diff options
Diffstat (limited to 'src/video_core')
37 files changed, 1025 insertions, 217 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index ccfed4f2e..db9332d00 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -29,6 +29,8 @@ add_library(video_core STATIC gpu_synch.h gpu_thread.cpp gpu_thread.h + guest_driver.cpp + guest_driver.h macro_interpreter.cpp macro_interpreter.h memory_manager.cpp @@ -154,6 +156,7 @@ if (ENABLE_VULKAN) renderer_vulkan/maxwell_to_vk.cpp renderer_vulkan/maxwell_to_vk.h renderer_vulkan/renderer_vulkan.h + renderer_vulkan/renderer_vulkan.cpp renderer_vulkan/vk_blit_screen.cpp renderer_vulkan/vk_blit_screen.h renderer_vulkan/vk_buffer_cache.cpp diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 0510ed777..186aca61d 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -101,7 +101,10 @@ public: void TickFrame() { ++epoch; while (!pending_destruction.empty()) { - if (pending_destruction.front()->GetEpoch() + 1 > epoch) { + // Delay at least 4 frames before destruction. + // This is due to triple buffering happening on some drivers. + static constexpr u64 epochs_to_destroy = 5; + if (pending_destruction.front()->GetEpoch() + epochs_to_destroy > epoch) { break; } pending_destruction.pop_front(); diff --git a/src/video_core/engines/const_buffer_engine_interface.h b/src/video_core/engines/const_buffer_engine_interface.h index 44b8b8d22..d56a47710 100644 --- a/src/video_core/engines/const_buffer_engine_interface.h +++ b/src/video_core/engines/const_buffer_engine_interface.h @@ -9,6 +9,7 @@ #include "common/common_types.h" #include "video_core/engines/shader_bytecode.h" #include "video_core/engines/shader_type.h" +#include "video_core/guest_driver.h" #include "video_core/textures/texture.h" namespace Tegra::Engines { @@ -106,6 +107,9 @@ public: virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer, u64 offset) const = 0; virtual u32 GetBoundBuffer() const = 0; + + virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0; + virtual const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const = 0; }; } // namespace Tegra::Engines diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index 110406f2f..4b824aa4e 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -94,6 +94,14 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con return result; } +VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() { + return rasterizer.AccessGuestDriverProfile(); +} + +const VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() const { + return rasterizer.AccessGuestDriverProfile(); +} + void KeplerCompute::ProcessLaunch() { const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address(); memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description, diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index 4ef3e0613..eeb79c56f 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -218,6 +218,10 @@ public: return regs.tex_cb_index; } + VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override; + + const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override; + private: Core::System& system; VideoCore::RasterizerInterface& rasterizer; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 58dfa8033..7cea146f0 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -784,4 +784,12 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b return result; } +VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() { + return rasterizer.AccessGuestDriverProfile(); +} + +const VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() const { + return rasterizer.AccessGuestDriverProfile(); +} + } // namespace Tegra::Engines diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index ee79260fc..8808bbf76 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1306,6 +1306,10 @@ public: return regs.tex_cb_index; } + VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override; + + const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override; + /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than /// we've seen used. using MacroMemory = std::array<u32, 0x40000>; diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 901adb97b..402869fde 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -1144,6 +1144,11 @@ union Instruction { } fset; union { + BitField<47, 1, u64> ftz; + BitField<48, 4, PredCondition> cond; + } fcmp; + + union { BitField<49, 1, u64> bf; BitField<35, 3, PredCondition> cond; BitField<50, 1, u64> ftz; @@ -1723,6 +1728,7 @@ public: BFE_C, BFE_R, BFE_IMM, + BFI_RC, BFI_IMM_R, BRA, BRX, @@ -1820,6 +1826,7 @@ public: ICMP_R, ICMP_CR, ICMP_IMM, + FCMP_R, MUFU, // Multi-Function Operator RRO_C, // Range Reduction Operator RRO_R, @@ -2124,6 +2131,7 @@ private: INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"), INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"), INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"), + INST("010110111010----", Id::FCMP_R, Type::Arithmetic, "FCMP_R"), INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"), INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"), INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"), @@ -2148,6 +2156,7 @@ private: INST("0100110000000---", Id::BFE_C, Type::Bfe, "BFE_C"), INST("0101110000000---", Id::BFE_R, Type::Bfe, "BFE_R"), INST("0011100-00000---", Id::BFE_IMM, Type::Bfe, "BFE_IMM"), + INST("0101001111110---", Id::BFI_RC, Type::Bfi, "BFI_RC"), INST("0011011-11110---", Id::BFI_IMM_R, Type::Bfi, "BFI_IMM_R"), INST("0100110001000---", Id::LOP_C, Type::ArithmeticInteger, "LOP_C"), INST("0101110001000---", Id::LOP_R, Type::ArithmeticInteger, "LOP_R"), diff --git a/src/video_core/guest_driver.cpp b/src/video_core/guest_driver.cpp new file mode 100644 index 000000000..6adef459e --- /dev/null +++ b/src/video_core/guest_driver.cpp @@ -0,0 +1,36 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <algorithm> +#include <limits> + +#include "video_core/guest_driver.h" + +namespace VideoCore { + +void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets) { + if (texture_handler_size_deduced) { + return; + } + const std::size_t size = bound_offsets.size(); + if (size < 2) { + return; + } + std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{}); + u32 min_val = std::numeric_limits<u32>::max(); + for (std::size_t i = 1; i < size; ++i) { + if (bound_offsets[i] == bound_offsets[i - 1]) { + continue; + } + const u32 new_min = bound_offsets[i] - bound_offsets[i - 1]; + min_val = std::min(min_val, new_min); + } + if (min_val > 2) { + return; + } + texture_handler_size_deduced = true; + texture_handler_size = min_texture_handler_size * min_val; +} + +} // namespace VideoCore diff --git a/src/video_core/guest_driver.h b/src/video_core/guest_driver.h new file mode 100644 index 000000000..fc1917347 --- /dev/null +++ b/src/video_core/guest_driver.h @@ -0,0 +1,41 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <vector> + +#include "common/common_types.h" + +namespace VideoCore { + +/** + * The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect + * information necessary for impossible to avoid HLE methods like shader tracks as they are + * Entscheidungsproblems. + */ +class GuestDriverProfile { +public: + void DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets); + + u32 GetTextureHandlerSize() const { + return texture_handler_size; + } + + bool TextureHandlerSizeKnown() const { + return texture_handler_size_deduced; + } + +private: + // Minimum size of texture handler any driver can use. + static constexpr u32 min_texture_handler_size = 4; + // This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily + // use 4 bytes instead. Thus, certain drivers may squish the size. + static constexpr u32 default_texture_handler_size = 8; + + u32 texture_handler_size = default_texture_handler_size; + bool texture_handler_size_deduced = false; +}; + +} // namespace VideoCore diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 5b0eca9e2..c586cd6fe 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -9,6 +9,7 @@ #include "common/common_types.h" #include "video_core/engines/fermi_2d.h" #include "video_core/gpu.h" +#include "video_core/guest_driver.h" namespace Tegra { class MemoryManager; @@ -78,5 +79,18 @@ public: /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(const std::atomic_bool& stop_loading = false, const DiskResourceLoadCallback& callback = {}) {} + + /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver. + GuestDriverProfile& AccessGuestDriverProfile() { + return guest_driver_profile; + } + + /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver. + const GuestDriverProfile& AccessGuestDriverProfile() const { + return guest_driver_profile; + } + +private: + GuestDriverProfile guest_driver_profile{}; }; } // namespace VideoCore diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index c428f06e4..46a7433ea 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -55,16 +55,20 @@ namespace { template <typename Engine, typename Entry> Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry, - Tegra::Engines::ShaderType shader_type) { + Tegra::Engines::ShaderType shader_type, + std::size_t index = 0) { if (entry.IsBindless()) { const Tegra::Texture::TextureHandle tex_handle = engine.AccessConstBuffer32(shader_type, entry.GetBuffer(), entry.GetOffset()); return engine.GetTextureInfo(tex_handle); } + const auto& gpu_profile = engine.AccessGuestDriverProfile(); + const u32 offset = + entry.GetOffset() + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize()); if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) { - return engine.GetStageTexture(shader_type, entry.GetOffset()); + return engine.GetStageTexture(shader_type, offset); } else { - return engine.GetTexture(entry.GetOffset()); + return engine.GetTexture(offset); } } @@ -244,9 +248,6 @@ void RasterizerOpenGL::SetupVertexInstances(GLuint vao) { } GLintptr RasterizerOpenGL::SetupIndexBuffer() { - if (accelerate_draw != AccelDraw::Indexed) { - return 0; - } MICROPROFILE_SCOPE(OpenGL_Index); const auto& regs = system.GPU().Maxwell3D().regs; const std::size_t size = CalculateIndexBufferSize(); @@ -542,7 +543,8 @@ void RasterizerOpenGL::Clear() { } } -void RasterizerOpenGL::DrawPrelude() { +void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { + MICROPROFILE_SCOPE(OpenGL_Drawing); auto& gpu = system.GPU().Maxwell3D(); SyncRasterizeEnable(state); @@ -563,9 +565,6 @@ void RasterizerOpenGL::DrawPrelude() { buffer_cache.Acquire(); - // Draw the vertex batch - const bool is_indexed = accelerate_draw == AccelDraw::Indexed; - std::size_t buffer_size = CalculateVertexArraysSize(); // Add space for index buffer @@ -592,7 +591,11 @@ void RasterizerOpenGL::DrawPrelude() { // Upload vertex and index data. SetupVertexBuffer(vao); SetupVertexInstances(vao); - index_buffer_offset = SetupIndexBuffer(); + + GLintptr index_buffer_offset; + if (is_indexed) { + index_buffer_offset = SetupIndexBuffer(); + } // Prepare packed bindings. bind_ubo_pushbuffer.Setup(); @@ -626,6 +629,7 @@ void RasterizerOpenGL::DrawPrelude() { // As all cached buffers are invalidated, we need to recheck their state. gpu.dirty.ResetVertexArrays(); } + gpu.dirty.memory_general = false; shader_program_manager->ApplyTo(state); state.Apply(); @@ -633,106 +637,33 @@ void RasterizerOpenGL::DrawPrelude() { if (texture_cache.TextureBarrier()) { glTextureBarrier(); } -} - -struct DrawParams { - bool is_indexed{}; - bool is_instanced{}; - GLenum primitive_mode{}; - GLint count{}; - GLint base_vertex{}; - - // Indexed settings - GLenum index_format{}; - GLintptr index_buffer_offset{}; - - // Instanced setting - GLint num_instances{}; - GLint base_instance{}; - - void DispatchDraw() { - if (is_indexed) { - const auto index_buffer_ptr = reinterpret_cast<const void*>(index_buffer_offset); - if (is_instanced) { - glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format, - index_buffer_ptr, num_instances, - base_vertex, base_instance); - } else { - glDrawElementsBaseVertex(primitive_mode, count, index_format, index_buffer_ptr, - base_vertex); - } - } else { - if (is_instanced) { - glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, count, num_instances, - base_instance); - } else { - glDrawArrays(primitive_mode, base_vertex, count); - } - } - } -}; - -bool RasterizerOpenGL::DrawBatch(bool is_indexed) { - accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; - - MICROPROFILE_SCOPE(OpenGL_Drawing); - - DrawPrelude(); - auto& maxwell3d = system.GPU().Maxwell3D(); - const auto& regs = maxwell3d.regs; - const auto current_instance = maxwell3d.state.current_instance; - DrawParams draw_call{}; - draw_call.is_indexed = is_indexed; - draw_call.num_instances = static_cast<GLint>(1); - draw_call.base_instance = static_cast<GLint>(current_instance); - draw_call.is_instanced = current_instance > 0; - draw_call.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); - if (draw_call.is_indexed) { - draw_call.count = static_cast<GLint>(regs.index_array.count); - draw_call.base_vertex = static_cast<GLint>(regs.vb_element_base); - draw_call.index_format = MaxwellToGL::IndexFormat(regs.index_array.format); - draw_call.index_buffer_offset = index_buffer_offset; + const GLuint base_instance = static_cast<GLuint>(gpu.regs.vb_base_instance); + const GLsizei num_instances = + static_cast<GLsizei>(is_instanced ? gpu.mme_draw.instance_count : 1); + if (is_indexed) { + const GLenum index_format = MaxwellToGL::IndexFormat(gpu.regs.index_array.format); + const GLint base_vertex = static_cast<GLint>(gpu.regs.vb_element_base); + const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.index_array.count); + glDrawElementsInstancedBaseVertexBaseInstance( + primitive_mode, num_vertices, index_format, + reinterpret_cast<const void*>(index_buffer_offset), num_instances, base_vertex, + base_instance); } else { - draw_call.count = static_cast<GLint>(regs.vertex_buffer.count); - draw_call.base_vertex = static_cast<GLint>(regs.vertex_buffer.first); + const GLint base_vertex = static_cast<GLint>(gpu.regs.vertex_buffer.first); + const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.vertex_buffer.count); + glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, num_vertices, num_instances, + base_instance); } - draw_call.DispatchDraw(); +} - maxwell3d.dirty.memory_general = false; - accelerate_draw = AccelDraw::Disabled; +bool RasterizerOpenGL::DrawBatch(bool is_indexed) { + Draw(is_indexed, false); return true; } bool RasterizerOpenGL::DrawMultiBatch(bool is_indexed) { - accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; - - MICROPROFILE_SCOPE(OpenGL_Drawing); - - DrawPrelude(); - - auto& maxwell3d = system.GPU().Maxwell3D(); - const auto& regs = maxwell3d.regs; - const auto& draw_setup = maxwell3d.mme_draw; - DrawParams draw_call{}; - draw_call.is_indexed = is_indexed; - draw_call.num_instances = static_cast<GLint>(draw_setup.instance_count); - draw_call.base_instance = static_cast<GLint>(regs.vb_base_instance); - draw_call.is_instanced = draw_setup.instance_count > 1; - draw_call.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); - if (draw_call.is_indexed) { - draw_call.count = static_cast<GLint>(regs.index_array.count); - draw_call.base_vertex = static_cast<GLint>(regs.vb_element_base); - draw_call.index_format = MaxwellToGL::IndexFormat(regs.index_array.format); - draw_call.index_buffer_offset = index_buffer_offset; - } else { - draw_call.count = static_cast<GLint>(regs.vertex_buffer.count); - draw_call.base_vertex = static_cast<GLint>(regs.vertex_buffer.first); - } - draw_call.DispatchDraw(); - - maxwell3d.dirty.memory_general = false; - accelerate_draw = AccelDraw::Disabled; + Draw(is_indexed, true); return true; } @@ -942,8 +873,15 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader& u32 binding = device.GetBaseBindings(stage_index).sampler; for (const auto& entry : shader->GetShaderEntries().samplers) { const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage_index); - const auto texture = GetTextureInfo(maxwell3d, entry, shader_type); - SetupTexture(binding++, texture, entry); + if (!entry.IsIndexed()) { + const auto texture = GetTextureInfo(maxwell3d, entry, shader_type); + SetupTexture(binding++, texture, entry); + } else { + for (std::size_t i = 0; i < entry.Size(); ++i) { + const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i); + SetupTexture(binding++, texture, entry); + } + } } } @@ -952,8 +890,17 @@ void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) { const auto& compute = system.GPU().KeplerCompute(); u32 binding = 0; for (const auto& entry : kernel->GetShaderEntries().samplers) { - const auto texture = GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute); - SetupTexture(binding++, texture, entry); + if (!entry.IsIndexed()) { + const auto texture = + GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute); + SetupTexture(binding++, texture, entry); + } else { + for (std::size_t i = 0; i < entry.Size(); ++i) { + const auto texture = + GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute, i); + SetupTexture(binding++, texture, entry); + } + } } } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 6a27cf497..0501f3828 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -103,7 +103,7 @@ private: std::size_t size); /// Syncs all the state, shaders, render targets and textures setting before a draw call. - void DrawPrelude(); + void Draw(bool is_indexed, bool is_instanced); /// Configures the current textures to use for the draw command. void SetupDrawTextures(std::size_t stage_index, const Shader& shader); @@ -220,12 +220,7 @@ private: GLintptr SetupIndexBuffer(); - GLintptr index_buffer_offset; - void SetupShaders(GLenum primitive_mode); - - enum class AccelDraw { Disabled, Arrays, Indexed }; - AccelDraw accelerate_draw = AccelDraw::Disabled; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 3c5bdd377..489eb143c 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -214,6 +214,7 @@ std::unique_ptr<ConstBufferLocker> MakeLocker(Core::System& system, ShaderType s } void FillLocker(ConstBufferLocker& locker, const ShaderDiskCacheUsage& usage) { + locker.SetBoundBuffer(usage.bound_buffer); for (const auto& key : usage.keys) { const auto [buffer, offset] = key.first; locker.InsertKey(buffer, offset, key.second); @@ -418,7 +419,8 @@ bool CachedShader::EnsureValidLockerVariant() { ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant, const ConstBufferLocker& locker) const { - return ShaderDiskCacheUsage{unique_identifier, variant, locker.GetKeys(), + return ShaderDiskCacheUsage{unique_identifier, variant, + locker.GetBoundBuffer(), locker.GetKeys(), locker.GetBoundSamplers(), locker.GetBindlessSamplers()}; } diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index a1ac3d7a9..4735000b5 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -391,6 +391,7 @@ public: DeclareVertex(); DeclareGeometry(); DeclareRegisters(); + DeclareCustomVariables(); DeclarePredicates(); DeclareLocalMemory(); DeclareInternalFlags(); @@ -503,6 +504,16 @@ private: } } + void DeclareCustomVariables() { + const u32 num_custom_variables = ir.GetNumCustomVariables(); + for (u32 i = 0; i < num_custom_variables; ++i) { + code.AddLine("float {} = 0.0f;", GetCustomVariable(i)); + } + if (num_custom_variables > 0) { + code.AddNewLine(); + } + } + void DeclarePredicates() { const auto& predicates = ir.GetPredicates(); for (const auto pred : predicates) { @@ -655,7 +666,8 @@ private: u32 binding = device.GetBaseBindings(stage).sampler; for (const auto& sampler : ir.GetSamplers()) { const std::string name = GetSampler(sampler); - const std::string description = fmt::format("layout (binding = {}) uniform", binding++); + const std::string description = fmt::format("layout (binding = {}) uniform", binding); + binding += sampler.IsIndexed() ? sampler.Size() : 1; std::string sampler_type = [&]() { if (sampler.IsBuffer()) { @@ -682,7 +694,11 @@ private: sampler_type += "Shadow"; } - code.AddLine("{} {} {};", description, sampler_type, name); + if (!sampler.IsIndexed()) { + code.AddLine("{} {} {};", description, sampler_type, name); + } else { + code.AddLine("{} {} {}[{}];", description, sampler_type, name, sampler.Size()); + } } if (!ir.GetSamplers().empty()) { code.AddNewLine(); @@ -775,6 +791,11 @@ private: return {GetRegister(index), Type::Float}; } + if (const auto cv = std::get_if<CustomVarNode>(&*node)) { + const u32 index = cv->GetIndex(); + return {GetCustomVariable(index), Type::Float}; + } + if (const auto immediate = std::get_if<ImmediateNode>(&*node)) { const u32 value = immediate->GetValue(); if (value < 10) { @@ -1098,7 +1119,11 @@ private: } else if (!meta->ptp.empty()) { expr += "Offsets"; } - expr += '(' + GetSampler(meta->sampler) + ", "; + if (!meta->sampler.IsIndexed()) { + expr += '(' + GetSampler(meta->sampler) + ", "; + } else { + expr += '(' + GetSampler(meta->sampler) + '[' + Visit(meta->index).AsUint() + "], "; + } expr += coord_constructors.at(count + (has_array ? 1 : 0) + (has_shadow && !separate_dc ? 1 : 0) - 1); expr += '('; @@ -1310,6 +1335,8 @@ private: const std::string final_offset = fmt::format("({} - {}) >> 2", real, base); target = {fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset), Type::Uint}; + } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) { + target = {GetCustomVariable(cv->GetIndex()), Type::Float}; } else { UNREACHABLE_MSG("Assign called without a proper target"); } @@ -2237,6 +2264,10 @@ private: return GetDeclarationWithSuffix(index, "gpr"); } + std::string GetCustomVariable(u32 index) const { + return GetDeclarationWithSuffix(index, "custom_var"); + } + std::string GetPredicate(Tegra::Shader::Pred pred) const { return GetDeclarationWithSuffix(static_cast<u32>(pred), "pred"); } diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp index cf874a09a..1fc204f6f 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp @@ -53,7 +53,7 @@ struct BindlessSamplerKey { Tegra::Engines::SamplerDescriptor sampler{}; }; -constexpr u32 NativeVersion = 11; +constexpr u32 NativeVersion = 12; // Making sure sizes doesn't change by accident static_assert(sizeof(ProgramVariant) == 20); @@ -186,7 +186,8 @@ ShaderDiskCacheOpenGL::LoadTransferable() { u32 num_bound_samplers{}; u32 num_bindless_samplers{}; if (file.ReadArray(&usage.unique_identifier, 1) != 1 || - file.ReadArray(&usage.variant, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 || + file.ReadArray(&usage.variant, 1) != 1 || + file.ReadArray(&usage.bound_buffer, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 || file.ReadArray(&num_bound_samplers, 1) != 1 || file.ReadArray(&num_bindless_samplers, 1) != 1) { LOG_ERROR(Render_OpenGL, error_loading); @@ -281,7 +282,9 @@ ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) { u32 num_bindless_samplers{}; ShaderDiskCacheUsage usage; if (!LoadObjectFromPrecompiled(usage.unique_identifier) || - !LoadObjectFromPrecompiled(usage.variant) || !LoadObjectFromPrecompiled(num_keys) || + !LoadObjectFromPrecompiled(usage.variant) || + !LoadObjectFromPrecompiled(usage.bound_buffer) || + !LoadObjectFromPrecompiled(num_keys) || !LoadObjectFromPrecompiled(num_bound_samplers) || !LoadObjectFromPrecompiled(num_bindless_samplers)) { return {}; @@ -393,6 +396,7 @@ void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) { if (file.WriteObject(TransferableEntryKind::Usage) != 1 || file.WriteObject(usage.unique_identifier) != 1 || file.WriteObject(usage.variant) != 1 || + file.WriteObject(usage.bound_buffer) != 1 || file.WriteObject(static_cast<u32>(usage.keys.size())) != 1 || file.WriteObject(static_cast<u32>(usage.bound_samplers.size())) != 1 || file.WriteObject(static_cast<u32>(usage.bindless_samplers.size())) != 1) { @@ -447,7 +451,7 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p }; if (!SaveObjectToPrecompiled(usage.unique_identifier) || - !SaveObjectToPrecompiled(usage.variant) || + !SaveObjectToPrecompiled(usage.variant) || !SaveObjectToPrecompiled(usage.bound_buffer) || !SaveObjectToPrecompiled(static_cast<u32>(usage.keys.size())) || !SaveObjectToPrecompiled(static_cast<u32>(usage.bound_samplers.size())) || !SaveObjectToPrecompiled(static_cast<u32>(usage.bindless_samplers.size()))) { diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h index 69a2fbdda..ef2371f6d 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h @@ -79,6 +79,7 @@ static_assert(std::is_trivially_copyable_v<ProgramVariant>); struct ShaderDiskCacheUsage { u64 unique_identifier{}; ProgramVariant variant; + u32 bound_buffer{}; VideoCommon::Shader::KeyMap keys; VideoCommon::Shader::BoundSamplerMap bound_samplers; VideoCommon::Shader::BindlessSamplerMap bindless_samplers; diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp new file mode 100644 index 000000000..d5032b432 --- /dev/null +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -0,0 +1,265 @@ +// Copyright 2018 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <memory> +#include <optional> +#include <vector> + +#include <fmt/format.h> + +#include "common/assert.h" +#include "common/logging/log.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/settings.h" +#include "core/telemetry_session.h" +#include "video_core/gpu.h" +#include "video_core/renderer_vulkan/declarations.h" +#include "video_core/renderer_vulkan/renderer_vulkan.h" +#include "video_core/renderer_vulkan/vk_blit_screen.h" +#include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/vk_memory_manager.h" +#include "video_core/renderer_vulkan/vk_rasterizer.h" +#include "video_core/renderer_vulkan/vk_resource_manager.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_swapchain.h" + +namespace Vulkan { + +namespace { + +VkBool32 DebugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity_, + VkDebugUtilsMessageTypeFlagsEXT type, + const VkDebugUtilsMessengerCallbackDataEXT* data, + [[maybe_unused]] void* user_data) { + const vk::DebugUtilsMessageSeverityFlagBitsEXT severity{severity_}; + const char* message{data->pMessage}; + + if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eError) { + LOG_CRITICAL(Render_Vulkan, "{}", message); + } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning) { + LOG_WARNING(Render_Vulkan, "{}", message); + } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo) { + LOG_INFO(Render_Vulkan, "{}", message); + } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose) { + LOG_DEBUG(Render_Vulkan, "{}", message); + } + return VK_FALSE; +} + +std::string GetReadableVersion(u32 version) { + return fmt::format("{}.{}.{}", VK_VERSION_MAJOR(version), VK_VERSION_MINOR(version), + VK_VERSION_PATCH(version)); +} + +std::string GetDriverVersion(const VKDevice& device) { + // Extracted from + // https://github.com/SaschaWillems/vulkan.gpuinfo.org/blob/5dddea46ea1120b0df14eef8f15ff8e318e35462/functions.php#L308-L314 + const u32 version = device.GetDriverVersion(); + + if (device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) { + const u32 major = (version >> 22) & 0x3ff; + const u32 minor = (version >> 14) & 0x0ff; + const u32 secondary = (version >> 6) & 0x0ff; + const u32 tertiary = version & 0x003f; + return fmt::format("{}.{}.{}.{}", major, minor, secondary, tertiary); + } + if (device.GetDriverID() == vk::DriverIdKHR::eIntelProprietaryWindows) { + const u32 major = version >> 14; + const u32 minor = version & 0x3fff; + return fmt::format("{}.{}", major, minor); + } + + return GetReadableVersion(version); +} + +std::string BuildCommaSeparatedExtensions(std::vector<std::string> available_extensions) { + std::sort(std::begin(available_extensions), std::end(available_extensions)); + + static constexpr std::size_t AverageExtensionSize = 64; + std::string separated_extensions; + separated_extensions.reserve(available_extensions.size() * AverageExtensionSize); + + const auto end = std::end(available_extensions); + for (auto extension = std::begin(available_extensions); extension != end; ++extension) { + if (const bool is_last = extension + 1 == end; is_last) { + separated_extensions += *extension; + } else { + separated_extensions += fmt::format("{},", *extension); + } + } + return separated_extensions; +} + +} // Anonymous namespace + +RendererVulkan::RendererVulkan(Core::Frontend::EmuWindow& window, Core::System& system) + : RendererBase(window), system{system} {} + +RendererVulkan::~RendererVulkan() { + ShutDown(); +} + +void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { + const auto& layout = render_window.GetFramebufferLayout(); + if (framebuffer && layout.width > 0 && layout.height > 0 && render_window.IsShown()) { + const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; + const bool use_accelerated = + rasterizer->AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); + const bool is_srgb = use_accelerated && screen_info.is_srgb; + if (swapchain->HasFramebufferChanged(layout) || swapchain->GetSrgbState() != is_srgb) { + swapchain->Create(layout.width, layout.height, is_srgb); + blit_screen->Recreate(); + } + + scheduler->WaitWorker(); + + swapchain->AcquireNextImage(); + const auto [fence, render_semaphore] = blit_screen->Draw(*framebuffer, use_accelerated); + + scheduler->Flush(false, render_semaphore); + + if (swapchain->Present(render_semaphore, fence)) { + blit_screen->Recreate(); + } + + render_window.SwapBuffers(); + rasterizer->TickFrame(); + } + + render_window.PollEvents(); +} + +bool RendererVulkan::Init() { + PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr{}; + render_window.RetrieveVulkanHandlers(&vkGetInstanceProcAddr, &instance, &surface); + const vk::DispatchLoaderDynamic dldi(instance, vkGetInstanceProcAddr); + + std::optional<vk::DebugUtilsMessengerEXT> callback; + if (Settings::values.renderer_debug && dldi.vkCreateDebugUtilsMessengerEXT) { + callback = CreateDebugCallback(dldi); + if (!callback) { + return false; + } + } + + if (!PickDevices(dldi)) { + if (callback) { + instance.destroy(*callback, nullptr, dldi); + } + return false; + } + debug_callback = UniqueDebugUtilsMessengerEXT( + *callback, vk::ObjectDestroy<vk::Instance, vk::DispatchLoaderDynamic>( + instance, nullptr, device->GetDispatchLoader())); + + Report(); + + memory_manager = std::make_unique<VKMemoryManager>(*device); + + resource_manager = std::make_unique<VKResourceManager>(*device); + + const auto& framebuffer = render_window.GetFramebufferLayout(); + swapchain = std::make_unique<VKSwapchain>(surface, *device); + swapchain->Create(framebuffer.width, framebuffer.height, false); + + scheduler = std::make_unique<VKScheduler>(*device, *resource_manager); + + rasterizer = std::make_unique<RasterizerVulkan>(system, render_window, screen_info, *device, + *resource_manager, *memory_manager, *scheduler); + + blit_screen = std::make_unique<VKBlitScreen>(system, render_window, *rasterizer, *device, + *resource_manager, *memory_manager, *swapchain, + *scheduler, screen_info); + + return true; +} + +void RendererVulkan::ShutDown() { + if (!device) { + return; + } + const auto dev = device->GetLogical(); + const auto& dld = device->GetDispatchLoader(); + if (dev && dld.vkDeviceWaitIdle) { + dev.waitIdle(dld); + } + + rasterizer.reset(); + blit_screen.reset(); + scheduler.reset(); + swapchain.reset(); + memory_manager.reset(); + resource_manager.reset(); + device.reset(); +} + +std::optional<vk::DebugUtilsMessengerEXT> RendererVulkan::CreateDebugCallback( + const vk::DispatchLoaderDynamic& dldi) { + const vk::DebugUtilsMessengerCreateInfoEXT callback_ci( + {}, + vk::DebugUtilsMessageSeverityFlagBitsEXT::eError | + vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning | + vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo | + vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose, + vk::DebugUtilsMessageTypeFlagBitsEXT::eGeneral | + vk::DebugUtilsMessageTypeFlagBitsEXT::eValidation | + vk::DebugUtilsMessageTypeFlagBitsEXT::ePerformance, + &DebugCallback, nullptr); + vk::DebugUtilsMessengerEXT callback; + if (instance.createDebugUtilsMessengerEXT(&callback_ci, nullptr, &callback, dldi) != + vk::Result::eSuccess) { + LOG_ERROR(Render_Vulkan, "Failed to create debug callback"); + return {}; + } + return callback; +} + +bool RendererVulkan::PickDevices(const vk::DispatchLoaderDynamic& dldi) { + const auto devices = instance.enumeratePhysicalDevices(dldi); + + // TODO(Rodrigo): Choose device from config file + const s32 device_index = Settings::values.vulkan_device; + if (device_index < 0 || device_index >= static_cast<s32>(devices.size())) { + LOG_ERROR(Render_Vulkan, "Invalid device index {}!", device_index); + return false; + } + const vk::PhysicalDevice physical_device = devices[device_index]; + + if (!VKDevice::IsSuitable(dldi, physical_device, surface)) { + return false; + } + + device = std::make_unique<VKDevice>(dldi, physical_device, surface); + return device->Create(dldi, instance); +} + +void RendererVulkan::Report() const { + const std::string vendor_name{device->GetVendorName()}; + const std::string model_name{device->GetModelName()}; + const std::string driver_version = GetDriverVersion(*device); + const std::string driver_name = fmt::format("{} {}", vendor_name, driver_version); + + const std::string api_version = GetReadableVersion(device->GetApiVersion()); + + const std::string extensions = BuildCommaSeparatedExtensions(device->GetAvailableExtensions()); + + LOG_INFO(Render_Vulkan, "Driver: {}", driver_name); + LOG_INFO(Render_Vulkan, "Device: {}", model_name); + LOG_INFO(Render_Vulkan, "Vulkan: {}", api_version); + + auto& telemetry_session = system.TelemetrySession(); + constexpr auto field = Telemetry::FieldType::UserSystem; + telemetry_session.AddField(field, "GPU_Vendor", vendor_name); + telemetry_session.AddField(field, "GPU_Model", model_name); + telemetry_session.AddField(field, "GPU_Vulkan_Driver", driver_name); + telemetry_session.AddField(field, "GPU_Vulkan_Version", api_version); + telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions); +} + +} // namespace Vulkan
\ No newline at end of file diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp index 939eebe83..9840f26e5 100644 --- a/src/video_core/renderer_vulkan/vk_device.cpp +++ b/src/video_core/renderer_vulkan/vk_device.cpp @@ -400,8 +400,10 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true); Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false); - Test(extension, nv_device_diagnostic_checkpoints, - VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true); + if (Settings::values.renderer_debug) { + Test(extension, nv_device_diagnostic_checkpoints, + VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true); + } } if (khr_shader_float16_int8) { diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index d2c6b1189..aada38702 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -571,7 +571,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() { color_attachments[rt] = texture_cache.GetColorBufferSurface(rt, true); } if (color_attachments[rt] && WalkAttachmentOverlaps(*color_attachments[rt])) { - texceptions.set(rt); + texceptions[rt] = true; } } @@ -579,7 +579,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() { zeta_attachment = texture_cache.GetDepthBufferSurface(true); } if (zeta_attachment && WalkAttachmentOverlaps(*zeta_attachment)) { - texceptions.set(ZETA_TEXCEPTION_INDEX); + texceptions[ZETA_TEXCEPTION_INDEX] = true; } texture_cache.GuardRenderTargets(false); @@ -1122,11 +1122,12 @@ RenderPassParams RasterizerVulkan::GetRenderPassParams(Texceptions texceptions) for (std::size_t rt = 0; rt < static_cast<std::size_t>(regs.rt_control.count); ++rt) { const auto& rendertarget = regs.rt[rt]; - if (rendertarget.Address() == 0 || rendertarget.format == Tegra::RenderTargetFormat::NONE) + if (rendertarget.Address() == 0 || rendertarget.format == Tegra::RenderTargetFormat::NONE) { continue; + } renderpass_params.color_attachments.push_back(RenderPassParams::ColorAttachment{ static_cast<u32>(rt), PixelFormatFromRenderTargetFormat(rendertarget.format), - texceptions.test(rt)}); + texceptions[rt]}); } renderpass_params.has_zeta = regs.zeta_enable; diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index 1ab22251e..24a658dce 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -353,6 +353,7 @@ private: DeclareFragment(); DeclareCompute(); DeclareRegisters(); + DeclareCustomVariables(); DeclarePredicates(); DeclareLocalMemory(); DeclareSharedMemory(); @@ -586,6 +587,15 @@ private: } } + void DeclareCustomVariables() { + const u32 num_custom_variables = ir.GetNumCustomVariables(); + for (u32 i = 0; i < num_custom_variables; ++i) { + const Id id = OpVariable(t_prv_float, spv::StorageClass::Private, v_float_zero); + Name(id, fmt::format("custom_var_{}", i)); + custom_variables.emplace(i, AddGlobalVariable(id)); + } + } + void DeclarePredicates() { for (const auto pred : ir.GetPredicates()) { const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false); @@ -982,6 +992,11 @@ private: return {OpLoad(t_float, registers.at(index)), Type::Float}; } + if (const auto cv = std::get_if<CustomVarNode>(&*node)) { + const u32 index = cv->GetIndex(); + return {OpLoad(t_float, custom_variables.at(index)), Type::Float}; + } + if (const auto immediate = std::get_if<ImmediateNode>(&*node)) { return {Constant(t_uint, immediate->GetValue()), Type::Uint}; } @@ -1333,6 +1348,9 @@ private: } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { target = {GetGlobalMemoryPointer(*gmem), Type::Uint}; + } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) { + target = {custom_variables.at(cv->GetIndex()), Type::Float}; + } else { UNIMPLEMENTED(); } @@ -2508,6 +2526,7 @@ private: Id out_vertex{}; Id in_vertex{}; std::map<u32, Id> registers; + std::map<u32, Id> custom_variables; std::map<Tegra::Shader::Pred, Id> predicates; std::map<u32, Id> flow_variables; Id local_memory{}; diff --git a/src/video_core/shader/ast.h b/src/video_core/shader/ast.h index a2f0044ba..cca13bcde 100644 --- a/src/video_core/shader/ast.h +++ b/src/video_core/shader/ast.h @@ -65,8 +65,8 @@ public: void DetachSegment(ASTNode start, ASTNode end); void Remove(ASTNode node); - ASTNode first{}; - ASTNode last{}; + ASTNode first; + ASTNode last; }; class ASTProgram { @@ -299,9 +299,9 @@ private: friend class ASTZipper; ASTData data; - ASTNode parent{}; - ASTNode next{}; - ASTNode previous{}; + ASTNode parent; + ASTNode next; + ASTNode previous; ASTZipper* manager{}; }; diff --git a/src/video_core/shader/const_buffer_locker.cpp b/src/video_core/shader/const_buffer_locker.cpp index a4a0319eb..0638be8cb 100644 --- a/src/video_core/shader/const_buffer_locker.cpp +++ b/src/video_core/shader/const_buffer_locker.cpp @@ -66,6 +66,18 @@ std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindle return value; } +std::optional<u32> ConstBufferLocker::ObtainBoundBuffer() { + if (bound_buffer_saved) { + return bound_buffer; + } + if (!engine) { + return std::nullopt; + } + bound_buffer_saved = true; + bound_buffer = engine->GetBoundBuffer(); + return bound_buffer; +} + void ConstBufferLocker::InsertKey(u32 buffer, u32 offset, u32 value) { keys.insert_or_assign({buffer, offset}, value); } @@ -78,6 +90,11 @@ void ConstBufferLocker::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDes bindless_samplers.insert_or_assign({buffer, offset}, sampler); } +void ConstBufferLocker::SetBoundBuffer(u32 buffer) { + bound_buffer_saved = true; + bound_buffer = buffer; +} + bool ConstBufferLocker::IsConsistent() const { if (!engine) { return false; diff --git a/src/video_core/shader/const_buffer_locker.h b/src/video_core/shader/const_buffer_locker.h index d32e2d657..d3ea11087 100644 --- a/src/video_core/shader/const_buffer_locker.h +++ b/src/video_core/shader/const_buffer_locker.h @@ -10,6 +10,7 @@ #include "common/hash.h" #include "video_core/engines/const_buffer_engine_interface.h" #include "video_core/engines/shader_type.h" +#include "video_core/guest_driver.h" namespace VideoCommon::Shader { @@ -40,6 +41,8 @@ public: std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset); + std::optional<u32> ObtainBoundBuffer(); + /// Inserts a key. void InsertKey(u32 buffer, u32 offset, u32 value); @@ -49,6 +52,9 @@ public: /// Inserts a bindless sampler key. void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler); + /// Set the bound buffer for this locker. + void SetBoundBuffer(u32 buffer); + /// Checks keys and samplers against engine's current const buffers. Returns true if they are /// the same value, false otherwise; bool IsConsistent() const; @@ -71,12 +77,27 @@ public: return bindless_samplers; } + /// Gets bound buffer used on this shader + u32 GetBoundBuffer() const { + return bound_buffer; + } + + /// Obtains access to the guest driver's profile. + VideoCore::GuestDriverProfile* AccessGuestDriverProfile() const { + if (engine) { + return &engine->AccessGuestDriverProfile(); + } + return nullptr; + } + private: const Tegra::Engines::ShaderType stage; Tegra::Engines::ConstBufferEngineInterface* engine = nullptr; KeyMap keys; BoundSamplerMap bound_samplers; BindlessSamplerMap bindless_samplers; + bool bound_buffer_saved{}; + u32 bound_buffer{}; }; } // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode.cpp b/src/video_core/shader/decode.cpp index 22c3e5120..6b697ed5d 100644 --- a/src/video_core/shader/decode.cpp +++ b/src/video_core/shader/decode.cpp @@ -3,6 +3,7 @@ // Refer to the license.txt file included. #include <cstring> +#include <limits> #include <set> #include <fmt/format.h> @@ -33,6 +34,52 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) { return (absolute_offset % SchedPeriod) == 0; } +void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver, + const std::list<Sampler>& used_samplers) { + if (gpu_driver == nullptr) { + LOG_CRITICAL(HW_GPU, "GPU driver profile has not been created yet"); + return; + } + if (gpu_driver->TextureHandlerSizeKnown() || used_samplers.size() <= 1) { + return; + } + u32 count{}; + std::vector<u32> bound_offsets; + for (const auto& sampler : used_samplers) { + if (sampler.IsBindless()) { + continue; + } + ++count; + bound_offsets.emplace_back(sampler.GetOffset()); + } + if (count > 1) { + gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets)); + } +} + +std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce, + VideoCore::GuestDriverProfile* gpu_driver, + const std::list<Sampler>& used_samplers) { + if (gpu_driver == nullptr) { + LOG_CRITICAL(HW_GPU, "GPU Driver profile has not been created yet"); + return std::nullopt; + } + const u32 base_offset = sampler_to_deduce.GetOffset(); + u32 max_offset{std::numeric_limits<u32>::max()}; + for (const auto& sampler : used_samplers) { + if (sampler.IsBindless()) { + continue; + } + if (sampler.GetOffset() > base_offset) { + max_offset = std::min(sampler.GetOffset(), max_offset); + } + } + if (max_offset == std::numeric_limits<u32>::max()) { + return std::nullopt; + } + return ((max_offset - base_offset) * 4) / gpu_driver->GetTextureHandlerSize(); +} + } // Anonymous namespace class ASTDecoder { @@ -315,4 +362,25 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) { return pc + 1; } +void ShaderIR::PostDecode() { + // Deduce texture handler size if needed + auto gpu_driver = locker.AccessGuestDriverProfile(); + DeduceTextureHandlerSize(gpu_driver, used_samplers); + // Deduce Indexed Samplers + if (!uses_indexed_samplers) { + return; + } + for (auto& sampler : used_samplers) { + if (!sampler.IsIndexed()) { + continue; + } + if (const auto size = TryDeduceSamplerSize(sampler, gpu_driver, used_samplers)) { + sampler.SetSize(*size); + } else { + LOG_CRITICAL(HW_GPU, "Failed to deduce size of indexed sampler"); + sampler.SetSize(1); + } + } +} + } // namespace VideoCommon::Shader diff --git a/src/video_core/shader/decode/arithmetic.cpp b/src/video_core/shader/decode/arithmetic.cpp index fcedd2af6..90240c765 100644 --- a/src/video_core/shader/decode/arithmetic.cpp +++ b/src/video_core/shader/decode/arithmetic.cpp @@ -21,7 +21,7 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) { Node op_a = GetRegister(instr.gpr8); - Node op_b = [&]() -> Node { + Node op_b = [&] { if (instr.is_b_imm) { return GetImmediate19(instr); } else if (instr.is_b_gpr) { @@ -141,6 +141,15 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) { SetRegister(bb, instr.gpr0, value); break; } + case OpCode::Id::FCMP_R: { + UNIMPLEMENTED_IF(instr.fcmp.ftz == 0); + Node op_c = GetRegister(instr.gpr39); + Node comp = GetPredicateComparisonFloat(instr.fcmp.cond, std::move(op_c), Immediate(0.0f)); + SetRegister( + bb, instr.gpr0, + Operation(OperationCode::Select, std::move(comp), std::move(op_a), std::move(op_b))); + break; + } case OpCode::Id::RRO_C: case OpCode::Id::RRO_R: case OpCode::Id::RRO_IMM: { diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp index 371fae127..e60875cc4 100644 --- a/src/video_core/shader/decode/arithmetic_integer.cpp +++ b/src/video_core/shader/decode/arithmetic_integer.cpp @@ -297,7 +297,7 @@ void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Nod const Node one = Immediate(1); const Node two = Immediate(2); - Node value{}; + Node value; for (u32 i = 0; i < lop_iterations; ++i) { const Node shift_amount = Immediate(i); diff --git a/src/video_core/shader/decode/bfi.cpp b/src/video_core/shader/decode/bfi.cpp index 8be1119df..f992bbe2a 100644 --- a/src/video_core/shader/decode/bfi.cpp +++ b/src/video_core/shader/decode/bfi.cpp @@ -17,10 +17,13 @@ u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; const auto opcode = OpCode::Decode(instr); - const auto [base, packed_shift] = [&]() -> std::tuple<Node, Node> { + const auto [packed_shift, base] = [&]() -> std::pair<Node, Node> { switch (opcode->get().GetId()) { + case OpCode::Id::BFI_RC: + return {GetRegister(instr.gpr39), + GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset)}; case OpCode::Id::BFI_IMM_R: - return {GetRegister(instr.gpr39), Immediate(instr.alu.GetSignedImm20_20())}; + return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)}; default: UNREACHABLE(); return {Immediate(0), Immediate(0)}; diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp index 3da833e81..b5fbc4d58 100644 --- a/src/video_core/shader/decode/memory.cpp +++ b/src/video_core/shader/decode/memory.cpp @@ -24,6 +24,7 @@ using Tegra::Shader::GlobalAtomicType; using Tegra::Shader::Instruction; using Tegra::Shader::OpCode; using Tegra::Shader::Register; +using Tegra::Shader::StoreType; namespace { @@ -63,6 +64,27 @@ u32 GetMemorySize(Tegra::Shader::UniformType uniform_type) { } } +Node ExtractUnaligned(Node value, Node address, u32 mask, u32 size) { + Node offset = Operation(OperationCode::UBitwiseAnd, address, Immediate(mask)); + offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3)); + return Operation(OperationCode::UBitfieldExtract, std::move(value), std::move(offset), + Immediate(size)); +} + +Node InsertUnaligned(Node dest, Node value, Node address, u32 mask, u32 size) { + Node offset = Operation(OperationCode::UBitwiseAnd, std::move(address), Immediate(mask)); + offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3)); + return Operation(OperationCode::UBitfieldInsert, std::move(dest), std::move(value), + std::move(offset), Immediate(size)); +} + +Node Sign16Extend(Node value) { + Node sign = Operation(OperationCode::UBitwiseAnd, value, Immediate(1U << 15)); + Node is_sign = Operation(OperationCode::LogicalUEqual, std::move(sign), Immediate(1U << 15)); + Node extend = Operation(OperationCode::Select, is_sign, Immediate(0xFFFF0000), Immediate(0)); + return Operation(OperationCode::UBitwiseOr, std::move(value), std::move(extend)); +} + } // Anonymous namespace u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { @@ -138,26 +160,31 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}", static_cast<u64>(instr.ld_l.unknown)); [[fallthrough]]; case OpCode::Id::LD_S: { - const auto GetMemory = [&](s32 offset) { + const auto GetAddress = [&](s32 offset) { ASSERT(offset % 4 == 0); const Node immediate_offset = Immediate(static_cast<s32>(instr.smem_imm) + offset); - const Node address = Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), - immediate_offset); - return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(address) - : GetLocalMemory(address); + return Operation(OperationCode::IAdd, GetRegister(instr.gpr8), immediate_offset); + }; + const auto GetMemory = [&](s32 offset) { + return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(GetAddress(offset)) + : GetLocalMemory(GetAddress(offset)); }; switch (instr.ldst_sl.type.Value()) { - case Tegra::Shader::StoreType::Bits32: - case Tegra::Shader::StoreType::Bits64: - case Tegra::Shader::StoreType::Bits128: { - const u32 count = [&]() { + case StoreType::Signed16: + SetRegister(bb, instr.gpr0, + Sign16Extend(ExtractUnaligned(GetMemory(0), GetAddress(0), 0b10, 16))); + break; + case StoreType::Bits32: + case StoreType::Bits64: + case StoreType::Bits128: { + const u32 count = [&] { switch (instr.ldst_sl.type.Value()) { - case Tegra::Shader::StoreType::Bits32: + case StoreType::Bits32: return 1; - case Tegra::Shader::StoreType::Bits64: + case StoreType::Bits64: return 2; - case Tegra::Shader::StoreType::Bits128: + case StoreType::Bits128: return 4; default: UNREACHABLE(); @@ -214,12 +241,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { // To handle unaligned loads get the bytes used to dereference global memory and extract // those bytes from the loaded u32. if (IsUnaligned(type)) { - Node mask = Immediate(GetUnalignedMask(type)); - Node offset = Operation(OperationCode::UBitwiseAnd, real_address, std::move(mask)); - offset = Operation(OperationCode::ULogicalShiftLeft, offset, Immediate(3)); - - gmem = Operation(OperationCode::UBitfieldExtract, std::move(gmem), - std::move(offset), Immediate(size)); + gmem = ExtractUnaligned(gmem, real_address, GetUnalignedMask(type), size); } SetTemporary(bb, i, gmem); @@ -271,21 +293,28 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { return Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), immediate); }; - const auto set_memory = opcode->get().GetId() == OpCode::Id::ST_L - ? &ShaderIR::SetLocalMemory - : &ShaderIR::SetSharedMemory; + const bool is_local = opcode->get().GetId() == OpCode::Id::ST_L; + const auto set_memory = is_local ? &ShaderIR::SetLocalMemory : &ShaderIR::SetSharedMemory; + const auto get_memory = is_local ? &ShaderIR::GetLocalMemory : &ShaderIR::GetSharedMemory; switch (instr.ldst_sl.type.Value()) { - case Tegra::Shader::StoreType::Bits128: + case StoreType::Bits128: (this->*set_memory)(bb, GetAddress(12), GetRegister(instr.gpr0.Value() + 3)); (this->*set_memory)(bb, GetAddress(8), GetRegister(instr.gpr0.Value() + 2)); [[fallthrough]]; - case Tegra::Shader::StoreType::Bits64: + case StoreType::Bits64: (this->*set_memory)(bb, GetAddress(4), GetRegister(instr.gpr0.Value() + 1)); [[fallthrough]]; - case Tegra::Shader::StoreType::Bits32: + case StoreType::Bits32: (this->*set_memory)(bb, GetAddress(0), GetRegister(instr.gpr0)); break; + case StoreType::Signed16: { + Node address = GetAddress(0); + Node memory = (this->*get_memory)(address); + (this->*set_memory)( + bb, address, InsertUnaligned(memory, GetRegister(instr.gpr0), address, 0b10, 16)); + break; + } default: UNIMPLEMENTED_MSG("{} unhandled type: {}", opcode->get().GetName(), static_cast<u32>(instr.ldst_sl.type.Value())); @@ -325,12 +354,8 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) { Node value = GetRegister(instr.gpr0.Value() + i); if (IsUnaligned(type)) { - Node mask = Immediate(GetUnalignedMask(type)); - Node offset = Operation(OperationCode::UBitwiseAnd, real_address, std::move(mask)); - offset = Operation(OperationCode::ULogicalShiftLeft, offset, Immediate(3)); - - value = Operation(OperationCode::UBitfieldInsert, gmem, std::move(value), offset, - Immediate(size)); + const u32 mask = GetUnalignedMask(type); + value = InsertUnaligned(gmem, std::move(value), real_address, mask, size); } bb.push_back(Operation(OperationCode::Assign, gmem, value)); diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp index 7321698b2..4944e9d69 100644 --- a/src/video_core/shader/decode/other.cpp +++ b/src/video_core/shader/decode/other.cpp @@ -69,13 +69,16 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { case OpCode::Id::MOV_SYS: { const Node value = [this, instr] { switch (instr.sys20) { + case SystemVariable::LaneId: + LOG_WARNING(HW_GPU, "MOV_SYS instruction with LaneId is incomplete"); + return Immediate(0U); case SystemVariable::InvocationId: return Operation(OperationCode::InvocationId); case SystemVariable::Ydirection: return Operation(OperationCode::YNegate); case SystemVariable::InvocationInfo: LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete"); - return Immediate(0u); + return Immediate(0U); case SystemVariable::Tid: { Node value = Immediate(0); value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdX), 0, 9); @@ -188,7 +191,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "SYNC condition code used: {}", static_cast<u32>(cc)); - if (disable_flow_stack) { + if (decompiled) { break; } @@ -200,7 +203,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { const Tegra::Shader::ConditionCode cc = instr.flow_condition_code; UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "BRK condition code used: {}", static_cast<u32>(cc)); - if (disable_flow_stack) { + if (decompiled) { break; } diff --git a/src/video_core/shader/decode/texture.cpp b/src/video_core/shader/decode/texture.cpp index 0b567e39d..351c8c2f1 100644 --- a/src/video_core/shader/decode/texture.cpp +++ b/src/video_core/shader/decode/texture.cpp @@ -144,7 +144,8 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, {}, depth_compare, aoffi, {}, {}, {}, {}, component, element}; + MetaTexture meta{sampler, {}, depth_compare, aoffi, {}, {}, + {}, {}, component, element, {}}; values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy)); } @@ -167,9 +168,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { const auto derivate_reg = instr.gpr20.Value(); const auto texture_type = instr.txd.texture_type.Value(); const auto coord_count = GetCoordCount(texture_type); - + Node index_var{}; const Sampler* sampler = - is_bindless ? GetBindlessSampler(base_reg, {{texture_type, is_array, false}}) + is_bindless ? GetBindlessSampler(base_reg, index_var, {{texture_type, is_array, false}}) : GetSampler(instr.sampler, {{texture_type, is_array, false}}); Node4 values; if (sampler == nullptr) { @@ -200,7 +201,8 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { } for (u32 element = 0; element < values.size(); ++element) { - MetaTexture meta{*sampler, array_node, {}, {}, {}, derivates, {}, {}, {}, element}; + MetaTexture meta{*sampler, array_node, {}, {}, {}, derivates, + {}, {}, {}, element, index_var}; values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords); } @@ -215,8 +217,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { // TODO: The new commits on the texture refactor, change the way samplers work. // Sadly, not all texture instructions specify the type of texture their sampler // uses. This must be fixed at a later instance. + Node index_var{}; const Sampler* sampler = - is_bindless ? GetBindlessSampler(instr.gpr8) : GetSampler(instr.sampler); + is_bindless ? GetBindlessSampler(instr.gpr8, index_var) : GetSampler(instr.sampler); if (sampler == nullptr) { u32 indexer = 0; @@ -240,7 +243,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { if (!instr.txq.IsComponentEnabled(element)) { continue; } - MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element}; + MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var}; const Node value = Operation(OperationCode::TextureQueryDimensions, meta, GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0))); @@ -266,8 +269,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { auto texture_type = instr.tmml.texture_type.Value(); const bool is_array = instr.tmml.array != 0; + Node index_var{}; const Sampler* sampler = - is_bindless ? GetBindlessSampler(instr.gpr20) : GetSampler(instr.sampler); + is_bindless ? GetBindlessSampler(instr.gpr20, index_var) : GetSampler(instr.sampler); if (sampler == nullptr) { u32 indexer = 0; @@ -309,7 +313,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) { continue; } auto params = coords; - MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element}; + MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, {}, element, index_var}; const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params)); SetTemporary(bb, indexer++, value); } @@ -383,37 +387,65 @@ const Sampler* ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, // Otherwise create a new mapping for this sampler const auto next_index = static_cast<u32>(used_samplers.size()); return &used_samplers.emplace_back(next_index, offset, info.type, info.is_array, info.is_shadow, - info.is_buffer); + info.is_buffer, false); } -const Sampler* ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg, +const Sampler* ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg, Node& index_var, std::optional<SamplerInfo> sampler_info) { const Node sampler_register = GetRegister(reg); - const auto [base_sampler, buffer, offset] = - TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size())); - ASSERT(base_sampler != nullptr); - if (base_sampler == nullptr) { + const auto [base_node, tracked_sampler_info] = + TrackBindlessSampler(sampler_register, global_code, static_cast<s64>(global_code.size())); + ASSERT(base_node != nullptr); + if (base_node == nullptr) { return nullptr; } - const auto info = GetSamplerInfo(sampler_info, offset, buffer); + if (const auto bindless_sampler_info = + std::get_if<BindlessSamplerNode>(&*tracked_sampler_info)) { + const u32 buffer = bindless_sampler_info->GetIndex(); + const u32 offset = bindless_sampler_info->GetOffset(); + const auto info = GetSamplerInfo(sampler_info, offset, buffer); + + // If this sampler has already been used, return the existing mapping. + const auto it = + std::find_if(used_samplers.begin(), used_samplers.end(), + [buffer = buffer, offset = offset](const Sampler& entry) { + return entry.GetBuffer() == buffer && entry.GetOffset() == offset; + }); + if (it != used_samplers.end()) { + ASSERT(it->IsBindless() && it->GetType() == info.type && + it->IsArray() == info.is_array && it->IsShadow() == info.is_shadow); + return &*it; + } - // If this sampler has already been used, return the existing mapping. - const auto it = - std::find_if(used_samplers.begin(), used_samplers.end(), - [buffer = buffer, offset = offset](const Sampler& entry) { - return entry.GetBuffer() == buffer && entry.GetOffset() == offset; - }); - if (it != used_samplers.end()) { - ASSERT(it->IsBindless() && it->GetType() == info.type && it->IsArray() == info.is_array && - it->IsShadow() == info.is_shadow); - return &*it; - } + // Otherwise create a new mapping for this sampler + const auto next_index = static_cast<u32>(used_samplers.size()); + return &used_samplers.emplace_back(next_index, offset, buffer, info.type, info.is_array, + info.is_shadow, info.is_buffer, false); + } else if (const auto array_sampler_info = + std::get_if<ArraySamplerNode>(&*tracked_sampler_info)) { + const u32 base_offset = array_sampler_info->GetBaseOffset() / 4; + index_var = GetCustomVariable(array_sampler_info->GetIndexVar()); + const auto info = GetSamplerInfo(sampler_info, base_offset); + + // If this sampler has already been used, return the existing mapping. + const auto it = std::find_if( + used_samplers.begin(), used_samplers.end(), + [base_offset](const Sampler& entry) { return entry.GetOffset() == base_offset; }); + if (it != used_samplers.end()) { + ASSERT(!it->IsBindless() && it->GetType() == info.type && + it->IsArray() == info.is_array && it->IsShadow() == info.is_shadow && + it->IsBuffer() == info.is_buffer && it->IsIndexed()); + return &*it; + } - // Otherwise create a new mapping for this sampler - const auto next_index = static_cast<u32>(used_samplers.size()); - return &used_samplers.emplace_back(next_index, offset, buffer, info.type, info.is_array, - info.is_shadow, info.is_buffer); + uses_indexed_samplers = true; + // Otherwise create a new mapping for this sampler + const auto next_index = static_cast<u32>(used_samplers.size()); + return &used_samplers.emplace_back(next_index, base_offset, info.type, info.is_array, + info.is_shadow, info.is_buffer, true); + } + return nullptr; } void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) { @@ -499,8 +531,9 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, "This method is not supported."); const SamplerInfo info{texture_type, is_array, is_shadow, false}; - const Sampler* sampler = - is_bindless ? GetBindlessSampler(*bindless_reg, info) : GetSampler(instr.sampler, info); + Node index_var{}; + const Sampler* sampler = is_bindless ? GetBindlessSampler(*bindless_reg, index_var, info) + : GetSampler(instr.sampler, info); Node4 values; if (sampler == nullptr) { for (u32 element = 0; element < values.size(); ++element) { @@ -548,7 +581,8 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, for (u32 element = 0; element < values.size(); ++element) { auto copy_coords = coords; - MetaTexture meta{*sampler, array, depth_compare, aoffi, {}, {}, bias, lod, {}, element}; + MetaTexture meta{*sampler, array, depth_compare, aoffi, {}, {}, bias, + lod, {}, element, index_var}; values[element] = Operation(read_method, meta, std::move(copy_coords)); } @@ -596,7 +630,7 @@ Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type, aoffi = GetAoffiCoordinates(GetRegister(parameter_register++), coord_count, false); } - Node dc{}; + Node dc; if (depth_compare) { // Depth is always stored in the register signaled by gpr20 or in the next register if lod // or bias are used @@ -632,7 +666,7 @@ Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type, const Node array = is_array ? GetRegister(array_register) : nullptr; - Node dc{}; + Node dc; if (depth_compare) { // Depth is always stored in the register signaled by gpr20 or in the next register if lod // or bias are used @@ -663,7 +697,8 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de u64 parameter_register = instr.gpr20.Value(); const SamplerInfo info{texture_type, is_array, depth_compare, false}; - const Sampler* sampler = is_bindless ? GetBindlessSampler(parameter_register++, info) + Node index_var{}; + const Sampler* sampler = is_bindless ? GetBindlessSampler(parameter_register++, index_var, info) : GetSampler(instr.sampler, info); Node4 values; if (sampler == nullptr) { @@ -692,7 +727,8 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; MetaTexture meta{ - *sampler, GetRegister(array_register), dc, aoffi, ptp, {}, {}, {}, component, element}; + *sampler, GetRegister(array_register), dc, aoffi, ptp, {}, {}, {}, component, element, + index_var}; values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy)); } @@ -725,7 +761,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) { Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, array_register, {}, {}, {}, {}, {}, lod, {}, element}; + MetaTexture meta{sampler, array_register, {}, {}, {}, {}, {}, lod, {}, element, {}}; values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy)); } @@ -775,7 +811,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is Node4 values; for (u32 element = 0; element < values.size(); ++element) { auto coords_copy = coords; - MetaTexture meta{sampler, array, {}, {}, {}, {}, {}, lod, {}, element}; + MetaTexture meta{sampler, array, {}, {}, {}, {}, {}, lod, {}, element, {}}; values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy)); } return values; diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h index 9af1f0228..a0a7b9111 100644 --- a/src/video_core/shader/node.h +++ b/src/video_core/shader/node.h @@ -212,6 +212,7 @@ enum class MetaStackClass { class OperationNode; class ConditionalNode; class GprNode; +class CustomVarNode; class ImmediateNode; class InternalFlagNode; class PredicateNode; @@ -223,26 +224,32 @@ class SmemNode; class GmemNode; class CommentNode; -using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode, +using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, CustomVarNode, ImmediateNode, InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode, LmemNode, SmemNode, GmemNode, CommentNode>; using Node = std::shared_ptr<NodeData>; using Node4 = std::array<Node, 4>; using NodeBlock = std::vector<Node>; +class BindlessSamplerNode; +class ArraySamplerNode; + +using TrackSamplerData = std::variant<BindlessSamplerNode, ArraySamplerNode>; +using TrackSampler = std::shared_ptr<TrackSamplerData>; + class Sampler { public: /// This constructor is for bound samplers constexpr explicit Sampler(u32 index, u32 offset, Tegra::Shader::TextureType type, - bool is_array, bool is_shadow, bool is_buffer) + bool is_array, bool is_shadow, bool is_buffer, bool is_indexed) : index{index}, offset{offset}, type{type}, is_array{is_array}, is_shadow{is_shadow}, - is_buffer{is_buffer} {} + is_buffer{is_buffer}, is_indexed{is_indexed} {} /// This constructor is for bindless samplers constexpr explicit Sampler(u32 index, u32 offset, u32 buffer, Tegra::Shader::TextureType type, - bool is_array, bool is_shadow, bool is_buffer) + bool is_array, bool is_shadow, bool is_buffer, bool is_indexed) : index{index}, offset{offset}, buffer{buffer}, type{type}, is_array{is_array}, - is_shadow{is_shadow}, is_buffer{is_buffer}, is_bindless{true} {} + is_shadow{is_shadow}, is_buffer{is_buffer}, is_bindless{true}, is_indexed{is_indexed} {} constexpr u32 GetIndex() const { return index; @@ -276,16 +283,72 @@ public: return is_bindless; } + constexpr bool IsIndexed() const { + return is_indexed; + } + + constexpr u32 Size() const { + return size; + } + + constexpr void SetSize(u32 new_size) { + size = new_size; + } + private: u32 index{}; ///< Emulated index given for the this sampler. u32 offset{}; ///< Offset in the const buffer from where the sampler is being read. u32 buffer{}; ///< Buffer where the bindless sampler is being read (unused on bound samplers). + u32 size{}; ///< Size of the sampler if indexed. Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc) bool is_array{}; ///< Whether the texture is being sampled as an array texture or not. bool is_shadow{}; ///< Whether the texture is being sampled as a depth texture or not. bool is_buffer{}; ///< Whether the texture is a texture buffer without sampler. bool is_bindless{}; ///< Whether this sampler belongs to a bindless texture or not. + bool is_indexed{}; ///< Whether this sampler is an indexed array of textures. +}; + +/// Represents a tracked bindless sampler into a direct const buffer +class ArraySamplerNode final { +public: + explicit ArraySamplerNode(u32 index, u32 base_offset, u32 bindless_var) + : index{index}, base_offset{base_offset}, bindless_var{bindless_var} {} + + constexpr u32 GetIndex() const { + return index; + } + + constexpr u32 GetBaseOffset() const { + return base_offset; + } + + constexpr u32 GetIndexVar() const { + return bindless_var; + } + +private: + u32 index; + u32 base_offset; + u32 bindless_var; +}; + +/// Represents a tracked bindless sampler into a direct const buffer +class BindlessSamplerNode final { +public: + explicit BindlessSamplerNode(u32 index, u32 offset) : index{index}, offset{offset} {} + + constexpr u32 GetIndex() const { + return index; + } + + constexpr u32 GetOffset() const { + return offset; + } + +private: + u32 index; + u32 offset; }; class Image final { @@ -380,8 +443,9 @@ struct MetaTexture { std::vector<Node> derivates; Node bias; Node lod; - Node component{}; + Node component; u32 element{}; + Node index; }; struct MetaImage { @@ -488,6 +552,19 @@ private: Tegra::Shader::Register index{}; }; +/// A custom variable +class CustomVarNode final { +public: + explicit constexpr CustomVarNode(u32 index) : index{index} {} + + constexpr u32 GetIndex() const { + return index; + } + +private: + u32 index{}; +}; + /// A 32-bits value that represents an immediate value class ImmediateNode final { public: diff --git a/src/video_core/shader/node_helper.h b/src/video_core/shader/node_helper.h index 0c2aa749b..11231bbea 100644 --- a/src/video_core/shader/node_helper.h +++ b/src/video_core/shader/node_helper.h @@ -45,6 +45,12 @@ Node MakeNode(Args&&... args) { return std::make_shared<NodeData>(T(std::forward<Args>(args)...)); } +template <typename T, typename... Args> +TrackSampler MakeTrackSampler(Args&&... args) { + static_assert(std::is_convertible_v<T, TrackSamplerData>); + return std::make_shared<TrackSamplerData>(T(std::forward<Args>(args)...)); +} + template <typename... Args> Node Operation(OperationCode code, Args&&... args) { if constexpr (sizeof...(args) == 0) { diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index 31eecb3f4..3a5d280a9 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -27,6 +27,7 @@ ShaderIR::ShaderIR(const ProgramCode& program_code, u32 main_offset, CompilerSet ConstBufferLocker& locker) : program_code{program_code}, main_offset{main_offset}, settings{settings}, locker{locker} { Decode(); + PostDecode(); } ShaderIR::~ShaderIR() = default; @@ -38,6 +39,10 @@ Node ShaderIR::GetRegister(Register reg) { return MakeNode<GprNode>(reg); } +Node ShaderIR::GetCustomVariable(u32 id) { + return MakeNode<CustomVarNode>(id); +} + Node ShaderIR::GetImmediate19(Instruction instr) { return Immediate(instr.alu.GetImm20_19()); } @@ -452,4 +457,8 @@ std::size_t ShaderIR::DeclareAmend(Node new_amend) { return id; } +u32 ShaderIR::NewCustomVariable() { + return num_custom_variables++; +} + } // namespace VideoCommon::Shader diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index ba1db4c11..b0851c3be 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -180,6 +180,10 @@ public: return amend_code[index]; } + u32 GetNumCustomVariables() const { + return num_custom_variables; + } + private: friend class ASTDecoder; @@ -191,6 +195,7 @@ private: }; void Decode(); + void PostDecode(); NodeBlock DecodeRange(u32 begin, u32 end); void DecodeRangeInner(NodeBlock& bb, u32 begin, u32 end); @@ -235,6 +240,8 @@ private: /// Generates a node for a passed register. Node GetRegister(Tegra::Shader::Register reg); + /// Generates a node for a custom variable + Node GetCustomVariable(u32 id); /// Generates a node representing a 19-bit immediate value Node GetImmediate19(Tegra::Shader::Instruction instr); /// Generates a node representing a 32-bit immediate value @@ -321,7 +328,7 @@ private: std::optional<SamplerInfo> sampler_info = std::nullopt); /// Accesses a texture sampler for a bindless texture. - const Sampler* GetBindlessSampler(Tegra::Shader::Register reg, + const Sampler* GetBindlessSampler(Tegra::Shader::Register reg, Node& index_var, std::optional<SamplerInfo> sampler_info = std::nullopt); /// Accesses an image. @@ -387,6 +394,9 @@ private: std::tuple<Node, u32, u32> TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const; + std::tuple<Node, TrackSampler> TrackBindlessSampler(Node tracked, const NodeBlock& code, + s64 cursor); + std::optional<u32> TrackImmediate(Node tracked, const NodeBlock& code, s64 cursor) const; std::pair<Node, s64> TrackRegister(const GprNode* tracked, const NodeBlock& code, @@ -399,6 +409,8 @@ private: /// Register new amending code and obtain the reference id. std::size_t DeclareAmend(Node new_amend); + u32 NewCustomVariable(); + const ProgramCode& program_code; const u32 main_offset; const CompilerSettings settings; @@ -414,6 +426,7 @@ private: NodeBlock global_code; ASTManager program_manager{true, true}; std::vector<Node> amend_code; + u32 num_custom_variables{}; std::set<u32> used_registers; std::set<Tegra::Shader::Pred> used_predicates; @@ -431,6 +444,7 @@ private: bool uses_instance_id{}; bool uses_vertex_id{}; bool uses_warps{}; + bool uses_indexed_samplers{}; Tegra::Shader::Header header; }; diff --git a/src/video_core/shader/track.cpp b/src/video_core/shader/track.cpp index 165c79330..face8c943 100644 --- a/src/video_core/shader/track.cpp +++ b/src/video_core/shader/track.cpp @@ -8,6 +8,7 @@ #include "common/common_types.h" #include "video_core/shader/node.h" +#include "video_core/shader/node_helper.h" #include "video_core/shader/shader_ir.h" namespace VideoCommon::Shader { @@ -35,8 +36,113 @@ std::pair<Node, s64> FindOperation(const NodeBlock& code, s64 cursor, } return {}; } + +std::optional<std::pair<Node, Node>> DecoupleIndirectRead(const OperationNode& operation) { + if (operation.GetCode() != OperationCode::UAdd) { + return std::nullopt; + } + Node gpr; + Node offset; + ASSERT(operation.GetOperandsCount() == 2); + for (std::size_t i = 0; i < operation.GetOperandsCount(); i++) { + Node operand = operation[i]; + if (std::holds_alternative<ImmediateNode>(*operand)) { + offset = operation[i]; + } else if (std::holds_alternative<GprNode>(*operand)) { + gpr = operation[i]; + } + } + if (offset && gpr) { + return std::make_pair(gpr, offset); + } + return std::nullopt; +} + +bool AmendNodeCv(std::size_t amend_index, Node node) { + if (const auto operation = std::get_if<OperationNode>(&*node)) { + operation->SetAmendIndex(amend_index); + return true; + } else if (const auto conditional = std::get_if<ConditionalNode>(&*node)) { + conditional->SetAmendIndex(amend_index); + return true; + } + return false; +} + } // Anonymous namespace +std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, const NodeBlock& code, + s64 cursor) { + if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) { + // Constant buffer found, test if it's an immediate + const auto offset = cbuf->GetOffset(); + if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { + auto track = + MakeTrackSampler<BindlessSamplerNode>(cbuf->GetIndex(), immediate->GetValue()); + return {tracked, track}; + } else if (const auto operation = std::get_if<OperationNode>(&*offset)) { + auto bound_buffer = locker.ObtainBoundBuffer(); + if (!bound_buffer) { + return {}; + } + if (*bound_buffer != cbuf->GetIndex()) { + return {}; + } + auto pair = DecoupleIndirectRead(*operation); + if (!pair) { + return {}; + } + auto [gpr, base_offset] = *pair; + const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset); + auto gpu_driver = locker.AccessGuestDriverProfile(); + if (gpu_driver == nullptr) { + return {}; + } + const u32 bindless_cv = NewCustomVariable(); + const Node op = Operation(OperationCode::UDiv, NO_PRECISE, gpr, + Immediate(gpu_driver->GetTextureHandlerSize())); + + const Node cv_node = GetCustomVariable(bindless_cv); + Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op)); + const std::size_t amend_index = DeclareAmend(amend_op); + AmendNodeCv(amend_index, code[cursor]); + // TODO Implement Bindless Index custom variable + auto track = MakeTrackSampler<ArraySamplerNode>(cbuf->GetIndex(), + offset_inm->GetValue(), bindless_cv); + return {tracked, track}; + } + return {}; + } + if (const auto gpr = std::get_if<GprNode>(&*tracked)) { + if (gpr->GetIndex() == Tegra::Shader::Register::ZeroIndex) { + return {}; + } + // Reduce the cursor in one to avoid infinite loops when the instruction sets the same + // register that it uses as operand + const auto [source, new_cursor] = TrackRegister(gpr, code, cursor - 1); + if (!source) { + return {}; + } + return TrackBindlessSampler(source, code, new_cursor); + } + if (const auto operation = std::get_if<OperationNode>(&*tracked)) { + for (std::size_t i = operation->GetOperandsCount(); i > 0; --i) { + if (auto found = TrackBindlessSampler((*operation)[i - 1], code, cursor); + std::get<0>(found)) { + // Cbuf found in operand. + return found; + } + } + return {}; + } + if (const auto conditional = std::get_if<ConditionalNode>(&*tracked)) { + const auto& conditional_code = conditional->GetCode(); + return TrackBindlessSampler(tracked, conditional_code, + static_cast<s64>(conditional_code.size())); + } + return {}; +} + std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& code, s64 cursor) const { if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) { diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index 8e947394c..a5f81a8a0 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -3,19 +3,32 @@ // Refer to the license.txt file included. #include <memory> +#include "common/logging/log.h" #include "core/core.h" #include "core/settings.h" #include "video_core/gpu_asynch.h" #include "video_core/gpu_synch.h" #include "video_core/renderer_base.h" #include "video_core/renderer_opengl/renderer_opengl.h" +#ifdef HAS_VULKAN +#include "video_core/renderer_vulkan/renderer_vulkan.h" +#endif #include "video_core/video_core.h" namespace VideoCore { std::unique_ptr<RendererBase> CreateRenderer(Core::Frontend::EmuWindow& emu_window, Core::System& system) { - return std::make_unique<OpenGL::RendererOpenGL>(emu_window, system); + switch (Settings::values.renderer_backend) { + case Settings::RendererBackend::OpenGL: + return std::make_unique<OpenGL::RendererOpenGL>(emu_window, system); +#ifdef HAS_VULKAN + case Settings::RendererBackend::Vulkan: + return std::make_unique<Vulkan::RendererVulkan>(emu_window, system); +#endif + default: + return nullptr; + } } std::unique_ptr<Tegra::GPU> CreateGPU(Core::System& system) { |