From 9ea8cffe35630f208024b4f547e9c261703edd06 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sun, 12 Jul 2020 05:03:05 -0300 Subject: maxwell_3d: Move code to separate functions Deduplicate some code and put it in separate functions so it's easier to understand and profile. --- src/video_core/engines/maxwell_3d.cpp | 267 +++++++++++++++------------------- 1 file changed, 116 insertions(+), 151 deletions(-) (limited to 'src/video_core/engines/maxwell_3d.cpp') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 57ebc785f..f7fa5fea7 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -124,6 +124,114 @@ void Maxwell3D::InitializeRegisterDefaults() { mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true; } +void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool is_last_call) { + if (executing_macro == 0) { + // A macro call must begin by writing the macro method's register, not its argument. + ASSERT_MSG((method % 2) == 0, + "Can't start macro execution by writing to the ARGS register"); + executing_macro = method; + } + + for (std::size_t i = 0; i < amount; i++) { + macro_params.push_back(base_start[i]); + } + + // Call the macro when there are no more parameters in the command buffer + if (is_last_call) { + CallMacroMethod(executing_macro, macro_params); + macro_params.clear(); + } +} + +u32 Maxwell3D::ProcessShadowRam(u32 method, u32 argument) { + // Keep track of the register value in shadow_state when requested. + const auto control = shadow_state.shadow_ram_control; + if (control == Regs::ShadowRamControl::Track || + control == Regs::ShadowRamControl::TrackWithFilter) { + shadow_state.reg_array[method] = argument; + return argument; + } + if (control == Regs::ShadowRamControl::Replay) { + return shadow_state.reg_array[method]; + } + return argument; +} + +void Maxwell3D::ProcessDirtyRegisters(u32 method, u32 argument) { + if (regs.reg_array[method] == argument) { + return; + } + regs.reg_array[method] = argument; + + for (const auto& table : dirty.tables) { + dirty.flags[table[method]] = true; + } +} + +void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argument, + bool is_last_call) { + switch (method) { + case MAXWELL3D_REG_INDEX(wait_for_idle): + return rasterizer->WaitForIdle(); + case MAXWELL3D_REG_INDEX(shadow_ram_control): + shadow_state.shadow_ram_control = static_cast(nonshadow_argument); + return; + case MAXWELL3D_REG_INDEX(macros.data): + return macro_engine->AddCode(regs.macros.upload_address, argument); + case MAXWELL3D_REG_INDEX(macros.bind): + return ProcessMacroBind(argument); + case MAXWELL3D_REG_INDEX(firmware[4]): + return ProcessFirmwareCall4(); + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): + return StartCBData(method); + case MAXWELL3D_REG_INDEX(cb_bind[0]): + return ProcessCBBind(0); + case MAXWELL3D_REG_INDEX(cb_bind[1]): + return ProcessCBBind(1); + case MAXWELL3D_REG_INDEX(cb_bind[2]): + return ProcessCBBind(2); + case MAXWELL3D_REG_INDEX(cb_bind[3]): + return ProcessCBBind(3); + case MAXWELL3D_REG_INDEX(cb_bind[4]): + return ProcessCBBind(4); + case MAXWELL3D_REG_INDEX(draw.vertex_end_gl): + return DrawArrays(); + case MAXWELL3D_REG_INDEX(clear_buffers): + return ProcessClearBuffers(); + case MAXWELL3D_REG_INDEX(query.query_get): + return ProcessQueryGet(); + case MAXWELL3D_REG_INDEX(condition.mode): + return ProcessQueryCondition(); + case MAXWELL3D_REG_INDEX(counter_reset): + return ProcessCounterReset(); + case MAXWELL3D_REG_INDEX(sync_info): + return ProcessSyncPoint(); + case MAXWELL3D_REG_INDEX(exec_upload): + return upload_state.ProcessExec(regs.exec_upload.linear != 0); + case MAXWELL3D_REG_INDEX(data_upload): + upload_state.ProcessData(argument, is_last_call); + if (is_last_call) { + OnMemoryWrite(); + } + return; + } +} + void Maxwell3D::CallMacroMethod(u32 method, const std::vector& parameters) { // Reset the current macro. executing_macro = 0; @@ -157,142 +265,16 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) { // Methods after 0xE00 are special, they're actually triggers for some microcode that was // uploaded to the GPU during initialization. if (method >= MacroRegistersStart) { - // We're trying to execute a macro - if (executing_macro == 0) { - // A macro call must begin by writing the macro method's register, not its argument. - ASSERT_MSG((method % 2) == 0, - "Can't start macro execution by writing to the ARGS register"); - executing_macro = method; - } - - macro_params.push_back(method_argument); - - // Call the macro when there are no more parameters in the command buffer - if (is_last_call) { - CallMacroMethod(executing_macro, macro_params); - macro_params.clear(); - } + ProcessMacro(method, &method_argument, 1, is_last_call); return; } ASSERT_MSG(method < Regs::NUM_REGS, "Invalid Maxwell3D register, increase the size of the Regs structure"); - u32 arg = method_argument; - // Keep track of the register value in shadow_state when requested. - if (shadow_state.shadow_ram_control == Regs::ShadowRamControl::Track || - shadow_state.shadow_ram_control == Regs::ShadowRamControl::TrackWithFilter) { - shadow_state.reg_array[method] = arg; - } else if (shadow_state.shadow_ram_control == Regs::ShadowRamControl::Replay) { - arg = shadow_state.reg_array[method]; - } - - if (regs.reg_array[method] != arg) { - regs.reg_array[method] = arg; - - for (const auto& table : dirty.tables) { - dirty.flags[table[method]] = true; - } - } - - switch (method) { - case MAXWELL3D_REG_INDEX(wait_for_idle): { - rasterizer->WaitForIdle(); - break; - } - case MAXWELL3D_REG_INDEX(shadow_ram_control): { - shadow_state.shadow_ram_control = static_cast(method_argument); - break; - } - case MAXWELL3D_REG_INDEX(macros.data): { - macro_engine->AddCode(regs.macros.upload_address, arg); - break; - } - case MAXWELL3D_REG_INDEX(macros.bind): { - ProcessMacroBind(arg); - break; - } - case MAXWELL3D_REG_INDEX(firmware[4]): { - ProcessFirmwareCall4(); - break; - } - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): { - StartCBData(method); - break; - } - case MAXWELL3D_REG_INDEX(cb_bind[0]): { - ProcessCBBind(0); - break; - } - case MAXWELL3D_REG_INDEX(cb_bind[1]): { - ProcessCBBind(1); - break; - } - case MAXWELL3D_REG_INDEX(cb_bind[2]): { - ProcessCBBind(2); - break; - } - case MAXWELL3D_REG_INDEX(cb_bind[3]): { - ProcessCBBind(3); - break; - } - case MAXWELL3D_REG_INDEX(cb_bind[4]): { - ProcessCBBind(4); - break; - } - case MAXWELL3D_REG_INDEX(draw.vertex_end_gl): { - DrawArrays(); - break; - } - case MAXWELL3D_REG_INDEX(clear_buffers): { - ProcessClearBuffers(); - break; - } - case MAXWELL3D_REG_INDEX(query.query_get): { - ProcessQueryGet(); - break; - } - case MAXWELL3D_REG_INDEX(condition.mode): { - ProcessQueryCondition(); - break; - } - case MAXWELL3D_REG_INDEX(counter_reset): { - ProcessCounterReset(); - break; - } - case MAXWELL3D_REG_INDEX(sync_info): { - ProcessSyncPoint(); - break; - } - case MAXWELL3D_REG_INDEX(exec_upload): { - upload_state.ProcessExec(regs.exec_upload.linear != 0); - break; - } - case MAXWELL3D_REG_INDEX(data_upload): { - upload_state.ProcessData(arg, is_last_call); - if (is_last_call) { - OnMemoryWrite(); - } - break; - } - default: - break; - } + const u32 argument = ProcessShadowRam(method, method_argument); + ProcessDirtyRegisters(method, argument); + ProcessMethodCall(method, argument, method_argument, is_last_call); } void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, @@ -300,23 +282,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, // Methods after 0xE00 are special, they're actually triggers for some microcode that was // uploaded to the GPU during initialization. if (method >= MacroRegistersStart) { - // We're trying to execute a macro - if (executing_macro == 0) { - // A macro call must begin by writing the macro method's register, not its argument. - ASSERT_MSG((method % 2) == 0, - "Can't start macro execution by writing to the ARGS register"); - executing_macro = method; - } - - for (std::size_t i = 0; i < amount; i++) { - macro_params.push_back(base_start[i]); - } - - // Call the macro when there are no more parameters in the command buffer - if (amount == methods_pending) { - CallMacroMethod(executing_macro, macro_params); - macro_params.clear(); - } + ProcessMacro(method, base_start, amount, amount == methods_pending); return; } switch (method) { @@ -335,15 +301,14 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): { + case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): ProcessCBMultiData(method, base_start, amount); break; - } - default: { + default: for (std::size_t i = 0; i < amount; i++) { CallMethod(method, base_start[i], methods_pending - static_cast(i) <= 1); } - } + break; } } -- cgit v1.2.3 From 622830f4e16a8f0eabeb1b81dea5fca613402d8c Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sun, 12 Jul 2020 05:05:04 -0300 Subject: maxwell_3d: Use insert instead of loop push_back This reduces the overhead of bounds checking on each element. It won't reduce the cost of allocation because usually this vector's capacity is usually large enough to hold whatever we push to it. --- src/video_core/engines/maxwell_3d.cpp | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) (limited to 'src/video_core/engines/maxwell_3d.cpp') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index f7fa5fea7..6287df633 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -132,9 +132,7 @@ void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool executing_macro = method; } - for (std::size_t i = 0; i < amount; i++) { - macro_params.push_back(base_start[i]); - } + macro_params.insert(macro_params.end(), base_start, base_start + amount); // Call the macro when there are no more parameters in the command buffer if (is_last_call) { -- cgit v1.2.3 From 4c5f5c9bf301d3626df104dbed6fed6f115cedc8 Mon Sep 17 00:00:00 2001 From: Lioncash Date: Mon, 7 Dec 2020 00:41:47 -0500 Subject: video_core: Remove unnecessary enum class casting in logging messages fmt now automatically prints the numeric value of an enum class member by default, so we don't need to use casts any more. Reduces the line noise a bit. --- src/video_core/engines/maxwell_3d.cpp | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) (limited to 'src/video_core/engines/maxwell_3d.cpp') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 6287df633..761962ed0 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -359,7 +359,7 @@ void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) { } void Maxwell3D::FlushMMEInlineDraw() { - LOG_TRACE(HW_GPU, "called, topology={}, count={}", static_cast(regs.draw.topology.Value()), + LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(), regs.vertex_buffer.count); ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?"); ASSERT(mme_draw.instance_count == mme_draw.gl_end_count); @@ -504,8 +504,7 @@ void Maxwell3D::ProcessCounterReset() { rasterizer->ResetCounter(QueryType::SamplesPassed); break; default: - LOG_DEBUG(Render_OpenGL, "Unimplemented counter reset={}", - static_cast(regs.counter_reset)); + LOG_DEBUG(Render_OpenGL, "Unimplemented counter reset={}", regs.counter_reset); break; } } @@ -520,7 +519,7 @@ void Maxwell3D::ProcessSyncPoint() { } void Maxwell3D::DrawArrays() { - LOG_TRACE(HW_GPU, "called, topology={}, count={}", static_cast(regs.draw.topology.Value()), + LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(), regs.vertex_buffer.count); ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?"); @@ -558,12 +557,12 @@ std::optional Maxwell3D::GetQueryResult() { return 0; case Regs::QuerySelect::SamplesPassed: // Deferred. - rasterizer->Query(regs.query.QueryAddress(), VideoCore::QueryType::SamplesPassed, + rasterizer->Query(regs.query.QueryAddress(), QueryType::SamplesPassed, system.GPU().GetTicks()); return std::nullopt; default: LOG_DEBUG(HW_GPU, "Unimplemented query select type {}", - static_cast(regs.query.query_get.select.Value())); + regs.query.query_get.select.Value()); return 1; } } -- cgit v1.2.3 From 9764c13d6d2977903f407761b27d847c0056e1c4 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Wed, 30 Dec 2020 02:25:23 -0300 Subject: video_core: Rewrite the texture cache The current texture cache has several points that hurt maintainability and performance. It's easy to break unrelated parts of the cache when doing minor changes. The cache can easily forget valuable information about the cached textures by CPU writes or simply by its normal usage.The current texture cache has several points that hurt maintainability and performance. It's easy to break unrelated parts of the cache when doing minor changes. The cache can easily forget valuable information about the cached textures by CPU writes or simply by its normal usage. This commit aims to address those issues. --- src/video_core/engines/maxwell_3d.cpp | 45 ++++++++++------------------------- 1 file changed, 12 insertions(+), 33 deletions(-) (limited to 'src/video_core/engines/maxwell_3d.cpp') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 761962ed0..9be651e24 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -2,7 +2,6 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include #include #include #include "common/assert.h" @@ -227,6 +226,10 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume OnMemoryWrite(); } return; + case MAXWELL3D_REG_INDEX(fragment_barrier): + return rasterizer->FragmentBarrier(); + case MAXWELL3D_REG_INDEX(tiled_cache_barrier): + return rasterizer->TiledCacheBarrier(); } } @@ -639,7 +642,7 @@ void Maxwell3D::FinishCBData() { } Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { - const GPUVAddr tic_address_gpu{regs.tic.TICAddress() + tic_index * sizeof(Texture::TICEntry)}; + const GPUVAddr tic_address_gpu{regs.tic.Address() + tic_index * sizeof(Texture::TICEntry)}; Texture::TICEntry tic_entry; memory_manager.ReadBlockUnsafe(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry)); @@ -648,43 +651,19 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { } Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const { - const GPUVAddr tsc_address_gpu{regs.tsc.TSCAddress() + tsc_index * sizeof(Texture::TSCEntry)}; + const GPUVAddr tsc_address_gpu{regs.tsc.Address() + tsc_index * sizeof(Texture::TSCEntry)}; Texture::TSCEntry tsc_entry; memory_manager.ReadBlockUnsafe(tsc_address_gpu, &tsc_entry, sizeof(Texture::TSCEntry)); return tsc_entry; } -Texture::FullTextureInfo Maxwell3D::GetTextureInfo(Texture::TextureHandle tex_handle) const { - return Texture::FullTextureInfo{GetTICEntry(tex_handle.tic_id), GetTSCEntry(tex_handle.tsc_id)}; -} - -Texture::FullTextureInfo Maxwell3D::GetStageTexture(ShaderType stage, std::size_t offset) const { - const auto stage_index = static_cast(stage); - const auto& shader = state.shader_stages[stage_index]; - const auto& tex_info_buffer = shader.const_buffers[regs.tex_cb_index]; - ASSERT(tex_info_buffer.enabled && tex_info_buffer.address != 0); - - const GPUVAddr tex_info_address = - tex_info_buffer.address + offset * sizeof(Texture::TextureHandle); - - ASSERT(tex_info_address < tex_info_buffer.address + tex_info_buffer.size); - - const Texture::TextureHandle tex_handle{memory_manager.Read(tex_info_address)}; - - return GetTextureInfo(tex_handle); -} - u32 Maxwell3D::GetRegisterValue(u32 method) const { ASSERT_MSG(method < Regs::NUM_REGS, "Invalid Maxwell3D register"); return regs.reg_array[method]; } void Maxwell3D::ProcessClearBuffers() { - ASSERT(regs.clear_buffers.R == regs.clear_buffers.G && - regs.clear_buffers.R == regs.clear_buffers.B && - regs.clear_buffers.R == regs.clear_buffers.A); - rasterizer->Clear(); } @@ -692,9 +671,7 @@ u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offse ASSERT(stage != ShaderType::Compute); const auto& shader_stage = state.shader_stages[static_cast(stage)]; const auto& buffer = shader_stage.const_buffers[const_buffer]; - u32 result; - std::memcpy(&result, memory_manager.GetPointer(buffer.address + offset), sizeof(u32)); - return result; + return memory_manager.Read(buffer.address + offset); } SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const { @@ -712,9 +689,11 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b SamplerDescriptor Maxwell3D::AccessSampler(u32 handle) const { const Texture::TextureHandle tex_handle{handle}; - const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle); - SamplerDescriptor result = SamplerDescriptor::FromTIC(tex_info.tic); - result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value()); + const Texture::TICEntry tic = GetTICEntry(tex_handle.tic_id); + const Texture::TSCEntry tsc = GetTSCEntry(tex_handle.tsc_id); + + SamplerDescriptor result = SamplerDescriptor::FromTIC(tic); + result.is_shadow.Assign(tsc.depth_compare_enabled.Value()); return result; } -- cgit v1.2.3