diff options
Diffstat (limited to 'src/video_core')
41 files changed, 472 insertions, 413 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 2f6cdd216..269db21a5 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -231,6 +231,7 @@ endif() target_include_directories(video_core PRIVATE ${FFmpeg_INCLUDE_DIR}) target_link_libraries(video_core PRIVATE ${FFmpeg_LIBRARIES}) +target_link_options(video_core PRIVATE ${FFmpeg_LDFLAGS}) add_dependencies(video_core host_shaders) target_include_directories(video_core PRIVATE ${HOST_SHADERS_INCLUDE}) diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 7bfd57369..d350c9b36 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -570,13 +570,12 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am ForEachWrittenRange(*cpu_src_address, amount, mirror); // This subtraction in this order is important for overlapping copies. common_ranges.subtract(subtract_interval); - bool atleast_1_download = tmp_intervals.size() != 0; - for (const IntervalType add_interval : tmp_intervals) { + const bool has_new_downloads = tmp_intervals.size() != 0; + for (const IntervalType& add_interval : tmp_intervals) { common_ranges.add(add_interval); } - runtime.CopyBuffer(dest_buffer, src_buffer, copies); - if (atleast_1_download) { + if (has_new_downloads) { dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount); } std::vector<u8> tmp_buffer(amount); diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/command_classes/codecs/codec.cpp index f798a0053..61966cbfe 100644 --- a/src/video_core/command_classes/codecs/codec.cpp +++ b/src/video_core/command_classes/codecs/codec.cpp @@ -5,6 +5,7 @@ #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/vp9.h" @@ -16,108 +17,146 @@ extern "C" { } namespace Tegra { -#if defined(LIBVA_FOUND) -// Hardware acceleration code from FFmpeg/doc/examples/hw_decode.c originally under MIT license namespace { -constexpr std::array<const char*, 2> VAAPI_DRIVERS = { - "i915", - "amdgpu", -}; +constexpr AVPixelFormat PREFERRED_GPU_FMT = AV_PIX_FMT_NV12; +constexpr AVPixelFormat PREFERRED_CPU_FMT = AV_PIX_FMT_YUV420P; + +void AVPacketDeleter(AVPacket* ptr) { + av_packet_free(&ptr); +} -AVPixelFormat GetHwFormat(AVCodecContext*, const AVPixelFormat* pix_fmts) { +using AVPacketPtr = std::unique_ptr<AVPacket, decltype(&AVPacketDeleter)>; + +AVPixelFormat GetGpuFormat(AVCodecContext* av_codec_ctx, const AVPixelFormat* pix_fmts) { for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) { - if (*p == AV_PIX_FMT_VAAPI) { - return AV_PIX_FMT_VAAPI; + if (*p == av_codec_ctx->pix_fmt) { + return av_codec_ctx->pix_fmt; } } LOG_INFO(Service_NVDRV, "Could not find compatible GPU AV format, falling back to CPU"); - return *pix_fmts; + av_buffer_unref(&av_codec_ctx->hw_device_ctx); + av_codec_ctx->pix_fmt = PREFERRED_CPU_FMT; + return PREFERRED_CPU_FMT; +} +} // namespace + +void AVFrameDeleter(AVFrame* ptr) { + av_frame_free(&ptr); } -bool CreateVaapiHwdevice(AVBufferRef** av_hw_device) { +Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs) + : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)), + vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {} + +Codec::~Codec() { + if (!initialized) { + return; + } + // Free libav memory + avcodec_free_context(&av_codec_ctx); + av_buffer_unref(&av_gpu_decoder); +} + +bool Codec::CreateGpuAvDevice() { +#if defined(LIBVA_FOUND) + static constexpr std::array<const char*, 3> VAAPI_DRIVERS = { + "i915", + "iHD", + "amdgpu", + }; AVDictionary* hwdevice_options = nullptr; av_dict_set(&hwdevice_options, "connection_type", "drm", 0); for (const auto& driver : VAAPI_DRIVERS) { av_dict_set(&hwdevice_options, "kernel_driver", driver, 0); - const int hwdevice_error = av_hwdevice_ctx_create(av_hw_device, AV_HWDEVICE_TYPE_VAAPI, + const int hwdevice_error = av_hwdevice_ctx_create(&av_gpu_decoder, AV_HWDEVICE_TYPE_VAAPI, nullptr, hwdevice_options, 0); if (hwdevice_error >= 0) { LOG_INFO(Service_NVDRV, "Using VA-API with {}", driver); av_dict_free(&hwdevice_options); + av_codec_ctx->pix_fmt = AV_PIX_FMT_VAAPI; return true; } LOG_DEBUG(Service_NVDRV, "VA-API av_hwdevice_ctx_create failed {}", hwdevice_error); } LOG_DEBUG(Service_NVDRV, "VA-API av_hwdevice_ctx_create failed for all drivers"); av_dict_free(&hwdevice_options); - return false; -} -} // namespace #endif - -void AVFrameDeleter(AVFrame* ptr) { - av_frame_free(&ptr); + static constexpr auto HW_CONFIG_METHOD = AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX; + static constexpr std::array GPU_DECODER_TYPES{ + AV_HWDEVICE_TYPE_CUDA, +#ifdef _WIN32 + AV_HWDEVICE_TYPE_D3D11VA, +#else + AV_HWDEVICE_TYPE_VDPAU, +#endif + }; + for (const auto& type : GPU_DECODER_TYPES) { + 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 {}", + av_hwdevice_get_type_name(type), hwdevice_res); + continue; + } + for (int i = 0;; i++) { + const AVCodecHWConfig* config = avcodec_get_hw_config(av_codec, i); + if (!config) { + LOG_DEBUG(Service_NVDRV, "{} decoder does not support device type {}.", + 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; + LOG_INFO(Service_NVDRV, "Using {} GPU decoder", av_hwdevice_get_type_name(type)); + return true; + } + } + } + return false; } -Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs) - : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)), - vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {} - -Codec::~Codec() { - if (!initialized) { - return; - } - // Free libav memory - avcodec_send_packet(av_codec_ctx, nullptr); - AVFrame* av_frame = av_frame_alloc(); - avcodec_receive_frame(av_codec_ctx, av_frame); - avcodec_flush_buffers(av_codec_ctx); - av_frame_free(&av_frame); - avcodec_close(av_codec_ctx); - av_buffer_unref(&av_hw_device); +void Codec::InitializeAvCodecContext() { + av_codec_ctx = avcodec_alloc_context3(av_codec); + av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0); } -void Codec::InitializeHwdec() { - // Prioritize integrated GPU to mitigate bandwidth bottlenecks -#if defined(LIBVA_FOUND) - if (CreateVaapiHwdevice(&av_hw_device)) { - const auto hw_device_ctx = av_buffer_ref(av_hw_device); - ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed"); - av_codec_ctx->hw_device_ctx = hw_device_ctx; - av_codec_ctx->get_format = GetHwFormat; +void Codec::InitializeGpuDecoder() { + if (!CreateGpuAvDevice()) { + av_buffer_unref(&av_gpu_decoder); return; } -#endif - // TODO more GPU accelerated decoders + auto* hw_device_ctx = av_buffer_ref(av_gpu_decoder); + ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed"); + av_codec_ctx->hw_device_ctx = hw_device_ctx; + av_codec_ctx->get_format = GetGpuFormat; } void Codec::Initialize() { - AVCodecID codec; - switch (current_codec) { - case NvdecCommon::VideoCodec::H264: - codec = AV_CODEC_ID_H264; - break; - case NvdecCommon::VideoCodec::Vp9: - codec = AV_CODEC_ID_VP9; - break; - default: - UNIMPLEMENTED_MSG("Unknown codec {}", current_codec); + const AVCodecID codec = [&] { + switch (current_codec) { + case NvdecCommon::VideoCodec::H264: + return AV_CODEC_ID_H264; + case NvdecCommon::VideoCodec::Vp9: + return AV_CODEC_ID_VP9; + default: + UNIMPLEMENTED_MSG("Unknown codec {}", current_codec); + return AV_CODEC_ID_NONE; + } + }(); + av_codec = avcodec_find_decoder(codec); + + InitializeAvCodecContext(); + if (Settings::values.nvdec_emulation.GetValue() == Settings::NvdecEmulation::GPU) { + InitializeGpuDecoder(); + } + if (const int res = avcodec_open2(av_codec_ctx, av_codec, nullptr); res < 0) { + LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed with result {}", res); + avcodec_free_context(&av_codec_ctx); + av_buffer_unref(&av_gpu_decoder); return; } - av_codec = avcodec_find_decoder(codec); - av_codec_ctx = avcodec_alloc_context3(av_codec); - av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0); - InitializeHwdec(); if (!av_codec_ctx->hw_device_ctx) { LOG_INFO(Service_NVDRV, "Using FFmpeg software decoding"); } - const auto av_error = avcodec_open2(av_codec_ctx, av_codec, nullptr); - if (av_error < 0) { - LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed."); - avcodec_close(av_codec_ctx); - av_buffer_unref(&av_hw_device); - return; - } initialized = true; } @@ -133,6 +172,9 @@ void Codec::Decode() { if (is_first_frame) { Initialize(); } + if (!initialized) { + return; + } bool vp9_hidden_frame = false; std::vector<u8> frame_data; if (current_codec == NvdecCommon::VideoCodec::H264) { @@ -141,50 +183,48 @@ void Codec::Decode() { frame_data = vp9_decoder->ComposeFrameHeader(state); vp9_hidden_frame = vp9_decoder->WasFrameHidden(); } - AVPacket packet{}; - av_init_packet(&packet); - packet.data = frame_data.data(); - packet.size = static_cast<s32>(frame_data.size()); - if (const int ret = avcodec_send_packet(av_codec_ctx, &packet); ret) { - LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", ret); + AVPacketPtr packet{av_packet_alloc(), AVPacketDeleter}; + if (!packet) { + LOG_ERROR(Service_NVDRV, "av_packet_alloc failed"); + return; + } + packet->data = frame_data.data(); + packet->size = static_cast<s32>(frame_data.size()); + if (const int res = avcodec_send_packet(av_codec_ctx, packet.get()); res != 0) { + LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", res); return; } // Only receive/store visible frames if (vp9_hidden_frame) { return; } - AVFrame* hw_frame = av_frame_alloc(); - AVFrame* sw_frame = hw_frame; - ASSERT_MSG(hw_frame, "av_frame_alloc hw_frame failed"); - if (const int ret = avcodec_receive_frame(av_codec_ctx, hw_frame); ret) { + AVFramePtr initial_frame{av_frame_alloc(), AVFrameDeleter}; + AVFramePtr final_frame{nullptr, AVFrameDeleter}; + ASSERT_MSG(initial_frame, "av_frame_alloc initial_frame failed"); + if (const int ret = avcodec_receive_frame(av_codec_ctx, initial_frame.get()); ret) { LOG_DEBUG(Service_NVDRV, "avcodec_receive_frame error {}", ret); - av_frame_free(&hw_frame); return; } - if (!hw_frame->width || !hw_frame->height) { + if (initial_frame->width == 0 || initial_frame->height == 0) { LOG_WARNING(Service_NVDRV, "Zero width or height in frame"); - av_frame_free(&hw_frame); return; } -#if defined(LIBVA_FOUND) - // Hardware acceleration code from FFmpeg/doc/examples/hw_decode.c under MIT license - if (hw_frame->format == AV_PIX_FMT_VAAPI) { - sw_frame = av_frame_alloc(); - ASSERT_MSG(sw_frame, "av_frame_alloc sw_frame failed"); + if (av_codec_ctx->hw_device_ctx) { + final_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter}; + ASSERT_MSG(final_frame, "av_frame_alloc final_frame failed"); // Can't use AV_PIX_FMT_YUV420P and share code with software decoding in vic.cpp // because Intel drivers crash unless using AV_PIX_FMT_NV12 - sw_frame->format = AV_PIX_FMT_NV12; - const int transfer_data_ret = av_hwframe_transfer_data(sw_frame, hw_frame, 0); - ASSERT_MSG(!transfer_data_ret, "av_hwframe_transfer_data error {}", transfer_data_ret); - av_frame_free(&hw_frame); + final_frame->format = PREFERRED_GPU_FMT; + const int ret = av_hwframe_transfer_data(final_frame.get(), initial_frame.get(), 0); + ASSERT_MSG(!ret, "av_hwframe_transfer_data error {}", ret); + } else { + final_frame = std::move(initial_frame); } -#endif - if (sw_frame->format != AV_PIX_FMT_YUV420P && sw_frame->format != AV_PIX_FMT_NV12) { - UNIMPLEMENTED_MSG("Unexpected video format from host graphics: {}", sw_frame->format); - av_frame_free(&sw_frame); + if (final_frame->format != PREFERRED_CPU_FMT && final_frame->format != PREFERRED_GPU_FMT) { + UNIMPLEMENTED_MSG("Unexpected video format: {}", final_frame->format); return; } - av_frames.push(AVFramePtr{sw_frame, AVFrameDeleter}); + av_frames.push(std::move(final_frame)); if (av_frames.size() > 10) { LOG_TRACE(Service_NVDRV, "av_frames.push overflow dropped frame"); av_frames.pop(); diff --git a/src/video_core/command_classes/codecs/codec.h b/src/video_core/command_classes/codecs/codec.h index 71936203f..f9a80886f 100644 --- a/src/video_core/command_classes/codecs/codec.h +++ b/src/video_core/command_classes/codecs/codec.h @@ -5,6 +5,7 @@ #pragma once #include <memory> +#include <string_view> #include <queue> #include "common/common_types.h" #include "video_core/command_classes/nvdec_common.h" @@ -50,18 +51,23 @@ public: /// Returns the value of current_codec [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const; + /// Return name of the current codec [[nodiscard]] std::string_view GetCurrentCodecName() const; private: - void InitializeHwdec(); + void InitializeAvCodecContext(); + + void InitializeGpuDecoder(); + + bool CreateGpuAvDevice(); bool initialized{}; NvdecCommon::VideoCodec current_codec{NvdecCommon::VideoCodec::None}; AVCodec* av_codec{nullptr}; - AVBufferRef* av_hw_device{nullptr}; AVCodecContext* av_codec_ctx{nullptr}; + AVBufferRef* av_gpu_decoder{nullptr}; GPU& gpu; const NvdecCommon::NvdecRegisters& state; diff --git a/src/video_core/command_classes/codecs/h264.cpp b/src/video_core/command_classes/codecs/h264.cpp index 5fb6d45ee..51ee14c13 100644 --- a/src/video_core/command_classes/codecs/h264.cpp +++ b/src/video_core/command_classes/codecs/h264.cpp @@ -95,7 +95,8 @@ const std::vector<u8>& H264::ComposeFrameHeader(const NvdecCommon::NvdecRegister const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units / (context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2); - writer.WriteUe(16); + // TODO (ameerj): Where do we get this number, it seems to be particular for each stream + writer.WriteUe(6); // Max number of reference frames writer.WriteBit(false); writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1); writer.WriteUe(pic_height - 1); diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 1aa43523a..7f4ca6282 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -475,10 +475,10 @@ public: // These values are used by Nouveau and some games. AddGL = 0x8006, - SubtractGL = 0x8007, - ReverseSubtractGL = 0x8008, - MinGL = 0x800a, - MaxGL = 0x800b + MinGL = 0x8007, + MaxGL = 0x8008, + SubtractGL = 0x800a, + ReverseSubtractGL = 0x800b }; enum class Factor : u32 { diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index c7ec1eac9..67388d980 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -82,41 +82,41 @@ void MaxwellDMA::Launch() { } void MaxwellDMA::CopyPitchToPitch() { - // When `multi_line_enable` bit is disabled the copy is performed as if we were copying a 1D - // buffer of length `line_length_in`. - // Otherwise we copy a 2D image of dimensions (line_length_in, line_count). - auto& accelerate = rasterizer->AccessAccelerateDMA(); - if (!regs.launch_dma.multi_line_enable) { - const bool is_buffer_clear = regs.launch_dma.remap_enable != 0 && - regs.remap_const.dst_x == RemapConst::Swizzle::CONST_A; - // TODO: allow multisized components. - if (is_buffer_clear) { - ASSERT(regs.remap_const.component_size_minus_one == 3); - accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value); - std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value); - memory_manager.WriteBlockUnsafe(regs.offset_out, - reinterpret_cast<u8*>(tmp_buffer.data()), - regs.line_length_in * sizeof(u32)); - return; - } - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); - if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) { - std::vector<u8> tmp_buffer(regs.line_length_in); - memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(), regs.line_length_in); - memory_manager.WriteBlock(regs.offset_out, tmp_buffer.data(), regs.line_length_in); + // When `multi_line_enable` bit is enabled we copy a 2D image of dimensions + // (line_length_in, line_count). + // Otherwise the copy is performed as if we were copying a 1D buffer of length line_length_in. + const bool remap_enabled = regs.launch_dma.remap_enable != 0; + if (regs.launch_dma.multi_line_enable) { + UNIMPLEMENTED_IF(remap_enabled); + + // Perform a line-by-line copy. + // We're going to take a subrect of size (line_length_in, line_count) from the source + // rectangle. There is no need to manually flush/invalidate the regions because CopyBlock + // does that for us. + for (u32 line = 0; line < regs.line_count; ++line) { + const GPUVAddr source_line = regs.offset_in + static_cast<size_t>(line) * regs.pitch_in; + const GPUVAddr dest_line = regs.offset_out + static_cast<size_t>(line) * regs.pitch_out; + memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in); } return; } - - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); - - // Perform a line-by-line copy. - // We're going to take a subrect of size (line_length_in, line_count) from the source rectangle. - // There is no need to manually flush/invalidate the regions because CopyBlock does that for us. - for (u32 line = 0; line < regs.line_count; ++line) { - const GPUVAddr source_line = regs.offset_in + static_cast<size_t>(line) * regs.pitch_in; - const GPUVAddr dest_line = regs.offset_out + static_cast<size_t>(line) * regs.pitch_out; - memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in); + // TODO: allow multisized components. + auto& accelerate = rasterizer->AccessAccelerateDMA(); + const bool is_const_a_dst = regs.remap_const.dst_x == RemapConst::Swizzle::CONST_A; + const bool is_buffer_clear = remap_enabled && is_const_a_dst; + if (is_buffer_clear) { + ASSERT(regs.remap_const.component_size_minus_one == 3); + accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value); + std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value); + memory_manager.WriteBlockUnsafe(regs.offset_out, reinterpret_cast<u8*>(tmp_buffer.data()), + regs.line_length_in * sizeof(u32)); + return; + } + UNIMPLEMENTED_IF(remap_enabled); + if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) { + std::vector<u8> tmp_buffer(regs.line_length_in); + memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(), regs.line_length_in); + memory_manager.WriteBlock(regs.offset_out, tmp_buffer.data(), regs.line_length_in); } } diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 9e457ae16..a04514425 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -175,7 +175,7 @@ public: static_assert(sizeof(LaunchDMA) == 4); struct RemapConst { - enum Swizzle : u32 { + enum class Swizzle : u32 { SRC_X = 0, SRC_Y = 1, SRC_Z = 2, diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index ff024f530..2ae3639b5 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -531,14 +531,6 @@ void GPU::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const { interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value); } -void GPU::ShutDown() { - // Signal that threads should no longer block on syncpoint fences - shutting_down.store(true, std::memory_order_relaxed); - sync_cv.notify_all(); - - gpu_thread.ShutDown(); -} - void GPU::OnCommandListEnd() { if (is_async) { // This command only applies to asynchronous GPU mode diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index a8e98e51b..e6a02a71b 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -219,9 +219,6 @@ public: return *shader_notify; } - // Stops the GPU execution and waits for the GPU to finish working - void ShutDown(); - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. void WaitFence(u32 syncpoint_id, u32 value); diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 46f642b19..9547f277a 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -17,9 +17,9 @@ namespace VideoCommon::GPUThread { /// Runs the GPU thread -static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, - Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher, - SynchState& state) { +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"; MicroProfileOnThreadCreate(name.c_str()); SCOPE_EXIT({ MicroProfileOnThreadExit(); }); @@ -28,20 +28,14 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, Common::SetCurrentThreadPriority(Common::ThreadPriority::High); system.RegisterHostThread(); - // Wait for first GPU command before acquiring the window context - state.queue.Wait(); - - // If emulation was stopped during disk shader loading, abort before trying to acquire context - if (!state.is_running) { - return; - } - auto current_context = context.Acquire(); VideoCore::RasterizerInterface* const rasterizer = renderer.ReadRasterizer(); - CommandDataContainer next; - while (state.is_running) { - next = state.queue.PopWait(); + while (!stop_token.stop_requested()) { + CommandDataContainer next = state.queue.PopWait(stop_token); + if (stop_token.stop_requested()) { + break; + } if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) { dma_pusher.Push(std::move(submit_list->entries)); dma_pusher.DispatchCalls(); @@ -55,8 +49,6 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, rasterizer->FlushRegion(flush->addr, flush->size); } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); - } else if (std::holds_alternative<EndProcessingCommand>(next.data)) { - ASSERT(state.is_running == false); } else { UNREACHABLE(); } @@ -73,16 +65,14 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, ThreadManager::ThreadManager(Core::System& system_, bool is_async_) : system{system_}, is_async{is_async_} {} -ThreadManager::~ThreadManager() { - ShutDown(); -} +ThreadManager::~ThreadManager() = default; void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher) { rasterizer = renderer.ReadRasterizer(); - thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), - std::ref(dma_pusher), std::ref(state)); + thread = std::jthread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), + std::ref(dma_pusher), std::ref(state)); } void ThreadManager::SubmitList(Tegra::CommandList&& entries) { @@ -117,26 +107,6 @@ void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) { rasterizer->OnCPUWrite(addr, size); } -void ThreadManager::ShutDown() { - if (!state.is_running) { - return; - } - - { - std::lock_guard lk(state.write_lock); - state.is_running = false; - state.cv.notify_all(); - } - - if (!thread.joinable()) { - return; - } - - // Notify GPU thread that a shutdown is pending - PushCommand(EndProcessingCommand()); - thread.join(); -} - void ThreadManager::OnCommandListEnd() { PushCommand(OnCommandListEndCommand()); } @@ -152,9 +122,8 @@ u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) { state.queue.Push(CommandDataContainer(std::move(command_data), fence, block)); if (block) { - state.cv.wait(lk, [this, fence] { - return fence <= state.signaled_fence.load(std::memory_order_relaxed) || - !state.is_running; + state.cv.wait(lk, thread.get_stop_token(), [this, fence] { + return fence <= state.signaled_fence.load(std::memory_order_relaxed); }); } diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 11a648f38..91bada925 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -33,9 +33,6 @@ class RendererBase; namespace VideoCommon::GPUThread { -/// Command to signal to the GPU thread that processing has ended -struct EndProcessingCommand final {}; - /// 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_)} {} @@ -83,7 +80,7 @@ struct OnCommandListEndCommand final {}; struct GPUTickCommand final {}; using CommandData = - std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand, + std::variant<std::monostate, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand, InvalidateRegionCommand, FlushAndInvalidateRegionCommand, OnCommandListEndCommand, GPUTickCommand>; @@ -100,14 +97,12 @@ struct CommandDataContainer { /// Struct used to synchronize the GPU thread struct SynchState final { - std::atomic_bool is_running{true}; - - using CommandQueue = Common::SPSCQueue<CommandDataContainer>; + using CommandQueue = Common::SPSCQueue<CommandDataContainer, true>; std::mutex write_lock; CommandQueue queue; u64 last_fence{}; std::atomic<u64> signaled_fence{}; - std::condition_variable cv; + std::condition_variable_any cv; }; /// Class used to manage the GPU thread @@ -149,7 +144,7 @@ private: VideoCore::RasterizerInterface* rasterizer = nullptr; SynchState state; - std::thread thread; + std::jthread thread; }; } // namespace VideoCommon::GPUThread diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index c9cff7450..20d748c12 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -6,7 +6,6 @@ set(SHADER_FILES convert_float_to_depth.frag full_screen_triangle.vert opengl_copy_bc4.comp - opengl_copy_bgra.comp opengl_present.frag opengl_present.vert pitch_unswizzle.comp diff --git a/src/video_core/host_shaders/opengl_copy_bgra.comp b/src/video_core/host_shaders/opengl_copy_bgra.comp deleted file mode 100644 index 2571a4abf..000000000 --- a/src/video_core/host_shaders/opengl_copy_bgra.comp +++ /dev/null @@ -1,15 +0,0 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#version 430 core - -layout (local_size_x = 4, local_size_y = 4) in; - -layout(binding = 0, rgba8) readonly uniform image2DArray bgr_input; -layout(binding = 1, rgba8) writeonly uniform image2DArray bgr_output; - -void main() { - vec4 color = imageLoad(bgr_input, ivec3(gl_GlobalInvocationID)); - imageStore(bgr_output, ivec3(gl_GlobalInvocationID), color.bgra); -} diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index c60ed6453..dce00e829 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -2,6 +2,8 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> + #include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 07a995f7d..187a28e4d 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -147,8 +147,7 @@ void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value) { glClearNamedBufferSubData(dest_buffer.Handle(), GL_R32UI, static_cast<GLintptr>(offset), - static_cast<GLsizeiptr>(size / sizeof(u32)), GL_RED, GL_UNSIGNED_INT, - &value); + static_cast<GLsizeiptr>(size), GL_RED, GL_UNSIGNED_INT, &value); } void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) { diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index b0e14182e..02682bd76 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -293,6 +293,8 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading, }}; LoadPipelines(stop_loading, shader_cache_filename, CACHE_VERSION, load_compute, load_graphics); + LOG_INFO(Render_OpenGL, "Total Pipeline Count: {}", state.total); + std::unique_lock lock{state.mutex}; callback(VideoCore::LoadCallbackStage::Build, 0, state.total); state.has_loaded = true; diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index b0aee6cc1..54dae2c41 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -461,7 +461,7 @@ bool TextureCacheRuntime::CanImageBeCopied(const Image& dst, const Image& src) { if (dst.info.type == ImageType::e3D && dst.info.format == PixelFormat::BC4_UNORM) { return false; } - if (IsPixelFormatBGR(dst.info.format) || IsPixelFormatBGR(src.info.format)) { + if (IsPixelFormatBGR(dst.info.format) != IsPixelFormatBGR(src.info.format)) { return false; } return true; @@ -473,7 +473,7 @@ void TextureCacheRuntime::EmulateCopyImage(Image& dst, Image& src, ASSERT(src.info.type == ImageType::e3D); util_shaders.CopyBC4(dst, src, copies); } else if (IsPixelFormatBGR(dst.info.format) || IsPixelFormatBGR(src.info.format)) { - util_shaders.CopyBGR(dst, src, copies); + bgr_copy_pass.CopyBGR(dst, src, copies); } else { UNREACHABLE(); } @@ -1112,4 +1112,37 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM framebuffer.handle = handle; } +void BGRCopyPass::CopyBGR(Image& dst_image, Image& src_image, + std::span<const VideoCommon::ImageCopy> copies) { + static constexpr VideoCommon::Offset3D zero_offset{0, 0, 0}; + const u32 requested_pbo_size = + std::max(src_image.unswizzled_size_bytes, dst_image.unswizzled_size_bytes); + + if (bgr_pbo_size < requested_pbo_size) { + bgr_pbo.Create(); + bgr_pbo_size = requested_pbo_size; + glNamedBufferData(bgr_pbo.handle, bgr_pbo_size, nullptr, GL_STREAM_COPY); + } + for (const ImageCopy& copy : copies) { + ASSERT(copy.src_offset == zero_offset); + ASSERT(copy.dst_offset == zero_offset); + + // Copy from source to PBO + glPixelStorei(GL_PACK_ALIGNMENT, 1); + glPixelStorei(GL_PACK_ROW_LENGTH, copy.extent.width); + glBindBuffer(GL_PIXEL_PACK_BUFFER, bgr_pbo.handle); + glGetTextureSubImage(src_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height, + copy.src_subresource.num_layers, src_image.GlFormat(), + src_image.GlType(), static_cast<GLsizei>(bgr_pbo_size), nullptr); + + // Copy from PBO to destination in desired GL format + glPixelStorei(GL_UNPACK_ALIGNMENT, 1); + glPixelStorei(GL_UNPACK_ROW_LENGTH, copy.extent.width); + glBindBuffer(GL_PIXEL_UNPACK_BUFFER, bgr_pbo.handle); + glTextureSubImage3D(dst_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height, + copy.dst_subresource.num_layers, dst_image.GlFormat(), + dst_image.GlType(), nullptr); + } +} + } // 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 4a4f6301c..c498a8a8f 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -47,6 +47,19 @@ struct FormatProperties { bool is_compressed; }; +class BGRCopyPass { +public: + BGRCopyPass() = default; + ~BGRCopyPass() = default; + + void CopyBGR(Image& dst_image, Image& src_image, + std::span<const VideoCommon::ImageCopy> copies); + +private: + OGLBuffer bgr_pbo; + size_t bgr_pbo_size{}; +}; + class TextureCacheRuntime { friend Framebuffer; friend Image; @@ -118,6 +131,7 @@ private: const Device& device; StateTracker& state_tracker; UtilShaders util_shaders; + BGRCopyPass bgr_copy_pass; std::array<std::unordered_map<GLenum, FormatProperties>, 3> format_properties; bool has_broken_texture_view_formats = false; @@ -162,6 +176,14 @@ public: return texture.handle; } + GLuint GlFormat() const noexcept { + return gl_format; + } + + GLuint GlType() const noexcept { + return gl_type; + } + private: void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index 672f94bfc..39158aa3e 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -52,7 +52,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT}, // BC6H_UFLOAT {GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT}, // BC6H_SFLOAT {GL_COMPRESSED_RGBA_ASTC_4x4_KHR}, // ASTC_2D_4X4_UNORM - {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM + {GL_RGBA8, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV}, // B8G8R8A8_UNORM {GL_RGBA32F, GL_RGBA, GL_FLOAT}, // R32G32B32A32_FLOAT {GL_RGBA32I, GL_RGBA_INTEGER, GL_INT}, // R32G32B32A32_SINT {GL_RG32F, GL_RG, GL_FLOAT}, // R32G32_FLOAT @@ -81,7 +81,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_RGBA_ASTC_8x8_KHR}, // ASTC_2D_8X8_UNORM {GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM {GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM - {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE}, // B8G8R8A8_SRGB + {GL_SRGB8_ALPHA8, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV}, // B8G8R8A8_SRGB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 333f35a1c..897c380b3 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -14,7 +14,6 @@ #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_copy_bc4_comp.h" -#include "video_core/host_shaders/opengl_copy_bgra_comp.h" #include "video_core/host_shaders/pitch_unswizzle_comp.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_shader_util.h" @@ -44,11 +43,6 @@ namespace { OGLProgram MakeProgram(std::string_view source) { return CreateProgram(source, GL_COMPUTE_SHADER); } - -size_t NumPixelsInCopy(const VideoCommon::ImageCopy& copy) { - return static_cast<size_t>(copy.extent.width * copy.extent.height * - copy.src_subresource.num_layers); -} } // Anonymous namespace UtilShaders::UtilShaders(ProgramManager& program_manager_) @@ -56,7 +50,6 @@ 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_bgra_program(MakeProgram(OPENGL_COPY_BGRA_COMP)), copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) { const auto swizzle_table = Tegra::Texture::MakeSwizzleTable(); swizzle_table_buffer.Create(); @@ -255,43 +248,6 @@ void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const Im program_manager.RestoreGuestCompute(); } -void UtilShaders::CopyBGR(Image& dst_image, Image& src_image, - std::span<const VideoCommon::ImageCopy> copies) { - static constexpr GLuint BINDING_INPUT_IMAGE = 0; - static constexpr GLuint BINDING_OUTPUT_IMAGE = 1; - static constexpr VideoCommon::Offset3D zero_offset{0, 0, 0}; - const u32 bytes_per_block = BytesPerBlock(dst_image.info.format); - switch (bytes_per_block) { - case 2: - // BGR565 copy - for (const ImageCopy& copy : copies) { - ASSERT(copy.src_offset == zero_offset); - ASSERT(copy.dst_offset == zero_offset); - bgr_copy_pass.Execute(dst_image, src_image, copy); - } - break; - case 4: { - // BGRA8 copy - program_manager.BindComputeProgram(copy_bgra_program.handle); - constexpr GLenum FORMAT = GL_RGBA8; - for (const ImageCopy& copy : copies) { - ASSERT(copy.src_offset == zero_offset); - ASSERT(copy.dst_offset == zero_offset); - glBindImageTexture(BINDING_INPUT_IMAGE, src_image.StorageHandle(), - copy.src_subresource.base_level, GL_FALSE, 0, GL_READ_ONLY, FORMAT); - glBindImageTexture(BINDING_OUTPUT_IMAGE, dst_image.StorageHandle(), - copy.dst_subresource.base_level, GL_FALSE, 0, GL_WRITE_ONLY, FORMAT); - glDispatchCompute(copy.extent.width, copy.extent.height, copy.extent.depth); - } - program_manager.RestoreGuestCompute(); - break; - } - default: - UNREACHABLE(); - break; - } -} - GLenum StoreFormat(u32 bytes_per_block) { switch (bytes_per_block) { case 1: @@ -309,36 +265,4 @@ GLenum StoreFormat(u32 bytes_per_block) { return GL_R8UI; } -void Bgr565CopyPass::Execute(const Image& dst_image, const Image& src_image, - const ImageCopy& copy) { - if (CopyBufferCreationNeeded(copy)) { - CreateNewCopyBuffer(copy, GL_TEXTURE_2D_ARRAY, GL_RGB565); - } - // Copy from source to PBO - glPixelStorei(GL_PACK_ALIGNMENT, 1); - glPixelStorei(GL_PACK_ROW_LENGTH, copy.extent.width); - glBindBuffer(GL_PIXEL_PACK_BUFFER, bgr16_pbo.handle); - glGetTextureSubImage(src_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height, - copy.src_subresource.num_layers, GL_RGB, GL_UNSIGNED_SHORT_5_6_5, - static_cast<GLsizei>(bgr16_pbo_size), nullptr); - - // Copy from PBO to destination in reverse order - glPixelStorei(GL_UNPACK_ALIGNMENT, 1); - glPixelStorei(GL_UNPACK_ROW_LENGTH, copy.extent.width); - glBindBuffer(GL_PIXEL_UNPACK_BUFFER, bgr16_pbo.handle); - glTextureSubImage3D(dst_image.Handle(), 0, 0, 0, 0, copy.extent.width, copy.extent.height, - copy.dst_subresource.num_layers, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, - nullptr); -} - -bool Bgr565CopyPass::CopyBufferCreationNeeded(const ImageCopy& copy) { - return bgr16_pbo_size < NumPixelsInCopy(copy) * sizeof(u16); -} - -void Bgr565CopyPass::CreateNewCopyBuffer(const ImageCopy& copy, GLenum target, GLuint format) { - bgr16_pbo.Create(); - bgr16_pbo_size = NumPixelsInCopy(copy) * sizeof(u16); - glNamedBufferData(bgr16_pbo.handle, bgr16_pbo_size, nullptr, GL_STREAM_COPY); -} - } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/util_shaders.h b/src/video_core/renderer_opengl/util_shaders.h index ef881e35f..5de95ea7a 100644 --- a/src/video_core/renderer_opengl/util_shaders.h +++ b/src/video_core/renderer_opengl/util_shaders.h @@ -19,22 +19,6 @@ class ProgramManager; struct ImageBufferMap; -class Bgr565CopyPass { -public: - Bgr565CopyPass() = default; - ~Bgr565CopyPass() = default; - - void Execute(const Image& dst_image, const Image& src_image, - const VideoCommon::ImageCopy& copy); - -private: - [[nodiscard]] bool CopyBufferCreationNeeded(const VideoCommon::ImageCopy& copy); - void CreateNewCopyBuffer(const VideoCommon::ImageCopy& copy, GLenum target, GLuint format); - - OGLBuffer bgr16_pbo; - size_t bgr16_pbo_size{}; -}; - class UtilShaders { public: explicit UtilShaders(ProgramManager& program_manager); @@ -55,9 +39,6 @@ public: void CopyBC4(Image& dst_image, Image& src_image, std::span<const VideoCommon::ImageCopy> copies); - void CopyBGR(Image& dst_image, Image& src_image, - std::span<const VideoCommon::ImageCopy> copies); - private: ProgramManager& program_manager; @@ -67,10 +48,7 @@ private: OGLProgram block_linear_unswizzle_2d_program; OGLProgram block_linear_unswizzle_3d_program; OGLProgram pitch_unswizzle_program; - OGLProgram copy_bgra_program; OGLProgram copy_bc4_program; - - Bgr565CopyPass bgr_copy_pass; }; GLenum StoreFormat(u32 bytes_per_block); diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 7c9b0d6db..74822814d 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -97,19 +97,14 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, std::unique_ptr<Core::Frontend::GraphicsContext> context_) try - : RendererBase(emu_window, std::move(context_)), - telemetry_session(telemetry_session_), - cpu_memory(cpu_memory_), - gpu(gpu_), - library(OpenLibrary()), + : RendererBase(emu_window, std::move(context_)), telemetry_session(telemetry_session_), + cpu_memory(cpu_memory_), gpu(gpu_), library(OpenLibrary()), instance(CreateInstance(library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type, true, Settings::values.renderer_debug.GetValue())), 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), + device(CreateDevice(instance, dld, *surface)), memory_allocator(device, false), + state_tracker(gpu), 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, @@ -149,7 +144,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); swapchain.Create(layout.width, layout.height, is_srgb); }; - if (swapchain.IsSubOptimal() || swapchain.HasColorSpaceChanged(is_srgb)) { + if (swapchain.NeedsRecreation(is_srgb)) { recreate_swapchain(); } bool is_outdated; @@ -164,7 +159,8 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { blit_screen.Recreate(); } const VkSemaphore render_semaphore = blit_screen.DrawToSwapchain(*framebuffer, use_accelerated); - scheduler.Flush(render_semaphore); + const VkSemaphore present_semaphore = swapchain.CurrentPresentSemaphore(); + scheduler.Flush(render_semaphore, present_semaphore); scheduler.WaitWorker(); swapchain.Present(render_semaphore); diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index cb0580182..888bc7392 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -358,7 +358,7 @@ void VKBlitScreen::CreateDescriptorPool() { void VKBlitScreen::CreateRenderPass() { const VkAttachmentDescription color_attachment{ .flags = 0, - .format = swapchain.GetImageFormat(), + .format = swapchain.GetImageViewFormat(), .samples = VK_SAMPLE_COUNT_1_BIT, .loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR, .storeOp = VK_ATTACHMENT_STORE_OP_STORE, diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp index 8e77e4796..d87da2a34 100644 --- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> #include <mutex> #include <span> #include <vector> @@ -18,7 +19,6 @@ namespace Vulkan { // Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines constexpr size_t SETS_GROW_RATE = 16; constexpr s32 SCORE_THRESHOLD = 3; -constexpr u32 SETS_PER_POOL = 64; struct DescriptorBank { DescriptorBankInfo info; @@ -58,11 +58,12 @@ static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) { static void AllocatePool(const Device& device, DescriptorBank& bank) { std::array<VkDescriptorPoolSize, 6> pool_sizes; size_t pool_cursor{}; + const u32 sets_per_pool = device.GetSetsPerPool(); const auto add = [&](VkDescriptorType type, u32 count) { if (count > 0) { pool_sizes[pool_cursor++] = { .type = type, - .descriptorCount = count * SETS_PER_POOL, + .descriptorCount = count * sets_per_pool, }; } }; @@ -77,7 +78,7 @@ static void AllocatePool(const Device& device, DescriptorBank& bank) { .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, .pNext = nullptr, .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, - .maxSets = SETS_PER_POOL, + .maxSets = sets_per_pool, .poolSizeCount = static_cast<u32>(pool_cursor), .pPoolSizes = std::data(pool_sizes), })); diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 7c0f91007..11cd41ad7 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -507,8 +507,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { vertex_attributes.push_back({ .location = static_cast<u32>(index), .binding = 0, - .format = type == 1 ? VK_FORMAT_R32_SFLOAT - : type == 2 ? VK_FORMAT_R32_SINT : VK_FORMAT_R32_UINT, + .format = type == 1 ? VK_FORMAT_R32_SFLOAT + : type == 2 ? VK_FORMAT_R32_SINT + : VK_FORMAT_R32_UINT, .offset = 0, }); } diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 31bfbcb06..eb8b4e08b 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -447,6 +447,8 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading VideoCommon::LoadPipelines(stop_loading, pipeline_cache_filename, CACHE_VERSION, load_compute, load_graphics); + LOG_INFO(Render_Vulkan, "Total Pipeline Count: {}", state.total); + std::unique_lock lock{state.mutex}; callback(VideoCore::LoadCallbackStage::Build, 0, state.total); state.has_loaded = true; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 3ac18ea54..3bcd6d6cc 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -228,9 +228,7 @@ void RasterizerVulkan::Clear() { }; const u32 color_attachment = regs.clear_buffers.RT; - const auto attachment_aspect_mask = framebuffer->ImageRanges()[color_attachment].aspectMask; - const bool is_color_rt = (attachment_aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; - if (use_color && is_color_rt) { + if (use_color && framebuffer->HasAspectColorBit(color_attachment)) { VkClearValue clear_value; std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); @@ -248,12 +246,15 @@ void RasterizerVulkan::Clear() { return; } VkImageAspectFlags aspect_flags = 0; - if (use_depth) { + if (use_depth && framebuffer->HasAspectDepthBit()) { aspect_flags |= VK_IMAGE_ASPECT_DEPTH_BIT; } - if (use_stencil) { + if (use_stencil && framebuffer->HasAspectStencilBit()) { aspect_flags |= VK_IMAGE_ASPECT_STENCIL_BIT; } + if (aspect_flags == 0) { + return; + } scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil, clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) { VkClearAttachment attachment; @@ -764,12 +765,7 @@ void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) { const Maxwell::StencilOp zpass = regs.stencil_front_op_zpass; const Maxwell::ComparisonOp compare = regs.stencil_front_func_func; if (regs.stencil_two_side_enable) { - scheduler.Record([fail, zfail, zpass, compare](vk::CommandBuffer cmdbuf) { - cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_AND_BACK, MaxwellToVK::StencilOp(fail), - MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail), - MaxwellToVK::ComparisonOp(compare)); - }); - } else { + // Separate stencil op per face const Maxwell::StencilOp back_fail = regs.stencil_back_op_fail; const Maxwell::StencilOp back_zfail = regs.stencil_back_op_zfail; const Maxwell::StencilOp back_zpass = regs.stencil_back_op_zpass; @@ -784,6 +780,13 @@ void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) { MaxwellToVK::StencilOp(back_zfail), MaxwellToVK::ComparisonOp(back_compare)); }); + } else { + // Front face defines the stencil op of both faces + scheduler.Record([fail, zfail, zpass, compare](vk::CommandBuffer cmdbuf) { + cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_AND_BACK, MaxwellToVK::StencilOp(fail), + MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail), + MaxwellToVK::ComparisonOp(compare)); + }); } } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 4840962de..0c11c814f 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -43,26 +43,19 @@ VKScheduler::VKScheduler(const Device& device_, StateTracker& state_tracker_) command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} { AcquireNewChunk(); AllocateWorkerCommandBuffer(); - worker_thread = std::thread(&VKScheduler::WorkerThread, this); + worker_thread = std::jthread([this](std::stop_token token) { WorkerThread(token); }); } -VKScheduler::~VKScheduler() { - { - std::lock_guard lock{work_mutex}; - quit = true; - } - work_cv.notify_all(); - worker_thread.join(); -} +VKScheduler::~VKScheduler() = default; -void VKScheduler::Flush(VkSemaphore semaphore) { - SubmitExecution(semaphore); +void VKScheduler::Flush(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { + SubmitExecution(signal_semaphore, wait_semaphore); AllocateNewContext(); } -void VKScheduler::Finish(VkSemaphore semaphore) { +void VKScheduler::Finish(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { const u64 presubmit_tick = CurrentTick(); - SubmitExecution(semaphore); + SubmitExecution(signal_semaphore, wait_semaphore); WaitWorker(); Wait(presubmit_tick); AllocateNewContext(); @@ -135,7 +128,7 @@ bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { return true; } -void VKScheduler::WorkerThread() { +void VKScheduler::WorkerThread(std::stop_token stop_token) { Common::SetCurrentThreadName("yuzu:VulkanWorker"); do { if (work_queue.empty()) { @@ -144,8 +137,8 @@ void VKScheduler::WorkerThread() { std::unique_ptr<CommandChunk> work; { std::unique_lock lock{work_mutex}; - work_cv.wait(lock, [this] { return !work_queue.empty() || quit; }); - if (quit) { + work_cv.wait(lock, stop_token, [this] { return !work_queue.empty(); }); + if (stop_token.stop_requested()) { continue; } work = std::move(work_queue.front()); @@ -158,7 +151,7 @@ void VKScheduler::WorkerThread() { } std::lock_guard reserve_lock{reserve_mutex}; chunk_reserve.push_back(std::move(work)); - } while (!quit); + } while (!stop_token.stop_requested()); } void VKScheduler::AllocateWorkerCommandBuffer() { @@ -171,37 +164,41 @@ void VKScheduler::AllocateWorkerCommandBuffer() { }); } -void VKScheduler::SubmitExecution(VkSemaphore semaphore) { +void VKScheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) { EndPendingOperations(); InvalidateState(); const u64 signal_value = master_semaphore->NextTick(); - Record([semaphore, signal_value, this](vk::CommandBuffer cmdbuf) { + Record([signal_semaphore, wait_semaphore, signal_value, this](vk::CommandBuffer cmdbuf) { cmdbuf.End(); - - const u32 num_signal_semaphores = semaphore ? 2U : 1U; - - const u64 wait_value = signal_value - 1; - const VkPipelineStageFlags wait_stage_mask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT; - const VkSemaphore timeline_semaphore = master_semaphore->Handle(); + + const u32 num_signal_semaphores = signal_semaphore ? 2U : 1U; const std::array signal_values{signal_value, u64(0)}; - const std::array signal_semaphores{timeline_semaphore, semaphore}; + const std::array signal_semaphores{timeline_semaphore, signal_semaphore}; + + const u32 num_wait_semaphores = wait_semaphore ? 2U : 1U; + const std::array wait_values{signal_value - 1, u64(1)}; + const std::array wait_semaphores{timeline_semaphore, wait_semaphore}; + static constexpr std::array<VkPipelineStageFlags, 2> wait_stage_masks{ + VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, + }; const VkTimelineSemaphoreSubmitInfoKHR timeline_si{ .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR, .pNext = nullptr, - .waitSemaphoreValueCount = 1, - .pWaitSemaphoreValues = &wait_value, + .waitSemaphoreValueCount = num_wait_semaphores, + .pWaitSemaphoreValues = wait_values.data(), .signalSemaphoreValueCount = num_signal_semaphores, .pSignalSemaphoreValues = signal_values.data(), }; const VkSubmitInfo submit_info{ .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO, .pNext = &timeline_si, - .waitSemaphoreCount = 1, - .pWaitSemaphores = &timeline_semaphore, - .pWaitDstStageMask = &wait_stage_mask, + .waitSemaphoreCount = num_wait_semaphores, + .pWaitSemaphores = wait_semaphores.data(), + .pWaitDstStageMask = wait_stage_masks.data(), .commandBufferCount = 1, .pCommandBuffers = cmdbuf.address(), .signalSemaphoreCount = num_signal_semaphores, diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index cf39a2363..85fc1712f 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -34,10 +34,10 @@ public: ~VKScheduler(); /// Sends the current execution context to the GPU. - void Flush(VkSemaphore semaphore = nullptr); + void Flush(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr); /// Sends the current execution context to the GPU and waits for it to complete. - void Finish(VkSemaphore semaphore = nullptr); + void Finish(VkSemaphore signal_semaphore = nullptr, VkSemaphore wait_semaphore = nullptr); /// Waits for the worker thread to finish executing everything. After this function returns it's /// safe to touch worker resources. @@ -187,11 +187,11 @@ private: GraphicsPipeline* graphics_pipeline = nullptr; }; - void WorkerThread(); + void WorkerThread(std::stop_token stop_token); void AllocateWorkerCommandBuffer(); - void SubmitExecution(VkSemaphore semaphore); + void SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore); void AllocateNewContext(); @@ -212,7 +212,6 @@ private: vk::CommandBuffer current_cmdbuf; std::unique_ptr<CommandChunk> chunk; - std::thread worker_thread; State state; @@ -224,9 +223,9 @@ private: std::vector<std::unique_ptr<CommandChunk>> chunk_reserve; std::mutex reserve_mutex; std::mutex work_mutex; - std::condition_variable work_cv; + std::condition_variable_any work_cv; std::condition_variable wait_cv; - std::atomic_bool quit{}; + std::jthread worker_thread; }; } // 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 5f78f6950..d90935f52 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -110,10 +110,6 @@ public: return Exchange(Dirty::DepthTestEnable, false); } - bool TouchDepthBoundsEnable() { - return Exchange(Dirty::DepthBoundsEnable, false); - } - bool TouchDepthWriteEnable() { return Exchange(Dirty::DepthWriteEnable, false); } diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index d990eefba..8972a6921 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -9,6 +9,7 @@ #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" @@ -20,16 +21,15 @@ namespace Vulkan { namespace { -VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats, bool srgb) { +VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats) { if (formats.size() == 1 && formats[0].format == VK_FORMAT_UNDEFINED) { VkSurfaceFormatKHR format; format.format = VK_FORMAT_B8G8R8A8_UNORM; format.colorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR; return format; } - const auto& found = std::find_if(formats.begin(), formats.end(), [srgb](const auto& format) { - const auto request_format = srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM; - return format.format == request_format && + const auto& found = std::find_if(formats.begin(), formats.end(), [](const auto& format) { + return format.format == VK_FORMAT_B8G8R8A8_UNORM && format.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR; }); return found != formats.end() ? *found : formats[0]; @@ -37,8 +37,19 @@ VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats, VkPresentModeKHR ChooseSwapPresentMode(vk::Span<VkPresentModeKHR> modes) { // Mailbox doesn't lock the application like fifo (vsync), prefer it - const auto found = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); - return found != modes.end() ? *found : VK_PRESENT_MODE_FIFO_KHR; + const auto found_mailbox = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); + if (found_mailbox != modes.end()) { + return VK_PRESENT_MODE_MAILBOX_KHR; + } + if (Settings::values.disable_fps_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); + if (found_imm != modes.end()) { + return VK_PRESENT_MODE_IMMEDIATE_KHR; + } + } + return VK_PRESENT_MODE_FIFO_KHR; } VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height) { @@ -107,14 +118,12 @@ void VKSwapchain::AcquireNextImage() { } void VKSwapchain::Present(VkSemaphore render_semaphore) { - const VkSemaphore present_semaphore{*present_semaphores[frame_index]}; - const std::array<VkSemaphore, 2> semaphores{present_semaphore, render_semaphore}; const auto present_queue{device.GetPresentQueue()}; const VkPresentInfoKHR present_info{ .sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR, .pNext = nullptr, - .waitSemaphoreCount = render_semaphore ? 2U : 1U, - .pWaitSemaphores = semaphores.data(), + .waitSemaphoreCount = render_semaphore ? 1U : 0U, + .pWaitSemaphores = &render_semaphore, .swapchainCount = 1, .pSwapchains = swapchain.address(), .pImageIndices = &image_index, @@ -145,8 +154,8 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, const auto formats{physical_device.GetSurfaceFormatsKHR(surface)}; const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)}; - const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats, srgb)}; - const VkPresentModeKHR present_mode{ChooseSwapPresentMode(present_modes)}; + const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats)}; + present_mode = ChooseSwapPresentMode(present_modes); u32 requested_image_count{capabilities.minImageCount + 1}; if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) { @@ -180,6 +189,17 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, swapchain_ci.queueFamilyIndexCount = static_cast<u32>(queue_indices.size()); swapchain_ci.pQueueFamilyIndices = queue_indices.data(); } + static constexpr std::array view_formats{VK_FORMAT_B8G8R8A8_UNORM, VK_FORMAT_B8G8R8A8_SRGB}; + VkImageFormatListCreateInfo format_list{ + .sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO_KHR, + .pNext = nullptr, + .viewFormatCount = static_cast<u32>(view_formats.size()), + .pViewFormats = view_formats.data(), + }; + if (device.IsKhrSwapchainMutableFormatEnabled()) { + format_list.pNext = std::exchange(swapchain_ci.pNext, &format_list); + swapchain_ci.flags |= VK_SWAPCHAIN_CREATE_MUTABLE_FORMAT_BIT_KHR; + } // Request the size again to reduce the possibility of a TOCTOU race condition. const auto updated_capabilities = physical_device.GetSurfaceCapabilitiesKHR(surface); swapchain_ci.imageExtent = ChooseSwapExtent(updated_capabilities, width, height); @@ -188,10 +208,11 @@ void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, extent = swapchain_ci.imageExtent; current_srgb = srgb; + current_fps_unlocked = Settings::values.disable_fps_limit.GetValue(); images = swapchain.GetImages(); image_count = static_cast<u32>(images.size()); - image_format = surface_format.format; + image_view_format = srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM; } void VKSwapchain::CreateSemaphores() { @@ -207,7 +228,7 @@ void VKSwapchain::CreateImageViews() { .flags = 0, .image = {}, .viewType = VK_IMAGE_VIEW_TYPE_2D, - .format = image_format, + .format = image_view_format, .components = { .r = VK_COMPONENT_SWIZZLE_IDENTITY, @@ -240,4 +261,14 @@ void VKSwapchain::Destroy() { swapchain.reset(); } +bool VKSwapchain::HasFpsUnlockChanged() const { + return current_fps_unlocked != Settings::values.disable_fps_limit.GetValue(); +} + +bool VKSwapchain::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(); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index 35c2cdc14..61a6d959e 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -33,6 +33,11 @@ public: /// Presents the rendered image to the swapchain. void Present(VkSemaphore render_semaphore); + /// Returns true when the swapchain needs to be recreated. + bool NeedsRecreation(bool is_srgb) const { + return HasColorSpaceChanged(is_srgb) || IsSubOptimal() || NeedsPresentModeUpdate(); + } + /// Returns true when the color space has changed. bool HasColorSpaceChanged(bool is_srgb) const { return current_srgb != is_srgb; @@ -68,8 +73,12 @@ public: return *image_views[index]; } - VkFormat GetImageFormat() const { - return image_format; + VkFormat GetImageViewFormat() const { + return image_view_format; + } + + VkSemaphore CurrentPresentSemaphore() const { + return *present_semaphores[frame_index]; } private: @@ -80,6 +89,10 @@ private: void Destroy(); + bool HasFpsUnlockChanged() const; + + bool NeedsPresentModeUpdate() const; + const VkSurfaceKHR surface; const Device& device; VKScheduler& scheduler; @@ -96,10 +109,12 @@ private: u32 image_index{}; u32 frame_index{}; - VkFormat image_format{}; + VkFormat image_view_format{}; VkExtent2D extent{}; + VkPresentModeKHR present_mode{}; bool current_srgb{}; + bool current_fps_unlocked{}; bool is_outdated{}; bool is_suboptimal{}; }; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 8f4df7122..3b87640b5 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -127,7 +127,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format); VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; if (info.type == ImageType::e2D && info.resources.layers >= 6 && - info.size.width == info.size.height) { + info.size.width == info.size.height && !device.HasBrokenCubeImageCompability()) { flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; } if (info.type == ImageType::e3D) { @@ -1186,9 +1186,12 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM renderpass_key.depth_format = depth_buffer->format; num_layers = std::max(num_layers, depth_buffer->range.extent.layers); images[num_images] = depth_buffer->ImageHandle(); - image_ranges[num_images] = MakeSubresourceRange(depth_buffer); + const VkImageSubresourceRange subresource_range = MakeSubresourceRange(depth_buffer); + image_ranges[num_images] = subresource_range; samples = depth_buffer->Samples(); ++num_images; + has_depth = (subresource_range.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0; + has_stencil = (subresource_range.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) != 0; } else { renderpass_key.depth_format = PixelFormat::Invalid; } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 5fe6b7ba3..6d5a68bfe 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -232,6 +232,18 @@ public: return image_ranges; } + [[nodiscard]] bool HasAspectColorBit(size_t index) const noexcept { + return (image_ranges.at(index).aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) != 0; + } + + [[nodiscard]] bool HasAspectDepthBit() const noexcept { + return has_depth; + } + + [[nodiscard]] bool HasAspectStencilBit() const noexcept { + return has_stencil; + } + private: vk::Framebuffer framebuffer; VkRenderPass renderpass{}; @@ -241,6 +253,8 @@ private: u32 num_images = 0; std::array<VkImage, 9> images{}; std::array<VkImageSubresourceRange, 9> image_ranges{}; + bool has_depth{}; + bool has_stencil{}; }; struct TextureCacheParams { diff --git a/src/video_core/shader_environment.cpp b/src/video_core/shader_environment.cpp index 8a4581c19..81a878bb2 100644 --- a/src/video_core/shader_environment.cpp +++ b/src/video_core/shader_environment.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <algorithm> #include <filesystem> #include <fstream> #include <memory> diff --git a/src/video_core/texture_cache/slot_vector.h b/src/video_core/texture_cache/slot_vector.h index 6180b8c0e..50df06409 100644 --- a/src/video_core/texture_cache/slot_vector.h +++ b/src/video_core/texture_cache/slot_vector.h @@ -4,6 +4,7 @@ #pragma once +#include <algorithm> #include <array> #include <bit> #include <concepts> @@ -30,8 +31,8 @@ struct SlotId { }; template <class T> -requires std::is_nothrow_move_assignable_v<T>&& - std::is_nothrow_move_constructible_v<T> class SlotVector { +requires std::is_nothrow_move_assignable_v<T> && std::is_nothrow_move_constructible_v<T> +class SlotVector { public: class Iterator { friend SlotVector<T>; diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index 3b575db4d..cae543a51 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -37,7 +37,8 @@ std::unique_ptr<VideoCore::RendererBase> CreateRenderer( namespace VideoCore { std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system) { - const bool use_nvdec = Settings::values.use_nvdec_emulation.GetValue(); + const auto nvdec_value = Settings::values.nvdec_emulation.GetValue(); + const bool use_nvdec = nvdec_value != Settings::NvdecEmulation::Off; const bool use_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); auto gpu = std::make_unique<Tegra::GPU>(system, use_async, use_nvdec); auto context = emu_window.CreateSharedContext(); diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.cpp b/src/video_core/vulkan_common/vulkan_debug_callback.cpp index 0f60765bb..cf94e1d39 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.cpp +++ b/src/video_core/vulkan_common/vulkan_debug_callback.cpp @@ -16,6 +16,7 @@ VkBool32 Callback(VkDebugUtilsMessageSeverityFlagBitsEXT severity, switch (static_cast<u32>(data->messageIdNumber)) { case 0x682a878au: // VUID-vkCmdBindVertexBuffers2EXT-pBuffers-parameter case 0x99fb7dfdu: // UNASSIGNED-RequiredParameter (vkCmdBindVertexBuffers2EXT pBuffers[0]) + case 0xe8616bf2u: // Bound VkDescriptorSet 0x0[] was destroyed. Likely push_descriptor related return VK_FALSE; default: break; diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 86ca4be54..6388ed2eb 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -368,8 +368,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR }; SetNext(next, demote); + VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8; if (is_int8_supported || is_float16_supported) { - VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8{ + float16_int8 = { .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR, .pNext = nullptr, .shaderFloat16 = is_float16_supported, @@ -587,6 +588,31 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR ext_extended_dynamic_state = false; } } + sets_per_pool = 64; + + const bool is_amd = + driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE; + if (is_amd) { + // AMD drivers need a higher amount of Sets per Pool in certain circunstances like in XC2. + sets_per_pool = 96; + // Disable VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT on AMD GCN4 and lower as it is broken. + if (!is_float16_supported) { + LOG_WARNING( + Render_Vulkan, + "AMD GCN4 and earlier do not properly support VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT"); + has_broken_cube_compatibility = true; + } + } + const bool is_amd_or_radv = is_amd || driver_id == VK_DRIVER_ID_MESA_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) { + LOG_WARNING(Render_Vulkan, + "Blacklisting AMD GCN4 and earlier for VK_EXT_sampler_filter_minmax"); + ext_sampler_filter_minmax = false; + } + } + if (ext_vertex_input_dynamic_state && driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) { LOG_WARNING(Render_Vulkan, "Blacklisting Intel for VK_EXT_vertex_input_dynamic_state"); ext_vertex_input_dynamic_state = false; @@ -839,6 +865,8 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { bool has_khr_shader_float16_int8{}; bool has_khr_workgroup_memory_explicit_layout{}; bool has_khr_pipeline_executable_properties{}; + bool has_khr_image_format_list{}; + bool has_khr_swapchain_mutable_format{}; bool has_ext_subgroup_size_control{}; bool has_ext_transform_feedback{}; bool has_ext_custom_border_color{}; @@ -888,6 +916,9 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { test(has_ext_shader_atomic_int64, VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME, false); test(has_khr_workgroup_memory_explicit_layout, VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME, false); + test(has_khr_image_format_list, VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME, false); + 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); if (Settings::values.enable_nsight_aftermath) { test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, @@ -1066,6 +1097,11 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) { khr_pipeline_executable_properties = true; } } + if (has_khr_image_format_list && has_khr_swapchain_mutable_format) { + extensions.push_back(VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME); + extensions.push_back(VK_KHR_SWAPCHAIN_MUTABLE_FORMAT_EXTENSION_NAME); + khr_swapchain_mutable_format = true; + } if (khr_push_descriptor) { VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor; push_descriptor.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR; diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 234d74129..d9e74f1aa 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -224,6 +224,11 @@ public: return khr_pipeline_executable_properties; } + /// Returns true if VK_KHR_swapchain_mutable_format is enabled. + bool IsKhrSwapchainMutableFormatEnabled() const { + return khr_swapchain_mutable_format; + } + /// Returns true if the device supports VK_KHR_workgroup_memory_explicit_layout. bool IsKhrWorkgroupMemoryExplicitLayoutSupported() const { return khr_workgroup_memory_explicit_layout; @@ -304,6 +309,11 @@ public: return has_renderdoc || has_nsight_graphics; } + /// Returns true when the device does not properly support cube compatibility. + bool HasBrokenCubeImageCompability() const { + return has_broken_cube_compatibility; + } + /// Returns the vendor name reported from Vulkan. std::string_view GetVendorName() const { return vendor_name; @@ -318,6 +328,10 @@ public: return device_access_memory; } + u32 GetSetsPerPool() const { + return sets_per_pool; + } + private: /// Checks if the physical device is suitable. void CheckSuitability(bool requires_swapchain) const; @@ -371,6 +385,7 @@ private: VkShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced. u64 device_access_memory{}; ///< Total size of device local memory in bytes. u32 max_push_descriptors{}; ///< Maximum number of push descriptors + u32 sets_per_pool{}; ///< Sets per Description Pool bool is_optimal_astc_supported{}; ///< Support for native ASTC. bool is_float16_supported{}; ///< Support for float16 arithmetic. bool is_int8_supported{}; ///< Support for int8 arithmetic. @@ -390,6 +405,7 @@ private: bool khr_workgroup_memory_explicit_layout{}; ///< Support for explicit workgroup layouts. bool khr_push_descriptor{}; ///< Support for VK_KHR_push_descritor. bool khr_pipeline_executable_properties{}; ///< Support for executable properties. + bool khr_swapchain_mutable_format{}; ///< Support for VK_KHR_swapchain_mutable_format. bool ext_index_type_uint8{}; ///< Support for VK_EXT_index_type_uint8. bool ext_sampler_filter_minmax{}; ///< Support for VK_EXT_sampler_filter_minmax. bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted. @@ -406,6 +422,7 @@ private: bool ext_conservative_rasterization{}; ///< Support for VK_EXT_conservative_rasterization. bool ext_provoking_vertex{}; ///< Support for VK_EXT_provoking_vertex. 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 |