017,7 @@ void BufferCache<P>::BindHostVertexBuffers() { } flags[Dirty::VertexBuffer0 + index] = false; - const u32 stride = maxwell3d.regs.vertex_array[index].stride; + const u32 stride = maxwell3d->regs.vertex_array[index].stride; const u32 offset = buffer.Offset(binding.cpu_addr); runtime.BindVertexBuffer(index, buffer, offset, binding.size, stride); } @@ -1131,7 +1157,7 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) { template <class P> void BufferCache<P>::BindHostTransformFeedbackBuffers() { - if (maxwell3d.regs.tfb_enabled == 0) { + if (maxwell3d->regs.tfb_enabled == 0) { return; } for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { @@ -1216,16 +1242,19 @@ void BufferCache<P>::BindHostComputeTextureBuffers() { template <class P> void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) { - if (is_indexed) { - UpdateIndexBuffer(); - } - UpdateVertexBuffers(); - UpdateTransformFeedbackBuffers(); - for (size_t stage = 0; stage < NUM_STAGES; ++stage) { - UpdateUniformBuffers(stage); - UpdateStorageBuffers(stage); - UpdateTextureBuffers(stage); - } + do { + has_deleted_buffers = false; + if (is_indexed) { + UpdateIndexBuffer(); + } + UpdateVertexBuffers(); + UpdateTransformFeedbackBuffers(); + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + UpdateUniformBuffers(stage); + UpdateStorageBuffers(stage); + UpdateTextureBuffers(stage); + } + } while (has_deleted_buffers); } template <class P> @@ -1239,8 +1268,8 @@ template <class P> void BufferCache<P>::UpdateIndexBuffer() { // We have to check for the dirty flags and index count // The index count is currently changed without updating the dirty flags - const auto& index_array = maxwell3d.regs.index_array; - auto& flags = maxwell3d.dirty.flags; + const auto& index_array = maxwell3d->regs.index_array; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::IndexBuffer] && last_index_count == index_array.count) { return; } @@ -1249,7 +1278,7 @@ void BufferCache<P>::UpdateIndexBuffer() { const GPUVAddr gpu_addr_begin = index_array.StartAddress(); const GPUVAddr gpu_addr_end = index_array.EndAddress(); - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin); const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); const u32 draw_size = (index_array.count + index_array.first) * index_array.FormatSizeInBytes(); const u32 size = std::min(address_size, draw_size); @@ -1266,8 +1295,8 @@ void BufferCache<P>::UpdateIndexBuffer() { template <class P> void BufferCache<P>::UpdateVertexBuffers() { - auto& flags = maxwell3d.dirty.flags; - if (!maxwell3d.dirty.flags[Dirty::VertexBuffers]) { + auto& flags = maxwell3d->dirty.flags; + if (!maxwell3d->dirty.flags[Dirty::VertexBuffers]) { return; } flags[Dirty::VertexBuffers] = false; @@ -1279,20 +1308,25 @@ void BufferCache<P>::UpdateVertexBuffers() { template <class P> void BufferCache<P>::UpdateVertexBuffer(u32 index) { - if (!maxwell3d.dirty.flags[Dirty::VertexBuffer0 + index]) { + if (!maxwell3d->dirty.flags[Dirty::VertexBuffer0 + index]) { return; } - const auto& array = maxwell3d.regs.vertex_array[index]; - const auto& limit = maxwell3d.regs.vertex_array_limit[index]; + const auto& array = maxwell3d->regs.vertex_array[index]; + const auto& limit = maxwell3d->regs.vertex_array_limit[index]; const GPUVAddr gpu_addr_begin = array.StartAddress(); const GPUVAddr gpu_addr_end = limit.LimitAddress() + 1; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin); - const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); - const u32 size = address_size; // TODO: Analyze stride and number of vertices - if (array.enable == 0 || size == 0 || !cpu_addr) { + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin); + u32 address_size = static_cast<u32>( + std::min(gpu_addr_end - gpu_addr_begin, static_cast<u64>(std::numeric_limits<u32>::max()))); + if (array.enable == 0 || address_size == 0 || !cpu_addr) { vertex_buffers[index] = NULL_BINDING; return; } + if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end)) { + address_size = + static_cast<u32>(gpu_memory->MaxContinousRange(gpu_addr_begin, address_size)); + } + const u32 size = address_size; // TODO: Analyze stride and number of vertices vertex_buffers[index] = Binding{ .cpu_addr = *cpu_addr, .size = size, @@ -1346,7 +1380,7 @@ void BufferCache<P>::UpdateTextureBuffers(size_t stage) { template <class P> void BufferCache<P>::UpdateTransformFeedbackBuffers() { - if (maxwell3d.regs.tfb_enabled == 0) { + if (maxwell3d->regs.tfb_enabled == 0) { return; } for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { @@ -1356,10 +1390,10 @@ void BufferCache<P>::UpdateTransformFeedbackBuffers() { template <class P> void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) { - const auto& binding = maxwell3d.regs.tfb_bindings[index]; + const auto& binding = maxwell3d->regs.tfb_bindings[index]; const GPUVAddr gpu_addr = binding.Address() + binding.buffer_offset; const u32 size = binding.buffer_size; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (binding.buffer_enable == 0 || size == 0 || !cpu_addr) { transform_feedback_buffers[index] = NULL_BINDING; return; @@ -1378,10 +1412,10 @@ void BufferCache<P>::UpdateComputeUniformBuffers() { ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { Binding& binding = compute_uniform_buffers[index]; binding = NULL_BINDING; - const auto& launch_desc = kepler_compute.launch_description; + const auto& launch_desc = kepler_compute->launch_description; if (((launch_desc.const_buffer_enable_mask >> index) & 1) != 0) { const auto& cbuf = launch_desc.const_buffer_config[index]; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(cbuf.Address()); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(cbuf.Address()); if (cpu_addr) { binding.cpu_addr = *cpu_addr; binding.size = cbuf.size; @@ -1436,7 +1470,7 @@ BufferId BufferCache<P>::FindBuffer(VAddr cpu_addr, u32 size) { if (cpu_addr == 0) { return NULL_BUFFER_ID; } - const u64 page = cpu_addr >> PAGE_BITS; + const u64 page = cpu_addr >> YUZU_PAGEBITS; const BufferId buffer_id = page_table[page]; if (!buffer_id) { return CreateBuffer(cpu_addr, size); @@ -1457,8 +1491,9 @@ typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(VAddr cpu VAddr end = cpu_addr + wanted_size; int stream_score = 0; bool has_stream_leap = false; - for (; cpu_addr >> PAGE_BITS < Common::DivCeil(end, PAGE_SIZE); cpu_addr += PAGE_SIZE) { - const BufferId overlap_id = page_table[cpu_addr >> PAGE_BITS]; + for (; cpu_addr >> YUZU_PAGEBITS < Common::DivCeil(end, YUZU_PAGESIZE); + cpu_addr += YUZU_PAGESIZE) { + const BufferId overlap_id = page_table[cpu_addr >> YUZU_PAGEBITS]; if (!overlap_id) { continue; } @@ -1469,19 +1504,27 @@ typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(VAddr cpu overlap_ids.push_back(overlap_id); overlap.Pick(); const VAddr overlap_cpu_addr = overlap.CpuAddr(); - if (overlap_cpu_addr < begin) { + const bool expands_left = overlap_cpu_addr < begin; + if (expands_left) { cpu_addr = begin = overlap_cpu_addr; } - end = std::max(end, overlap_cpu_addr + overlap.SizeBytes()); - + const VAddr overlap_end = overlap_cpu_addr + overlap.SizeBytes(); + const bool expands_right = overlap_end > end; + if (overlap_end > end) { + end = overlap_end; + } stream_score += overlap.StreamScore(); if (stream_score > STREAM_LEAP_THRESHOLD && !has_stream_leap) { // When this memory region has been joined a bunch of times, we assume it's being used // as a stream buffer. Increase the size to skip constantly recreating buffers. has_stream_leap = true; - begin -= PAGE_SIZE * 256; - cpu_addr = begin; - end += PAGE_SIZE * 256; + if (expands_right) { + begin -= YUZU_PAGESIZE * 256; + cpu_addr = begin; + } + if (expands_left) { + end += YUZU_PAGESIZE * 256; + } } } return OverlapResult{ @@ -1522,6 +1565,8 @@ BufferId BufferCache<P>::CreateBuffer(VAddr cpu_addr, u32 wanted_size) { const OverlapResult overlap = ResolveOverlaps(cpu_addr, wanted_size); const u32 size = static_cast<u32>(overlap.end - overlap.begin); const BufferId new_buffer_id = slot_buffers.insert(runtime, rasterizer, overlap.begin, size); + auto& new_buffer = slot_buffers[new_buffer_id]; + runtime.ClearBuffer(new_buffer, 0, new_buffer.SizeBytes(), 0); for (const BufferId overlap_id : overlap.ids) { JoinOverlap(new_buffer_id, overlap_id, !overlap.has_stream_leap); } @@ -1554,8 +1599,8 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) { } const VAddr cpu_addr_begin = buffer.CpuAddr(); const VAddr cpu_addr_end = cpu_addr_begin + size; - const u64 page_begin = cpu_addr_begin / PAGE_SIZE; - const u64 page_end = Common::DivCeil(cpu_addr_end, PAGE_SIZE); + const u64 page_begin = cpu_addr_begin / YUZU_PAGESIZE; + const u64 page_end = Common::DivCeil(cpu_addr_end, YUZU_PAGESIZE); for (u64 page = page_begin; page != page_end; ++page) { if constexpr (insert) { page_table[page] = buffer_id; @@ -1650,7 +1695,7 @@ void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, template <class P> bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, - std::span<u8> inlined_buffer) { + std::span<const u8> inlined_buffer) { const bool is_dirty = IsRegionRegistered(dest_address, copy_size); if (!is_dirty) { return false; @@ -1786,7 +1831,7 @@ void BufferCache<P>::NotifyBufferDeletion() { dirty_uniform_buffers.fill(~u32{0}); uniform_buffer_binding_sizes.fill({}); } - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; flags[Dirty::IndexBuffer] = true; flags[Dirty::VertexBuffers] = true; for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { @@ -1796,16 +1841,18 @@ void BufferCache<P>::NotifyBufferDeletion() { } template <class P> -typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr) const { - const GPUVAddr gpu_addr = gpu_memory.Read<u64>(ssbo_addr); - const u32 size = gpu_memory.Read<u32>(ssbo_addr + 8); - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); +typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, + bool is_written) const { + const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr); + const u32 size = gpu_memory->Read<u32>(ssbo_addr + 8); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr || size == 0) { return NULL_BINDING; } + const VAddr cpu_end = Common::AlignUp(*cpu_addr + size, Core::Memory::YUZU_PAGESIZE); const Binding binding{ .cpu_addr = *cpu_addr, - .size = size, + .size = is_written ? size : static_cast<u32>(cpu_end - *cpu_addr), .buffer_id = BufferId{}, }; return binding; @@ -1814,7 +1861,7 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s template <class P> typename BufferCache<P>::TextureBufferBinding BufferCache<P>::GetTextureBufferBinding( GPUVAddr gpu_addr, u32 size, PixelFormat format) { - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); TextureBufferBinding binding; if (!cpu_addr || size == 0) { binding.cpu_addr = 0; diff --git a/src/video_core/cdma_pusher.cpp b/src/video_core/cdma_pusher.cpp index a8c4b4415..28a2d2090 100644 --- a/src/video_core/cdma_pusher.cpp +++ b/src/video_core/cdma_pusher.cpp @@ -1,40 +1,23 @@ -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: MIT #include <bit> -#include "command_classes/host1x.h" -#include "command_classes/nvdec.h" -#include "command_classes/vic.h" #include "video_core/cdma_pusher.h" -#include "video_core/command_classes/nvdec_common.h" -#include "video_core/command_classes/sync_manager.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" +#include "video_core/host1x/control.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" +#include "video_core/host1x/nvdec_common.h" +#include "video_core/host1x/sync_manager.h" +#include "video_core/host1x/vic.h" #include "video_core/memory_manager.h" namespace Tegra { -CDmaPusher::CDmaPusher(GPU& gpu_) - : gpu{gpu_}, nvdec_processor(std::make_shared<Nvdec>(gpu)), - vic_processor(std::make_unique<Vic>(gpu, nvdec_processor)), - host1x_processor(std::make_unique<Host1x>(gpu)), - sync_manager(std::make_unique<SyncptIncrManager>(gpu)) {} +CDmaPusher::CDmaPusher(Host1x::Host1x& host1x_) + : host1x{host1x_}, nvdec_processor(std::make_shared<Host1x::Nvdec>(host1x)), + vic_processor(std::make_unique<Host1x::Vic>(host1x, nvdec_processor)), + host1x_processor(std::make_unique<Host1x::Control>(host1x)), + sync_manager(std::make_unique<Host1x::SyncptIncrManager>(host1x)) {} CDmaPusher::~CDmaPusher() = default; @@ -128,16 +111,17 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) { case ThiMethod::SetMethod1: LOG_DEBUG(Service_NVDRV, "VIC method 0x{:X}, Args=({})", static_cast<u32>(vic_thi_state.method_0), data); - vic_processor->ProcessMethod(static_cast<Vic::Method>(vic_thi_state.method_0), data); + vic_processor->ProcessMethod(static_cast<Host1x::Vic::Method>(vic_thi_state.method_0), + data); break; default: break; } break; - case ChClassId::Host1x: + case ChClassId::Control: // This device is mainly for syncpoint synchronization LOG_DEBUG(Service_NVDRV, "Host1X Class Method"); - host1x_processor->ProcessMethod(static_cast<Host1x::Method>(offset), data); + host1x_processor->ProcessMethod(static_cast<Host1x::Control::Method>(offset), data); break; default: UNIMPLEMENTED_MSG("Current class not implemented {:X}", static_cast<u32>(current_class)); diff --git a/src/video_core/cdma_pusher.h b/src/video_core/cdma_pusher.h index 87b49d6ea..83112dfce 100644 --- a/src/video_core/cdma_pusher.h +++ b/src/video_core/cdma_pusher.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,15 +7,18 @@ #include <vector> #include "common/bit_field.h" +#include "common/common_funcs.h" #include "common/common_types.h" namespace Tegra { -class GPU; +namespace Host1x { +class Control; class Host1x; class Nvdec; class SyncptIncrManager; class Vic; +} // namespace Host1x enum class ChSubmissionMode : u32 { SetClass = 0, @@ -30,7 +32,7 @@ enum class ChSubmissionMode : u32 { enum class ChClassId : u32 { NoClass = 0x0, - Host1x = 0x1, + Control = 0x1, VideoEncodeMpeg = 0x20, VideoEncodeNvEnc = 0x21, VideoStreamingVi = 0x30, @@ -88,7 +90,7 @@ enum class ThiMethod : u32 { class CDmaPusher { public: - explicit CDmaPusher(GPU& gpu_); + explicit CDmaPusher(Host1x::Host1x& host1x); ~CDmaPusher(); /// Process the command entry @@ -101,11 +103,11 @@ private: /// Write arguments value to the ThiRegisters member at the specified offset void ThiStateWrite(ThiRegisters& state, u32 offset, u32 argument); - GPU& gpu; - std::shared_ptr<Tegra::Nvdec> nvdec_processor; - std::unique_ptr<Tegra::Vic> vic_processor; - std::unique_ptr<Tegra::Host1x> host1x_processor; - std::unique_ptr<SyncptIncrManager> sync_manager; + Host1x::Host1x& host1x; + std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor; + std::unique_ptr<Tegra::Host1x::Vic> vic_processor; + std::unique_ptr<Tegra::Host1x::Control> host1x_processor; + std::unique_ptr<Host1x::SyncptIncrManager> sync_manager; ChClassId current_class{}; ThiRegisters vic_thi_state{}; ThiRegisters nvdec_thi_state{}; diff --git a/src/video_core/command_classes/host1x.cpp b/src/video_core/command_classes/host1x.cpp deleted file mode 100644 index b12494528..000000000 --- a/src/video_core/command_classes/host1x.cpp +++ /dev/null @@ -1,30 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/assert.h" -#include "video_core/command_classes/host1x.h" -#include "video_core/gpu.h" - -Tegra::Host1x::Host1x(GPU& gpu_) : gpu(gpu_) {} - -Tegra::Host1x::~Host1x() = default; - -void Tegra::Host1x::ProcessMethod(Method method, u32 argument) { - switch (method) { - case Method::LoadSyncptPayload32: - syncpoint_value = argument; - break; - case Method::WaitSyncpt: - case Method::WaitSyncpt32: - Execute(argument); - break; - default: - UNIMPLEMENTED_MSG("Host1x method 0x{:X}", static_cast<u32>(method)); - break; - } -} - -void Tegra::Host1x::Execute(u32 data) { - gpu.WaitFence(data, syncpoint_value); -} diff --git a/src/video_core/command_classes/sync_manager.cpp b/src/video_core/command_classes/sync_manager.cpp deleted file mode 100644 index 19dc9e0ab..000000000 --- a/src/video_core/command_classes/sync_manager.cpp +++ /dev/null @@ -1,60 +0,0 @@ -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// - -#include <algorithm> -#include "sync_manager.h" -#include "video_core/gpu.h" - -namespace Tegra { -SyncptIncrManager::SyncptIncrManager(GPU& gpu_) : gpu(gpu_) {} -SyncptIncrManager::~SyncptIncrManager() = default; - -void SyncptIncrManager::Increment(u32 id) { - increments.emplace_back(0, 0, id, true); - IncrementAllDone(); -} - -u32 SyncptIncrManager::IncrementWhenDone(u32 class_id, u32 id) { - const u32 handle = current_id++; - increments.emplace_back(handle, class_id, id); - return handle; -} - -void SyncptIncrManager::SignalDone(u32 handle) { - const auto done_incr = - std::find_if(increments.begin(), increments.end(), - [handle](const SyncptIncr& incr) { return incr.id == handle; }); - if (done_incr != increments.cend()) { - done_incr->complete = true; - } - IncrementAllDone(); -} - -void SyncptIncrManager::IncrementAllDone() { - std::size_t done_count = 0; - for (; done_count < increments.size(); ++done_count) { - if (!increments[done_count].complete) { - break; - } - gpu.IncrementSyncPoint(increments[done_count].syncpt_id); - } - increments.erase(increments.begin(), increments.begin() + done_count); -} -} // namespace Tegra diff --git a/src/video_core/command_classes/sync_manager.h b/src/video_core/command_classes/sync_manager.h deleted file mode 100644 index 2c321ec58..000000000 --- a/src/video_core/command_classes/sync_manager.h +++ /dev/null @@ -1,64 +0,0 @@ -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// - -#pragma once - -#include <mutex> -#include <vector> -#include "common/common_types.h" - -namespace Tegra { -class GPU; -struct SyncptIncr { - u32 id; - u32 class_id; - u32 syncpt_id; - bool complete; - - SyncptIncr(u32 id_, u32 class_id_, u32 syncpt_id_, bool done = false) - : id(id_), class_id(class_id_), syncpt_id(syncpt_id_), complete(done) {} -}; - -class SyncptIncrManager { -public: - explicit SyncptIncrManager(GPU& gpu); - ~SyncptIncrManager(); - - /// Add syncpoint id and increment all - void Increment(u32 id); - - /// Returns a handle to increment later - u32 IncrementWhenDone(u32 class_id, u32 id); - - /// IncrememntAllDone, including handle - void SignalDone(u32 handle); - - /// Increment all sequential pending increments that are already done. - void IncrementAllDone(); - -private: - std::vector<SyncptIncr> increments; - std::mutex increment_lock; - u32 current_id{}; - - GPU& gpu; -}; - -} // namespace Tegra diff --git a/src/video_core/compatible_formats.cpp b/src/video_core/compatible_formats.cpp index 8317d0636..4e75f33ca 100644 --- a/src/video_core/compatible_formats.cpp +++ b/src/video_core/compatible_formats.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <cstddef> @@ -132,9 +131,12 @@ constexpr std::array VIEW_CLASS_ASTC_8x8_RGBA{ // PixelFormat::ASTC_2D_10X5_SRGB // Missing formats: -// PixelFormat::ASTC_2D_10X6_UNORM // PixelFormat::ASTC_2D_10X6_SRGB +constexpr std::array VIEW_CLASS_ASTC_10x6_RGBA{ + PixelFormat::ASTC_2D_10X6_UNORM, +}; + constexpr std::array VIEW_CLASS_ASTC_10x8_RGBA{ PixelFormat::ASTC_2D_10X8_UNORM, PixelFormat::ASTC_2D_10X8_SRGB, @@ -227,6 +229,7 @@ constexpr Table MakeViewTable() { EnableRange(view, VIEW_CLASS_ASTC_6x6_RGBA); EnableRange(view, VIEW_CLASS_ASTC_8x5_RGBA); EnableRange(view, VIEW_CLASS_ASTC_8x8_RGBA); + EnableRange(view, VIEW_CLASS_ASTC_10x6_RGBA); EnableRange(view, VIEW_CLASS_ASTC_10x8_RGBA); EnableRange(view, VIEW_CLASS_ASTC_10x10_RGBA); EnableRange(view, VIEW_CLASS_ASTC_12x12_RGBA); diff --git a/src/video_core/compatible_formats.h b/src/video_core/compatible_formats.h index 55745e042..a7ce2c198 100644 --- a/src/video_core/compatible_formats.h +++ b/src/video_core/compatible_formats.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/control/channel_state.cpp b/src/video_core/control/channel_state.cpp new file mode 100644 index 000000000..cdecc3a91 --- /dev/null +++ b/src/video_core/control/channel_state.cpp @@ -0,0 +1,40 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "video_core/control/channel_state.h" +#include "video_core/dma_pusher.h" +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/kepler_memory.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_dma.h" +#include "video_core/engines/puller.h" +#include "video_core/memory_manager.h" + +namespace Tegra::Control { + +ChannelState::ChannelState(s32 bind_id_) : bind_id{bind_id_}, initialized{} {} + +void ChannelState::Init(Core::System& system, GPU& gpu) { + ASSERT(memory_manager); + dma_pusher = std::make_unique<Tegra::DmaPusher>(system, gpu, *memory_manager, *this); + maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, *memory_manager); + fermi_2d = std::make_unique<Engines::Fermi2D>(); + kepler_compute = std::make_unique<Engines::KeplerCompute>(system, *memory_manager); + maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, *memory_manager); + kepler_memory = std::make_unique<Engines::KeplerMemory>(system, *memory_manager); + initialized = true; +} + +void ChannelState::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) { + dma_pusher->BindRasterizer(rasterizer); + memory_manager->BindRasterizer(rasterizer); + maxwell_3d->BindRasterizer(rasterizer); + fermi_2d->BindRasterizer(rasterizer); + kepler_memory->BindRasterizer(rasterizer); + kepler_compute->BindRasterizer(rasterizer); + maxwell_dma->BindRasterizer(rasterizer); +} + +} // namespace Tegra::Control diff --git a/src/video_core/control/channel_state.h b/src/video_core/control/channel_state.h new file mode 100644 index 000000000..3a7b9872c --- /dev/null +++ b/src/video_core/control/channel_state.h @@ -0,0 +1,68 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <memory> + +#include "common/common_types.h" + +namespace Core { +class System; +} + +namespace VideoCore { +class RasterizerInterface; +} + +namespace Tegra { + +class GPU; + +namespace Engines { +class Puller; +class Fermi2D; +class Maxwell3D; +class MaxwellDMA; +class KeplerCompute; +class KeplerMemory; +} // namespace Engines + +class MemoryManager; +class DmaPusher; + +namespace Control { + +struct ChannelState { + explicit ChannelState(s32 bind_id); + ChannelState(const ChannelState& state) = delete; + ChannelState& operator=(const ChannelState&) = delete; + ChannelState(ChannelState&& other) noexcept = default; + ChannelState& operator=(ChannelState&& other) noexcept = default; + + void Init(Core::System& system, GPU& gpu); + + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + + s32 bind_id = -1; + /// 3D engine + std::unique_ptr<Engines::Maxwell3D> maxwell_3d; + /// 2D engine + std::unique_ptr<Engines::Fermi2D> fermi_2d; + /// Compute engine + std::unique_ptr<Engines::KeplerCompute> kepler_compute; + /// DMA engine + std::unique_ptr<Engines::MaxwellDMA> maxwell_dma; + /// Inline memory engine + std::unique_ptr<Engines::KeplerMemory> kepler_memory; + + std::shared_ptr<MemoryManager> memory_manager; + + std::unique_ptr<DmaPusher> dma_pusher; + + bool initialized{}; +}; + +} // namespace Control + +} // namespace Tegra diff --git a/src/video_core/control/channel_state_cache.cpp b/src/video_core/control/channel_state_cache.cpp new file mode 100644 index 000000000..4ebeb6356 --- /dev/null +++ b/src/video_core/control/channel_state_cache.cpp @@ -0,0 +1,14 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "video_core/control/channel_state_cache.inc" + +namespace VideoCommon { + +ChannelInfo::ChannelInfo(Tegra::Control::ChannelState& channel_state) + : maxwell3d{*channel_state.maxwell_3d}, kepler_compute{*channel_state.kepler_compute}, + gpu_memory{*channel_state.memory_manager} {} + +template class VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo>; + +} // namespace VideoCommon diff --git a/src/video_core/control/channel_state_cache.h b/src/video_core/control/channel_state_cache.h new file mode 100644 index 000000000..584a0c26c --- /dev/null +++ b/src/video_core/control/channel_state_cache.h @@ -0,0 +1,101 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <deque> +#include <limits> +#include <mutex> +#include <optional> +#include <unordered_map> +#include <vector> + +#include "common/common_types.h" + +namespace Tegra { + +namespace Engines { +class Maxwell3D; +class KeplerCompute; +} // namespace Engines + +class MemoryManager; + +namespace Control { +struct ChannelState; +} + +} // namespace Tegra + +namespace VideoCommon { + +class ChannelInfo { +public: + ChannelInfo() = delete; + explicit ChannelInfo(Tegra::Control::ChannelState& state); + ChannelInfo(const ChannelInfo& state) = delete; + ChannelInfo& operator=(const ChannelInfo&) = delete; + ChannelInfo(ChannelInfo&& other) = default; + ChannelInfo& operator=(ChannelInfo&& other) = default; + + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager& gpu_memory; +}; + +template <class P> +class ChannelSetupCaches { +public: + /// Operations for seting the channel of execution. + virtual ~ChannelSetupCaches(); + + /// Create channel state. + virtual void CreateChannel(Tegra::Control::ChannelState& channel); + + /// Bind a channel for execution. + void BindToChannel(s32 id); + + /// Erase channel's state. + void EraseChannel(s32 id); + + Tegra::MemoryManager* GetFromID(size_t id) const { + std::unique_lock<std::mutex> lk(config_mutex); + const auto ref = address_spaces.find(id); + return ref->second.gpu_memory; + } + + std::optional<size_t> getStorageID(size_t id) const { + std::unique_lock<std::mutex> lk(config_mutex); + const auto ref = address_spaces.find(id); + if (ref == address_spaces.end()) { + return std::nullopt; + } + return ref->second.storage_id; + } + +protected: + static constexpr size_t UNSET_CHANNEL{std::numeric_limits<size_t>::max()}; + + P* channel_state; + size_t current_channel_id{UNSET_CHANNEL}; + size_t current_address_space{}; + Tegra::Engines::Maxwell3D* maxwell3d; + Tegra::Engines::KeplerCompute* kepler_compute; + Tegra::MemoryManager* gpu_memory; + + std::deque<P> channel_storage; + std::deque<size_t> free_channel_ids; + std::unordered_map<s32, size_t> channel_map; + std::vector<size_t> active_channel_ids; + struct AddresSpaceRef { + size_t ref_count; + size_t storage_id; + Tegra::MemoryManager* gpu_memory; + }; + std::unordered_map<size_t, AddresSpaceRef> address_spaces; + mutable std::mutex config_mutex; + + virtual void OnGPUASRegister([[maybe_unused]] size_t map_id) {} +}; + +} // namespace VideoCommon diff --git a/src/video_core/control/channel_state_cache.inc b/src/video_core/control/channel_state_cache.inc new file mode 100644 index 000000000..460313893 --- /dev/null +++ b/src/video_core/control/channel_state_cache.inc @@ -0,0 +1,86 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <algorithm> + +#include "video_core/control/channel_state.h" +#include "video_core/control/channel_state_cache.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" + +namespace VideoCommon { + +template <class P> +ChannelSetupCaches<P>::~ChannelSetupCaches() = default; + +template <class P> +void ChannelSetupCaches<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) { + std::unique_lock<std::mutex> lk(config_mutex); + ASSERT(channel_map.find(channel.bind_id) == channel_map.end() && channel.bind_id >= 0); + auto new_id = [this, &channel]() { + if (!free_channel_ids.empty()) { + auto id = free_channel_ids.front(); + free_channel_ids.pop_front(); + new (&channel_storage[id]) P(channel); + return id; + } + channel_storage.emplace_back(channel); + return channel_storage.size() - 1; + }(); + channel_map.emplace(channel.bind_id, new_id); + if (current_channel_id != UNSET_CHANNEL) { + channel_state = &channel_storage[current_channel_id]; + } + active_channel_ids.push_back(new_id); + auto as_it = address_spaces.find(channel.memory_manager->GetID()); + if (as_it != address_spaces.end()) { + as_it->second.ref_count++; + return; + } + AddresSpaceRef new_gpu_mem_ref{ + .ref_count = 1, + .storage_id = address_spaces.size(), + .gpu_memory = channel.memory_manager.get(), + }; + address_spaces.emplace(channel.memory_manager->GetID(), new_gpu_mem_ref); + OnGPUASRegister(channel.memory_manager->GetID()); +} + +/// Bind a channel for execution. +template <class P> +void ChannelSetupCaches<P>::BindToChannel(s32 id) { + std::unique_lock<std::mutex> lk(config_mutex); + auto it = channel_map.find(id); + ASSERT(it != channel_map.end() && id >= 0); + current_channel_id = it->second; + channel_state = &channel_storage[current_channel_id]; + maxwell3d = &channel_state->maxwell3d; + kepler_compute = &channel_state->kepler_compute; + gpu_memory = &channel_state->gpu_memory; + current_address_space = gpu_memory->GetID(); +} + +/// Erase channel's channel_state. +template <class P> +void ChannelSetupCaches<P>::EraseChannel(s32 id) { + std::unique_lock<std::mutex> lk(config_mutex); + const auto it = channel_map.find(id); + ASSERT(it != channel_map.end() && id >= 0); + const auto this_id = it->second; + free_channel_ids.push_back(this_id); + channel_map.erase(it); + if (this_id == current_channel_id) { + current_channel_id = UNSET_CHANNEL; + channel_state = nullptr; + maxwell3d = nullptr; + kepler_compute = nullptr; + gpu_memory = nullptr; + } else if (current_channel_id != UNSET_CHANNEL) { + channel_state = &channel_storage[current_channel_id]; + } + active_channel_ids.erase( + std::find(active_channel_ids.begin(), active_channel_ids.end(), this_id)); +} + +} // namespace VideoCommon diff --git a/src/video_core/control/scheduler.cpp b/src/video_core/control/scheduler.cpp new file mode 100644 index 000000000..f7cbe204e --- /dev/null +++ b/src/video_core/control/scheduler.cpp @@ -0,0 +1,32 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <memory> + +#include "common/assert.h" +#include "video_core/control/channel_state.h" +#include "video_core/control/scheduler.h" +#include "video_core/gpu.h" + +namespace Tegra::Control { +Scheduler::Scheduler(GPU& gpu_) : gpu{gpu_} {} + +Scheduler::~Scheduler() = default; + +void Scheduler::Push(s32 channel, CommandList&& entries) { + std::unique_lock lk(scheduling_guard); + auto it = channels.find(channel); + ASSERT(it != channels.end()); + auto channel_state = it->second; + gpu.BindChannel(channel_state->bind_id); + channel_state->dma_pusher->Push(std::move(entries)); + channel_state->dma_pusher->DispatchCalls(); +} + +void Scheduler::DeclareChannel(std::shared_ptr<ChannelState> new_channel) { + s32 channel = new_channel->bind_id; + std::unique_lock lk(scheduling_guard); + channels.emplace(channel, new_channel); +} + +} // namespace Tegra::Control diff --git a/src/video_core/control/scheduler.h b/src/video_core/control/scheduler.h new file mode 100644 index 000000000..44addf61c --- /dev/null +++ b/src/video_core/control/scheduler.h @@ -0,0 +1,37 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <memory> +#include <mutex> +#include <unordered_map> + +#include "video_core/dma_pusher.h" + +namespace Tegra { + +class GPU; + +namespace Control { + +struct ChannelState; + +class Scheduler { +public: + explicit Scheduler(GPU& gpu_); + ~Scheduler(); + + void Push(s32 channel, CommandList&& entries); + + void DeclareChannel(std::shared_ptr<ChannelState> new_channel); + +private: + std::unordered_map<s32, std::shared_ptr<ChannelState>> channels; + std::mutex scheduling_guard; + GPU& gpu; +}; + +} // namespace Control + +} // namespace Tegra diff --git a/src/video_core/delayed_destruction_ring.h b/src/video_core/delayed_destruction_ring.h index 4f1d29c04..d13ee622b 100644 --- a/src/video_core/delayed_destruction_ring.h +++ b/src/video_core/delayed_destruction_ring.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/dirty_flags.cpp b/src/video_core/dirty_flags.cpp index b1be065c3..9dc4341f0 100644 --- a/src/video_core/dirty_flags.cpp +++ b/src/video_core/dirty_flags.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <cstddef> diff --git a/src/video_core/dirty_flags.h b/src/video_core/dirty_flags.h index d63ad5a35..736082d83 100644 --- a/src/video_core/dirty_flags.h +++ b/src/video_core/dirty_flags.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 8d28bd884..9835e3ac1 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -1,12 +1,10 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/cityhash.h" #include "common/microprofile.h" #include "common/settings.h" #include "core/core.h" -#include "core/memory.h" #include "video_core/dma_pusher.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/gpu.h" @@ -14,7 +12,10 @@ namespace Tegra { -DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_) : gpu{gpu_}, system{system_} {} +DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_, MemoryManager& memory_manager_, + Control::ChannelState& channel_state_) + : gpu{gpu_}, system{system_}, memory_manager{memory_manager_}, puller{gpu_, memory_manager_, + *this, channel_state_} {} DmaPusher::~DmaPusher() = default; @@ -23,8 +24,6 @@ MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128, void DmaPusher::DispatchCalls() { MICROPROFILE_SCOPE(DispatchCalls); - gpu.SyncGuestHost(); - dma_pushbuffer_subindex = 0; dma_state.is_last_call = true; @@ -35,7 +34,6 @@ void DmaPusher::DispatchCalls() { } } gpu.FlushCommands(); - gpu.SyncGuestHost(); gpu.OnCommandListEnd(); } @@ -78,11 +76,11 @@ bool DmaPusher::Step() { // Push buffer non-empty, read a word command_headers.resize(command_list_header.size); if (Settings::IsGPULevelHigh()) { - gpu.MemoryManager().ReadBlock(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + memory_manager.ReadBlock(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); } else { - gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + memory_manager.ReadBlockUnsafe(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); } } for (std::size_t index = 0; index < command_headers.size();) { @@ -156,7 +154,7 @@ void DmaPusher::SetState(const CommandHeader& command_header) { void DmaPusher::CallMethod(u32 argument) const { if (dma_state.method < non_puller_methods) { - gpu.CallMethod(GPU::MethodCall{ + puller.CallPullerMethod(Engines::Puller::MethodCall{ dma_state.method, argument, dma_state.subchannel, @@ -170,12 +168,16 @@ void DmaPusher::CallMethod(u32 argument) const { void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const { if (dma_state.method < non_puller_methods) { - gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods, - dma_state.method_count); + puller.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods, + dma_state.method_count); } else { subchannels[dma_state.subchannel]->CallMultiMethod(dma_state.method, base_start, num_methods, dma_state.method_count); } } +void DmaPusher::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) { + puller.BindRasterizer(rasterizer); +} + } // namespace Tegra diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h index 19f286fa7..938f0f11c 100644 --- a/src/video_core/dma_pusher.h +++ b/src/video_core/dma_pusher.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -11,6 +10,7 @@ #include "common/bit_field.h" #include "common/common_types.h" #include "video_core/engines/engine_interface.h" +#include "video_core/engines/puller.h" namespace Core { class System; @@ -18,7 +18,12 @@ class System; namespace Tegra { +namespace Control { +struct ChannelState; +} + class GPU; +class MemoryManager; enum class SubmissionMode : u32 { IncreasingOld = 0, @@ -32,24 +37,32 @@ enum class SubmissionMode : u32 { // Note that, traditionally, methods are treated as 4-byte addressable locations, and hence // their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4. // So the values you see in docs might be multiplied by 4. +// Register documentation: +// https://github.com/NVIDIA/open-gpu-doc/blob/ab27fc22db5de0d02a4cabe08e555663b62db4d4/classes/host/cla26f.h +// +// Register Description (approx): +// https://github.com/NVIDIA/open-gpu-doc/blob/ab27fc22db5de0d02a4cabe08e555663b62db4d4/manuals/volta/gv100/dev_pbdma.ref.txt enum class BufferMethods : u32 { BindObject = 0x0, + Illegal = 0x1, Nop = 0x2, SemaphoreAddressHigh = 0x4, SemaphoreAddressLow = 0x5, - SemaphoreSequence = 0x6, - SemaphoreTrigger = 0x7, - NotifyIntr = 0x8, + SemaphoreSequencePayload = 0x6, + SemaphoreOperation = 0x7, + NonStallInterrupt = 0x8, WrcacheFlush = 0x9, - Unk28 = 0xA, - UnkCacheFlush = 0xB, + MemOpA = 0xA, + MemOpB = 0xB, + MemOpC = 0xC, + MemOpD = 0xD, RefCnt = 0x14, SemaphoreAcquire = 0x1A, SemaphoreRelease = 0x1B, - FenceValue = 0x1C, - FenceAction = 0x1D, - WaitForInterrupt = 0x1E, - Unk7c = 0x1F, + SyncpointPayload = 0x1C, + SyncpointOperation = 0x1D, + WaitForIdle = 0x1E, + CRCCheck = 0x1F, Yield = 0x20, NonPullerMethods = 0x40, }; @@ -103,7 +116,8 @@ struct CommandList final { */ class DmaPusher final { public: - explicit DmaPusher(Core::System& system_, GPU& gpu_); + explicit DmaPusher(Core::System& system_, GPU& gpu_, MemoryManager& memory_manager_, + Control::ChannelState& channel_state_); ~DmaPusher(); void Push(CommandList&& entries) { @@ -116,6 +130,8 @@ public: subchannels[subchannel_id] = engine; } + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + private: static constexpr u32 non_puller_methods = 0x40; static constexpr u32 max_subchannels = 8; @@ -149,6 +165,8 @@ private: GPU& gpu; Core::System& system; + MemoryManager& memory_manager; + mutable Engines::Puller puller; }; } // namespace Tegra diff --git a/src/video_core/engines/const_buffer_info.h b/src/video_core/engines/const_buffer_info.h index d8f672462..abafcf090 100644 --- a/src/video_core/engines/const_buffer_info.h +++ b/src/video_core/engines/const_buffer_info.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/engines/engine_interface.h b/src/video_core/engines/engine_interface.h index c7ffd68c5..26cde8584 100644 --- a/src/video_core/engines/engine_interface.h +++ b/src/video_core/engines/engine_interface.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/engines/engine_upload.cpp b/src/video_core/engines/engine_upload.cpp index 351b110fe..a34819234 100644 --- a/src/video_core/engines/engine_upload.cpp +++ b/src/video_core/engines/engine_upload.cpp @@ -1,9 +1,9 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstring> +#include "common/algorithm.h" #include "common/assert.h" #include "video_core/engines/engine_upload.h" #include "video_core/memory_manager.h" @@ -35,21 +35,48 @@ void State::ProcessData(const u32 data, const bool is_last_call) { if (!is_last_call) { return; } + ProcessData(inner_buffer); +} + +void State::ProcessData(const u32* data, size_t num_data) { + std::span<const u8> read_buffer(reinterpret_cast<const u8*>(data), num_data * sizeof(u32)); + ProcessData(read_buffer); +} + +void State::ProcessData(std::span<const u8> read_buffer) { const GPUVAddr address{regs.dest.Address()}; if (is_linear) { - rasterizer->AccelerateInlineToMemory(address, copy_size, inner_buffer); + if (regs.line_count == 1) { + rasterizer->AccelerateInlineToMemory(address, copy_size, read_buffer); + } else { + for (u32 line = 0; line < regs.line_count; ++line) { + const GPUVAddr dest_line = address + static_cast<size_t>(line) * regs.dest.pitch; + memory_manager.WriteBlockUnsafe( + dest_line, read_buffer.data() + static_cast<size_t>(line) * regs.line_length_in, + regs.line_length_in); + } + memory_manager.InvalidateRegion(address, regs.dest.pitch * regs.line_count); + } } else { - UNIMPLEMENTED_IF(regs.dest.z != 0); - UNIMPLEMENTED_IF(regs.dest.depth != 1); - UNIMPLEMENTED_IF(regs.dest.BlockWidth() != 0); - UNIMPLEMENTED_IF(regs.dest.BlockDepth() != 0); + u32 width = regs.dest.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = regs.dest.x; + const u32 bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(address)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + const u32 bytes_per_pixel = 1U << bpp_shift; const std::size_t dst_size = Tegra::Texture::CalculateSize( - true, 1, regs.dest.width, regs.dest.height, 1, regs.dest.BlockHeight(), 0); + true, bytes_per_pixel, width, regs.dest.height, regs.dest.depth, + regs.dest.BlockHeight(), regs.dest.BlockDepth()); tmp_buffer.resize(dst_size); memory_manager.ReadBlock(address, tmp_buffer.data(), dst_size); - Tegra::Texture::SwizzleKepler(regs.dest.width, regs.dest.height, regs.dest.x, regs.dest.y, - regs.dest.BlockHeight(), copy_size, inner_buffer.data(), - tmp_buffer.data()); + Tegra::Texture::SwizzleSubrect(tmp_buffer, read_buffer, bytes_per_pixel, width, + regs.dest.height, regs.dest.depth, x_offset, regs.dest.y, + x_elements, regs.line_count, regs.dest.BlockHeight(), + regs.dest.BlockDepth(), regs.line_length_in); memory_manager.WriteBlock(address, tmp_buffer.data(), dst_size); } } diff --git a/src/video_core/engines/engine_upload.h b/src/video_core/engines/engine_upload.h index c9c5ec8c3..f08f6e36a 100644 --- a/src/video_core/engines/engine_upload.h +++ b/src/video_core/engines/engine_upload.h @@ -1,9 +1,9 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once +#include <span> #include <vector> #include "common/bit_field.h" #include "common/common_types.h" @@ -34,7 +34,7 @@ struct Registers { u32 width; u32 height; u32 depth; - u32 z; + u32 layer; u32 x; u32 y; @@ -63,11 +63,14 @@ public: void ProcessExec(bool is_linear_); void ProcessData(u32 data, bool is_last_call); + void ProcessData(const u32* data, size_t num_data); /// Binds a rasterizer to this engine. void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); private: + void ProcessData(std::span<const u8> read_buffer); + u32 write_offset = 0; u32 copy_size = 0; std::vector<u8> inner_buffer; diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index f26530ede..453e0fb01 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" #include "common/logging/log.h" diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h index d76c5ed56..1229aa35b 100644 --- a/src/video_core/engines/fermi_2d.h +++ b/src/video_core/engines/fermi_2d.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,7 +8,6 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" -#include "common/math_util.h" #include "video_core/engines/engine_interface.h" #include "video_core/gpu.h" diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index 5a1c12076..7c50bdbe0 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <bitset> #include "common/assert.h" @@ -10,7 +9,6 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" -#include "video_core/renderer_base.h" #include "video_core/textures/decoders.h" namespace Tegra::Engines { @@ -38,8 +36,6 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal } case KEPLER_COMPUTE_REG_INDEX(data_upload): { upload_state.ProcessData(method_argument, is_last_call); - if (is_last_call) { - } break; } case KEPLER_COMPUTE_REG_INDEX(launch): @@ -52,8 +48,15 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { - for (std::size_t i = 0; i < amount; i++) { - CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + switch (method) { + case KEPLER_COMPUTE_REG_INDEX(data_upload): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; + default: + for (std::size_t i = 0; i < amount; i++) { + CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + } + break; } } diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index f8b8d06ac..aab309ecc 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -12,7 +11,6 @@ #include "common/common_types.h" #include "video_core/engines/engine_interface.h" #include "video_core/engines/engine_upload.h" -#include "video_core/gpu.h" #include "video_core/textures/texture.h" namespace Core { diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index 8aed16caa..a3fbab1e5 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" #include "common/logging/log.h" @@ -9,8 +8,6 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" -#include "video_core/renderer_base.h" -#include "video_core/textures/decoders.h" namespace Tegra::Engines { @@ -36,8 +33,6 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call } case KEPLERMEMORY_REG_INDEX(data): { upload_state.ProcessData(method_argument, is_last_call); - if (is_last_call) { - } break; } } @@ -45,8 +40,15 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { - for (std::size_t i = 0; i < amount; i++) { - CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + switch (method) { + case KEPLERMEMORY_REG_INDEX(data): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; + default: + for (std::size_t i = 0; i < amount; i++) { + CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + } + break; } } diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h index 949e2fae1..5fe7489f0 100644 --- a/src/video_core/engines/kepler_memory.h +++ b/src/video_core/engines/kepler_memory.h @@ -1,18 +1,15 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> #include <cstddef> -#include <vector> #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/engines/engine_interface.h" #include "video_core/engines/engine_upload.h" -#include "video_core/gpu.h" namespace Core { class System; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 5d6d217bb..3c6e44a25 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -1,12 +1,12 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstring> #include <optional> #include "common/assert.h" #include "core/core.h" #include "core/core_timing.h" +#include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/gpu.h" #include "video_core/memory_manager.h" @@ -173,6 +173,8 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume case MAXWELL3D_REG_INDEX(shadow_ram_control): shadow_state.shadow_ram_control = static_cast<Regs::ShadowRamControl>(nonshadow_argument); return; + case MAXWELL3D_REG_INDEX(macros.upload_address): + return macro_engine->ClearCode(regs.macros.upload_address); case MAXWELL3D_REG_INDEX(macros.data): return macro_engine->AddCode(regs.macros.upload_address, argument); case MAXWELL3D_REG_INDEX(macros.bind): @@ -195,7 +197,7 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13: case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14: case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15: - return StartCBData(method); + return ProcessCBData(argument); case MAXWELL3D_REG_INDEX(cb_bind[0]): return ProcessCBBind(0); case MAXWELL3D_REG_INDEX(cb_bind[1]): @@ -208,6 +210,21 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume return ProcessCBBind(4); case MAXWELL3D_REG_INDEX(draw.vertex_end_gl): return DrawArrays(); + case MAXWELL3D_REG_INDEX(small_index): + regs.index_array.count = regs.small_index.count; + regs.index_array.first = regs.small_index.first; + dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; + return DrawArrays(); + case MAXWELL3D_REG_INDEX(small_index_2): + regs.index_array.count = regs.small_index_2.count; + regs.index_array.first = regs.small_index_2.first; + dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; + // a macro calls this one over and over, should it increase instancing? + // Used by Hades and likely other Vulkan games. + return DrawArrays(); + case MAXWELL3D_REG_INDEX(topology_override): + use_topology_override = true; + return; case MAXWELL3D_REG_INDEX(clear_buffers): return ProcessClearBuffers(); case MAXWELL3D_REG_INDEX(query.query_get): @@ -222,11 +239,12 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume return upload_state.ProcessExec(regs.exec_upload.linear != 0); case MAXWELL3D_REG_INDEX(data_upload): upload_state.ProcessData(argument, is_last_call); - if (is_last_call) { - } return; case MAXWELL3D_REG_INDEX(fragment_barrier): return rasterizer->FragmentBarrier(); + case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache): + rasterizer->InvalidateGPUCache(); + return rasterizer->WaitForIdle(); case MAXWELL3D_REG_INDEX(tiled_cache_barrier): return rasterizer->TiledCacheBarrier(); } @@ -248,14 +266,6 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters) } void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) { - if (method == cb_data_state.current) { - regs.reg_array[method] = method_argument; - ProcessCBData(method_argument); - return; - } else if (cb_data_state.current != null_cb_data) { - FinishCBData(); - } - // It is an error to write to a register other than the current macro's ARG register before it // has finished execution. if (executing_macro != 0) { @@ -302,8 +312,11 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13: case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14: case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15: - ProcessCBMultiData(method, base_start, amount); + ProcessCBMultiData(base_start, amount); break; + case MAXWELL3D_REG_INDEX(data_upload): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; default: for (std::size_t i = 0; i < amount; i++) { CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); @@ -360,6 +373,35 @@ void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) { } } +void Maxwell3D::ProcessTopologyOverride() { + using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology; + using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride; + + PrimitiveTopology topology{}; + + switch (regs.topology_override) { + case PrimitiveTopologyOverride::None: + topology = regs.draw.topology; + break; + case PrimitiveTopologyOverride::Points: + topology = PrimitiveTopology::Points; + break; + case PrimitiveTopologyOverride::Lines: + topology = PrimitiveTopology::Lines; + break; + case PrimitiveTopologyOverride::LineStrip: + topology = PrimitiveTopology::LineStrip; + break; + default: + topology = static_cast<PrimitiveTopology>(regs.topology_override); + break; + } + + if (use_topology_override) { + regs.draw.topology.Assign(topology); + } +} + void Maxwell3D::FlushMMEInlineDraw() { LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(), regs.vertex_buffer.count); @@ -370,6 +412,8 @@ void Maxwell3D::FlushMMEInlineDraw() { ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont, "Illegal combination of instancing parameters"); + ProcessTopologyOverride(); + const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed; if (ShouldExecute()) { rasterizer->Draw(is_indexed, true); @@ -409,18 +453,10 @@ void Maxwell3D::ProcessFirmwareCall4() { } void Maxwell3D::StampQueryResult(u64 payload, bool long_query) { - struct LongQueryResult { - u64_le value; - u64_le timestamp; - }; - static_assert(sizeof(LongQueryResult) == 16, "LongQueryResult has wrong size"); const GPUVAddr sequence_address{regs.query.QueryAddress()}; if (long_query) { - // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast - // GPU, this command may actually take a while to complete in real hardware due to GPU - // wait queues. - LongQueryResult query_result{payload, system.GPU().GetTicks()}; - memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); + memory_manager.Write<u64>(sequence_address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); } else { memory_manager.Write<u32>(sequence_address, static_cast<u32>(payload)); } @@ -434,10 +470,25 @@ void Maxwell3D::ProcessQueryGet() { switch (regs.query.query_get.operation) { case Regs::QueryOperation::Release: - if (regs.query.query_get.fence == 1) { - rasterizer->SignalSemaphore(regs.query.QueryAddress(), regs.query.query_sequence); + if (regs.query.query_get.fence == 1 || regs.query.query_get.short_query != 0) { + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); } else { - StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0); + struct LongQueryResult { + u64_le value; + u64_le timestamp; + }; + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u64>(sequence_address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); + }); + rasterizer->SyncOperation(std::move(operation)); } break; case Regs::QueryOperation::Acquire: @@ -529,6 +580,8 @@ void Maxwell3D::DrawArrays() { ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont, "Illegal combination of instancing parameters"); + ProcessTopologyOverride(); + if (regs.draw.instance_next) { // Increment the current instance *before* drawing. state.current_instance += 1; @@ -555,8 +608,8 @@ void Maxwell3D::DrawArrays() { std::optional<u64> Maxwell3D::GetQueryResult() { switch (regs.query.query_get.select) { - case Regs::QuerySelect::Zero: - return 0; + case Regs::QuerySelect::Payload: + return regs.query.query_sequence; case Regs::QuerySelect::SamplesPassed: // Deferred. rasterizer->Query(regs.query.QueryAddress(), QueryType::SamplesPassed, @@ -587,46 +640,7 @@ void Maxwell3D::ProcessCBBind(size_t stage_index) { rasterizer->BindGraphicsUniformBuffer(stage_index, bind_data.index, gpu_addr, size); } -void Maxwell3D::ProcessCBData(u32 value) { - const u32 id = cb_data_state.id; - cb_data_state.buffer[id][cb_data_state.counter] = value; - // Increment the current buffer position. - regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4; - cb_data_state.counter++; -} - -void Maxwell3D::StartCBData(u32 method) { - constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data); - cb_data_state.start_pos = regs.const_buffer.cb_pos; - cb_data_state.id = method - first_cb_data; - cb_data_state.current = method; - cb_data_state.counter = 0; - ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]); -} - -void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount) { - if (cb_data_state.current != method) { - if (cb_data_state.current != null_cb_data) { - FinishCBData(); - } - constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data); - cb_data_state.start_pos = regs.const_buffer.cb_pos; - cb_data_state.id = method - first_cb_data; - cb_data_state.current = method; - cb_data_state.counter = 0; - } - const std::size_t id = cb_data_state.id; - const std::size_t size = amount; - std::size_t i = 0; - for (; i < size; i++) { - cb_data_state.buffer[id][cb_data_state.counter] = start_base[i]; - cb_data_state.counter++; - } - // Increment the current buffer position. - regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4 * amount; -} - -void Maxwell3D::FinishCBData() { +void Maxwell3D::ProcessCBMultiData(const u32* start_base, u32 amount) { // Write the input value to the current const buffer at the current position. const GPUVAddr buffer_address = regs.const_buffer.BufferAddress(); ASSERT(buffer_address != 0); @@ -634,14 +648,16 @@ void Maxwell3D::FinishCBData() { // Don't allow writing past the end of the buffer. ASSERT(regs.const_buffer.cb_pos <= regs.const_buffer.cb_size); - const GPUVAddr address{buffer_address + cb_data_state.start_pos}; - const std::size_t size = regs.const_buffer.cb_pos - cb_data_state.start_pos; + const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos}; + const size_t copy_size = amount * sizeof(u32); + memory_manager.WriteBlock(address, start_base, copy_size); - const u32 id = cb_data_state.id; - memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size); + // Increment the current buffer position. + regs.const_buffer.cb_pos += static_cast<u32>(copy_size); +} - cb_data_state.id = null_cb_data; - cb_data_state.current = null_cb_data; +void Maxwell3D::ProcessCBData(u32 value) { + ProcessCBMultiData(&value, 1); } Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index dc9df6c8b..5f9eb208c 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -10,7 +9,6 @@ #include <limits> #include <optional> #include <type_traits> -#include <unordered_map> #include <vector> #include "common/assert.h" @@ -95,7 +93,7 @@ public: }; enum class QuerySelect : u32 { - Zero = 0, + Payload = 0, TimeElapsed = 2, TransformFeedbackPrimitivesGenerated = 11, PrimitivesGenerated = 18, @@ -204,7 +202,7 @@ public: case Size::Size_11_11_10: return 3; default: - UNREACHABLE(); + ASSERT(false); return 1; } } @@ -240,7 +238,7 @@ public: case Size::Size_11_11_10: return 4; default: - UNREACHABLE(); + ASSERT(false); return 1; } } @@ -276,7 +274,7 @@ public: case Size::Size_11_11_10: return "11_11_10"; default: - UNREACHABLE(); + ASSERT(false); return {}; } } @@ -298,7 +296,7 @@ public: case Type::Float: return "FLOAT"; } - UNREACHABLE(); + ASSERT(false); return {}; } @@ -338,7 +336,7 @@ public: case 3: return {x3, y3}; default: - UNREACHABLE(); + ASSERT(false); return {0, 0}; } } @@ -367,6 +365,22 @@ public: Patches = 0xe, }; + // Constants as from NVC0_3D_UNK1970_D3D + // https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/drivers/nouveau/nvc0/nvc0_3d.xml.h#L1598 + enum class PrimitiveTopologyOverride : u32 { + None = 0x0, + Points = 0x1, + Lines = 0x2, + LineStrip = 0x3, + Triangles = 0x4, + TriangleStrip = 0x5, + LinesAdjacency = 0xa, + LineStripAdjacency = 0xb, + TrianglesAdjacency = 0xc, + TriangleStripAdjacency = 0xd, + Patches = 0xe, + }; + enum class IndexFormat : u32 { UnsignedByte = 0x0, UnsignedShort = 0x1, @@ -1179,7 +1193,7 @@ public: case IndexFormat::UnsignedInt: return 4; } - UNREACHABLE(); + ASSERT(false); return 1; } @@ -1200,7 +1214,17 @@ public: } } index_array; - INSERT_PADDING_WORDS_NOINIT(0x7); + union { + BitField<0, 16, u32> first; + BitField<16, 16, u32> count; + } small_index; + + union { + BitField<0, 16, u32> first; + BitField<16, 16, u32> count; + } small_index_2; + + INSERT_PADDING_WORDS_NOINIT(0x5); INSERT_PADDING_WORDS_NOINIT(0x1F); @@ -1244,7 +1268,11 @@ public: BitField<11, 1, u32> depth_clamp_disabled; } view_volume_clip_control; - INSERT_PADDING_WORDS_NOINIT(0x1F); + INSERT_PADDING_WORDS_NOINIT(0xC); + + PrimitiveTopologyOverride topology_override; + + INSERT_PADDING_WORDS_NOINIT(0x12); u32 depth_bounds_enable; @@ -1520,10 +1548,8 @@ private: void ProcessSyncPoint(); /// Handles a write to the CB_DATA[i] register. - void StartCBData(u32 method); void ProcessCBData(u32 value); - void ProcessCBMultiData(u32 method, const u32* start_base, u32 amount); - void FinishCBData(); + void ProcessCBMultiData(const u32* start_base, u32 amount); /// Handles a write to the CB_BIND register. void ProcessCBBind(size_t stage_index); @@ -1531,6 +1557,9 @@ private: /// Handles a write to the VERTEX_END_GL register, triggering a draw. void DrawArrays(); + /// Handles use of topology overrides (e.g., to avoid using a topology assigned from a macro) + void ProcessTopologyOverride(); + // Handles a instance drawcall from MME void StepInstance(MMEDrawMode expected_mode, u32 count); @@ -1555,20 +1584,10 @@ private: /// Interpreter for the macro codes uploaded to the GPU. std::unique_ptr<MacroEngine> macro_engine; - static constexpr u32 null_cb_data = 0xFFFFFFFF; - struct CBDataState { - static constexpr size_t inline_size = 0x4000; - std::array<std::array<u32, inline_size>, 16> buffer; - u32 current{null_cb_data}; - u32 id{null_cb_data}; - u32 start_pos{}; - u32 counter{}; - }; - CBDataState cb_data_state; - Upload::State upload_state; bool execute_on{true}; + bool use_topology_override{false}; }; #define ASSERT_REG_POSITION(field_name, position) \ @@ -1685,6 +1704,7 @@ ASSERT_REG_POSITION(draw, 0x585); ASSERT_REG_POSITION(primitive_restart, 0x591); ASSERT_REG_POSITION(provoking_vertex_last, 0x5A1); ASSERT_REG_POSITION(index_array, 0x5F2); +ASSERT_REG_POSITION(small_index, 0x5F9); ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F); ASSERT_REG_POSITION(instanced_arrays, 0x620); ASSERT_REG_POSITION(vp_point_size, 0x644); @@ -1694,6 +1714,7 @@ ASSERT_REG_POSITION(cull_face, 0x648); ASSERT_REG_POSITION(pixel_center_integer, 0x649); ASSERT_REG_POSITION(viewport_transform_enabled, 0x64B); ASSERT_REG_POSITION(view_volume_clip_control, 0x64F); +ASSERT_REG_POSITION(topology_override, 0x65C); ASSERT_REG_POSITION(depth_bounds_enable, 0x66F); ASSERT_REG_POSITION(logic_op, 0x671); ASSERT_REG_POSITION(clear_buffers, 0x674); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 67388d980..3909d36c1 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -1,7 +1,7 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later +#include "common/algorithm.h" #include "common/assert.h" #include "common/logging/log.h" #include "common/microprofile.h" @@ -53,18 +53,15 @@ void MaxwellDMA::Launch() { // TODO(Subv): Perform more research and implement all features of this engine. const LaunchDMA& launch = regs.launch_dma; - ASSERT(launch.semaphore_type == LaunchDMA::SemaphoreType::NONE); ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE); ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED); - ASSERT(regs.dst_params.origin.x == 0); - ASSERT(regs.dst_params.origin.y == 0); const bool is_src_pitch = launch.src_memory_layout == LaunchDMA::MemoryLayout::PITCH; const bool is_dst_pitch = launch.dst_memory_layout == LaunchDMA::MemoryLayout::PITCH; if (!is_src_pitch && !is_dst_pitch) { // If both the source and the destination are in block layout, assert. - UNREACHABLE_MSG("Tiled->Tiled DMA transfers are not yet implemented"); + UNIMPLEMENTED_MSG("Tiled->Tiled DMA transfers are not yet implemented"); return; } @@ -79,6 +76,7 @@ void MaxwellDMA::Launch() { CopyPitchToBlockLinear(); } } + ReleaseSemaphore(); } void MaxwellDMA::CopyPitchToPitch() { @@ -122,22 +120,40 @@ void MaxwellDMA::CopyPitchToPitch() { void MaxwellDMA::CopyBlockLinearToPitch() { UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0); - UNIMPLEMENTED_IF(regs.src_params.block_size.depth != 0); UNIMPLEMENTED_IF(regs.src_params.layer != 0); + const bool is_remapping = regs.launch_dma.remap_enable != 0; + // Optimized path for micro copies. const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; - if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && + if (!is_remapping && dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && regs.src_params.height > GOB_SIZE_Y) { FastCopyBlockLinearToPitch(); return; } // Deswizzle the input and copy it over. - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); - const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in; const Parameters& src_params = regs.src_params; - const u32 width = src_params.width; + + const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; + const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; + + const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size; + + u32 width = src_params.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = src_params.origin.x; + u32 bpp_shift = 0U; + if (!is_remapping) { + bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(regs.offset_in)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + } + + const u32 bytes_per_pixel = base_bpp << bpp_shift; const u32 height = src_params.height; const u32 depth = src_params.depth; const u32 block_height = src_params.block_size.height; @@ -155,29 +171,45 @@ void MaxwellDMA::CopyBlockLinearToPitch() { memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); - UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, width, bytes_per_pixel, - block_height, src_params.origin.x, src_params.origin.y, write_buffer.data(), - read_buffer.data()); + UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, + src_params.origin.y, x_elements, regs.line_count, block_height, block_depth, + regs.pitch_out); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } void MaxwellDMA::CopyPitchToBlockLinear() { UNIMPLEMENTED_IF_MSG(regs.dst_params.block_size.width != 0, "Block width is not one"); - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); + UNIMPLEMENTED_IF(regs.dst_params.layer != 0); + + const bool is_remapping = regs.launch_dma.remap_enable != 0; + const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; + const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; const auto& dst_params = regs.dst_params; - const u32 bytes_per_pixel = regs.pitch_in / regs.line_length_in; - const u32 width = dst_params.width; + + const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size; + + u32 width = dst_params.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = dst_params.origin.x; + u32 bpp_shift = 0U; + if (!is_remapping) { + bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(regs.offset_out)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + } + + const u32 bytes_per_pixel = base_bpp << bpp_shift; const u32 height = dst_params.height; const u32 depth = dst_params.depth; const u32 block_height = dst_params.block_size.height; const u32 block_depth = dst_params.block_size.depth; const size_t dst_size = CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); - const size_t dst_layer_size = - CalculateSize(true, bytes_per_pixel, width, height, 1, block_height, block_depth); - const size_t src_size = static_cast<size_t>(regs.pitch_in) * regs.line_count; if (read_buffer.size() < src_size) { @@ -187,31 +219,23 @@ void MaxwellDMA::CopyPitchToBlockLinear() { write_buffer.resize(dst_size); } + memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); if (Settings::IsGPULevelExtreme()) { - memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); } else { - memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); } // If the input is linear and the output is tiled, swizzle the input and copy it over. - if (regs.dst_params.block_size.depth > 0) { - ASSERT(dst_params.layer == 0); - SwizzleSliceToVoxel(regs.line_length_in, regs.line_count, regs.pitch_in, width, height, - bytes_per_pixel, block_height, block_depth, dst_params.origin.x, - dst_params.origin.y, write_buffer.data(), read_buffer.data()); - } else { - SwizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_in, width, bytes_per_pixel, - write_buffer.data() + dst_layer_size * dst_params.layer, read_buffer.data(), - block_height, dst_params.origin.x, dst_params.origin.y); - } + SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, + dst_params.origin.y, x_elements, regs.line_count, block_height, block_depth, + regs.pitch_in); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } void MaxwellDMA::FastCopyBlockLinearToPitch() { - const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in; + const u32 bytes_per_pixel = 1U; const size_t src_size = GOB_SIZE; const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; u32 pos_x = regs.src_params.origin.x; @@ -237,11 +261,38 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() { memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); } - UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, regs.src_params.width, - bytes_per_pixel, regs.src_params.block_size.height, pos_x, pos_y, - write_buffer.data(), read_buffer.data()); + UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, regs.src_params.width, + regs.src_params.height, 1, pos_x, pos_y, regs.line_length_in, regs.line_count, + regs.src_params.block_size.height, regs.src_params.block_size.depth, + regs.pitch_out); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } +void MaxwellDMA::ReleaseSemaphore() { + const auto type = regs.launch_dma.semaphore_type; + const GPUVAddr address = regs.semaphore.address; + const u32 payload = regs.semaphore.payload; + switch (type) { + case LaunchDMA::SemaphoreType::NONE: + break; + case LaunchDMA::SemaphoreType::RELEASE_ONE_WORD_SEMAPHORE: { + std::function<void()> operation( + [this, address, payload] { memory_manager.Write<u32>(address, payload); }); + rasterizer->SignalFence(std::move(operation)); + break; + } + case LaunchDMA::SemaphoreType::RELEASE_FOUR_WORD_SEMAPHORE: { + std::function<void()> operation([this, address, payload] { + memory_manager.Write<u64>(address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(address, payload); + }); + rasterizer->SignalFence(std::move(operation)); + break; + } + default: + ASSERT_MSG(false, "Unknown semaphore type: {}", static_cast<u32>(type.Value())); + } +} + } // namespace Tegra::Engines diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index a04514425..bc48320ce 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,10 +7,8 @@ #include <cstddef> #include <vector> #include "common/bit_field.h" -#include "common/common_funcs.h" #include "common/common_types.h" #include "video_core/engines/engine_interface.h" -#include "video_core/gpu.h" namespace Core { class System; @@ -192,10 +189,16 @@ public: BitField<4, 3, Swizzle> dst_y; BitField<8, 3, Swizzle> dst_z; BitField<12, 3, Swizzle> dst_w; + BitField<0, 12, u32> dst_components_raw; BitField<16, 2, u32> component_size_minus_one; BitField<20, 2, u32> num_src_components_minus_one; BitField<24, 2, u32> num_dst_components_minus_one; }; + + Swizzle GetComponent(size_t i) const { + const u32 raw = dst_components_raw; + return static_cast<Swizzle>((raw >> (i * 3)) & 0x7); + } }; static_assert(sizeof(RemapConst) == 12); @@ -224,6 +227,8 @@ private: void FastCopyBlockLinearToPitch(); + void ReleaseSemaphore(); + Core::System& system; MemoryManager& memory_manager; diff --git a/src/video_core/engines/puller.cpp b/src/video_core/engines/puller.cpp new file mode 100644 index 000000000..cca890792 --- /dev/null +++ b/src/video_core/engines/puller.cpp @@ -0,0 +1,306 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "common/logging/log.h" +#include "common/settings.h" +#include "core/core.h" +#include "video_core/control/channel_state.h" +#include "video_core/dma_pusher.h" +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/kepler_memory.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_dma.h" +#include "video_core/engines/puller.h" +#include "video_core/gpu.h" +#include "video_core/memory_manager.h" +#include "video_core/rasterizer_interface.h" + +namespace Tegra::Engines { + +Puller::Puller(GPU& gpu_, MemoryManager& memory_manager_, DmaPusher& dma_pusher_, + Control::ChannelState& channel_state_) + : gpu{gpu_}, memory_manager{memory_manager_}, dma_pusher{dma_pusher_}, channel_state{ + channel_state_} {} + +Puller::~Puller() = default; + +void Puller::ProcessBindMethod(const MethodCall& method_call) { + // Bind the current subchannel to the desired engine id. + LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, + method_call.argument); + const auto engine_id = static_cast<EngineID>(method_call.argument); + bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id); + switch (engine_id) { + case EngineID::FERMI_TWOD_A: + dma_pusher.BindSubchannel(channel_state.fermi_2d.get(), method_call.subchannel); + break; + case EngineID::MAXWELL_B: + dma_pusher.BindSubchannel(channel_state.maxwell_3d.get(), method_call.subchannel); + break; + case EngineID::KEPLER_COMPUTE_B: + dma_pusher.BindSubchannel(channel_state.kepler_compute.get(), method_call.subchannel); + break; + case EngineID::MAXWELL_DMA_COPY_A: + dma_pusher.BindSubchannel(channel_state.maxwell_dma.get(), method_call.subchannel); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + dma_pusher.BindSubchannel(channel_state.kepler_memory.get(), method_call.subchannel); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", engine_id); + } +} + +void Puller::ProcessFenceActionMethod() { + switch (regs.fence_action.op) { + case Puller::FenceOperation::Acquire: + // UNIMPLEMENTED_MSG("Channel Scheduling pending."); + // WaitFence(regs.fence_action.syncpoint_id, regs.fence_value); + rasterizer->ReleaseFences(); + break; + case Puller::FenceOperation::Increment: + rasterizer->SignalSyncPoint(regs.fence_action.syncpoint_id); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented operation {}", regs.fence_action.op.Value()); + } +} + +void Puller::ProcessSemaphoreTriggerMethod() { + const auto semaphoreOperationMask = 0xF; + const auto op = + static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask); + if (op == GpuSemaphoreOperation::WriteLong) { + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u64>(sequence_address + sizeof(u64), gpu.GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); + } else { + do { + const u32 word{memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress())}; + regs.acquire_source = true; + regs.acquire_value = regs.semaphore_sequence; + if (op == GpuSemaphoreOperation::AcquireEqual) { + regs.acquire_active = true; + regs.acquire_mode = false; + if (word != regs.acquire_value) { + rasterizer->ReleaseFences(); + continue; + } + } else if (op == GpuSemaphoreOperation::AcquireGequal) { + regs.acquire_active = true; + regs.acquire_mode = true; + if (word < regs.acquire_value) { + rasterizer->ReleaseFences(); + continue; + } + } else if (op == GpuSemaphoreOperation::AcquireMask) { + if (word && regs.semaphore_sequence == 0) { + rasterizer->ReleaseFences(); + continue; + } + } else { + LOG_ERROR(HW_GPU, "Invalid semaphore operation"); + } + } while (false); + } +} + +void Puller::ProcessSemaphoreRelease() { + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_release; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SyncOperation(std::move(operation)); +} + +void Puller::ProcessSemaphoreAcquire() { + u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); + const auto value = regs.semaphore_acquire; + while (word != value) { + regs.acquire_active = true; + regs.acquire_value = value; + std::this_thread::sleep_for(std::chrono::milliseconds(1)); + rasterizer->ReleaseFences(); + word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); + // TODO(kemathe73) figure out how to do the acquire_timeout + regs.acquire_mode = false; + regs.acquire_source = false; + } +} + +/// Calls a GPU puller method. +void Puller::CallPullerMethod(const MethodCall& method_call) { + regs.reg_array[method_call.method] = method_call.argument; + const auto method = static_cast<BufferMethods>(method_call.method); + + switch (method) { + case BufferMethods::BindObject: { + ProcessBindMethod(method_call); + break; + } + case BufferMethods::Nop: + case BufferMethods::SemaphoreAddressHigh: + case BufferMethods::SemaphoreAddressLow: + case BufferMethods::SemaphoreSequencePayload: + case BufferMethods::SyncpointPayload: + break; + case BufferMethods::WrcacheFlush: + case BufferMethods::RefCnt: + rasterizer->SignalReference(); + break; + case BufferMethods::SyncpointOperation: + ProcessFenceActionMethod(); + break; + case BufferMethods::WaitForIdle: + rasterizer->WaitForIdle(); + break; + case BufferMethods::SemaphoreOperation: { + ProcessSemaphoreTriggerMethod(); + break; + } + case BufferMethods::NonStallInterrupt: { + LOG_ERROR(HW_GPU, "Special puller engine method NonStallInterrupt not implemented"); + break; + } + case BufferMethods::MemOpA: { + LOG_ERROR(HW_GPU, "Memory Operation A"); + break; + } + case BufferMethods::MemOpB: { + // Implement this better. + rasterizer->InvalidateGPUCache(); + break; + } + case BufferMethods::MemOpC: + case BufferMethods::MemOpD: { + LOG_ERROR(HW_GPU, "Memory Operation C,D"); + break; + } + case BufferMethods::SemaphoreAcquire: { + ProcessSemaphoreAcquire(); + break; + } + case BufferMethods::SemaphoreRelease: { + ProcessSemaphoreRelease(); + break; + } + case BufferMethods::Yield: { + // TODO(Kmather73): Research and implement this method. + LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented"); + break; + } + default: + LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented", method); + break; + } +} + +/// Calls a GPU engine method. +void Puller::CallEngineMethod(const MethodCall& method_call) { + const EngineID engine = bound_engines[method_call.subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + channel_state.fermi_2d->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::MAXWELL_B: + channel_state.maxwell_3d->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::KEPLER_COMPUTE_B: + channel_state.kepler_compute->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::MAXWELL_DMA_COPY_A: + channel_state.maxwell_dma->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + channel_state.kepler_memory->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + +/// Calls a GPU engine multivalue method. +void Puller::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { + const EngineID engine = bound_engines[subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + channel_state.fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_B: + channel_state.maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_COMPUTE_B: + channel_state.kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_DMA_COPY_A: + channel_state.maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + channel_state.kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + +/// Calls a GPU method. +void Puller::CallMethod(const MethodCall& method_call) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method_call.method, + method_call.subchannel); + + ASSERT(method_call.subchannel < bound_engines.size()); + + if (ExecuteMethodOnEngine(method_call.method)) { + CallEngineMethod(method_call); + } else { + CallPullerMethod(method_call); + } +} + +/// Calls a GPU multivalue method. +void Puller::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel); + + ASSERT(subchannel < bound_engines.size()); + + if (ExecuteMethodOnEngine(method)) { + CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); + } else { + for (std::size_t i = 0; i < amount; i++) { + CallPullerMethod(MethodCall{ + method, + base_start[i], + subchannel, + methods_pending - static_cast<u32>(i), + }); + } + } +} + +void Puller::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; +} + +/// Determines where the method should be executed. +[[nodiscard]] bool Puller::ExecuteMethodOnEngine(u32 method) { + const auto buffer_method = static_cast<BufferMethods>(method); + return buffer_method >= BufferMethods::NonPullerMethods; +} + +} // namespace Tegra::Engines diff --git a/src/video_core/engines/puller.h b/src/video_core/engines/puller.h new file mode 100644 index 000000000..d4175ee94 --- /dev/null +++ b/src/video_core/engines/puller.h @@ -0,0 +1,177 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <array> +#include <cstddef> +#include <vector> +#include "common/bit_field.h" +#include "common/common_funcs.h" +#include "common/common_types.h" +#include "video_core/engines/engine_interface.h" + +namespace Core { +class System; +} + +namespace Tegra { +class MemoryManager; +class DmaPusher; + +enum class EngineID { + FERMI_TWOD_A = 0x902D, // 2D Engine + MAXWELL_B = 0xB197, // 3D Engine + KEPLER_COMPUTE_B = 0xB1C0, + KEPLER_INLINE_TO_MEMORY_B = 0xA140, + MAXWELL_DMA_COPY_A = 0xB0B5, +}; + +namespace Control { +struct ChannelState; +} +} // namespace Tegra + +namespace VideoCore { +class RasterizerInterface; +} + +namespace Tegra::Engines { + +class Puller final { +public: + struct MethodCall { + u32 method{}; + u32 argument{}; + u32 subchannel{}; + u32 method_count{}; + + explicit MethodCall(u32 method_, u32 argument_, u32 subchannel_ = 0, u32 method_count_ = 0) + : method(method_), argument(argument_), subchannel(subchannel_), + method_count(method_count_) {} + + [[nodiscard]] bool IsLastCall() const { + return method_count <= 1; + } + }; + + enum class FenceOperation : u32 { + Acquire = 0, + Increment = 1, + }; + + union FenceAction { + u32 raw; + BitField<0, 1, FenceOperation> op; + BitField<8, 24, u32> syncpoint_id; + }; + + explicit Puller(GPU& gpu_, MemoryManager& memory_manager_, DmaPusher& dma_pusher, + Control::ChannelState& channel_state); + ~Puller(); + + void CallMethod(const MethodCall& method_call); + + void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); + + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + + void CallPullerMethod(const MethodCall& method_call); + + void CallEngineMethod(const MethodCall& method_call); + + void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); + +private: + Tegra::GPU& gpu; + + MemoryManager& memory_manager; + DmaPusher& dma_pusher; + Control::ChannelState& channel_state; + VideoCore::RasterizerInterface* rasterizer = nullptr; + + static constexpr std::size_t NUM_REGS = 0x800; + struct Regs { + static constexpr size_t NUM_REGS = 0x40; + + union { + struct { + INSERT_PADDING_WORDS_NOINIT(0x4); + struct { + u32 address_high; + u32 address_low; + + [[nodiscard]] GPUVAddr SemaphoreAddress() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | + address_low); + } + } semaphore_address; + + u32 semaphore_sequence; + u32 semaphore_trigger; + INSERT_PADDING_WORDS_NOINIT(0xC); + + // The pusher and the puller share the reference counter, the pusher only has read + // access + u32 reference_count; + INSERT_PADDING_WORDS_NOINIT(0x5); + + u32 semaphore_acquire; + u32 semaphore_release; + u32 fence_value; + FenceAction fence_action; + INSERT_PADDING_WORDS_NOINIT(0xE2); + + // Puller state + u32 acquire_mode; + u32 acquire_source; + u32 acquire_active; + u32 acquire_timeout; + u32 acquire_value; + }; + std::array<u32, NUM_REGS> reg_array; + }; + } regs{}; + + void ProcessBindMethod(const MethodCall& method_call); + void ProcessFenceActionMethod(); + void ProcessSemaphoreAcquire(); + void ProcessSemaphoreRelease(); + void ProcessSemaphoreTriggerMethod(); + [[nodiscard]] bool ExecuteMethodOnEngine(u32 method); + + /// Mapping of command subchannels to their bound engine ids + std::array<EngineID, 8> bound_engines{}; + + enum class GpuSemaphoreOperation { + AcquireEqual = 0x1, + WriteLong = 0x2, + AcquireGequal = 0x4, + AcquireMask = 0x8, + }; + +#define ASSERT_REG_POSITION(field_name, position) \ + static_assert(offsetof(Regs, field_name) == position * 4, \ + "Field " #field_name " has invalid position") + + ASSERT_REG_POSITION(semaphore_address, 0x4); + ASSERT_REG_POSITION(semaphore_sequence, 0x6); + ASSERT_REG_POSITION(semaphore_trigger, 0x7); + ASSERT_REG_POSITION(reference_count, 0x14); + ASSERT_REG_POSITION(semaphore_acquire, 0x1A); + ASSERT_REG_POSITION(semaphore_release, 0x1B); + ASSERT_REG_POSITION(fence_value, 0x1C); + ASSERT_REG_POSITION(fence_action, 0x1D); + + ASSERT_REG_POSITION(acquire_mode, 0x100); + ASSERT_REG_POSITION(acquire_source, 0x101); + ASSERT_REG_POSITION(acquire_active, 0x102); + ASSERT_REG_POSITION(acquire_timeout, 0x103); + ASSERT_REG_POSITION(acquire_value, 0x104); + +#undef ASSERT_REG_POSITION +}; + +} // namespace Tegra::Engines diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 34dc6c596..c390ac91b 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -1,46 +1,27 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <algorithm> +#include <cstring> +#include <deque> +#include <functional> +#include <memory> #include <queue> #include "common/common_types.h" -#include "common/settings.h" -#include "core/core.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" #include "video_core/rasterizer_interface.h" namespace VideoCommon { class FenceBase { public: - explicit FenceBase(u32 payload_, bool is_stubbed_) - : address{}, payload{payload_}, is_semaphore{false}, is_stubbed{is_stubbed_} {} - - explicit FenceBase(GPUVAddr address_, u32 payload_, bool is_stubbed_) - : address{address_}, payload{payload_}, is_semaphore{true}, is_stubbed{is_stubbed_} {} - - GPUVAddr GetAddress() const { - return address; - } - - u32 GetPayload() const { - return payload; - } - - bool IsSemaphore() const { - return is_semaphore; - } - -private: - GPUVAddr address; - u32 payload; - bool is_semaphore; + explicit FenceBase(bool is_stubbed_) : is_stubbed{is_stubbed_} {} protected: bool is_stubbed; @@ -60,30 +41,28 @@ public: buffer_cache.AccumulateFlushes(); } - void SignalSemaphore(GPUVAddr addr, u32 value) { + void SyncOperation(std::function<void()>&& func) { + uncommitted_operations.emplace_back(std::move(func)); + } + + void SignalFence(std::function<void()>&& func) { TryReleasePendingFences(); const bool should_flush = ShouldFlush(); CommitAsyncFlushes(); - TFence new_fence = CreateFence(addr, value, !should_flush); + uncommitted_operations.emplace_back(std::move(func)); + CommitOperations(); + TFence new_fence = CreateFence(!should_flush); fences.push(new_fence); QueueFence(new_fence); if (should_flush) { rasterizer.FlushCommands(); } - rasterizer.SyncGuestHost(); } void SignalSyncPoint(u32 value) { - TryReleasePendingFences(); - const bool should_flush = ShouldFlush(); - CommitAsyncFlushes(); - TFence new_fence = CreateFence(value, !should_flush); - fences.push(new_fence); - QueueFence(new_fence); - if (should_flush) { - rasterizer.FlushCommands(); - } - rasterizer.SyncGuestHost(); + syncpoint_manager.IncrementGuest(value); + std::function<void()> func([this, value] { syncpoint_manager.IncrementHost(value); }); + SignalFence(std::move(func)); } void WaitPendingFences() { @@ -93,11 +72,10 @@ public: WaitFence(current_fence); } PopAsyncFlushes(); - if (current_fence->IsSemaphore()) { - gpu_memory.template Write<u32>(current_fence->GetAddress(), - current_fence->GetPayload()); - } else { - gpu.IncrementSyncPoint(current_fence->GetPayload()); + auto operations = std::move(pending_operations.front()); + pending_operations.pop_front(); + for (auto& operation : operations) { + operation(); } PopFence(); } @@ -107,16 +85,14 @@ protected: explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, TTextureCache& texture_cache_, TTBufferCache& buffer_cache_, TQueryCache& query_cache_) - : rasterizer{rasterizer_}, gpu{gpu_}, gpu_memory{gpu.MemoryManager()}, + : rasterizer{rasterizer_}, gpu{gpu_}, syncpoint_manager{gpu.Host1x().GetSyncpointManager()}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {} virtual ~FenceManager() = default; - /// Creates a Sync Point Fence Interface, does not create a backend fence if 'is_stubbed' is + /// Creates a Fence Interface, does not create a backend fence if 'is_stubbed' is /// true - virtual TFence CreateFence(u32 value, bool is_stubbed) = 0; - /// Creates a Semaphore Fence Interface, does not create a backend fence if 'is_stubbed' is true - virtual TFence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) = 0; + virtual TFence CreateFence(bool is_stubbed) = 0; /// Queues a fence into the backend if the fence isn't stubbed. virtual void QueueFence(TFence& fence) = 0; /// Notifies that the backend fence has been signaled/reached in host GPU. @@ -126,7 +102,7 @@ protected: VideoCore::RasterizerInterface& rasterizer; Tegra::GPU& gpu; - Tegra::MemoryManager& gpu_memory; + Tegra::Host1x::SyncpointManager& syncpoint_manager; TTextureCache& texture_cache; TTBufferCache& buffer_cache; TQueryCache& query_cache; @@ -139,11 +115,10 @@ private: return; } PopAsyncFlushes(); - if (current_fence->IsSemaphore()) { - gpu_memory.template Write<u32>(current_fence->GetAddress(), - current_fence->GetPayload()); - } else { - gpu.IncrementSyncPoint(current_fence->GetPayload()); + auto operations = std::move(pending_operations.front()); + pending_operations.pop_front(); + for (auto& operation : operations) { + operation(); } PopFence(); } @@ -162,16 +137,20 @@ private: } void PopAsyncFlushes() { - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - texture_cache.PopAsyncFlushes(); - buffer_cache.PopAsyncFlushes(); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.PopAsyncFlushes(); + buffer_cache.PopAsyncFlushes(); + } query_cache.PopAsyncFlushes(); } void CommitAsyncFlushes() { - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - texture_cache.CommitAsyncFlushes(); - buffer_cache.CommitAsyncFlushes(); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CommitAsyncFlushes(); + buffer_cache.CommitAsyncFlushes(); + } query_cache.CommitAsyncFlushes(); } @@ -180,7 +159,13 @@ private: fences.pop(); } + void CommitOperations() { + pending_operations.emplace_back(std::move(uncommitted_operations)); + } + std::queue<TFence> fences; + std::deque<std::function<void()>> uncommitted_operations; + std::deque<std::deque<std::function<void()>>> pending_operations; DelayedDestructionRing<TFence, 6> delayed_destruction_ring; }; diff --git a/src/video_core/framebuffer_config.h b/src/video_core/framebuffer_config.h index b1d455e30..d93f5a37f 100644 --- a/src/video_core/framebuffer_config.h +++ b/src/video_core/framebuffer_config.h @@ -1,46 +1,26 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include "common/common_types.h" #include "common/math_util.h" +#include "core/hle/service/nvflinger/buffer_transform_flags.h" +#include "core/hle/service/nvflinger/pixel_format.h" namespace Tegra { + /** * Struct describing framebuffer configuration */ struct FramebufferConfig { - enum class PixelFormat : u32 { - A8B8G8R8_UNORM = 1, - RGB565_UNORM = 4, - B8G8R8A8_UNORM = 5, - }; - - enum class TransformFlags : u32 { - /// No transform flags are set - Unset = 0x00, - /// Flip source image horizontally (around the vertical axis) - FlipH = 0x01, - /// Flip source image vertically (around the horizontal axis) - FlipV = 0x02, - /// Rotate source image 90 degrees clockwise - Rotate90 = 0x04, - /// Rotate source image 180 degrees - Rotate180 = 0x03, - /// Rotate source image 270 degrees clockwise - Rotate270 = 0x07, - }; - VAddr address{}; u32 offset{}; u32 width{}; u32 height{}; u32 stride{}; - PixelFormat pixel_format{}; - - TransformFlags transform_flags{}; + Service::android::PixelFormat pixel_format{}; + Service::android::BufferTransformFlags transform_flags{}; Common::Rectangle<int> crop_rect; }; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index ba9ba082f..28b38273e 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <atomic> @@ -15,10 +14,11 @@ #include "core/core.h" #include "core/core_timing.h" #include "core/frontend/emu_window.h" -#include "core/hardware_interrupt_manager.h" #include "core/hle/service/nvdrv/nvdata.h" #include "core/perf_stats.h" #include "video_core/cdma_pusher.h" +#include "video_core/control/channel_state.h" +#include "video_core/control/scheduler.h" #include "video_core/dma_pusher.h" #include "video_core/engines/fermi_2d.h" #include "video_core/engines/kepler_compute.h" @@ -27,75 +27,64 @@ #include "video_core/engines/maxwell_dma.h" #include "video_core/gpu.h" #include "video_core/gpu_thread.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" #include "video_core/memory_manager.h" #include "video_core/renderer_base.h" #include "video_core/shader_notify.h" namespace Tegra { -MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); - struct GPU::Impl { explicit Impl(GPU& gpu_, Core::System& system_, bool is_async_, bool use_nvdec_) - : gpu{gpu_}, system{system_}, memory_manager{std::make_unique<Tegra::MemoryManager>( - system)}, - dma_pusher{std::make_unique<Tegra::DmaPusher>(system, gpu)}, use_nvdec{use_nvdec_}, - maxwell_3d{std::make_unique<Engines::Maxwell3D>(system, *memory_manager)}, - fermi_2d{std::make_unique<Engines::Fermi2D>()}, - kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)}, - maxwell_dma{std::make_unique<Engines::MaxwellDMA>(system, *memory_manager)}, - kepler_memory{std::make_unique<Engines::KeplerMemory>(system, *memory_manager)}, + : gpu{gpu_}, system{system_}, host1x{system.Host1x()}, use_nvdec{use_nvdec_}, shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_}, - gpu_thread{system_, is_async_} {} + gpu_thread{system_, is_async_}, scheduler{std::make_unique<Control::Scheduler>(gpu)} {} ~Impl() = default; - /// Binds a renderer to the GPU. - void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { - renderer = std::move(renderer_); - rasterizer = renderer->ReadRasterizer(); - - memory_manager->BindRasterizer(rasterizer); - maxwell_3d->BindRasterizer(rasterizer); - fermi_2d->BindRasterizer(rasterizer); - kepler_compute->BindRasterizer(rasterizer); - kepler_memory->BindRasterizer(rasterizer); - maxwell_dma->BindRasterizer(rasterizer); + std::shared_ptr<Control::ChannelState> CreateChannel(s32 channel_id) { + auto channel_state = std::make_shared<Tegra::Control::ChannelState>(channel_id); + channels.emplace(channel_id, channel_state); + scheduler->DeclareChannel(channel_state); + return channel_state; } - /// Calls a GPU method. - void CallMethod(const GPU::MethodCall& method_call) { - LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method_call.method, - method_call.subchannel); + void BindChannel(s32 channel_id) { + if (bound_channel == channel_id) { + return; + } + auto it = channels.find(channel_id); + ASSERT(it != channels.end()); + bound_channel = channel_id; + current_channel = it->second.get(); - ASSERT(method_call.subchannel < bound_engines.size()); + rasterizer->BindChannel(*current_channel); + } - if (ExecuteMethodOnEngine(method_call.method)) { - CallEngineMethod(method_call); - } else { - CallPullerMethod(method_call); - } + std::shared_ptr<Control::ChannelState> AllocateChannel() { + return CreateChannel(new_channel_id++); } - /// Calls a GPU multivalue method. - void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel); + void InitChannel(Control::ChannelState& to_init) { + to_init.Init(system, gpu); + to_init.BindRasterizer(rasterizer); + rasterizer->InitializeChannel(to_init); + } - ASSERT(subchannel < bound_engines.size()); + void InitAddressSpace(Tegra::MemoryManager& memory_manager) { + memory_manager.BindRasterizer(rasterizer); + } - if (ExecuteMethodOnEngine(method)) { - CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); - } else { - for (std::size_t i = 0; i < amount; i++) { - CallPullerMethod(GPU::MethodCall{ - method, - base_start[i], - subchannel, - methods_pending - static_cast<u32>(i), - }); - } - } + void ReleaseChannel(Control::ChannelState& to_release) { + UNIMPLEMENTED(); + } + + /// Binds a renderer to the GPU. + void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { + renderer = std::move(renderer_); + rasterizer = renderer->ReadRasterizer(); + host1x.MemoryManager().BindRasterizer(rasterizer); } /// Flush all current written commands into the host GPU for execution. @@ -104,85 +93,82 @@ struct GPU::Impl { } /// Synchronizes CPU writes with Host GPU memory. - void SyncGuestHost() { - rasterizer->SyncGuestHost(); + void InvalidateGPUCache() { + rasterizer->InvalidateGPUCache(); } /// Signal the ending of command list. void OnCommandListEnd() { - if (is_async) { - // This command only applies to asynchronous GPU mode - gpu_thread.OnCommandListEnd(); - } + gpu_thread.OnCommandListEnd(); } /// Request a host GPU memory flush from the CPU. - [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size) { - std::unique_lock lck{flush_request_mutex}; - const u64 fence = ++last_flush_fence; - flush_requests.emplace_back(fence, addr, size); + template <typename Func> + [[nodiscard]] u64 RequestSyncOperation(Func&& action) { + std::unique_lock lck{sync_request_mutex}; + const u64 fence = ++last_sync_fence; + sync_requests.emplace_back(action); return fence; } /// Obtains current flush request fence id. - [[nodiscard]] u64 CurrentFlushRequestFence() const { - return current_flush_fence.load(std::memory_order_relaxed); + [[nodiscard]] u64 CurrentSyncRequestFence() const { + return current_sync_fence.load(std::memory_order_relaxed); + } + + void WaitForSyncOperation(const u64 fence) { + std::unique_lock lck{sync_request_mutex}; + sync_request_cv.wait(lck, [this, fence] { return CurrentSyncRequestFence() >= fence; }); } /// Tick pending requests within the GPU. void TickWork() { - std::unique_lock lck{flush_request_mutex}; - while (!flush_requests.empty()) { - auto& request = flush_requests.front(); - const u64 fence = request.fence; - const VAddr addr = request.addr; - const std::size_t size = request.size; - flush_requests.pop_front(); - flush_request_mutex.unlock(); - rasterizer->FlushRegion(addr, size); - current_flush_fence.store(fence); - flush_request_mutex.lock(); + std::unique_lock lck{sync_request_mutex}; + while (!sync_requests.empty()) { + auto request = std::move(sync_requests.front()); + sync_requests.pop_front(); + sync_request_mutex.unlock(); + request(); + current_sync_fence.fetch_add(1, std::memory_order_release); + sync_request_mutex.lock(); + sync_request_cv.notify_all(); } } /// Returns a reference to the Maxwell3D GPU engine. [[nodiscard]] Engines::Maxwell3D& Maxwell3D() { - return *maxwell_3d; + ASSERT(current_channel); + return *current_channel->maxwell_3d; } /// Returns a const reference to the Maxwell3D GPU engine. [[nodiscard]] const Engines::Maxwell3D& Maxwell3D() const { - return *maxwell_3d; + ASSERT(current_channel); + return *current_channel->maxwell_3d; } /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] Engines::KeplerCompute& KeplerCompute() { - return *kepler_compute; + ASSERT(current_channel); + return *current_channel->kepler_compute; } /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] const Engines::KeplerCompute& KeplerCompute() const { - return *kepler_compute; - } - - /// Returns a reference to the GPU memory manager. - [[nodiscard]] Tegra::MemoryManager& MemoryManager() { - return *memory_manager; - } - - /// Returns a const reference to the GPU memory manager. - [[nodiscard]] const Tegra::MemoryManager& MemoryManager() const { - return *memory_manager; + ASSERT(current_channel); + return *current_channel->kepler_compute; } /// Returns a reference to the GPU DMA pusher. [[nodiscard]] Tegra::DmaPusher& DmaPusher() { - return *dma_pusher; + ASSERT(current_channel); + return *current_channel->dma_pusher; } /// Returns a const reference to the GPU DMA pusher. [[nodiscard]] const Tegra::DmaPusher& DmaPusher() const { - return *dma_pusher; + ASSERT(current_channel); + return *current_channel->dma_pusher; } /// Returns a reference to the underlying renderer. @@ -205,77 +191,6 @@ struct GPU::Impl { return *shader_notify; } - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. - void WaitFence(u32 syncpoint_id, u32 value) { - // Synced GPU, is always in sync - if (!is_async) { - return; - } - if (syncpoint_id == UINT32_MAX) { - // TODO: Research what this does. - LOG_ERROR(HW_GPU, "Waiting for syncpoint -1 not implemented"); - return; - } - MICROPROFILE_SCOPE(GPU_wait); - std::unique_lock lock{sync_mutex}; - sync_cv.wait(lock, [=, this] { - if (shutting_down.load(std::memory_order_relaxed)) { - // We're shutting down, ensure no threads continue to wait for the next syncpoint - return true; - } - return syncpoints.at(syncpoint_id).load() >= value; - }); - } - - void IncrementSyncPoint(u32 syncpoint_id) { - auto& syncpoint = syncpoints.at(syncpoint_id); - syncpoint++; - std::lock_guard lock{sync_mutex}; - sync_cv.notify_all(); - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - if (!interrupt.empty()) { - u32 value = syncpoint.load(); - auto it = interrupt.begin(); - while (it != interrupt.end()) { - if (value >= *it) { - TriggerCpuInterrupt(syncpoint_id, *it); - it = interrupt.erase(it); - continue; - } - it++; - } - } - } - - [[nodiscard]] u32 GetSyncpointValue(u32 syncpoint_id) const { - return syncpoints.at(syncpoint_id).load(); - } - - void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value) { - std::lock_guard lock{sync_mutex}; - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - bool contains = std::any_of(interrupt.begin(), interrupt.end(), - [value](u32 in_value) { return in_value == value; }); - if (contains) { - return; - } - interrupt.emplace_back(value); - } - - [[nodiscard]] bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value) { - std::lock_guard lock{sync_mutex}; - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - const auto iter = - std::find_if(interrupt.begin(), interrupt.end(), - [value](u32 interrupt_value) { return value == interrupt_value; }); - - if (iter == interrupt.end()) { - return false; - } - interrupt.erase(iter); - return true; - } - [[nodiscard]] u64 GetTicks() const { // This values were reversed engineered by fincs from NVN // The gpu clock is reported in units of 385/625 nanoseconds @@ -307,7 +222,7 @@ struct GPU::Impl { /// This can be used to launch any necessary threads and register any necessary /// core timing events. void Start() { - gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher); + gpu_thread.StartThread(*renderer, renderer->Context(), *scheduler); cpu_context = renderer->GetRenderWindow().CreateSharedContext(); cpu_context->MakeCurrent(); } @@ -329,8 +244,8 @@ struct GPU::Impl { } /// Push GPU command entries to be processed - void PushGPUEntries(Tegra::CommandList&& entries) { - gpu_thread.SubmitList(std::move(entries)); + void PushGPUEntries(s32 channel, Tegra::CommandList&& entries) { + gpu_thread.SubmitList(channel, std::move(entries)); } /// Push GPU command buffer entries to be processed @@ -340,7 +255,7 @@ struct GPU::Impl { } if (!cdma_pushers.contains(id)) { - cdma_pushers.insert_or_assign(id, std::make_unique<Tegra::CDmaPusher>(gpu)); + cdma_pushers.insert_or_assign(id, std::make_unique<Tegra::CDmaPusher>(host1x)); } // SubmitCommandBuffer would make the nvdec operations async, this is not currently working @@ -377,308 +292,55 @@ struct GPU::Impl { gpu_thread.FlushAndInvalidateRegion(addr, size); } - void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const { - auto& interrupt_manager = system.InterruptManager(); - interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value); - } - - void ProcessBindMethod(const GPU::MethodCall& method_call) { - // Bind the current subchannel to the desired engine id. - LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, - method_call.argument); - const auto engine_id = static_cast<EngineID>(method_call.argument); - bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id); - switch (engine_id) { - case EngineID::FERMI_TWOD_A: - dma_pusher->BindSubchannel(fermi_2d.get(), method_call.subchannel); - break; - case EngineID::MAXWELL_B: - dma_pusher->BindSubchannel(maxwell_3d.get(), method_call.subchannel); - break; - case EngineID::KEPLER_COMPUTE_B: - dma_pusher->BindSubchannel(kepler_compute.get(), method_call.subchannel); - break; - case EngineID::MAXWELL_DMA_COPY_A: - dma_pusher->BindSubchannel(maxwell_dma.get(), method_call.subchannel); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - dma_pusher->BindSubchannel(kepler_memory.get(), method_call.subchannel); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", engine_id); - } - } - - void ProcessFenceActionMethod() { - switch (regs.fence_action.op) { - case GPU::FenceOperation::Acquire: - WaitFence(regs.fence_action.syncpoint_id, regs.fence_value); - break; - case GPU::FenceOperation::Increment: - IncrementSyncPoint(regs.fence_action.syncpoint_id); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented operation {}", regs.fence_action.op.Value()); - } - } - - void ProcessWaitForInterruptMethod() { - // TODO(bunnei) ImplementMe - LOG_WARNING(HW_GPU, "(STUBBED) called"); - } - - void ProcessSemaphoreTriggerMethod() { - const auto semaphoreOperationMask = 0xF; - const auto op = - static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask); - if (op == GpuSemaphoreOperation::WriteLong) { - struct Block { - u32 sequence; - u32 zeros = 0; - u64 timestamp; - }; - - Block block{}; - block.sequence = regs.semaphore_sequence; - // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of - // CoreTiming - block.timestamp = GetTicks(); - memory_manager->WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block, - sizeof(block)); - } else { - const u32 word{memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress())}; - if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) || - (op == GpuSemaphoreOperation::AcquireGequal && - static_cast<s32>(word - regs.semaphore_sequence) > 0) || - (op == GpuSemaphoreOperation::AcquireMask && (word & regs.semaphore_sequence))) { - // Nothing to do in this case + void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) { + size_t current_request_counter{}; + { + std::unique_lock<std::mutex> lk(request_swap_mutex); + if (free_swap_counters.empty()) { + current_request_counter = request_swap_counters.size(); + request_swap_counters.emplace_back(num_fences); } else { - regs.acquire_source = true; - regs.acquire_value = regs.semaphore_sequence; - if (op == GpuSemaphoreOperation::AcquireEqual) { - regs.acquire_active = true; - regs.acquire_mode = false; - } else if (op == GpuSemaphoreOperation::AcquireGequal) { - regs.acquire_active = true; - regs.acquire_mode = true; - } else if (op == GpuSemaphoreOperation::AcquireMask) { - // TODO(kemathe) The acquire mask operation waits for a value that, ANDed with - // semaphore_sequence, gives a non-0 result - LOG_ERROR(HW_GPU, "Invalid semaphore operation AcquireMask not implemented"); - } else { - LOG_ERROR(HW_GPU, "Invalid semaphore operation"); - } + current_request_counter = free_swap_counters.front(); + request_swap_counters[current_request_counter] = num_fences; + free_swap_counters.pop_front(); } } - } - - void ProcessSemaphoreRelease() { - memory_manager->Write<u32>(regs.semaphore_address.SemaphoreAddress(), - regs.semaphore_release); - } - - void ProcessSemaphoreAcquire() { - const u32 word = memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress()); - const auto value = regs.semaphore_acquire; - if (word != value) { - regs.acquire_active = true; - regs.acquire_value = value; - // TODO(kemathe73) figure out how to do the acquire_timeout - regs.acquire_mode = false; - regs.acquire_source = false; - } - } - - /// Calls a GPU puller method. - void CallPullerMethod(const GPU::MethodCall& method_call) { - regs.reg_array[method_call.method] = method_call.argument; - const auto method = static_cast<BufferMethods>(method_call.method); - - switch (method) { - case BufferMethods::BindObject: { - ProcessBindMethod(method_call); - break; - } - case BufferMethods::Nop: - case BufferMethods::SemaphoreAddressHigh: - case BufferMethods::SemaphoreAddressLow: - case BufferMethods::SemaphoreSequence: - break; - case BufferMethods::UnkCacheFlush: - rasterizer->SyncGuestHost(); - break; - case BufferMethods::WrcacheFlush: - rasterizer->SignalReference(); - break; - case BufferMethods::FenceValue: - break; - case BufferMethods::RefCnt: - rasterizer->SignalReference(); - break; - case BufferMethods::FenceAction: - ProcessFenceActionMethod(); - break; - case BufferMethods::WaitForInterrupt: - rasterizer->WaitForIdle(); - break; - case BufferMethods::SemaphoreTrigger: { - ProcessSemaphoreTriggerMethod(); - break; - } - case BufferMethods::NotifyIntr: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented"); - break; - } - case BufferMethods::Unk28: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented"); - break; - } - case BufferMethods::SemaphoreAcquire: { - ProcessSemaphoreAcquire(); - break; - } - case BufferMethods::SemaphoreRelease: { - ProcessSemaphoreRelease(); - break; - } - case BufferMethods::Yield: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented"); - break; - } - default: - LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented", method); - break; - } - } - - /// Calls a GPU engine method. - void CallEngineMethod(const GPU::MethodCall& method_call) { - const EngineID engine = bound_engines[method_call.subchannel]; - - switch (engine) { - case EngineID::FERMI_TWOD_A: - fermi_2d->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::MAXWELL_B: - maxwell_3d->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::KEPLER_COMPUTE_B: - kepler_compute->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::MAXWELL_DMA_COPY_A: - maxwell_dma->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - kepler_memory->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine"); - } - } - - /// Calls a GPU engine multivalue method. - void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - const EngineID engine = bound_engines[subchannel]; - - switch (engine) { - case EngineID::FERMI_TWOD_A: - fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::MAXWELL_B: - maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::KEPLER_COMPUTE_B: - kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::MAXWELL_DMA_COPY_A: - maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine"); - } - } - - /// Determines where the method should be executed. - [[nodiscard]] bool ExecuteMethodOnEngine(u32 method) { - const auto buffer_method = static_cast<BufferMethods>(method); - return buffer_method >= BufferMethods::NonPullerMethods; - } - - struct Regs { - static constexpr size_t NUM_REGS = 0x40; - - union { - struct { - INSERT_PADDING_WORDS_NOINIT(0x4); - struct { - u32 address_high; - u32 address_low; - - [[nodiscard]] GPUVAddr SemaphoreAddress() const { - return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | - address_low); + const auto wait_fence = + RequestSyncOperation([this, current_request_counter, framebuffer, fences, num_fences] { + auto& syncpoint_manager = host1x.GetSyncpointManager(); + if (num_fences == 0) { + renderer->SwapBuffers(framebuffer); + } + const auto executer = [this, current_request_counter, + framebuffer_copy = *framebuffer]() { + { + std::unique_lock<std::mutex> lk(request_swap_mutex); + if (--request_swap_counters[current_request_counter] != 0) { + return; + } + free_swap_counters.push_back(current_request_counter); } - } semaphore_address; - - u32 semaphore_sequence; - u32 semaphore_trigger; - INSERT_PADDING_WORDS_NOINIT(0xC); - - // The pusher and the puller share the reference counter, the pusher only has read - // access - u32 reference_count; - INSERT_PADDING_WORDS_NOINIT(0x5); - - u32 semaphore_acquire; - u32 semaphore_release; - u32 fence_value; - GPU::FenceAction fence_action; - INSERT_PADDING_WORDS_NOINIT(0xE2); - - // Puller state - u32 acquire_mode; - u32 acquire_source; - u32 acquire_active; - u32 acquire_timeout; - u32 acquire_value; - }; - std::array<u32, NUM_REGS> reg_array; - }; - } regs{}; + renderer->SwapBuffers(&framebuffer_copy); + }; + for (size_t i = 0; i < num_fences; i++) { + syncpoint_manager.RegisterGuestAction(fences[i].id, fences[i].value, executer); + } + }); + gpu_thread.TickGPU(); + WaitForSyncOperation(wait_fence); + } GPU& gpu; Core::System& system; - std::unique_ptr<Tegra::MemoryManager> memory_manager; - std::unique_ptr<Tegra::DmaPusher> dma_pusher; + Host1x::Host1x& host1x; + std::map<u32, std::unique_ptr<Tegra::CDmaPusher>> cdma_pushers; std::unique_ptr<VideoCore::RendererBase> renderer; VideoCore::RasterizerInterface* rasterizer = nullptr; const bool use_nvdec; - /// Mapping of command subchannels to their bound engine ids - std::array<EngineID, 8> bound_engines{}; - /// 3D engine - std::unique_ptr<Engines::Maxwell3D> maxwell_3d; - /// 2D engine - std::unique_ptr<Engines::Fermi2D> fermi_2d; - /// Compute engine - std::unique_ptr<Engines::KeplerCompute> kepler_compute; - /// DMA engine - std::unique_ptr<Engines::MaxwellDMA> maxwell_dma; - /// Inline memory engine - std::unique_ptr<Engines::KeplerMemory> kepler_memory; + s32 new_channel_id{1}; /// Shader build notifier std::unique_ptr<VideoCore::ShaderNotify> shader_notify; /// When true, we are about to shut down emulation session, so terminate outstanding tasks @@ -693,51 +355,25 @@ struct GPU::Impl { std::condition_variable sync_cv; - struct FlushRequest { - explicit FlushRequest(u64 fence_, VAddr addr_, std::size_t size_) - : fence{fence_}, addr{addr_}, size{size_} {} - u64 fence; - VAddr addr; - std::size_t size; - }; - - std::list<FlushRequest> flush_requests; - std::atomic<u64> current_flush_fence{}; - u64 last_flush_fence{}; - std::mutex flush_request_mutex; + std::list<std::function<void()>> sync_requests; + std::atomic<u64> current_sync_fence{}; + u64 last_sync_fence{}; + std::mutex sync_request_mutex; + std::condition_variable sync_request_cv; const bool is_async; VideoCommon::GPUThread::ThreadManager gpu_thread; std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context; -#define ASSERT_REG_POSITION(field_name, position) \ - static_assert(offsetof(Regs, field_name) == position * 4, \ - "Field " #field_name " has invalid position") - - ASSERT_REG_POSITION(semaphore_address, 0x4); - ASSERT_REG_POSITION(semaphore_sequence, 0x6); - ASSERT_REG_POSITION(semaphore_trigger, 0x7); - ASSERT_REG_POSITION(reference_count, 0x14); - ASSERT_REG_POSITION(semaphore_acquire, 0x1A); - ASSERT_REG_POSITION(semaphore_release, 0x1B); - ASSERT_REG_POSITION(fence_value, 0x1C); - ASSERT_REG_POSITION(fence_action, 0x1D); - - ASSERT_REG_POSITION(acquire_mode, 0x100); - ASSERT_REG_POSITION(acquire_source, 0x101); - ASSERT_REG_POSITION(acquire_active, 0x102); - ASSERT_REG_POSITION(acquire_timeout, 0x103); - ASSERT_REG_POSITION(acquire_value, 0x104); - -#undef ASSERT_REG_POSITION - - enum class GpuSemaphoreOperation { - AcquireEqual = 0x1, - WriteLong = 0x2, - AcquireGequal = 0x4, - AcquireMask = 0x8, - }; + std::unique_ptr<Tegra::Control::Scheduler> scheduler; + std::unordered_map<s32, std::shared_ptr<Tegra::Control::ChannelState>> channels; + Tegra::Control::ChannelState* current_channel; + s32 bound_channel{-1}; + + std::deque<size_t> free_swap_counters; + std::deque<size_t> request_swap_counters; + std::mutex request_swap_mutex; }; GPU::GPU(Core::System& system, bool is_async, bool use_nvdec) @@ -745,25 +381,36 @@ GPU::GPU(Core::System& system, bool is_async, bool use_nvdec) GPU::~GPU() = default; -void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer) { - impl->BindRenderer(std::move(renderer)); +std::shared_ptr<Control::ChannelState> GPU::AllocateChannel() { + return impl->AllocateChannel(); +} + +void GPU::InitChannel(Control::ChannelState& to_init) { + impl->InitChannel(to_init); +} + +void GPU::BindChannel(s32 channel_id) { + impl->BindChannel(channel_id); } -void GPU::CallMethod(const MethodCall& method_call) { - impl->CallMethod(method_call); +void GPU::ReleaseChannel(Control::ChannelState& to_release) { + impl->ReleaseChannel(to_release); } -void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - impl->CallMultiMethod(method, subchannel, base_start, amount, methods_pending); +void GPU::InitAddressSpace(Tegra::MemoryManager& memory_manager) { + impl->InitAddressSpace(memory_manager); +} + +void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer) { + impl->BindRenderer(std::move(renderer)); } void GPU::FlushCommands() { impl->FlushCommands(); } -void GPU::SyncGuestHost() { - impl->SyncGuestHost(); +void GPU::InvalidateGPUCache() { + impl->InvalidateGPUCache(); } void GPU::OnCommandListEnd() { @@ -771,17 +418,32 @@ void GPU::OnCommandListEnd() { } u64 GPU::RequestFlush(VAddr addr, std::size_t size) { - return impl->RequestFlush(addr, size); + return impl->RequestSyncOperation( + [this, addr, size]() { impl->rasterizer->FlushRegion(addr, size); }); } -u64 GPU::CurrentFlushRequestFence() const { - return impl->CurrentFlushRequestFence(); +u64 GPU::CurrentSyncRequestFence() const { + return impl->CurrentSyncRequestFence(); +} + +void GPU::WaitForSyncOperation(u64 fence) { + return impl->WaitForSyncOperation(fence); } void GPU::TickWork() { impl->TickWork(); } +/// Gets a mutable reference to the Host1x interface +Host1x::Host1x& GPU::Host1x() { + return impl->host1x; +} + +/// Gets an immutable reference to the Host1x interface. +const Host1x::Host1x& GPU::Host1x() const { + return impl->host1x; +} + Engines::Maxwell3D& GPU::Maxwell3D() { return impl->Maxwell3D(); } @@ -798,14 +460,6 @@ const Engines::KeplerCompute& GPU::KeplerCompute() const { return impl->KeplerCompute(); } -Tegra::MemoryManager& GPU::MemoryManager() { - return impl->MemoryManager(); -} - -const Tegra::MemoryManager& GPU::MemoryManager() const { - return impl->MemoryManager(); -} - Tegra::DmaPusher& GPU::DmaPusher() { return impl->DmaPusher(); } @@ -830,24 +484,9 @@ const VideoCore::ShaderNotify& GPU::ShaderNotify() const { return impl->ShaderNotify(); } -void GPU::WaitFence(u32 syncpoint_id, u32 value) { - impl->WaitFence(syncpoint_id, value); -} - -void GPU::IncrementSyncPoint(u32 syncpoint_id) { - impl->IncrementSyncPoint(syncpoint_id); -} - -u32 GPU::GetSyncpointValue(u32 syncpoint_id) const { - return impl->GetSyncpointValue(syncpoint_id); -} - -void GPU::RegisterSyncptInterrupt(u32 syncpoint_id, u32 value) { - impl->RegisterSyncptInterrupt(syncpoint_id, value); -} - -bool GPU::CancelSyncptInterrupt(u32 syncpoint_id, u32 value) { - return impl->CancelSyncptInterrupt(syncpoint_id, value); +void GPU::RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) { + impl->RequestSwapBuffers(framebuffer, fences, num_fences); } u64 GPU::GetTicks() const { @@ -882,8 +521,8 @@ void GPU::ReleaseContext() { impl->ReleaseContext(); } -void GPU::PushGPUEntries(Tegra::CommandList&& entries) { - impl->PushGPUEntries(std::move(entries)); +void GPU::PushGPUEntries(s32 channel, Tegra::CommandList&& entries) { + impl->PushGPUEntries(channel, std::move(entries)); } void GPU::PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries) { diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index 26b8ea233..0a4a8b14f 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,6 +7,7 @@ #include "common/bit_field.h" #include "common/common_types.h" +#include "core/hle/service/nvdrv/nvdata.h" #include "video_core/cdma_pusher.h" #include "video_core/framebuffer_config.h" @@ -89,73 +89,58 @@ class Maxwell3D; class KeplerCompute; } // namespace Engines -enum class EngineID { - FERMI_TWOD_A = 0x902D, // 2D Engine - MAXWELL_B = 0xB197, // 3D Engine - KEPLER_COMPUTE_B = 0xB1C0, - KEPLER_INLINE_TO_MEMORY_B = 0xA140, - MAXWELL_DMA_COPY_A = 0xB0B5, -}; +namespace Control { +struct ChannelState; +} + +namespace Host1x { +class Host1x; +} // namespace Host1x class MemoryManager; class GPU final { public: - struct MethodCall { - u32 method{}; - u32 argument{}; - u32 subchannel{}; - u32 method_count{}; - - explicit MethodCall(u32 method_, u32 argument_, u32 subchannel_ = 0, u32 method_count_ = 0) - : method(method_), argument(argument_), subchannel(subchannel_), - method_count(method_count_) {} - - [[nodiscard]] bool IsLastCall() const { - return method_count <= 1; - } - }; - - enum class FenceOperation : u32 { - Acquire = 0, - Increment = 1, - }; - - union FenceAction { - u32 raw; - BitField<0, 1, FenceOperation> op; - BitField<8, 24, u32> syncpoint_id; - }; - explicit GPU(Core::System& system, bool is_async, bool use_nvdec); ~GPU(); /// Binds a renderer to the GPU. void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer); - /// Calls a GPU method. - void CallMethod(const MethodCall& method_call); - - /// Calls a GPU multivalue method. - void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending); - /// Flush all current written commands into the host GPU for execution. void FlushCommands(); /// Synchronizes CPU writes with Host GPU memory. - void SyncGuestHost(); + void InvalidateGPUCache(); /// Signal the ending of command list. void OnCommandListEnd(); + std::shared_ptr<Control::ChannelState> AllocateChannel(); + + void InitChannel(Control::ChannelState& to_init); + + void BindChannel(s32 channel_id); + + void ReleaseChannel(Control::ChannelState& to_release); + + void InitAddressSpace(Tegra::MemoryManager& memory_manager); + /// Request a host GPU memory flush from the CPU. [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size); /// Obtains current flush request fence id. - [[nodiscard]] u64 CurrentFlushRequestFence() const; + [[nodiscard]] u64 CurrentSyncRequestFence() const; + + void WaitForSyncOperation(u64 fence); /// Tick pending requests within the GPU. void TickWork(); + /// Gets a mutable reference to the Host1x interface + [[nodiscard]] Host1x::Host1x& Host1x(); + + /// Gets an immutable reference to the Host1x interface. + [[nodiscard]] const Host1x::Host1x& Host1x() const; + /// Returns a reference to the Maxwell3D GPU engine. [[nodiscard]] Engines::Maxwell3D& Maxwell3D(); @@ -168,12 +153,6 @@ public: /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] const Engines::KeplerCompute& KeplerCompute() const; - /// Returns a reference to the GPU memory manager. - [[nodiscard]] Tegra::MemoryManager& MemoryManager(); - - /// Returns a const reference to the GPU memory manager. - [[nodiscard]] const Tegra::MemoryManager& MemoryManager() const; - /// Returns a reference to the GPU DMA pusher. [[nodiscard]] Tegra::DmaPusher& DmaPusher(); @@ -192,17 +171,6 @@ public: /// Returns a const reference to the shader notifier. [[nodiscard]] const VideoCore::ShaderNotify& ShaderNotify() const; - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. - void WaitFence(u32 syncpoint_id, u32 value); - - void IncrementSyncPoint(u32 syncpoint_id); - - [[nodiscard]] u32 GetSyncpointValue(u32 syncpoint_id) const; - - void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value); - - [[nodiscard]] bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value); - [[nodiscard]] u64 GetTicks() const; [[nodiscard]] bool IsAsync() const; @@ -211,6 +179,9 @@ public: void RendererFrameEndNotify(); + void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences); + /// Performs any additional setup necessary in order to begin GPU emulation. /// This can be used to launch any necessary threads and register any necessary /// core timing events. @@ -226,7 +197,7 @@ public: void ReleaseContext(); /// Push GPU command entries to be processed - void PushGPUEntries(Tegra::CommandList&& entries); + void PushGPUEntries(s32 channel, Tegra::CommandList&& entries); /// Push GPU command buffer entries to be processed void PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries); @@ -248,7 +219,7 @@ public: private: struct Impl; - std::unique_ptr<Impl> impl; + mutable std::unique_ptr<Impl> impl; }; } // namespace Tegra diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 9547f277a..1bd477011 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" #include "common/microprofile.h" @@ -9,6 +8,7 @@ #include "common/thread.h" #include "core/core.h" #include "core/frontend/emu_window.h" +#include "video_core/control/scheduler.h" #include "video_core/dma_pusher.h" #include "video_core/gpu.h" #include "video_core/gpu_thread.h" @@ -19,8 +19,8 @@ namespace VideoCommon::GPUThread { /// Runs the GPU thread static void RunThread(std::stop_token stop_token, Core::System& system, VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher, SynchState& state) { - std::string name = "yuzu:GPU"; + Tegra::Control::Scheduler& scheduler, SynchState& state) { + std::string name = "GPU"; MicroProfileOnThreadCreate(name.c_str()); SCOPE_EXIT({ MicroProfileOnThreadExit(); }); @@ -37,8 +37,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system, break; } if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) { - dma_pusher.Push(std::move(submit_list->entries)); - dma_pusher.DispatchCalls(); + scheduler.Push(submit_list->channel, std::move(submit_list->entries)); } else if (const auto* data = std::get_if<SwapBuffersCommand>(&next.data)) { renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr); } else if (std::holds_alternative<OnCommandListEndCommand>(next.data)) { @@ -50,13 +49,13 @@ static void RunThread(std::stop_token stop_token, Core::System& system, } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); } else { - UNREACHABLE(); + ASSERT(false); } state.signaled_fence.store(next.fence); if (next.block) { // We have to lock the write_lock to ensure that the condition_variable wait not get a // race between the check and the lock itself. - std::lock_guard lk(state.write_lock); + std::scoped_lock lk{state.write_lock}; state.cv.notify_all(); } } @@ -69,14 +68,14 @@ ThreadManager::~ThreadManager() = default; void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher) { + Tegra::Control::Scheduler& scheduler) { rasterizer = renderer.ReadRasterizer(); thread = std::jthread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), - std::ref(dma_pusher), std::ref(state)); + std::ref(scheduler), std::ref(state)); } -void ThreadManager::SubmitList(Tegra::CommandList&& entries) { - PushCommand(SubmitListCommand(std::move(entries))); +void ThreadManager::SubmitList(s32 channel, Tegra::CommandList&& entries) { + PushCommand(SubmitListCommand(channel, std::move(entries))); } void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { @@ -94,8 +93,12 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) { } auto& gpu = system.GPU(); u64 fence = gpu.RequestFlush(addr, size); - PushCommand(GPUTickCommand(), true); - ASSERT(fence <= gpu.CurrentFlushRequestFence()); + TickGPU(); + gpu.WaitForSyncOperation(fence); +} + +void ThreadManager::TickGPU() { + PushCommand(GPUTickCommand()); } void ThreadManager::InvalidateRegion(VAddr addr, u64 size) { diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 00984188e..64628d3e3 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -16,7 +15,9 @@ namespace Tegra { struct FramebufferConfig; -class DmaPusher; +namespace Control { +class Scheduler; +} } // namespace Tegra namespace Core { @@ -35,8 +36,10 @@ namespace VideoCommon::GPUThread { /// Command to signal to the GPU thread that a command list is ready for processing struct SubmitListCommand final { - explicit SubmitListCommand(Tegra::CommandList&& entries_) : entries{std::move(entries_)} {} + explicit SubmitListCommand(s32 channel_, Tegra::CommandList&& entries_) + : channel{channel_}, entries{std::move(entries_)} {} + s32 channel; Tegra::CommandList entries; }; @@ -97,7 +100,7 @@ struct CommandDataContainer { /// Struct used to synchronize the GPU thread struct SynchState final { - using CommandQueue = Common::SPSCQueue<CommandDataContainer, true>; + using CommandQueue = Common::MPSCQueue<CommandDataContainer, true>; std::mutex write_lock; CommandQueue queue; u64 last_fence{}; @@ -113,10 +116,10 @@ public: /// Creates and starts the GPU thread. void StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher); + Tegra::Control::Scheduler& scheduler); /// Push GPU command entries to be processed - void SubmitList(Tegra::CommandList&& entries); + void SubmitList(s32 channel, Tegra::CommandList&& entries); /// Swap buffers (render frame) void SwapBuffers(const Tegra::FramebufferConfig* framebuffer); @@ -132,6 +135,8 @@ public: void OnCommandListEnd(); + void TickGPU(); + private: /// Pushes a command to be executed by the GPU thread u64 PushCommand(CommandData&& command_data, bool block = false); diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/host1x/codecs/codec.cpp index 04d0f3a2f..42e7d6e4f 100644 --- a/src/video_core/command_classes/codecs/codec.cpp +++ b/src/video_core/host1x/codecs/codec.cpp @@ -1,18 +1,16 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> -#include <cstdio> #include <fstream> #include <vector> #include "common/assert.h" #include "common/settings.h" -#include "video_core/command_classes/codecs/codec.h" -#include "video_core/command_classes/codecs/h264.h" -#include "video_core/command_classes/codecs/vp8.h" -#include "video_core/command_classes/codecs/vp9.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/codec.h" +#include "video_core/host1x/codecs/h264.h" +#include "video_core/host1x/codecs/vp8.h" +#include "video_core/host1x/codecs/vp9.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" extern "C" { @@ -57,16 +55,28 @@ AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pi av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT; return PREFERRED_CPU_FMT; } + +// List all the currently available hwcontext in ffmpeg +std::vector<AVHWDeviceType> ListSupportedContexts() { + std::vector<AVHWDeviceType> contexts{}; + AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE; + do { + current_device_type = av_hwdevice_iterate_types(current_device_type); + contexts.push_back(current_device_type); + } while (current_device_type != AV_HWDEVICE_TYPE_NONE); + return contexts; +} + } // namespace void AVFrameDeleter(AVFrame* ptr) { av_frame_free(&ptr); } -Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs) - : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)), - vp8_decoder(std::make_unique<Decoder::VP8>(gpu)), - vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {} +Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs) + : host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)), + vp8_decoder(std::make_unique<Decoder::VP8>(host1x)), + vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {} Codec::~Codec() { if (!initialized) { @@ -77,17 +87,6 @@ Codec::~Codec() { av_buffer_unref(&av_gpu_decoder); } -// List all the currently available hwcontext in ffmpeg -static std::vector<AVHWDeviceType> ListSupportedContexts() { - std::vector<AVHWDeviceType> contexts{}; - AVHWDeviceType current_device_type = AV_HWDEVICE_TYPE_NONE; - do { - current_device_type = av_hwdevice_iterate_types(current_device_type); - contexts.push_back(current_device_type); - } while (current_device_type != AV_HWDEVICE_TYPE_NONE); - return contexts; -} - bool Codec::CreateGpuAvDevice() { static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX; static const auto supported_contexts = ListSupportedContexts(); @@ -97,6 +96,8 @@ bool Codec::CreateGpuAvDevice() { LOG_DEBUG(Service_NVDRV, "{} explicitly unsupported", av_hwdevice_get_type_name(type)); continue; } + // Avoid memory leak from not cleaning up after av_hwdevice_ctx_create + av_buffer_unref(&av_gpu_decoder); const int hwdevice_res = av_hwdevice_ctx_create(&av_gpu_decoder, type, nullptr, nullptr, 0); if (hwdevice_res < 0) { LOG_DEBUG(Service_NVDRV, "{} av_hwdevice_ctx_create failed {}", @@ -128,15 +129,19 @@ bool Codec::CreateGpuAvDevice() { av_codec->name, av_hwdevice_get_type_name(type)); break; } - if (config->methods & HW_CONFIG_METHOD && config->device_type == type) { - av_codec_ctx->pix_fmt = config->pix_fmt; - if (config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) { + if ((config->methods & HW_CONFIG_METHOD) != 0 && config->device_type == type) { +#if defined(__unix__) + // Some linux decoding backends are reported to crash with this config method + // TODO(ameerj): Properly support this method + if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_FRAMES_CTX) != 0) { // skip zero-copy decoders, we don't currently support them LOG_DEBUG(Service_NVDRV, "Skipping decoder {} with unsupported capability {}.", av_hwdevice_get_type_name(type), config->methods); continue; } +#endif LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type)); + av_codec_ctx->pix_fmt = config->pix_fmt; return true; } } @@ -163,11 +168,11 @@ void Codec::InitializeGpuDecoder() { void Codec::Initialize() { const AVCodecID codec = [&] { switch (current_codec) { - case NvdecCommon::VideoCodec::H264: + case Host1x::NvdecCommon::VideoCodec::H264: return AV_CODEC_ID_H264; - case NvdecCommon::VideoCodec::VP8: + case Host1x::NvdecCommon::VideoCodec::VP8: return AV_CODEC_ID_VP8; - case NvdecCommon::VideoCodec::VP9: + case Host1x::NvdecCommon::VideoCodec::VP9: return AV_CODEC_ID_VP9; default: UNIMPLEMENTED_MSG("Unknown codec {}", current_codec); @@ -192,7 +197,7 @@ void Codec::Initialize() { initialized = true; } -void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) { +void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) { if (current_codec != codec) { current_codec = codec; LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", GetCurrentCodecName()); @@ -210,16 +215,16 @@ void Codec::Decode() { bool vp9_hidden_frame = false; const auto& frame_data = [&]() { switch (current_codec) { - case Tegra::NvdecCommon::VideoCodec::H264: + case Tegra::Host1x::NvdecCommon::VideoCodec::H264: return h264_decoder->ComposeFrame(state, is_first_frame); - case Tegra::NvdecCommon::VideoCodec::VP8: + case Tegra::Host1x::NvdecCommon::VideoCodec::VP8: return vp8_decoder->ComposeFrame(state); - case Tegra::NvdecCommon::VideoCodec::VP9: + case Tegra::Host1x::NvdecCommon::VideoCodec::VP9: vp9_decoder->ComposeFrame(state); vp9_hidden_frame = vp9_decoder->WasFrameHidden(); return vp9_decoder->GetFrameBytes(); default: - UNREACHABLE(); + ASSERT(false); return std::vector<u8>{}; } }(); @@ -282,21 +287,21 @@ AVFramePtr Codec::GetCurrentFrame() { return frame; } -NvdecCommon::VideoCodec Codec::GetCurrentCodec() const { +Host1x::NvdecCommon::VideoCodec Codec::GetCurrentCodec() const { return current_codec; } std::string_view Codec::GetCurrentCodecName() const { switch (current_codec) { - case NvdecCommon::VideoCodec::None: + case Host1x::NvdecCommon::VideoCodec::None: return "None"; - case NvdecCommon::VideoCodec::H264: + case Host1x::NvdecCommon::VideoCodec::H264: return "H264"; - case NvdecCommon::VideoCodec::VP8: + case Host1x::NvdecCommon::VideoCodec::VP8: return "VP8"; - case NvdecCommon::VideoCodec::H265: + case Host1x::NvdecCommon::VideoCodec::H265: return "H265"; - case NvdecCommon::VideoCodec::VP9: + case Host1x::NvdecCommon::VideoCodec::VP9: return "VP9"; default: return "Unknown"; diff --git a/src/video_core/command_classes/codecs/codec.h b/src/video_core/host1x/codecs/codec.h index de5672155..0d45fb7fe 100644 --- a/src/video_core/command_classes/codecs/codec.h +++ b/src/video_core/host1x/codecs/codec.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,7 +7,7 @@ #include <string_view> #include <queue> #include "common/common_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/nvdec_common.h" extern "C" { #if defined(__GNUC__) || defined(__clang__) @@ -22,7 +21,6 @@ extern "C" { } namespace Tegra { -class GPU; void AVFrameDeleter(AVFrame* ptr); using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>; @@ -33,16 +31,20 @@ class VP8; class VP9; } // namespace Decoder +namespace Host1x { +class Host1x; +} // namespace Host1x + class Codec { public: - explicit Codec(GPU& gpu, const NvdecCommon::NvdecRegisters& regs); + explicit Codec(Host1x::Host1x& host1x, const Host1x::NvdecCommon::NvdecRegisters& regs); ~Codec(); /// Initialize the codec, returning success or failure void Initialize(); /// Sets NVDEC video stream codec - void SetTargetCodec(NvdecCommon::VideoCodec codec); + void SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec); /// Call decoders to construct headers, decode AVFrame with ffmpeg void Decode(); @@ -51,7 +53,7 @@ public: [[nodiscard]] AVFramePtr GetCurrentFrame(); /// Returns the value of current_codec - [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const; + [[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const; /// Return name of the current codec [[nodiscard]] std::string_view GetCurrentCodecName() const; @@ -64,14 +66,14 @@ private: bool CreateGpuAvDevice(); bool initialized{}; - NvdecCommon::VideoCodec current_codec{NvdecCommon::VideoCodec::None}; + Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None}; const AVCodec* av_codec{nullptr}; AVCodecContext* av_codec_ctx{nullptr}; AVBufferRef* av_gpu_decoder{nullptr}; - GPU& gpu; - const NvdecCommon::NvdecRegisters& state; + Host1x::Host1x& host1x; + const Host1x::NvdecCommon::NvdecRegisters& state; std::unique_ptr<Decoder::H264> h264_decoder; std::unique_ptr<Decoder::VP8> vp8_decoder; std::unique_ptr<Decoder::VP9> vp9_decoder; diff --git a/src/video_core/command_classes/codecs/h264.cpp b/src/video_core/host1x/codecs/h264.cpp index 84f1fa938..e87bd65fa 100644 --- a/src/video_core/command_classes/codecs/h264.cpp +++ b/src/video_core/host1x/codecs/h264.cpp @@ -1,29 +1,12 @@ -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: MIT #include <array> #include <bit> #include "common/settings.h" -#include "video_core/command_classes/codecs/h264.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/h264.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { @@ -41,19 +24,20 @@ constexpr std::array<u8, 16> zig_zag_scan{ }; } // Anonymous namespace -H264::H264(GPU& gpu_) : gpu(gpu_) {} +H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {} H264::~H264() = default; -const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& state, +const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame) { H264DecoderContext context; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext)); + host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context, + sizeof(H264DecoderContext)); const s64 frame_number = context.h264_parameter_set.frame_number.Value(); if (!is_first_frame && frame_number != 0) { frame.resize(context.stream_len); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); return frame; } @@ -172,8 +156,8 @@ const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& sta frame.resize(encoded_header.size() + context.stream_len); std::memcpy(frame.data(), encoded_header.data(), encoded_header.size()); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, - frame.data() + encoded_header.size(), context.stream_len); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, + frame.data() + encoded_header.size(), context.stream_len); return frame; } diff --git a/src/video_core/command_classes/codecs/h264.h b/src/video_core/host1x/codecs/h264.h index 1899d8e7f..5cc86454e 100644 --- a/src/video_core/command_classes/codecs/h264.h +++ b/src/video_core/host1x/codecs/h264.h @@ -1,22 +1,5 @@ -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: MIT #pragma once @@ -25,10 +8,14 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { class H264BitWriter { @@ -72,16 +59,16 @@ private: class H264 { public: - explicit H264(GPU& gpu); + explicit H264(Host1x::Host1x& host1x); ~H264(); /// Compose the H264 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state, - bool is_first_frame = false); + [[nodiscard]] const std::vector<u8>& ComposeFrame( + const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame = false); private: std::vector<u8> frame; - GPU& gpu; + Host1x::Host1x& host1x; struct H264ParameterSet { s32 log2_max_pic_order_cnt_lsb_minus4; ///< 0x00 diff --git a/src/video_core/command_classes/codecs/vp8.cpp b/src/video_core/host1x/codecs/vp8.cpp index 32ad0ec16..28fb12cb8 100644 --- a/src/video_core/command_classes/codecs/vp8.cpp +++ b/src/video_core/host1x/codecs/vp8.cpp @@ -1,22 +1,20 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <array> #include <vector> -#include "video_core/command_classes/codecs/vp8.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/vp8.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { -VP8::VP8(GPU& gpu_) : gpu(gpu_) {} +VP8::VP8(Host1x::Host1x& host1x_) : host1x{host1x_} {} VP8::~VP8() = default; -const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { +const std::vector<u8>& VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { VP8PictureInfo info; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo)); + host1x.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo)); const bool is_key_frame = info.key_frame == 1u; const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size); @@ -47,7 +45,7 @@ const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& stat frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f)); } const u64 bitstream_offset = state.frame_bitstream_offset; - gpu.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size); + host1x.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size); return frame; } diff --git a/src/video_core/command_classes/codecs/vp8.h b/src/video_core/host1x/codecs/vp8.h index 41fc7b403..5bf07ecab 100644 --- a/src/video_core/command_classes/codecs/vp8.h +++ b/src/video_core/host1x/codecs/vp8.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,23 +8,28 @@ #include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { class VP8 { public: - explicit VP8(GPU& gpu); + explicit VP8(Host1x::Host1x& host1x); ~VP8(); /// Compose the VP8 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] const std::vector<u8>& ComposeFrame( + const Host1x::NvdecCommon::NvdecRegisters& state); private: std::vector<u8> frame; - GPU& gpu; + Host1x::Host1x& host1x; struct VP8PictureInfo { INSERT_PADDING_WORDS_NOINIT(14); diff --git a/src/video_core/command_classes/codecs/vp9.cpp b/src/video_core/host1x/codecs/vp9.cpp index 2c00181fa..cf40c9012 100644 --- a/src/video_core/command_classes/codecs/vp9.cpp +++ b/src/video_core/host1x/codecs/vp9.cpp @@ -1,12 +1,11 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> // for std::copy #include <numeric> #include "common/assert.h" -#include "video_core/command_classes/codecs/vp9.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/vp9.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { @@ -154,7 +153,7 @@ constexpr Vp9EntropyProbs default_probs{ .high_precision{128, 128}, }; -constexpr std::array<s32, 256> norm_lut{ +constexpr std::array<u8, 256> norm_lut{ 0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, @@ -165,7 +164,7 @@ constexpr std::array<s32, 256> norm_lut{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }; -constexpr std::array<s32, 254> map_lut{ +constexpr std::array<u8, 254> map_lut{ 20, 21, 22, 23, 24, 25, 0, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 1, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 2, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 3, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, @@ -233,11 +232,11 @@ constexpr std::array<s32, 254> map_lut{ std::max(0, RecenterNonNeg(0xff - 1 - new_prob, 0xff - 1 - old_prob) - 1)); } - return map_lut[index]; + return static_cast<s32>(map_lut[index]); } } // Anonymous namespace -VP9::VP9(GPU& gpu_) : gpu{gpu_} {} +VP9::VP9(Host1x::Host1x& host1x_) : host1x{host1x_} {} VP9::~VP9() = default; @@ -356,9 +355,9 @@ void VP9::WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_ } } -Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state) { +Vp9PictureInfo VP9::GetVp9PictureInfo(const Host1x::NvdecCommon::NvdecRegisters& state) { PictureInfo picture_info; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo)); + host1x.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo)); Vp9PictureInfo vp9_info = picture_info.Convert(); InsertEntropy(state.vp9_entropy_probs_offset, vp9_info.entropy); @@ -373,18 +372,19 @@ Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state) void VP9::InsertEntropy(u64 offset, Vp9EntropyProbs& dst) { EntropyProbs entropy; - gpu.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs)); + host1x.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs)); entropy.Convert(dst); } -Vp9FrameContainer VP9::GetCurrentFrame(const NvdecCommon::NvdecRegisters& state) { +Vp9FrameContainer VP9::GetCurrentFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { Vp9FrameContainer current_frame{}; { - gpu.SyncGuestHost(); + // gpu.SyncGuestHost(); epic, why? current_frame.info = GetVp9PictureInfo(state); current_frame.bit_stream.resize(current_frame.info.bitstream_size); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, current_frame.bit_stream.data(), - current_frame.info.bitstream_size); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, + current_frame.bit_stream.data(), + current_frame.info.bitstream_size); } if (!next_frame.bit_stream.empty()) { Vp9FrameContainer temp{ @@ -770,7 +770,7 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() { return uncomp_writer; } -void VP9::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { +void VP9::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { std::vector<u8> bitstream; { Vp9FrameContainer curr_frame = GetCurrentFrame(state); @@ -820,7 +820,7 @@ void VpxRangeEncoder::Write(bool bit, s32 probability) { local_range = range - split; } - s32 shift = norm_lut[local_range]; + s32 shift = static_cast<s32>(norm_lut[local_range]); local_range <<= shift; count += shift; diff --git a/src/video_core/command_classes/codecs/vp9.h b/src/video_core/host1x/codecs/vp9.h index 2e735c792..d4083e8d3 100644 --- a/src/video_core/command_classes/codecs/vp9.h +++ b/src/video_core/host1x/codecs/vp9.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,11 +8,15 @@ #include "common/common_types.h" #include "common/stream.h" -#include "video_core/command_classes/codecs/vp9_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/codecs/vp9_types.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { /// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the @@ -107,7 +110,7 @@ private: class VP9 { public: - explicit VP9(GPU& gpu_); + explicit VP9(Host1x::Host1x& host1x); ~VP9(); VP9(const VP9&) = delete; @@ -118,7 +121,7 @@ public: /// Composes the VP9 frame from the GPU state information. /// Based on the official VP9 spec documentation - void ComposeFrame(const NvdecCommon::NvdecRegisters& state); + void ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state); /// Returns true if the most recent frame was a hidden frame. [[nodiscard]] bool WasFrameHidden() const { @@ -163,19 +166,21 @@ private: void WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob); /// Returns VP9 information from NVDEC provided offset and size - [[nodiscard]] Vp9PictureInfo GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] Vp9PictureInfo GetVp9PictureInfo( + const Host1x::NvdecCommon::NvdecRegisters& state); /// Read and convert NVDEC provided entropy probs to Vp9EntropyProbs struct void InsertEntropy(u64 offset, Vp9EntropyProbs& dst); /// Returns frame to be decoded after buffering - [[nodiscard]] Vp9FrameContainer GetCurrentFrame(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] Vp9FrameContainer GetCurrentFrame( + const Host1x::NvdecCommon::NvdecRegisters& state); /// Use NVDEC providied information to compose the headers for the current frame [[nodiscard]] std::vector<u8> ComposeCompressedHeader(); [[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader(); - GPU& gpu; + Host1x::Host1x& host1x; std::vector<u8> frame; std::array<s8, 4> loop_filter_ref_deltas{}; diff --git a/src/video_core/command_classes/codecs/vp9_types.h b/src/video_core/host1x/codecs/vp9_types.h index 3b1ed4b3a..adad8ed7e 100644 --- a/src/video_core/command_classes/codecs/vp9_types.h +++ b/src/video_core/host1x/codecs/vp9_types.h @@ -1,17 +1,14 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> -#include <cstring> #include <vector> #include "common/common_funcs.h" #include "common/common_types.h" namespace Tegra { -class GPU; namespace Decoder { struct Vp9FrameDimensions { diff --git a/src/video_core/host1x/control.cpp b/src/video_core/host1x/control.cpp new file mode 100644 index 000000000..dceefdb7f --- /dev/null +++ b/src/video_core/host1x/control.cpp @@ -0,0 +1,33 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "video_core/host1x/control.h" +#include "video_core/host1x/host1x.h" + +namespace Tegra::Host1x { + +Control::Control(Host1x& host1x_) : host1x(host1x_) {} + +Control::~Control() = default; + +void Control::ProcessMethod(Method method, u32 argument) { + switch (method) { + case Method::LoadSyncptPayload32: + syncpoint_value = argument; + break; + case Method::WaitSyncpt: + case Method::WaitSyncpt32: + Execute(argument); + break; + default: + UNIMPLEMENTED_MSG("Control method 0x{:X}", static_cast<u32>(method)); + break; + } +} + +void Control::Execute(u32 data) { + host1x.GetSyncpointManager().WaitHost(data, syncpoint_value); +} + +} // namespace Tegra::Host1x diff --git a/src/video_core/command_classes/host1x.h b/src/video_core/host1x/control.h index 7e94799dd..e117888a3 100644 --- a/src/video_core/command_classes/host1x.h +++ b/src/video_core/host1x/control.h @@ -1,18 +1,19 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once -#include <vector> -#include "common/common_funcs.h" #include "common/common_types.h" namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec; -class Host1x { +class Control { public: enum class Method : u32 { WaitSyncpt = 0x8, @@ -20,8 +21,8 @@ public: WaitSyncpt32 = 0x50, }; - explicit Host1x(GPU& gpu); - ~Host1x(); + explicit Control(Host1x& host1x); + ~Control(); /// Writes the method into the state, Invoke Execute() if encountered void ProcessMethod(Method method, u32 argument); @@ -31,7 +32,9 @@ private: void Execute(u32 data); u32 syncpoint_value{}; - GPU& gpu; + Host1x& host1x; }; +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/host1x/host1x.cpp b/src/video_core/host1x/host1x.cpp new file mode 100644 index 000000000..7c317a85d --- /dev/null +++ b/src/video_core/host1x/host1x.cpp @@ -0,0 +1,17 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "core/core.h" +#include "video_core/host1x/host1x.h" + +namespace Tegra { + +namespace Host1x { + +Host1x::Host1x(Core::System& system_) + : system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12}, + allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {} + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/host1x/host1x.h b/src/video_core/host1x/host1x.h new file mode 100644 index 000000000..57082ae54 --- /dev/null +++ b/src/video_core/host1x/host1x.h @@ -0,0 +1,57 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include "common/common_types.h" + +#include "common/address_space.h" +#include "video_core/host1x/syncpoint_manager.h" +#include "video_core/memory_manager.h" + +namespace Core { +class System; +} // namespace Core + +namespace Tegra { + +namespace Host1x { + +class Host1x { +public: + explicit Host1x(Core::System& system); + + SyncpointManager& GetSyncpointManager() { + return syncpoint_manager; + } + + const SyncpointManager& GetSyncpointManager() const { + return syncpoint_manager; + } + + Tegra::MemoryManager& MemoryManager() { + return memory_manager; + } + + const Tegra::MemoryManager& MemoryManager() const { + return memory_manager; + } + + Common::FlatAllocator<u32, 0, 32>& Allocator() { + return *allocator; + } + + const Common::FlatAllocator<u32, 0, 32>& Allocator() const { + return *allocator; + } + +private: + Core::System& system; + SyncpointManager syncpoint_manager; + Tegra::MemoryManager memory_manager; + std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator; +}; + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/command_classes/nvdec.cpp b/src/video_core/host1x/nvdec.cpp index 9aaf5247e..a4bd5b79f 100644 --- a/src/video_core/command_classes/nvdec.cpp +++ b/src/video_core/host1x/nvdec.cpp @@ -1,17 +1,17 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" -#include "video_core/command_classes/nvdec.h" -#include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" -namespace Tegra { +namespace Tegra::Host1x { #define NVDEC_REG_INDEX(field_name) \ (offsetof(NvdecCommon::NvdecRegisters, field_name) / sizeof(u64)) -Nvdec::Nvdec(GPU& gpu_) : gpu(gpu_), state{}, codec(std::make_unique<Codec>(gpu, state)) {} +Nvdec::Nvdec(Host1x& host1x_) + : host1x(host1x_), state{}, codec(std::make_unique<Codec>(host1x, state)) {} Nvdec::~Nvdec() = default; @@ -45,4 +45,4 @@ void Nvdec::Execute() { } } -} // namespace Tegra +} // namespace Tegra::Host1x diff --git a/src/video_core/command_classes/nvdec.h b/src/video_core/host1x/nvdec.h index 6e1da0b04..3949d5181 100644 --- a/src/video_core/command_classes/nvdec.h +++ b/src/video_core/host1x/nvdec.h @@ -1,20 +1,22 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <memory> #include <vector> #include "common/common_types.h" -#include "video_core/command_classes/codecs/codec.h" +#include "video_core/host1x/codecs/codec.h" namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec { public: - explicit Nvdec(GPU& gpu); + explicit Nvdec(Host1x& host1x); ~Nvdec(); /// Writes the method into the state, Invoke Execute() if encountered @@ -27,8 +29,11 @@ private: /// Invoke codec to decode a frame void Execute(); - GPU& gpu; + Host1x& host1x; NvdecCommon::NvdecRegisters state; std::unique_ptr<Codec> codec; }; + +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/command_classes/nvdec_common.h b/src/video_core/host1x/nvdec_common.h index 8a35c44a1..49d67ebbe 100644 --- a/src/video_core/command_classes/nvdec_common.h +++ b/src/video_core/host1x/nvdec_common.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,7 +7,7 @@ #include "common/common_funcs.h" #include "common/common_types.h" -namespace Tegra::NvdecCommon { +namespace Tegra::Host1x::NvdecCommon { enum class VideoCodec : u64 { None = 0x0, @@ -95,4 +94,4 @@ ASSERT_REG_POSITION(vp9_curr_frame_mvs_offset, 0x176); #undef ASSERT_REG_POSITION -} // namespace Tegra::NvdecCommon +} // namespace Tegra::Host1x::NvdecCommon diff --git a/src/video_core/host1x/sync_manager.cpp b/src/video_core/host1x/sync_manager.cpp new file mode 100644 index 000000000..5ef9ea217 --- /dev/null +++ b/src/video_core/host1x/sync_manager.cpp @@ -0,0 +1,50 @@ +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: MIT + +#include <algorithm> +#include "sync_manager.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" + +namespace Tegra { +namespace Host1x { + +SyncptIncrManager::SyncptIncrManager(Host1x& host1x_) : host1x(host1x_) {} +SyncptIncrManager::~SyncptIncrManager() = default; + +void SyncptIncrManager::Increment(u32 id) { + increments.emplace_back(0, 0, id, true); + IncrementAllDone(); +} + +u32 SyncptIncrManager::IncrementWhenDone(u32 class_id, u32 id) { + const u32 handle = current_id++; + increments.emplace_back(handle, class_id, id); + return handle; +} + +void SyncptIncrManager::SignalDone(u32 handle) { + const auto done_incr = + std::find_if(increments.begin(), increments.end(), + [handle](const SyncptIncr& incr) { return incr.id == handle; }); + if (done_incr != increments.cend()) { + done_incr->complete = true; + } + IncrementAllDone(); +} + +void SyncptIncrManager::IncrementAllDone() { + std::size_t done_count = 0; + for (; done_count < increments.size(); ++done_count) { + if (!increments[done_count].complete) { + break; + } + auto& syncpoint_manager = host1x.GetSyncpointManager(); + syncpoint_manager.IncrementGuest(increments[done_count].syncpt_id); + syncpoint_manager.IncrementHost(increments[done_count].syncpt_id); + } + increments.erase(increments.begin(), increments.begin() + done_count); +} + +} // namespace Host1x +} // namespace Tegra diff --git a/src/video_core/host1x/sync_manager.h b/src/video_core/host1x/sync_manager.h new file mode 100644 index 000000000..7bb77fa27 --- /dev/null +++ b/src/video_core/host1x/sync_manager.h @@ -0,0 +1,53 @@ +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: MIT + +#pragma once + +#include <mutex> +#include <vector> +#include "common/common_types.h" + +namespace Tegra { + +namespace Host1x { + +class Host1x; + +struct SyncptIncr { + u32 id; + u32 class_id; + u32 syncpt_id; + bool complete; + + SyncptIncr(u32 id_, u32 class_id_, u32 syncpt_id_, bool done = false) + : id(id_), class_id(class_id_), syncpt_id(syncpt_id_), complete(done) {} +}; + +class SyncptIncrManager { +public: + explicit SyncptIncrManager(Host1x& host1x); + ~SyncptIncrManager(); + + /// Add syncpoint id and increment all + void Increment(u32 id); + + /// Returns a handle to increment later + u32 IncrementWhenDone(u32 class_id, u32 id); + + /// IncrememntAllDone, including handle + void SignalDone(u32 handle); + + /// Increment all sequential pending increments that are already done. + void IncrementAllDone(); + +private: + std::vector<SyncptIncr> increments; + std::mutex increment_lock; + u32 current_id{}; + + Host1x& host1x; +}; + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/host1x/syncpoint_manager.cpp b/src/video_core/host1x/syncpoint_manager.cpp new file mode 100644 index 000000000..326e8355a --- /dev/null +++ b/src/video_core/host1x/syncpoint_manager.cpp @@ -0,0 +1,96 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/microprofile.h" +#include "video_core/host1x/syncpoint_manager.h" + +namespace Tegra { + +namespace Host1x { + +MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); + +SyncpointManager::ActionHandle SyncpointManager::RegisterAction( + std::atomic<u32>& syncpoint, std::list<RegisteredAction>& action_storage, u32 expected_value, + std::function<void()>&& action) { + if (syncpoint.load(std::memory_order_acquire) >= expected_value) { + action(); + return {}; + } + + std::unique_lock lk(guard); + if (syncpoint.load(std::memory_order_relaxed) >= expected_value) { + action(); + return {}; + } + auto it = action_storage.begin(); + while (it != action_storage.end()) { + if (it->expected_value >= expected_value) { + break; + } + ++it; + } + return action_storage.emplace(it, expected_value, std::move(action)); +} + +void SyncpointManager::DeregisterAction(std::list<RegisteredAction>& action_storage, + ActionHandle& handle) { + std::unique_lock lk(guard); + action_storage.erase(handle); +} + +void SyncpointManager::DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle) { + DeregisterAction(guest_action_storage[syncpoint_id], handle); +} + +void SyncpointManager::DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle) { + DeregisterAction(host_action_storage[syncpoint_id], handle); +} + +void SyncpointManager::IncrementGuest(u32 syncpoint_id) { + Increment(syncpoints_guest[syncpoint_id], wait_guest_cv, guest_action_storage[syncpoint_id]); +} + +void SyncpointManager::IncrementHost(u32 syncpoint_id) { + Increment(syncpoints_host[syncpoint_id], wait_host_cv, host_action_storage[syncpoint_id]); +} + +void SyncpointManager::WaitGuest(u32 syncpoint_id, u32 expected_value) { + Wait(syncpoints_guest[syncpoint_id], wait_guest_cv, expected_value); +} + +void SyncpointManager::WaitHost(u32 syncpoint_id, u32 expected_value) { + MICROPROFILE_SCOPE(GPU_wait); + Wait(syncpoints_host[syncpoint_id], wait_host_cv, expected_value); +} + +void SyncpointManager::Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + std::list<RegisteredAction>& action_storage) { + auto new_value{syncpoint.fetch_add(1, std::memory_order_acq_rel) + 1}; + + std::unique_lock lk(guard); + auto it = action_storage.begin(); + while (it != action_storage.end()) { + if (it->expected_value > new_value) { + break; + } + it->action(); + it = action_storage.erase(it); + } + wait_cv.notify_all(); +} + +void SyncpointManager::Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + u32 expected_value) { + const auto pred = [&]() { return syncpoint.load(std::memory_order_acquire) >= expected_value; }; + if (pred()) { + return; + } + + std::unique_lock lk(guard); + wait_cv.wait(lk, pred); +} + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/host1x/syncpoint_manager.h b/src/video_core/host1x/syncpoint_manager.h new file mode 100644 index 000000000..50a264e23 --- /dev/null +++ b/src/video_core/host1x/syncpoint_manager.h @@ -0,0 +1,98 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <array> +#include <atomic> +#include <condition_variable> +#include <functional> +#include <list> +#include <mutex> + +#include "common/common_types.h" + +namespace Tegra { + +namespace Host1x { + +class SyncpointManager { +public: + u32 GetGuestSyncpointValue(u32 id) const { + return syncpoints_guest[id].load(std::memory_order_acquire); + } + + u32 GetHostSyncpointValue(u32 id) const { + return syncpoints_host[id].load(std::memory_order_acquire); + } + + struct RegisteredAction { + explicit RegisteredAction(u32 expected_value_, std::function<void()>&& action_) + : expected_value{expected_value_}, action{std::move(action_)} {} + u32 expected_value; + std::function<void()> action; + }; + using ActionHandle = std::list<RegisteredAction>::iterator; + + template <typename Func> + ActionHandle RegisterGuestAction(u32 syncpoint_id, u32 expected_value, Func&& action) { + std::function<void()> func(action); + return RegisterAction(syncpoints_guest[syncpoint_id], guest_action_storage[syncpoint_id], + expected_value, std::move(func)); + } + + template <typename Func> + ActionHandle RegisterHostAction(u32 syncpoint_id, u32 expected_value, Func&& action) { + std::function<void()> func(action); + return RegisterAction(syncpoints_host[syncpoint_id], host_action_storage[syncpoint_id], + expected_value, std::move(func)); + } + + void DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle); + + void DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle); + + void IncrementGuest(u32 syncpoint_id); + + void IncrementHost(u32 syncpoint_id); + + void WaitGuest(u32 syncpoint_id, u32 expected_value); + + void WaitHost(u32 syncpoint_id, u32 expected_value); + + bool IsReadyGuest(u32 syncpoint_id, u32 expected_value) const { + return syncpoints_guest[syncpoint_id].load(std::memory_order_acquire) >= expected_value; + } + + bool IsReadyHost(u32 syncpoint_id, u32 expected_value) const { + return syncpoints_host[syncpoint_id].load(std::memory_order_acquire) >= expected_value; + } + +private: + void Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + std::list<RegisteredAction>& action_storage); + + ActionHandle RegisterAction(std::atomic<u32>& syncpoint, + std::list<RegisteredAction>& action_storage, u32 expected_value, + std::function<void()>&& action); + + void DeregisterAction(std::list<RegisteredAction>& action_storage, ActionHandle& handle); + + void Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, u32 expected_value); + + static constexpr size_t NUM_MAX_SYNCPOINTS = 192; + + std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_guest{}; + std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_host{}; + + std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> guest_action_storage; + std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> host_action_storage; + + std::mutex guard; + std::condition_variable wait_guest_cv; + std::condition_variable wait_host_cv; +}; + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/host1x/vic.cpp index 051616124..ac0b7d20e 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/host1x/vic.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> @@ -19,14 +18,17 @@ extern "C" { #include "common/bit_field.h" #include "common/logging/log.h" -#include "video_core/command_classes/nvdec.h" -#include "video_core/command_classes/vic.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" +#include "video_core/host1x/vic.h" #include "video_core/memory_manager.h" #include "video_core/textures/decoders.h" namespace Tegra { + +namespace Host1x { + namespace { enum class VideoPixelFormat : u64_le { RGBA8 = 0x1f, @@ -47,8 +49,8 @@ union VicConfig { BitField<46, 14, u64_le> surface_height_minus1; }; -Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_) - : gpu(gpu_), +Vic::Vic(Host1x& host1x_, std::shared_ptr<Nvdec> nvdec_processor_) + : host1x(host1x_), nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free} {} Vic::~Vic() = default; @@ -79,7 +81,7 @@ void Vic::Execute() { LOG_ERROR(Service_NVDRV, "VIC Luma address not set."); return; } - const VicConfig config{gpu.MemoryManager().Read<u64>(config_struct_address + 0x20)}; + const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)}; const AVFramePtr frame_ptr = nvdec_processor->GetFrame(); const auto* frame = frame_ptr.get(); if (!frame) { @@ -154,15 +156,16 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) { const u32 block_height = static_cast<u32>(config.block_linear_height_log2); const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0); luma_buffer.resize(size); - Texture::SwizzleSubrect(width, height, width * 4, width, 4, luma_buffer.data(), - converted_frame_buf_addr, block_height, 0, 0); + std::span<const u8> frame_buff(converted_frame_buf_addr, 4 * width * height); + Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height, + block_height, 0, width * 4); - gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size); } else { // send pitch linear frame const size_t linear_size = width * height * 4; - gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, - linear_size); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, + linear_size); } } @@ -190,8 +193,8 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { luma_buffer[dst + x] = luma_src[src + x]; } } - gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), - luma_buffer.size()); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), + luma_buffer.size()); // Chroma const std::size_t half_height = frame_height / 2; @@ -229,11 +232,13 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { break; } default: - UNREACHABLE(); + ASSERT(false); break; } - gpu.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(), - chroma_buffer.size()); + host1x.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(), + chroma_buffer.size()); } +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/command_classes/vic.h b/src/video_core/host1x/vic.h index 6d4cdfd57..2b78786e8 100644 --- a/src/video_core/command_classes/vic.h +++ b/src/video_core/host1x/vic.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -11,7 +10,10 @@ struct SwsContext; namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec; union VicConfig; @@ -26,7 +28,7 @@ public: SetOutputSurfaceChromaUnusedOffset = 0x1ca }; - explicit Vic(GPU& gpu, std::shared_ptr<Nvdec> nvdec_processor); + explicit Vic(Host1x& host1x, std::shared_ptr<Nvdec> nvdec_processor); ~Vic(); @@ -40,8 +42,8 @@ private: void WriteYUVFrame(const AVFrame* frame, const VicConfig& config); - GPU& gpu; - std::shared_ptr<Tegra::Nvdec> nvdec_processor; + Host1x& host1x; + std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor; /// Avoid reallocation of the following buffers every frame, as their /// size does not change during a stream @@ -59,4 +61,6 @@ private: s32 scaler_height{}; }; +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index fd3e41434..2149ab93e 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -1,3 +1,6 @@ +# SPDX-FileCopyrightText: 2018 yuzu Emulator Project +# SPDX-License-Identifier: GPL-2.0-or-later + set(FIDELITYFX_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/externals/FidelityFX-FSR/ffx-fsr) set(GLSL_INCLUDES @@ -14,9 +17,11 @@ set(SHADER_FILES convert_d24s8_to_abgr8.frag convert_depth_to_float.frag convert_float_to_depth.frag + convert_s8d24_to_abgr8.frag full_screen_triangle.vert fxaa.frag fxaa.vert + opengl_convert_s8d24.comp opengl_copy_bc4.comp opengl_present.frag opengl_present.vert diff --git a/src/video_core/host_shaders/StringShaderHeader.cmake b/src/video_core/host_shaders/StringShaderHeader.cmake index 1b4bc6103..9f7525535 100644 --- a/src/video_core/host_shaders/StringShaderHeader.cmake +++ b/src/video_core/host_shaders/StringShaderHeader.cmake @@ -1,3 +1,6 @@ +# SPDX-FileCopyrightText: 2020 yuzu Emulator Project +# SPDX-License-Identifier: GPL-2.0-or-later + set(SOURCE_FILE ${CMAKE_ARGV3}) set(HEADER_FILE ${CMAKE_ARGV4}) set(INPUT_FILE ${CMAKE_ARGV5}) diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index 3a10578cb..d608678a3 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 @@ -1066,7 +1065,7 @@ TexelWeightParams DecodeBlockInfo() { void FillError(ivec3 coord) { for (uint j = 0; j < block_dims.y; j++) { for (uint i = 0; i < block_dims.x; i++) { - imageStore(dest_image, coord + ivec3(i, j, 0), vec4(1.0, 1.0, 0.0, 1.0)); + imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0)); } } } diff --git a/src/video_core/host_shaders/block_linear_unswizzle_2d.comp b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp index a131be79e..121f60cbd 100644 --- a/src/video_core/host_shaders/block_linear_unswizzle_2d.comp +++ b/src/video_core/host_shaders/block_linear_unswizzle_2d.comp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 diff --git a/src/video_core/host_shaders/block_linear_unswizzle_3d.comp b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp index bb6872e6b..7fc98accb 100644 --- a/src/video_core/host_shaders/block_linear_unswizzle_3d.comp +++ b/src/video_core/host_shaders/block_linear_unswizzle_3d.comp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 diff --git a/src/video_core/host_shaders/convert_abgr8_to_d24s8.frag b/src/video_core/host_shaders/convert_abgr8_to_d24s8.frag index ea055ddad..ff102e582 100644 --- a/src/video_core/host_shaders/convert_abgr8_to_d24s8.frag +++ b/src/video_core/host_shaders/convert_abgr8_to_d24s8.frag @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 #extension GL_ARB_shader_stencil_export : require diff --git a/src/video_core/host_shaders/convert_d24s8_to_abgr8.frag b/src/video_core/host_shaders/convert_d24s8_to_abgr8.frag index 94368fb59..d33131d7c 100644 --- a/src/video_core/host_shaders/convert_d24s8_to_abgr8.frag +++ b/src/video_core/host_shaders/convert_d24s8_to_abgr8.frag @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 diff --git a/src/video_core/host_shaders/convert_depth_to_float.frag b/src/video_core/host_shaders/convert_depth_to_float.frag index 624c58509..57f0ce5a6 100644 --- a/src/video_core/host_shaders/convert_depth_to_float.frag +++ b/src/video_core/host_shaders/convert_depth_to_float.frag @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 diff --git a/src/video_core/host_shaders/convert_float_to_depth.frag b/src/video_core/host_shaders/convert_float_to_depth.frag index d86c795f4..5d403b1ec 100644 --- a/src/video_core/host_shaders/convert_float_to_depth.frag +++ b/src/video_core/host_shaders/convert_float_to_depth.frag @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 diff --git a/src/video_core/host_shaders/convert_s8d24_to_abgr8.frag b/src/video_core/host_shaders/convert_s8d24_to_abgr8.frag new file mode 100644 index 000000000..31db7d426 --- /dev/null +++ b/src/video_core/host_shaders/convert_s8d24_to_abgr8.frag @@ -0,0 +1,22 @@ +// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#version 450 + +layout(binding = 0) uniform sampler2D depth_tex; +layout(binding = 1) uniform isampler2D stencil_tex; + +layout(location = 0) out vec4 color; + +void main() { + ivec2 coord = ivec2(gl_FragCoord.xy); + uint depth = uint(textureLod(depth_tex, coord, 0).r * (exp2(24.0) - 1.0f)); + uint stencil = uint(textureLod(stencil_tex, coord, 0).r); + + highp uint depth_val = + uint(textureLod(depth_tex, coord, 0).r * (exp2(32.0) - 1.0)); + lowp uint stencil_val = textureLod(stencil_tex, coord, 0).r; + highp uvec4 components = + uvec4((uvec3(depth_val) >> uvec3(24u, 16u, 8u)) & 0x000000FFu, stencil_val); + color.rgba = vec4(components) / (exp2(8.0) - 1.0); +} diff --git a/src/video_core/host_shaders/fidelityfx_fsr.comp b/src/video_core/host_shaders/fidelityfx_fsr.comp index 6b97f789d..f91b1aa9f 100644 --- a/src/video_core/host_shaders/fidelityfx_fsr.comp +++ b/src/video_core/host_shaders/fidelityfx_fsr.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later //!#version 460 core #extension GL_ARB_separate_shader_objects : enable diff --git a/src/video_core/host_shaders/full_screen_triangle.vert b/src/video_core/host_shaders/full_screen_triangle.vert index 452ad6502..2c976b19f 100644 --- a/src/video_core/host_shaders/full_screen_triangle.vert +++ b/src/video_core/host_shaders/full_screen_triangle.vert @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 diff --git a/src/video_core/host_shaders/fxaa.frag b/src/video_core/host_shaders/fxaa.frag index 02f4068d1..9bffc20d5 100644 --- a/src/video_core/host_shaders/fxaa.frag +++ b/src/video_core/host_shaders/fxaa.frag @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later // Source code is adapted from // https://www.geeks3d.com/20110405/fxaa-fast-approximate-anti-aliasing-demo-glsl-opengl-test-radeon-geforce/3/ diff --git a/src/video_core/host_shaders/fxaa.vert b/src/video_core/host_shaders/fxaa.vert index ac20c04e9..c2717d90d 100644 --- a/src/video_core/host_shaders/fxaa.vert +++ b/src/video_core/host_shaders/fxaa.vert @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 diff --git a/src/video_core/host_shaders/opengl_convert_s8d24.comp b/src/video_core/host_shaders/opengl_convert_s8d24.comp new file mode 100644 index 000000000..970203214 --- /dev/null +++ b/src/video_core/host_shaders/opengl_convert_s8d24.comp @@ -0,0 +1,17 @@ +// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#version 430 core + +layout(local_size_x = 16, local_size_y = 8) in; + +layout(binding = 0, rgba8ui) restrict uniform uimage2D destination; +layout(location = 0) uniform uvec3 size; + +void main() { + if (any(greaterThanEqual(gl_GlobalInvocationID, size))) { + return; + } + uvec4 components = imageLoad(destination, ivec2(gl_GlobalInvocationID.xy)); + imageStore(destination, ivec2(gl_GlobalInvocationID.xy), components.wxyz); +} diff --git a/src/video_core/host_shaders/opengl_copy_bc4.comp b/src/video_core/host_shaders/opengl_copy_bc4.comp index 7b8e20fbe..8d3ae5a97 100644 --- a/src/video_core/host_shaders/opengl_copy_bc4.comp +++ b/src/video_core/host_shaders/opengl_copy_bc4.comp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 core #extension GL_ARB_gpu_shader_int64 : require diff --git a/src/video_core/host_shaders/opengl_present.frag b/src/video_core/host_shaders/opengl_present.frag index 84b818227..5fd7ad297 100644 --- a/src/video_core/host_shaders/opengl_present.frag +++ b/src/video_core/host_shaders/opengl_present.frag @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 core diff --git a/src/video_core/host_shaders/opengl_present.vert b/src/video_core/host_shaders/opengl_present.vert index c3b5adbba..cc2f40163 100644 --- a/src/video_core/host_shaders/opengl_present.vert +++ b/src/video_core/host_shaders/opengl_present.vert @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 core diff --git a/src/video_core/host_shaders/opengl_present_scaleforce.frag b/src/video_core/host_shaders/opengl_present_scaleforce.frag index 71ff9e1e3..a780373e3 100644 --- a/src/video_core/host_shaders/opengl_present_scaleforce.frag +++ b/src/video_core/host_shaders/opengl_present_scaleforce.frag @@ -1,24 +1,5 @@ -// MIT License -// -// Copyright (c) 2020 BreadFish64 -// -// Permission is hereby granted, free of charge, to any person obtaining a copy -// of this software and associated documentation files (the "Software"), to deal -// in the Software without restriction, including without limitation the rights -// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -// copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all -// copies or substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE -// SOFTWARE. +// SPDX-FileCopyrightText: 2020 BreadFish64 +// SPDX-License-Identifier: MIT // Adapted from https://github.com/BreadFish64/ScaleFish/tree/master/scaleforce diff --git a/src/video_core/host_shaders/pitch_unswizzle.comp b/src/video_core/host_shaders/pitch_unswizzle.comp index cb48ec170..7d23f594c 100644 --- a/src/video_core/host_shaders/pitch_unswizzle.comp +++ b/src/video_core/host_shaders/pitch_unswizzle.comp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 430 diff --git a/src/video_core/host_shaders/present_bicubic.frag b/src/video_core/host_shaders/present_bicubic.frag index 902b70c2b..c57dd2851 100644 --- a/src/video_core/host_shaders/present_bicubic.frag +++ b/src/video_core/host_shaders/present_bicubic.frag @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core diff --git a/src/video_core/host_shaders/present_gaussian.frag b/src/video_core/host_shaders/present_gaussian.frag index 66fed3238..5f54b71b6 100644 --- a/src/video_core/host_shaders/present_gaussian.frag +++ b/src/video_core/host_shaders/present_gaussian.frag @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later // Code adapted from the following sources: // - https://learnopengl.com/Advanced-Lighting/Bloom diff --git a/src/video_core/host_shaders/source_shader.h.in b/src/video_core/host_shaders/source_shader.h.in index 929dec39b..f189ee06b 100644 --- a/src/video_core/host_shaders/source_shader.h.in +++ b/src/video_core/host_shaders/source_shader.h.in @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + #pragma once #include <string_view> diff --git a/src/video_core/host_shaders/vulkan_blit_color_float.frag b/src/video_core/host_shaders/vulkan_blit_color_float.frag index 4a6aae410..c0c832296 100644 --- a/src/video_core/host_shaders/vulkan_blit_color_float.frag +++ b/src/video_core/host_shaders/vulkan_blit_color_float.frag @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 diff --git a/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag b/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag index 19bb23a5a..484b99773 100644 --- a/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag +++ b/src/video_core/host_shaders/vulkan_blit_depth_stencil.frag @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 450 #extension GL_ARB_shader_stencil_export : require diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp index 1c96a7905..00af13726 100644 --- a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp index f4daff739..13d783fa8 100644 --- a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp index 6b6796dd1..331549d96 100644 --- a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp index f785eebf3..013ca0014 100644 --- a/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp +++ b/src/video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32.comp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_present.frag b/src/video_core/host_shaders/vulkan_present.frag index 0979ff3e6..97e098ced 100644 --- a/src/video_core/host_shaders/vulkan_present.frag +++ b/src/video_core/host_shaders/vulkan_present.frag @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core diff --git a/src/video_core/host_shaders/vulkan_present.vert b/src/video_core/host_shaders/vulkan_present.vert index 00b868958..89dc80468 100644 --- a/src/video_core/host_shaders/vulkan_present.vert +++ b/src/video_core/host_shaders/vulkan_present.vert @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core diff --git a/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag b/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag index 924c03060..3dc9c0df5 100644 --- a/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag +++ b/src/video_core/host_shaders/vulkan_present_scaleforce_fp16.frag @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + #version 460 #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag b/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag index a594b83ca..77ed07552 100644 --- a/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag +++ b/src/video_core/host_shaders/vulkan_present_scaleforce_fp32.frag @@ -1,3 +1,6 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + #version 460 #extension GL_GOOGLE_include_directive : enable diff --git a/src/video_core/host_shaders/vulkan_quad_indexed.comp b/src/video_core/host_shaders/vulkan_quad_indexed.comp index 8655591d0..a412f30ff 100644 --- a/src/video_core/host_shaders/vulkan_quad_indexed.comp +++ b/src/video_core/host_shaders/vulkan_quad_indexed.comp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core diff --git a/src/video_core/host_shaders/vulkan_uint8.comp b/src/video_core/host_shaders/vulkan_uint8.comp index 872291670..5aa6abdc8 100644 --- a/src/video_core/host_shaders/vulkan_uint8.comp +++ b/src/video_core/host_shaders/vulkan_uint8.comp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #version 460 core #extension GL_EXT_shader_16bit_storage : require diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index 0aeda4ce8..f61d5998e 100644 --- a/src/video_core/macro/macro.cpp +++ b/src/video_core/macro/macro.cpp @@ -1,13 +1,17 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstring> +#include <fstream> #include <optional> +#include <span> #include <boost/container_hash/hash.hpp> +#include <fstream> #include "common/assert.h" +#include "common/fs/fs.h" +#include "common/fs/path_util.h" #include "common/settings.h" #include "video_core/macro/macro.h" #include "video_core/macro/macro_hle.h" @@ -16,6 +20,23 @@ namespace Tegra { +static void Dump(u64 hash, std::span<const u32> code) { + const auto base_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::DumpDir)}; + const auto macro_dir{base_dir / "macros"}; + if (!Common::FS::CreateDir(base_dir) || !Common::FS::CreateDir(macro_dir)) { + LOG_ERROR(Common_Filesystem, "Failed to create macro dump directories"); + return; + } + const auto name{macro_dir / fmt::format("{:016x}.macro", hash)}; + std::fstream macro_file(name, std::ios::out | std::ios::binary); + if (!macro_file) { + LOG_ERROR(Common_Filesystem, "Unable to open or create file at {}", + Common::FS::PathToUTF8String(name)); + return; + } + macro_file.write(reinterpret_cast<const char*>(code.data()), code.size_bytes()); +} + MacroEngine::MacroEngine(Engines::Maxwell3D& maxwell3d) : hle_macros{std::make_unique<Tegra::HLEMacro>(maxwell3d)} {} @@ -25,6 +46,11 @@ void MacroEngine::AddCode(u32 method, u32 data) { uploaded_macro_code[method].push_back(data); } +void MacroEngine::ClearCode(u32 method) { + macro_cache.erase(method); + uploaded_macro_code.erase(method); +} + void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { auto compiled_macro = macro_cache.find(method); if (compiled_macro != macro_cache.end()) { @@ -46,7 +72,7 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { } } if (!mid_method.has_value()) { - UNREACHABLE_MSG("Macro 0x{0:x} was not uploaded", method); + ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method); return; } } @@ -55,6 +81,9 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { if (!mid_method.has_value()) { cache_info.lle_program = Compile(macro_code->second); cache_info.hash = boost::hash_value(macro_code->second); + if (Settings::values.dump_macros) { + Dump(cache_info.hash, macro_code->second); + } } else { const auto& macro_cached = uploaded_macro_code[mid_method.value()]; const auto rebased_method = method - mid_method.value(); @@ -64,6 +93,9 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { code.size() * sizeof(u32)); cache_info.hash = boost::hash_value(code); cache_info.lle_program = Compile(code); + if (Settings::values.dump_macros) { + Dump(cache_info.hash, code); + } } if (auto hle_program = hle_macros->GetHLEProgram(cache_info.hash)) { diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h index 7aaa49286..07d97ba39 100644 --- a/src/video_core/macro/macro.h +++ b/src/video_core/macro/macro.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -118,6 +117,9 @@ public: // Store the uploaded macro code to compile them when they're called. void AddCode(u32 method, u32 data); + // Clear the code associated with a method. + void ClearCode(u32 method); + // Compiles the macro if its not in the cache, and executes the compiled macro void Execute(u32 method, const std::vector<u32>& parameters); diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp index 900ad23c9..cabe8dcbf 100644 --- a/src/video_core/macro/macro_hle.cpp +++ b/src/video_core/macro/macro_hle.cpp @@ -1,9 +1,10 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <vector> +#include "common/scope_exit.h" +#include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/macro/macro.h" #include "video_core/macro/macro_hle.h" @@ -59,6 +60,7 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& maxwell3d.regs.index_array.first = parameters[3]; maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base? maxwell3d.regs.index_array.count = parameters[1]; + maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; maxwell3d.regs.vb_element_base = element_base; maxwell3d.regs.vb_base_instance = base_instance; maxwell3d.mme_draw.instance_count = instance_count; @@ -81,10 +83,67 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined; } -constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{ +// Multidraw Indirect +void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) { + SCOPE_EXIT({ + // Clean everything. + maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base? + maxwell3d.regs.index_array.count = 0; + maxwell3d.regs.vb_element_base = 0x0; + maxwell3d.regs.vb_base_instance = 0x0; + maxwell3d.mme_draw.instance_count = 0; + maxwell3d.CallMethodFromMME(0x8e3, 0x640); + maxwell3d.CallMethodFromMME(0x8e4, 0x0); + maxwell3d.CallMethodFromMME(0x8e5, 0x0); + maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined; + maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; + }); + const u32 start_indirect = parameters[0]; + const u32 end_indirect = parameters[1]; + if (start_indirect >= end_indirect) { + // Nothing to do. + return; + } + const auto topology = + static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]); + maxwell3d.regs.draw.topology.Assign(topology); + const u32 padding = parameters[3]; + const std::size_t max_draws = parameters[4]; + + const u32 indirect_words = 5 + padding; + const std::size_t first_draw = start_indirect; + const std::size_t effective_draws = end_indirect - start_indirect; + const std::size_t last_draw = start_indirect + std::min(effective_draws, max_draws); + + for (std::size_t index = first_draw; index < last_draw; index++) { + const std::size_t base = index * indirect_words + 5; + const u32 num_vertices = parameters[base]; + const u32 instance_count = parameters[base + 1]; + const u32 first_index = parameters[base + 2]; + const u32 base_vertex = parameters[base + 3]; + const u32 base_instance = parameters[base + 4]; + maxwell3d.regs.index_array.first = first_index; + maxwell3d.regs.reg_array[0x446] = base_vertex; + maxwell3d.regs.index_array.count = num_vertices; + maxwell3d.regs.vb_element_base = base_vertex; + maxwell3d.regs.vb_base_instance = base_instance; + maxwell3d.mme_draw.instance_count = instance_count; + maxwell3d.CallMethodFromMME(0x8e3, 0x640); + maxwell3d.CallMethodFromMME(0x8e4, base_vertex); + maxwell3d.CallMethodFromMME(0x8e5, base_instance); + maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; + if (maxwell3d.ShouldExecute()) { + maxwell3d.Rasterizer().Draw(true, true); + } + maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined; + } +} + +constexpr std::array<std::pair<u64, HLEFunction>, 4> hle_funcs{{ {0x771BB18C62444DA0, &HLE_771BB18C62444DA0}, {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD}, {0x0217920100488FF7, &HLE_0217920100488FF7}, + {0x3F5E74B9C9A50164, &HLE_3F5E74B9C9A50164}, }}; class HLEMacroImpl final : public CachedMacro { @@ -100,6 +159,7 @@ private: Engines::Maxwell3D& maxwell3d; HLEFunction func; }; + } // Anonymous namespace HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {} diff --git a/src/video_core/macro/macro_hle.h b/src/video_core/macro/macro_hle.h index b86ba84a1..625332c9d 100644 --- a/src/video_core/macro/macro_hle.h +++ b/src/video_core/macro/macro_hle.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/macro/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp index fba755448..f670b1bca 100644 --- a/src/video_core/macro/macro_interpreter.cpp +++ b/src/video_core/macro/macro_interpreter.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <optional> @@ -309,7 +308,6 @@ bool MacroInterpreterImpl::EvaluateBranchCondition(Macro::BranchCondition cond, return value != 0; } UNREACHABLE(); - return true; } Macro::Opcode MacroInterpreterImpl::GetOpcode() const { diff --git a/src/video_core/macro/macro_interpreter.h b/src/video_core/macro/macro_interpreter.h index 8a9648e46..f5eeb0b76 100644 --- a/src/video_core/macro/macro_interpreter.h +++ b/src/video_core/macro/macro_interpreter.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp index 47b28ad16..a302a9603 100644 --- a/src/video_core/macro/macro_jit_x64.cpp +++ b/src/video_core/macro/macro_jit_x64.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <bitset> @@ -24,7 +23,8 @@ MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255 namespace Tegra { namespace { constexpr Xbyak::Reg64 STATE = Xbyak::util::rbx; -constexpr Xbyak::Reg32 RESULT = Xbyak::util::ebp; +constexpr Xbyak::Reg32 RESULT = Xbyak::util::r10d; +constexpr Xbyak::Reg64 MAX_PARAMETER = Xbyak::util::r11; constexpr Xbyak::Reg64 PARAMETERS = Xbyak::util::r12; constexpr Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d; constexpr Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15; @@ -32,6 +32,7 @@ constexpr Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15; constexpr std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({ STATE, RESULT, + MAX_PARAMETER, PARAMETERS, METHOD_ADDRESS, BRANCH_HOLDER, @@ -81,7 +82,7 @@ private: u32 carry_flag{}; }; static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0"); - using ProgramType = void (*)(JITState*, const u32*); + using ProgramType = void (*)(JITState*, const u32*, const u32*); struct OptimizerState { bool can_skip_carry{}; @@ -113,7 +114,7 @@ void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) { JITState state{}; state.maxwell3d = &maxwell3d; state.registers = {}; - program(&state, parameters.data()); + program(&state, parameters.data(), parameters.data() + parameters.size()); } void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) { @@ -278,28 +279,13 @@ void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) { auto dst = Compile_GetRegister(opcode.src_a, RESULT); auto src = Compile_GetRegister(opcode.src_b, eax); - if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) { - shr(src, opcode.bf_src_bit); - } else if (opcode.bf_src_bit == 31) { - xor_(src, src); - } - // Don't bother masking the whole register since we're using a 32 bit register - if (opcode.bf_size != 31 && opcode.bf_size != 0) { - and_(src, opcode.GetBitfieldMask()); - } else if (opcode.bf_size == 0) { - xor_(src, src); - } - if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) { - shl(src, opcode.bf_dst_bit); - } else if (opcode.bf_dst_bit == 31) { - xor_(src, src); - } - const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit); - if (mask != 0xffffffff) { - and_(dst, mask); - } + and_(dst, mask); + shr(src, opcode.bf_src_bit); + and_(src, opcode.GetBitfieldMask()); + shl(src, opcode.bf_dst_bit); or_(dst, src); + Compile_ProcessResult(opcode.result_operation, opcode.dst); } @@ -308,17 +294,9 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) { const auto src = Compile_GetRegister(opcode.src_b, RESULT); shr(src, dst.cvt8()); - if (opcode.bf_size != 0 && opcode.bf_size != 31) { - and_(src, opcode.GetBitfieldMask()); - } else if (opcode.bf_size == 0) { - xor_(src, src); - } + and_(src, opcode.GetBitfieldMask()); + shl(src, opcode.bf_dst_bit); - if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) { - shl(src, opcode.bf_dst_bit); - } else if (opcode.bf_dst_bit == 31) { - xor_(src, src); - } Compile_ProcessResult(opcode.result_operation, opcode.dst); } @@ -326,13 +304,8 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) { const auto dst = Compile_GetRegister(opcode.src_a, ecx); const auto src = Compile_GetRegister(opcode.src_b, RESULT); - if (opcode.bf_src_bit != 0) { - shr(src, opcode.bf_src_bit); - } - - if (opcode.bf_size != 31) { - and_(src, opcode.GetBitfieldMask()); - } + shr(src, opcode.bf_src_bit); + and_(src, opcode.GetBitfieldMask()); shl(src, dst.cvt8()); Compile_ProcessResult(opcode.result_operation, opcode.dst); @@ -410,7 +383,7 @@ void MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) { Xbyak::Label end; auto value = Compile_GetRegister(opcode.src_a, eax); - test(value, value); + cmp(value, 0); // test(value, value); if (optimizer.has_delayed_pc) { switch (opcode.branch_condition) { case Macro::BranchCondition::Zero: @@ -428,17 +401,11 @@ void MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) { Xbyak::Label handle_post_exit{}; Xbyak::Label skip{}; jmp(skip, T_NEAR); - if (opcode.is_exit) { - L(handle_post_exit); - // Execute 1 instruction - mov(BRANCH_HOLDER, end_of_code); - // Jump to next instruction to skip delay slot check - jmp(labels[jump_address], T_NEAR); - } else { - L(handle_post_exit); - xor_(BRANCH_HOLDER, BRANCH_HOLDER); - jmp(labels[jump_address], T_NEAR); - } + + L(handle_post_exit); + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + jmp(labels[jump_address], T_NEAR); + L(skip); mov(BRANCH_HOLDER, handle_post_exit); jmp(delay_skip[pc], T_NEAR); @@ -489,6 +456,7 @@ void MacroJITx64Impl::Compile() { // JIT state mov(STATE, Common::X64::ABI_PARAM1); mov(PARAMETERS, Common::X64::ABI_PARAM2); + mov(MAX_PARAMETER, Common::X64::ABI_PARAM3); xor_(RESULT, RESULT); xor_(METHOD_ADDRESS, METHOD_ADDRESS); xor_(BRANCH_HOLDER, BRANCH_HOLDER); @@ -599,7 +567,22 @@ bool MacroJITx64Impl::Compile_NextInstruction() { return true; } +static void WarnInvalidParameter(uintptr_t parameter, uintptr_t max_parameter) { + LOG_CRITICAL(HW_GPU, + "Macro JIT: invalid parameter access 0x{:x} (0x{:x} is the last parameter)", + parameter, max_parameter - sizeof(u32)); +} + Xbyak::Reg32 MacroJITx64Impl::Compile_FetchParameter() { + Xbyak::Label parameter_ok{}; + cmp(PARAMETERS, MAX_PARAMETER); + jb(parameter_ok, T_NEAR); + Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0); + mov(Common::X64::ABI_PARAM1, PARAMETERS); + mov(Common::X64::ABI_PARAM2, MAX_PARAMETER); + Common::X64::CallFarFunction(*this, &WarnInvalidParameter); + Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0); + L(parameter_ok); mov(eax, dword[PARAMETERS]); add(PARAMETERS, sizeof(u32)); return eax; diff --git a/src/video_core/macro/macro_jit_x64.h b/src/video_core/macro/macro_jit_x64.h index 773b037ae..99ee1b9e6 100644 --- a/src/video_core/macro/macro_jit_x64.h +++ b/src/video_core/macro/macro_jit_x64.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index 4ff3fa268..cca401c74 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu emulator team -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> @@ -8,182 +7,208 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/device_memory.h" #include "core/hle/kernel/k_page_table.h" #include "core/hle/kernel/k_process.h" #include "core/memory.h" -#include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_base.h" namespace Tegra { -MemoryManager::MemoryManager(Core::System& system_) - : system{system_}, page_table(page_table_size) {} +std::atomic<size_t> MemoryManager::unique_identifier_generator{}; + +MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64 big_page_bits_, + u64 page_bits_) + : system{system_}, memory{system.Memory()}, device_memory{system.DeviceMemory()}, + address_space_bits{address_space_bits_}, page_bits{page_bits_}, big_page_bits{big_page_bits_}, + entries{}, big_entries{}, page_table{address_space_bits, address_space_bits + page_bits - 38, + page_bits != big_page_bits ? page_bits : 0}, + unique_identifier{unique_identifier_generator.fetch_add(1, std::memory_order_acq_rel)} { + address_space_size = 1ULL << address_space_bits; + page_size = 1ULL << page_bits; + page_mask = page_size - 1ULL; + big_page_size = 1ULL << big_page_bits; + big_page_mask = big_page_size - 1ULL; + const u64 page_table_bits = address_space_bits - page_bits; + const u64 big_page_table_bits = address_space_bits - big_page_bits; + const u64 page_table_size = 1ULL << page_table_bits; + const u64 big_page_table_size = 1ULL << big_page_table_bits; + page_table_mask = page_table_size - 1; + big_page_table_mask = big_page_table_size - 1; + + big_entries.resize(big_page_table_size / 32, 0); + big_page_table_cpu.resize(big_page_table_size); + big_page_continous.resize(big_page_table_size / continous_bits, 0); + entries.resize(page_table_size / 32, 0); +} MemoryManager::~MemoryManager() = default; -void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { - rasterizer = rasterizer_; -} - -GPUVAddr MemoryManager::UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size) { - u64 remaining_size{size}; - for (u64 offset{}; offset < size; offset += page_size) { - if (remaining_size < page_size) { - SetPageEntry(gpu_addr + offset, page_entry + offset, remaining_size); - } else { - SetPageEntry(gpu_addr + offset, page_entry + offset); - } - remaining_size -= page_size; +template <bool is_big_page> +MemoryManager::EntryType MemoryManager::GetEntry(size_t position) const { + if constexpr (is_big_page) { + position = position >> big_page_bits; + const u64 entry_mask = big_entries[position / 32]; + const size_t sub_index = position % 32; + return static_cast<EntryType>((entry_mask >> (2 * sub_index)) & 0x03ULL); + } else { + position = position >> page_bits; + const u64 entry_mask = entries[position / 32]; + const size_t sub_index = position % 32; + return static_cast<EntryType>((entry_mask >> (2 * sub_index)) & 0x03ULL); } - return gpu_addr; } -GPUVAddr MemoryManager::Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size) { - const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first); - if (it != map_ranges.end() && it->first == gpu_addr) { - it->second = size; +template <bool is_big_page> +void MemoryManager::SetEntry(size_t position, MemoryManager::EntryType entry) { + if constexpr (is_big_page) { + position = position >> big_page_bits; + const u64 entry_mask = big_entries[position / 32]; + const size_t sub_index = position % 32; + big_entries[position / 32] = + (~(3ULL << sub_index * 2) & entry_mask) | (static_cast<u64>(entry) << sub_index * 2); } else { - map_ranges.insert(it, MapRange{gpu_addr, size}); + position = position >> page_bits; + const u64 entry_mask = entries[position / 32]; + const size_t sub_index = position % 32; + entries[position / 32] = + (~(3ULL << sub_index * 2) & entry_mask) | (static_cast<u64>(entry) << sub_index * 2); } - return UpdateRange(gpu_addr, cpu_addr, size); } -GPUVAddr MemoryManager::MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align) { - return Map(cpu_addr, *FindFreeRange(size, align), size); +inline bool MemoryManager::IsBigPageContinous(size_t big_page_index) const { + const u64 entry_mask = big_page_continous[big_page_index / continous_bits]; + const size_t sub_index = big_page_index % continous_bits; + return ((entry_mask >> sub_index) & 0x1ULL) != 0; } -GPUVAddr MemoryManager::MapAllocate32(VAddr cpu_addr, std::size_t size) { - const std::optional<GPUVAddr> gpu_addr = FindFreeRange(size, 1, true); - ASSERT(gpu_addr); - return Map(cpu_addr, *gpu_addr, size); +inline void MemoryManager::SetBigPageContinous(size_t big_page_index, bool value) { + const u64 continous_mask = big_page_continous[big_page_index / continous_bits]; + const size_t sub_index = big_page_index % continous_bits; + big_page_continous[big_page_index / continous_bits] = + (~(1ULL << sub_index) & continous_mask) | (value ? 1ULL << sub_index : 0); } -void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) { - if (size == 0) { - return; - } - const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first); - if (it != map_ranges.end()) { - ASSERT(it->first == gpu_addr); - map_ranges.erase(it); - } else { - UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr); - } - const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); - - for (const auto& [map_addr, map_size] : submapped_ranges) { - // Flush and invalidate through the GPU interface, to be asynchronous if possible. - const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr); - ASSERT(cpu_addr); - - rasterizer->UnmapMemory(*cpu_addr, map_size); +template <MemoryManager::EntryType entry_type> +GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, + size_t size) { + u64 remaining_size{size}; + if constexpr (entry_type == EntryType::Mapped) { + page_table.ReserveRange(gpu_addr, size); } - - UpdateRange(gpu_addr, PageEntry::State::Unmapped, size); -} - -std::optional<GPUVAddr> MemoryManager::AllocateFixed(GPUVAddr gpu_addr, std::size_t size) { for (u64 offset{}; offset < size; offset += page_size) { - if (!GetPageEntry(gpu_addr + offset).IsUnmapped()) { - return std::nullopt; + const GPUVAddr current_gpu_addr = gpu_addr + offset; + [[maybe_unused]] const auto current_entry_type = GetEntry<false>(current_gpu_addr); + SetEntry<false>(current_gpu_addr, entry_type); + if (current_entry_type != entry_type) { + rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, page_size); + } + if constexpr (entry_type == EntryType::Mapped) { + const VAddr current_cpu_addr = cpu_addr + offset; + const auto index = PageEntryIndex<false>(current_gpu_addr); + const u32 sub_value = static_cast<u32>(current_cpu_addr >> cpu_page_bits); + page_table[index] = sub_value; } + remaining_size -= page_size; } - - return UpdateRange(gpu_addr, PageEntry::State::Allocated, size); -} - -GPUVAddr MemoryManager::Allocate(std::size_t size, std::size_t align) { - return *AllocateFixed(*FindFreeRange(size, align), size); + return gpu_addr; } -void MemoryManager::TryLockPage(PageEntry page_entry, std::size_t size) { - if (!page_entry.IsValid()) { - return; +template <MemoryManager::EntryType entry_type> +GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, + size_t size) { + u64 remaining_size{size}; + for (u64 offset{}; offset < size; offset += big_page_size) { + const GPUVAddr current_gpu_addr = gpu_addr + offset; + [[maybe_unused]] const auto current_entry_type = GetEntry<true>(current_gpu_addr); + SetEntry<true>(current_gpu_addr, entry_type); + if (current_entry_type != entry_type) { + rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, big_page_size); + } + if constexpr (entry_type == EntryType::Mapped) { + const VAddr current_cpu_addr = cpu_addr + offset; + const auto index = PageEntryIndex<true>(current_gpu_addr); + const u32 sub_value = static_cast<u32>(current_cpu_addr >> cpu_page_bits); + big_page_table_cpu[index] = sub_value; + const bool is_continous = ([&] { + uintptr_t base_ptr{ + reinterpret_cast<uintptr_t>(memory.GetPointerSilent(current_cpu_addr))}; + if (base_ptr == 0) { + return false; + } + for (VAddr start_cpu = current_cpu_addr + page_size; + start_cpu < current_cpu_addr + big_page_size; start_cpu += page_size) { + base_ptr += page_size; + auto next_ptr = reinterpret_cast<uintptr_t>(memory.GetPointerSilent(start_cpu)); + if (next_ptr == 0 || base_ptr != next_ptr) { + return false; + } + } + return true; + })(); + SetBigPageContinous(index, is_continous); + } + remaining_size -= big_page_size; } - - ASSERT(system.CurrentProcess() - ->PageTable() - .LockForDeviceAddressSpace(page_entry.ToAddress(), size) - .IsSuccess()); + return gpu_addr; } -void MemoryManager::TryUnlockPage(PageEntry page_entry, std::size_t size) { - if (!page_entry.IsValid()) { - return; - } - - ASSERT(system.CurrentProcess() - ->PageTable() - .UnlockForDeviceAddressSpace(page_entry.ToAddress(), size) - .IsSuccess()); +void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } -PageEntry MemoryManager::GetPageEntry(GPUVAddr gpu_addr) const { - return page_table[PageEntryIndex(gpu_addr)]; +GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, + bool is_big_pages) { + if (is_big_pages) [[likely]] { + return BigPageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size); + } + return PageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size); } -void MemoryManager::SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size) { - // TODO(bunnei): We should lock/unlock device regions. This currently causes issues due to - // improper tracking, but should be fixed in the future. - - //// Unlock the old page - // TryUnlockPage(page_table[PageEntryIndex(gpu_addr)], size); - - //// Lock the new page - // TryLockPage(page_entry, size); - auto& current_page = page_table[PageEntryIndex(gpu_addr)]; - - if ((!current_page.IsValid() && page_entry.IsValid()) || - current_page.ToAddress() != page_entry.ToAddress()) { - rasterizer->ModifyGPUMemory(gpu_addr, size); +GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) { + if (is_big_pages) [[likely]] { + return BigPageTableOp<EntryType::Reserved>(gpu_addr, 0, size); } - - current_page = page_entry; + return PageTableOp<EntryType::Reserved>(gpu_addr, 0, size); } -std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align, - bool start_32bit_address) const { - if (!align) { - align = page_size; - } else { - align = Common::AlignUp(align, page_size); +void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) { + if (size == 0) { + return; } + const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); - u64 available_size{}; - GPUVAddr gpu_addr{start_32bit_address ? address_space_start_low : address_space_start}; - while (gpu_addr + available_size < address_space_size) { - if (GetPageEntry(gpu_addr + available_size).IsUnmapped()) { - available_size += page_size; - - if (available_size >= size) { - return gpu_addr; - } - } else { - gpu_addr += available_size + page_size; - available_size = 0; + for (const auto& [map_addr, map_size] : submapped_ranges) { + // Flush and invalidate through the GPU interface, to be asynchronous if possible. + const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr); + ASSERT(cpu_addr); - const auto remainder{gpu_addr % align}; - if (remainder) { - gpu_addr = (gpu_addr - remainder) + align; - } - } + rasterizer->UnmapMemory(*cpu_addr, map_size); } - return std::nullopt; + BigPageTableOp<EntryType::Free>(gpu_addr, 0, size); + PageTableOp<EntryType::Free>(gpu_addr, 0, size); } std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const { - if (gpu_addr == 0) { + if (!IsWithinGPUAddressRange(gpu_addr)) [[unlikely]] { return std::nullopt; } - const auto page_entry{GetPageEntry(gpu_addr)}; - if (!page_entry.IsValid()) { - return std::nullopt; + if (GetEntry<true>(gpu_addr) != EntryType::Mapped) [[unlikely]] { + if (GetEntry<false>(gpu_addr) != EntryType::Mapped) { + return std::nullopt; + } + + const VAddr cpu_addr_base = static_cast<VAddr>(page_table[PageEntryIndex<false>(gpu_addr)]) + << cpu_page_bits; + return cpu_addr_base + (gpu_addr & page_mask); } - return page_entry.ToAddress() + (gpu_addr & page_mask); + const VAddr cpu_addr_base = + static_cast<VAddr>(big_page_table_cpu[PageEntryIndex<true>(gpu_addr)]) << cpu_page_bits; + return cpu_addr_base + (gpu_addr & big_page_mask); } std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t size) const { @@ -191,7 +216,7 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t s const size_t page_last{(addr + size + page_size - 1) >> page_bits}; while (page_index < page_last) { const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { + if (page_addr) { return page_addr; } ++page_index; @@ -208,7 +233,7 @@ T MemoryManager::Read(GPUVAddr addr) const { return value; } - UNREACHABLE(); + ASSERT(false); return {}; } @@ -221,7 +246,7 @@ void MemoryManager::Write(GPUVAddr addr, T data) { return; } - UNREACHABLE(); + ASSERT(false); } template u8 MemoryManager::Read<u8>(GPUVAddr addr) const; @@ -234,126 +259,298 @@ template void MemoryManager::Write<u32>(GPUVAddr addr, u32 data); template void MemoryManager::Write<u64>(GPUVAddr addr, u64 data); u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) { - if (!GetPageEntry(gpu_addr).IsValid()) { - return {}; - } - const auto address{GpuToCpuAddress(gpu_addr)}; if (!address) { return {}; } - return system.Memory().GetPointer(*address); + return memory.GetPointer(*address); } const u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) const { - if (!GetPageEntry(gpu_addr).IsValid()) { - return {}; - } - const auto address{GpuToCpuAddress(gpu_addr)}; if (!address) { return {}; } - return system.Memory().GetPointer(*address); -} - -size_t MemoryManager::BytesToMapEnd(GPUVAddr gpu_addr) const noexcept { - auto it = std::ranges::upper_bound(map_ranges, gpu_addr, {}, &MapRange::first); - --it; - return it->second - (gpu_addr - it->first); -} - -void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, - bool is_safe) const { + return memory.GetPointer(*address); +} + +#ifdef _MSC_VER // no need for gcc / clang but msvc's compiler is more conservative with inlining. +#pragma inline_recursion(on) +#endif + +template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped> +inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, + FuncMapped&& func_mapped, FuncReserved&& func_reserved, + FuncUnmapped&& func_unmapped) const { + static constexpr bool BOOL_BREAK_MAPPED = std::is_same_v<FuncMapped, bool>; + static constexpr bool BOOL_BREAK_RESERVED = std::is_same_v<FuncReserved, bool>; + static constexpr bool BOOL_BREAK_UNMAPPED = std::is_same_v<FuncUnmapped, bool>; + u64 used_page_size; + u64 used_page_mask; + u64 used_page_bits; + if constexpr (is_big_pages) { + used_page_size = big_page_size; + used_page_mask = big_page_mask; + used_page_bits = big_page_bits; + } else { + used_page_size = page_size; + used_page_mask = page_mask; + used_page_bits = page_bits; + } std::size_t remaining_size{size}; - std::size_t page_index{gpu_src_addr >> page_bits}; - std::size_t page_offset{gpu_src_addr & page_mask}; + std::size_t page_index{gpu_src_addr >> used_page_bits}; + std::size_t page_offset{gpu_src_addr & used_page_mask}; + GPUVAddr current_address = gpu_src_addr; while (remaining_size > 0) { const std::size_t copy_amount{ - std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { - const auto src_addr{*page_addr + page_offset}; - if (is_safe) { - // Flush must happen on the rasterizer interface, such that memory is always - // synchronous when it is read (even when in asynchronous GPU mode). - // Fixes Dead Cells title menu. - rasterizer->FlushRegion(src_addr, copy_amount); + std::min(static_cast<std::size_t>(used_page_size) - page_offset, remaining_size)}; + auto entry = GetEntry<is_big_pages>(current_address); + if (entry == EntryType::Mapped) [[likely]] { + if constexpr (BOOL_BREAK_MAPPED) { + if (func_mapped(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_mapped(page_index, page_offset, copy_amount); } - system.Memory().ReadBlockUnsafe(src_addr, dest_buffer, copy_amount); - } else { - std::memset(dest_buffer, 0, copy_amount); - } + } else if (entry == EntryType::Reserved) { + if constexpr (BOOL_BREAK_RESERVED) { + if (func_reserved(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_reserved(page_index, page_offset, copy_amount); + } + + } else [[unlikely]] { + if constexpr (BOOL_BREAK_UNMAPPED) { + if (func_unmapped(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_unmapped(page_index, page_offset, copy_amount); + } + } page_index++; page_offset = 0; - dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; remaining_size -= copy_amount; + current_address += copy_amount; } } +template <bool is_safe> +void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, + std::size_t size) const { + auto set_to_zero = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, std::size_t copy_amount) { + std::memset(dest_buffer, 0, copy_amount); + dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; + }; + auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + if constexpr (is_safe) { + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + } + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(dest_buffer, physical, copy_amount); + dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; + }; + auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + if constexpr (is_safe) { + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + } + if (!IsBigPageContinous(page_index)) [[unlikely]] { + memory.ReadBlockUnsafe(cpu_addr_base, dest_buffer, copy_amount); + } else { + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(dest_buffer, physical, copy_amount); + } + dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; + }; + auto read_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, set_to_zero, set_to_zero); + }; + MemoryOperation<true>(gpu_src_addr, size, mapped_big, set_to_zero, read_short_pages); +} + void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const { - ReadBlockImpl(gpu_src_addr, dest_buffer, size, true); + ReadBlockImpl<true>(gpu_src_addr, dest_buffer, size); } void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, const std::size_t size) const { - ReadBlockImpl(gpu_src_addr, dest_buffer, size, false); + ReadBlockImpl<false>(gpu_src_addr, dest_buffer, size); } -void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, - bool is_safe) { - std::size_t remaining_size{size}; - std::size_t page_index{gpu_dest_addr >> page_bits}; - std::size_t page_offset{gpu_dest_addr & page_mask}; - - while (remaining_size > 0) { - const std::size_t copy_amount{ - std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { - const auto dest_addr{*page_addr + page_offset}; - - if (is_safe) { - // Invalidate must happen on the rasterizer interface, such that memory is always - // synchronous when it is written (even when in asynchronous GPU mode). - rasterizer->InvalidateRegion(dest_addr, copy_amount); - } - system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount); +template <bool is_safe> +void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, + std::size_t size) { + auto just_advance = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, std::size_t copy_amount) { + src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; + }; + auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + if constexpr (is_safe) { + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); } - - page_index++; - page_offset = 0; + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(physical, src_buffer, copy_amount); src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; - remaining_size -= copy_amount; - } + }; + auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + if constexpr (is_safe) { + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); + } + if (!IsBigPageContinous(page_index)) [[unlikely]] { + memory.WriteBlockUnsafe(cpu_addr_base, src_buffer, copy_amount); + } else { + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(physical, src_buffer, copy_amount); + } + src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; + }; + auto write_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, just_advance, just_advance); + }; + MemoryOperation<true>(gpu_dest_addr, size, mapped_big, just_advance, write_short_pages); } void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) { - WriteBlockImpl(gpu_dest_addr, src_buffer, size, true); + WriteBlockImpl<true>(gpu_dest_addr, src_buffer, size); } void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) { - WriteBlockImpl(gpu_dest_addr, src_buffer, size, false); + WriteBlockImpl<false>(gpu_dest_addr, src_buffer, size); } void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size) const { - size_t remaining_size{size}; - size_t page_index{gpu_addr >> page_bits}; - size_t page_offset{gpu_addr & page_mask}; - while (remaining_size > 0) { - const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; - if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) { - rasterizer->FlushRegion(*page_addr + page_offset, num_bytes); + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) {}; + + auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + }; + auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + }; + auto flush_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, flush_short_pages); +} + +bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size) const { + bool result = false; + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { return false; }; + + auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + result |= rasterizer->MustFlushRegion(cpu_addr_base, copy_amount); + return result; + }; + auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + result |= rasterizer->MustFlushRegion(cpu_addr_base, copy_amount); + return result; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + return result; + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, check_short_pages); + return result; +} + +size_t MemoryManager::MaxContinousRange(GPUVAddr gpu_addr, size_t size) const { + std::optional<VAddr> old_page_addr{}; + size_t range_so_far = 0; + bool result{false}; + auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + std::size_t copy_amount) { + result = true; + return true; + }; + auto short_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + if (old_page_addr && *old_page_addr != cpu_addr_base) { + result = true; + return true; } - ++page_index; - page_offset = 0; - remaining_size -= num_bytes; - } + range_so_far += copy_amount; + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto big_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + if (old_page_addr && *old_page_addr != cpu_addr_base) { + return true; + } + range_so_far += copy_amount; + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, short_check, fail, fail); + return result; + }; + MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages); + return range_so_far; +} + +void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size) const { + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) {}; + + auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); + }; + auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); + }; + auto invalidate_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, invalidate_short_pages); } void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size) { @@ -367,87 +564,134 @@ void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std } bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const { - const auto cpu_addr{GpuToCpuAddress(gpu_addr)}; - if (!cpu_addr) { + if (GetEntry<true>(gpu_addr) == EntryType::Mapped) [[likely]] { + size_t page_index = gpu_addr >> big_page_bits; + if (IsBigPageContinous(page_index)) [[likely]] { + const std::size_t page{(page_index & big_page_mask) + size}; + return page <= big_page_size; + } + const std::size_t page{(gpu_addr & Core::Memory::YUZU_PAGEMASK) + size}; + return page <= Core::Memory::YUZU_PAGESIZE; + } + if (GetEntry<false>(gpu_addr) != EntryType::Mapped) { return false; } - const std::size_t page{(*cpu_addr & Core::Memory::PAGE_MASK) + size}; - return page <= Core::Memory::PAGE_SIZE; + const std::size_t page{(gpu_addr & Core::Memory::YUZU_PAGEMASK) + size}; + return page <= Core::Memory::YUZU_PAGESIZE; } bool MemoryManager::IsContinousRange(GPUVAddr gpu_addr, std::size_t size) const { - size_t page_index{gpu_addr >> page_bits}; - const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits}; std::optional<VAddr> old_page_addr{}; - while (page_index != page_last) { - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (!page_addr || *page_addr == 0) { - return false; + bool result{true}; + auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + std::size_t copy_amount) { + result = false; + return true; + }; + auto short_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + if (old_page_addr && *old_page_addr != cpu_addr_base) { + result = false; + return true; } - if (old_page_addr) { - if (*old_page_addr + page_size != *page_addr) { - return false; - } + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto big_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + if (old_page_addr && *old_page_addr != cpu_addr_base) { + result = false; + return true; } - old_page_addr = page_addr; - ++page_index; - } - return true; + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, short_check, fail, fail); + return !result; + }; + MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages); + return result; } bool MemoryManager::IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const { - size_t page_index{gpu_addr >> page_bits}; - const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits}; - while (page_index < page_last) { - if (!page_table[page_index].IsValid() || page_table[page_index].ToAddress() == 0) { - return false; - } - ++page_index; - } - return true; + bool result{true}; + auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { + result = false; + return true; + }; + auto pass = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { return false; }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, pass, pass, fail); + return !result; + }; + MemoryOperation<true>(gpu_addr, size, pass, fail, check_short_pages); + return result; } std::vector<std::pair<GPUVAddr, std::size_t>> MemoryManager::GetSubmappedRange( GPUVAddr gpu_addr, std::size_t size) const { std::vector<std::pair<GPUVAddr, std::size_t>> result{}; - size_t page_index{gpu_addr >> page_bits}; - size_t remaining_size{size}; - size_t page_offset{gpu_addr & page_mask}; std::optional<std::pair<GPUVAddr, std::size_t>> last_segment{}; std::optional<VAddr> old_page_addr{}; - const auto extend_size = [&last_segment, &page_index, &page_offset](std::size_t bytes) { - if (!last_segment) { - const GPUVAddr new_base_addr = (page_index << page_bits) + page_offset; - last_segment = {new_base_addr, bytes}; - } else { - last_segment->second += bytes; - } - }; - const auto split = [&last_segment, &result] { + const auto split = [&last_segment, &result]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { if (last_segment) { result.push_back(*last_segment); last_segment = std::nullopt; } }; - while (remaining_size > 0) { - const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (!page_addr || *page_addr == 0) { - split(); - } else if (old_page_addr) { - if (*old_page_addr + page_size != *page_addr) { - split(); + const auto extend_size_big = [this, &split, &old_page_addr, + &last_segment](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset; + if (old_page_addr) { + if (*old_page_addr != cpu_addr_base) { + split(0, 0, 0); + } + } + old_page_addr = {cpu_addr_base + copy_amount}; + if (!last_segment) { + const GPUVAddr new_base_addr = (page_index << big_page_bits) + offset; + last_segment = {new_base_addr, copy_amount}; + } else { + last_segment->second += copy_amount; + } + }; + const auto extend_size_short = [this, &split, &old_page_addr, + &last_segment](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + const VAddr cpu_addr_base = + (static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset; + if (old_page_addr) { + if (*old_page_addr != cpu_addr_base) { + split(0, 0, 0); } - extend_size(num_bytes); + } + old_page_addr = {cpu_addr_base + copy_amount}; + if (!last_segment) { + const GPUVAddr new_base_addr = (page_index << page_bits) + offset; + last_segment = {new_base_addr, copy_amount}; } else { - extend_size(num_bytes); + last_segment->second += copy_amount; } - ++page_index; - page_offset = 0; - remaining_size -= num_bytes; - old_page_addr = page_addr; - } - split(); + }; + auto do_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, extend_size_short, split, split); + }; + MemoryOperation<true>(gpu_addr, size, extend_size_big, split, do_short_pages); + split(0, 0, 0); return result; } diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 61bfe47c7..f992e29f3 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -1,76 +1,41 @@ -// Copyright 2018 yuzu emulator team -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once +#include <atomic> #include <map> #include <optional> #include <vector> #include "common/common_types.h" +#include "common/multi_level_page_table.h" +#include "common/virtual_buffer.h" namespace VideoCore { class RasterizerInterface; } namespace Core { +class DeviceMemory; +namespace Memory { +class Memory; +} // namespace Memory class System; -} +} // namespace Core namespace Tegra { -class PageEntry final { -public: - enum class State : u32 { - Unmapped = static_cast<u32>(-1), - Allocated = static_cast<u32>(-2), - }; - - constexpr PageEntry() = default; - constexpr PageEntry(State state_) : state{state_} {} - constexpr PageEntry(VAddr addr) : state{static_cast<State>(addr >> ShiftBits)} {} - - [[nodiscard]] constexpr bool IsUnmapped() const { - return state == State::Unmapped; - } - - [[nodiscard]] constexpr bool IsAllocated() const { - return state == State::Allocated; - } - - [[nodiscard]] constexpr bool IsValid() const { - return !IsUnmapped() && !IsAllocated(); - } - - [[nodiscard]] constexpr VAddr ToAddress() const { - if (!IsValid()) { - return {}; - } - - return static_cast<VAddr>(state) << ShiftBits; - } - - [[nodiscard]] constexpr PageEntry operator+(u64 offset) const { - // If this is a reserved value, offsets do not apply - if (!IsValid()) { - return *this; - } - return PageEntry{(static_cast<VAddr>(state) << ShiftBits) + offset}; - } - -private: - static constexpr std::size_t ShiftBits{12}; - - State state{State::Unmapped}; -}; -static_assert(sizeof(PageEntry) == 4, "PageEntry is too large"); - class MemoryManager final { public: - explicit MemoryManager(Core::System& system_); + explicit MemoryManager(Core::System& system_, u64 address_space_bits_ = 40, + u64 big_page_bits_ = 16, u64 page_bits_ = 12); ~MemoryManager(); + size_t GetID() const { + return unique_identifier; + } + /// Binds a renderer to the memory manager. void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); @@ -87,9 +52,6 @@ public: [[nodiscard]] u8* GetPointer(GPUVAddr addr); [[nodiscard]] const u8* GetPointer(GPUVAddr addr) const; - /// Returns the number of bytes until the end of the memory map containing the given GPU address - [[nodiscard]] size_t BytesToMapEnd(GPUVAddr gpu_addr) const noexcept; - /** * ReadBlock and WriteBlock are full read and write operations over virtual * GPU Memory. It's important to use these when GPU memory may not be continuous @@ -136,54 +98,95 @@ public: std::vector<std::pair<GPUVAddr, std::size_t>> GetSubmappedRange(GPUVAddr gpu_addr, std::size_t size) const; - [[nodiscard]] GPUVAddr Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size); - [[nodiscard]] GPUVAddr MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align); - [[nodiscard]] GPUVAddr MapAllocate32(VAddr cpu_addr, std::size_t size); - [[nodiscard]] std::optional<GPUVAddr> AllocateFixed(GPUVAddr gpu_addr, std::size_t size); - [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align); + GPUVAddr Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, bool is_big_pages = true); + GPUVAddr MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages = true); void Unmap(GPUVAddr gpu_addr, std::size_t size); void FlushRegion(GPUVAddr gpu_addr, size_t size) const; + void InvalidateRegion(GPUVAddr gpu_addr, size_t size) const; + + bool IsMemoryDirty(GPUVAddr gpu_addr, size_t size) const; + + size_t MaxContinousRange(GPUVAddr gpu_addr, size_t size) const; + + bool IsWithinGPUAddressRange(GPUVAddr gpu_addr) const { + return gpu_addr < address_space_size; + } + private: - [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const; - void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size); - GPUVAddr UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size); - [[nodiscard]] std::optional<GPUVAddr> FindFreeRange(std::size_t size, std::size_t align, - bool start_32bit_address = false) const; - - void TryLockPage(PageEntry page_entry, std::size_t size); - void TryUnlockPage(PageEntry page_entry, std::size_t size); - - void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, - bool is_safe) const; - void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, - bool is_safe); - - [[nodiscard]] static constexpr std::size_t PageEntryIndex(GPUVAddr gpu_addr) { - return (gpu_addr >> page_bits) & page_table_mask; + template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped> + inline void MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, FuncMapped&& func_mapped, + FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const; + + template <bool is_safe> + void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const; + + template <bool is_safe> + void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size); + + template <bool is_big_page> + [[nodiscard]] std::size_t PageEntryIndex(GPUVAddr gpu_addr) const { + if constexpr (is_big_page) { + return (gpu_addr >> big_page_bits) & big_page_table_mask; + } else { + return (gpu_addr >> page_bits) & page_table_mask; + } } - static constexpr u64 address_space_size = 1ULL << 40; - static constexpr u64 address_space_start = 1ULL << 32; - static constexpr u64 address_space_start_low = 1ULL << 16; - static constexpr u64 page_bits{16}; - static constexpr u64 page_size{1 << page_bits}; - static constexpr u64 page_mask{page_size - 1}; - static constexpr u64 page_table_bits{24}; - static constexpr u64 page_table_size{1 << page_table_bits}; - static constexpr u64 page_table_mask{page_table_size - 1}; + inline bool IsBigPageContinous(size_t big_page_index) const; + inline void SetBigPageContinous(size_t big_page_index, bool value); Core::System& system; + Core::Memory::Memory& memory; + Core::DeviceMemory& device_memory; + + const u64 address_space_bits; + const u64 page_bits; + u64 address_space_size; + u64 page_size; + u64 page_mask; + u64 page_table_mask; + static constexpr u64 cpu_page_bits{12}; + + const u64 big_page_bits; + u64 big_page_size; + u64 big_page_mask; + u64 big_page_table_mask; VideoCore::RasterizerInterface* rasterizer = nullptr; - std::vector<PageEntry> page_table; + enum class EntryType : u64 { + Free = 0, + Reserved = 1, + Mapped = 2, + }; + + std::vector<u64> entries; + std::vector<u64> big_entries; + + template <EntryType entry_type> + GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size); + + template <EntryType entry_type> + GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size); + + template <bool is_big_page> + inline EntryType GetEntry(size_t position) const; + + template <bool is_big_page> + inline void SetEntry(size_t position, EntryType entry); + + Common::MultiLevelPageTable<u32> page_table; + Common::VirtualBuffer<u32> big_page_table_cpu; + + std::vector<u64> big_page_continous; + + constexpr static size_t continous_bits = 64; - using MapRange = std::pair<GPUVAddr, size_t>; - std::vector<MapRange> map_ranges; + const size_t unique_identifier; - std::vector<std::pair<VAddr, std::size_t>> cache_invalidate_queue; + static std::atomic<size_t> unique_identifier_generator; }; } // namespace Tegra diff --git a/src/video_core/query_cache.h b/src/video_core/query_cache.h index 392f82eb7..b0ebe71b7 100644 --- a/src/video_core/query_cache.h +++ b/src/video_core/query_cache.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -18,9 +17,8 @@ #include "common/assert.h" #include "common/settings.h" -#include "core/core.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" @@ -93,13 +91,10 @@ private: }; template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter> -class QueryCacheBase { +class QueryCacheBase : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: - explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_) - : rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - gpu_memory{gpu_memory_}, streams{{CounterStream{static_cast<QueryCache&>(*this), + explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_) + : rasterizer{rasterizer_}, streams{{CounterStream{static_cast<QueryCache&>(*this), VideoCore::QueryType::SamplesPassed}}} {} void InvalidateRegion(VAddr addr, std::size_t size) { @@ -120,13 +115,13 @@ public: */ void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) { std::unique_lock lock{mutex}; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); CachedQuery* query = TryGet(*cpu_addr); if (!query) { ASSERT_OR_EXECUTE(cpu_addr, return;); - u8* const host_ptr = gpu_memory.GetPointer(gpu_addr); + u8* const host_ptr = gpu_memory->GetPointer(gpu_addr); query = Register(type, *cpu_addr, host_ptr, timestamp.has_value()); } @@ -140,8 +135,10 @@ public: /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch. void UpdateCounters() { std::unique_lock lock{mutex}; - const auto& regs = maxwell3d.regs; - Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable); + if (maxwell3d) { + const auto& regs = maxwell3d->regs; + Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable); + } } /// Resets a counter to zero. It doesn't disable the query after resetting. @@ -217,8 +214,8 @@ private: return cache_begin < addr_end && addr_begin < cache_end; }; - const u64 page_end = addr_end >> PAGE_BITS; - for (u64 page = addr_begin >> PAGE_BITS; page <= page_end; ++page) { + const u64 page_end = addr_end >> YUZU_PAGEBITS; + for (u64 page = addr_begin >> YUZU_PAGEBITS; page <= page_end; ++page) { const auto& it = cached_queries.find(page); if (it == std::end(cached_queries)) { continue; @@ -238,14 +235,14 @@ private: /// Registers the passed parameters as cached and returns a pointer to the stored cached query. CachedQuery* Register(VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr, bool timestamp) { rasterizer.UpdatePagesCachedCount(cpu_addr, CachedQuery::SizeInBytes(timestamp), 1); - const u64 page = static_cast<u64>(cpu_addr) >> PAGE_BITS; + const u64 page = static_cast<u64>(cpu_addr) >> YUZU_PAGEBITS; return &cached_queries[page].emplace_back(static_cast<QueryCache&>(*this), type, cpu_addr, host_ptr); } /// Tries to a get a cached query. Returns nullptr on failure. CachedQuery* TryGet(VAddr addr) { - const u64 page = static_cast<u64>(addr) >> PAGE_BITS; + const u64 page = static_cast<u64>(addr) >> YUZU_PAGEBITS; const auto it = cached_queries.find(page); if (it == std::end(cached_queries)) { return nullptr; @@ -263,12 +260,10 @@ private: uncommitted_flushes->push_back(addr); } - static constexpr std::uintptr_t PAGE_SIZE = 4096; - static constexpr unsigned PAGE_BITS = 12; + static constexpr std::uintptr_t YUZU_PAGESIZE = 4096; + static constexpr unsigned YUZU_PAGEBITS = 12; VideoCore::RasterizerInterface& rasterizer; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::MemoryManager& gpu_memory; std::recursive_mutex mutex; diff --git a/src/video_core/rasterizer_accelerated.cpp b/src/video_core/rasterizer_accelerated.cpp index 4c9524702..4a197d65d 100644 --- a/src/video_core/rasterizer_accelerated.cpp +++ b/src/video_core/rasterizer_accelerated.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <atomic> @@ -25,8 +24,8 @@ void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int del u64 cache_bytes = 0; std::atomic_thread_fence(std::memory_order_acquire); - const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); - for (u64 page = addr >> PAGE_BITS; page != page_end; ++page) { + const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE); + for (u64 page = addr >> YUZU_PAGEBITS; page != page_end; ++page) { std::atomic_uint16_t& count = cached_pages.at(page >> 2).Count(page); if (delta > 0) { @@ -45,26 +44,27 @@ void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int del if (uncache_bytes == 0) { uncache_begin = page; } - uncache_bytes += PAGE_SIZE; + uncache_bytes += YUZU_PAGESIZE; } else if (uncache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(uncache_begin << PAGE_BITS, uncache_bytes, false); + cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, + false); uncache_bytes = 0; } if (count.load(std::memory_order::relaxed) == 1 && delta > 0) { if (cache_bytes == 0) { cache_begin = page; } - cache_bytes += PAGE_SIZE; + cache_bytes += YUZU_PAGESIZE; } else if (cache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(cache_begin << PAGE_BITS, cache_bytes, true); + cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); cache_bytes = 0; } } if (uncache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(uncache_begin << PAGE_BITS, uncache_bytes, false); + cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, false); } if (cache_bytes > 0) { - cpu_memory.RasterizerMarkRegionCached(cache_begin << PAGE_BITS, cache_bytes, true); + cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true); } } diff --git a/src/video_core/rasterizer_accelerated.h b/src/video_core/rasterizer_accelerated.h index 249644e50..7118b8aff 100644 --- a/src/video_core/rasterizer_accelerated.h +++ b/src/video_core/rasterizer_accelerated.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 1f1f12291..d2d40884c 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -17,6 +16,9 @@ class MemoryManager; namespace Engines { class AccelerateDMAInterface; } +namespace Control { +struct ChannelState; +} } // namespace Tegra namespace VideoCore { @@ -60,7 +62,10 @@ public: virtual void DisableGraphicsUniformBuffer(size_t stage, u32 index) = 0; /// Signal a GPU based semaphore as a fence - virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0; + virtual void SignalFence(std::function<void()>&& func) = 0; + + /// Send an operation to be done after a certain amount of flushes. + virtual void SyncOperation(std::function<void()>&& func) = 0; /// Signal a GPU based syncpoint as a fence virtual void SignalSyncPoint(u32 value) = 0; @@ -87,13 +92,13 @@ public: virtual void OnCPUWrite(VAddr addr, u64 size) = 0; /// Sync memory between guest and host. - virtual void SyncGuestHost() = 0; + virtual void InvalidateGPUCache() = 0; /// Unmap memory range virtual void UnmapMemory(VAddr addr, u64 size) = 0; /// Remap GPU memory range. This means underneath backing memory changed - virtual void ModifyGPUMemory(GPUVAddr addr, u64 size) = 0; + virtual void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) = 0; /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory /// and invalidated @@ -124,7 +129,7 @@ public: [[nodiscard]] virtual Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() = 0; virtual void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) = 0; + std::span<const u8> memory) = 0; /// Attempt to use a faster method to display the framebuffer to screen [[nodiscard]] virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config, @@ -138,5 +143,11 @@ public: /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading, const DiskResourceLoadCallback& callback) {} + + virtual void InitializeChannel(Tegra::Control::ChannelState& channel) {} + + virtual void BindChannel(Tegra::Control::ChannelState& channel) {} + + virtual void ReleaseChannel(s32 channel_id) {} }; } // namespace VideoCore diff --git a/src/video_core/renderer_base.cpp b/src/video_core/renderer_base.cpp index a99c33c37..45791aa75 100644 --- a/src/video_core/renderer_base.cpp +++ b/src/video_core/renderer_base.cpp @@ -1,9 +1,7 @@ -// Copyright 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2015 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/logging/log.h" -#include "common/settings.h" #include "core/frontend/emu_window.h" #include "video_core/renderer_base.h" @@ -27,6 +25,10 @@ void RendererBase::UpdateCurrentFramebufferLayout() { render_window.UpdateCurrentFramebufferLayout(layout.width, layout.height); } +bool RendererBase::IsScreenshotPending() const { + return renderer_settings.screenshot_requested; +} + void RendererBase::RequestScreenshot(void* data, std::function<void(bool)> callback, const Layout::FramebufferLayout& layout) { if (renderer_settings.screenshot_requested) { diff --git a/src/video_core/renderer_base.h b/src/video_core/renderer_base.h index c5f974080..8d20cbece 100644 --- a/src/video_core/renderer_base.h +++ b/src/video_core/renderer_base.h @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -83,6 +82,9 @@ public: /// Refreshes the settings common to all renderers void RefreshBaseSettings(); + /// Returns true if a screenshot is being processed + bool IsScreenshotPending() const; + /// Request a screenshot of the next frame void RequestScreenshot(void* data, std::function<void(bool)> callback, const Layout::FramebufferLayout& layout); diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index d4dd10bb6..08f4d69ab 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <span> @@ -135,6 +134,20 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_) buffer.Create(); glNamedBufferData(buffer.handle, 0x10'000, nullptr, GL_STREAM_COPY); } + + device_access_memory = [this]() -> u64 { + if (device.CanReportMemoryUsage()) { + return device.GetCurrentDedicatedVideoMemory() + 512_MiB; + } + return 2_GiB; // Return minimum requirements + }(); +} + +u64 BufferCacheRuntime::GetDeviceMemoryUsage() const { + if (device.CanReportMemoryUsage()) { + return device_access_memory - device.GetCurrentDedicatedVideoMemory(); + } + return 2_GiB; } void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, @@ -155,7 +168,7 @@ void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) { if (has_unified_vertex_buffers) { buffer.MakeResident(GL_READ_ONLY); glBufferAddressRangeNV(GL_ELEMENT_ARRAY_ADDRESS_NV, 0, buffer.HostGpuAddr() + offset, - static_cast<GLsizeiptr>(size)); + static_cast<GLsizeiptr>(Common::AlignUp(size, 4))); } else { glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer.Handle()); index_buffer_offset = offset; diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index 060d36427..a8c3f8b67 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -1,15 +1,12 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> #include <span> -#include "common/alignment.h" #include "common/common_types.h" -#include "common/dynamic_library.h" #include "video_core/buffer_cache/buffer_cache.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_opengl/gl_device.h" @@ -91,6 +88,8 @@ public: void BindImageBuffer(Buffer& buffer, u32 offset, u32 size, VideoCore::Surface::PixelFormat format); + u64 GetDeviceMemoryUsage() const; + void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) { const GLuint handle = fast_uniforms[stage][binding_index].handle; const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size); @@ -153,6 +152,14 @@ public: use_storage_buffers = use_storage_buffers_; } + u64 GetDeviceLocalMemory() const { + return device_access_memory; + } + + bool CanReportMemoryUsage() const { + return device.CanReportMemoryUsage(); + } + private: static constexpr std::array PABO_LUT{ GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, @@ -186,6 +193,8 @@ private: std::array<OGLBuffer, VideoCommon::NUM_COMPUTE_UNIFORM_BUFFERS> copy_compute_uniforms; u32 index_buffer_offset = 0; + + u64 device_access_memory; }; struct BufferCacheParams { diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp index 5c1f21c65..26d066004 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstring> @@ -29,12 +28,11 @@ bool ComputePipelineKey::operator==(const ComputePipelineKey& rhs) const noexcep } ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::KeplerCompute& kepler_compute_, - ProgramManager& program_manager_, const Shader::Info& info_, - std::string code, std::vector<u32> code_v) - : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, gpu_memory{gpu_memory_}, - kepler_compute{kepler_compute_}, program_manager{program_manager_}, info{info_} { + BufferCache& buffer_cache_, ProgramManager& program_manager_, + const Shader::Info& info_, std::string code, + std::vector<u32> code_v) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, + program_manager{program_manager_}, info{info_} { switch (device.GetShaderBackend()) { case Settings::ShaderBackend::GLSL: source_program = CreateProgram(code, GL_COMPUTE_SHADER); @@ -87,7 +85,7 @@ void ComputePipeline::Configure() { GLsizei texture_binding{}; GLsizei image_binding{}; - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const auto& cbufs{qmd.const_buffer_config}; const bool via_header_index{qmd.linked_tsc != 0}; const auto read_handle{[&](const auto& desc, u32 index) { @@ -102,12 +100,13 @@ void ComputePipeline::Configure() { const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + secondary_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; return TexturePair(lhs_raw | rhs_raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) { for (u32 index = 0; index < desc.count; ++index) { diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.h b/src/video_core/renderer_opengl/gl_compute_pipeline.h index 50c676365..6534dec32 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.h +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.h @@ -1,12 +1,10 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> #include <type_traits> -#include <utility> #include "common/common_types.h" #include "shader_recompiler/shader_info.h" @@ -51,10 +49,8 @@ static_assert(std::is_trivially_constructible_v<ComputePipelineKey>); class ComputePipeline { public: explicit ComputePipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::KeplerCompute& kepler_compute_, - ProgramManager& program_manager_, const Shader::Info& info_, - std::string code, std::vector<u32> code_v); + BufferCache& buffer_cache_, ProgramManager& program_manager_, + const Shader::Info& info_, std::string code, std::vector<u32> code_v); void Configure(); @@ -62,11 +58,17 @@ public: return writes_global_memory; } + void SetEngine(Tegra::Engines::KeplerCompute* kepler_compute_, + Tegra::MemoryManager* gpu_memory_) { + kepler_compute = kepler_compute_; + gpu_memory = gpu_memory_; + } + private: TextureCache& texture_cache; BufferCache& buffer_cache; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager* gpu_memory; + Tegra::Engines::KeplerCompute* kepler_compute; ProgramManager& program_manager; Shader::Info info; diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index e62912a22..1663e277d 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -1,13 +1,10 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <cstddef> #include <cstdlib> -#include <cstring> -#include <limits> #include <optional> #include <span> #include <stdexcept> @@ -15,13 +12,15 @@ #include <glad/glad.h> +#include "common/literals.h" #include "common/logging/log.h" -#include "common/scope_exit.h" #include "common/settings.h" #include "shader_recompiler/stage.h" #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" +using namespace Common::Literals; + namespace OpenGL { namespace { constexpr std::array LIMIT_UBOS = { @@ -168,6 +167,7 @@ Device::Device() { has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2; warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel; need_fastmath_off = is_nvidia; + can_report_memory = GLAD_GL_NVX_gpu_memory_info; // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive // uniform buffers as "push constants" @@ -279,4 +279,10 @@ void main() { })"); } +u64 Device::GetCurrentDedicatedVideoMemory() const { + GLint cur_avail_mem_kb = 0; + glGetIntegerv(GL_GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX, &cur_avail_mem_kb); + return static_cast<u64>(cur_avail_mem_kb) * 1_KiB; +} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 95c2e8d38..5ef51ebcf 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -20,6 +19,8 @@ public: [[nodiscard]] std::string GetVendorName() const; + u64 GetCurrentDedicatedVideoMemory() const; + u32 GetMaxUniformBuffers(Shader::Stage stage) const noexcept { return max_uniform_buffers[static_cast<size_t>(stage)]; } @@ -168,6 +169,10 @@ public: return vendor_name == "ATI Technologies Inc."; } + bool CanReportMemoryUsage() const { + return can_report_memory; + } + private: static bool TestVariableAoffi(); static bool TestPreciseBug(); @@ -210,6 +215,7 @@ private: bool need_fastmath_off{}; bool has_cbuf_ftou_bug{}; bool has_bool_ref_bug{}; + bool can_report_memory{}; std::string vendor_name; }; diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp index 151290101..91463f854 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.cpp +++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" @@ -11,10 +10,7 @@ namespace OpenGL { -GLInnerFence::GLInnerFence(u32 payload_, bool is_stubbed_) : FenceBase{payload_, is_stubbed_} {} - -GLInnerFence::GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_) - : FenceBase{address_, payload_, is_stubbed_} {} +GLInnerFence::GLInnerFence(bool is_stubbed_) : FenceBase{is_stubbed_} {} GLInnerFence::~GLInnerFence() = default; @@ -31,9 +27,8 @@ bool GLInnerFence::IsSignaled() const { return true; } ASSERT(sync_object.handle != 0); - GLsizei length; GLint sync_status; - glGetSynciv(sync_object.handle, GL_SYNC_STATUS, sizeof(GLint), &length, &sync_status); + glGetSynciv(sync_object.handle, GL_SYNC_STATUS, 1, nullptr, &sync_status); return sync_status == GL_SIGNALED; } @@ -50,12 +45,8 @@ FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterize BufferCache& buffer_cache_, QueryCache& query_cache_) : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {} -Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) { - return std::make_shared<GLInnerFence>(value, is_stubbed); -} - -Fence FenceManagerOpenGL::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) { - return std::make_shared<GLInnerFence>(addr, value, is_stubbed); +Fence FenceManagerOpenGL::CreateFence(bool is_stubbed) { + return std::make_shared<GLInnerFence>(is_stubbed); } void FenceManagerOpenGL::QueueFence(Fence& fence) { diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h index e714aa115..f1446e732 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.h +++ b/src/video_core/renderer_opengl/gl_fence_manager.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -17,8 +16,7 @@ namespace OpenGL { class GLInnerFence : public VideoCommon::FenceBase { public: - explicit GLInnerFence(u32 payload_, bool is_stubbed_); - explicit GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_); + explicit GLInnerFence(bool is_stubbed_); ~GLInnerFence(); void Queue(); @@ -41,8 +39,7 @@ public: QueryCache& query_cache); protected: - Fence CreateFence(u32 value, bool is_stubbed) override; - Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override; + Fence CreateFence(bool is_stubbed) override; void QueueFence(Fence& fence) override; bool IsFenceSignaled(Fence& fence) const override; void WaitFence(Fence& fence) override; diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index f8495896c..41493a7da 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -49,7 +48,7 @@ GLenum Stage(size_t stage_index) { case 4: return GL_FRAGMENT_SHADER; } - UNREACHABLE_MSG("{}", stage_index); + ASSERT_MSG(false, "{}", stage_index); return GL_NONE; } @@ -66,7 +65,7 @@ GLenum AssemblyStage(size_t stage_index) { case 4: return GL_FRAGMENT_PROGRAM_NV; } - UNREACHABLE_MSG("{}", stage_index); + ASSERT_MSG(false, "{}", stage_index); return GL_NONE; } @@ -170,15 +169,15 @@ ConfigureFuncPtr ConfigureFunc(const std::array<Shader::Info, 5>& infos, u32 ena } } // Anonymous namespace -GraphicsPipeline::GraphicsPipeline( - const Device& device, TextureCache& texture_cache_, BufferCache& buffer_cache_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - ProgramManager& program_manager_, StateTracker& state_tracker_, ShaderWorker* thread_worker, - VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources, - std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos, - const GraphicsPipelineKey& key_) - : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, - gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, program_manager{program_manager_}, +GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_cache_, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, ShaderWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, + std::array<std::string, 5> sources, + std::array<std::vector<u32>, 5> sources_spirv, + const std::array<const Shader::Info*, 5>& infos, + const GraphicsPipelineKey& key_) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_}, state_tracker{state_tracker_}, key{key_} { if (shader_notify) { shader_notify->MarkShaderBuilding(); @@ -243,10 +242,6 @@ GraphicsPipeline::GraphicsPipeline( case Settings::ShaderBackend::GLASM: if (!sources[stage].empty()) { assembly_programs[stage] = CompileProgram(sources[stage], AssemblyStage(stage)); - if (in_parallel) { - // Make sure program is built before continuing when building in parallel - glGetString(GL_PROGRAM_ERROR_STRING_NV); - } } break; case Settings::ShaderBackend::SPIRV: @@ -256,20 +251,18 @@ GraphicsPipeline::GraphicsPipeline( break; } } - if (in_parallel && backend != Settings::ShaderBackend::GLASM) { - // Make sure programs have built if we are building shaders in parallel - for (OGLProgram& program : source_programs) { - if (program.handle != 0) { - GLint status{}; - glGetProgramiv(program.handle, GL_LINK_STATUS, &status); - } - } + if (in_parallel) { + std::scoped_lock lock{built_mutex}; + built_fence.Create(); + // Flush this context to ensure compilation commands and fence are in the GPU pipe. + glFlush(); + built_condvar.notify_one(); + } else { + is_built = true; } if (shader_notify) { shader_notify->MarkShaderComplete(); } - is_built = true; - built_condvar.notify_one(); }}; if (thread_worker) { thread_worker->QueueWork(std::move(func)); @@ -292,7 +285,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.runtime.SetBaseStorageBindings(base_storage_bindings); buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers); - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { const Shader::Info& info{stage_infos[stage]}; @@ -306,7 +299,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { ++ssbo_index; } } - const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; @@ -319,13 +312,14 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + second_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; const u32 raw{lhs_raw | rhs_raw}; return TexturePair(raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { @@ -440,7 +434,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.UpdateGraphicsBuffers(is_indexed); buffer_cache.BindHostGeometryBuffers(is_indexed); - if (!is_built.load(std::memory_order::relaxed)) { + if (!IsBuilt()) { WaitForBuild(); } const bool use_assembly{assembly_programs[0].handle != 0}; @@ -585,8 +579,26 @@ void GraphicsPipeline::GenerateTransformFeedbackState() { } void GraphicsPipeline::WaitForBuild() { - std::unique_lock lock{built_mutex}; - built_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); + if (built_fence.handle == 0) { + std::unique_lock lock{built_mutex}; + built_condvar.wait(lock, [this] { return built_fence.handle != 0; }); + } + ASSERT(glClientWaitSync(built_fence.handle, 0, GL_TIMEOUT_IGNORED) != GL_WAIT_FAILED); + is_built = true; +} + +bool GraphicsPipeline::IsBuilt() noexcept { + if (is_built) { + return true; + } + if (built_fence.handle == 0) { + return false; + } + // Timeout of zero means this is non-blocking + const auto sync_status = glClientWaitSync(built_fence.handle, 0, 0); + ASSERT(sync_status != GL_WAIT_FAILED); + is_built = sync_status != GL_TIMEOUT_EXPIRED; + return is_built; } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h index 4e28d9a42..a0f0e63cb 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -14,7 +13,6 @@ #include "common/common_types.h" #include "shader_recompiler/shader_info.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_texture_cache.h" @@ -73,10 +71,9 @@ static_assert(std::is_trivially_constructible_v<GraphicsPipelineKey>); class GraphicsPipeline { public: explicit GraphicsPipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::Maxwell3D& maxwell3d_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - ShaderWorker* thread_worker, VideoCore::ShaderNotify* shader_notify, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, ShaderWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources, std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos, @@ -100,9 +97,7 @@ public: return writes_global_memory; } - [[nodiscard]] bool IsBuilt() const noexcept { - return is_built.load(std::memory_order::relaxed); - } + [[nodiscard]] bool IsBuilt() noexcept; template <typename Spec> static auto MakeConfigureSpecFunc() { @@ -111,6 +106,11 @@ public: }; } + void SetEngine(Tegra::Engines::Maxwell3D* maxwell3d_, Tegra::MemoryManager* gpu_memory_) { + maxwell3d = maxwell3d_; + gpu_memory = gpu_memory_; + } + private: template <typename Spec> void ConfigureImpl(bool is_indexed); @@ -123,8 +123,8 @@ private: TextureCache& texture_cache; BufferCache& buffer_cache; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::MemoryManager* gpu_memory; + Tegra::Engines::Maxwell3D* maxwell3d; ProgramManager& program_manager; StateTracker& state_tracker; const GraphicsPipelineKey key; @@ -154,7 +154,8 @@ private: std::mutex built_mutex; std::condition_variable built_condvar; - std::atomic_bool is_built{false}; + OGLSync built_fence{}; + bool is_built{false}; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_query_cache.cpp b/src/video_core/renderer_opengl/gl_query_cache.cpp index acebbf5f4..5070db441 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.cpp +++ b/src/video_core/renderer_opengl/gl_query_cache.cpp @@ -1,17 +1,13 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> -#include <cstring> #include <memory> -#include <unordered_map> #include <utility> #include <vector> #include <glad/glad.h> -#include "common/assert.h" #include "core/core.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" @@ -30,9 +26,8 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) { } // Anonymous namespace -QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_) - : QueryCacheBase(rasterizer_, maxwell3d_, gpu_memory_), gl_rasterizer{rasterizer_} {} +QueryCache::QueryCache(RasterizerOpenGL& rasterizer_) + : QueryCacheBase(rasterizer_), gl_rasterizer{rasterizer_} {} QueryCache::~QueryCache() = default; diff --git a/src/video_core/renderer_opengl/gl_query_cache.h b/src/video_core/renderer_opengl/gl_query_cache.h index 7bbe5cfe9..14ce59990 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.h +++ b/src/video_core/renderer_opengl/gl_query_cache.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -29,8 +28,7 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>; class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_); + explicit QueryCache(RasterizerOpenGL& rasterizer_); ~QueryCache(); OGLQuery AllocateQuery(VideoCore::QueryType type); diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 142412a8e..c2d80605d 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -1,14 +1,11 @@ -// Copyright 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2015 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <bitset> #include <memory> -#include <string> #include <string_view> -#include <tuple> #include <utility> #include <glad/glad.h> @@ -17,8 +14,9 @@ #include "common/logging/log.h" #include "common/math_util.h" #include "common/microprofile.h" +#include "common/scope_exit.h" #include "common/settings.h" -#include "core/memory.h" +#include "video_core/control/channel_state.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" @@ -58,22 +56,20 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra Core::Memory::Memory& cpu_memory_, const Device& device_, ScreenInfo& screen_info_, ProgramManager& program_manager_, StateTracker& state_tracker_) - : RasterizerAccelerated(cpu_memory_), gpu(gpu_), maxwell3d(gpu.Maxwell3D()), - kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_), - screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_), + : RasterizerAccelerated(cpu_memory_), gpu(gpu_), device(device_), screen_info(screen_info_), + program_manager(program_manager_), state_tracker(state_tracker_), texture_cache_runtime(device, program_manager, state_tracker), - texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), - buffer_cache_runtime(device), - buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - shader_cache(*this, emu_window_, maxwell3d, kepler_compute, gpu_memory, device, texture_cache, - buffer_cache, program_manager, state_tracker, gpu.ShaderNotify()), - query_cache(*this, maxwell3d, gpu_memory), accelerate_dma(buffer_cache), + texture_cache(texture_cache_runtime, *this), buffer_cache_runtime(device), + buffer_cache(*this, cpu_memory_, buffer_cache_runtime), + shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager, + state_tracker, gpu.ShaderNotify()), + query_cache(*this), accelerate_dma(buffer_cache), fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {} RasterizerOpenGL::~RasterizerOpenGL() = default; void RasterizerOpenGL::SyncVertexFormats() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::VertexFormats]) { return; } @@ -91,7 +87,7 @@ void RasterizerOpenGL::SyncVertexFormats() { } flags[Dirty::VertexFormat0 + index] = false; - const auto attrib = maxwell3d.regs.vertex_attrib_format[index]; + const auto attrib = maxwell3d->regs.vertex_attrib_format[index]; const auto gl_index = static_cast<GLuint>(index); // Disable constant attributes. @@ -115,13 +111,13 @@ void RasterizerOpenGL::SyncVertexFormats() { } void RasterizerOpenGL::SyncVertexInstances() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::VertexInstances]) { return; } flags[Dirty::VertexInstances] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) { if (!flags[Dirty::VertexInstance0 + index]) { continue; @@ -142,11 +138,11 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_load void RasterizerOpenGL::Clear() { MICROPROFILE_SCOPE(OpenGL_Clears); - if (!maxwell3d.ShouldExecute()) { + if (!maxwell3d->ShouldExecute()) { return; } - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; bool use_color{}; bool use_depth{}; bool use_stencil{}; @@ -212,28 +208,33 @@ void RasterizerOpenGL::Clear() { void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { MICROPROFILE_SCOPE(OpenGL_Drawing); + SCOPE_EXIT({ gpu.TickWork(); }); query_cache.UpdateCounters(); GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()}; if (!pipeline) { return; } + + gpu.TickWork(); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + pipeline->SetEngine(maxwell3d, gpu_memory); pipeline->Configure(is_indexed); SyncState(); - const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); + const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d->regs.draw.topology); BeginTransformFeedback(pipeline, primitive_mode); - const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance); + const GLuint base_instance = static_cast<GLuint>(maxwell3d->regs.vb_base_instance); const GLsizei num_instances = - static_cast<GLsizei>(is_instanced ? maxwell3d.mme_draw.instance_count : 1); + static_cast<GLsizei>(is_instanced ? maxwell3d->mme_draw.instance_count : 1); if (is_indexed) { - const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vb_element_base); - const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.index_array.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.vb_element_base); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.index_array.count); const GLvoid* const offset = buffer_cache_runtime.IndexOffset(); - const GLenum format = MaxwellToGL::IndexFormat(maxwell3d.regs.index_array.format); + const GLenum format = MaxwellToGL::IndexFormat(maxwell3d->regs.index_array.format); if (num_instances == 1 && base_instance == 0 && base_vertex == 0) { glDrawElements(primitive_mode, num_vertices, format, offset); } else if (num_instances == 1 && base_instance == 0) { @@ -252,8 +253,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { base_instance); } } else { - const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vertex_buffer.first); - const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.vertex_buffer.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.vertex_buffer.first); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.vertex_buffer.count); if (num_instances == 1 && base_instance == 0) { glDrawArrays(primitive_mode, base_vertex, num_vertices); } else if (base_instance == 0) { @@ -267,8 +268,6 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { ++num_queued_commands; has_written_global_memory |= pipeline->WritesGlobalMemory(); - - gpu.TickWork(); } void RasterizerOpenGL::DispatchCompute() { @@ -276,8 +275,9 @@ void RasterizerOpenGL::DispatchCompute() { if (!pipeline) { return; } + pipeline->SetEngine(kepler_compute, gpu_memory); pipeline->Configure(); - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z); ++num_queued_commands; has_written_global_memory |= pipeline->WritesGlobalMemory(); @@ -362,7 +362,7 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { } } -void RasterizerOpenGL::SyncGuestHost() { +void RasterizerOpenGL::InvalidateGPUCache() { MICROPROFILE_SCOPE(OpenGL_CacheManagement); shader_cache.SyncGuestHost(); { @@ -383,40 +383,30 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { shader_cache.OnCPUWrite(addr, size); } -void RasterizerOpenGL::ModifyGPUMemory(GPUVAddr addr, u64 size) { +void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { { std::scoped_lock lock{texture_cache.mutex}; - texture_cache.UnmapGPUMemory(addr, size); + texture_cache.UnmapGPUMemory(as_id, addr, size); } } -void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) { - if (!gpu.IsAsync()) { - gpu_memory.Write<u32>(addr, value); - return; - } - fence_manager.SignalSemaphore(addr, value); +void RasterizerOpenGL::SignalFence(std::function<void()>&& func) { + fence_manager.SignalFence(std::move(func)); +} + +void RasterizerOpenGL::SyncOperation(std::function<void()>&& func) { + fence_manager.SyncOperation(std::move(func)); } void RasterizerOpenGL::SignalSyncPoint(u32 value) { - if (!gpu.IsAsync()) { - gpu.IncrementSyncPoint(value); - return; - } fence_manager.SignalSyncPoint(value); } void RasterizerOpenGL::SignalReference() { - if (!gpu.IsAsync()) { - return; - } fence_manager.SignalOrdering(); } void RasterizerOpenGL::ReleaseFences() { - if (!gpu.IsAsync()) { - return; - } fence_manager.WaitPendingFences(); } @@ -433,6 +423,7 @@ void RasterizerOpenGL::WaitForIdle() { } void RasterizerOpenGL::FragmentBarrier() { + glTextureBarrier(); glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT); } @@ -485,13 +476,13 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerOpenGL::AccessAccelerateDMA() } void RasterizerOpenGL::AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) { - auto cpu_addr = gpu_memory.GpuToCpuAddress(address); + std::span<const u8> memory) { + auto cpu_addr = gpu_memory->GpuToCpuAddress(address); if (!cpu_addr) [[unlikely]] { - gpu_memory.WriteBlock(address, memory.data(), copy_size); + gpu_memory->WriteBlock(address, memory.data(), copy_size); return; } - gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size); + gpu_memory->WriteBlockUnsafe(address, memory.data(), copy_size); { std::unique_lock<std::mutex> lock{buffer_cache.mutex}; if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) { @@ -522,6 +513,8 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, // ASSERT_MSG(image_view->size.width == config.width, "Framebuffer width is different"); // ASSERT_MSG(image_view->size.height == config.height, "Framebuffer height is different"); + screen_info.texture.width = image_view->size.width; + screen_info.texture.height = image_view->size.height; screen_info.display_texture = image_view->Handle(Shader::TextureType::Color2D); screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); return true; @@ -552,19 +545,25 @@ void RasterizerOpenGL::SyncState() { } void RasterizerOpenGL::SyncViewport() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; const bool rescale_viewports = flags[VideoCommon::Dirty::RescaleViewports]; const bool dirty_viewport = flags[Dirty::Viewports] || rescale_viewports; const bool dirty_clip_control = flags[Dirty::ClipControl]; - if (dirty_clip_control || flags[Dirty::FrontFace]) { + if (dirty_viewport || dirty_clip_control || flags[Dirty::FrontFace]) { flags[Dirty::FrontFace] = false; GLenum mode = MaxwellToGL::FrontFace(regs.front_face); - if (regs.screen_y_control.triangle_rast_flip != 0 && - regs.viewport_transform[0].scale_y < 0.0f) { + bool flip_faces = true; + if (regs.screen_y_control.triangle_rast_flip != 0) { + flip_faces = !flip_faces; + } + if (regs.viewport_transform[0].scale_y < 0.0f) { + flip_faces = !flip_faces; + } + if (flip_faces) { switch (mode) { case GL_CW: mode = GL_CCW; @@ -652,23 +651,23 @@ void RasterizerOpenGL::SyncViewport() { } void RasterizerOpenGL::SyncDepthClamp() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::DepthClampEnabled]) { return; } flags[Dirty::DepthClampEnabled] = false; - oglEnable(GL_DEPTH_CLAMP, maxwell3d.regs.view_volume_clip_control.depth_clamp_disabled == 0); + oglEnable(GL_DEPTH_CLAMP, maxwell3d->regs.view_volume_clip_control.depth_clamp_disabled == 0); } void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::ClipDistances] && !flags[VideoCommon::Dirty::Shaders]) { return; } flags[Dirty::ClipDistances] = false; - clip_mask &= maxwell3d.regs.clip_distance_enabled; + clip_mask &= maxwell3d->regs.clip_distance_enabled; if (clip_mask == last_clip_distance_mask) { return; } @@ -684,8 +683,8 @@ void RasterizerOpenGL::SyncClipCoef() { } void RasterizerOpenGL::SyncCullMode() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::CullTest]) { flags[Dirty::CullTest] = false; @@ -700,23 +699,23 @@ void RasterizerOpenGL::SyncCullMode() { } void RasterizerOpenGL::SyncPrimitiveRestart() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PrimitiveRestart]) { return; } flags[Dirty::PrimitiveRestart] = false; - if (maxwell3d.regs.primitive_restart.enabled) { + if (maxwell3d->regs.primitive_restart.enabled) { glEnable(GL_PRIMITIVE_RESTART); - glPrimitiveRestartIndex(maxwell3d.regs.primitive_restart.index); + glPrimitiveRestartIndex(maxwell3d->regs.primitive_restart.index); } else { glDisable(GL_PRIMITIVE_RESTART); } } void RasterizerOpenGL::SyncDepthTestState() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::DepthMask]) { flags[Dirty::DepthMask] = false; @@ -735,13 +734,13 @@ void RasterizerOpenGL::SyncDepthTestState() { } void RasterizerOpenGL::SyncStencilTestState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::StencilTest]) { return; } flags[Dirty::StencilTest] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_STENCIL_TEST, regs.stencil_enable); glStencilFuncSeparate(GL_FRONT, MaxwellToGL::ComparisonOp(regs.stencil_front_func_func), @@ -766,23 +765,23 @@ void RasterizerOpenGL::SyncStencilTestState() { } void RasterizerOpenGL::SyncRasterizeEnable() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::RasterizeEnable]) { return; } flags[Dirty::RasterizeEnable] = false; - oglEnable(GL_RASTERIZER_DISCARD, maxwell3d.regs.rasterize_enable == 0); + oglEnable(GL_RASTERIZER_DISCARD, maxwell3d->regs.rasterize_enable == 0); } void RasterizerOpenGL::SyncPolygonModes() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PolygonModes]) { return; } flags[Dirty::PolygonModes] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.fill_rectangle) { if (!GLAD_GL_NV_fill_rectangle) { LOG_ERROR(Render_OpenGL, "GL_NV_fill_rectangle used and not supported"); @@ -815,7 +814,7 @@ void RasterizerOpenGL::SyncPolygonModes() { } void RasterizerOpenGL::SyncColorMask() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::ColorMasks]) { return; } @@ -824,7 +823,7 @@ void RasterizerOpenGL::SyncColorMask() { const bool force = flags[Dirty::ColorMaskCommon]; flags[Dirty::ColorMaskCommon] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.color_mask_common) { if (!force && !flags[Dirty::ColorMask0]) { return; @@ -849,30 +848,30 @@ void RasterizerOpenGL::SyncColorMask() { } void RasterizerOpenGL::SyncMultiSampleState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::MultisampleControl]) { return; } flags[Dirty::MultisampleControl] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_SAMPLE_ALPHA_TO_COVERAGE, regs.multisample_control.alpha_to_coverage); oglEnable(GL_SAMPLE_ALPHA_TO_ONE, regs.multisample_control.alpha_to_one); } void RasterizerOpenGL::SyncFragmentColorClampState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::FragmentClampColor]) { return; } flags[Dirty::FragmentClampColor] = false; - glClampColor(GL_CLAMP_FRAGMENT_COLOR, maxwell3d.regs.frag_color_clamp ? GL_TRUE : GL_FALSE); + glClampColor(GL_CLAMP_FRAGMENT_COLOR, maxwell3d->regs.frag_color_clamp ? GL_TRUE : GL_FALSE); } void RasterizerOpenGL::SyncBlendState() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::BlendColor]) { flags[Dirty::BlendColor] = false; @@ -929,13 +928,13 @@ void RasterizerOpenGL::SyncBlendState() { } void RasterizerOpenGL::SyncLogicOpState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::LogicOp]) { return; } flags[Dirty::LogicOp] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.logic_op.enable) { glEnable(GL_COLOR_LOGIC_OP); glLogicOp(MaxwellToGL::LogicOp(regs.logic_op.operation)); @@ -945,7 +944,7 @@ void RasterizerOpenGL::SyncLogicOpState() { } void RasterizerOpenGL::SyncScissorTest() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::Scissors] && !flags[VideoCommon::Dirty::RescaleScissors]) { return; } @@ -954,7 +953,7 @@ void RasterizerOpenGL::SyncScissorTest() { const bool force = flags[VideoCommon::Dirty::RescaleScissors]; flags[VideoCommon::Dirty::RescaleScissors] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; const auto& resolution = Settings::values.resolution_info; const bool is_rescaling{texture_cache.IsRescaling()}; @@ -990,39 +989,39 @@ void RasterizerOpenGL::SyncScissorTest() { } void RasterizerOpenGL::SyncPointState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PointSize]) { return; } flags[Dirty::PointSize] = false; - oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); - oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable); + oglEnable(GL_POINT_SPRITE, maxwell3d->regs.point_sprite_enable); + oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d->regs.vp_point_size.enable); const bool is_rescaling{texture_cache.IsRescaling()}; const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; - glPointSize(std::max(1.0f, maxwell3d.regs.point_size * scale)); + glPointSize(std::max(1.0f, maxwell3d->regs.point_size * scale)); } void RasterizerOpenGL::SyncLineState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::LineWidth]) { return; } flags[Dirty::LineWidth] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_LINE_SMOOTH, regs.line_smooth_enable); glLineWidth(regs.line_smooth_enable ? regs.line_width_smooth : regs.line_width_aliased); } void RasterizerOpenGL::SyncPolygonOffset() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PolygonOffset]) { return; } flags[Dirty::PolygonOffset] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_POLYGON_OFFSET_FILL, regs.polygon_offset_fill_enable); oglEnable(GL_POLYGON_OFFSET_LINE, regs.polygon_offset_line_enable); oglEnable(GL_POLYGON_OFFSET_POINT, regs.polygon_offset_point_enable); @@ -1036,13 +1035,13 @@ void RasterizerOpenGL::SyncPolygonOffset() { } void RasterizerOpenGL::SyncAlphaTest() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::AlphaTest]) { return; } flags[Dirty::AlphaTest] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.alpha_test_enabled) { glEnable(GL_ALPHA_TEST); glAlphaFunc(MaxwellToGL::ComparisonOp(regs.alpha_test_func), regs.alpha_test_ref); @@ -1052,17 +1051,17 @@ void RasterizerOpenGL::SyncAlphaTest() { } void RasterizerOpenGL::SyncFramebufferSRGB() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::FramebufferSRGB]) { return; } flags[Dirty::FramebufferSRGB] = false; - oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d.regs.framebuffer_srgb); + oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d->regs.framebuffer_srgb); } void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum primitive_mode) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -1081,11 +1080,48 @@ void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum } void RasterizerOpenGL::EndTransformFeedback() { - if (maxwell3d.regs.tfb_enabled != 0) { + if (maxwell3d->regs.tfb_enabled != 0) { glEndTransformFeedback(); } } +void RasterizerOpenGL::InitializeChannel(Tegra::Control::ChannelState& channel) { + CreateChannel(channel); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CreateChannel(channel); + buffer_cache.CreateChannel(channel); + } + shader_cache.CreateChannel(channel); + query_cache.CreateChannel(channel); + state_tracker.SetupTables(channel); +} + +void RasterizerOpenGL::BindChannel(Tegra::Control::ChannelState& channel) { + const s32 channel_id = channel.bind_id; + BindToChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.BindToChannel(channel_id); + buffer_cache.BindToChannel(channel_id); + } + shader_cache.BindToChannel(channel_id); + query_cache.BindToChannel(channel_id); + state_tracker.ChangeChannel(channel); + state_tracker.InvalidateState(); +} + +void RasterizerOpenGL::ReleaseChannel(s32 channel_id) { + EraseChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.EraseChannel(channel_id); + buffer_cache.EraseChannel(channel_id); + } + shader_cache.EraseChannel(channel_id); + query_cache.EraseChannel(channel_id); +} + AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 98f6fd342..45131b785 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -1,24 +1,18 @@ -// Copyright 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2015 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> -#include <atomic> #include <cstddef> -#include <memory> #include <optional> -#include <tuple> -#include <utility> #include <boost/container/static_vector.hpp> #include <glad/glad.h> #include "common/common_types.h" -#include "video_core/engines/const_buffer_info.h" -#include "video_core/engines/maxwell_3d.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/engines/maxwell_dma.h" #include "video_core/rasterizer_accelerated.h" #include "video_core/rasterizer_interface.h" @@ -26,12 +20,8 @@ #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_fence_manager.h" #include "video_core/renderer_opengl/gl_query_cache.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_cache.h" -#include "video_core/renderer_opengl/gl_shader_manager.h" -#include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_texture_cache.h" -#include "video_core/textures/texture.h" namespace Core::Memory { class Memory; @@ -69,7 +59,8 @@ private: BufferCache& buffer_cache; }; -class RasterizerOpenGL : public VideoCore::RasterizerAccelerated { +class RasterizerOpenGL : public VideoCore::RasterizerAccelerated, + protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, Core::Memory::Memory& cpu_memory_, const Device& device_, @@ -89,10 +80,11 @@ public: bool MustFlushRegion(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override; - void SyncGuestHost() override; + void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; - void ModifyGPUMemory(GPUVAddr addr, u64 size) override; - void SignalSemaphore(GPUVAddr addr, u32 value) override; + void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; + void SignalFence(std::function<void()>&& func) override; + void SyncOperation(std::function<void()>&& func) override; void SignalSyncPoint(u32 value) override; void SignalReference() override; void ReleaseFences() override; @@ -107,7 +99,7 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) override; + std::span<const u8> memory) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; void LoadDiskResources(u64 title_id, std::stop_token stop_loading, @@ -118,6 +110,12 @@ public: return num_queued_commands > 0; } + void InitializeChannel(Tegra::Control::ChannelState& channel) override; + + void BindChannel(Tegra::Control::ChannelState& channel) override; + + void ReleaseChannel(s32 channel_id) override; + private: static constexpr size_t MAX_TEXTURES = 192; static constexpr size_t MAX_IMAGES = 48; @@ -202,9 +200,6 @@ private: void EndTransformFeedback(); Tegra::GPU& gpu; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; const Device& device; ScreenInfo& screen_info; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index 5e7101d28..3a664fdec 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -1,11 +1,8 @@ -// Copyright 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2015 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <string_view> -#include <utility> #include <glad/glad.h> -#include "common/common_types.h" #include "common/microprofile.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_util.h" diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index 84e07f8bd..bc05ba4bd 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -1,6 +1,5 @@ -// Copyright 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2015 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index f71e01a34..5a29a41d2 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <atomic> #include <fstream> @@ -14,10 +13,8 @@ #include "common/fs/fs.h" #include "common/fs/path_util.h" #include "common/logging/log.h" -#include "common/scope_exit.h" #include "common/settings.h" #include "common/thread_worker.h" -#include "core/core.h" #include "shader_recompiler/backend/glasm/emit_glasm.h" #include "shader_recompiler/backend/glsl/emit_glsl.h" #include "shader_recompiler/backend/spirv/emit_spirv.h" @@ -29,7 +26,6 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/renderer_opengl/gl_rasterizer.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_cache.h" #include "video_core/renderer_opengl/gl_shader_util.h" #include "video_core/renderer_opengl/gl_state_tracker.h" @@ -53,7 +49,7 @@ using VideoCommon::LoadPipelines; using VideoCommon::SerializePipeline; using Context = ShaderContext::Context; -constexpr u32 CACHE_VERSION = 5; +constexpr u32 CACHE_VERSION = 6; template <typename Container> auto MakeSpan(Container& container) { @@ -89,7 +85,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key, case Maxwell::TessellationPrimitive::Quads: return Shader::TessPrimitive::Quads; } - UNREACHABLE(); + ASSERT(false); return Shader::TessPrimitive::Triangles; }(); info.tess_spacing = [&] { @@ -101,7 +97,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key, case Maxwell::TessellationSpacing::FractionalEven: return Shader::TessSpacing::FractionalEven; } - UNREACHABLE(); + ASSERT(false); return Shader::TessSpacing::Equal; }(); break; @@ -155,16 +151,13 @@ void SetXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs } // Anonymous namespace ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - TextureCache& texture_cache_, BufferCache& buffer_cache_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - VideoCore::ShaderNotify& shader_notify_) - : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, - emu_window{emu_window_}, device{device_}, texture_cache{texture_cache_}, - buffer_cache{buffer_cache_}, program_manager{program_manager_}, state_tracker{state_tracker_}, - shader_notify{shader_notify_}, use_asynchronous_shaders{device.UseAsynchronousShaders()}, + const Device& device_, TextureCache& texture_cache_, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, VideoCore::ShaderNotify& shader_notify_) + : VideoCommon::ShaderCache{rasterizer_}, emu_window{emu_window_}, device{device_}, + texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_}, + state_tracker{state_tracker_}, shader_notify{shader_notify_}, + use_asynchronous_shaders{device.UseAsynchronousShaders()}, profile{ .supported_spirv = 0x00010000, @@ -261,7 +254,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, [this, key, env = std::move(env), &state, &callback](Context* ctx) mutable { ctx->pools.ReleaseContents(); auto pipeline{CreateComputePipeline(ctx->pools, key, env)}; - std::lock_guard lock{state.mutex}; + std::scoped_lock lock{state.mutex}; if (pipeline) { compute_cache.emplace(key, std::move(pipeline)); } @@ -283,7 +276,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, } ctx->pools.ReleaseContents(); auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false)}; - std::lock_guard lock{state.mutex}; + std::scoped_lock lock{state.mutex}; if (pipeline) { graphics_cache.emplace(key, std::move(pipeline)); } @@ -303,7 +296,7 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, state.has_loaded = true; lock.unlock(); - workers->WaitForRequests(); + workers->WaitForRequests(stop_loading); if (!use_asynchronous_shaders) { workers.reset(); } @@ -314,7 +307,7 @@ GraphicsPipeline* ShaderCache::CurrentGraphicsPipeline() { current_pipeline = nullptr; return nullptr; } - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; graphics_key.raw = 0; graphics_key.early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0); graphics_key.gs_input_topology.Assign(graphics_key.unique_hashes[4] != 0 @@ -355,13 +348,13 @@ GraphicsPipeline* ShaderCache::BuiltPipeline(GraphicsPipeline* pipeline) const n } // If something is using depth, we can assume that games are not rendering anything which // will be used one time. - if (maxwell3d.regs.zeta_enable) { + if (maxwell3d->regs.zeta_enable) { return nullptr; } // If games are using a small index count, we can assume these are full screen quads. // Usually these shaders are only used once for building textures so we can assume they // can't be built async - if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + if (maxwell3d->regs.index_array.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) { return pipeline; } return nullptr; @@ -372,7 +365,7 @@ ComputePipeline* ShaderCache::CurrentComputePipeline() { if (!shader) { return nullptr; } - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const ComputePipelineKey key{ .unique_hash = shader->unique_hash, .shared_memory_size = qmd.shared_alloc, @@ -484,9 +477,9 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( previous_program = &program; } auto* const thread_worker{build_in_parallel ? workers.get() : nullptr}; - return std::make_unique<GraphicsPipeline>( - device, texture_cache, buffer_cache, gpu_memory, maxwell3d, program_manager, state_tracker, - thread_worker, &shader_notify, sources, sources_spirv, infos, key); + return std::make_unique<GraphicsPipeline>(device, texture_cache, buffer_cache, program_manager, + state_tracker, thread_worker, &shader_notify, sources, + sources_spirv, infos, key); } catch (Shader::Exception& exception) { LOG_ERROR(Render_OpenGL, "{}", exception.what()); @@ -495,9 +488,9 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( const ComputePipelineKey& key, const VideoCommon::ShaderInfo* shader) { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; env.SetCachedSize(shader->size_bytes); main_pools.ReleaseContents(); @@ -540,9 +533,8 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( break; } - return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, gpu_memory, - kepler_compute, program_manager, program.info, code, - code_spirv); + return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, program_manager, + program.info, code, code_spirv); } catch (Shader::Exception& exception) { LOG_ERROR(Render_OpenGL, "{}", exception.what()); return nullptr; @@ -550,7 +542,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( std::unique_ptr<ShaderWorker> ShaderCache::CreateWorkers() const { return std::make_unique<ShaderWorker>(std::max(std::thread::hardware_concurrency(), 2U) - 1, - "yuzu:ShaderBuilder", + "GlShaderBuilder", [this] { return Context{emu_window}; }); } diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index a34110b37..89f181fe3 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -1,21 +1,15 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <array> #include <filesystem> #include <stop_token> #include <unordered_map> -#include <glad/glad.h> - #include "common/common_types.h" #include "common/thread_worker.h" -#include "shader_recompiler/frontend/ir/value.h" #include "shader_recompiler/host_translate_info.h" -#include "shader_recompiler/object_pool.h" #include "shader_recompiler/profile.h" #include "video_core/renderer_opengl/gl_compute_pipeline.h" #include "video_core/renderer_opengl/gl_graphics_pipeline.h" @@ -36,12 +30,9 @@ using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>; class ShaderCache : public VideoCommon::ShaderCache { public: explicit ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - TextureCache& texture_cache_, BufferCache& buffer_cache_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - VideoCore::ShaderNotify& shader_notify_); + const Device& device_, TextureCache& texture_cache_, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, VideoCore::ShaderNotify& shader_notify_); ~ShaderCache(); void LoadDiskResources(u64 title_id, std::stop_token stop_loading, diff --git a/src/video_core/renderer_opengl/gl_shader_context.h b/src/video_core/renderer_opengl/gl_shader_context.h index 6ff34e5d6..ca2bd8e8e 100644 --- a/src/video_core/renderer_opengl/gl_shader_context.h +++ b/src/video_core/renderer_opengl/gl_shader_context.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp index 399959afb..d9c29d8b7 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -1,3 +1,2 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index d7ef0775d..a84f5aeb3 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_opengl/gl_shader_util.cpp b/src/video_core/renderer_opengl/gl_shader_util.cpp index d432072ad..a0d9d10ef 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.cpp +++ b/src/video_core/renderer_opengl/gl_shader_util.cpp @@ -1,12 +1,10 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <string_view> #include <vector> #include <glad/glad.h> -#include "common/assert.h" #include "common/logging/log.h" #include "common/settings.h" #include "video_core/renderer_opengl/gl_shader_util.h" @@ -19,6 +17,7 @@ static OGLProgram LinkSeparableProgram(GLuint shader) { glProgramParameteri(program.handle, GL_PROGRAM_SEPARABLE, GL_TRUE); glAttachShader(program.handle, shader); glLinkProgram(program.handle); + glDetachShader(program.handle, shader); if (!Settings::values.renderer_debug) { return program; } diff --git a/src/video_core/renderer_opengl/gl_shader_util.h b/src/video_core/renderer_opengl/gl_shader_util.h index 4e1a2a8e1..43ebcdeba 100644 --- a/src/video_core/renderer_opengl/gl_shader_util.h +++ b/src/video_core/renderer_opengl/gl_shader_util.h @@ -1,18 +1,13 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <span> -#include <string> #include <string_view> -#include <vector> #include <glad/glad.h> -#include "common/assert.h" -#include "common/logging/log.h" #include "video_core/renderer_opengl/gl_resource_manager.h" namespace OpenGL { diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index 586da84e3..a8f3a0f57 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -8,8 +7,8 @@ #include "common/common_types.h" #include "core/core.h" +#include "video_core/control/channel_state.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" #include "video_core/renderer_opengl/gl_state_tracker.h" #define OFF(field_name) MAXWELL3D_REG_INDEX(field_name) @@ -203,9 +202,8 @@ void SetupDirtyMisc(Tables& tables) { } // Anonymous namespace -StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} { - auto& dirty = gpu.Maxwell3D().dirty; - auto& tables = dirty.tables; +void StateTracker::SetupTables(Tegra::Control::ChannelState& channel_state) { + auto& tables{channel_state.maxwell_3d->dirty.tables}; SetupDirtyFlags(tables); SetupDirtyColorMasks(tables); SetupDirtyViewports(tables); @@ -231,4 +229,14 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} SetupDirtyMisc(tables); } +void StateTracker::ChangeChannel(Tegra::Control::ChannelState& channel_state) { + flags = &channel_state.maxwell_3d->dirty.flags; +} + +void StateTracker::InvalidateState() { + flags->set(); +} + +StateTracker::StateTracker() : flags{&default_flags} {} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h index 5864c7c07..19bcf3f35 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,13 +8,14 @@ #include <glad/glad.h> #include "common/common_types.h" -#include "core/core.h" #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" namespace Tegra { -class GPU; +namespace Control { +struct ChannelState; } +} // namespace Tegra namespace OpenGL { @@ -85,7 +85,7 @@ static_assert(Last <= std::numeric_limits<u8>::max()); class StateTracker { public: - explicit StateTracker(Tegra::GPU& gpu); + explicit StateTracker(); void BindIndexBuffer(GLuint new_index_buffer) { if (index_buffer == new_index_buffer) { @@ -123,94 +123,107 @@ public: } void NotifyScreenDrawVertexArray() { - flags[OpenGL::Dirty::VertexFormats] = true; - flags[OpenGL::Dirty::VertexFormat0 + 0] = true; - flags[OpenGL::Dirty::VertexFormat0 + 1] = true; + (*flags)[OpenGL::Dirty::VertexFormats] = true; + (*flags)[OpenGL::Dirty::VertexFormat0 + 0] = true; + (*flags)[OpenGL::Dirty::VertexFormat0 + 1] = true; - flags[VideoCommon::Dirty::VertexBuffers] = true; - flags[VideoCommon::Dirty::VertexBuffer0] = true; + (*flags)[VideoCommon::Dirty::VertexBuffers] = true; + (*flags)[VideoCommon::Dirty::VertexBuffer0] = true; - flags[OpenGL::Dirty::VertexInstances] = true; - flags[OpenGL::Dirty::VertexInstance0 + 0] = true; - flags[OpenGL::Dirty::VertexInstance0 + 1] = true; + (*flags)[OpenGL::Dirty::VertexInstances] = true; + (*flags)[OpenGL::Dirty::VertexInstance0 + 0] = true; + (*flags)[OpenGL::Dirty::VertexInstance0 + 1] = true; } void NotifyPolygonModes() { - flags[OpenGL::Dirty::PolygonModes] = true; - flags[OpenGL::Dirty::PolygonModeFront] = true; - flags[OpenGL::Dirty::PolygonModeBack] = true; + (*flags)[OpenGL::Dirty::PolygonModes] = true; + (*flags)[OpenGL::Dirty::PolygonModeFront] = true; + (*flags)[OpenGL::Dirty::PolygonModeBack] = true; } void NotifyViewport0() { - flags[OpenGL::Dirty::Viewports] = true; - flags[OpenGL::Dirty::Viewport0] = true; + (*flags)[OpenGL::Dirty::Viewports] = true; + (*flags)[OpenGL::Dirty::Viewport0] = true; } void NotifyScissor0() { - flags[OpenGL::Dirty::Scissors] = true; - flags[OpenGL::Dirty::Scissor0] = true; + (*flags)[OpenGL::Dirty::Scissors] = true; + (*flags)[OpenGL::Dirty::Scissor0] = true; } void NotifyColorMask(size_t index) { - flags[OpenGL::Dirty::ColorMasks] = true; - flags[OpenGL::Dirty::ColorMask0 + index] = true; + (*flags)[OpenGL::Dirty::ColorMasks] = true; + (*flags)[OpenGL::Dirty::ColorMask0 + index] = true; } void NotifyBlend0() { - flags[OpenGL::Dirty::BlendStates] = true; - flags[OpenGL::Dirty::BlendState0] = true; + (*flags)[OpenGL::Dirty::BlendStates] = true; + (*flags)[OpenGL::Dirty::BlendState0] = true; } void NotifyFramebuffer() { - flags[VideoCommon::Dirty::RenderTargets] = true; + (*flags)[VideoCommon::Dirty::RenderTargets] = true; } void NotifyFrontFace() { - flags[OpenGL::Dirty::FrontFace] = true; + (*flags)[OpenGL::Dirty::FrontFace] = true; } void NotifyCullTest() { - flags[OpenGL::Dirty::CullTest] = true; + (*flags)[OpenGL::Dirty::CullTest] = true; } void NotifyDepthMask() { - flags[OpenGL::Dirty::DepthMask] = true; + (*flags)[OpenGL::Dirty::DepthMask] = true; } void NotifyDepthTest() { - flags[OpenGL::Dirty::DepthTest] = true; + (*flags)[OpenGL::Dirty::DepthTest] = true; } void NotifyStencilTest() { - flags[OpenGL::Dirty::StencilTest] = true; + (*flags)[OpenGL::Dirty::StencilTest] = true; } void NotifyPolygonOffset() { - flags[OpenGL::Dirty::PolygonOffset] = true; + (*flags)[OpenGL::Dirty::PolygonOffset] = true; } void NotifyRasterizeEnable() { - flags[OpenGL::Dirty::RasterizeEnable] = true; + (*flags)[OpenGL::Dirty::RasterizeEnable] = true; } void NotifyFramebufferSRGB() { - flags[OpenGL::Dirty::FramebufferSRGB] = true; + (*flags)[OpenGL::Dirty::FramebufferSRGB] = true; } void NotifyLogicOp() { - flags[OpenGL::Dirty::LogicOp] = true; + (*flags)[OpenGL::Dirty::LogicOp] = true; } void NotifyClipControl() { - flags[OpenGL::Dirty::ClipControl] = true; + (*flags)[OpenGL::Dirty::ClipControl] = true; } void NotifyAlphaTest() { - flags[OpenGL::Dirty::AlphaTest] = true; + (*flags)[OpenGL::Dirty::AlphaTest] = true; } + void NotifyRange(u8 start, u8 end) { + for (auto flag = start; flag <= end; flag++) { + (*flags)[flag] = true; + } + } + + void SetupTables(Tegra::Control::ChannelState& channel_state); + + void ChangeChannel(Tegra::Control::ChannelState& channel_state); + + void InvalidateState(); + private: - Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags* flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags default_flags{}; GLuint framebuffer = 0; GLuint index_buffer = 0; diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp index 77b3ee0fe..2005c8993 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <memory> diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h index 2e67922a6..8fe927aaf 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.h +++ b/src/video_core/renderer_opengl/gl_stream_buffer.h @@ -1,11 +1,9 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> -#include <memory> #include <span> #include <utility> diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 3c1f79a27..99cd11d1e 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -84,7 +83,7 @@ GLenum ImageTarget(const VideoCommon::ImageInfo& info) { case ImageType::Buffer: return GL_TEXTURE_BUFFER; } - UNREACHABLE_MSG("Invalid image type={}", info.type); + ASSERT_MSG(false, "Invalid image type={}", info.type); return GL_NONE; } @@ -94,6 +93,7 @@ GLenum ImageTarget(Shader::TextureType type, int num_samples = 1) { case Shader::TextureType::Color1D: return GL_TEXTURE_1D; case Shader::TextureType::Color2D: + case Shader::TextureType::Color2DRect: return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE : GL_TEXTURE_2D; case Shader::TextureType::ColorCube: return GL_TEXTURE_CUBE_MAP; @@ -108,7 +108,7 @@ GLenum ImageTarget(Shader::TextureType type, int num_samples = 1) { case Shader::TextureType::Buffer: return GL_TEXTURE_BUFFER; } - UNREACHABLE_MSG("Invalid image view type={}", type); + ASSERT_MSG(false, "Invalid image view type={}", type); return GL_NONE; } @@ -120,7 +120,7 @@ GLenum TextureMode(PixelFormat format, bool is_first) { case PixelFormat::S8_UINT_D24_UNORM: return is_first ? GL_STENCIL_INDEX : GL_DEPTH_COMPONENT; default: - UNREACHABLE(); + ASSERT(false); return GL_DEPTH_COMPONENT; } } @@ -141,7 +141,7 @@ GLint Swizzle(SwizzleSource source) { case SwizzleSource::OneFloat: return GL_ONE; } - UNREACHABLE_MSG("Invalid swizzle source={}", source); + ASSERT_MSG(false, "Invalid swizzle source={}", source); return GL_NONE; } @@ -182,6 +182,26 @@ GLenum AttachmentType(PixelFormat format) { } } +GLint ConvertA5B5G5R1_UNORM(SwizzleSource source) { + switch (source) { + case SwizzleSource::Zero: + return GL_ZERO; + case SwizzleSource::R: + return GL_ALPHA; + case SwizzleSource::G: + return GL_BLUE; + case SwizzleSource::B: + return GL_GREEN; + case SwizzleSource::A: + return GL_RED; + case SwizzleSource::OneInt: + case SwizzleSource::OneFloat: + return GL_ONE; + } + ASSERT_MSG(false, "Invalid swizzle source={}", source); + return GL_NONE; +} + void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4> swizzle) { switch (format) { case PixelFormat::D24_UNORM_S8_UINT: @@ -192,6 +212,12 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4 TextureMode(format, swizzle[0] == SwizzleSource::R)); std::ranges::transform(swizzle, swizzle.begin(), ConvertGreenRed); break; + case PixelFormat::A5B5G5R1_UNORM: { + std::array<GLint, 4> gl_swizzle; + std::ranges::transform(swizzle, gl_swizzle.begin(), ConvertA5B5G5R1_UNORM); + glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data()); + return; + } default: break; } @@ -356,10 +382,10 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form glTextureStorage3D(handle, gl_num_levels, gl_internal_format, width, height, depth); break; case GL_TEXTURE_BUFFER: - UNREACHABLE(); + ASSERT(false); break; default: - UNREACHABLE_MSG("Invalid target=0x{:x}", target); + ASSERT_MSG(false, "Invalid target=0x{:x}", target); break; } return texture; @@ -395,7 +421,7 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form case Shader::ImageFormat::R32G32B32A32_UINT: return GL_RGBA32UI; } - UNREACHABLE_MSG("Invalid image format={}", format); + ASSERT_MSG(false, "Invalid image format={}", format); return GL_R32UI; } @@ -409,8 +435,8 @@ ImageBufferMap::~ImageBufferMap() { TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& program_manager, StateTracker& state_tracker_) - : device{device_}, state_tracker{state_tracker_}, - util_shaders(program_manager), resolution{Settings::values.resolution_info} { + : device{device_}, state_tracker{state_tracker_}, util_shaders(program_manager), + format_conversion_pass{util_shaders}, resolution{Settings::values.resolution_info} { static constexpr std::array TARGETS{GL_TEXTURE_1D_ARRAY, GL_TEXTURE_2D_ARRAY, GL_TEXTURE_3D}; for (size_t i = 0; i < TARGETS.size(); ++i) { const GLenum target = TARGETS[i]; @@ -477,6 +503,7 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& set_view(Shader::TextureType::ColorArray1D, null_image_1d_array.handle); set_view(Shader::TextureType::ColorArray2D, null_image_view_2d_array.handle); set_view(Shader::TextureType::ColorArrayCube, null_image_cube_array.handle); + set_view(Shader::TextureType::Color2DRect, null_image_view_2d.handle); if (resolution.active) { for (size_t i = 0; i < rescale_draw_fbos.size(); ++i) { @@ -484,6 +511,13 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& rescale_read_fbos[i].Create(); } } + + device_access_memory = [this]() -> u64 { + if (device.CanReportMemoryUsage()) { + return device.GetCurrentDedicatedVideoMemory() + 512_MiB; + } + return 2_GiB; // Return minimum requirements + }(); } TextureCacheRuntime::~TextureCacheRuntime() = default; @@ -500,13 +534,11 @@ ImageBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) { return download_buffers.RequestMap(size, false); } -u64 TextureCacheRuntime::GetDeviceLocalMemory() const { - if (GLAD_GL_NVX_gpu_memory_info) { - GLint cur_avail_mem_kb = 0; - glGetIntegerv(GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &cur_avail_mem_kb); - return static_cast<u64>(cur_avail_mem_kb) * 1_KiB; +u64 TextureCacheRuntime::GetDeviceMemoryUsage() const { + if (device.CanReportMemoryUsage()) { + return device_access_memory - device.GetCurrentDedicatedVideoMemory(); } - return 2_GiB; // Return minimum requirements + return 2_GiB; } void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image, @@ -549,7 +581,7 @@ void TextureCacheRuntime::EmulateCopyImage(Image& dst, Image& src, } else if (IsPixelFormatBGR(dst.info.format) || IsPixelFormatBGR(src.info.format)) { format_conversion_pass.ConvertImage(dst, src, copies); } else { - UNREACHABLE(); + ASSERT(false); } } @@ -590,7 +622,7 @@ void TextureCacheRuntime::AccelerateImageUpload(Image& image, const ImageBufferM case ImageType::Linear: return util_shaders.PitchUpload(image, map, swizzles); default: - UNREACHABLE(); + ASSERT(false); break; } } @@ -609,7 +641,7 @@ FormatProperties TextureCacheRuntime::FormatInfo(ImageType type, GLenum internal case ImageType::e3D: return format_properties[2].at(internal_format); default: - UNREACHABLE(); + ASSERT(false); return FormatProperties{}; } } @@ -686,6 +718,7 @@ Image::Image(TextureCacheRuntime& runtime_, const VideoCommon::ImageInfo& info_, } if (IsConverted(runtime->device, info.format, info.type)) { flags |= ImageFlagBits::Converted; + flags |= ImageFlagBits::CostlyLoad; gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; gl_format = GL_RGBA; gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; @@ -857,7 +890,7 @@ void Image::CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t b } break; default: - UNREACHABLE(); + ASSERT(false); } } @@ -893,7 +926,7 @@ void Image::CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t b depth = copy.image_extent.depth; break; default: - UNREACHABLE(); + ASSERT(false); } // Compressed formats don't have a pixel format or type const bool is_compressed = gl_format == GL_NONE; @@ -919,7 +952,7 @@ void Image::Scale(bool up_scale) { case SurfaceType::DepthStencil: return GL_DEPTH_STENCIL_ATTACHMENT; default: - UNREACHABLE(); + ASSERT(false); return GL_COLOR_ATTACHMENT0; } }(); @@ -934,7 +967,7 @@ void Image::Scale(bool up_scale) { case SurfaceType::DepthStencil: return GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; default: - UNREACHABLE(); + ASSERT(false); return GL_COLOR_BUFFER_BIT; } }(); @@ -949,7 +982,7 @@ void Image::Scale(bool up_scale) { case SurfaceType::DepthStencil: return 3; default: - UNREACHABLE(); + ASSERT(false); return 0; } }(); @@ -1014,7 +1047,7 @@ bool Image::ScaleUp(bool ignore) { return false; } if (info.type == ImageType::Linear) { - UNREACHABLE(); + ASSERT(false); return false; } flags |= ImageFlagBits::Rescaled; @@ -1079,6 +1112,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI flat_range.extent.layers = 1; [[fallthrough]]; case ImageViewType::e2D: + case ImageViewType::Rect: if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) { // 2D and 2D array views on a 3D textures are used exclusively for render targets ASSERT(info.range.extent.levels == 1); @@ -1104,11 +1138,8 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI SetupView(Shader::TextureType::ColorCube); SetupView(Shader::TextureType::ColorArrayCube); break; - case ImageViewType::Rect: - UNIMPLEMENTED(); - break; case ImageViewType::Buffer: - UNREACHABLE(); + ASSERT(false); break; } switch (info.type) { @@ -1119,6 +1150,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI default_handle = Handle(Shader::TextureType::ColorArray1D); break; case ImageViewType::e2D: + case ImageViewType::Rect: default_handle = Handle(Shader::TextureType::Color2D); break; case ImageViewType::e2DArray: @@ -1179,6 +1211,7 @@ GLuint ImageView::MakeView(Shader::TextureType view_type, GLenum view_format) { case Shader::TextureType::Color1D: case Shader::TextureType::Color2D: case Shader::TextureType::ColorCube: + case Shader::TextureType::Color2DRect: view_range = flat_range; break; case Shader::TextureType::ColorArray1D: @@ -1219,7 +1252,6 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) { const GLint seamless = config.cubemap_interface_filtering ? GL_TRUE : GL_FALSE; UNIMPLEMENTED_IF(config.cubemap_anisotropy != 1); - UNIMPLEMENTED_IF(config.float_coord_normalization != 0); sampler.Create(); const GLuint handle = sampler.handle; @@ -1288,7 +1320,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM buffer_bits |= GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; break; default: - UNREACHABLE(); + ASSERT(false); buffer_bits |= GL_DEPTH_BUFFER_BIT; break; } @@ -1319,6 +1351,9 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM Framebuffer::~Framebuffer() = default; +FormatConversionPass::FormatConversionPass(UtilShaders& util_shaders_) + : util_shaders{util_shaders_} {} + void FormatConversionPass::ConvertImage(Image& dst_image, Image& src_image, std::span<const VideoCommon::ImageCopy> copies) { const GLenum dst_target = ImageTarget(dst_image.info); @@ -1351,6 +1386,12 @@ void FormatConversionPass::ConvertImage(Image& dst_image, Image& src_image, dst_origin.z, region.width, region.height, region.depth, dst_image.GlFormat(), dst_image.GlType(), nullptr); } + + // Swap component order of S8D24 to ABGR8 reinterprets + if (src_image.info.format == PixelFormat::D24_UNORM_S8_UINT && + dst_image.info.format == PixelFormat::A8B8G8R8_UNORM) { + util_shaders.ConvertS8D24(dst_image, copies); + } } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 7f425631f..113528e9b 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -10,6 +9,7 @@ #include <glad/glad.h> #include "shader_recompiler/shader_info.h" +#include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/util_shaders.h" #include "video_core/texture_cache/image_view_base.h" @@ -21,7 +21,6 @@ struct ResolutionScalingInfo; namespace OpenGL { -class Device; class ProgramManager; class StateTracker; @@ -55,13 +54,14 @@ struct FormatProperties { class FormatConversionPass { public: - FormatConversionPass() = default; + explicit FormatConversionPass(UtilShaders& util_shaders); ~FormatConversionPass() = default; void ConvertImage(Image& dst_image, Image& src_image, std::span<const VideoCommon::ImageCopy> copies); private: + UtilShaders& util_shaders; OGLBuffer intermediate_pbo; size_t pbo_size{}; }; @@ -83,7 +83,15 @@ public: ImageBufferMap DownloadStagingBuffer(size_t size); - u64 GetDeviceLocalMemory() const; + u64 GetDeviceLocalMemory() const { + return device_access_memory; + } + + u64 GetDeviceMemoryUsage() const; + + bool CanReportMemoryUsage() const { + return device.CanReportMemoryUsage(); + } bool ShouldReinterpret([[maybe_unused]] Image& dst, [[maybe_unused]] Image& src) { return true; @@ -172,6 +180,7 @@ private: std::array<OGLFramebuffer, 4> rescale_draw_fbos; std::array<OGLFramebuffer, 4> rescale_read_fbos; const Settings::ResolutionScalingInfo& resolution; + u64 device_access_memory; }; class Image : public VideoCommon::ImageBase { diff --git a/src/video_core/renderer_opengl/gl_texture_cache_base.cpp b/src/video_core/renderer_opengl/gl_texture_cache_base.cpp index 385358fea..64416fd51 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache_base.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache_base.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "video_core/renderer_opengl/gl_texture_cache.h" #include "video_core/texture_cache/texture_cache.h" diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index daba42ed9..004421236 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -30,6 +29,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UNORM {GL_RGB10_A2UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT_2_10_10_10_REV}, // A2B10G10R10_UINT {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV}, // A1B5G5R5_UNORM + {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_5_5_5_1}, // A5B5G5R1_UNORM {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R8_UNORM {GL_R8_SNORM, GL_RED, GL_BYTE}, // R8_SNORM {GL_R8I, GL_RED_INTEGER, GL_BYTE}, // R8_SINT @@ -87,6 +87,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM + {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // G4R4_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB @@ -97,6 +98,9 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB + {GL_COMPRESSED_RGBA_ASTC_10x6_KHR}, // ASTC_2D_10X6_UNORM + {GL_COMPRESSED_RGBA_ASTC_10x5_KHR}, // ASTC_2D_10X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x5_KHR}, // ASTC_2D_10X5_SRGB {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM @@ -184,6 +188,8 @@ inline GLenum VertexFormat(Maxwell::VertexAttribute attrib) { case Maxwell::VertexAttribute::Size::Size_32_32_32: case Maxwell::VertexAttribute::Size::Size_32_32_32_32: return GL_FLOAT; + case Maxwell::VertexAttribute::Size::Size_11_11_10: + return GL_UNSIGNED_INT_10F_11F_11F_REV; default: break; } @@ -203,7 +209,7 @@ inline GLenum IndexFormat(Maxwell::IndexFormat index_format) { case Maxwell::IndexFormat::UnsignedInt: return GL_UNSIGNED_INT; } - UNREACHABLE_MSG("Invalid index_format={}", index_format); + ASSERT_MSG(false, "Invalid index_format={}", index_format); return {}; } @@ -240,7 +246,7 @@ inline GLenum PrimitiveTopology(Maxwell::PrimitiveTopology topology) { case Maxwell::PrimitiveTopology::Patches: return GL_PATCHES; } - UNREACHABLE_MSG("Invalid topology={}", topology); + ASSERT_MSG(false, "Invalid topology={}", topology); return GL_POINTS; } @@ -268,8 +274,8 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode, } break; } - UNREACHABLE_MSG("Invalid texture filter mode={} and mipmap filter mode={}", filter_mode, - mipmap_filter_mode); + ASSERT_MSG(false, "Invalid texture filter mode={} and mipmap filter mode={}", filter_mode, + mipmap_filter_mode); return GL_NEAREST; } @@ -547,7 +553,7 @@ inline GLenum PolygonMode(Maxwell::PolygonMode polygon_mode) { case Maxwell::PolygonMode::Fill: return GL_FILL; } - UNREACHABLE_MSG("Invalid polygon mode={}", polygon_mode); + ASSERT_MSG(false, "Invalid polygon mode={}", polygon_mode); return GL_FILL; } @@ -560,7 +566,7 @@ inline GLenum ReductionFilter(Tegra::Texture::SamplerReduction filter) { case Tegra::Texture::SamplerReduction::Max: return GL_MAX; } - UNREACHABLE_MSG("Invalid reduction filter={}", static_cast<int>(filter)); + ASSERT_MSG(false, "Invalid reduction filter={}", static_cast<int>(filter)); return GL_WEIGHTED_AVERAGE_ARB; } diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index f81c1b233..8bd5eba7e 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -1,11 +1,9 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <cstddef> #include <cstdlib> -#include <cstring> #include <memory> #include <glad/glad.h> @@ -15,11 +13,9 @@ #include "common/microprofile.h" #include "common/settings.h" #include "common/telemetry.h" -#include "core/core.h" #include "core/core_timing.h" #include "core/frontend/emu_window.h" #include "core/memory.h" -#include "core/perf_stats.h" #include "core/telemetry_session.h" #include "video_core/host_shaders/fxaa_frag.h" #include "video_core/host_shaders/fxaa_vert.h" @@ -82,7 +78,7 @@ const char* GetSource(GLenum source) { case GL_DEBUG_SOURCE_OTHER: return "OTHER"; default: - UNREACHABLE(); + ASSERT(false); return "Unknown source"; } } @@ -104,7 +100,7 @@ const char* GetType(GLenum type) { case GL_DEBUG_TYPE_MARKER: return "MARKER"; default: - UNREACHABLE(); + ASSERT(false); return "Unknown type"; } } @@ -135,7 +131,7 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, std::unique_ptr<Core::Frontend::GraphicsContext> context_) : RendererBase{emu_window_, std::move(context_)}, telemetry_session{telemetry_session_}, - emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, state_tracker{gpu}, + emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, state_tracker{}, program_manager{device}, rasterizer(emu_window, gpu, cpu_memory, device, screen_info, program_manager, state_tracker) { if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) { @@ -211,6 +207,8 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf // Framebuffer orientation handling framebuffer_transform_flags = framebuffer.transform_flags; framebuffer_crop_rect = framebuffer.crop_rect; + framebuffer_width = framebuffer.width; + framebuffer_height = framebuffer.height; const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset}; screen_info.was_accelerated = @@ -326,12 +324,12 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, GLint internal_format; switch (framebuffer.pixel_format) { - case Tegra::FramebufferConfig::PixelFormat::A8B8G8R8_UNORM: + case Service::android::PixelFormat::Rgba8888: internal_format = GL_RGBA8; texture.gl_format = GL_RGBA; texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; break; - case Tegra::FramebufferConfig::PixelFormat::RGB565_UNORM: + case Service::android::PixelFormat::Rgb565: internal_format = GL_RGB565; texture.gl_format = GL_RGB; texture.gl_type = GL_UNSIGNED_SHORT_5_6_5; @@ -467,8 +465,8 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { const auto& texcoords = screen_info.display_texcoords; auto left = texcoords.left; auto right = texcoords.right; - if (framebuffer_transform_flags != Tegra::FramebufferConfig::TransformFlags::Unset) { - if (framebuffer_transform_flags == Tegra::FramebufferConfig::TransformFlags::FlipV) { + if (framebuffer_transform_flags != Service::android::BufferTransformFlags::Unset) { + if (framebuffer_transform_flags == Service::android::BufferTransformFlags::FlipV) { // Flip the framebuffer vertically left = texcoords.right; right = texcoords.left; @@ -480,12 +478,18 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { } } - ASSERT_MSG(framebuffer_crop_rect.top == 0, "Unimplemented"); ASSERT_MSG(framebuffer_crop_rect.left == 0, "Unimplemented"); + f32 left_start{}; + if (framebuffer_crop_rect.Top() > 0) { + left_start = static_cast<f32>(framebuffer_crop_rect.Top()) / + static_cast<f32>(framebuffer_crop_rect.Bottom()); + } + f32 scale_u = static_cast<f32>(framebuffer_width) / static_cast<f32>(screen_info.texture.width); + f32 scale_v = + static_cast<f32>(framebuffer_height) / static_cast<f32>(screen_info.texture.height); // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering // (e.g. handheld mode) on a 1920x1080 framebuffer. - f32 scale_u = 1.f, scale_v = 1.f; if (framebuffer_crop_rect.GetWidth() > 0) { scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) / static_cast<f32>(screen_info.texture.width); @@ -502,10 +506,14 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { const auto& screen = layout.screen; const std::array vertices = { - ScreenRectVertex(screen.left, screen.top, texcoords.top * scale_u, left * scale_v), - ScreenRectVertex(screen.right, screen.top, texcoords.bottom * scale_u, left * scale_v), - ScreenRectVertex(screen.left, screen.bottom, texcoords.top * scale_u, right * scale_v), - ScreenRectVertex(screen.right, screen.bottom, texcoords.bottom * scale_u, right * scale_v), + ScreenRectVertex(screen.left, screen.top, texcoords.top * scale_u, + left_start + left * scale_v), + ScreenRectVertex(screen.right, screen.top, texcoords.bottom * scale_u, + left_start + left * scale_v), + ScreenRectVertex(screen.left, screen.bottom, texcoords.top * scale_u, + left_start + right * scale_v), + ScreenRectVertex(screen.right, screen.bottom, texcoords.bottom * scale_u, + left_start + right * scale_v), }; glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), std::data(vertices)); diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index cda333cad..1a32e739d 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,10 +7,12 @@ #include <glad/glad.h> #include "common/common_types.h" #include "common/math_util.h" + #include "video_core/renderer_base.h" #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" namespace Core { @@ -44,7 +45,7 @@ struct TextureInfo { GLsizei height; GLenum gl_format; GLenum gl_type; - Tegra::FramebufferConfig::PixelFormat pixel_format; + Service::android::PixelFormat pixel_format; }; /// Structure used for storing information about the display target for the Switch screen @@ -133,8 +134,10 @@ private: std::vector<u8> gl_framebuffer_data; /// Used for transforming the framebuffer orientation - Tegra::FramebufferConfig::TransformFlags framebuffer_transform_flags{}; + Service::android::BufferTransformFlags framebuffer_transform_flags{}; Common::Rectangle<int> framebuffer_crop_rect; + u32 framebuffer_width; + u32 framebuffer_height; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 897c380b3..404def62e 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <span> #include <string_view> @@ -13,6 +12,7 @@ #include "video_core/host_shaders/astc_decoder_comp.h" #include "video_core/host_shaders/block_linear_unswizzle_2d_comp.h" #include "video_core/host_shaders/block_linear_unswizzle_3d_comp.h" +#include "video_core/host_shaders/opengl_convert_s8d24_comp.h" #include "video_core/host_shaders/opengl_copy_bc4_comp.h" #include "video_core/host_shaders/pitch_unswizzle_comp.h" #include "video_core/renderer_opengl/gl_shader_manager.h" @@ -50,7 +50,8 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_) block_linear_unswizzle_2d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_2D_COMP)), block_linear_unswizzle_3d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_3D_COMP)), pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)), - copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) { + copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)), + convert_s8d24_program(MakeProgram(OPENGL_CONVERT_S8D24_COMP)) { const auto swizzle_table = Tegra::Texture::MakeSwizzleTable(); swizzle_table_buffer.Create(); glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0); @@ -248,6 +249,26 @@ void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const Im program_manager.RestoreGuestCompute(); } +void UtilShaders::ConvertS8D24(Image& dst_image, std::span<const ImageCopy> copies) { + static constexpr GLuint BINDING_DESTINATION = 0; + static constexpr GLuint LOC_SIZE = 0; + + program_manager.BindComputeProgram(convert_s8d24_program.handle); + for (const ImageCopy& copy : copies) { + ASSERT(copy.src_subresource.base_layer == 0); + ASSERT(copy.src_subresource.num_layers == 1); + ASSERT(copy.dst_subresource.base_layer == 0); + ASSERT(copy.dst_subresource.num_layers == 1); + + glUniform3ui(LOC_SIZE, copy.extent.width, copy.extent.height, copy.extent.depth); + glBindImageTexture(BINDING_DESTINATION, dst_image.StorageHandle(), + copy.dst_subresource.base_level, GL_TRUE, 0, GL_READ_WRITE, GL_RGBA8UI); + glDispatchCompute(Common::DivCeil(copy.extent.width, 16u), + Common::DivCeil(copy.extent.height, 8u), copy.extent.depth); + } + program_manager.RestoreGuestCompute(); +} + GLenum StoreFormat(u32 bytes_per_block) { switch (bytes_per_block) { case 1: @@ -261,7 +282,7 @@ GLenum StoreFormat(u32 bytes_per_block) { case 16: return GL_RGBA32UI; } - UNREACHABLE(); + ASSERT(false); return GL_R8UI; } diff --git a/src/video_core/renderer_opengl/util_shaders.h b/src/video_core/renderer_opengl/util_shaders.h index 5de95ea7a..44efb6ecf 100644 --- a/src/video_core/renderer_opengl/util_shaders.h +++ b/src/video_core/renderer_opengl/util_shaders.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -39,6 +38,8 @@ public: void CopyBC4(Image& dst_image, Image& src_image, std::span<const VideoCommon::ImageCopy> copies); + void ConvertS8D24(Image& dst_image, std::span<const VideoCommon::ImageCopy> copies); + private: ProgramManager& program_manager; @@ -49,6 +50,7 @@ private: OGLProgram block_linear_unswizzle_3d_program; OGLProgram pitch_unswizzle_program; OGLProgram copy_bc4_program; + OGLProgram convert_s8d24_program; }; GLenum StoreFormat(u32 bytes_per_block); diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index 2c3914459..3f2b139e0 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> @@ -9,6 +8,7 @@ #include "video_core/host_shaders/convert_d24s8_to_abgr8_frag_spv.h" #include "video_core/host_shaders/convert_depth_to_float_frag_spv.h" #include "video_core/host_shaders/convert_float_to_depth_frag_spv.h" +#include "video_core/host_shaders/convert_s8d24_to_abgr8_frag_spv.h" #include "video_core/host_shaders/full_screen_triangle_vert_spv.h" #include "video_core/host_shaders/vulkan_blit_color_float_frag_spv.h" #include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h" @@ -349,7 +349,7 @@ VkExtent2D GetConversionExtent(const ImageView& src_image_view) { } } // Anonymous namespace -BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_, +BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_, StateTracker& state_tracker_, DescriptorPool& descriptor_pool) : device{device_}, scheduler{scheduler_}, state_tracker{state_tracker_}, one_texture_set_layout(device.GetLogical().CreateDescriptorSetLayout( @@ -366,16 +366,14 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_, PipelineLayoutCreateInfo(two_textures_set_layout.address()))), full_screen_vert(BuildShader(device, FULL_SCREEN_TRIANGLE_VERT_SPV)), blit_color_to_color_frag(BuildShader(device, VULKAN_BLIT_COLOR_FLOAT_FRAG_SPV)), + blit_depth_stencil_frag(BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV)), convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)), convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)), convert_abgr8_to_d24s8_frag(BuildShader(device, CONVERT_ABGR8_TO_D24S8_FRAG_SPV)), convert_d24s8_to_abgr8_frag(BuildShader(device, CONVERT_D24S8_TO_ABGR8_FRAG_SPV)), + convert_s8d24_to_abgr8_frag(BuildShader(device, CONVERT_S8D24_TO_ABGR8_FRAG_SPV)), linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)), - nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) { - if (device.IsExtShaderStencilExportSupported()) { - blit_depth_stencil_frag = BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV); - } -} + nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {} BlitImageHelper::~BlitImageHelper() = default; @@ -474,6 +472,13 @@ void BlitImageHelper::ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer, ConvertDepthStencil(*convert_d24s8_to_abgr8_pipeline, dst_framebuffer, src_image_view); } +void BlitImageHelper::ConvertS8D24ToABGR8(const Framebuffer* dst_framebuffer, + ImageView& src_image_view) { + ConvertPipelineColorTargetEx(convert_s8d24_to_abgr8_pipeline, dst_framebuffer->RenderPass(), + convert_s8d24_to_abgr8_frag); + ConvertDepthStencil(*convert_s8d24_to_abgr8_pipeline, dst_framebuffer, src_image_view); +} + void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, const ImageView& src_image_view) { const VkPipelineLayout layout = *one_texture_pipeline_layout; diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 85e7dca5b..5df679fb4 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -1,11 +1,8 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <compare> - #include "video_core/engines/fermi_2d.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/texture_cache/types.h" @@ -19,7 +16,7 @@ class Device; class Framebuffer; class ImageView; class StateTracker; -class VKScheduler; +class Scheduler; struct BlitImagePipelineKey { constexpr auto operator<=>(const BlitImagePipelineKey&) const noexcept = default; @@ -30,7 +27,7 @@ struct BlitImagePipelineKey { class BlitImageHelper { public: - explicit BlitImageHelper(const Device& device, VKScheduler& scheduler, + explicit BlitImageHelper(const Device& device, Scheduler& scheduler, StateTracker& state_tracker, DescriptorPool& descriptor_pool); ~BlitImageHelper(); @@ -56,6 +53,8 @@ public: void ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer, ImageView& src_image_view); + void ConvertS8D24ToABGR8(const Framebuffer* dst_framebuffer, ImageView& src_image_view); + private: void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, const ImageView& src_image_view); @@ -83,7 +82,7 @@ private: vk::ShaderModule& module); const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; StateTracker& state_tracker; vk::DescriptorSetLayout one_texture_set_layout; @@ -99,6 +98,7 @@ private: vk::ShaderModule convert_float_to_depth_frag; vk::ShaderModule convert_abgr8_to_d24s8_frag; vk::ShaderModule convert_d24s8_to_abgr8_frag; + vk::ShaderModule convert_s8d24_to_abgr8_frag; vk::Sampler linear_sampler; vk::Sampler nearest_sampler; @@ -112,6 +112,7 @@ private: vk::Pipeline convert_r16_to_d16_pipeline; vk::Pipeline convert_abgr8_to_d24s8_pipeline; vk::Pipeline convert_d24s8_to_abgr8_pipeline; + vk::Pipeline convert_s8d24_to_abgr8_pipeline; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp index d70153df3..733b454de 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp @@ -1,12 +1,8 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <cstring> -#include <tuple> - -#include <boost/functional/hash.hpp> #include "common/bit_cast.h" #include "common/cityhash.h" diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h index c9be37935..9d60756e5 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index 751e4792b..e7104d377 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <iterator> @@ -26,7 +25,7 @@ VkFilter Filter(Tegra::Texture::TextureFilter filter) { case Tegra::Texture::TextureFilter::Linear: return VK_FILTER_LINEAR; } - UNREACHABLE_MSG("Invalid sampler filter={}", filter); + ASSERT_MSG(false, "Invalid sampler filter={}", filter); return {}; } @@ -43,7 +42,7 @@ VkSamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter case Tegra::Texture::TextureMipmapFilter::Linear: return VK_SAMPLER_MIPMAP_MODE_LINEAR; } - UNREACHABLE_MSG("Invalid sampler mipmap mode={}", mipmap_filter); + ASSERT_MSG(false, "Invalid sampler mipmap mode={}", mipmap_filter); return {}; } @@ -71,7 +70,7 @@ VkSamplerAddressMode WrapMode(const Device& device, Tegra::Texture::WrapMode wra case Tegra::Texture::TextureFilter::Linear: return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER; } - UNREACHABLE(); + ASSERT(false); return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE; case Tegra::Texture::WrapMode::MirrorOnceClampToEdge: return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE; @@ -127,6 +126,7 @@ struct FormatTuple { {VK_FORMAT_A2B10G10R10_UNORM_PACK32, Attachable | Storage}, // A2B10G10R10_UNORM {VK_FORMAT_A2B10G10R10_UINT_PACK32, Attachable | Storage}, // A2B10G10R10_UINT {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable}, // A1B5G5R5_UNORM (flipped with swizzle) + {VK_FORMAT_R5G5B5A1_UNORM_PACK16}, // A5B5G5R1_UNORM (specially swizzled) {VK_FORMAT_R8_UNORM, Attachable | Storage}, // R8_UNORM {VK_FORMAT_R8_SNORM, Attachable | Storage}, // R8_SNORM {VK_FORMAT_R8_SINT, Attachable | Storage}, // R8_SINT @@ -172,7 +172,7 @@ struct FormatTuple { {VK_FORMAT_R8G8_SINT, Attachable | Storage}, // R8G8_SINT {VK_FORMAT_R8G8_UINT, Attachable | Storage}, // R8G8_UINT {VK_FORMAT_R32G32_UINT, Attachable | Storage}, // R32G32_UINT - {VK_FORMAT_UNDEFINED}, // R16G16B16X16_FLOAT + {VK_FORMAT_R16G16B16A16_SFLOAT, Attachable | Storage}, // R16G16B16X16_FLOAT {VK_FORMAT_R32_UINT, Attachable | Storage}, // R32_UINT {VK_FORMAT_R32_SINT, Attachable | Storage}, // R32_SINT {VK_FORMAT_ASTC_8x8_UNORM_BLOCK}, // ASTC_2D_8X8_UNORM @@ -184,6 +184,7 @@ struct FormatTuple { {VK_FORMAT_BC3_SRGB_BLOCK}, // BC3_SRGB {VK_FORMAT_BC7_SRGB_BLOCK}, // BC7_SRGB {VK_FORMAT_R4G4B4A4_UNORM_PACK16, Attachable}, // A4B4G4R4_UNORM + {VK_FORMAT_R4G4_UNORM_PACK8}, // G4R4_UNORM {VK_FORMAT_ASTC_4x4_SRGB_BLOCK}, // ASTC_2D_4X4_SRGB {VK_FORMAT_ASTC_8x8_SRGB_BLOCK}, // ASTC_2D_8X8_SRGB {VK_FORMAT_ASTC_8x5_SRGB_BLOCK}, // ASTC_2D_8X5_SRGB @@ -194,6 +195,9 @@ struct FormatTuple { {VK_FORMAT_ASTC_10x8_SRGB_BLOCK}, // ASTC_2D_10X8_SRGB {VK_FORMAT_ASTC_6x6_UNORM_BLOCK}, // ASTC_2D_6X6_UNORM {VK_FORMAT_ASTC_6x6_SRGB_BLOCK}, // ASTC_2D_6X6_SRGB + {VK_FORMAT_ASTC_10x6_UNORM_BLOCK}, // ASTC_2D_10X6_UNORM + {VK_FORMAT_ASTC_10x5_UNORM_BLOCK}, // ASTC_2D_10X5_UNORM + {VK_FORMAT_ASTC_10x5_SRGB_BLOCK}, // ASTC_2D_10X5_SRGB {VK_FORMAT_ASTC_10x10_UNORM_BLOCK}, // ASTC_2D_10X10_UNORM {VK_FORMAT_ASTC_10x10_SRGB_BLOCK}, // ASTC_2D_10X10_SRGB {VK_FORMAT_ASTC_12x12_UNORM_BLOCK}, // ASTC_2D_12X12_UNORM @@ -315,193 +319,204 @@ VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const Device& device, } } -VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) { - switch (type) { - case Maxwell::VertexAttribute::Type::UnsignedNorm: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_UNORM; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_UNORM; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_UNORM; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_UNORM; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_UNORM; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_UNORM; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_UNORM; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_UNORM; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_UNORM_PACK32; - default: +VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type, + Maxwell::VertexAttribute::Size size) { + const VkFormat format{([&]() { + switch (type) { + case Maxwell::VertexAttribute::Type::UnsignedNorm: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_UNORM; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_UNORM; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_UNORM; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_UNORM; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_UNORM; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_UNORM; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_UNORM; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_UNORM; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_UNORM_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::SignedNorm: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_SNORM; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_SNORM; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_SNORM; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_SNORM; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_SNORM; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_SNORM; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_SNORM; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_SNORM; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_SNORM_PACK32; - default: + case Maxwell::VertexAttribute::Type::SignedNorm: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_SNORM; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_SNORM; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_SNORM; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_SNORM; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_SNORM; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_SNORM; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_SNORM; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_SNORM; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_SNORM_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::UnsignedScaled: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_USCALED; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_USCALED; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_USCALED; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_USCALED; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_USCALED; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_USCALED; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_USCALED; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_USCALED; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_USCALED_PACK32; - default: + case Maxwell::VertexAttribute::Type::UnsignedScaled: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_USCALED; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_USCALED; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_USCALED; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_USCALED; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_USCALED; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_USCALED; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_USCALED; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_USCALED; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_USCALED_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::SignedScaled: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_SSCALED; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_SSCALED; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_SSCALED; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_SSCALED; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_SSCALED; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_SSCALED; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_SSCALED; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_SSCALED; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_SSCALED_PACK32; - default: + case Maxwell::VertexAttribute::Type::SignedScaled: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_SSCALED; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_SSCALED; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_SSCALED; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_SSCALED; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_SSCALED; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_SSCALED; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_SSCALED; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_SSCALED; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_SSCALED_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::UnsignedInt: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_UINT; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_UINT; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_UINT; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_UINT; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_UINT; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_UINT; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_UINT; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_UINT; - case Maxwell::VertexAttribute::Size::Size_32: - return VK_FORMAT_R32_UINT; - case Maxwell::VertexAttribute::Size::Size_32_32: - return VK_FORMAT_R32G32_UINT; - case Maxwell::VertexAttribute::Size::Size_32_32_32: - return VK_FORMAT_R32G32B32_UINT; - case Maxwell::VertexAttribute::Size::Size_32_32_32_32: - return VK_FORMAT_R32G32B32A32_UINT; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_UINT_PACK32; - default: + case Maxwell::VertexAttribute::Type::UnsignedInt: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_UINT; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_UINT; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_UINT; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_UINT; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_UINT; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_UINT; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_UINT; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_UINT; + case Maxwell::VertexAttribute::Size::Size_32: + return VK_FORMAT_R32_UINT; + case Maxwell::VertexAttribute::Size::Size_32_32: + return VK_FORMAT_R32G32_UINT; + case Maxwell::VertexAttribute::Size::Size_32_32_32: + return VK_FORMAT_R32G32B32_UINT; + case Maxwell::VertexAttribute::Size::Size_32_32_32_32: + return VK_FORMAT_R32G32B32A32_UINT; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_UINT_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::SignedInt: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_8: - return VK_FORMAT_R8_SINT; - case Maxwell::VertexAttribute::Size::Size_8_8: - return VK_FORMAT_R8G8_SINT; - case Maxwell::VertexAttribute::Size::Size_8_8_8: - return VK_FORMAT_R8G8B8_SINT; - case Maxwell::VertexAttribute::Size::Size_8_8_8_8: - return VK_FORMAT_R8G8B8A8_SINT; - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_SINT; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_SINT; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_SINT; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_SINT; - case Maxwell::VertexAttribute::Size::Size_32: - return VK_FORMAT_R32_SINT; - case Maxwell::VertexAttribute::Size::Size_32_32: - return VK_FORMAT_R32G32_SINT; - case Maxwell::VertexAttribute::Size::Size_32_32_32: - return VK_FORMAT_R32G32B32_SINT; - case Maxwell::VertexAttribute::Size::Size_32_32_32_32: - return VK_FORMAT_R32G32B32A32_SINT; - case Maxwell::VertexAttribute::Size::Size_10_10_10_2: - return VK_FORMAT_A2B10G10R10_SINT_PACK32; - default: + case Maxwell::VertexAttribute::Type::SignedInt: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_8: + return VK_FORMAT_R8_SINT; + case Maxwell::VertexAttribute::Size::Size_8_8: + return VK_FORMAT_R8G8_SINT; + case Maxwell::VertexAttribute::Size::Size_8_8_8: + return VK_FORMAT_R8G8B8_SINT; + case Maxwell::VertexAttribute::Size::Size_8_8_8_8: + return VK_FORMAT_R8G8B8A8_SINT; + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_SINT; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_SINT; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_SINT; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_SINT; + case Maxwell::VertexAttribute::Size::Size_32: + return VK_FORMAT_R32_SINT; + case Maxwell::VertexAttribute::Size::Size_32_32: + return VK_FORMAT_R32G32_SINT; + case Maxwell::VertexAttribute::Size::Size_32_32_32: + return VK_FORMAT_R32G32B32_SINT; + case Maxwell::VertexAttribute::Size::Size_32_32_32_32: + return VK_FORMAT_R32G32B32A32_SINT; + case Maxwell::VertexAttribute::Size::Size_10_10_10_2: + return VK_FORMAT_A2B10G10R10_SINT_PACK32; + default: + break; + } break; - } - break; - case Maxwell::VertexAttribute::Type::Float: - switch (size) { - case Maxwell::VertexAttribute::Size::Size_16: - return VK_FORMAT_R16_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_16_16: - return VK_FORMAT_R16G16_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_16_16_16: - return VK_FORMAT_R16G16B16_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_16_16_16_16: - return VK_FORMAT_R16G16B16A16_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_32: - return VK_FORMAT_R32_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_32_32: - return VK_FORMAT_R32G32_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_32_32_32: - return VK_FORMAT_R32G32B32_SFLOAT; - case Maxwell::VertexAttribute::Size::Size_32_32_32_32: - return VK_FORMAT_R32G32B32A32_SFLOAT; - default: + case Maxwell::VertexAttribute::Type::Float: + switch (size) { + case Maxwell::VertexAttribute::Size::Size_16: + return VK_FORMAT_R16_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_16_16: + return VK_FORMAT_R16G16_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_16_16_16: + return VK_FORMAT_R16G16B16_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_16_16_16_16: + return VK_FORMAT_R16G16B16A16_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_32: + return VK_FORMAT_R32_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_32_32: + return VK_FORMAT_R32G32_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_32_32_32: + return VK_FORMAT_R32G32B32_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_32_32_32_32: + return VK_FORMAT_R32G32B32A32_SFLOAT; + case Maxwell::VertexAttribute::Size::Size_11_11_10: + return VK_FORMAT_B10G11R11_UFLOAT_PACK32; + default: + break; + } break; } - break; + return VK_FORMAT_UNDEFINED; + })()}; + + if (format == VK_FORMAT_UNDEFINED) { + UNIMPLEMENTED_MSG("Unimplemented vertex format of type={} and size={}", type, size); } - UNIMPLEMENTED_MSG("Unimplemented vertex format of type={} and size={}", type, size); - return {}; + + return device.GetSupportedFormat(format, VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT, + FormatType::Buffer); } VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison) { @@ -741,7 +756,7 @@ VkViewportCoordinateSwizzleNV ViewportSwizzle(Maxwell::ViewportSwizzle swizzle) case Maxwell::ViewportSwizzle::NegativeW: return VK_VIEWPORT_COORDINATE_SWIZZLE_NEGATIVE_W_NV; } - UNREACHABLE_MSG("Invalid swizzle={}", swizzle); + ASSERT_MSG(false, "Invalid swizzle={}", swizzle); return {}; } @@ -754,7 +769,7 @@ VkSamplerReductionMode SamplerReduction(Tegra::Texture::SamplerReduction reducti case Tegra::Texture::SamplerReduction::Max: return VK_SAMPLER_REDUCTION_MODE_MAX_EXT; } - UNREACHABLE_MSG("Invalid sampler mode={}", static_cast<int>(reduction)); + ASSERT_MSG(false, "Invalid sampler mode={}", static_cast<int>(reduction)); return VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT; } @@ -777,7 +792,7 @@ VkSampleCountFlagBits MsaaMode(Tegra::Texture::MsaaMode msaa_mode) { case Tegra::Texture::MsaaMode::Msaa4x4: return VK_SAMPLE_COUNT_16_BIT; default: - UNREACHABLE_MSG("Invalid msaa_mode={}", static_cast<int>(msaa_mode)); + ASSERT_MSG(false, "Invalid msaa_mode={}", static_cast<int>(msaa_mode)); return VK_SAMPLE_COUNT_1_BIT; } } diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.h b/src/video_core/renderer_vulkan/maxwell_to_vk.h index 8a9616039..356d46292 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.h +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h @@ -1,10 +1,8 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include "common/common_types.h" #include "shader_recompiler/stage.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/surface.h" @@ -50,7 +48,8 @@ VkShaderStageFlagBits ShaderStage(Shader::Stage stage); VkPrimitiveTopology PrimitiveTopology(const Device& device, Maxwell::PrimitiveTopology topology); -VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size); +VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type, + Maxwell::VertexAttribute::Size size); VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison); diff --git a/src/video_core/renderer_vulkan/pipeline_helper.h b/src/video_core/renderer_vulkan/pipeline_helper.h index 11c160570..b24f3424a 100644 --- a/src/video_core/renderer_vulkan/pipeline_helper.h +++ b/src/video_core/renderer_vulkan/pipeline_helper.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,7 +7,6 @@ #include <boost/container/small_vector.hpp> -#include "common/assert.h" #include "common/common_types.h" #include "shader_recompiler/backend/spirv/emit_spirv.h" #include "shader_recompiler/shader_info.h" @@ -16,7 +14,6 @@ #include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/texture_cache/texture_cache.h" #include "video_core/texture_cache/types.h" -#include "video_core/textures/texture.h" #include "video_core/vulkan_common/vulkan_device.h" namespace Vulkan { @@ -171,7 +168,7 @@ private: }; inline void PushImageDescriptors(TextureCache& texture_cache, - VKUpdateDescriptorQueue& update_descriptor_queue, + UpdateDescriptorQueue& update_descriptor_queue, const Shader::Info& info, RescalingPushConstant& rescaling, const VkSampler*& samplers, const VideoCommon::ImageViewInOut*& views) { diff --git a/src/video_core/renderer_vulkan/pipeline_statistics.cpp b/src/video_core/renderer_vulkan/pipeline_statistics.cpp index bfec931a6..fcd160a32 100644 --- a/src/video_core/renderer_vulkan/pipeline_statistics.cpp +++ b/src/video_core/renderer_vulkan/pipeline_statistics.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <string_view> @@ -57,7 +56,7 @@ void PipelineStatistics::Collect(VkPipeline pipeline) { stage_stats.basic_block_count = GetUint64(statistic); } } - std::lock_guard lock{mutex}; + std::scoped_lock lock{mutex}; collected_stats.push_back(stage_stats); } } @@ -66,7 +65,7 @@ void PipelineStatistics::Report() const { double num{}; Stats total; { - std::lock_guard lock{mutex}; + std::scoped_lock lock{mutex}; for (const Stats& stats : collected_stats) { total.code_size += stats.code_size; total.register_count += stats.register_count; diff --git a/src/video_core/renderer_vulkan/pipeline_statistics.h b/src/video_core/renderer_vulkan/pipeline_statistics.h index b61840107..197bb0936 100644 --- a/src/video_core/renderer_vulkan/pipeline_statistics.h +++ b/src/video_core/renderer_vulkan/pipeline_statistics.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 74822814d..d8131232a 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -13,16 +12,15 @@ #include <fmt/format.h> #include "common/logging/log.h" +#include "common/scope_exit.h" #include "common/settings.h" #include "common/telemetry.h" -#include "core/core.h" #include "core/core_timing.h" #include "core/frontend/emu_window.h" #include "core/telemetry_session.h" #include "video_core/gpu.h" #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" -#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_state_tracker.h" @@ -104,13 +102,13 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, debug_callback(Settings::values.renderer_debug ? CreateDebugCallback(instance) : nullptr), surface(CreateSurface(instance, render_window)), device(CreateDevice(instance, dld, *surface)), memory_allocator(device, false), - state_tracker(gpu), scheduler(device, state_tracker), + state_tracker(), scheduler(device, state_tracker), swapchain(*surface, device, scheduler, render_window.GetFramebufferLayout().width, render_window.GetFramebufferLayout().height, false), blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, scheduler, screen_info), - rasterizer(render_window, gpu, gpu.MemoryManager(), cpu_memory, screen_info, device, - memory_allocator, state_tracker, scheduler) { + rasterizer(render_window, gpu, cpu_memory, screen_info, device, memory_allocator, + state_tracker, scheduler) { Report(); } catch (const vk::Exception& exception) { LOG_ERROR(Render_Vulkan, "Vulkan initialization failed with error: {}", exception.what()); @@ -129,6 +127,11 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { if (!render_window.IsShown()) { return; } + // Update screen info if the framebuffer size has changed. + if (screen_info.width != framebuffer->width || screen_info.height != framebuffer->height) { + screen_info.width = framebuffer->width; + screen_info.height = framebuffer->height; + } const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; const bool use_accelerated = rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); @@ -139,7 +142,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { const auto recreate_swapchain = [&] { if (!has_been_recreated) { has_been_recreated = true; - scheduler.WaitWorker(); + scheduler.Finish(); } const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); swapchain.Create(layout.width, layout.height, is_srgb); diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index 6dc985109..e7bfecb20 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -1,12 +1,10 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <memory> #include <string> -#include <vector> #include "common/dynamic_library.h" #include "video_core/renderer_base.h" @@ -67,14 +65,14 @@ private: vk::DebugUtilsMessenger debug_callback; vk::SurfaceKHR surface; - VKScreenInfo screen_info; + ScreenInfo screen_info; Device device; MemoryAllocator memory_allocator; StateTracker state_tracker; - VKScheduler scheduler; - VKSwapchain swapchain; - VKBlitScreen blit_screen; + Scheduler scheduler; + Swapchain swapchain; + BlitScreen blit_screen; RasterizerVulkan rasterizer; }; diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index c71a1f44d..cb7fa2078 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -1,12 +1,10 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <cstring> #include <memory> -#include <tuple> #include <vector> #include "common/assert.h" @@ -28,7 +26,6 @@ #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" #include "video_core/renderer_vulkan/vk_fsr.h" -#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_shader_util.h" #include "video_core/renderer_vulkan/vk_swapchain.h" @@ -96,10 +93,12 @@ std::size_t GetSizeInBytes(const Tegra::FramebufferConfig& framebuffer) { VkFormat GetFormat(const Tegra::FramebufferConfig& framebuffer) { switch (framebuffer.pixel_format) { - case Tegra::FramebufferConfig::PixelFormat::A8B8G8R8_UNORM: + case Service::android::PixelFormat::Rgba8888: return VK_FORMAT_A8B8G8R8_UNORM_PACK32; - case Tegra::FramebufferConfig::PixelFormat::RGB565_UNORM: + case Service::android::PixelFormat::Rgb565: return VK_FORMAT_R5G6B5_UNORM_PACK16; + case Service::android::PixelFormat::Bgra8888: + return VK_FORMAT_B8G8R8A8_UNORM; default: UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}", static_cast<u32>(framebuffer.pixel_format)); @@ -109,7 +108,7 @@ VkFormat GetFormat(const Tegra::FramebufferConfig& framebuffer) { } // Anonymous namespace -struct VKBlitScreen::BufferData { +struct BlitScreen::BufferData { struct { std::array<f32, 4 * 4> modelview_matrix; } uniform; @@ -119,10 +118,9 @@ struct VKBlitScreen::BufferData { // Unaligned image data goes here }; -VKBlitScreen::VKBlitScreen(Core::Memory::Memory& cpu_memory_, - Core::Frontend::EmuWindow& render_window_, const Device& device_, - MemoryAllocator& memory_allocator_, VKSwapchain& swapchain_, - VKScheduler& scheduler_, const VKScreenInfo& screen_info_) +BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWindow& render_window_, + const Device& device_, MemoryAllocator& memory_allocator_, + Swapchain& swapchain_, Scheduler& scheduler_, const ScreenInfo& screen_info_) : cpu_memory{cpu_memory_}, render_window{render_window_}, device{device_}, memory_allocator{memory_allocator_}, swapchain{swapchain_}, scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} { @@ -132,21 +130,26 @@ VKBlitScreen::VKBlitScreen(Core::Memory::Memory& cpu_memory_, CreateDynamicResources(); } -VKBlitScreen::~VKBlitScreen() = default; +BlitScreen::~BlitScreen() = default; -void VKBlitScreen::Recreate() { +void BlitScreen::Recreate() { CreateDynamicResources(); } -VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, - const VkFramebuffer& host_framebuffer, - const Layout::FramebufferLayout layout, VkExtent2D render_area, - bool use_accelerated) { +VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, + const VkFramebuffer& host_framebuffer, + const Layout::FramebufferLayout layout, VkExtent2D render_area, + bool use_accelerated) { RefreshResources(framebuffer); // Finish any pending renderpass scheduler.RequestOutsideRenderPassOperationContext(); + if (const auto swapchain_images = swapchain.GetImageCount(); swapchain_images != image_count) { + image_count = swapchain_images; + Recreate(); + } + const std::size_t image_index = swapchain.GetImageIndex(); scheduler.Wait(resource_ticks[image_index]); @@ -171,11 +174,12 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, // TODO(Rodrigo): Read this from HLE constexpr u32 block_height_log2 = 4; const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer); - const u64 size_bytes{Tegra::Texture::CalculateSize(true, bytes_per_pixel, + const u64 linear_size{GetSizeInBytes(framebuffer)}; + const u64 tiled_size{Tegra::Texture::CalculateSize(true, bytes_per_pixel, framebuffer.stride, framebuffer.height, 1, block_height_log2, 0)}; Tegra::Texture::UnswizzleTexture( - mapped_span.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes), + mapped_span.subspan(image_offset, linear_size), std::span(host_ptr, tiled_size), bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0); const VkBufferImageCopy copy{ @@ -420,20 +424,20 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, return *semaphores[image_index]; } -VkSemaphore VKBlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer, - bool use_accelerated) { +VkSemaphore BlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer, + bool use_accelerated) { const std::size_t image_index = swapchain.GetImageIndex(); const VkExtent2D render_area = swapchain.GetSize(); const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); return Draw(framebuffer, *framebuffers[image_index], layout, render_area, use_accelerated); } -vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) { +vk::Framebuffer BlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) { return CreateFramebuffer(image_view, extent, renderpass); } -vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent, - vk::RenderPass& rd) { +vk::Framebuffer BlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent, + vk::RenderPass& rd) { return device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, .pNext = nullptr, @@ -447,17 +451,17 @@ vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, V }); } -void VKBlitScreen::CreateStaticResources() { +void BlitScreen::CreateStaticResources() { CreateShaders(); + CreateSampler(); +} + +void BlitScreen::CreateDynamicResources() { CreateSemaphores(); CreateDescriptorPool(); CreateDescriptorSetLayout(); CreateDescriptorSets(); CreatePipelineLayout(); - CreateSampler(); -} - -void VKBlitScreen::CreateDynamicResources() { CreateRenderPass(); CreateFramebuffers(); CreateGraphicsPipeline(); @@ -467,7 +471,7 @@ void VKBlitScreen::CreateDynamicResources() { } } -void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) { +void BlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) { if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) { if (!fsr) { CreateFSR(); @@ -487,7 +491,7 @@ void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) CreateRawImages(framebuffer); } -void VKBlitScreen::CreateShaders() { +void BlitScreen::CreateShaders() { vertex_shader = BuildShader(device, VULKAN_PRESENT_VERT_SPV); fxaa_vertex_shader = BuildShader(device, FXAA_VERT_SPV); fxaa_fragment_shader = BuildShader(device, FXAA_FRAG_SPV); @@ -501,12 +505,12 @@ void VKBlitScreen::CreateShaders() { } } -void VKBlitScreen::CreateSemaphores() { +void BlitScreen::CreateSemaphores() { semaphores.resize(image_count); std::ranges::generate(semaphores, [this] { return device.GetLogical().CreateSemaphore(); }); } -void VKBlitScreen::CreateDescriptorPool() { +void BlitScreen::CreateDescriptorPool() { const std::array<VkDescriptorPoolSize, 2> pool_sizes{{ { .type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, @@ -546,11 +550,11 @@ void VKBlitScreen::CreateDescriptorPool() { aa_descriptor_pool = device.GetLogical().CreateDescriptorPool(ci_aa); } -void VKBlitScreen::CreateRenderPass() { +void BlitScreen::CreateRenderPass() { renderpass = CreateRenderPassImpl(swapchain.GetImageViewFormat()); } -vk::RenderPass VKBlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present) { +vk::RenderPass BlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present) { const VkAttachmentDescription color_attachment{ .flags = 0, .format = format, @@ -606,7 +610,7 @@ vk::RenderPass VKBlitScreen::CreateRenderPassImpl(VkFormat format, bool is_prese return device.GetLogical().CreateRenderPass(renderpass_ci); } -void VKBlitScreen::CreateDescriptorSetLayout() { +void BlitScreen::CreateDescriptorSetLayout() { const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings{{ { .binding = 0, @@ -661,7 +665,7 @@ void VKBlitScreen::CreateDescriptorSetLayout() { aa_descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci_aa); } -void VKBlitScreen::CreateDescriptorSets() { +void BlitScreen::CreateDescriptorSets() { const std::vector layouts(image_count, *descriptor_set_layout); const std::vector layouts_aa(image_count, *aa_descriptor_set_layout); @@ -685,7 +689,7 @@ void VKBlitScreen::CreateDescriptorSets() { aa_descriptor_sets = aa_descriptor_pool.Allocate(ai_aa); } -void VKBlitScreen::CreatePipelineLayout() { +void BlitScreen::CreatePipelineLayout() { const VkPipelineLayoutCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, .pNext = nullptr, @@ -708,7 +712,7 @@ void VKBlitScreen::CreatePipelineLayout() { aa_pipeline_layout = device.GetLogical().CreatePipelineLayout(ci_aa); } -void VKBlitScreen::CreateGraphicsPipeline() { +void BlitScreen::CreateGraphicsPipeline() { const std::array<VkPipelineShaderStageCreateInfo, 2> bilinear_shader_stages{{ { .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, @@ -981,7 +985,7 @@ void VKBlitScreen::CreateGraphicsPipeline() { scaleforce_pipeline = device.GetLogical().CreateGraphicsPipeline(scaleforce_pipeline_ci); } -void VKBlitScreen::CreateSampler() { +void BlitScreen::CreateSampler() { const VkSamplerCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, .pNext = nullptr, @@ -1028,7 +1032,7 @@ void VKBlitScreen::CreateSampler() { nn_sampler = device.GetLogical().CreateSampler(ci_nn); } -void VKBlitScreen::CreateFramebuffers() { +void BlitScreen::CreateFramebuffers() { const VkExtent2D size{swapchain.GetSize()}; framebuffers.resize(image_count); @@ -1038,7 +1042,7 @@ void VKBlitScreen::CreateFramebuffers() { } } -void VKBlitScreen::ReleaseRawImages() { +void BlitScreen::ReleaseRawImages() { for (const u64 tick : resource_ticks) { scheduler.Wait(tick); } @@ -1053,7 +1057,7 @@ void VKBlitScreen::ReleaseRawImages() { buffer_commit = MemoryCommit{}; } -void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer) { +void BlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer) { const VkBufferCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, @@ -1070,7 +1074,7 @@ void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuff buffer_commit = memory_allocator.Commit(buffer, MemoryUsage::Upload); } -void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) { +void BlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) { raw_images.resize(image_count); raw_image_views.resize(image_count); raw_buffer_commits.resize(image_count); @@ -1295,8 +1299,8 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) aa_pipeline = device.GetLogical().CreateGraphicsPipeline(fxaa_pipeline_ci); } -void VKBlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, - bool nn) const { +void BlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { const VkDescriptorImageInfo image_info{ .sampler = nn ? *nn_sampler : *sampler, .imageView = image_view, @@ -1332,8 +1336,8 @@ void VKBlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView im device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, sampler_write_2}, {}); } -void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, - bool nn) const { +void BlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { const VkDescriptorBufferInfo buffer_info{ .buffer = *buffer, .offset = offsetof(BufferData, uniform), @@ -1375,13 +1379,13 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag device.GetLogical().UpdateDescriptorSets(std::array{ubo_write, sampler_write}, {}); } -void VKBlitScreen::SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const { +void BlitScreen::SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const { data.uniform.modelview_matrix = MakeOrthographicMatrix(static_cast<f32>(layout.width), static_cast<f32>(layout.height)); } -void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer, - const Layout::FramebufferLayout layout) const { +void BlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer, + const Layout::FramebufferLayout layout) const { const auto& framebuffer_transform_flags = framebuffer.transform_flags; const auto& framebuffer_crop_rect = framebuffer.crop_rect; @@ -1390,9 +1394,9 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi auto right = texcoords.right; switch (framebuffer_transform_flags) { - case Tegra::FramebufferConfig::TransformFlags::Unset: + case Service::android::BufferTransformFlags::Unset: break; - case Tegra::FramebufferConfig::TransformFlags::FlipV: + case Service::android::BufferTransformFlags::FlipV: // Flip the framebuffer vertically left = texcoords.right; right = texcoords.left; @@ -1403,11 +1407,15 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi break; } - UNIMPLEMENTED_IF(framebuffer_crop_rect.top != 0); UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0); - f32 scale_u = 1.0f; - f32 scale_v = 1.0f; + f32 left_start{}; + if (framebuffer_crop_rect.Top() > 0) { + left_start = static_cast<f32>(framebuffer_crop_rect.Top()) / + static_cast<f32>(framebuffer_crop_rect.Bottom()); + } + f32 scale_u = static_cast<f32>(framebuffer.width) / static_cast<f32>(screen_info.width); + f32 scale_v = static_cast<f32>(framebuffer.height) / static_cast<f32>(screen_info.height); // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering // (e.g. handheld mode) on a 1920x1080 framebuffer. if (!fsr) { @@ -1426,13 +1434,16 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi const auto y = static_cast<f32>(screen.top); const auto w = static_cast<f32>(screen.GetWidth()); const auto h = static_cast<f32>(screen.GetHeight()); - data.vertices[0] = ScreenRectVertex(x, y, texcoords.top * scale_u, left * scale_v); - data.vertices[1] = ScreenRectVertex(x + w, y, texcoords.bottom * scale_u, left * scale_v); - data.vertices[2] = ScreenRectVertex(x, y + h, texcoords.top * scale_u, right * scale_v); - data.vertices[3] = ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, right * scale_v); + data.vertices[0] = ScreenRectVertex(x, y, texcoords.top * scale_u, left_start + left * scale_v); + data.vertices[1] = + ScreenRectVertex(x + w, y, texcoords.bottom * scale_u, left_start + left * scale_v); + data.vertices[2] = + ScreenRectVertex(x, y + h, texcoords.top * scale_u, left_start + right * scale_v); + data.vertices[3] = + ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, left_start + right * scale_v); } -void VKBlitScreen::CreateFSR() { +void BlitScreen::CreateFSR() { const auto& layout = render_window.GetFramebufferLayout(); const VkExtent2D fsr_size{ .width = layout.screen.GetWidth(), @@ -1441,12 +1452,12 @@ void VKBlitScreen::CreateFSR() { fsr = std::make_unique<FSR>(device, memory_allocator, image_count, fsr_size); } -u64 VKBlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const { +u64 BlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const { return sizeof(BufferData) + GetSizeInBytes(framebuffer) * image_count; } -u64 VKBlitScreen::GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer, - std::size_t image_index) const { +u64 BlitScreen::GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer, + std::size_t image_index) const { constexpr auto first_image_offset = static_cast<u64>(sizeof(BufferData)); return first_image_offset + GetSizeInBytes(framebuffer) * image_index; } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index bbca71af3..29e2ea925 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -36,23 +35,22 @@ struct ScreenInfo; class Device; class FSR; class RasterizerVulkan; -class VKScheduler; -class VKSwapchain; +class Scheduler; +class Swapchain; -struct VKScreenInfo { +struct ScreenInfo { VkImageView image_view{}; u32 width{}; u32 height{}; bool is_srgb{}; }; -class VKBlitScreen { +class BlitScreen { public: - explicit VKBlitScreen(Core::Memory::Memory& cpu_memory, - Core::Frontend::EmuWindow& render_window, const Device& device, - MemoryAllocator& memory_manager, VKSwapchain& swapchain, - VKScheduler& scheduler, const VKScreenInfo& screen_info); - ~VKBlitScreen(); + explicit BlitScreen(Core::Memory::Memory& cpu_memory, Core::Frontend::EmuWindow& render_window, + const Device& device, MemoryAllocator& memory_manager, Swapchain& swapchain, + Scheduler& scheduler, const ScreenInfo& screen_info); + ~BlitScreen(); void Recreate(); @@ -109,10 +107,10 @@ private: Core::Frontend::EmuWindow& render_window; const Device& device; MemoryAllocator& memory_allocator; - VKSwapchain& swapchain; - VKScheduler& scheduler; - const std::size_t image_count; - const VKScreenInfo& screen_info; + Swapchain& swapchain; + Scheduler& scheduler; + std::size_t image_count; + const ScreenInfo& screen_info; vk::ShaderModule vertex_shader; vk::ShaderModule fxaa_vertex_shader; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 5ffd93499..558b8db56 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -47,7 +46,7 @@ size_t BytesPerIndex(VkIndexType index_type) { case VK_INDEX_TYPE_UINT32: return 4; default: - UNREACHABLE_MSG("Invalid index type={}", index_type); + ASSERT_MSG(false, "Invalid index type={}", index_type); return 1; } } @@ -125,8 +124,8 @@ VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat } BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, - VKScheduler& scheduler_, StagingBufferPool& staging_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, + Scheduler& scheduler_, StagingBufferPool& staging_pool_, + UpdateDescriptorQueue& update_descriptor_queue_, DescriptorPool& descriptor_pool) : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_}, @@ -141,6 +140,18 @@ StagingBufferRef BufferCacheRuntime::DownloadStagingBuffer(size_t size) { return staging_pool.Request(size, MemoryUsage::Download); } +u64 BufferCacheRuntime::GetDeviceLocalMemory() const { + return device.GetDeviceLocalMemory(); +} + +u64 BufferCacheRuntime::GetDeviceMemoryUsage() const { + return device.GetDeviceMemoryUsage(); +} + +bool BufferCacheRuntime::CanReportMemoryUsage() const { + return device.CanReportMemoryUsage(); +} + void BufferCacheRuntime::Finish() { scheduler.Finish(); } @@ -355,7 +366,7 @@ void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle std::memcpy(staging_data, MakeQuadIndices<u32>(quad, first).data(), quad_size); break; default: - UNREACHABLE(); + ASSERT(false); break; } staging_data += quad_size; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 1ee0d8420..a15c8b39b 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -17,7 +16,7 @@ namespace Vulkan { class Device; class DescriptorPool; -class VKScheduler; +class Scheduler; class BufferCacheRuntime; @@ -59,12 +58,18 @@ class BufferCacheRuntime { public: explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_, - VKScheduler& scheduler_, StagingBufferPool& staging_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, + Scheduler& scheduler_, StagingBufferPool& staging_pool_, + UpdateDescriptorQueue& update_descriptor_queue_, DescriptorPool& descriptor_pool); void Finish(); + u64 GetDeviceLocalMemory() const; + + u64 GetDeviceMemoryUsage() const; + + bool CanReportMemoryUsage() const; + [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size); [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); @@ -119,9 +124,9 @@ private: const Device& device; MemoryAllocator& memory_allocator; - VKScheduler& scheduler; + Scheduler& scheduler; StagingBufferPool& staging_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; vk::Buffer quad_array_lut; MemoryCommit quad_array_lut_commit; diff --git a/src/video_core/renderer_vulkan/vk_command_pool.cpp b/src/video_core/renderer_vulkan/vk_command_pool.cpp index d8e92ac0e..2f09de1c1 100644 --- a/src/video_core/renderer_vulkan/vk_command_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_command_pool.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstddef> diff --git a/src/video_core/renderer_vulkan/vk_command_pool.h b/src/video_core/renderer_vulkan/vk_command_pool.h index 61c26a22a..ec1647f01 100644 --- a/src/video_core/renderer_vulkan/vk_command_pool.h +++ b/src/video_core/renderer_vulkan/vk_command_pool.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 3e96c0f60..241d7573e 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -1,13 +1,11 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <cstring> +#include <array> #include <memory> #include <optional> #include <utility> -#include "common/alignment.h" #include "common/assert.h" #include "common/common_types.h" #include "common/div_ceil.h" @@ -22,15 +20,12 @@ #include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/texture_cache/accelerated_swizzle.h" #include "video_core/texture_cache/types.h" -#include "video_core/textures/astc.h" #include "video_core/textures/decoders.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { -using Tegra::Texture::SWIZZLE_TABLE; - namespace { constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0; @@ -203,9 +198,9 @@ ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool, ComputePass::~ComputePass() = default; -Uint8Pass::Uint8Pass(const Device& device_, VKScheduler& scheduler_, - DescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_) +Uint8Pass::Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool, + StagingBufferPool& staging_buffer_pool_, + UpdateDescriptorQueue& update_descriptor_queue_) : ComputePass(device_, descriptor_pool, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, {}, VULKAN_UINT8_COMP_SPV), @@ -244,10 +239,10 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer return {staging.buffer, staging.offset}; } -QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, +QuadIndexedPass::QuadIndexedPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_) + UpdateDescriptorQueue& update_descriptor_queue_) : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 2>, VULKAN_QUAD_INDEXED_COMP_SPV), @@ -268,7 +263,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedInt: return 2; } - UNREACHABLE(); + ASSERT(false); return 2; }(); const u32 input_size = num_vertices << index_shift; @@ -292,7 +287,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, .dstAccessMask = VK_ACCESS_INDEX_READ_BIT, }; - const std::array push_constants{base_vertex, index_shift}; + const std::array<u32, 2> push_constants{base_vertex, index_shift}; const VkDescriptorSet set = descriptor_allocator.Commit(); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); @@ -306,10 +301,10 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( return {staging.buffer, staging.offset}; } -ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_, +ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, + UpdateDescriptorQueue& update_descriptor_queue_, MemoryAllocator& memory_allocator_) : ComputePass(device_, descriptor_pool_, ASTC_DESCRIPTOR_SET_BINDINGS, ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY, ASTC_BANK_INFO, @@ -331,31 +326,32 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, const VkImageAspectFlags aspect_mask = image.AspectMask(); const VkImage vk_image = image.Handle(); const bool is_initialized = image.ExchangeInitialization(); - scheduler.Record( - [vk_pipeline, vk_image, aspect_mask, is_initialized](vk::CommandBuffer cmdbuf) { - const VkImageMemoryBarrier image_barrier{ - .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = is_initialized ? VK_ACCESS_SHADER_WRITE_BIT : VkAccessFlags{}, - .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, - .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, - .newLayout = VK_IMAGE_LAYOUT_GENERAL, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .image = vk_image, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - }; - cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT - : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); - }); + scheduler.Record([vk_pipeline, vk_image, aspect_mask, + is_initialized](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier image_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = static_cast<VkAccessFlags>(is_initialized ? VK_ACCESS_SHADER_WRITE_BIT + : VK_ACCESS_NONE), + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = vk_image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + cmdbuf.PipelineBarrier(is_initialized ? VK_PIPELINE_STAGE_ALL_COMMANDS_BIT + : VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline); + }); for (const VideoCommon::SwizzleParameters& swizzle : swizzles) { const size_t input_offset = swizzle.buffer_offset + map.offset; const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U); diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index c7b92cce0..dcc691a8e 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -21,8 +20,8 @@ namespace Vulkan { class Device; class StagingBufferPool; -class VKScheduler; -class VKUpdateDescriptorQueue; +class Scheduler; +class UpdateDescriptorQueue; class Image; struct StagingBufferRef; @@ -49,9 +48,9 @@ private: class Uint8Pass final : public ComputePass { public: - explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_, + explicit Uint8Pass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_); + UpdateDescriptorQueue& update_descriptor_queue_); ~Uint8Pass(); /// Assemble uint8 indices into an uint16 index buffer @@ -60,17 +59,17 @@ public: u32 src_offset); private: - VKScheduler& scheduler; + Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; }; class QuadIndexedPass final : public ComputePass { public: - explicit QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, + explicit QuadIndexedPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_); + UpdateDescriptorQueue& update_descriptor_queue_); ~QuadIndexedPass(); std::pair<VkBuffer, VkDeviceSize> Assemble( @@ -78,17 +77,17 @@ public: u32 base_vertex, VkBuffer src_buffer, u32 src_offset); private: - VKScheduler& scheduler; + Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; }; class ASTCDecoderPass final : public ComputePass { public: - explicit ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_, + explicit ASTCDecoderPass(const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, + UpdateDescriptorQueue& update_descriptor_queue_, MemoryAllocator& memory_allocator_); ~ASTCDecoderPass(); @@ -96,9 +95,9 @@ public: std::span<const VideoCommon::SwizzleParameters> swizzles); private: - VKScheduler& scheduler; + Scheduler& scheduler; StagingBufferPool& staging_buffer_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; MemoryAllocator& memory_allocator; }; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index de36bcdb7..7906e11a8 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <vector> @@ -26,7 +25,7 @@ using Shader::Backend::SPIRV::RESCALING_LAYOUT_WORDS_OFFSET; using Tegra::Texture::TexturePair; ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue_, + UpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* thread_worker, PipelineStatistics* pipeline_statistics, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_, @@ -77,7 +76,7 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript if (pipeline_statistics) { pipeline_statistics->Collect(*pipeline); } - std::lock_guard lock{build_mutex}; + std::scoped_lock lock{build_mutex}; is_built = true; build_condvar.notify_one(); if (shader_notify) { @@ -92,7 +91,7 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript } void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, - Tegra::MemoryManager& gpu_memory, VKScheduler& scheduler, + Tegra::MemoryManager& gpu_memory, Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache) { update_descriptor_queue.Acquire(); @@ -127,8 +126,8 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + secondary_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr) << desc.secondary_shift_left}; return TexturePair(lhs_raw | rhs_raw, via_header_index); } } diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.h b/src/video_core/renderer_vulkan/vk_compute_pipeline.h index 8c4b0a301..9879735fe 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -11,7 +10,6 @@ #include "common/common_types.h" #include "common/thread_worker.h" #include "shader_recompiler/shader_info.h" -#include "video_core/memory_manager.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" @@ -26,12 +24,12 @@ namespace Vulkan { class Device; class PipelineStatistics; -class VKScheduler; +class Scheduler; class ComputePipeline { public: explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue, + UpdateDescriptorQueue& update_descriptor_queue, Common::ThreadWorker* thread_worker, PipelineStatistics* pipeline_statistics, VideoCore::ShaderNotify* shader_notify, const Shader::Info& info, @@ -44,11 +42,11 @@ public: ComputePipeline(const ComputePipeline&) = delete; void Configure(Tegra::Engines::KeplerCompute& kepler_compute, Tegra::MemoryManager& gpu_memory, - VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache); + Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache); private: const Device& device; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; Shader::Info info; VideoCommon::ComputeUniformBufferSizes uniform_buffer_sizes{}; diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index d87da2a34..c7196b64e 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <mutex> @@ -122,7 +121,7 @@ vk::DescriptorSets DescriptorAllocator::AllocateDescriptors(size_t count) { throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY); } -DescriptorPool::DescriptorPool(const Device& device_, VKScheduler& scheduler) +DescriptorPool::DescriptorPool(const Device& device_, Scheduler& scheduler) : device{device_}, master_semaphore{scheduler.GetMasterSemaphore()} {} DescriptorPool::~DescriptorPool() = default; diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.h b/src/video_core/renderer_vulkan/vk_descriptor_pool.h index 59466aac5..bd6696b07 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.h +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -15,7 +14,7 @@ namespace Vulkan { class Device; -class VKScheduler; +class Scheduler; struct DescriptorBank; @@ -63,7 +62,7 @@ private: class DescriptorPool { public: - explicit DescriptorPool(const Device& device, VKScheduler& scheduler); + explicit DescriptorPool(const Device& device, Scheduler& scheduler); ~DescriptorPool(); DescriptorPool& operator=(const DescriptorPool&) = delete; diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp index 3bec48d14..0214b103a 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <memory> @@ -9,15 +8,11 @@ #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/vulkan_common/vulkan_device.h" -#include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { -InnerFence::InnerFence(VKScheduler& scheduler_, u32 payload_, bool is_stubbed_) - : FenceBase{payload_, is_stubbed_}, scheduler{scheduler_} {} - -InnerFence::InnerFence(VKScheduler& scheduler_, GPUVAddr address_, u32 payload_, bool is_stubbed_) - : FenceBase{address_, payload_, is_stubbed_}, scheduler{scheduler_} {} +InnerFence::InnerFence(Scheduler& scheduler_, bool is_stubbed_) + : FenceBase{is_stubbed_}, scheduler{scheduler_} {} InnerFence::~InnerFence() = default; @@ -44,30 +39,25 @@ void InnerFence::Wait() { scheduler.Wait(wait_tick); } -VKFenceManager::VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, - TextureCache& texture_cache_, BufferCache& buffer_cache_, - VKQueryCache& query_cache_, const Device& device_, - VKScheduler& scheduler_) +FenceManager::FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, + TextureCache& texture_cache_, BufferCache& buffer_cache_, + QueryCache& query_cache_, const Device& device_, Scheduler& scheduler_) : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_}, scheduler{scheduler_} {} -Fence VKFenceManager::CreateFence(u32 value, bool is_stubbed) { - return std::make_shared<InnerFence>(scheduler, value, is_stubbed); -} - -Fence VKFenceManager::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) { - return std::make_shared<InnerFence>(scheduler, addr, value, is_stubbed); +Fence FenceManager::CreateFence(bool is_stubbed) { + return std::make_shared<InnerFence>(scheduler, is_stubbed); } -void VKFenceManager::QueueFence(Fence& fence) { +void FenceManager::QueueFence(Fence& fence) { fence->Queue(); } -bool VKFenceManager::IsFenceSignaled(Fence& fence) const { +bool FenceManager::IsFenceSignaled(Fence& fence) const { return fence->IsSignaled(); } -void VKFenceManager::WaitFence(Fence& fence) { +void FenceManager::WaitFence(Fence& fence) { fence->Wait(); } diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h index 2f8322d29..7fe2afcd9 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.h +++ b/src/video_core/renderer_vulkan/vk_fence_manager.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,7 +8,6 @@ #include "video_core/fence_manager.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" -#include "video_core/vulkan_common/vulkan_wrapper.h" namespace Core { class System; @@ -22,13 +20,12 @@ class RasterizerInterface; namespace Vulkan { class Device; -class VKQueryCache; -class VKScheduler; +class QueryCache; +class Scheduler; class InnerFence : public VideoCommon::FenceBase { public: - explicit InnerFence(VKScheduler& scheduler_, u32 payload_, bool is_stubbed_); - explicit InnerFence(VKScheduler& scheduler_, GPUVAddr address_, u32 payload_, bool is_stubbed_); + explicit InnerFence(Scheduler& scheduler_, bool is_stubbed_); ~InnerFence(); void Queue(); @@ -38,30 +35,27 @@ public: void Wait(); private: - VKScheduler& scheduler; + Scheduler& scheduler; u64 wait_tick = 0; }; using Fence = std::shared_ptr<InnerFence>; -using GenericFenceManager = - VideoCommon::FenceManager<Fence, TextureCache, BufferCache, VKQueryCache>; +using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>; -class VKFenceManager final : public GenericFenceManager { +class FenceManager final : public GenericFenceManager { public: - explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, - TextureCache& texture_cache, BufferCache& buffer_cache, - VKQueryCache& query_cache, const Device& device, - VKScheduler& scheduler); + explicit FenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + TextureCache& texture_cache, BufferCache& buffer_cache, + QueryCache& query_cache, const Device& device, Scheduler& scheduler); protected: - Fence CreateFence(u32 value, bool is_stubbed) override; - Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override; + Fence CreateFence(bool is_stubbed) override; void QueueFence(Fence& fence) override; bool IsFenceSignaled(Fence& fence) const override; void WaitFence(Fence& fence) override; private: - VKScheduler& scheduler; + Scheduler& scheduler; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_fsr.cpp b/src/video_core/renderer_vulkan/vk_fsr.cpp index b630090e8..dd450169e 100644 --- a/src/video_core/renderer_vulkan/vk_fsr.cpp +++ b/src/video_core/renderer_vulkan/vk_fsr.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cmath> #include "common/bit_cast.h" @@ -173,7 +172,7 @@ FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image CreatePipeline(); } -VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, +VkImageView FSR::Draw(Scheduler& scheduler, size_t image_index, VkImageView image_view, VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect) { UpdateDescriptorSet(image_index, image_view); diff --git a/src/video_core/renderer_vulkan/vk_fsr.h b/src/video_core/renderer_vulkan/vk_fsr.h index 6bbec3d36..5d872861f 100644 --- a/src/video_core/renderer_vulkan/vk_fsr.h +++ b/src/video_core/renderer_vulkan/vk_fsr.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -11,13 +10,13 @@ namespace Vulkan { class Device; -class VKScheduler; +class Scheduler; class FSR { public: explicit FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, VkExtent2D output_size); - VkImageView Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, + VkImageView Draw(Scheduler& scheduler, size_t image_index, VkImageView image_view, VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect); private: diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index d514b71d0..f47786f48 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <span> @@ -216,15 +215,14 @@ ConfigureFuncPtr ConfigureFunc(const std::array<vk::ShaderModule, NUM_STAGES>& m } // Anonymous namespace GraphicsPipeline::GraphicsPipeline( - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, - VKScheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, + Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, + UpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages, const std::array<const Shader::Info*, NUM_STAGES>& infos) - : key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_}, - texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_}, + : key{key_}, device{device_}, texture_cache{texture_cache_}, + buffer_cache{buffer_cache_}, scheduler{scheduler_}, update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} { if (shader_notify) { shader_notify->MarkShaderBuilding(); @@ -258,7 +256,7 @@ GraphicsPipeline::GraphicsPipeline( pipeline_statistics->Collect(*pipeline); } - std::lock_guard lock{build_mutex}; + std::scoped_lock lock{build_mutex}; is_built = true; build_condvar.notify_one(); if (shader_notify) { @@ -289,7 +287,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.SetUniformBuffersState(enabled_uniform_buffer_masks, &uniform_buffer_sizes); - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { const Shader::Info& info{stage_infos[stage]}; @@ -303,7 +301,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { ++ssbo_index; } } - const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; @@ -316,13 +314,14 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + second_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; const u32 raw{lhs_raw | rhs_raw}; return TexturePair(raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { @@ -560,7 +559,7 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { vertex_attributes.push_back({ .location = static_cast<u32>(index), .binding = attribute.buffer, - .format = MaxwellToVK::VertexFormat(attribute.Type(), attribute.Size()), + .format = MaxwellToVK::VertexFormat(device, attribute.Type(), attribute.Size()), .offset = attribute.offset, }); } diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index a0c1d8f07..85602592b 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -63,22 +62,23 @@ class Device; class PipelineStatistics; class RenderPassCache; class RescalingPushConstant; -class VKScheduler; -class VKUpdateDescriptorQueue; +class Scheduler; +class UpdateDescriptorQueue; class GraphicsPipeline { static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage; public: - explicit GraphicsPipeline( - Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, - VKScheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache, - VideoCore::ShaderNotify* shader_notify, const Device& device, - DescriptorPool& descriptor_pool, VKUpdateDescriptorQueue& update_descriptor_queue, - Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, - RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key, - std::array<vk::ShaderModule, NUM_STAGES> stages, - const std::array<const Shader::Info*, NUM_STAGES>& infos); + explicit GraphicsPipeline(Scheduler& scheduler, BufferCache& buffer_cache, + TextureCache& texture_cache, VideoCore::ShaderNotify* shader_notify, + const Device& device, DescriptorPool& descriptor_pool, + UpdateDescriptorQueue& update_descriptor_queue, + Common::ThreadWorker* worker_thread, + PipelineStatistics* pipeline_statistics, + RenderPassCache& render_pass_cache, + const GraphicsPipelineCacheKey& key, + std::array<vk::ShaderModule, NUM_STAGES> stages, + const std::array<const Shader::Info*, NUM_STAGES>& infos); GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete; GraphicsPipeline(GraphicsPipeline&&) noexcept = delete; @@ -110,6 +110,11 @@ public: return [](GraphicsPipeline* pl, bool is_indexed) { pl->ConfigureImpl<Spec>(is_indexed); }; } + void SetEngine(Tegra::Engines::Maxwell3D* maxwell3d_, Tegra::MemoryManager* gpu_memory_) { + maxwell3d = maxwell3d_; + gpu_memory = gpu_memory_; + } + private: template <typename Spec> void ConfigureImpl(bool is_indexed); @@ -121,13 +126,13 @@ private: void Validate(); const GraphicsPipelineCacheKey key; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D* maxwell3d; + Tegra::MemoryManager* gpu_memory; const Device& device; TextureCache& texture_cache; BufferCache& buffer_cache; - VKScheduler& scheduler; - VKUpdateDescriptorQueue& update_descriptor_queue; + Scheduler& scheduler; + UpdateDescriptorQueue& update_descriptor_queue; void (*configure_func)(GraphicsPipeline*, bool){}; diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp index 6852c11b0..4e81d3d28 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <thread> diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h index 9be9c9bed..362ed579a 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.h +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index a633b73e5..732e7b6f2 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <cstddef> @@ -16,13 +15,11 @@ #include "common/microprofile.h" #include "common/thread_worker.h" #include "core/core.h" -#include "core/memory.h" #include "shader_recompiler/backend/spirv/emit_spirv.h" #include "shader_recompiler/environment.h" #include "shader_recompiler/frontend/maxwell/control_flow.h" #include "shader_recompiler/frontend/maxwell/translate_program.h" #include "shader_recompiler/program_header.h" -#include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" @@ -56,7 +53,7 @@ using VideoCommon::FileEnvironment; using VideoCommon::GenericEnvironment; using VideoCommon::GraphicsEnvironment; -constexpr u32 CACHE_VERSION = 5; +constexpr u32 CACHE_VERSION = 6; template <typename Container> auto MakeSpan(Container& container) { @@ -177,7 +174,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> program case Maxwell::TessellationPrimitive::Quads: return Shader::TessPrimitive::Quads; } - UNREACHABLE(); + ASSERT(false); return Shader::TessPrimitive::Triangles; }(); info.tess_spacing = [&] { @@ -190,7 +187,7 @@ Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> program case Maxwell::TessellationSpacing::FractionalEven: return Shader::TessSpacing::FractionalEven; } - UNREACHABLE(); + ASSERT(false); return Shader::TessSpacing::Equal; }(); break; @@ -262,20 +259,18 @@ bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) c return std::memcmp(&rhs, this, Size()) == 0; } -PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - VKScheduler& scheduler_, DescriptorPool& descriptor_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_, +PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_, + Scheduler& scheduler_, DescriptorPool& descriptor_pool_, + UpdateDescriptorQueue& update_descriptor_queue_, RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_, TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_) - : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, - device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, - update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_}, - buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_}, + : VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_}, + descriptor_pool{descriptor_pool_}, update_descriptor_queue{update_descriptor_queue_}, + render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_}, + texture_cache{texture_cache_}, shader_notify{shader_notify_}, use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, - workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"), - serialization_thread(1, "yuzu:PipelineSerialization") { + workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "VkPipelineBuilder"), + serialization_thread(1, "VkPipelineSerialization") { const auto& float_control{device.FloatControlProperties()}; const VkDriverIdKHR driver_id{device.GetDriverID()}; profile = Shader::Profile{ @@ -340,7 +335,7 @@ GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() { current_pipeline = nullptr; return nullptr; } - graphics_key.state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported(), + graphics_key.state.Refresh(*maxwell3d, device.IsExtExtendedDynamicStateSupported(), device.IsExtVertexInputDynamicStateSupported()); if (current_pipeline) { @@ -360,7 +355,7 @@ ComputePipeline* PipelineCache::CurrentComputePipeline() { if (!shader) { return nullptr; } - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const ComputePipelineCacheKey key{ .unique_hash = shader->unique_hash, .shared_memory_size = qmd.shared_alloc, @@ -406,7 +401,7 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading workers.QueueWork([this, key, env = std::move(env), &state, &callback]() mutable { ShaderPools pools; auto pipeline{CreateComputePipeline(pools, key, env, state.statistics.get(), false)}; - std::lock_guard lock{state.mutex}; + std::scoped_lock lock{state.mutex}; if (pipeline) { compute_cache.emplace(key, std::move(pipeline)); } @@ -436,8 +431,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs), state.statistics.get(), false)}; - std::lock_guard lock{state.mutex}; - graphics_cache.emplace(key, std::move(pipeline)); + std::scoped_lock lock{state.mutex}; + if (pipeline) { + graphics_cache.emplace(key, std::move(pipeline)); + } ++state.built; if (state.has_loaded) { callback(VideoCore::LoadCallbackStage::Build, state.built, state.total); @@ -455,7 +452,7 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading state.has_loaded = true; lock.unlock(); - workers.WaitForRequests(); + workers.WaitForRequests(stop_loading); if (state.statistics) { state.statistics->Report(); @@ -487,13 +484,13 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const } // If something is using depth, we can assume that games are not rendering anything which // will be used one time. - if (maxwell3d.regs.zeta_enable) { + if (maxwell3d->regs.zeta_enable) { return nullptr; } // If games are using a small index count, we can assume these are full screen quads. // Usually these shaders are only used once for building textures so we can assume they // can't be built async - if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + if (maxwell3d->regs.index_array.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) { return pipeline; } return nullptr; @@ -558,10 +555,10 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( previous_stage = &program; } Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; - return std::make_unique<GraphicsPipeline>( - maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device, - descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key, - std::move(modules), infos); + return std::make_unique<GraphicsPipeline>(scheduler, buffer_cache, texture_cache, + &shader_notify, device, descriptor_pool, + update_descriptor_queue, thread_worker, statistics, + render_pass_cache, key, std::move(modules), infos); } catch (const Shader::Exception& exception) { LOG_ERROR(Render_Vulkan, "{}", exception.what()); @@ -593,9 +590,9 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() { std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline( const ComputePipelineCacheKey& key, const ShaderInfo* shader) { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; env.SetCachedSize(shader->size_bytes); main_pools.ReleaseContents(); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index 4c135b5dd..61f9e9366 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -1,17 +1,14 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> #include <cstddef> #include <filesystem> -#include <iosfwd> #include <memory> #include <type_traits> #include <unordered_map> -#include <utility> #include <vector> #include "common/common_types.h" @@ -29,7 +26,6 @@ #include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/shader_cache.h" -#include "video_core/vulkan_common/vulkan_wrapper.h" namespace Core { class System; @@ -85,8 +81,8 @@ class Device; class PipelineStatistics; class RasterizerVulkan; class RenderPassCache; -class VKScheduler; -class VKUpdateDescriptorQueue; +class Scheduler; +class UpdateDescriptorQueue; using VideoCommon::ShaderInfo; @@ -104,11 +100,9 @@ struct ShaderPools { class PipelineCache : public VideoCommon::ShaderCache { public: - explicit PipelineCache(RasterizerVulkan& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::KeplerCompute& kepler_compute, - Tegra::MemoryManager& gpu_memory, const Device& device, - VKScheduler& scheduler, DescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue, + explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler, + DescriptorPool& descriptor_pool, + UpdateDescriptorQueue& update_descriptor_queue, RenderPassCache& render_pass_cache, BufferCache& buffer_cache, TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_); ~PipelineCache(); @@ -142,9 +136,9 @@ private: bool build_in_parallel); const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; DescriptorPool& descriptor_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; + UpdateDescriptorQueue& update_descriptor_queue; RenderPassCache& render_pass_cache; BufferCache& buffer_cache; TextureCache& texture_cache; diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 259cba156..7cb02631c 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <cstddef> @@ -27,7 +26,7 @@ constexpr VkQueryType GetTarget(QueryType type) { } // Anonymous namespace -QueryPool::QueryPool(const Device& device_, VKScheduler& scheduler, QueryType type_) +QueryPool::QueryPool(const Device& device_, Scheduler& scheduler, QueryType type_) : ResourcePool{scheduler.GetMasterSemaphore(), GROW_STEP}, device{device_}, type{type_} {} QueryPool::~QueryPool() = default; @@ -66,15 +65,14 @@ void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) { usage[pool_index * GROW_STEP + static_cast<std::ptrdiff_t>(query.second)] = false; } -VKQueryCache::VKQueryCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, - const Device& device_, VKScheduler& scheduler_) - : QueryCacheBase{rasterizer_, maxwell3d_, gpu_memory_}, device{device_}, scheduler{scheduler_}, +QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_, + Scheduler& scheduler_) + : QueryCacheBase{rasterizer_}, device{device_}, scheduler{scheduler_}, query_pools{ QueryPool{device_, scheduler_, QueryType::SamplesPassed}, } {} -VKQueryCache::~VKQueryCache() { +QueryCache::~QueryCache() { // TODO(Rodrigo): This is a hack to destroy all HostCounter instances before the base class // destructor is called. The query cache should be redesigned to have a proper ownership model // instead of using shared pointers. @@ -85,15 +83,15 @@ VKQueryCache::~VKQueryCache() { } } -std::pair<VkQueryPool, u32> VKQueryCache::AllocateQuery(QueryType type) { +std::pair<VkQueryPool, u32> QueryCache::AllocateQuery(QueryType type) { return query_pools[static_cast<std::size_t>(type)].Commit(); } -void VKQueryCache::Reserve(QueryType type, std::pair<VkQueryPool, u32> query) { +void QueryCache::Reserve(QueryType type, std::pair<VkQueryPool, u32> query) { query_pools[static_cast<std::size_t>(type)].Reserve(query); } -HostCounter::HostCounter(VKQueryCache& cache_, std::shared_ptr<HostCounter> dependency_, +HostCounter::HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_, QueryType type_) : HostCounterBase{std::move(dependency_)}, cache{cache_}, type{type_}, query{cache_.AllocateQuery(type_)}, tick{cache_.GetScheduler().CurrentTick()} { diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h index 7190946b9..26762ee09 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.h +++ b/src/video_core/renderer_vulkan/vk_query_cache.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -23,14 +22,14 @@ namespace Vulkan { class CachedQuery; class Device; class HostCounter; -class VKQueryCache; -class VKScheduler; +class QueryCache; +class Scheduler; -using CounterStream = VideoCommon::CounterStreamBase<VKQueryCache, HostCounter>; +using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>; class QueryPool final : public ResourcePool { public: - explicit QueryPool(const Device& device, VKScheduler& scheduler, VideoCore::QueryType type); + explicit QueryPool(const Device& device, Scheduler& scheduler, VideoCore::QueryType type); ~QueryPool() override; std::pair<VkQueryPool, u32> Commit(); @@ -50,13 +49,12 @@ private: std::vector<bool> usage; }; -class VKQueryCache final - : public VideoCommon::QueryCacheBase<VKQueryCache, CachedQuery, CounterStream, HostCounter> { +class QueryCache final + : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit VKQueryCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, - const Device& device_, VKScheduler& scheduler_); - ~VKQueryCache(); + explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_, + Scheduler& scheduler_); + ~QueryCache(); std::pair<VkQueryPool, u32> AllocateQuery(VideoCore::QueryType type); @@ -66,19 +64,19 @@ public: return device; } - VKScheduler& GetScheduler() const noexcept { + Scheduler& GetScheduler() const noexcept { return scheduler; } private: const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; std::array<QueryPool, VideoCore::NumQueryTypes> query_pools; }; -class HostCounter final : public VideoCommon::HostCounterBase<VKQueryCache, HostCounter> { +class HostCounter final : public VideoCommon::HostCounterBase<QueryCache, HostCounter> { public: - explicit HostCounter(VKQueryCache& cache_, std::shared_ptr<HostCounter> dependency_, + explicit HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_, VideoCore::QueryType type_); ~HostCounter(); @@ -87,7 +85,7 @@ public: private: u64 BlockingQuery() const override; - VKQueryCache& cache; + QueryCache& cache; const VideoCore::QueryType type; const std::pair<VkQueryPool, u32> query; const u64 tick; @@ -95,7 +93,7 @@ private: class CachedQuery : public VideoCommon::CachedQueryBase<HostCounter> { public: - explicit CachedQuery(VKQueryCache&, VideoCore::QueryType, VAddr cpu_addr_, u8* host_ptr_) + explicit CachedQuery(QueryCache&, VideoCore::QueryType, VAddr cpu_addr_, u8* host_ptr_) : CachedQueryBase{cpu_addr_, host_ptr_} {} }; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 2227d9197..acfd5da7d 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -1,20 +1,17 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <memory> #include <mutex> -#include <vector> -#include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" #include "common/microprofile.h" #include "common/scope_exit.h" #include "common/settings.h" -#include "core/core.h" +#include "video_core/control/channel_state.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/blit_image.h" @@ -73,10 +70,17 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in const float width = conv(src.scale_x * 2.0f); float y = conv(src.translate_y - src.scale_y); float height = conv(src.scale_y * 2.0f); - if (regs.screen_y_control.y_negate) { + bool y_negate = regs.screen_y_control.y_negate; + + if (!device.IsNvViewportSwizzleSupported()) { + y_negate = y_negate != (src.swizzle.y == Maxwell::ViewportSwizzle::NegativeY); + } + + if (y_negate) { y += height; height = -height; } + const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f; VkViewport viewport{ .x = x, @@ -145,14 +149,11 @@ DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instan } // Anonymous namespace RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, - Tegra::MemoryManager& gpu_memory_, - Core::Memory::Memory& cpu_memory_, VKScreenInfo& screen_info_, + Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_, const Device& device_, MemoryAllocator& memory_allocator_, - StateTracker& state_tracker_, VKScheduler& scheduler_) - : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, - gpu_memory{gpu_memory_}, maxwell3d{gpu.Maxwell3D()}, kepler_compute{gpu.KeplerCompute()}, - screen_info{screen_info_}, device{device_}, memory_allocator{memory_allocator_}, - state_tracker{state_tracker_}, scheduler{scheduler_}, + StateTracker& state_tracker_, Scheduler& scheduler_) + : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, screen_info{screen_info_}, device{device_}, + memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_}, staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler), update_descriptor_queue(device, scheduler), blit_image(device, scheduler, state_tracker, descriptor_pool), @@ -162,14 +163,13 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra memory_allocator, staging_pool, blit_image, astc_decoder_pass, render_pass_cache}, - texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), + texture_cache(texture_cache_runtime, *this), buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool, update_descriptor_queue, descriptor_pool), - buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - pipeline_cache(*this, maxwell3d, kepler_compute, gpu_memory, device, scheduler, - descriptor_pool, update_descriptor_queue, render_pass_cache, buffer_cache, - texture_cache, gpu.ShaderNotify()), - query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{buffer_cache}, + buffer_cache(*this, cpu_memory_, buffer_cache_runtime), + pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue, + render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()), + query_cache{*this, device, scheduler}, accelerate_dma{buffer_cache}, fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler), wfi_event(device.GetLogical().CreateEvent()) { scheduler.SetQueryCache(query_cache); @@ -190,14 +190,16 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { return; } std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + // update engine as channel may be different. + pipeline->SetEngine(maxwell3d, gpu_memory); pipeline->Configure(is_indexed); BeginTransformFeedback(); UpdateDynamicStates(); - const auto& regs{maxwell3d.regs}; - const u32 num_instances{maxwell3d.mme_draw.instance_count}; + const auto& regs{maxwell3d->regs}; + const u32 num_instances{maxwell3d->mme_draw.instance_count}; const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_instanced, is_indexed)}; scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) { if (draw_params.is_indexed) { @@ -215,14 +217,14 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { void RasterizerVulkan::Clear() { MICROPROFILE_SCOPE(Vulkan_Clearing); - if (!maxwell3d.ShouldExecute()) { + if (!maxwell3d->ShouldExecute()) { return; } FlushWork(); query_cache.UpdateCounters(); - auto& regs = maxwell3d.regs; + auto& regs = maxwell3d->regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; const bool use_depth = regs.clear_buffers.Z; @@ -245,8 +247,15 @@ void RasterizerVulkan::Clear() { } UpdateViewportsState(regs); + VkRect2D default_scissor; + default_scissor.offset.x = 0; + default_scissor.offset.y = 0; + default_scissor.extent.width = std::numeric_limits<s32>::max(); + default_scissor.extent.height = std::numeric_limits<s32>::max(); + VkClearRect clear_rect{ - .rect = GetScissorState(regs, 0, up_scale, down_shift), + .rect = regs.clear_flags.scissor ? GetScissorState(regs, 0, up_scale, down_shift) + : default_scissor, .baseArrayLayer = regs.clear_buffers.layer, .layerCount = 1, }; @@ -336,9 +345,9 @@ void RasterizerVulkan::DispatchCompute() { return; } std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex}; - pipeline->Configure(kepler_compute, gpu_memory, scheduler, buffer_cache, texture_cache); + pipeline->Configure(*kepler_compute, *gpu_memory, scheduler, buffer_cache, texture_cache); - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z}; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); }); @@ -419,7 +428,7 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { } } -void RasterizerVulkan::SyncGuestHost() { +void RasterizerVulkan::InvalidateGPUCache() { pipeline_cache.SyncGuestHost(); { std::scoped_lock lock{buffer_cache.mutex}; @@ -439,40 +448,30 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { pipeline_cache.OnCPUWrite(addr, size); } -void RasterizerVulkan::ModifyGPUMemory(GPUVAddr addr, u64 size) { +void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { { std::scoped_lock lock{texture_cache.mutex}; - texture_cache.UnmapGPUMemory(addr, size); + texture_cache.UnmapGPUMemory(as_id, addr, size); } } -void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) { - if (!gpu.IsAsync()) { - gpu_memory.Write<u32>(addr, value); - return; - } - fence_manager.SignalSemaphore(addr, value); +void RasterizerVulkan::SignalFence(std::function<void()>&& func) { + fence_manager.SignalFence(std::move(func)); +} + +void RasterizerVulkan::SyncOperation(std::function<void()>&& func) { + fence_manager.SyncOperation(std::move(func)); } void RasterizerVulkan::SignalSyncPoint(u32 value) { - if (!gpu.IsAsync()) { - gpu.IncrementSyncPoint(value); - return; - } fence_manager.SignalSyncPoint(value); } void RasterizerVulkan::SignalReference() { - if (!gpu.IsAsync()) { - return; - } fence_manager.SignalOrdering(); } void RasterizerVulkan::ReleaseFences() { - if (!gpu.IsAsync()) { - return; - } fence_manager.WaitPendingFences(); } @@ -549,13 +548,13 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerVulkan::AccessAccelerateDMA() } void RasterizerVulkan::AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) { - auto cpu_addr = gpu_memory.GpuToCpuAddress(address); + std::span<const u8> memory) { + auto cpu_addr = gpu_memory->GpuToCpuAddress(address); if (!cpu_addr) [[unlikely]] { - gpu_memory.WriteBlock(address, memory.data(), copy_size); + gpu_memory->WriteBlock(address, memory.data(), copy_size); return; } - gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size); + gpu_memory->WriteBlockUnsafe(address, memory.data(), copy_size); { std::unique_lock<std::mutex> lock{buffer_cache.mutex}; if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) { @@ -624,7 +623,7 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 } void RasterizerVulkan::UpdateDynamicStates() { - auto& regs = maxwell3d.regs; + auto& regs = maxwell3d->regs; UpdateViewportsState(regs); UpdateScissorsState(regs); UpdateDepthBias(regs); @@ -648,7 +647,7 @@ void RasterizerVulkan::UpdateDynamicStates() { } void RasterizerVulkan::BeginTransformFeedback() { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -664,7 +663,7 @@ void RasterizerVulkan::BeginTransformFeedback() { } void RasterizerVulkan::EndTransformFeedback() { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -788,8 +787,8 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs) }); } else { // Front face defines both faces - scheduler.Record([ref = regs.stencil_back_func_ref, write_mask = regs.stencil_back_mask, - test_mask = regs.stencil_back_func_mask](vk::CommandBuffer cmdbuf) { + scheduler.Record([ref = regs.stencil_front_func_ref, write_mask = regs.stencil_front_mask, + test_mask = regs.stencil_front_func_mask](vk::CommandBuffer cmdbuf) { cmdbuf.SetStencilReference(VK_STENCIL_FACE_FRONT_AND_BACK, ref); cmdbuf.SetStencilWriteMask(VK_STENCIL_FACE_FRONT_AND_BACK, write_mask); cmdbuf.SetStencilCompareMask(VK_STENCIL_FACE_FRONT_AND_BACK, test_mask); @@ -914,7 +913,7 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& } void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) { - auto& dirty{maxwell3d.dirty.flags}; + auto& dirty{maxwell3d->dirty.flags}; if (!dirty[Dirty::VertexInput]) { return; } @@ -943,7 +942,7 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) .pNext = nullptr, .location = static_cast<u32>(index), .binding = binding, - .format = MaxwellToVK::VertexFormat(attribute.type, attribute.size), + .format = MaxwellToVK::VertexFormat(device, attribute.type, attribute.size), .offset = attribute.offset, }); } @@ -971,4 +970,41 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) }); } +void RasterizerVulkan::InitializeChannel(Tegra::Control::ChannelState& channel) { + CreateChannel(channel); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CreateChannel(channel); + buffer_cache.CreateChannel(channel); + } + pipeline_cache.CreateChannel(channel); + query_cache.CreateChannel(channel); + state_tracker.SetupTables(channel); +} + +void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) { + const s32 channel_id = channel.bind_id; + BindToChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.BindToChannel(channel_id); + buffer_cache.BindToChannel(channel_id); + } + pipeline_cache.BindToChannel(channel_id); + query_cache.BindToChannel(channel_id); + state_tracker.ChangeChannel(channel); + state_tracker.InvalidateState(); +} + +void RasterizerVulkan::ReleaseChannel(s32 channel_id) { + EraseChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.EraseChannel(channel_id); + buffer_cache.EraseChannel(channel_id); + } + pipeline_cache.EraseChannel(channel_id); + query_cache.EraseChannel(channel_id); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 5af2e275b..4cde3c983 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -1,30 +1,24 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> -#include <bitset> -#include <memory> -#include <utility> -#include <vector> #include <boost/container/static_vector.hpp> #include "common/common_types.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/engines/maxwell_dma.h" #include "video_core/rasterizer_accelerated.h" #include "video_core/rasterizer_interface.h" #include "video_core/renderer_vulkan/blit_image.h" -#include "video_core/renderer_vulkan/fixed_pipeline_state.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_fence_manager.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_query_cache.h" #include "video_core/renderer_vulkan/vk_render_pass_cache.h" -#include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" @@ -45,7 +39,7 @@ class Maxwell3D; namespace Vulkan { -struct VKScreenInfo; +struct ScreenInfo; class StateTracker; @@ -61,13 +55,13 @@ private: BufferCache& buffer_cache; }; -class RasterizerVulkan final : public VideoCore::RasterizerAccelerated { +class RasterizerVulkan final : public VideoCore::RasterizerAccelerated, + protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - VKScreenInfo& screen_info_, const Device& device_, - MemoryAllocator& memory_allocator_, StateTracker& state_tracker_, - VKScheduler& scheduler_); + Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_, + const Device& device_, MemoryAllocator& memory_allocator_, + StateTracker& state_tracker_, Scheduler& scheduler_); ~RasterizerVulkan() override; void Draw(bool is_indexed, bool is_instanced) override; @@ -82,10 +76,11 @@ public: bool MustFlushRegion(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override; - void SyncGuestHost() override; + void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; - void ModifyGPUMemory(GPUVAddr addr, u64 size) override; - void SignalSemaphore(GPUVAddr addr, u32 value) override; + void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override; + void SignalFence(std::function<void()>&& func) override; + void SyncOperation(std::function<void()>&& func) override; void SignalSyncPoint(u32 value) override; void SignalReference() override; void ReleaseFences() override; @@ -100,12 +95,18 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) override; + std::span<const u8> memory) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; void LoadDiskResources(u64 title_id, std::stop_token stop_loading, const VideoCore::DiskResourceLoadCallback& callback) override; + void InitializeChannel(Tegra::Control::ChannelState& channel) override; + + void BindChannel(Tegra::Control::ChannelState& channel) override; + + void ReleaseChannel(s32 channel_id) override; + private: static constexpr size_t MAX_TEXTURES = 192; static constexpr size_t MAX_IMAGES = 48; @@ -141,19 +142,16 @@ private: void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs); Tegra::GPU& gpu; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - VKScreenInfo& screen_info; + ScreenInfo& screen_info; const Device& device; MemoryAllocator& memory_allocator; StateTracker& state_tracker; - VKScheduler& scheduler; + Scheduler& scheduler; StagingBufferPool staging_pool; DescriptorPool descriptor_pool; - VKUpdateDescriptorQueue update_descriptor_queue; + UpdateDescriptorQueue update_descriptor_queue; BlitImageHelper blit_image; ASTCDecoderPass astc_decoder_pass; RenderPassCache render_pass_cache; @@ -163,9 +161,9 @@ private: BufferCacheRuntime buffer_cache_runtime; BufferCache buffer_cache; PipelineCache pipeline_cache; - VKQueryCache query_cache; + QueryCache query_cache; AccelerateDMA accelerate_dma; - VKFenceManager fence_manager; + FenceManager fence_manager; vk::Event wfi_event; diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp index 451ffe019..ae9f1de64 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <unordered_map> @@ -36,7 +35,7 @@ VkAttachmentDescription AttachmentDescription(const Device& device, PixelFormat RenderPassCache::RenderPassCache(const Device& device_) : device{&device_} {} VkRenderPass RenderPassCache::Get(const RenderPassKey& key) { - std::lock_guard lock{mutex}; + std::scoped_lock lock{mutex}; const auto [pair, is_new] = cache.try_emplace(key); if (!is_new) { return *pair->second; diff --git a/src/video_core/renderer_vulkan/vk_render_pass_cache.h b/src/video_core/renderer_vulkan/vk_render_pass_cache.h index eaa0ed775..dc21b7e69 100644 --- a/src/video_core/renderer_vulkan/vk_render_pass_cache.h +++ b/src/video_core/renderer_vulkan/vk_render_pass_cache.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.cpp b/src/video_core/renderer_vulkan/vk_resource_pool.cpp index 2dd514968..6c8ac22f4 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_resource_pool.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <optional> diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.h b/src/video_core/renderer_vulkan/vk_resource_pool.h index f0b80ad59..a39a3b881 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.h +++ b/src/video_core/renderer_vulkan/vk_resource_pool.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 7d9d4f7ba..d96720b80 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -1,10 +1,8 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <memory> #include <mutex> -#include <optional> #include <thread> #include <utility> @@ -23,7 +21,7 @@ namespace Vulkan { MICROPROFILE_DECLARE(Vulkan_WaitForWorker); -void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) { +void Scheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) { auto command = first; while (command != nullptr) { auto next = command->GetNext(); @@ -37,7 +35,7 @@ void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) { last = nullptr; } -VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_) +Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_) : device{device_}, state_tracker{state_tracker_}, master_semaphore{std::make_unique<MasterSemaphore>(device)}, command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} { @@ -46,14 +44,14 @@ VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_) worker_thread = std::jthread([this](std::stop_token token) { WorkerThread(token); }); } -VKScheduler::~VKScheduler() = default; +Scheduler::~Scheduler() = default; -void VKScheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { +void Scheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { SubmitExecution(signal_semaphore, wait_semaphore); AllocateNewContext(); } -void VKScheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { +void Scheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { const u64 presubmit_tick = CurrentTick(); SubmitExecution(signal_semaphore, wait_semaphore); WaitWorker(); @@ -61,7 +59,7 @@ void VKScheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphor AllocateNewContext(); } -void VKScheduler::WaitWorker() { +void Scheduler::WaitWorker() { MICROPROFILE_SCOPE(Vulkan_WaitForWorker); DispatchWork(); @@ -69,19 +67,19 @@ void VKScheduler::WaitWorker() { wait_cv.wait(lock, [this] { return work_queue.empty(); }); } -void VKScheduler::DispatchWork() { +void Scheduler::DispatchWork() { if (chunk->Empty()) { return; } { - std::lock_guard lock{work_mutex}; + std::scoped_lock lock{work_mutex}; work_queue.push(std::move(chunk)); } work_cv.notify_one(); AcquireNewChunk(); } -void VKScheduler::RequestRenderpass(const Framebuffer* framebuffer) { +void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) { const VkRenderPass renderpass = framebuffer->RenderPass(); const VkFramebuffer framebuffer_handle = framebuffer->Handle(); const VkExtent2D render_area = framebuffer->RenderArea(); @@ -116,11 +114,11 @@ void VKScheduler::RequestRenderpass(const Framebuffer* framebuffer) { renderpass_image_ranges = framebuffer->ImageRanges(); } -void VKScheduler::RequestOutsideRenderPassOperationContext() { +void Scheduler::RequestOutsideRenderPassOperationContext() { EndRenderPass(); } -bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { +bool Scheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { if (state.graphics_pipeline == pipeline) { return false; } @@ -128,7 +126,7 @@ bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { return true; } -bool VKScheduler::UpdateRescaling(bool is_rescaling) { +bool Scheduler::UpdateRescaling(bool is_rescaling) { if (state.rescaling_defined && is_rescaling == state.is_rescaling) { return false; } @@ -137,8 +135,8 @@ bool VKScheduler::UpdateRescaling(bool is_rescaling) { return true; } -void VKScheduler::WorkerThread(std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:VulkanWorker"); +void Scheduler::WorkerThread(std::stop_token stop_token) { + Common::SetCurrentThreadName("VulkanWorker"); do { std::unique_ptr<CommandChunk> work; { @@ -158,12 +156,12 @@ void VKScheduler::WorkerThread(std::stop_token stop_token) { if (has_submit) { AllocateWorkerCommandBuffer(); } - std::lock_guard reserve_lock{reserve_mutex}; + std::scoped_lock reserve_lock{reserve_mutex}; chunk_reserve.push_back(std::move(work)); } while (!stop_token.stop_requested()); } -void VKScheduler::AllocateWorkerCommandBuffer() { +void Scheduler::AllocateWorkerCommandBuffer() { current_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader()); current_cmdbuf.Begin({ .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO, @@ -173,7 +171,7 @@ void VKScheduler::AllocateWorkerCommandBuffer() { }); } -void VKScheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { +void Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { EndPendingOperations(); InvalidateState(); @@ -227,25 +225,25 @@ void VKScheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait DispatchWork(); } -void VKScheduler::AllocateNewContext() { +void Scheduler::AllocateNewContext() { // Enable counters once again. These are disabled when a command buffer is finished. if (query_cache) { query_cache->UpdateCounters(); } } -void VKScheduler::InvalidateState() { +void Scheduler::InvalidateState() { state.graphics_pipeline = nullptr; state.rescaling_defined = false; state_tracker.InvalidateCommandBufferState(); } -void VKScheduler::EndPendingOperations() { +void Scheduler::EndPendingOperations() { query_cache->DisableStreams(); EndRenderPass(); } -void VKScheduler::EndRenderPass() { +void Scheduler::EndRenderPass() { if (!state.renderpass) { return; } @@ -282,8 +280,8 @@ void VKScheduler::EndRenderPass() { num_renderpass_images = 0; } -void VKScheduler::AcquireNewChunk() { - std::lock_guard lock{reserve_mutex}; +void Scheduler::AcquireNewChunk() { + std::scoped_lock lock{reserve_mutex}; if (chunk_reserve.empty()) { chunk = std::make_unique<CommandChunk>(); return; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index e69aa136b..c04aad08f 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -1,10 +1,8 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <atomic> #include <condition_variable> #include <cstddef> #include <memory> @@ -24,14 +22,14 @@ class Device; class Framebuffer; class GraphicsPipeline; class StateTracker; -class VKQueryCache; +class QueryCache; /// The scheduler abstracts command buffer and fence management with an interface that's able to do /// OpenGL-like operations on Vulkan command buffers. -class VKScheduler { +class Scheduler { public: - explicit VKScheduler(const Device& device, StateTracker& state_tracker); - ~VKScheduler(); + explicit Scheduler(const Device& device, StateTracker& state_tracker); + ~Scheduler(); /// Sends the current execution context to the GPU. void Flush(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr); @@ -63,7 +61,7 @@ public: void InvalidateState(); /// Assigns the query cache. - void SetQueryCache(VKQueryCache& query_cache_) { + void SetQueryCache(QueryCache& query_cache_) { query_cache = &query_cache_; } @@ -214,7 +212,7 @@ private: std::unique_ptr<MasterSemaphore> master_semaphore; std::unique_ptr<CommandPool> command_pool; - VKQueryCache* query_cache = nullptr; + QueryCache* query_cache = nullptr; vk::CommandBuffer current_cmdbuf; diff --git a/src/video_core/renderer_vulkan/vk_shader_util.cpp b/src/video_core/renderer_vulkan/vk_shader_util.cpp index aaad4f292..7a0a2b154 100644 --- a/src/video_core/renderer_vulkan/vk_shader_util.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_util.cpp @@ -1,11 +1,8 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <cstring> -#include <memory> -#include "common/assert.h" #include "common/common_types.h" #include "video_core/renderer_vulkan/vk_shader_util.h" #include "video_core/vulkan_common/vulkan_device.h" diff --git a/src/video_core/renderer_vulkan/vk_shader_util.h b/src/video_core/renderer_vulkan/vk_shader_util.h index 9517cbe84..2f7c9f25c 100644 --- a/src/video_core/renderer_vulkan/vk_shader_util.h +++ b/src/video_core/renderer_vulkan/vk_shader_util.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 5d5329abf..7fb256953 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <utility> @@ -27,20 +26,39 @@ using namespace Common::Literals; constexpr VkDeviceSize MAX_ALIGNMENT = 256; // Maximum size to put elements in the stream buffer constexpr VkDeviceSize MAX_STREAM_BUFFER_REQUEST_SIZE = 8_MiB; -// Stream buffer size in bytes -constexpr VkDeviceSize STREAM_BUFFER_SIZE = 128_MiB; -constexpr VkDeviceSize REGION_SIZE = STREAM_BUFFER_SIZE / StagingBufferPool::NUM_SYNCS; constexpr VkMemoryPropertyFlags HOST_FLAGS = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; constexpr VkMemoryPropertyFlags STREAM_FLAGS = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | HOST_FLAGS; -bool IsStreamHeap(VkMemoryHeap heap) noexcept { - return STREAM_BUFFER_SIZE < (heap.size * 2) / 3; +static bool IsStreamHeap(VkMemoryHeap heap, size_t staging_buffer_size) noexcept { + return staging_buffer_size < (heap.size * 2) / 3; +} + +static bool HasLargeDeviceLocalHostVisibleMemory(const VkPhysicalDeviceMemoryProperties& props) { + const auto flags{VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT}; + + for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) { + const auto& memory_type{props.memoryTypes[type_index]}; + + if ((memory_type.propertyFlags & flags) != flags) { + // Memory must be device local and host visible + continue; + } + + const auto& heap{props.memoryHeaps[memory_type.heapIndex]}; + if (heap.size >= 7168_MiB) { + // This is the right type of memory + return true; + } + } + + return false; } std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask, - VkMemoryPropertyFlags flags) noexcept { + VkMemoryPropertyFlags flags, + size_t staging_buffer_size) noexcept { for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) { if (((type_mask >> type_index) & 1) == 0) { // Memory type is incompatible @@ -51,7 +69,7 @@ std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& p // Memory type doesn't have the flags we want continue; } - if (!IsStreamHeap(props.memoryHeaps[memory_type.heapIndex])) { + if (!IsStreamHeap(props.memoryHeaps[memory_type.heapIndex], staging_buffer_size)) { // Memory heap is not suitable for streaming continue; } @@ -62,17 +80,17 @@ std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& p } u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask, - bool try_device_local) { + bool try_device_local, size_t staging_buffer_size) { std::optional<u32> type; if (try_device_local) { // Try to find a DEVICE_LOCAL_BIT type, Nvidia and AMD have a dedicated heap for this - type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS); + type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS, staging_buffer_size); if (type) { return *type; } } // Otherwise try without the DEVICE_LOCAL_BIT - type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS); + type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS, staging_buffer_size); if (type) { return *type; } @@ -80,20 +98,32 @@ u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_ throw vk::Exception(VK_ERROR_OUT_OF_DEVICE_MEMORY); } -size_t Region(size_t iterator) noexcept { - return iterator / REGION_SIZE; +size_t Region(size_t iterator, size_t region_size) noexcept { + return iterator / region_size; } } // Anonymous namespace StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_, - VKScheduler& scheduler_) + Scheduler& scheduler_) : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} { + + const auto memory_properties{device.GetPhysical().GetMemoryProperties().memoryProperties}; + if (HasLargeDeviceLocalHostVisibleMemory(memory_properties)) { + // Possible on many integrated and newer discrete cards + staging_buffer_size = 1_GiB; + } else { + // Well-supported default size used by most Vulkan PC games + staging_buffer_size = 256_MiB; + } + + region_size = staging_buffer_size / StagingBufferPool::NUM_SYNCS; + const vk::Device& dev = device.GetLogical(); stream_buffer = dev.CreateBuffer(VkBufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .size = STREAM_BUFFER_SIZE, + .size = staging_buffer_size, .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, @@ -118,19 +148,18 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem .image = nullptr, .buffer = *stream_buffer, }; - const auto memory_properties = device.GetPhysical().GetMemoryProperties(); VkMemoryAllocateInfo stream_memory_info{ .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, .pNext = make_dedicated ? &dedicated_info : nullptr, .allocationSize = requirements.size, - .memoryTypeIndex = - FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, true), + .memoryTypeIndex = FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, true, + staging_buffer_size), }; stream_memory = dev.TryAllocateMemory(stream_memory_info); if (!stream_memory) { LOG_INFO(Render_Vulkan, "Dynamic memory allocation failed, trying with system memory"); - stream_memory_info.memoryTypeIndex = - FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, false); + stream_memory_info.memoryTypeIndex = FindMemoryTypeIndex( + memory_properties, requirements.memoryTypeBits, false, staging_buffer_size); stream_memory = dev.AllocateMemory(stream_memory_info); } @@ -138,7 +167,7 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem stream_memory.SetObjectNameEXT("Stream Buffer Memory"); } stream_buffer.BindMemory(*stream_memory, 0); - stream_pointer = stream_memory.Map(0, STREAM_BUFFER_SIZE); + stream_pointer = stream_memory.Map(0, staging_buffer_size); } StagingBufferPool::~StagingBufferPool() = default; @@ -159,25 +188,25 @@ void StagingBufferPool::TickFrame() { } StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { - if (AreRegionsActive(Region(free_iterator) + 1, - std::min(Region(iterator + size) + 1, NUM_SYNCS))) { + if (AreRegionsActive(Region(free_iterator, region_size) + 1, + std::min(Region(iterator + size, region_size) + 1, NUM_SYNCS))) { // Avoid waiting for the previous usages to be free return GetStagingBuffer(size, MemoryUsage::Upload); } const u64 current_tick = scheduler.CurrentTick(); - std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + Region(iterator), - current_tick); + std::fill(sync_ticks.begin() + Region(used_iterator, region_size), + sync_ticks.begin() + Region(iterator, region_size), current_tick); used_iterator = iterator; free_iterator = std::max(free_iterator, iterator + size); - if (iterator + size >= STREAM_BUFFER_SIZE) { - std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + NUM_SYNCS, - current_tick); + if (iterator + size >= staging_buffer_size) { + std::fill(sync_ticks.begin() + Region(used_iterator, region_size), + sync_ticks.begin() + NUM_SYNCS, current_tick); used_iterator = 0; iterator = 0; free_iterator = size; - if (AreRegionsActive(0, Region(size) + 1)) { + if (AreRegionsActive(0, Region(size, region_size) + 1)) { // Avoid waiting for the previous usages to be free return GetStagingBuffer(size, MemoryUsage::Upload); } @@ -264,7 +293,7 @@ StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage case MemoryUsage::Download: return download_cache; default: - UNREACHABLE_MSG("Invalid memory usage={}", usage); + ASSERT_MSG(false, "Invalid memory usage={}", usage); return upload_cache; } } diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h index 69f7618de..90c67177f 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -15,7 +14,7 @@ namespace Vulkan { class Device; -class VKScheduler; +class Scheduler; struct StagingBufferRef { VkBuffer buffer; @@ -28,7 +27,7 @@ public: static constexpr size_t NUM_SYNCS = 16; explicit StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator, - VKScheduler& scheduler); + Scheduler& scheduler); ~StagingBufferPool(); StagingBufferRef Request(size_t size, MemoryUsage usage); @@ -83,7 +82,7 @@ private: const Device& device; MemoryAllocator& memory_allocator; - VKScheduler& scheduler; + Scheduler& scheduler; vk::Buffer stream_buffer; vk::DeviceMemory stream_memory; @@ -94,6 +93,9 @@ private: size_t free_iterator = 0; std::array<u64, NUM_SYNCS> sync_ticks{}; + size_t staging_buffer_size = 0; + size_t region_size = 0; + StagingBuffersCache device_local_cache; StagingBuffersCache upload_cache; StagingBuffersCache download_cache; diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index c00913f55..f234e1a31 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -1,17 +1,15 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <cstddef> -#include <iterator> #include "common/common_types.h" #include "core/core.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" #include "video_core/renderer_vulkan/vk_state_tracker.h" #define OFF(field_name) MAXWELL3D_REG_INDEX(field_name) @@ -176,9 +174,8 @@ void SetupDirtyVertexBindings(Tables& tables) { } } // Anonymous namespace -StateTracker::StateTracker(Tegra::GPU& gpu) - : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} { - auto& tables{gpu.Maxwell3D().dirty.tables}; +void StateTracker::SetupTables(Tegra::Control::ChannelState& channel_state) { + auto& tables{channel_state.maxwell_3d->dirty.tables}; SetupDirtyFlags(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); @@ -201,4 +198,15 @@ StateTracker::StateTracker(Tegra::GPU& gpu) SetupDirtyVertexBindings(tables); } +void StateTracker::ChangeChannel(Tegra::Control::ChannelState& channel_state) { + flags = &channel_state.maxwell_3d->dirty.flags; +} + +void StateTracker::InvalidateState() { + flags->set(); +} + +StateTracker::StateTracker() + : flags{&default_flags}, default_flags{}, invalidation_flags{MakeInvalidationFlags()} {} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 40a149832..2296dea60 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -8,10 +7,15 @@ #include <limits> #include "common/common_types.h" -#include "core/core.h" #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" +namespace Tegra { +namespace Control { +struct ChannelState; +} +} // namespace Tegra + namespace Vulkan { namespace Dirty { @@ -55,19 +59,19 @@ class StateTracker { using Maxwell = Tegra::Engines::Maxwell3D::Regs; public: - explicit StateTracker(Tegra::GPU& gpu); + explicit StateTracker(); void InvalidateCommandBufferState() { - flags |= invalidation_flags; + (*flags) |= invalidation_flags; current_topology = INVALID_TOPOLOGY; } void InvalidateViewports() { - flags[Dirty::Viewports] = true; + (*flags)[Dirty::Viewports] = true; } void InvalidateScissors() { - flags[Dirty::Scissors] = true; + (*flags)[Dirty::Scissors] = true; } bool TouchViewports() { @@ -141,16 +145,23 @@ public: return has_changed; } + void SetupTables(Tegra::Control::ChannelState& channel_state); + + void ChangeChannel(Tegra::Control::ChannelState& channel_state); + + void InvalidateState(); + private: static constexpr auto INVALID_TOPOLOGY = static_cast<Maxwell::PrimitiveTopology>(~0u); bool Exchange(std::size_t id, bool new_value) const noexcept { - const bool is_dirty = flags[id]; - flags[id] = new_value; + const bool is_dirty = (*flags)[id]; + (*flags)[id] = new_value; return is_dirty; } - Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags* flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags default_flags; Tegra::Engines::Maxwell3D::DirtyState::Flags invalidation_flags; Maxwell::PrimitiveTopology current_topology = INVALID_TOPOLOGY; }; diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index 8972a6921..706d9ba74 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -1,17 +1,14 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> #include <limits> #include <vector> -#include "common/assert.h" #include "common/logging/log.h" #include "common/settings.h" #include "core/core.h" -#include "core/frontend/framebuffer_layout.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_swapchain.h" #include "video_core/vulkan_common/vulkan_device.h" @@ -36,12 +33,14 @@ VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats) } VkPresentModeKHR ChooseSwapPresentMode(vk::Span<VkPresentModeKHR> modes) { - // Mailbox doesn't lock the application like fifo (vsync), prefer it + // Mailbox (triple buffering) doesn't lock the application like fifo (vsync), + // prefer it if vsync option is not selected const auto found_mailbox = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); - if (found_mailbox != modes.end()) { + if (Settings::values.fullscreen_mode.GetValue() == Settings::FullscreenMode::Borderless && + found_mailbox != modes.end() && !Settings::values.use_vsync.GetValue()) { return VK_PRESENT_MODE_MAILBOX_KHR; } - if (Settings::values.disable_fps_limit.GetValue()) { + if (!Settings::values.use_speed_limit.GetValue()) { // FIFO present mode locks the framerate to the monitor's refresh rate, // Find an alternative to surpass this limitation if FPS is unlocked. const auto found_imm = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_IMMEDIATE_KHR); @@ -67,15 +66,15 @@ VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 wi } // Anonymous namespace -VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const Device& device_, VKScheduler& scheduler_, - u32 width, u32 height, bool srgb) +Swapchain::Swapchain(VkSurfaceKHR surface_, const Device& device_, Scheduler& scheduler_, u32 width, + u32 height, bool srgb) : surface{surface_}, device{device_}, scheduler{scheduler_} { Create(width, height, srgb); } -VKSwapchain::~VKSwapchain() = default; +Swapchain::~Swapchain() = default; -void VKSwapchain::Create(u32 width, u32 height, bool srgb) { +void Swapchain::Create(u32 width, u32 height, bool srgb) { is_outdated = false; is_suboptimal = false; @@ -96,7 +95,7 @@ void VKSwapchain::Create(u32 width, u32 height, bool srgb) { resource_ticks.resize(image_count); } -void VKSwapchain::AcquireNextImage() { +void Swapchain::AcquireNextImage() { const VkResult result = device.GetLogical().AcquireNextImageKHR( *swapchain, std::numeric_limits<u64>::max(), *present_semaphores[frame_index], VK_NULL_HANDLE, &image_index); @@ -117,7 +116,7 @@ void VKSwapchain::AcquireNextImage() { resource_ticks[image_index] = scheduler.CurrentTick(); } -void VKSwapchain::Present(VkSemaphore render_semaphore) { +void Swapchain::Present(VkSemaphore render_semaphore) { const auto present_queue{device.GetPresentQueue()}; const VkPresentInfoKHR present_info{ .sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR, @@ -148,8 +147,8 @@ void VKSwapchain::Present(VkSemaphore render_semaphore) { } } -void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, - u32 height, bool srgb) { +void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height, + bool srgb) { const auto physical_device{device.GetPhysical()}; const auto formats{physical_device.GetSurfaceFormatsKHR(surface)}; const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)}; @@ -158,8 +157,16 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, present_mode = ChooseSwapPresentMode(present_modes); u32 requested_image_count{capabilities.minImageCount + 1}; - if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) { - requested_image_count = capabilities.maxImageCount; + // Ensure Tripple buffering if possible. + if (capabilities.maxImageCount > 0) { + if (requested_image_count > capabilities.maxImageCount) { + requested_image_count = capabilities.maxImageCount; + } else { + requested_image_count = + std::max(requested_image_count, std::min(3U, capabilities.maxImageCount)); + } + } else { + requested_image_count = std::max(requested_image_count, 3U); } VkSwapchainCreateInfoKHR swapchain_ci{ .sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR, @@ -208,20 +215,20 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, extent = swapchain_ci.imageExtent; current_srgb = srgb; - current_fps_unlocked = Settings::values.disable_fps_limit.GetValue(); + current_fps_unlocked = !Settings::values.use_speed_limit.GetValue(); images = swapchain.GetImages(); image_count = static_cast<u32>(images.size()); image_view_format = srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM; } -void VKSwapchain::CreateSemaphores() { +void Swapchain::CreateSemaphores() { present_semaphores.resize(image_count); std::ranges::generate(present_semaphores, [this] { return device.GetLogical().CreateSemaphore(); }); } -void VKSwapchain::CreateImageViews() { +void Swapchain::CreateImageViews() { VkImageViewCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = nullptr, @@ -253,7 +260,7 @@ void VKSwapchain::CreateImageViews() { } } -void VKSwapchain::Destroy() { +void Swapchain::Destroy() { frame_index = 0; present_semaphores.clear(); framebuffers.clear(); @@ -261,11 +268,11 @@ void VKSwapchain::Destroy() { swapchain.reset(); } -bool VKSwapchain::HasFpsUnlockChanged() const { - return current_fps_unlocked != Settings::values.disable_fps_limit.GetValue(); +bool Swapchain::HasFpsUnlockChanged() const { + return current_fps_unlocked != !Settings::values.use_speed_limit.GetValue(); } -bool VKSwapchain::NeedsPresentModeUpdate() const { +bool Swapchain::NeedsPresentModeUpdate() const { // Mailbox present mode is the ideal for all scenarios. If it is not available, // A different present mode is needed to support unlocked FPS above the monitor's refresh rate. return present_mode != VK_PRESENT_MODE_MAILBOX_KHR && HasFpsUnlockChanged(); diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index 61a6d959e..111b3902d 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -16,13 +15,13 @@ struct FramebufferLayout; namespace Vulkan { class Device; -class VKScheduler; +class Scheduler; -class VKSwapchain { +class Swapchain { public: - explicit VKSwapchain(VkSurfaceKHR surface, const Device& device, VKScheduler& scheduler, - u32 width, u32 height, bool srgb); - ~VKSwapchain(); + explicit Swapchain(VkSurfaceKHR surface, const Device& device, Scheduler& scheduler, u32 width, + u32 height, bool srgb); + ~Swapchain(); /// Creates (or recreates) the swapchain with a given size. void Create(u32 width, u32 height, bool srgb); @@ -95,7 +94,7 @@ private: const VkSurfaceKHR surface; const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; vk::SwapchainKHR swapchain; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 0ba56ff1e..305ad8aee 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -71,7 +70,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case ImageType::Buffer: break; } - UNREACHABLE_MSG("Invalid image type={}", type); + ASSERT_MSG(false, "Invalid image type={}", type); return {}; } @@ -88,7 +87,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case 16: return VK_SAMPLE_COUNT_16_BIT; default: - UNREACHABLE_MSG("Invalid number of samples={}", num_samples); + ASSERT_MSG(false, "Invalid number of samples={}", num_samples); return VK_SAMPLE_COUNT_1_BIT; } } @@ -108,7 +107,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; break; default: - UNREACHABLE_MSG("Invalid surface type"); + ASSERT_MSG(false, "Invalid surface type"); } } if (info.storage) { @@ -180,7 +179,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case VideoCore::Surface::SurfaceType::DepthStencil: return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; default: - UNREACHABLE_MSG("Invalid surface type"); + ASSERT_MSG(false, "Invalid surface type"); return VkImageAspectFlags{}; } } @@ -222,7 +221,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case SwizzleSource::OneInt: return VK_COMPONENT_SWIZZLE_ONE; } - UNREACHABLE_MSG("Invalid swizzle={}", swizzle); + ASSERT_MSG(false, "Invalid swizzle={}", swizzle); return VK_COMPONENT_SWIZZLE_ZERO; } @@ -231,6 +230,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case Shader::TextureType::Color1D: return VK_IMAGE_VIEW_TYPE_1D; case Shader::TextureType::Color2D: + case Shader::TextureType::Color2DRect: return VK_IMAGE_VIEW_TYPE_2D; case Shader::TextureType::ColorCube: return VK_IMAGE_VIEW_TYPE_CUBE; @@ -243,10 +243,10 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case Shader::TextureType::ColorArrayCube: return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY; case Shader::TextureType::Buffer: - UNREACHABLE_MSG("Texture buffers can't be image views"); + ASSERT_MSG(false, "Texture buffers can't be image views"); return VK_IMAGE_VIEW_TYPE_1D; } - UNREACHABLE_MSG("Invalid image view type={}", type); + ASSERT_MSG(false, "Invalid image view type={}", type); return VK_IMAGE_VIEW_TYPE_2D; } @@ -255,6 +255,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { case VideoCommon::ImageViewType::e1D: return VK_IMAGE_VIEW_TYPE_1D; case VideoCommon::ImageViewType::e2D: + case VideoCommon::ImageViewType::Rect: return VK_IMAGE_VIEW_TYPE_2D; case VideoCommon::ImageViewType::Cube: return VK_IMAGE_VIEW_TYPE_CUBE; @@ -266,14 +267,11 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { return VK_IMAGE_VIEW_TYPE_2D_ARRAY; case VideoCommon::ImageViewType::CubeArray: return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY; - case VideoCommon::ImageViewType::Rect: - UNIMPLEMENTED_MSG("Rect image view"); - return VK_IMAGE_VIEW_TYPE_2D; case VideoCommon::ImageViewType::Buffer: - UNREACHABLE_MSG("Texture buffers can't be image views"); + ASSERT_MSG(false, "Texture buffers can't be image views"); return VK_IMAGE_VIEW_TYPE_1D; } - UNREACHABLE_MSG("Invalid image view type={}", type); + ASSERT_MSG(false, "Invalid image view type={}", type); return VK_IMAGE_VIEW_TYPE_2D; } @@ -438,6 +436,32 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { } } +[[nodiscard]] SwizzleSource SwapGreenRed(SwizzleSource value) { + switch (value) { + case SwizzleSource::R: + return SwizzleSource::G; + case SwizzleSource::G: + return SwizzleSource::R; + default: + return value; + } +} + +[[nodiscard]] SwizzleSource SwapSpecial(SwizzleSource value) { + switch (value) { + case SwizzleSource::A: + return SwizzleSource::R; + case SwizzleSource::R: + return SwizzleSource::A; + case SwizzleSource::G: + return SwizzleSource::B; + case SwizzleSource::B: + return SwizzleSource::G; + default: + return value; + } +} + void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image, VkImageAspectFlags aspect_mask, bool is_initialized, std::span<const VkBufferImageCopy> copies) { @@ -554,12 +578,25 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im }; } -[[nodiscard]] bool IsFormatFlipped(PixelFormat format) { +void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4>& swizzle, + bool emulate_bgr565) { switch (format) { case PixelFormat::A1B5G5R5_UNORM: - return true; + std::ranges::transform(swizzle, swizzle.begin(), SwapBlueRed); + break; + case PixelFormat::B5G6R5_UNORM: + if (emulate_bgr565) { + std::ranges::transform(swizzle, swizzle.begin(), SwapBlueRed); + } + break; + case PixelFormat::A5B5G5R1_UNORM: + std::ranges::transform(swizzle, swizzle.begin(), SwapSpecial); + break; + case PixelFormat::G4R4_UNORM: + std::ranges::transform(swizzle, swizzle.begin(), SwapGreenRed); + break; default: - return false; + break; } } @@ -606,11 +643,11 @@ struct RangedBarrierRange { case Shader::ImageFormat::R32G32B32A32_UINT: return VK_FORMAT_R32G32B32A32_UINT; } - UNREACHABLE_MSG("Invalid image format={}", format); + ASSERT_MSG(false, "Invalid image format={}", format); return VK_FORMAT_R32_UINT; } -void BlitScale(VKScheduler& scheduler, VkImage src_image, VkImage dst_image, const ImageInfo& info, +void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const ImageInfo& info, VkImageAspectFlags aspect_mask, const Settings::ResolutionScalingInfo& resolution, bool up_scaling = true) { const bool is_2d = info.type == ImageType::e2D; @@ -750,7 +787,7 @@ void BlitScale(VKScheduler& scheduler, VkImage src_image, VkImage dst_image, con } } // Anonymous namespace -TextureCacheRuntime::TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, +TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& scheduler_, MemoryAllocator& memory_allocator_, StagingBufferPool& staging_buffer_pool_, BlitImageHelper& blit_image_helper_, @@ -779,11 +816,6 @@ bool TextureCacheRuntime::ShouldReinterpret(Image& dst, Image& src) { !device.IsExtShaderStencilExportSupported()) { return true; } - if (VideoCore::Surface::GetFormatType(src.info.format) == - VideoCore::Surface::SurfaceType::DepthStencil && - !device.IsExtShaderStencilExportSupported()) { - return true; - } if (dst.info.format == PixelFormat::D32_FLOAT_S8_UINT || src.info.format == PixelFormat::D32_FLOAT_S8_UINT) { return true; @@ -1068,6 +1100,9 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) { return blit_image_helper.ConvertD24S8ToABGR8(dst, src_view); } + if (src_view.format == PixelFormat::D24_UNORM_S8_UINT) { + return blit_image_helper.ConvertS8D24ToABGR8(dst, src_view); + } break; case PixelFormat::R32_FLOAT: if (src_view.format == PixelFormat::D32_FLOAT) { @@ -1189,6 +1224,14 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const { return device.GetDeviceLocalMemory(); } +u64 TextureCacheRuntime::GetDeviceMemoryUsage() const { + return device.GetDeviceMemoryUsage(); +} + +bool TextureCacheRuntime::CanReportMemoryUsage() const { + return device.CanReportMemoryUsage(); +} + void TextureCacheRuntime::TickFrame() {} Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_, @@ -1203,6 +1246,7 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu } else { flags |= VideoCommon::ImageFlagBits::Converted; } + flags |= VideoCommon::ImageFlagBits::CostlyLoad; } if (runtime->device.HasDebuggingToolAttached()) { original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); @@ -1430,22 +1474,23 @@ bool Image::BlitScaleHelper(bool scale_up) { }; const VkExtent2D extent{ .width = std::max(scaled_width, info.size.width), - .height = std::max(scaled_height, info.size.width), + .height = std::max(scaled_height, info.size.height), }; auto* view_ptr = blit_view.get(); if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) { if (!blit_framebuffer) { - blit_framebuffer = std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent); + blit_framebuffer = + std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent, scale_up); } const auto color_view = blit_view->Handle(Shader::TextureType::Color2D); runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), color_view, dst_region, src_region, operation, BLIT_OPERATION); - } else if (!runtime->device.IsBlitDepthStencilSupported() && - aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { + } else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { if (!blit_framebuffer) { - blit_framebuffer = std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent); + blit_framebuffer = + std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent, scale_up); } runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), blit_view->DepthView(), blit_view->StencilView(), dst_region, @@ -1488,9 +1533,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI }; if (!info.IsRenderTarget()) { swizzle = info.Swizzle(); - if (IsFormatFlipped(format)) { - std::ranges::transform(swizzle, swizzle.begin(), SwapBlueRed); - } + TryTransformSwizzleIfNeeded(format, swizzle, device->MustEmulateBGR565()); if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) { std::ranges::transform(swizzle, swizzle.begin(), ConvertGreenRed); } @@ -1537,6 +1580,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI break; case VideoCommon::ImageViewType::e2D: case VideoCommon::ImageViewType::e2DArray: + case VideoCommon::ImageViewType::Rect: create(TextureType::Color2D, 1); create(TextureType::ColorArray2D, std::nullopt); render_target = Handle(Shader::TextureType::ColorArray2D); @@ -1550,11 +1594,8 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI create(TextureType::ColorCube, 6); create(TextureType::ColorArrayCube, std::nullopt); break; - case VideoCommon::ImageViewType::Rect: - UNIMPLEMENTED(); - break; case VideoCommon::ImageViewType::Buffer: - UNREACHABLE(); + ASSERT(false); break; } } @@ -1576,6 +1617,9 @@ ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParam ImageView::~ImageView() = default; VkImageView ImageView::DepthView() { + if (!image_handle) { + return VK_NULL_HANDLE; + } if (depth_view) { return *depth_view; } @@ -1585,6 +1629,9 @@ VkImageView ImageView::DepthView() { } VkImageView ImageView::StencilView() { + if (!image_handle) { + return VK_NULL_HANDLE; + } if (stencil_view) { return *stencil_view; } @@ -1594,6 +1641,9 @@ VkImageView ImageView::StencilView() { } VkImageView ImageView::ColorView() { + if (!image_handle) { + return VK_NULL_HANDLE; + } if (color_view) { return *color_view; } @@ -1603,6 +1653,9 @@ VkImageView ImageView::ColorView() { VkImageView ImageView::StorageView(Shader::TextureType texture_type, Shader::ImageFormat image_format) { + if (!image_handle) { + return VK_NULL_HANDLE; + } if (image_format == Shader::ImageFormat::Typeless) { return Handle(texture_type); } @@ -1705,34 +1758,42 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM .width = key.size.width, .height = key.size.height, }} { - CreateFramebuffer(runtime, color_buffers, depth_buffer); + CreateFramebuffer(runtime, color_buffers, depth_buffer, key.is_rescaled); if (runtime.device.HasDebuggingToolAttached()) { framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); } } Framebuffer::Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, - ImageView* depth_buffer, VkExtent2D extent) + ImageView* depth_buffer, VkExtent2D extent, bool is_rescaled) : render_area{extent} { std::array<ImageView*, NUM_RT> color_buffers{color_buffer}; - CreateFramebuffer(runtime, color_buffers, depth_buffer); + CreateFramebuffer(runtime, color_buffers, depth_buffer, is_rescaled); } Framebuffer::~Framebuffer() = default; void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, - ImageView* depth_buffer) { + ImageView* depth_buffer, bool is_rescaled) { std::vector<VkImageView> attachments; RenderPassKey renderpass_key{}; s32 num_layers = 1; + const auto& resolution = runtime.resolution; + + u32 width = 0; + u32 height = 0; for (size_t index = 0; index < NUM_RT; ++index) { const ImageView* const color_buffer = color_buffers[index]; if (!color_buffer) { renderpass_key.color_formats[index] = PixelFormat::Invalid; continue; } + width = std::max(width, is_rescaled ? resolution.ScaleUp(color_buffer->size.width) + : color_buffer->size.width); + height = std::max(height, is_rescaled ? resolution.ScaleUp(color_buffer->size.height) + : color_buffer->size.height); attachments.push_back(color_buffer->RenderTarget()); renderpass_key.color_formats[index] = color_buffer->format; num_layers = std::max(num_layers, color_buffer->range.extent.layers); @@ -1743,6 +1804,10 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, } const size_t num_colors = attachments.size(); if (depth_buffer) { + width = std::max(width, is_rescaled ? resolution.ScaleUp(depth_buffer->size.width) + : depth_buffer->size.width); + height = std::max(height, is_rescaled ? resolution.ScaleUp(depth_buffer->size.height) + : depth_buffer->size.height); attachments.push_back(depth_buffer->RenderTarget()); renderpass_key.depth_format = depth_buffer->format; num_layers = std::max(num_layers, depth_buffer->range.extent.layers); @@ -1759,6 +1824,8 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, renderpass_key.samples = samples; renderpass = runtime.render_pass_cache.Get(renderpass_key); + render_area.width = std::min(render_area.width, width); + render_area.height = std::min(render_area.height, height); num_color_buffers = static_cast<u32>(num_colors); framebuffer = runtime.device.GetLogical().CreateFramebuffer({ @@ -1780,7 +1847,7 @@ void TextureCacheRuntime::AccelerateImageUpload( if (IsPixelFormatASTC(image.info.format)) { return astc_decoder_pass.Assemble(image, map, swizzles); } - UNREACHABLE(); + ASSERT(false); } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index c81130dd2..0b7ac0df1 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -34,11 +33,11 @@ class ImageView; class Framebuffer; class RenderPassCache; class StagingBufferPool; -class VKScheduler; +class Scheduler; class TextureCacheRuntime { public: - explicit TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, + explicit TextureCacheRuntime(const Device& device_, Scheduler& scheduler_, MemoryAllocator& memory_allocator_, StagingBufferPool& staging_buffer_pool_, BlitImageHelper& blit_image_helper_, @@ -55,6 +54,10 @@ public: u64 GetDeviceLocalMemory() const; + u64 GetDeviceMemoryUsage() const; + + bool CanReportMemoryUsage() const; + void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, @@ -90,7 +93,7 @@ public: [[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size); const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; MemoryAllocator& memory_allocator; StagingBufferPool& staging_buffer_pool; BlitImageHelper& blit_image_helper; @@ -151,7 +154,7 @@ private: bool NeedsScaleHelper() const; - VKScheduler* scheduler{}; + Scheduler* scheduler{}; TextureCacheRuntime* runtime{}; vk::Image original_image; @@ -265,7 +268,7 @@ public: ImageView* depth_buffer, const VideoCommon::RenderTargets& key); explicit Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, - ImageView* depth_buffer, VkExtent2D extent); + ImageView* depth_buffer, VkExtent2D extent, bool is_rescaled); ~Framebuffer(); @@ -276,7 +279,8 @@ public: Framebuffer& operator=(Framebuffer&&) = default; void CreateFramebuffer(TextureCacheRuntime& runtime, - std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer); + std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, + bool is_rescaled = false); [[nodiscard]] VkFramebuffer Handle() const noexcept { return *framebuffer; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache_base.cpp b/src/video_core/renderer_vulkan/vk_texture_cache_base.cpp index 44e688342..7e1c3a863 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache_base.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache_base.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/texture_cache/texture_cache.h" diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.cpp b/src/video_core/renderer_vulkan/vk_update_descriptor.cpp index 0df3a7fe9..4d4a6753b 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.cpp +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.cpp @@ -1,11 +1,9 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <variant> #include <boost/container/static_vector.hpp> -#include "common/assert.h" #include "common/logging/log.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" @@ -14,18 +12,18 @@ namespace Vulkan { -VKUpdateDescriptorQueue::VKUpdateDescriptorQueue(const Device& device_, VKScheduler& scheduler_) +UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_, Scheduler& scheduler_) : device{device_}, scheduler{scheduler_} { payload_cursor = payload.data(); } -VKUpdateDescriptorQueue::~VKUpdateDescriptorQueue() = default; +UpdateDescriptorQueue::~UpdateDescriptorQueue() = default; -void VKUpdateDescriptorQueue::TickFrame() { +void UpdateDescriptorQueue::TickFrame() { payload_cursor = payload.data(); } -void VKUpdateDescriptorQueue::Acquire() { +void UpdateDescriptorQueue::Acquire() { // Minimum number of entries required. // This is the maximum number of entries a single draw call migth use. static constexpr size_t MIN_ENTRIES = 0x400; diff --git a/src/video_core/renderer_vulkan/vk_update_descriptor.h b/src/video_core/renderer_vulkan/vk_update_descriptor.h index d7de4c490..625bcc809 100644 --- a/src/video_core/renderer_vulkan/vk_update_descriptor.h +++ b/src/video_core/renderer_vulkan/vk_update_descriptor.h @@ -1,18 +1,16 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <array> -#include "common/common_types.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { class Device; -class VKScheduler; +class Scheduler; struct DescriptorUpdateEntry { struct Empty {}; @@ -30,10 +28,10 @@ struct DescriptorUpdateEntry { }; }; -class VKUpdateDescriptorQueue final { +class UpdateDescriptorQueue final { public: - explicit VKUpdateDescriptorQueue(const Device& device_, VKScheduler& scheduler_); - ~VKUpdateDescriptorQueue(); + explicit UpdateDescriptorQueue(const Device& device_, Scheduler& scheduler_); + ~UpdateDescriptorQueue(); void TickFrame(); @@ -73,7 +71,7 @@ public: private: const Device& device; - VKScheduler& scheduler; + Scheduler& scheduler; DescriptorUpdateEntry* payload_cursor = nullptr; const DescriptorUpdateEntry* upload_start = nullptr; diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index 87636857d..f53066579 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> @@ -9,6 +8,7 @@ #include "common/assert.h" #include "shader_recompiler/frontend/maxwell/control_flow.h" #include "shader_recompiler/object_pool.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" @@ -25,7 +25,7 @@ void ShaderCache::InvalidateRegion(VAddr addr, size_t size) { } void ShaderCache::OnCPUWrite(VAddr addr, size_t size) { - std::lock_guard lock{invalidation_mutex}; + std::scoped_lock lock{invalidation_mutex}; InvalidatePagesInRegion(addr, size); } @@ -34,29 +34,25 @@ void ShaderCache::SyncGuestHost() { RemovePendingShaders(); } -ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_) - : gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, - rasterizer{rasterizer_} {} +ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_) : rasterizer{rasterizer_} {} bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { - auto& dirty{maxwell3d.dirty.flags}; + auto& dirty{maxwell3d->dirty.flags}; if (!dirty[VideoCommon::Dirty::Shaders]) { return last_shaders_valid; } dirty[VideoCommon::Dirty::Shaders] = false; - const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + const GPUVAddr base_addr{maxwell3d->regs.code_address.CodeAddress()}; for (size_t index = 0; index < Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; ++index) { - if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { + if (!maxwell3d->regs.IsShaderConfigEnabled(index)) { unique_hashes[index] = 0; continue; } - const auto& shader_config{maxwell3d.regs.shader_config[index]}; + const auto& shader_config{maxwell3d->regs.shader_config[index]}; const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; const GPUVAddr shader_addr{base_addr + shader_config.offset}; - const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)}; if (!cpu_shader_addr) { LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); last_shaders_valid = false; @@ -65,7 +61,7 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { const ShaderInfo* shader_info{TryGet(*cpu_shader_addr)}; if (!shader_info) { const u32 start_address{shader_config.offset}; - GraphicsEnvironment env{maxwell3d, gpu_memory, program, base_addr, start_address}; + GraphicsEnvironment env{*maxwell3d, *gpu_memory, program, base_addr, start_address}; shader_info = MakeShaderInfo(env, *cpu_shader_addr); } shader_infos[index] = shader_info; @@ -76,10 +72,10 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { } const ShaderInfo* ShaderCache::ComputeShader() { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; const GPUVAddr shader_addr{program_base + qmd.program_start}; - const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)}; if (!cpu_shader_addr) { LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); return nullptr; @@ -87,22 +83,22 @@ const ShaderInfo* ShaderCache::ComputeShader() { if (const ShaderInfo* const shader = TryGet(*cpu_shader_addr)) { return shader; } - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; return MakeShaderInfo(env, *cpu_shader_addr); } void ShaderCache::GetGraphicsEnvironments(GraphicsEnvironments& result, const std::array<u64, NUM_PROGRAMS>& unique_hashes) { size_t env_index{}; - const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + const GPUVAddr base_addr{maxwell3d->regs.code_address.CodeAddress()}; for (size_t index = 0; index < NUM_PROGRAMS; ++index) { if (unique_hashes[index] == 0) { continue; } const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; auto& env{result.envs[index]}; - const u32 start_address{maxwell3d.regs.shader_config[index].offset}; - env = GraphicsEnvironment{maxwell3d, gpu_memory, program, base_addr, start_address}; + const u32 start_address{maxwell3d->regs.shader_config[index].offset}; + env = GraphicsEnvironment{*maxwell3d, *gpu_memory, program, base_addr, start_address}; env.SetCachedSize(shader_infos[index]->size_bytes); result.env_ptrs[env_index++] = &env; } @@ -124,8 +120,8 @@ void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t const VAddr addr_end = addr + size; Entry* const entry = NewEntry(addr, addr_end, data.get()); - const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { + const u64 page_end = (addr_end + YUZU_PAGESIZE - 1) >> YUZU_PAGEBITS; + for (u64 page = addr >> YUZU_PAGEBITS; page < page_end; ++page) { invalidation_cache[page].push_back(entry); } @@ -136,8 +132,8 @@ void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) { const VAddr addr_end = addr + size; - const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { + const u64 page_end = (addr_end + YUZU_PAGESIZE - 1) >> YUZU_PAGEBITS; + for (u64 page = addr >> YUZU_PAGEBITS; page < page_end; ++page) { auto it = invalidation_cache.find(page); if (it == invalidation_cache.end()) { continue; @@ -190,8 +186,8 @@ void ShaderCache::InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr } void ShaderCache::RemoveEntryFromInvalidationCache(const Entry* entry) { - const u64 page_end = (entry->addr_end + PAGE_SIZE - 1) >> PAGE_BITS; - for (u64 page = entry->addr_start >> PAGE_BITS; page < page_end; ++page) { + const u64 page_end = (entry->addr_end + YUZU_PAGESIZE - 1) >> YUZU_PAGEBITS; + for (u64 page = entry->addr_start >> YUZU_PAGEBITS; page < page_end; ++page) { const auto entries_it = invalidation_cache.find(page); ASSERT(entries_it != invalidation_cache.end()); std::vector<Entry*>& entries = entries_it->second; diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h index 8836bc8c6..a4391202d 100644 --- a/src/video_core/shader_cache.h +++ b/src/video_core/shader_cache.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -13,6 +12,7 @@ #include <vector> #include "common/common_types.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/rasterizer_interface.h" #include "video_core/shader_environment.h" @@ -20,6 +20,10 @@ namespace Tegra { class MemoryManager; } +namespace Tegra::Control { +struct ChannelState; +} + namespace VideoCommon { class GenericEnvironment; @@ -29,9 +33,9 @@ struct ShaderInfo { size_t size_bytes{}; }; -class ShaderCache { - static constexpr u64 PAGE_BITS = 14; - static constexpr u64 PAGE_SIZE = u64(1) << PAGE_BITS; +class ShaderCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { + static constexpr u64 YUZU_PAGEBITS = 14; + static constexpr u64 YUZU_PAGESIZE = u64(1) << YUZU_PAGEBITS; static constexpr size_t NUM_PROGRAMS = 6; @@ -72,9 +76,7 @@ protected: } }; - explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_); + explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_); /// @brief Update the hashes and information of shader stages /// @param unique_hashes Shader hashes to store into when a stage is enabled @@ -89,10 +91,6 @@ protected: void GetGraphicsEnvironments(GraphicsEnvironments& result, const std::array<u64, NUM_PROGRAMS>& unique_hashes); - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - std::array<const ShaderInfo*, NUM_PROGRAMS> shader_infos{}; bool last_shaders_valid = false; diff --git a/src/video_core/shader_environment.cpp b/src/video_core/shader_environment.cpp index 3e673c437..5f7625947 100644 --- a/src/video_core/shader_environment.cpp +++ b/src/video_core/shader_environment.cpp @@ -1,9 +1,7 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> -#include <bit> #include <filesystem> #include <fstream> #include <memory> @@ -41,7 +39,8 @@ static Shader::TextureType ConvertType(const Tegra::Texture::TICEntry& entry) { return Shader::TextureType::Color1D; case Tegra::Texture::TextureType::Texture2D: case Tegra::Texture::TextureType::Texture2DNoMipmap: - return Shader::TextureType::Color2D; + return entry.normalized_coords ? Shader::TextureType::Color2D + : Shader::TextureType::Color2DRect; case Tegra::Texture::TextureType::Texture3D: return Shader::TextureType::Color3D; case Tegra::Texture::TextureType::TextureCubemap: @@ -55,7 +54,8 @@ static Shader::TextureType ConvertType(const Tegra::Texture::TICEntry& entry) { case Tegra::Texture::TextureType::TextureCubeArray: return Shader::TextureType::ColorArrayCube; default: - throw Shader::NotImplementedException("Unknown texture type"); + UNIMPLEMENTED(); + return Shader::TextureType::Color2D; } } @@ -190,11 +190,11 @@ void GenericEnvironment::Serialize(std::ofstream& file) const { .write(reinterpret_cast<const char*>(&cached_highest), sizeof(cached_highest)) .write(reinterpret_cast<const char*>(&stage), sizeof(stage)) .write(reinterpret_cast<const char*>(code.data()), code_size); - for (const auto [key, type] : texture_types) { + for (const auto& [key, type] : texture_types) { file.write(reinterpret_cast<const char*>(&key), sizeof(key)) .write(reinterpret_cast<const char*>(&type), sizeof(type)); } - for (const auto [key, type] : cbuf_values) { + for (const auto& [key, type] : cbuf_values) { file.write(reinterpret_cast<const char*>(&key), sizeof(key)) .write(reinterpret_cast<const char*>(&type), sizeof(type)); } @@ -282,7 +282,7 @@ GraphicsEnvironment::GraphicsEnvironment(Tegra::Engines::Maxwell3D& maxwell3d_, stage_index = 4; break; default: - UNREACHABLE_MSG("Invalid program={}", program); + ASSERT_MSG(false, "Invalid program={}", program); break; } const u64 local_size{sph.LocalMemorySize()}; diff --git a/src/video_core/shader_environment.h b/src/video_core/shader_environment.h index aae762b27..5a145f33a 100644 --- a/src/video_core/shader_environment.h +++ b/src/video_core/shader_environment.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/shader_notify.cpp b/src/video_core/shader_notify.cpp index bcaf5f575..631891ec2 100644 --- a/src/video_core/shader_notify.cpp +++ b/src/video_core/shader_notify.cpp @@ -1,10 +1,8 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <atomic> #include <chrono> -#include <optional> #include "video_core/shader_notify.h" diff --git a/src/video_core/shader_notify.h b/src/video_core/shader_notify.h index 4d8d52071..696a51384 100644 --- a/src/video_core/shader_notify.h +++ b/src/video_core/shader_notify.h @@ -1,12 +1,10 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <atomic> #include <chrono> -#include <optional> namespace VideoCore { class ShaderNotify { diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index a36015c8c..a2bf08294 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/common_types.h" #include "common/math_util.h" @@ -29,7 +28,7 @@ SurfaceTarget SurfaceTargetFromTextureType(Tegra::Texture::TextureType texture_t return SurfaceTarget::Texture2DArray; default: LOG_CRITICAL(HW_GPU, "Unimplemented texture_type={}", texture_type); - UNREACHABLE(); + ASSERT(false); return SurfaceTarget::Texture2D; } } @@ -48,7 +47,7 @@ bool SurfaceTargetIsLayered(SurfaceTarget target) { return true; default: LOG_CRITICAL(HW_GPU, "Unimplemented surface_target={}", target); - UNREACHABLE(); + ASSERT(false); return false; } } @@ -67,7 +66,7 @@ bool SurfaceTargetIsArray(SurfaceTarget target) { return true; default: LOG_CRITICAL(HW_GPU, "Unimplemented surface_target={}", target); - UNREACHABLE(); + ASSERT(false); return false; } } @@ -190,13 +189,13 @@ PixelFormat PixelFormatFromRenderTargetFormat(Tegra::RenderTargetFormat format) } } -PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat format) { +PixelFormat PixelFormatFromGPUPixelFormat(Service::android::PixelFormat format) { switch (format) { - case Tegra::FramebufferConfig::PixelFormat::A8B8G8R8_UNORM: + case Service::android::PixelFormat::Rgba8888: return PixelFormat::A8B8G8R8_UNORM; - case Tegra::FramebufferConfig::PixelFormat::RGB565_UNORM: + case Service::android::PixelFormat::Rgb565: return PixelFormat::R5G6B5_UNORM; - case Tegra::FramebufferConfig::PixelFormat::B8G8R8A8_UNORM: + case Service::android::PixelFormat::Bgra8888: return PixelFormat::B8G8R8A8_UNORM; default: UNIMPLEMENTED_MSG("Unimplemented format={}", format); @@ -247,6 +246,9 @@ bool IsPixelFormatASTC(PixelFormat format) { case PixelFormat::ASTC_2D_10X8_SRGB: case PixelFormat::ASTC_2D_6X6_UNORM: case PixelFormat::ASTC_2D_6X6_SRGB: + case PixelFormat::ASTC_2D_10X6_UNORM: + case PixelFormat::ASTC_2D_10X5_UNORM: + case PixelFormat::ASTC_2D_10X5_SRGB: case PixelFormat::ASTC_2D_10X10_UNORM: case PixelFormat::ASTC_2D_10X10_SRGB: case PixelFormat::ASTC_2D_12X12_UNORM: @@ -276,6 +278,7 @@ bool IsPixelFormatSRGB(PixelFormat format) { case PixelFormat::ASTC_2D_5X5_SRGB: case PixelFormat::ASTC_2D_10X8_SRGB: case PixelFormat::ASTC_2D_6X6_SRGB: + case PixelFormat::ASTC_2D_10X5_SRGB: case PixelFormat::ASTC_2D_10X10_SRGB: case PixelFormat::ASTC_2D_12X12_SRGB: case PixelFormat::ASTC_2D_8X6_SRGB: diff --git a/src/video_core/surface.h b/src/video_core/surface.h index 33e8d24ab..57ca7f597 100644 --- a/src/video_core/surface.h +++ b/src/video_core/surface.h @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -25,6 +24,7 @@ enum class PixelFormat { A2B10G10R10_UNORM, A2B10G10R10_UINT, A1B5G5R5_UNORM, + A5B5G5R1_UNORM, R8_UNORM, R8_SNORM, R8_SINT, @@ -82,6 +82,7 @@ enum class PixelFormat { BC3_SRGB, BC7_SRGB, A4B4G4R4_UNORM, + G4R4_UNORM, ASTC_2D_4X4_SRGB, ASTC_2D_8X8_SRGB, ASTC_2D_8X5_SRGB, @@ -92,6 +93,9 @@ enum class PixelFormat { ASTC_2D_10X8_SRGB, ASTC_2D_6X6_UNORM, ASTC_2D_6X6_SRGB, + ASTC_2D_10X6_UNORM, + ASTC_2D_10X5_UNORM, + ASTC_2D_10X5_SRGB, ASTC_2D_10X10_UNORM, ASTC_2D_10X10_SRGB, ASTC_2D_12X12_UNORM, @@ -145,7 +149,7 @@ enum class SurfaceTarget { TextureCubeArray, }; -constexpr std::array<u32, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ +constexpr std::array<u8, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 1, // A8B8G8R8_UNORM 1, // A8B8G8R8_SNORM 1, // A8B8G8R8_SINT @@ -156,6 +160,7 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 1, // A2B10G10R10_UNORM 1, // A2B10G10R10_UINT 1, // A1B5G5R5_UNORM + 1, // A5B5G5R1_UNORM 1, // R8_UNORM 1, // R8_SNORM 1, // R8_SINT @@ -213,6 +218,7 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 4, // BC3_SRGB 4, // BC7_SRGB 1, // A4B4G4R4_UNORM + 1, // G4R4_UNORM 4, // ASTC_2D_4X4_SRGB 8, // ASTC_2D_8X8_SRGB 8, // ASTC_2D_8X5_SRGB @@ -223,6 +229,9 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 10, // ASTC_2D_10X8_SRGB 6, // ASTC_2D_6X6_UNORM 6, // ASTC_2D_6X6_SRGB + 10, // ASTC_2D_10X6_UNORM + 10, // ASTC_2D_10X5_UNORM + 10, // ASTC_2D_10X5_SRGB 10, // ASTC_2D_10X10_UNORM 10, // ASTC_2D_10X10_SRGB 12, // ASTC_2D_12X12_UNORM @@ -245,7 +254,7 @@ constexpr u32 DefaultBlockWidth(PixelFormat format) { return BLOCK_WIDTH_TABLE[static_cast<std::size_t>(format)]; } -constexpr std::array<u32, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ +constexpr std::array<u8, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 1, // A8B8G8R8_UNORM 1, // A8B8G8R8_SNORM 1, // A8B8G8R8_SINT @@ -256,6 +265,7 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 1, // A2B10G10R10_UNORM 1, // A2B10G10R10_UINT 1, // A1B5G5R5_UNORM + 1, // A5B5G5R1_UNORM 1, // R8_UNORM 1, // R8_SNORM 1, // R8_SINT @@ -313,6 +323,7 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 4, // BC3_SRGB 4, // BC7_SRGB 1, // A4B4G4R4_UNORM + 1, // G4R4_UNORM 4, // ASTC_2D_4X4_SRGB 8, // ASTC_2D_8X8_SRGB 5, // ASTC_2D_8X5_SRGB @@ -323,6 +334,9 @@ constexpr std::array<u32, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 8, // ASTC_2D_10X8_SRGB 6, // ASTC_2D_6X6_UNORM 6, // ASTC_2D_6X6_SRGB + 6, // ASTC_2D_10X6_UNORM + 5, // ASTC_2D_10X5_UNORM + 5, // ASTC_2D_10X5_SRGB 10, // ASTC_2D_10X10_UNORM 10, // ASTC_2D_10X10_SRGB 12, // ASTC_2D_12X12_UNORM @@ -345,7 +359,7 @@ constexpr u32 DefaultBlockHeight(PixelFormat format) { return BLOCK_HEIGHT_TABLE[static_cast<std::size_t>(format)]; } -constexpr std::array<u32, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ +constexpr std::array<u8, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 32, // A8B8G8R8_UNORM 32, // A8B8G8R8_SNORM 32, // A8B8G8R8_SINT @@ -356,6 +370,7 @@ constexpr std::array<u32, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 32, // A2B10G10R10_UNORM 32, // A2B10G10R10_UINT 16, // A1B5G5R5_UNORM + 16, // A5B5G5R1_UNORM 8, // R8_UNORM 8, // R8_SNORM 8, // R8_SINT @@ -413,6 +428,7 @@ constexpr std::array<u32, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 128, // BC3_SRGB 128, // BC7_UNORM 16, // A4B4G4R4_UNORM + 8, // G4R4_UNORM 128, // ASTC_2D_4X4_SRGB 128, // ASTC_2D_8X8_SRGB 128, // ASTC_2D_8X5_SRGB @@ -423,6 +439,9 @@ constexpr std::array<u32, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 128, // ASTC_2D_10X8_SRGB 128, // ASTC_2D_6X6_UNORM 128, // ASTC_2D_6X6_SRGB + 128, // ASTC_2D_10X6_UNORM + 128, // ASTC_2D_10X5_UNORM + 128, // ASTC_2D_10X5_SRGB 128, // ASTC_2D_10X10_UNORM 128, // ASTC_2D_10X10_SRGB 128, // ASTC_2D_12X12_UNORM @@ -460,7 +479,7 @@ PixelFormat PixelFormatFromDepthFormat(Tegra::DepthFormat format); PixelFormat PixelFormatFromRenderTargetFormat(Tegra::RenderTargetFormat format); -PixelFormat PixelFormatFromGPUPixelFormat(Tegra::FramebufferConfig::PixelFormat format); +PixelFormat PixelFormatFromGPUPixelFormat(Service::android::PixelFormat format); SurfaceType GetFormatType(PixelFormat pixel_format); diff --git a/src/video_core/texture_cache/accelerated_swizzle.cpp b/src/video_core/texture_cache/accelerated_swizzle.cpp index 15585caeb..70be1657e 100644 --- a/src/video_core/texture_cache/accelerated_swizzle.cpp +++ b/src/video_core/texture_cache/accelerated_swizzle.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <bit> diff --git a/src/video_core/texture_cache/accelerated_swizzle.h b/src/video_core/texture_cache/accelerated_swizzle.h index a11c924e1..d4250da68 100644 --- a/src/video_core/texture_cache/accelerated_swizzle.h +++ b/src/video_core/texture_cache/accelerated_swizzle.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/decode_bc4.cpp b/src/video_core/texture_cache/decode_bc4.cpp index 017327975..ef98afdca 100644 --- a/src/video_core/texture_cache/decode_bc4.cpp +++ b/src/video_core/texture_cache/decode_bc4.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> diff --git a/src/video_core/texture_cache/decode_bc4.h b/src/video_core/texture_cache/decode_bc4.h index 63fb23508..ab2f735be 100644 --- a/src/video_core/texture_cache/decode_bc4.h +++ b/src/video_core/texture_cache/decode_bc4.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/descriptor_table.h b/src/video_core/texture_cache/descriptor_table.h index 3a03b786f..b18e3838f 100644 --- a/src/video_core/texture_cache/descriptor_table.h +++ b/src/video_core/texture_cache/descriptor_table.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,7 +8,6 @@ #include "common/common_types.h" #include "common/div_ceil.h" -#include "common/logging/log.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp index afa807d5d..ad935d386 100644 --- a/src/video_core/texture_cache/format_lookup_table.cpp +++ b/src/video_core/texture_cache/format_lookup_table.cpp @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/common_types.h" #include "common/logging/log.h" @@ -63,6 +62,10 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, return PixelFormat::A1B5G5R5_UNORM; case Hash(TextureFormat::A4B4G4R4, UNORM): return PixelFormat::A4B4G4R4_UNORM; + case Hash(TextureFormat::G4R4, UNORM): + return PixelFormat::G4R4_UNORM; + case Hash(TextureFormat::A5B5G5R1, UNORM): + return PixelFormat::A5B5G5R1_UNORM; case Hash(TextureFormat::R8, UNORM): return PixelFormat::R8_UNORM; case Hash(TextureFormat::R8, SNORM): @@ -143,6 +146,8 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, return PixelFormat::S8_UINT_D24_UNORM; case Hash(TextureFormat::R8G24, UINT, UNORM, UNORM, UNORM, LINEAR): return PixelFormat::S8_UINT_D24_UNORM; + case Hash(TextureFormat::D24S8, UNORM, UINT, UINT, UINT, LINEAR): + return PixelFormat::D24_UNORM_S8_UINT; case Hash(TextureFormat::D32S8, FLOAT, UINT, UNORM, UNORM, LINEAR): return PixelFormat::D32_FLOAT_S8_UINT; case Hash(TextureFormat::BC1_RGBA, UNORM, LINEAR): @@ -201,6 +206,12 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, return PixelFormat::ASTC_2D_6X6_UNORM; case Hash(TextureFormat::ASTC_2D_6X6, UNORM, SRGB): return PixelFormat::ASTC_2D_6X6_SRGB; + case Hash(TextureFormat::ASTC_2D_10X6, UNORM, LINEAR): + return PixelFormat::ASTC_2D_10X6_UNORM; + case Hash(TextureFormat::ASTC_2D_10X5, UNORM, LINEAR): + return PixelFormat::ASTC_2D_10X5_UNORM; + case Hash(TextureFormat::ASTC_2D_10X5, UNORM, SRGB): + return PixelFormat::ASTC_2D_10X5_SRGB; case Hash(TextureFormat::ASTC_2D_10X10, UNORM, LINEAR): return PixelFormat::ASTC_2D_10X10_UNORM; case Hash(TextureFormat::ASTC_2D_10X10, UNORM, SRGB): diff --git a/src/video_core/texture_cache/format_lookup_table.h b/src/video_core/texture_cache/format_lookup_table.h index 729533999..f504a6cd0 100644 --- a/src/video_core/texture_cache/format_lookup_table.h +++ b/src/video_core/texture_cache/format_lookup_table.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/formatter.cpp b/src/video_core/texture_cache/formatter.cpp index 249cc4d0f..ee4f2d406 100644 --- a/src/video_core/texture_cache/formatter.cpp +++ b/src/video_core/texture_cache/formatter.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <string> diff --git a/src/video_core/texture_cache/formatter.h b/src/video_core/texture_cache/formatter.h index b2c81057b..acc854715 100644 --- a/src/video_core/texture_cache/formatter.h +++ b/src/video_core/texture_cache/formatter.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -38,6 +37,8 @@ struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::str return "A2B10G10R10_UINT"; case PixelFormat::A1B5G5R5_UNORM: return "A1B5G5R5_UNORM"; + case PixelFormat::A5B5G5R1_UNORM: + return "A5B5G5R1_UNORM"; case PixelFormat::R8_UNORM: return "R8_UNORM"; case PixelFormat::R8_SNORM: @@ -152,6 +153,8 @@ struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::str return "BC7_SRGB"; case PixelFormat::A4B4G4R4_UNORM: return "A4B4G4R4_UNORM"; + case PixelFormat::G4R4_UNORM: + return "G4R4_UNORM"; case PixelFormat::ASTC_2D_4X4_SRGB: return "ASTC_2D_4X4_SRGB"; case PixelFormat::ASTC_2D_8X8_SRGB: @@ -172,6 +175,12 @@ struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::str return "ASTC_2D_6X6_UNORM"; case PixelFormat::ASTC_2D_6X6_SRGB: return "ASTC_2D_6X6_SRGB"; + case PixelFormat::ASTC_2D_10X6_UNORM: + return "ASTC_2D_10X6_UNORM"; + case PixelFormat::ASTC_2D_10X5_UNORM: + return "ASTC_2D_10X5_UNORM"; + case PixelFormat::ASTC_2D_10X5_SRGB: + return "ASTC_2D_10X5_SRGB"; case PixelFormat::ASTC_2D_10X10_UNORM: return "ASTC_2D_10X10_UNORM"; case PixelFormat::ASTC_2D_10X10_SRGB: diff --git a/src/video_core/texture_cache/image_base.cpp b/src/video_core/texture_cache/image_base.cpp index 3db2fdf34..91512022f 100644 --- a/src/video_core/texture_cache/image_base.cpp +++ b/src/video_core/texture_cache/image_base.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <optional> @@ -8,6 +7,7 @@ #include <vector> #include "common/common_types.h" +#include "common/div_ceil.h" #include "video_core/surface.h" #include "video_core/texture_cache/formatter.h" #include "video_core/texture_cache/image_base.h" @@ -183,10 +183,6 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i }; const bool is_lhs_compressed = lhs_block.width > 1 || lhs_block.height > 1; const bool is_rhs_compressed = rhs_block.width > 1 || rhs_block.height > 1; - if (is_lhs_compressed && is_rhs_compressed) { - LOG_ERROR(HW_GPU, "Compressed to compressed image aliasing is not implemented"); - return; - } const s32 lhs_mips = lhs.info.resources.levels; const s32 rhs_mips = rhs.info.resources.levels; const s32 num_mips = std::min(lhs_mips - base->level, rhs_mips); @@ -200,12 +196,12 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i Extent3D lhs_size = MipSize(lhs.info.size, base->level + mip_level); Extent3D rhs_size = MipSize(rhs.info.size, mip_level); if (is_lhs_compressed) { - lhs_size.width /= lhs_block.width; - lhs_size.height /= lhs_block.height; + lhs_size.width = Common::DivCeil(lhs_size.width, lhs_block.width); + lhs_size.height = Common::DivCeil(lhs_size.height, lhs_block.height); } if (is_rhs_compressed) { - rhs_size.width /= rhs_block.width; - rhs_size.height /= rhs_block.height; + rhs_size.width = Common::DivCeil(rhs_size.width, rhs_block.width); + rhs_size.height = Common::DivCeil(rhs_size.height, rhs_block.height); } const Extent3D copy_size{ .width = std::min(lhs_size.width, rhs_size.width), diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h index 89c111c00..620565684 100644 --- a/src/video_core/texture_cache/image_base.h +++ b/src/video_core/texture_cache/image_base.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -33,11 +32,12 @@ enum class ImageFlagBits : u32 { ///< garbage collection priority Alias = 1 << 11, ///< This image has aliases and has priority on garbage ///< collection + CostlyLoad = 1 << 12, ///< Protected from low-tier GC as it is costly to load back. // Rescaler - Rescaled = 1 << 12, - CheckingRescalable = 1 << 13, - IsRescalable = 1 << 14, + Rescaled = 1 << 13, + CheckingRescalable = 1 << 14, + IsRescalable = 1 << 15, }; DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits) @@ -88,6 +88,9 @@ struct ImageBase { u32 scale_rating = 0; u64 scale_tick = 0; bool has_scaled = false; + + size_t channel = 0; + ImageFlagBits flags = ImageFlagBits::CpuModified; GPUVAddr gpu_addr = 0; diff --git a/src/video_core/texture_cache/image_info.cpp b/src/video_core/texture_cache/image_info.cpp index afb94082b..6c073ee57 100644 --- a/src/video_core/texture_cache/image_info.cpp +++ b/src/video_core/texture_cache/image_info.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" #include "video_core/surface.h" @@ -95,7 +94,7 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept { resources.layers = 1; break; default: - UNREACHABLE_MSG("Invalid texture_type={}", static_cast<int>(config.texture_type.Value())); + ASSERT_MSG(false, "Invalid texture_type={}", static_cast<int>(config.texture_type.Value())); break; } if (type != ImageType::Linear) { diff --git a/src/video_core/texture_cache/image_info.h b/src/video_core/texture_cache/image_info.h index 5932dcaba..93755e15e 100644 --- a/src/video_core/texture_cache/image_info.h +++ b/src/video_core/texture_cache/image_info.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/image_view_base.cpp b/src/video_core/texture_cache/image_view_base.cpp index c7b4fc231..04fb84bfa 100644 --- a/src/video_core/texture_cache/image_view_base.cpp +++ b/src/video_core/texture_cache/image_view_base.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> diff --git a/src/video_core/texture_cache/image_view_base.h b/src/video_core/texture_cache/image_view_base.h index 9c24c5359..69c9776e7 100644 --- a/src/video_core/texture_cache/image_view_base.h +++ b/src/video_core/texture_cache/image_view_base.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/image_view_info.cpp b/src/video_core/texture_cache/image_view_info.cpp index e751f26c7..f47885147 100644 --- a/src/video_core/texture_cache/image_view_info.cpp +++ b/src/video_core/texture_cache/image_view_info.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <limits> @@ -72,7 +71,7 @@ ImageViewInfo::ImageViewInfo(const TICEntry& config, s32 base_layer) noexcept range.extent.layers = config.Depth() * 6; break; default: - UNREACHABLE_MSG("Invalid texture_type={}", static_cast<int>(config.texture_type.Value())); + ASSERT_MSG(false, "Invalid texture_type={}", static_cast<int>(config.texture_type.Value())); break; } } diff --git a/src/video_core/texture_cache/image_view_info.h b/src/video_core/texture_cache/image_view_info.h index 0c1f99117..921f88988 100644 --- a/src/video_core/texture_cache/image_view_info.h +++ b/src/video_core/texture_cache/image_view_info.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/render_targets.h b/src/video_core/texture_cache/render_targets.h index 0cb227d69..1efbd6507 100644 --- a/src/video_core/texture_cache/render_targets.h +++ b/src/video_core/texture_cache/render_targets.h @@ -1,12 +1,10 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <algorithm> #include <span> -#include <utility> #include "common/bit_cast.h" #include "video_core/texture_cache/types.h" @@ -28,6 +26,7 @@ struct RenderTargets { ImageViewId depth_buffer_id{}; std::array<u8, NUM_RT> draw_buffers{}; Extent2D size{}; + bool is_rescaled{}; }; } // namespace VideoCommon diff --git a/src/video_core/texture_cache/samples_helper.h b/src/video_core/texture_cache/samples_helper.h index 04539a43c..d552bccf0 100644 --- a/src/video_core/texture_cache/samples_helper.h +++ b/src/video_core/texture_cache/samples_helper.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -24,7 +23,7 @@ namespace VideoCommon { case 16: return {2, 2}; } - UNREACHABLE_MSG("Invalid number of samples={}", num_samples); + ASSERT_MSG(false, "Invalid number of samples={}", num_samples); return {1, 1}; } @@ -48,7 +47,7 @@ namespace VideoCommon { case MsaaMode::Msaa4x4: return 16; } - UNREACHABLE_MSG("Invalid MSAA mode={}", static_cast<int>(msaa_mode)); + ASSERT_MSG(false, "Invalid MSAA mode={}", static_cast<int>(msaa_mode)); return 1; } diff --git a/src/video_core/texture_cache/slot_vector.h b/src/video_core/texture_cache/slot_vector.h index 50df06409..46e8a86e6 100644 --- a/src/video_core/texture_cache/slot_vector.h +++ b/src/video_core/texture_cache/slot_vector.h @@ -1,13 +1,10 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <algorithm> -#include <array> #include <bit> -#include <concepts> #include <numeric> #include <type_traits> #include <utility> diff --git a/src/video_core/texture_cache/texture_cache.cpp b/src/video_core/texture_cache/texture_cache.cpp new file mode 100644 index 000000000..8a9a32f44 --- /dev/null +++ b/src/video_core/texture_cache/texture_cache.cpp @@ -0,0 +1,15 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "video_core/control/channel_state_cache.inc" +#include "video_core/texture_cache/texture_cache_base.h" + +namespace VideoCommon { + +TextureCacheChannelInfo::TextureCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept + : ChannelInfo(state), graphics_image_table{gpu_memory}, graphics_sampler_table{gpu_memory}, + compute_image_table{gpu_memory}, compute_sampler_table{gpu_memory} {} + +template class VideoCommon::ChannelSetupCaches<VideoCommon::TextureCacheChannelInfo>; + +} // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 198bb0cfb..eaf4a1c95 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once @@ -8,6 +7,7 @@ #include "common/alignment.h" #include "common/settings.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/texture_cache/image_view_base.h" @@ -30,12 +30,8 @@ using VideoCore::Surface::SurfaceType; using namespace Common::Literals; template <class P> -TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_) - : runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_} { +TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_) + : runtime{runtime_}, rasterizer{rasterizer_} { // Configure null sampler TSCEntry sampler_descriptor{}; sampler_descriptor.min_filter.Assign(Tegra::Texture::TextureFilter::Linear); @@ -50,14 +46,20 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& void(slot_samplers.insert(runtime, sampler_descriptor)); if constexpr (HAS_DEVICE_MEMORY_INFO) { - const auto device_memory = runtime.GetDeviceLocalMemory(); - const u64 possible_expected_memory = (device_memory * 4) / 10; - const u64 possible_critical_memory = (device_memory * 7) / 10; - expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY - 256_MiB); - critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY - 512_MiB); - minimum_memory = 0; + const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory()); + const s64 min_spacing_expected = device_memory - 1_GiB - 512_MiB; + const s64 min_spacing_critical = device_memory - 1_GiB; + const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD); + const s64 min_vacancy_expected = (6 * mem_threshold) / 10; + const s64 min_vacancy_critical = (3 * mem_threshold) / 10; + expected_memory = static_cast<u64>( + std::max(std::min(device_memory - min_vacancy_expected, min_spacing_expected), + DEFAULT_EXPECTED_MEMORY)); + critical_memory = static_cast<u64>( + std::max(std::min(device_memory - min_vacancy_critical, min_spacing_critical), + DEFAULT_CRITICAL_MEMORY)); + minimum_memory = static_cast<u64>((device_memory - mem_threshold) / 2); } else { - // On OpenGL we can be more conservatives as the driver takes care. expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB; critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB; minimum_memory = 0; @@ -66,18 +68,21 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& template <class P> void TextureCache<P>::RunGarbageCollector() { - const bool high_priority_mode = total_used_memory >= expected_memory; - const bool aggressive_mode = total_used_memory >= critical_memory; - const u64 ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 100ULL; - size_t num_iterations = aggressive_mode ? 300 : (high_priority_mode ? 50 : 10); - const auto clean_up = [this, &num_iterations, high_priority_mode](ImageId image_id) { + bool high_priority_mode = total_used_memory >= expected_memory; + bool aggressive_mode = total_used_memory >= critical_memory; + const u64 ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 50ULL; + size_t num_iterations = aggressive_mode ? 40 : (high_priority_mode ? 20 : 10); + const auto clean_up = [this, &num_iterations, &high_priority_mode, + &aggressive_mode](ImageId image_id) { if (num_iterations == 0) { return true; } --num_iterations; auto& image = slot_images[image_id]; - const bool must_download = image.IsSafeDownload(); - if (!high_priority_mode && must_download) { + const bool must_download = + image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap); + if (!high_priority_mode && + (must_download || True(image.flags & ImageFlagBits::CostlyLoad))) { return false; } if (must_download) { @@ -85,13 +90,25 @@ void TextureCache<P>::RunGarbageCollector() { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); } if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, image_id); } UnregisterImage(image_id); DeleteImage(image_id, image.scale_tick > frame_tick + 5); + if (total_used_memory < critical_memory) { + if (aggressive_mode) { + // Sink the aggresiveness. + num_iterations >>= 2; + aggressive_mode = false; + return false; + } + if (high_priority_mode && total_used_memory < expected_memory) { + num_iterations >>= 1; + high_priority_mode = false; + } + } return false; }; lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up); @@ -99,6 +116,10 @@ void TextureCache<P>::RunGarbageCollector() { template <class P> void TextureCache<P>::TickFrame() { + // If we can obtain the memory info, use it instead of the estimate. + if (runtime.CanReportMemoryUsage()) { + total_used_memory = runtime.GetDeviceMemoryUsage(); + } if (total_used_memory > minimum_memory) { RunGarbageCollector(); } @@ -106,6 +127,7 @@ void TextureCache<P>::TickFrame() { sentenced_framebuffers.Tick(); sentenced_image_view.Tick(); runtime.TickFrame(); + critical_gc = 0; ++frame_tick; } @@ -127,22 +149,24 @@ void TextureCache<P>::MarkModification(ImageId id) noexcept { template <class P> template <bool has_blacklists> void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) { - FillImageViews<has_blacklists>(graphics_image_table, graphics_image_view_ids, views); + FillImageViews<has_blacklists>(channel_state->graphics_image_table, + channel_state->graphics_image_view_ids, views); } template <class P> void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) { - FillImageViews<true>(compute_image_table, compute_image_view_ids, views); + FillImageViews<true>(channel_state->compute_image_table, channel_state->compute_image_view_ids, + views); } template <class P> typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { - if (index > graphics_sampler_table.Limit()) { + if (index > channel_state->graphics_sampler_table.Limit()) { LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } - const auto [descriptor, is_new] = graphics_sampler_table.Read(index); - SamplerId& id = graphics_sampler_ids[index]; + const auto [descriptor, is_new] = channel_state->graphics_sampler_table.Read(index); + SamplerId& id = channel_state->graphics_sampler_ids[index]; if (is_new) { id = FindSampler(descriptor); } @@ -151,12 +175,12 @@ typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { template <class P> typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { - if (index > compute_sampler_table.Limit()) { + if (index > channel_state->compute_sampler_table.Limit()) { LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } - const auto [descriptor, is_new] = compute_sampler_table.Read(index); - SamplerId& id = compute_sampler_ids[index]; + const auto [descriptor, is_new] = channel_state->compute_sampler_table.Read(index); + SamplerId& id = channel_state->compute_sampler_ids[index]; if (is_new) { id = FindSampler(descriptor); } @@ -166,34 +190,36 @@ typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { template <class P> void TextureCache<P>::SynchronizeGraphicsDescriptors() { using SamplerIndex = Tegra::Engines::Maxwell3D::Regs::SamplerIndex; - const bool linked_tsc = maxwell3d.regs.sampler_index == SamplerIndex::ViaHeaderIndex; - const u32 tic_limit = maxwell3d.regs.tic.limit; - const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d.regs.tsc.limit; - if (graphics_sampler_table.Synchornize(maxwell3d.regs.tsc.Address(), tsc_limit)) { - graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); + const bool linked_tsc = maxwell3d->regs.sampler_index == SamplerIndex::ViaHeaderIndex; + const u32 tic_limit = maxwell3d->regs.tic.limit; + const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tsc.limit; + if (channel_state->graphics_sampler_table.Synchornize(maxwell3d->regs.tsc.Address(), + tsc_limit)) { + channel_state->graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); } - if (graphics_image_table.Synchornize(maxwell3d.regs.tic.Address(), tic_limit)) { - graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); + if (channel_state->graphics_image_table.Synchornize(maxwell3d->regs.tic.Address(), tic_limit)) { + channel_state->graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); } } template <class P> void TextureCache<P>::SynchronizeComputeDescriptors() { - const bool linked_tsc = kepler_compute.launch_description.linked_tsc; - const u32 tic_limit = kepler_compute.regs.tic.limit; - const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute.regs.tsc.limit; - const GPUVAddr tsc_gpu_addr = kepler_compute.regs.tsc.Address(); - if (compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) { - compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); + const bool linked_tsc = kepler_compute->launch_description.linked_tsc; + const u32 tic_limit = kepler_compute->regs.tic.limit; + const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit; + const GPUVAddr tsc_gpu_addr = kepler_compute->regs.tsc.Address(); + if (channel_state->compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) { + channel_state->compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); } - if (compute_image_table.Synchornize(kepler_compute.regs.tic.Address(), tic_limit)) { - compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); + if (channel_state->compute_image_table.Synchornize(kepler_compute->regs.tic.Address(), + tic_limit)) { + channel_state->compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); } } template <class P> bool TextureCache<P>::RescaleRenderTargets(bool is_clear) { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; u32 scale_rating = 0; bool rescaled = false; std::array<ImageId, NUM_RT> tmp_color_images{}; @@ -290,7 +316,7 @@ bool TextureCache<P>::RescaleRenderTargets(bool is_clear) { template <class P> void TextureCache<P>::UpdateRenderTargets(bool is_clear) { using namespace VideoCommon::Dirty; - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::RenderTargets]) { for (size_t index = 0; index < NUM_RT; ++index) { ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index]; @@ -317,7 +343,7 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); for (size_t index = 0; index < NUM_RT; ++index) { - render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d.regs.rt_control.Map(index)); + render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d->regs.rt_control.Map(index)); } u32 up_scale = 1; u32 down_shift = 0; @@ -326,9 +352,10 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { down_shift = Settings::values.resolution_info.down_shift; } render_targets.size = Extent2D{ - (maxwell3d.regs.render_area.width * up_scale) >> down_shift, - (maxwell3d.regs.render_area.height * up_scale) >> down_shift, + (maxwell3d->regs.render_area.width * up_scale) >> down_shift, + (maxwell3d->regs.render_area.height * up_scale) >> down_shift, }; + render_targets.is_rescaled = is_rescaling; flags[Dirty::DepthBiasGlobal] = true; } @@ -343,7 +370,7 @@ template <bool has_blacklists> void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table, std::span<ImageViewId> cached_image_view_ids, std::span<ImageViewInOut> views) { - bool has_blacklisted; + bool has_blacklisted = false; do { has_deleted_images = false; if constexpr (has_blacklists) { @@ -433,7 +460,7 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); } } @@ -452,12 +479,20 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) { } template <class P> -void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) { +void TextureCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size) { std::vector<ImageId> deleted_images; - ForEachImageInRegionGPU(gpu_addr, size, + ForEachImageInRegionGPU(as_id, gpu_addr, size, [&](ImageId id, Image&) { deleted_images.push_back(id); }); for (const ImageId id : deleted_images) { Image& image = slot_images[id]; + if (True(image.flags & ImageFlagBits::CpuModified)) { + return; + } + image.flags |= ImageFlagBits::CpuModified; + if (True(image.flags & ImageFlagBits::Tracked)) { + UntrackImage(image, id); + } + /* if (True(image.flags & ImageFlagBits::Remapped)) { continue; } @@ -465,6 +500,7 @@ void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) { if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, id); } + */ } } @@ -564,7 +600,7 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, template <class P> typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_addr) { // TODO: Properly implement this - const auto it = page_table.find(cpu_addr >> PAGE_BITS); + const auto it = page_table.find(cpu_addr >> YUZU_PAGEBITS); if (it == page_table.end()) { return nullptr; } @@ -630,7 +666,7 @@ void TextureCache<P>::PopAsyncFlushes() { for (const ImageId image_id : download_ids) { const ImageBase& image = slot_images[image_id]; const auto copies = FullDownloadCopies(image.info); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, download_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span); download_map.offset += image.unswizzled_size_bytes; download_span = download_span.subspan(image.unswizzled_size_bytes); } @@ -689,26 +725,26 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging) const GPUVAddr gpu_addr = image.gpu_addr; if (True(image.flags & ImageFlagBits::AcceleratedUpload)) { - gpu_memory.ReadBlockUnsafe(gpu_addr, mapped_span.data(), mapped_span.size_bytes()); + gpu_memory->ReadBlockUnsafe(gpu_addr, mapped_span.data(), mapped_span.size_bytes()); const auto uploads = FullUploadSwizzles(image.info); runtime.AccelerateImageUpload(image, staging, uploads); } else if (True(image.flags & ImageFlagBits::Converted)) { std::vector<u8> unswizzled_data(image.unswizzled_size_bytes); - auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, unswizzled_data); + auto copies = UnswizzleImage(*gpu_memory, gpu_addr, image.info, unswizzled_data); ConvertImage(unswizzled_data, image.info, mapped_span, copies); image.UploadMemory(staging, copies); } else { - const auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, mapped_span); + const auto copies = UnswizzleImage(*gpu_memory, gpu_addr, image.info, mapped_span); image.UploadMemory(staging, copies); } } template <class P> ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) { - if (!IsValidEntry(gpu_memory, config)) { + if (!IsValidEntry(*gpu_memory, config)) { return NULL_IMAGE_VIEW_ID; } - const auto [pair, is_new] = image_views.try_emplace(config); + const auto [pair, is_new] = channel_state->image_views.try_emplace(config); ImageViewId& image_view_id = pair->second; if (is_new) { image_view_id = CreateImageView(config); @@ -752,9 +788,9 @@ ImageId TextureCache<P>::FindOrInsertImage(const ImageInfo& info, GPUVAddr gpu_a template <class P> ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr, RelaxedOptions options) { - std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr) { - cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info)); + cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info)); if (!cpu_addr) { return ImageId{}; } @@ -835,7 +871,7 @@ void TextureCache<P>::InvalidateScale(Image& image) { image.scale_tick = frame_tick + 1; } const std::span<const ImageViewId> image_view_ids = image.image_view_ids; - auto& dirty = maxwell3d.dirty.flags; + auto& dirty = maxwell3d->dirty.flags; dirty[Dirty::RenderTargets] = true; dirty[Dirty::ZetaBuffer] = true; for (size_t rt = 0; rt < NUM_RT; ++rt) { @@ -855,12 +891,15 @@ void TextureCache<P>::InvalidateScale(Image& image) { } image.image_view_ids.clear(); image.image_view_infos.clear(); - if constexpr (ENABLE_VALIDATION) { - std::ranges::fill(graphics_image_view_ids, CORRUPT_ID); - std::ranges::fill(compute_image_view_ids, CORRUPT_ID); + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + if constexpr (ENABLE_VALIDATION) { + std::ranges::fill(channel_info.graphics_image_view_ids, CORRUPT_ID); + std::ranges::fill(channel_info.compute_image_view_ids, CORRUPT_ID); + } + channel_info.graphics_image_table.Invalidate(); + channel_info.compute_image_table.Invalidate(); } - graphics_image_table.Invalidate(); - compute_image_table.Invalidate(); has_deleted_images = true; } @@ -904,10 +943,10 @@ bool TextureCache<P>::ScaleDown(Image& image) { template <class P> ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr, RelaxedOptions options) { - std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr) { const auto size = CalculateGuestSizeInBytes(info); - cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, size); + cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, size); if (!cpu_addr) { const VAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space; virtual_invalid_space += Common::AlignUp(size, 32); @@ -1025,7 +1064,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr); Image& new_image = slot_images[new_image_id]; - if (!gpu_memory.IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) { + if (!gpu_memory->IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) { new_image.flags |= ImageFlagBits::Sparse; } @@ -1052,6 +1091,9 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA for (const ImageId overlap_id : overlap_ids) { Image& overlap = slot_images[overlap_id]; + if (True(overlap.flags & ImageFlagBits::GpuModified)) { + new_image.flags |= ImageFlagBits::GpuModified; + } if (overlap.info.num_samples != new_image.info.num_samples) { LOG_WARNING(HW_GPU, "Copying between images with different samples is not implemented"); } else { @@ -1164,7 +1206,7 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config) { if (std::ranges::all_of(config.raw, [](u64 value) { return value == 0; })) { return NULL_SAMPLER_ID; } - const auto [pair, is_new] = samplers.try_emplace(config); + const auto [pair, is_new] = channel_state->samplers.try_emplace(config); if (is_new) { pair->second = slot_samplers.insert(runtime, config); } @@ -1173,7 +1215,7 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config) { template <class P> ImageViewId TextureCache<P>::FindColorBuffer(size_t index, bool is_clear) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (index >= regs.rt_control.count) { return ImageViewId{}; } @@ -1191,7 +1233,7 @@ ImageViewId TextureCache<P>::FindColorBuffer(size_t index, bool is_clear) { template <class P> ImageViewId TextureCache<P>::FindDepthBuffer(bool is_clear) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (!regs.zeta_enable) { return ImageViewId{}; } @@ -1288,11 +1330,17 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f template <class P> template <typename Func> -void TextureCache<P>::ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func) { +void TextureCache<P>::ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, size_t size, + Func&& func) { using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type; static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>; boost::container::small_vector<ImageId, 8> images; - ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) { + auto storage_id = getStorageID(as_id); + if (!storage_id) { + return; + } + auto& gpu_page_table = gpu_page_table_storage[*storage_id]; + ForEachGPUPage(gpu_addr, size, [this, gpu_page_table, &images, gpu_addr, size, func](u64 page) { const auto it = gpu_page_table.find(page); if (it == gpu_page_table.end()) { if constexpr (BOOL_BREAK) { @@ -1375,9 +1423,9 @@ template <typename Func> void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) { using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type; static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>; - const auto segments = gpu_memory.GetSubmappedRange(image.gpu_addr, image.guest_size_bytes); + const auto segments = gpu_memory->GetSubmappedRange(image.gpu_addr, image.guest_size_bytes); for (const auto& [gpu_addr, size] : segments) { - std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); if constexpr (RETURNS_BOOL) { if (func(gpu_addr, *cpu_addr, size)) { @@ -1414,10 +1462,15 @@ void TextureCache<P>::RegisterImage(ImageId image_id) { tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format); } total_used_memory += Common::AlignUp(tentative_size, 1024); + if (total_used_memory > critical_memory && critical_gc < GC_EMERGENCY_COUNTS) { + RunGarbageCollector(); + critical_gc++; + } image.lru_index = lru_cache.Insert(image_id, frame_tick); - ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, - [this, image_id](u64 page) { gpu_page_table[page].push_back(image_id); }); + ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) { + (*channel_state->gpu_page_table)[page].push_back(image_id); + }); if (False(image.flags & ImageFlagBits::Sparse)) { auto map_id = slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id); @@ -1448,38 +1501,39 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) { image.flags &= ~ImageFlagBits::BadOverlap; lru_cache.Free(image.lru_index); const auto& clear_page_table = - [this, image_id]( - u64 page, - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>>& selected_page_table) { + [this, image_id](u64 page, + std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>& + selected_page_table) { const auto page_it = selected_page_table.find(page); if (page_it == selected_page_table.end()) { - UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS); + ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS); return; } std::vector<ImageId>& image_ids = page_it->second; const auto vector_it = std::ranges::find(image_ids, image_id); if (vector_it == image_ids.end()) { - UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}", - page << PAGE_BITS); + ASSERT_MSG(false, "Unregistering unregistered image in page=0x{:x}", + page << YUZU_PAGEBITS); return; } image_ids.erase(vector_it); }; - ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, - [this, &clear_page_table](u64 page) { clear_page_table(page, gpu_page_table); }); + ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) { + clear_page_table(page, (*channel_state->gpu_page_table)); + }); if (False(image.flags & ImageFlagBits::Sparse)) { const auto map_id = image.map_view_id; ForEachCPUPage(image.cpu_addr, image.guest_size_bytes, [this, map_id](u64 page) { const auto page_it = page_table.find(page); if (page_it == page_table.end()) { - UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS); + ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS); return; } std::vector<ImageMapId>& image_map_ids = page_it->second; const auto vector_it = std::ranges::find(image_map_ids, map_id); if (vector_it == image_map_ids.end()) { - UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}", - page << PAGE_BITS); + ASSERT_MSG(false, "Unregistering unregistered image in page=0x{:x}", + page << YUZU_PAGEBITS); return; } image_map_ids.erase(vector_it); @@ -1500,7 +1554,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) { ForEachCPUPage(cpu_addr, size, [this, image_id](u64 page) { const auto page_it = page_table.find(page); if (page_it == page_table.end()) { - UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS); + ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS); return; } std::vector<ImageMapId>& image_map_ids = page_it->second; @@ -1584,22 +1638,22 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) { const GPUVAddr gpu_addr = image.gpu_addr; const auto alloc_it = image_allocs_table.find(gpu_addr); if (alloc_it == image_allocs_table.end()) { - UNREACHABLE_MSG("Trying to delete an image alloc that does not exist in address 0x{:x}", - gpu_addr); + ASSERT_MSG(false, "Trying to delete an image alloc that does not exist in address 0x{:x}", + gpu_addr); return; } const ImageAllocId alloc_id = alloc_it->second; std::vector<ImageId>& alloc_images = slot_image_allocs[alloc_id].images; const auto alloc_image_it = std::ranges::find(alloc_images, image_id); if (alloc_image_it == alloc_images.end()) { - UNREACHABLE_MSG("Trying to delete an image that does not exist"); + ASSERT_MSG(false, "Trying to delete an image that does not exist"); return; } ASSERT_MSG(False(image.flags & ImageFlagBits::Tracked), "Image was not untracked"); ASSERT_MSG(False(image.flags & ImageFlagBits::Registered), "Image was not unregistered"); // Mark render targets as dirty - auto& dirty = maxwell3d.dirty.flags; + auto& dirty = maxwell3d->dirty.flags; dirty[Dirty::RenderTargets] = true; dirty[Dirty::ZetaBuffer] = true; for (size_t rt = 0; rt < NUM_RT; ++rt) { @@ -1649,24 +1703,30 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) { if (alloc_images.empty()) { image_allocs_table.erase(alloc_it); } - if constexpr (ENABLE_VALIDATION) { - std::ranges::fill(graphics_image_view_ids, CORRUPT_ID); - std::ranges::fill(compute_image_view_ids, CORRUPT_ID); + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + if constexpr (ENABLE_VALIDATION) { + std::ranges::fill(channel_info.graphics_image_view_ids, CORRUPT_ID); + std::ranges::fill(channel_info.compute_image_view_ids, CORRUPT_ID); + } + channel_info.graphics_image_table.Invalidate(); + channel_info.compute_image_table.Invalidate(); } - graphics_image_table.Invalidate(); - compute_image_table.Invalidate(); has_deleted_images = true; } template <class P> void TextureCache<P>::RemoveImageViewReferences(std::span<const ImageViewId> removed_views) { - auto it = image_views.begin(); - while (it != image_views.end()) { - const auto found = std::ranges::find(removed_views, it->second); - if (found != removed_views.end()) { - it = image_views.erase(it); - } else { - ++it; + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + auto it = channel_info.image_views.begin(); + while (it != channel_info.image_views.end()) { + const auto found = std::ranges::find(removed_views, it->second); + if (found != removed_views.end()) { + it = channel_info.image_views.erase(it); + } else { + ++it; + } } } } @@ -1697,6 +1757,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { boost::container::small_vector<const AliasedImage*, 1> aliased_images; Image& image = slot_images[image_id]; bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled); + bool any_modified = True(image.flags & ImageFlagBits::GpuModified); u64 most_recent_tick = image.modification_tick; for (const AliasedImage& aliased : image.aliased_images) { ImageBase& aliased_image = slot_images[aliased.id]; @@ -1704,6 +1765,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick); aliased_images.push_back(&aliased); any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled); + any_modified |= True(aliased_image.flags & ImageFlagBits::GpuModified); } } if (aliased_images.empty()) { @@ -1718,6 +1780,9 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { } } image.modification_tick = most_recent_tick; + if (any_modified) { + image.flags |= ImageFlagBits::GpuModified; + } std::ranges::sort(aliased_images, [this](const AliasedImage* lhs, const AliasedImage* rhs) { const ImageBase& lhs_image = slot_images[lhs->id]; const ImageBase& rhs_image = slot_images[rhs->id]; @@ -1725,7 +1790,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { }); const auto& resolution = Settings::values.resolution_info; for (const AliasedImage* const aliased : aliased_images) { - if (!resolution.active | !any_rescaled) { + if (!resolution.active || !any_rescaled) { CopyImage(image_id, aliased->id, aliased->copies); continue; } @@ -1736,19 +1801,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { continue; } ScaleUp(aliased_image); - - const bool both_2d{image.info.type == ImageType::e2D && - aliased_image.info.type == ImageType::e2D}; - auto copies = aliased->copies; - for (auto copy : copies) { - copy.extent.width = std::max<u32>( - (copy.extent.width * resolution.up_scale) >> resolution.down_shift, 1); - if (both_2d) { - copy.extent.height = std::max<u32>( - (copy.extent.height * resolution.up_scale) >> resolution.down_shift, 1); - } - } - CopyImage(image_id, aliased->id, copies); + CopyImage(image_id, aliased->id, aliased->copies); } } @@ -1908,6 +1961,7 @@ std::pair<FramebufferId, ImageViewId> TextureCache<P>::RenderTargetFromImage( .color_buffer_ids = {color_view_id}, .depth_buffer_id = depth_view_id, .size = {extent.width >> samples_x, extent.height >> samples_y}, + .is_rescaled = is_rescaled, }); return {framebuffer_id, view_id}; } @@ -1920,7 +1974,7 @@ bool TextureCache<P>::IsFullClear(ImageViewId id) { const ImageViewBase& image_view = slot_image_views[id]; const ImageBase& image = slot_images[image_view.image_id]; const Extent3D size = image_view.size; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; const auto& scissor = regs.scissor_test[0]; if (image.info.resources.levels > 1 || image.info.resources.layers > 1) { // Images with multiple resources can't be cleared in a single call @@ -1935,4 +1989,19 @@ bool TextureCache<P>::IsFullClear(ImageViewId id) { scissor.max_y >= size.height; } +template <class P> +void TextureCache<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) { + VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo>::CreateChannel(channel); + const auto it = channel_map.find(channel.bind_id); + auto* this_state = &channel_storage[it->second]; + const auto& this_as_ref = address_spaces[channel.memory_manager->GetID()]; + this_state->gpu_page_table = &gpu_page_table_storage[this_as_ref.storage_id]; +} + +/// Bind a channel for execution. +template <class P> +void TextureCache<P>::OnGPUASRegister([[maybe_unused]] size_t map_id) { + gpu_page_table_storage.emplace_back(); +} + } // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 7107887a6..2fa8445eb 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -1,9 +1,10 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once +#include <deque> +#include <limits> #include <mutex> #include <span> #include <type_traits> @@ -12,9 +13,11 @@ #include <queue> #include "common/common_types.h" +#include "common/hash.h" #include "common/literals.h" #include "common/lru_cache.h" #include "video_core/compatible_formats.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/engines/fermi_2d.h" #include "video_core/surface.h" @@ -22,12 +25,15 @@ #include "video_core/texture_cache/image_base.h" #include "video_core/texture_cache/image_info.h" #include "video_core/texture_cache/image_view_base.h" -#include "video_core/texture_cache/image_view_info.h" #include "video_core/texture_cache/render_targets.h" #include "video_core/texture_cache/slot_vector.h" #include "video_core/texture_cache/types.h" #include "video_core/textures/texture.h" +namespace Tegra::Control { +struct ChannelState; +} + namespace VideoCommon { using Tegra::Texture::SwizzleSource; @@ -46,10 +52,37 @@ struct ImageViewInOut { ImageViewId id{}; }; +using TextureCacheGPUMap = std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>; + +class TextureCacheChannelInfo : public ChannelInfo { +public: + TextureCacheChannelInfo() = delete; + TextureCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept; + TextureCacheChannelInfo(const TextureCacheChannelInfo& state) = delete; + TextureCacheChannelInfo& operator=(const TextureCacheChannelInfo&) = delete; + TextureCacheChannelInfo(TextureCacheChannelInfo&& other) noexcept = default; + TextureCacheChannelInfo& operator=(TextureCacheChannelInfo&& other) noexcept = default; + + DescriptorTable<TICEntry> graphics_image_table{gpu_memory}; + DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory}; + std::vector<SamplerId> graphics_sampler_ids; + std::vector<ImageViewId> graphics_image_view_ids; + + DescriptorTable<TICEntry> compute_image_table{gpu_memory}; + DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory}; + std::vector<SamplerId> compute_sampler_ids; + std::vector<ImageViewId> compute_image_view_ids; + + std::unordered_map<TICEntry, ImageViewId> image_views; + std::unordered_map<TSCEntry, SamplerId> samplers; + + TextureCacheGPUMap* gpu_page_table; +}; + template <class P> -class TextureCache { +class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo> { /// Address shift for caching images into a hash table - static constexpr u64 PAGE_BITS = 20; + static constexpr u64 YUZU_PAGEBITS = 20; /// Enables debugging features to the texture cache static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION; @@ -60,8 +93,12 @@ class TextureCache { /// True when the API can provide info about the memory of the device. static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO; - static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB; - static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB; + static constexpr size_t UNSET_CHANNEL{std::numeric_limits<size_t>::max()}; + + static constexpr s64 TARGET_THRESHOLD = 4_GiB; + static constexpr s64 DEFAULT_EXPECTED_MEMORY = 1_GiB + 125_MiB; + static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB + 625_MiB; + static constexpr size_t GC_EMERGENCY_COUNTS = 2; using Runtime = typename P::Runtime; using Image = typename P::Image; @@ -77,16 +114,8 @@ class TextureCache { PixelFormat src_format; }; - template <typename T> - struct IdentityHash { - [[nodiscard]] size_t operator()(T value) const noexcept { - return static_cast<size_t>(value); - } - }; - public: - explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&, Tegra::Engines::Maxwell3D&, - Tegra::Engines::KeplerCompute&, Tegra::MemoryManager&); + explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&); /// Notify the cache that a new frame has been queued void TickFrame(); @@ -142,7 +171,7 @@ public: void UnmapMemory(VAddr cpu_addr, size_t size); /// Remove images in a region - void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size); + void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size); /// Blit an image with the given parameters void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, @@ -171,6 +200,9 @@ public: [[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept; + /// Create channel state. + void CreateChannel(Tegra::Control::ChannelState& channel) final override; + std::mutex mutex; private: @@ -178,8 +210,8 @@ private: template <typename Func> static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) { static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>; - const u64 page_end = (addr + size - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) { + const u64 page_end = (addr + size - 1) >> YUZU_PAGEBITS; + for (u64 page = addr >> YUZU_PAGEBITS; page <= page_end; ++page) { if constexpr (RETURNS_BOOL) { if (func(page)) { break; @@ -193,8 +225,8 @@ private: template <typename Func> static void ForEachGPUPage(GPUVAddr addr, size_t size, Func&& func) { static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>; - const u64 page_end = (addr + size - 1) >> PAGE_BITS; - for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) { + const u64 page_end = (addr + size - 1) >> YUZU_PAGEBITS; + for (u64 page = addr >> YUZU_PAGEBITS; page <= page_end; ++page) { if constexpr (RETURNS_BOOL) { if (func(page)) { break; @@ -205,6 +237,8 @@ private: } } + void OnGPUASRegister(size_t map_id) final override; + /// Runs the Garbage Collector. void RunGarbageCollector(); @@ -273,7 +307,7 @@ private: void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func); template <typename Func> - void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func); + void ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, size_t size, Func&& func); template <typename Func> void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func); @@ -338,31 +372,16 @@ private: u64 GetScaledImageSizeBytes(ImageBase& image); Runtime& runtime; - VideoCore::RasterizerInterface& rasterizer; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; - DescriptorTable<TICEntry> graphics_image_table{gpu_memory}; - DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory}; - std::vector<SamplerId> graphics_sampler_ids; - std::vector<ImageViewId> graphics_image_view_ids; - - DescriptorTable<TICEntry> compute_image_table{gpu_memory}; - DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory}; - std::vector<SamplerId> compute_sampler_ids; - std::vector<ImageViewId> compute_image_view_ids; + VideoCore::RasterizerInterface& rasterizer; + std::deque<TextureCacheGPUMap> gpu_page_table_storage; RenderTargets render_targets; - std::unordered_map<TICEntry, ImageViewId> image_views; - std::unordered_map<TSCEntry, SamplerId> samplers; std::unordered_map<RenderTargets, FramebufferId> framebuffers; - std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table; - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table; - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table; - + std::unordered_map<u64, std::vector<ImageMapId>, Common::IdentityHash<u64>> page_table; + std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>> sparse_page_table; std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views; VAddr virtual_invalid_space{}; @@ -373,6 +392,7 @@ private: u64 minimum_memory; u64 expected_memory; u64 critical_memory; + size_t critical_gc; SlotVector<Image> slot_images; SlotVector<ImageMapView> slot_map_views; diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h index 5ac27b3a7..0453456b4 100644 --- a/src/video_core/texture_cache/types.h +++ b/src/video_core/texture_cache/types.h @@ -1,6 +1,5 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index d8e19cb2f..1223df5a0 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -1,31 +1,11 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-FileCopyrightText: Ryujinx Team and Contributors +// SPDX-License-Identifier: GPL-2.0-or-later AND MIT // This files contains code from Ryujinx // A copy of the code can be obtained from https://github.com/Ryujinx/Ryujinx // The sections using code from Ryujinx are marked with a link to the original version -// MIT License -// -// Copyright (c) Ryujinx Team and Contributors -// -// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and -// associated documentation files (the "Software"), to deal in the Software without restriction, -// including without limitation the rights to use, copy, modify, merge, publish, distribute, -// sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is -// furnished to do so, subject to the following conditions: -// -// The above copyright notice and this permission notice shall be included in all copies or -// substantial portions of the Software. -// -// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT -// NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, -// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. -// - #include <algorithm> #include <array> #include <numeric> @@ -537,7 +517,6 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr const u32 host_bytes_per_layer = num_blocks_per_layer * bytes_per_block; UNIMPLEMENTED_IF(info.tile_width_spacing > 0); - UNIMPLEMENTED_IF(copy.image_offset.x != 0); UNIMPLEMENTED_IF(copy.image_offset.y != 0); UNIMPLEMENTED_IF(copy.image_offset.z != 0); @@ -775,7 +754,7 @@ bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config if (address == 0) { return false; } - if (address > (1ULL << 48)) { + if (address >= (1ULL << 40)) { return false; } if (gpu_memory.GpuToCpuAddress(address).has_value()) { diff --git a/src/video_core/texture_cache/util.h b/src/video_core/texture_cache/util.h index 7af52de2e..5e28f4ab3 100644 --- a/src/video_core/texture_cache/util.h +++ b/src/video_core/texture_cache/util.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -9,10 +8,8 @@ #include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" #include "video_core/surface.h" #include "video_core/texture_cache/image_base.h" -#include "video_core/texture_cache/image_view_base.h" #include "video_core/texture_cache/types.h" #include "video_core/textures/texture.h" diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 25161df1f..15b9d4182 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -1,21 +1,11 @@ -// Copyright 2016 The University of North Carolina at Chapel Hill -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. -// +// SPDX-FileCopyrightText: 2016 The University of North Carolina at Chapel Hill +// SPDX-License-Identifier: Apache-2.0 + // Please send all BUG REPORTS to <pavel@cs.unc.edu>. // <http://gamma.cs.unc.edu/FasTC/> #include <algorithm> +#include <bit> #include <cassert> #include <cstring> #include <span> @@ -23,7 +13,9 @@ #include <boost/container/static_vector.hpp> +#include "common/alignment.h" #include "common/common_types.h" +#include "common/thread_worker.h" #include "video_core/textures/astc.h" class InputBitStream { @@ -1421,7 +1413,7 @@ static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 b static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) { for (u32 j = 0; j < blockHeight; j++) { for (u32 i = 0; i < blockWidth; i++) { - outBuf[j * blockWidth + i] = 0xFFFF00FF; + outBuf[j * blockWidth + i] = 0x00000000; } } } @@ -1660,29 +1652,41 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth, void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth, uint32_t block_width, uint32_t block_height, std::span<uint8_t> output) { - u32 block_index = 0; - std::size_t depth_offset = 0; - for (u32 z = 0; z < depth; z++) { - for (u32 y = 0; y < height; y += block_height) { - for (u32 x = 0; x < width; x += block_width) { - const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)}; - - // Blocks can be at most 12x12 - std::array<u32, 12 * 12> uncompData; - DecompressBlock(blockPtr, block_width, block_height, uncompData); - - u32 decompWidth = std::min(block_width, width - x); - u32 decompHeight = std::min(block_height, height - y); - - const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4); - for (u32 jj = 0; jj < decompHeight; jj++) { - std::memcpy(outRow.data() + jj * width * 4, - uncompData.data() + jj * block_width, decompWidth * 4); + const u32 rows = Common::DivideUp(height, block_height); + const u32 cols = Common::DivideUp(width, block_width); + + Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2, + "ASTCDecompress"}; + + for (u32 z = 0; z < depth; ++z) { + const u32 depth_offset = z * height * width * 4; + for (u32 y_index = 0; y_index < rows; ++y_index) { + auto decompress_stride = [data, width, height, depth, block_width, block_height, output, + rows, cols, z, depth_offset, y_index] { + const u32 y = y_index * block_height; + for (u32 x_index = 0; x_index < cols; ++x_index) { + const u32 block_index = (z * rows * cols) + (y_index * cols) + x_index; + const u32 x = x_index * block_width; + + const std::span<const u8, 16> blockPtr{data.subspan(block_index * 16, 16)}; + + // Blocks can be at most 12x12 + std::array<u32, 12 * 12> uncompData; + DecompressBlock(blockPtr, block_width, block_height, uncompData); + + u32 decompWidth = std::min(block_width, width - x); + u32 decompHeight = std::min(block_height, height - y); + + const std::span<u8> outRow = output.subspan(depth_offset + (y * width + x) * 4); + for (u32 h = 0; h < decompHeight; ++h) { + std::memcpy(outRow.data() + h * width * 4, + uncompData.data() + h * block_width, decompWidth * 4); + } } - ++block_index; - } + }; + workers.QueueWork(std::move(decompress_stride)); } - depth_offset += height * width * 4; + workers.WaitForRequests(); } } diff --git a/src/video_core/textures/astc.h b/src/video_core/textures/astc.h index 14d2beec0..afd3933c3 100644 --- a/src/video_core/textures/astc.h +++ b/src/video_core/textures/astc.h @@ -1,12 +1,8 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <bit> -#include "common/common_types.h" - namespace Tegra::Texture::ASTC { void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth, diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 24e943e4c..52d067a2d 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -1,12 +1,10 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <array> #include <cmath> #include <cstring> #include <span> -#include <utility> #include "common/alignment.h" #include "common/assert.h" @@ -14,13 +12,30 @@ #include "common/div_ceil.h" #include "video_core/gpu.h" #include "video_core/textures/decoders.h" -#include "video_core/textures/texture.h" namespace Tegra::Texture { namespace { +template <u32 mask> +constexpr u32 pdep(u32 value) { + u32 result = 0; + u32 m = mask; + for (u32 bit = 1; m; bit += bit) { + if (value & bit) + result |= m & -m; + m &= m - 1; + } + return result; +} + +template <u32 mask, u32 incr_amount> +void incrpdep(u32& value) { + constexpr u32 swizzled_incr = pdep<mask>(incr_amount); + value = ((value | ~mask) + swizzled_incr) & mask; +} + template <bool TO_LINEAR, u32 BYTES_PER_PIXEL> void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth, - u32 block_height, u32 block_depth, u32 stride_alignment) { + u32 block_height, u32 block_depth, u32 stride) { // The origin of the transformation can be configured here, leave it as zero as the current API // doesn't expose it. static constexpr u32 origin_x = 0; @@ -30,7 +45,6 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 // We can configure here a custom pitch // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch. const u32 pitch = width * BYTES_PER_PIXEL; - const u32 stride = Common::AlignUpLog2(width, stride_alignment) * BYTES_PER_PIXEL; const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT); const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); @@ -47,18 +61,20 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); for (u32 line = 0; line < height; ++line) { const u32 y = line + origin_y; - const auto& table = SWIZZLE_TABLE[y % GOB_SIZE_Y]; + const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(y); const u32 block_y = y >> GOB_SIZE_Y_SHIFT; const u32 offset_y = (block_y >> block_height) * block_size + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); - for (u32 column = 0; column < width; ++column) { + u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL); + for (u32 column = 0; column < width; + ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) { const u32 x = (column + origin_x) * BYTES_PER_PIXEL; const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; const u32 base_swizzled_offset = offset_z + offset_y + offset_x; - const u32 swizzled_offset = base_swizzled_offset + table[x % GOB_SIZE_X]; + const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y); const u32 unswizzled_offset = slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL; @@ -72,6 +88,69 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 } } +template <bool TO_LINEAR, u32 BYTES_PER_PIXEL> +void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, + u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 num_lines, + u32 block_height, u32 block_depth, u32 pitch_linear) { + // The origin of the transformation can be configured here, leave it as zero as the current API + // doesn't expose it. + static constexpr u32 origin_z = 0; + + // We can configure here a custom pitch + // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch. + const u32 pitch = pitch_linear; + const u32 stride = Common::AlignUpLog2(width * BYTES_PER_PIXEL, GOB_SIZE_X_SHIFT); + + const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT); + const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); + const u32 slice_size = + Common::DivCeilLog2(height, block_height + GOB_SIZE_Y_SHIFT) * block_size; + + const u32 block_height_mask = (1U << block_height) - 1; + const u32 block_depth_mask = (1U << block_depth) - 1; + const u32 x_shift = GOB_SIZE_SHIFT + block_height + block_depth; + + u32 unprocessed_lines = num_lines; + u32 extent_y = std::min(num_lines, height - origin_y); + + for (u32 slice = 0; slice < depth; ++slice) { + const u32 z = slice + origin_z; + const u32 offset_z = (z >> block_depth) * slice_size + + ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); + const u32 lines_in_y = std::min(unprocessed_lines, extent_y); + for (u32 line = 0; line < lines_in_y; ++line) { + const u32 y = line + origin_y; + const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(y); + + const u32 block_y = y >> GOB_SIZE_Y_SHIFT; + const u32 offset_y = (block_y >> block_height) * block_size + + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); + + u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL); + for (u32 column = 0; column < extent_x; + ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) { + const u32 x = (column + origin_x) * BYTES_PER_PIXEL; + const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; + + const u32 base_swizzled_offset = offset_z + offset_y + offset_x; + const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y); + + const u32 unswizzled_offset = + slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL; + + u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset]; + const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset]; + + std::memcpy(dst, src, BYTES_PER_PIXEL); + } + } + unprocessed_lines -= lines_in_y; + if (unprocessed_lines == 0) { + return; + } + } +} + template <bool TO_LINEAR> void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { @@ -90,118 +169,43 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe BPP_CASE(16) #undef BPP_CASE default: - UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel); - } -} - -template <u32 BYTES_PER_PIXEL> -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u8* swizzled_data, const u8* unswizzled_data, u32 block_height_bit, - u32 offset_x, u32 offset_y) { - const u32 block_height = 1U << block_height_bit; - const u32 image_width_in_gobs = - (swizzled_width * BYTES_PER_PIXEL + (GOB_SIZE_X - 1)) / GOB_SIZE_X; - for (u32 line = 0; line < subrect_height; ++line) { - const u32 dst_y = line + offset_y; - const u32 gob_address_y = - (dst_y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs + - ((dst_y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE; - const auto& table = SWIZZLE_TABLE[dst_y % GOB_SIZE_Y]; - for (u32 x = 0; x < subrect_width; ++x) { - const u32 dst_x = x + offset_x; - const u32 gob_address = - gob_address_y + (dst_x * BYTES_PER_PIXEL / GOB_SIZE_X) * GOB_SIZE * block_height; - const u32 swizzled_offset = gob_address + table[(dst_x * BYTES_PER_PIXEL) % GOB_SIZE_X]; - const u32 unswizzled_offset = line * source_pitch + x * BYTES_PER_PIXEL; - - const u8* const source_line = unswizzled_data + unswizzled_offset; - u8* const dest_addr = swizzled_data + swizzled_offset; - std::memcpy(dest_addr, source_line, BYTES_PER_PIXEL); - } + ASSERT_MSG(false, "Invalid bytes_per_pixel={}", bytes_per_pixel); } } -template <u32 BYTES_PER_PIXEL> -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 block_height, - u32 origin_x, u32 origin_y, u8* output, const u8* input) { - const u32 stride = width * BYTES_PER_PIXEL; - const u32 gobs_in_x = (stride + GOB_SIZE_X - 1) / GOB_SIZE_X; - const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height); - - const u32 block_height_mask = (1U << block_height) - 1; - const u32 x_shift = GOB_SIZE_SHIFT + block_height; - - for (u32 line = 0; line < line_count; ++line) { - const u32 src_y = line + origin_y; - const auto& table = SWIZZLE_TABLE[src_y % GOB_SIZE_Y]; - - const u32 block_y = src_y >> GOB_SIZE_Y_SHIFT; - const u32 src_offset_y = (block_y >> block_height) * block_size + - ((block_y & block_height_mask) << GOB_SIZE_SHIFT); - for (u32 column = 0; column < line_length_in; ++column) { - const u32 src_x = (column + origin_x) * BYTES_PER_PIXEL; - const u32 src_offset_x = (src_x >> GOB_SIZE_X_SHIFT) << x_shift; - - const u32 swizzled_offset = src_offset_y + src_offset_x + table[src_x % GOB_SIZE_X]; - const u32 unswizzled_offset = line * pitch + column * BYTES_PER_PIXEL; - - std::memcpy(output + unswizzled_offset, input + swizzled_offset, BYTES_PER_PIXEL); - } - } -} - -template <u32 BYTES_PER_PIXEL> -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 block_height, u32 block_depth, u32 origin_x, u32 origin_y, u8* output, - const u8* input) { - UNIMPLEMENTED_IF(origin_x > 0); - UNIMPLEMENTED_IF(origin_y > 0); - - const u32 stride = width * BYTES_PER_PIXEL; - const u32 gobs_in_x = (stride + GOB_SIZE_X - 1) / GOB_SIZE_X; - const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); - - const u32 block_height_mask = (1U << block_height) - 1; - const u32 x_shift = static_cast<u32>(GOB_SIZE_SHIFT) + block_height + block_depth; - - for (u32 line = 0; line < line_count; ++line) { - const auto& table = SWIZZLE_TABLE[line % GOB_SIZE_Y]; - const u32 block_y = line / GOB_SIZE_Y; - const u32 dst_offset_y = - (block_y >> block_height) * block_size + (block_y & block_height_mask) * GOB_SIZE; - for (u32 x = 0; x < line_length_in; ++x) { - const u32 dst_offset = - ((x / GOB_SIZE_X) << x_shift) + dst_offset_y + table[x % GOB_SIZE_X]; - const u32 src_offset = x * BYTES_PER_PIXEL + line * pitch; - std::memcpy(output + dst_offset, input + src_offset, BYTES_PER_PIXEL); - } - } -} } // Anonymous namespace void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { + const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel; + const u32 new_bpp = std::min(4U, static_cast<u32>(std::countr_zero(width * bytes_per_pixel))); + width = (width * bytes_per_pixel) >> new_bpp; + bytes_per_pixel = 1U << new_bpp; Swizzle<false>(output, input, bytes_per_pixel, width, height, depth, block_height, block_depth, - stride_alignment); + stride); } void SwizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { + const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel; + const u32 new_bpp = std::min(4U, static_cast<u32>(std::countr_zero(width * bytes_per_pixel))); + width = (width * bytes_per_pixel) >> new_bpp; + bytes_per_pixel = 1U << new_bpp; Swizzle<true>(output, input, bytes_per_pixel, width, height, depth, block_height, block_depth, - stride_alignment); + stride); } -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, u8* swizzled_data, const u8* unswizzled_data, - u32 block_height_bit, u32 offset_x, u32 offset_y) { +void SwizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, + u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 extent_y, + u32 block_height, u32 block_depth, u32 pitch_linear) { switch (bytes_per_pixel) { #define BPP_CASE(x) \ case x: \ - return SwizzleSubrect<x>(subrect_width, subrect_height, source_pitch, swizzled_width, \ - swizzled_data, unswizzled_data, block_height_bit, offset_x, \ - offset_y); + return SwizzleSubrectImpl<true, x>(output, input, width, height, depth, origin_x, \ + origin_y, extent_x, extent_y, block_height, \ + block_depth, pitch_linear); BPP_CASE(1) BPP_CASE(2) BPP_CASE(3) @@ -212,17 +216,19 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 BPP_CASE(16) #undef BPP_CASE default: - UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel); + ASSERT_MSG(false, "Invalid bytes_per_pixel={}", bytes_per_pixel); } } -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 bytes_per_pixel, - u32 block_height, u32 origin_x, u32 origin_y, u8* output, const u8* input) { +void UnswizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, + u32 extent_y, u32 block_height, u32 block_depth, u32 pitch_linear) { switch (bytes_per_pixel) { #define BPP_CASE(x) \ case x: \ - return UnswizzleSubrect<x>(line_length_in, line_count, pitch, width, block_height, \ - origin_x, origin_y, output, input); + return SwizzleSubrectImpl<false, x>(output, input, width, height, depth, origin_x, \ + origin_y, extent_x, extent_y, block_height, \ + block_depth, pitch_linear); BPP_CASE(1) BPP_CASE(2) BPP_CASE(3) @@ -233,54 +239,7 @@ void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, BPP_CASE(16) #undef BPP_CASE default: - UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel); - } -} - -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 bytes_per_pixel, u32 block_height, u32 block_depth, u32 origin_x, - u32 origin_y, u8* output, const u8* input) { - switch (bytes_per_pixel) { -#define BPP_CASE(x) \ - case x: \ - return SwizzleSliceToVoxel<x>(line_length_in, line_count, pitch, width, height, \ - block_height, block_depth, origin_x, origin_y, output, \ - input); - BPP_CASE(1) - BPP_CASE(2) - BPP_CASE(3) - BPP_CASE(4) - BPP_CASE(6) - BPP_CASE(8) - BPP_CASE(12) - BPP_CASE(16) -#undef BPP_CASE - default: - UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel); - } -} - -void SwizzleKepler(const u32 width, const u32 height, const u32 dst_x, const u32 dst_y, - const u32 block_height_bit, const std::size_t copy_size, const u8* source_data, - u8* swizzle_data) { - const u32 block_height = 1U << block_height_bit; - const u32 image_width_in_gobs{(width + GOB_SIZE_X - 1) / GOB_SIZE_X}; - std::size_t count = 0; - for (std::size_t y = dst_y; y < height && count < copy_size; ++y) { - const std::size_t gob_address_y = - (y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs + - ((y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE; - const auto& table = SWIZZLE_TABLE[y % GOB_SIZE_Y]; - for (std::size_t x = dst_x; x < width && count < copy_size; ++x) { - const std::size_t gob_address = - gob_address_y + (x / GOB_SIZE_X) * GOB_SIZE * block_height; - const std::size_t swizzled_offset = gob_address + table[x % GOB_SIZE_X]; - const u8* source_line = source_data + count; - u8* dest_addr = swizzle_data + swizzled_offset; - count++; - - *dest_addr = *source_line; - } + ASSERT_MSG(false, "Invalid bytes_per_pixel={}", bytes_per_pixel); } } diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 4c14cefbf..e70407692 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once @@ -21,6 +20,9 @@ constexpr u32 GOB_SIZE_Y_SHIFT = 3; constexpr u32 GOB_SIZE_Z_SHIFT = 0; constexpr u32 GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT; +constexpr u32 SWIZZLE_X_BITS = 0b100101111; +constexpr u32 SWIZZLE_Y_BITS = 0b011010000; + using SwizzleTable = std::array<std::array<u32, GOB_SIZE_X>, GOB_SIZE_Y>; /** @@ -38,7 +40,6 @@ constexpr SwizzleTable MakeSwizzleTable() { } return table; } -constexpr SwizzleTable SWIZZLE_TABLE = MakeSwizzleTable(); /// Unswizzles a block linear texture into linear memory. void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, @@ -55,34 +56,14 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height u32 block_height, u32 block_depth); /// Copies an untiled subrectangle into a tiled surface. -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, u8* swizzled_data, const u8* unswizzled_data, - u32 block_height_bit, u32 offset_x, u32 offset_y); +void SwizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, + u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 extent_y, + u32 block_height, u32 block_depth, u32 pitch_linear); /// Copies a tiled subrectangle into a linear surface. -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 bytes_per_pixel, - u32 block_height, u32 origin_x, u32 origin_y, u8* output, const u8* input); - -/// @brief Swizzles a 2D array of pixels into a 3D texture -/// @param line_length_in Number of pixels per line -/// @param line_count Number of lines -/// @param pitch Number of bytes per line -/// @param width Width of the swizzled texture -/// @param height Height of the swizzled texture -/// @param bytes_per_pixel Number of bytes used per pixel -/// @param block_height Block height shift -/// @param block_depth Block depth shift -/// @param origin_x Column offset in pixels of the swizzled texture -/// @param origin_y Row offset in pixels of the swizzled texture -/// @param output Pointer to the pixels of the swizzled texture -/// @param input Pointer to the 2D array of pixels used as input -/// @pre input and output points to an array large enough to hold the number of bytes used -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 bytes_per_pixel, u32 block_height, u32 block_depth, u32 origin_x, - u32 origin_y, u8* output, const u8* input); - -void SwizzleKepler(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height, - std::size_t copy_size, const u8* source_data, u8* swizzle_data); +void UnswizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, + u32 extent_y, u32 block_height, u32 block_depth, u32 pitch_linear); /// Obtains the offset of the gob for positions 'dst_x' & 'dst_y' u64 GetGOBOffset(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height, diff --git a/src/video_core/textures/texture.cpp b/src/video_core/textures/texture.cpp index 06954963d..b8327c88a 100644 --- a/src/video_core/textures/texture.cpp +++ b/src/video_core/textures/texture.cpp @@ -1,8 +1,6 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <algorithm> #include <array> #include "common/cityhash.h" diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index 7994cb859..7c4553a53 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/transform_feedback.cpp b/src/video_core/transform_feedback.cpp index ba26ac3f1..7e605981c 100644 --- a/src/video_core/transform_feedback.cpp +++ b/src/video_core/transform_feedback.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <array> diff --git a/src/video_core/transform_feedback.h b/src/video_core/transform_feedback.h index 8f6946d65..a519adb59 100644 --- a/src/video_core/transform_feedback.h +++ b/src/video_core/transform_feedback.h @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index 329bf4def..04ac4af11 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <memory> @@ -50,6 +49,7 @@ std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Cor gpu->BindRenderer(std::move(renderer)); return gpu; } catch (const std::runtime_error& exception) { + scope.Cancel(); LOG_ERROR(HW_GPU, "Failed to initialize GPU: {}", exception.what()); return nullptr; } diff --git a/src/video_core/video_core.h b/src/video_core/video_core.h index 084df641f..f8e2444f3 100644 --- a/src/video_core/video_core.h +++ b/src/video_core/video_core.h @@ -1,6 +1,5 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: 2014 Citra Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp index fdd1a5081..85f1d13e0 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #ifdef HAS_NSIGHT_AFTERMATH diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.h b/src/video_core/vulkan_common/nsight_aftermath_tracker.h index eae1891dd..9c1fb7720 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.h +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.h @@ -1,23 +1,24 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <filesystem> -#include <mutex> #include <span> -#include <string> -#include <vector> #include "common/common_types.h" -#include "common/dynamic_library.h" -#include "video_core/vulkan_common/vulkan_wrapper.h" #ifdef HAS_NSIGHT_AFTERMATH +#include <filesystem> +#include <mutex> + +// Vulkan headers must be included before Aftermath +#include "video_core/vulkan_common/vulkan_wrapper.h" + #include <GFSDK_Aftermath_Defines.h> #include <GFSDK_Aftermath_GpuCrashDump.h> #include <GFSDK_Aftermath_GpuCrashDumpDecoding.h> + +#include "common/dynamic_library.h" #endif namespace Vulkan { diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.cpp b/src/video_core/vulkan_common/vulkan_debug_callback.cpp index cf94e1d39..736474009 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.cpp +++ b/src/video_core/vulkan_common/vulkan_debug_callback.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <string_view> #include "common/logging/log.h" diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.h b/src/video_core/vulkan_common/vulkan_debug_callback.h index b0519f132..71b1f69ec 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.h +++ b/src/video_core/vulkan_common/vulkan_debug_callback.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 153702c0b..ddecfca13 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1,24 +1,24 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <bitset> #include <chrono> #include <optional> -#include <string_view> #include <thread> #include <unordered_set> #include <utility> #include <vector> #include "common/assert.h" +#include "common/literals.h" #include "common/settings.h" #include "video_core/vulkan_common/nsight_aftermath_tracker.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { +using namespace Common::Literals; namespace { namespace Alternatives { constexpr std::array STENCIL8_UINT{ @@ -39,6 +39,32 @@ constexpr std::array DEPTH16_UNORM_STENCIL8_UINT{ VK_FORMAT_D32_SFLOAT_S8_UINT, VK_FORMAT_UNDEFINED, }; + +constexpr std::array B5G6R5_UNORM_PACK16{ + VK_FORMAT_R5G6B5_UNORM_PACK16, + VK_FORMAT_UNDEFINED, +}; + +constexpr std::array R4G4_UNORM_PACK8{ + VK_FORMAT_R8_UNORM, + VK_FORMAT_UNDEFINED, +}; + +constexpr std::array R16G16B16_SFLOAT{ + VK_FORMAT_R16G16B16A16_SFLOAT, + VK_FORMAT_UNDEFINED, +}; + +constexpr std::array R16G16B16_SSCALED{ + VK_FORMAT_R16G16B16A16_SSCALED, + VK_FORMAT_UNDEFINED, +}; + +constexpr std::array R8G8B8_SSCALED{ + VK_FORMAT_R8G8B8A8_SSCALED, + VK_FORMAT_UNDEFINED, +}; + } // namespace Alternatives enum class NvidiaArchitecture { @@ -87,6 +113,16 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) { return Alternatives::DEPTH24_UNORM_STENCIL8_UINT.data(); case VK_FORMAT_D16_UNORM_S8_UINT: return Alternatives::DEPTH16_UNORM_STENCIL8_UINT.data(); + case VK_FORMAT_B5G6R5_UNORM_PACK16: + return Alternatives::B5G6R5_UNORM_PACK16.data(); + case VK_FORMAT_R4G4_UNORM_PACK8: + return Alternatives::R4G4_UNORM_PACK8.data(); + case VK_FORMAT_R16G16B16_SFLOAT: + return Alternatives::R16G16B16_SFLOAT.data(); + case VK_FORMAT_R16G16B16_SSCALED: + return Alternatives::R16G16B16_SSCALED.data(); + case VK_FORMAT_R8G8B8_SSCALED: + return Alternatives::R8G8B8_SSCALED.data(); default: return nullptr; } @@ -107,105 +143,142 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::PhysicalDevice physical) { static constexpr std::array formats{ - VK_FORMAT_A8B8G8R8_UNORM_PACK32, - VK_FORMAT_A8B8G8R8_UINT_PACK32, - VK_FORMAT_A8B8G8R8_SNORM_PACK32, + VK_FORMAT_A1R5G5B5_UNORM_PACK16, + VK_FORMAT_A2B10G10R10_SINT_PACK32, + VK_FORMAT_A2B10G10R10_SNORM_PACK32, + VK_FORMAT_A2B10G10R10_SSCALED_PACK32, + VK_FORMAT_A2B10G10R10_UINT_PACK32, + VK_FORMAT_A2B10G10R10_UNORM_PACK32, + VK_FORMAT_A2B10G10R10_USCALED_PACK32, VK_FORMAT_A8B8G8R8_SINT_PACK32, + VK_FORMAT_A8B8G8R8_SNORM_PACK32, VK_FORMAT_A8B8G8R8_SRGB_PACK32, - VK_FORMAT_R5G6B5_UNORM_PACK16, + VK_FORMAT_A8B8G8R8_UINT_PACK32, + VK_FORMAT_A8B8G8R8_UNORM_PACK32, + VK_FORMAT_ASTC_10x10_SRGB_BLOCK, + VK_FORMAT_ASTC_10x10_UNORM_BLOCK, + VK_FORMAT_ASTC_10x5_SRGB_BLOCK, + VK_FORMAT_ASTC_10x5_UNORM_BLOCK, + VK_FORMAT_ASTC_10x6_SRGB_BLOCK, + VK_FORMAT_ASTC_10x6_UNORM_BLOCK, + VK_FORMAT_ASTC_10x8_SRGB_BLOCK, + VK_FORMAT_ASTC_10x8_UNORM_BLOCK, + VK_FORMAT_ASTC_12x10_SRGB_BLOCK, + VK_FORMAT_ASTC_12x10_UNORM_BLOCK, + VK_FORMAT_ASTC_12x12_SRGB_BLOCK, + VK_FORMAT_ASTC_12x12_UNORM_BLOCK, + VK_FORMAT_ASTC_4x4_SRGB_BLOCK, + VK_FORMAT_ASTC_4x4_UNORM_BLOCK, + VK_FORMAT_ASTC_5x4_SRGB_BLOCK, + VK_FORMAT_ASTC_5x4_UNORM_BLOCK, + VK_FORMAT_ASTC_5x5_SRGB_BLOCK, + VK_FORMAT_ASTC_5x5_UNORM_BLOCK, + VK_FORMAT_ASTC_6x5_SRGB_BLOCK, + VK_FORMAT_ASTC_6x5_UNORM_BLOCK, + VK_FORMAT_ASTC_6x6_SRGB_BLOCK, + VK_FORMAT_ASTC_6x6_UNORM_BLOCK, + VK_FORMAT_ASTC_8x5_SRGB_BLOCK, + VK_FORMAT_ASTC_8x5_UNORM_BLOCK, + VK_FORMAT_ASTC_8x6_SRGB_BLOCK, + VK_FORMAT_ASTC_8x6_UNORM_BLOCK, + VK_FORMAT_ASTC_8x8_SRGB_BLOCK, + VK_FORMAT_ASTC_8x8_UNORM_BLOCK, + VK_FORMAT_B10G11R11_UFLOAT_PACK32, + VK_FORMAT_B4G4R4A4_UNORM_PACK16, + VK_FORMAT_B5G5R5A1_UNORM_PACK16, VK_FORMAT_B5G6R5_UNORM_PACK16, - VK_FORMAT_A2B10G10R10_UNORM_PACK32, - VK_FORMAT_A2B10G10R10_UINT_PACK32, - VK_FORMAT_A1R5G5B5_UNORM_PACK16, - VK_FORMAT_R32G32B32A32_SFLOAT, - VK_FORMAT_R32G32B32A32_SINT, - VK_FORMAT_R32G32B32A32_UINT, - VK_FORMAT_R32G32_SFLOAT, - VK_FORMAT_R32G32_SINT, - VK_FORMAT_R32G32_UINT, + VK_FORMAT_B8G8R8A8_SRGB, + VK_FORMAT_B8G8R8A8_UNORM, + VK_FORMAT_BC1_RGBA_SRGB_BLOCK, + VK_FORMAT_BC1_RGBA_UNORM_BLOCK, + VK_FORMAT_BC2_SRGB_BLOCK, + VK_FORMAT_BC2_UNORM_BLOCK, + VK_FORMAT_BC3_SRGB_BLOCK, + VK_FORMAT_BC3_UNORM_BLOCK, + VK_FORMAT_BC4_SNORM_BLOCK, + VK_FORMAT_BC4_UNORM_BLOCK, + VK_FORMAT_BC5_SNORM_BLOCK, + VK_FORMAT_BC5_UNORM_BLOCK, + VK_FORMAT_BC6H_SFLOAT_BLOCK, + VK_FORMAT_BC6H_UFLOAT_BLOCK, + VK_FORMAT_BC7_SRGB_BLOCK, + VK_FORMAT_BC7_UNORM_BLOCK, + VK_FORMAT_D16_UNORM, + VK_FORMAT_D16_UNORM_S8_UINT, + VK_FORMAT_D24_UNORM_S8_UINT, + VK_FORMAT_D32_SFLOAT, + VK_FORMAT_D32_SFLOAT_S8_UINT, + VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, + VK_FORMAT_R16G16B16A16_SFLOAT, VK_FORMAT_R16G16B16A16_SINT, - VK_FORMAT_R16G16B16A16_UINT, VK_FORMAT_R16G16B16A16_SNORM, + VK_FORMAT_R16G16B16A16_SSCALED, + VK_FORMAT_R16G16B16A16_UINT, VK_FORMAT_R16G16B16A16_UNORM, - VK_FORMAT_R16G16_UNORM, - VK_FORMAT_R16G16_SNORM, + VK_FORMAT_R16G16B16A16_USCALED, + VK_FORMAT_R16G16B16_SFLOAT, + VK_FORMAT_R16G16B16_SINT, + VK_FORMAT_R16G16B16_SNORM, + VK_FORMAT_R16G16B16_SSCALED, + VK_FORMAT_R16G16B16_UINT, + VK_FORMAT_R16G16B16_UNORM, + VK_FORMAT_R16G16B16_USCALED, VK_FORMAT_R16G16_SFLOAT, - VK_FORMAT_R16G16_UINT, VK_FORMAT_R16G16_SINT, - VK_FORMAT_R16_UNORM, + VK_FORMAT_R16G16_SNORM, + VK_FORMAT_R16G16_SSCALED, + VK_FORMAT_R16G16_UINT, + VK_FORMAT_R16G16_UNORM, + VK_FORMAT_R16G16_USCALED, + VK_FORMAT_R16_SFLOAT, + VK_FORMAT_R16_SINT, VK_FORMAT_R16_SNORM, + VK_FORMAT_R16_SSCALED, VK_FORMAT_R16_UINT, + VK_FORMAT_R16_UNORM, + VK_FORMAT_R16_USCALED, + VK_FORMAT_R32G32B32A32_SFLOAT, + VK_FORMAT_R32G32B32A32_SINT, + VK_FORMAT_R32G32B32A32_UINT, + VK_FORMAT_R32G32B32_SFLOAT, + VK_FORMAT_R32G32B32_SINT, + VK_FORMAT_R32G32B32_UINT, + VK_FORMAT_R32G32_SFLOAT, + VK_FORMAT_R32G32_SINT, + VK_FORMAT_R32G32_UINT, + VK_FORMAT_R32_SFLOAT, + VK_FORMAT_R32_SINT, + VK_FORMAT_R32_UINT, + VK_FORMAT_R4G4B4A4_UNORM_PACK16, + VK_FORMAT_R4G4_UNORM_PACK8, + VK_FORMAT_R5G5B5A1_UNORM_PACK16, + VK_FORMAT_R5G6B5_UNORM_PACK16, + VK_FORMAT_R8G8B8A8_SINT, + VK_FORMAT_R8G8B8A8_SNORM, VK_FORMAT_R8G8B8A8_SRGB, - VK_FORMAT_R8G8_UNORM, - VK_FORMAT_R8G8_SNORM, + VK_FORMAT_R8G8B8A8_SSCALED, + VK_FORMAT_R8G8B8A8_UINT, + VK_FORMAT_R8G8B8A8_UNORM, + VK_FORMAT_R8G8B8A8_USCALED, + VK_FORMAT_R8G8B8_SINT, + VK_FORMAT_R8G8B8_SNORM, + VK_FORMAT_R8G8B8_SSCALED, + VK_FORMAT_R8G8B8_UINT, + VK_FORMAT_R8G8B8_UNORM, + VK_FORMAT_R8G8B8_USCALED, VK_FORMAT_R8G8_SINT, + VK_FORMAT_R8G8_SNORM, + VK_FORMAT_R8G8_SSCALED, VK_FORMAT_R8G8_UINT, - VK_FORMAT_R8_UNORM, - VK_FORMAT_R8_SNORM, + VK_FORMAT_R8G8_UNORM, + VK_FORMAT_R8G8_USCALED, VK_FORMAT_R8_SINT, + VK_FORMAT_R8_SNORM, + VK_FORMAT_R8_SSCALED, VK_FORMAT_R8_UINT, - VK_FORMAT_B10G11R11_UFLOAT_PACK32, - VK_FORMAT_R32_SFLOAT, - VK_FORMAT_R32_UINT, - VK_FORMAT_R32_SINT, - VK_FORMAT_R16_SFLOAT, - VK_FORMAT_R16G16B16A16_SFLOAT, - VK_FORMAT_B8G8R8A8_UNORM, - VK_FORMAT_B8G8R8A8_SRGB, - VK_FORMAT_R4G4B4A4_UNORM_PACK16, - VK_FORMAT_D32_SFLOAT, - VK_FORMAT_D16_UNORM, + VK_FORMAT_R8_UNORM, + VK_FORMAT_R8_USCALED, VK_FORMAT_S8_UINT, - VK_FORMAT_D16_UNORM_S8_UINT, - VK_FORMAT_D24_UNORM_S8_UINT, - VK_FORMAT_D32_SFLOAT_S8_UINT, - VK_FORMAT_BC1_RGBA_UNORM_BLOCK, - VK_FORMAT_BC2_UNORM_BLOCK, - VK_FORMAT_BC3_UNORM_BLOCK, - VK_FORMAT_BC4_UNORM_BLOCK, - VK_FORMAT_BC4_SNORM_BLOCK, - VK_FORMAT_BC5_UNORM_BLOCK, - VK_FORMAT_BC5_SNORM_BLOCK, - VK_FORMAT_BC7_UNORM_BLOCK, - VK_FORMAT_BC6H_UFLOAT_BLOCK, - VK_FORMAT_BC6H_SFLOAT_BLOCK, - VK_FORMAT_BC1_RGBA_SRGB_BLOCK, - VK_FORMAT_BC2_SRGB_BLOCK, - VK_FORMAT_BC3_SRGB_BLOCK, - VK_FORMAT_BC7_SRGB_BLOCK, - VK_FORMAT_ASTC_4x4_UNORM_BLOCK, - VK_FORMAT_ASTC_4x4_SRGB_BLOCK, - VK_FORMAT_ASTC_5x4_UNORM_BLOCK, - VK_FORMAT_ASTC_5x4_SRGB_BLOCK, - VK_FORMAT_ASTC_5x5_UNORM_BLOCK, - VK_FORMAT_ASTC_5x5_SRGB_BLOCK, - VK_FORMAT_ASTC_6x5_UNORM_BLOCK, - VK_FORMAT_ASTC_6x5_SRGB_BLOCK, - VK_FORMAT_ASTC_6x6_UNORM_BLOCK, - VK_FORMAT_ASTC_6x6_SRGB_BLOCK, - VK_FORMAT_ASTC_8x5_UNORM_BLOCK, - VK_FORMAT_ASTC_8x5_SRGB_BLOCK, - VK_FORMAT_ASTC_8x6_UNORM_BLOCK, - VK_FORMAT_ASTC_8x6_SRGB_BLOCK, - VK_FORMAT_ASTC_8x8_UNORM_BLOCK, - VK_FORMAT_ASTC_8x8_SRGB_BLOCK, - VK_FORMAT_ASTC_10x5_UNORM_BLOCK, - VK_FORMAT_ASTC_10x5_SRGB_BLOCK, - VK_FORMAT_ASTC_10x6_UNORM_BLOCK, - VK_FORMAT_ASTC_10x6_SRGB_BLOCK, - VK_FORMAT_ASTC_10x8_UNORM_BLOCK, - VK_FORMAT_ASTC_10x8_SRGB_BLOCK, - VK_FORMAT_ASTC_10x10_UNORM_BLOCK, - VK_FORMAT_ASTC_10x10_SRGB_BLOCK, - VK_FORMAT_ASTC_12x10_UNORM_BLOCK, - VK_FORMAT_ASTC_12x10_SRGB_BLOCK, - VK_FORMAT_ASTC_12x12_UNORM_BLOCK, - VK_FORMAT_ASTC_12x12_SRGB_BLOCK, - VK_FORMAT_ASTC_8x6_UNORM_BLOCK, - VK_FORMAT_ASTC_8x6_SRGB_BLOCK, - VK_FORMAT_ASTC_6x5_UNORM_BLOCK, - VK_FORMAT_ASTC_6x5_SRGB_BLOCK, - VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, }; std::unordered_map<VkFormat, VkFormatProperties> format_properties; for (const auto format : formats) { @@ -224,9 +297,14 @@ std::vector<std::string> GetSupportedExtensions(vk::PhysicalDevice physical) { return supported_extensions; } +bool IsExtensionSupported(std::span<const std::string> supported_extensions, + std::string_view extension) { + return std::ranges::find(supported_extensions, extension) != supported_extensions.end(); +} + NvidiaArchitecture GetNvidiaArchitecture(vk::PhysicalDevice physical, std::span<const std::string> exts) { - if (std::ranges::find(exts, VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME) != exts.end()) { + if (IsExtensionSupported(exts, VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME)) { VkPhysicalDeviceFragmentShadingRatePropertiesKHR shading_rate_props{}; shading_rate_props.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADING_RATE_PROPERTIES_KHR; @@ -239,7 +317,7 @@ NvidiaArchitecture GetNvidiaArchitecture(vk::PhysicalDevice physical, return NvidiaArchitecture::AmpereOrNewer; } } - if (std::ranges::find(exts, VK_NV_SHADING_RATE_IMAGE_EXTENSION_NAME) != exts.end()) { + if (IsExtensionSupported(exts, VK_NV_SHADING_RATE_IMAGE_EXTENSION_NAME)) { return NvidiaArchitecture::Turing; } return NvidiaArchitecture::VoltaOrOlder; @@ -542,7 +620,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR } VkPhysicalDeviceWorkgroupMemoryExplicitLayoutFeaturesKHR workgroup_layout; - if (khr_workgroup_memory_explicit_layout) { + if (khr_workgroup_memory_explicit_layout && is_shader_int16_supported) { workgroup_layout = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_FEATURES_KHR, @@ -553,6 +631,11 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR .workgroupMemoryExplicitLayout16BitAccess = VK_TRUE, }; SetNext(next, workgroup_layout); + } else if (khr_workgroup_memory_explicit_layout) { + // TODO(lat9nq): Find a proper fix for this + LOG_WARNING(Render_Vulkan, "Disabling VK_KHR_workgroup_memory_explicit_layout due to a " + "yuzu bug when host driver does not support 16-bit integers"); + khr_workgroup_memory_explicit_layout = false; } VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR executable_properties; @@ -585,6 +668,11 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR } logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld); + is_integrated = properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU; + is_virtual = properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_VIRTUAL_GPU; + is_non_gpu = properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_OTHER || + properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_CPU; + CollectPhysicalMemoryInfo(); CollectTelemetryParameters(); CollectToolingInfo(); @@ -603,8 +691,14 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR khr_push_descriptor = false; break; } + const u32 nv_major_version = (properties.driverVersion >> 22) & 0x3ff; + if (nv_major_version >= 510) { + LOG_WARNING(Render_Vulkan, "NVIDIA Drivers >= 510 do not support MSAA image blits"); + cant_blit_msaa = true; + } } - if (ext_extended_dynamic_state && driver_id == VK_DRIVER_ID_MESA_RADV) { + const bool is_radv = driver_id == VK_DRIVER_ID_MESA_RADV; + if (ext_extended_dynamic_state && is_radv) { // Mask driver version variant const u32 version = (properties.driverVersion << 3) >> 3; if (version < VK_MAKE_API_VERSION(0, 21, 2, 0)) { @@ -613,6 +707,17 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR ext_extended_dynamic_state = false; } } + if (ext_vertex_input_dynamic_state && is_radv) { + // TODO(ameerj): Blacklist only offending driver versions + // TODO(ameerj): Confirm if RDNA1 is affected + const bool is_rdna2 = + IsExtensionSupported(supported_extensions, VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME); + if (is_rdna2) { + LOG_WARNING(Render_Vulkan, + "RADV has broken VK_EXT_vertex_input_dynamic_state on RDNA2 hardware"); + ext_vertex_input_dynamic_state = false; + } + } sets_per_pool = 64; const bool is_amd = @@ -628,7 +733,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR has_broken_cube_compatibility = true; } } - const bool is_amd_or_radv = is_amd || driver_id == VK_DRIVER_ID_MESA_RADV; + const bool is_amd_or_radv = is_amd || is_radv; if (ext_sampler_filter_minmax && is_amd_or_radv) { // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken. if (!is_float16_supported) { @@ -639,6 +744,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR } const bool is_intel_windows = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS; + const bool is_intel_anv = driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA; if (ext_vertex_input_dynamic_state && is_intel_windows) { LOG_WARNING(Render_Vulkan, "Blacklisting Intel for VK_EXT_vertex_input_dynamic_state"); ext_vertex_input_dynamic_state = false; @@ -652,6 +758,10 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_WARNING(Render_Vulkan, "Intel proprietary drivers do not support MSAA image blits"); cant_blit_msaa = true; } + if (is_intel_anv) { + LOG_WARNING(Render_Vulkan, "ANV driver does not support native BGR format"); + must_emulate_bgr565 = true; + } supports_d24_depth = IsFormatSupported(VK_FORMAT_D24_UNORM_S8_UINT, @@ -671,9 +781,10 @@ VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags // The wanted format is not supported by hardware, search for alternatives const VkFormat* alternatives = GetFormatAlternatives(wanted_format); if (alternatives == nullptr) { - UNREACHABLE_MSG("Format={} with usage={} and type={} has no defined alternatives and host " - "hardware does not support it", - wanted_format, wanted_usage, format_type); + ASSERT_MSG(false, + "Format={} with usage={} and type={} has no defined alternatives and host " + "hardware does not support it", + wanted_format, wanted_usage, format_type); return wanted_format; } @@ -682,21 +793,22 @@ VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags if (!IsFormatSupported(alternative, wanted_usage, format_type)) { continue; } - LOG_WARNING(Render_Vulkan, - "Emulating format={} with alternative format={} with usage={} and type={}", - wanted_format, alternative, wanted_usage, format_type); + LOG_DEBUG(Render_Vulkan, + "Emulating format={} with alternative format={} with usage={} and type={}", + wanted_format, alternative, wanted_usage, format_type); return alternative; } // No alternatives found, panic - UNREACHABLE_MSG("Format={} with usage={} and type={} is not supported by the host hardware and " - "doesn't support any of the alternatives", - wanted_format, wanted_usage, format_type); + ASSERT_MSG(false, + "Format={} with usage={} and type={} is not supported by the host hardware and " + "doesn't support any of the alternatives", + wanted_format, wanted_usage, format_type); return wanted_format; } void Device::ReportLoss() const { - LOG_CRITICAL(Render_Vulkan, "Device loss occured!"); + LOG_CRITICAL(Render_Vulkan, "Device loss occurred!"); // Wait for the log to flush and for Nsight Aftermath to dump the results std::this_thread::sleep_for(std::chrono::seconds{15}); @@ -957,6 +1069,7 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { test(has_khr_swapchain_mutable_format, VK_KHR_SWAPCHAIN_MUTABLE_FORMAT_EXTENSION_NAME, false); test(has_ext_line_rasterization, VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME, false); + test(ext_memory_budget, VK_EXT_MEMORY_BUDGET_EXTENSION_NAME, true); if (Settings::values.enable_nsight_aftermath) { test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true); @@ -969,7 +1082,7 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { VkPhysicalDeviceFeatures2KHR features{}; features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR; - VkPhysicalDeviceProperties2KHR physical_properties; + VkPhysicalDeviceProperties2KHR physical_properties{}; physical_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR; if (has_khr_shader_float16_int8) { @@ -1239,15 +1352,50 @@ void Device::CollectTelemetryParameters() { vendor_name = driver.driverName; } +u64 Device::GetDeviceMemoryUsage() const { + VkPhysicalDeviceMemoryBudgetPropertiesEXT budget; + budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT; + budget.pNext = nullptr; + physical.GetMemoryProperties(&budget); + u64 result{}; + for (const size_t heap : valid_heap_memory) { + result += budget.heapUsage[heap]; + } + return result; +} + void Device::CollectPhysicalMemoryInfo() { - const auto mem_properties = physical.GetMemoryProperties(); + VkPhysicalDeviceMemoryBudgetPropertiesEXT budget{}; + budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT; + const auto mem_info = physical.GetMemoryProperties(ext_memory_budget ? &budget : nullptr); + const auto& mem_properties = mem_info.memoryProperties; const size_t num_properties = mem_properties.memoryHeapCount; device_access_memory = 0; + u64 device_initial_usage = 0; + u64 local_memory = 0; for (size_t element = 0; element < num_properties; ++element) { - if ((mem_properties.memoryHeaps[element].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) != 0) { - device_access_memory += mem_properties.memoryHeaps[element].size; + const bool is_heap_local = + (mem_properties.memoryHeaps[element].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) != 0; + if (!is_integrated && !is_heap_local) { + continue; + } + valid_heap_memory.push_back(element); + if (is_heap_local) { + local_memory += mem_properties.memoryHeaps[element].size; + } + if (ext_memory_budget) { + device_initial_usage += budget.heapUsage[element]; + device_access_memory += budget.heapBudget[element]; + continue; } + device_access_memory += mem_properties.memoryHeaps[element].size; + } + if (!is_integrated) { + return; } + const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage); + device_access_memory = static_cast<u64>(std::max<s64>( + std::min<s64>(available_memory - 8_GiB, 4_GiB), static_cast<s64>(local_memory))); } void Device::CollectToolingInfo() { diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 37d140ebd..d7cc6c593 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -1,12 +1,10 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <span> #include <string> -#include <string_view> #include <unordered_map> #include <vector> @@ -342,6 +340,12 @@ public: return device_access_memory; } + bool CanReportMemoryUsage() const { + return ext_memory_budget; + } + + u64 GetDeviceMemoryUsage() const; + u32 GetSetsPerPool() const { return sets_per_pool; } @@ -354,6 +358,10 @@ public: return cant_blit_msaa; } + bool MustEmulateBGR565() const { + return must_emulate_bgr565; + } + private: /// Checks if the physical device is suitable. void CheckSuitability(bool requires_swapchain) const; @@ -418,6 +426,9 @@ private: bool is_topology_list_restart_supported{}; ///< Support for primitive restart with list ///< topologies. bool is_patch_list_restart_supported{}; ///< Support for primitive restart with list patch. + bool is_integrated{}; ///< Is GPU an iGPU. + bool is_virtual{}; ///< Is GPU a virtual GPU. + bool is_non_gpu{}; ///< Is SoftwareRasterizer, FPGA, non-GPU device. bool nv_viewport_swizzle{}; ///< Support for VK_NV_viewport_swizzle. bool nv_viewport_array2{}; ///< Support for VK_NV_viewport_array2. bool nv_geometry_shader_passthrough{}; ///< Support for VK_NV_geometry_shader_passthrough. @@ -442,16 +453,19 @@ private: bool ext_shader_atomic_int64{}; ///< Support for VK_KHR_shader_atomic_int64. bool ext_conservative_rasterization{}; ///< Support for VK_EXT_conservative_rasterization. bool ext_provoking_vertex{}; ///< Support for VK_EXT_provoking_vertex. + bool ext_memory_budget{}; ///< Support for VK_EXT_memory_budget. bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. bool has_broken_cube_compatibility{}; ///< Has broken cube compatiblity bit bool has_renderdoc{}; ///< Has RenderDoc attached bool has_nsight_graphics{}; ///< Has Nsight Graphics attached bool supports_d24_depth{}; ///< Supports D24 depth buffers. bool cant_blit_msaa{}; ///< Does not support MSAA<->MSAA blitting. + bool must_emulate_bgr565{}; ///< Emulates BGR565 by swizzling RGB565 format. // Telemetry parameters std::string vendor_name; ///< Device's driver name. std::vector<std::string> supported_extensions; ///< Reported Vulkan extensions. + std::vector<size_t> valid_heap_memory; ///< Heaps used. /// Format properties dictionary. std::unordered_map<VkFormat, VkFormatProperties> format_properties; diff --git a/src/video_core/vulkan_common/vulkan_instance.cpp b/src/video_core/vulkan_common/vulkan_instance.cpp index bfd6e6add..a082e3059 100644 --- a/src/video_core/vulkan_common/vulkan_instance.cpp +++ b/src/video_core/vulkan_common/vulkan_instance.cpp @@ -1,12 +1,9 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <algorithm> #include <future> #include <optional> #include <span> -#include <utility> #include <vector> #include "common/common_types.h" diff --git a/src/video_core/vulkan_common/vulkan_instance.h b/src/video_core/vulkan_common/vulkan_instance.h index e5e3a7144..40419d802 100644 --- a/src/video_core/vulkan_common/vulkan_instance.h +++ b/src/video_core/vulkan_common/vulkan_instance.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/vulkan_common/vulkan_library.cpp b/src/video_core/vulkan_common/vulkan_library.cpp index 22833fa56..4eb3913ee 100644 --- a/src/video_core/vulkan_common/vulkan_library.cpp +++ b/src/video_core/vulkan_common/vulkan_library.cpp @@ -1,17 +1,17 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <cstdlib> #include <string> #include "common/dynamic_library.h" #include "common/fs/path_util.h" +#include "common/logging/log.h" #include "video_core/vulkan_common/vulkan_library.h" namespace Vulkan { Common::DynamicLibrary OpenLibrary() { + LOG_DEBUG(Render_Vulkan, "Looking for a Vulkan library"); Common::DynamicLibrary library; #ifdef __APPLE__ // Check if a path to a specific Vulkan library has been specified. @@ -24,9 +24,11 @@ Common::DynamicLibrary OpenLibrary() { } #else std::string filename = Common::DynamicLibrary::GetVersionedFilename("vulkan", 1); + LOG_DEBUG(Render_Vulkan, "Trying Vulkan library: {}", filename); if (!library.Open(filename.c_str())) { // Android devices may not have libvulkan.so.1, only libvulkan.so. filename = Common::DynamicLibrary::GetVersionedFilename("vulkan"); + LOG_DEBUG(Render_Vulkan, "Trying Vulkan library (second attempt): {}", filename); void(library.Open(filename.c_str())); } #endif diff --git a/src/video_core/vulkan_common/vulkan_library.h b/src/video_core/vulkan_common/vulkan_library.h index 8b28b0e17..364ca979b 100644 --- a/src/video_core/vulkan_common/vulkan_library.h +++ b/src/video_core/vulkan_common/vulkan_library.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index 300a61205..6442898bd 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -1,6 +1,5 @@ -// Copyright 2018 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <bit> @@ -50,7 +49,7 @@ struct Range { return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT; } - UNREACHABLE_MSG("Invalid memory usage={}", usage); + ASSERT_MSG(false, "Invalid memory usage={}", usage); return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; } @@ -227,7 +226,7 @@ void MemoryCommit::Release() { } MemoryAllocator::MemoryAllocator(const Device& device_, bool export_allocations_) - : device{device_}, properties{device_.GetPhysical().GetMemoryProperties()}, + : device{device_}, properties{device_.GetPhysical().GetMemoryProperties().memoryProperties}, export_allocations{export_allocations_}, buffer_image_granularity{ device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {} @@ -326,7 +325,7 @@ VkMemoryPropertyFlags MemoryAllocator::MemoryPropertyFlags(u32 type_mask, // Remove device local, if it's not supported by the requested resource return MemoryPropertyFlags(type_mask, flags & ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT); } - UNREACHABLE_MSG("No compatible memory types found"); + ASSERT_MSG(false, "No compatible memory types found"); return 0; } @@ -350,7 +349,7 @@ bool IsHostVisible(MemoryUsage usage) noexcept { case MemoryUsage::Download: return true; } - UNREACHABLE_MSG("Invalid memory usage={}", usage); + ASSERT_MSG(false, "Invalid memory usage={}", usage); return false; } diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.h b/src/video_core/vulkan_common/vulkan_memory_allocator.h index 86e8ed119..a5bff03fe 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.h +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.h @@ -1,12 +1,10 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <memory> #include <span> -#include <utility> #include <vector> #include "common/common_types.h" #include "video_core/vulkan_common/vulkan_wrapper.h" diff --git a/src/video_core/vulkan_common/vulkan_surface.cpp b/src/video_core/vulkan_common/vulkan_surface.cpp index 3c3238f96..69f9c494b 100644 --- a/src/video_core/vulkan_common/vulkan_surface.cpp +++ b/src/video_core/vulkan_common/vulkan_surface.cpp @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/logging/log.h" #include "core/frontend/emu_window.h" diff --git a/src/video_core/vulkan_common/vulkan_surface.h b/src/video_core/vulkan_common/vulkan_surface.h index 05a169e32..5725143e6 100644 --- a/src/video_core/vulkan_common/vulkan_surface.h +++ b/src/video_core/vulkan_common/vulkan_surface.h @@ -1,6 +1,5 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index a9faa4807..2ad98dcfe 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -1,12 +1,9 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> -#include <exception> #include <memory> #include <optional> -#include <string_view> #include <utility> #include <vector> @@ -239,8 +236,8 @@ bool Load(VkInstance instance, InstanceDispatch& dld) noexcept { return X(vkCreateDevice) && X(vkDestroyDevice) && X(vkDestroyDevice) && X(vkEnumerateDeviceExtensionProperties) && X(vkEnumeratePhysicalDevices) && X(vkGetDeviceProcAddr) && X(vkGetPhysicalDeviceFormatProperties) && - X(vkGetPhysicalDeviceMemoryProperties) && X(vkGetPhysicalDeviceProperties) && - X(vkGetPhysicalDeviceQueueFamilyProperties); + X(vkGetPhysicalDeviceMemoryProperties) && X(vkGetPhysicalDeviceMemoryProperties2) && + X(vkGetPhysicalDeviceProperties) && X(vkGetPhysicalDeviceQueueFamilyProperties); #undef X } @@ -328,6 +325,8 @@ const char* ToString(VkResult result) noexcept { return "VK_PIPELINE_COMPILE_REQUIRED_EXT"; case VkResult::VK_RESULT_MAX_ENUM: return "VK_RESULT_MAX_ENUM"; + case VkResult::VK_ERROR_COMPRESSION_EXHAUSTED_EXT: + return "VK_ERROR_COMPRESSION_EXHAUSTED_EXT"; } return "Unknown"; } @@ -928,9 +927,12 @@ std::vector<VkPresentModeKHR> PhysicalDevice::GetSurfacePresentModesKHR( return modes; } -VkPhysicalDeviceMemoryProperties PhysicalDevice::GetMemoryProperties() const noexcept { - VkPhysicalDeviceMemoryProperties properties; - dld->vkGetPhysicalDeviceMemoryProperties(physical_device, &properties); +VkPhysicalDeviceMemoryProperties2 PhysicalDevice::GetMemoryProperties( + void* next_structures) const noexcept { + VkPhysicalDeviceMemoryProperties2 properties{}; + properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_PROPERTIES_2; + properties.pNext = next_structures; + dld->vkGetPhysicalDeviceMemoryProperties2(physical_device, &properties); return properties; } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index b7ae01c6c..1b3f493bd 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -1,11 +1,9 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once #include <exception> -#include <iterator> #include <limits> #include <memory> #include <optional> @@ -173,6 +171,7 @@ struct InstanceDispatch { PFN_vkGetPhysicalDeviceFeatures2KHR vkGetPhysicalDeviceFeatures2KHR{}; PFN_vkGetPhysicalDeviceFormatProperties vkGetPhysicalDeviceFormatProperties{}; PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties{}; + PFN_vkGetPhysicalDeviceMemoryProperties2 vkGetPhysicalDeviceMemoryProperties2{}; PFN_vkGetPhysicalDeviceProperties vkGetPhysicalDeviceProperties{}; PFN_vkGetPhysicalDeviceProperties2KHR vkGetPhysicalDeviceProperties2KHR{}; PFN_vkGetPhysicalDeviceQueueFamilyProperties vkGetPhysicalDeviceQueueFamilyProperties{}; @@ -520,9 +519,7 @@ public: dld{rhs.dld} {} /// Assign an allocation transfering ownership from another allocation. - /// Releases any previously held allocation. PoolAllocations& operator=(PoolAllocations&& rhs) noexcept { - Release(); allocations = std::move(rhs.allocations); num = rhs.num; device = rhs.device; @@ -531,11 +528,6 @@ public: return *this; } - /// Destroys any held allocation. - ~PoolAllocations() { - Release(); - } - /// Returns the number of allocations. std::size_t size() const noexcept { return num; @@ -558,19 +550,6 @@ public: } private: - /// Destroys the held allocations if they exist. - void Release() noexcept { - if (!allocations) { - return; - } - const Span<AllocationType> span(allocations.get(), num); - const VkResult result = Free(device, pool, span, *dld); - // There's no way to report errors from a destructor. - if (result != VK_SUCCESS) { - std::terminate(); - } - } - std::unique_ptr<AllocationType[]> allocations; std::size_t num = 0; VkDevice device = nullptr; @@ -951,7 +930,8 @@ public: std::vector<VkPresentModeKHR> GetSurfacePresentModesKHR(VkSurfaceKHR) const; - VkPhysicalDeviceMemoryProperties GetMemoryProperties() const noexcept; + VkPhysicalDeviceMemoryProperties2 GetMemoryProperties( + void* next_structures = nullptr) const noexcept; private: VkPhysicalDevice physical_device = nullptr; |