From ba02d564f8a0b0167b96f247b6ad9d2bde05b6c8 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 15 Sep 2019 11:48:54 -0400 Subject: Video Core: initial Implementation of InstanceDraw Packaging --- src/video_core/engines/maxwell_3d.cpp | 74 +++++++++++++++++++++++++++++++++++ src/video_core/engines/maxwell_3d.h | 23 ++++++++++- 2 files changed, 96 insertions(+), 1 deletion(-) (limited to 'src/video_core/engines') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index c7a3c85a0..48fc1a9e1 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -92,6 +92,10 @@ void Maxwell3D::InitializeRegisterDefaults() { // Some games (like Super Mario Odyssey) assume that SRGB is enabled. regs.framebuffer_srgb = 1; + mme_inline[MAXWELL3D_REG_INDEX(draw.vertex_end_gl)] = true; + mme_inline[MAXWELL3D_REG_INDEX(draw.vertex_begin_gl)] = true; + mme_inline[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true; + mme_inline[MAXWELL3D_REG_INDEX(index_array.count)] = true; } #define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name)) @@ -416,6 +420,76 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) { } } +void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) { + const u32 method = method_call.method; + if (mme_inline[method]) { + regs.reg_array[method] = method_call.argument; + if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) || + method == MAXWELL3D_REG_INDEX(index_array.count)) { + MMMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count) + ? MMMEDrawMode::Array + : MMMEDrawMode::Indexed; + u32 count = method_call.argument; + while (true) { + if (mme_draw.current_mode == MMMEDrawMode::Undefined) { + mme_draw.current_mode = expected_mode; + mme_draw.current_count = count; + mme_draw.instance_count = 1; + break; + } else { + if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count) { + mme_draw.instance_count++; + break; + } else { + FlushMMEInlineDraw(); + } + } + } + } + } else { + if (mme_draw.current_mode != MMMEDrawMode::Undefined) { + FlushMMEInlineDraw(); + } + CallMethod(method_call); + } +} + +void Maxwell3D::FlushMMEInlineDraw() { + LOG_DEBUG(HW_GPU, "called, topology={}, count={}", static_cast(regs.draw.topology.Value()), + regs.vertex_buffer.count); + ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?"); + + auto debug_context = system.GetGPUDebugContext(); + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::IncomingPrimitiveBatch, nullptr); + } + + // Both instance configuration registers can not be set at the same time. + ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont, + "Illegal combination of instancing parameters"); + + const bool is_indexed = mme_draw.current_mode == MMMEDrawMode::Indexed; + rasterizer.AccelerateDrawMultiBatch(is_indexed); + + if (debug_context) { + debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); + } + + // TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if + // the game is trying to draw indexed or direct mode. This needs to be verified on HW still - + // it's possible that it is incorrect and that there is some other register used to specify the + // drawing mode. + if (is_indexed) { + regs.index_array.count = 0; + } else { + regs.vertex_buffer.count = 0; + } + mme_draw.current_mode = MMMEDrawMode::Undefined; + mme_draw.current_count = 0; + mme_draw.instance_count = 0; +} + void Maxwell3D::ProcessMacroUpload(u32 data) { ASSERT_MSG(regs.macros.upload_address < macro_memory.size(), "upload_address exceeded macro_memory size!"); diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index e5ec90717..1547d930e 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -811,8 +811,9 @@ public: INSERT_PADDING_WORDS(0x21); u32 vb_element_base; + u32 vb_base_instance; - INSERT_PADDING_WORDS(0x36); + INSERT_PADDING_WORDS(0x35); union { BitField<0, 1, u32> c0; @@ -1238,6 +1239,11 @@ public: /// Write the value to the register identified by method. void CallMethod(const GPU::MethodCall& method_call); + /// Write the value to the register identified by method. + void CallMethodFromMME(const GPU::MethodCall& method_call); + + void FlushMMEInlineDraw(); + /// Given a Texture Handle, returns the TSC and TIC entries. Texture::FullTextureInfo GetTextureInfo(const Texture::TextureHandle tex_handle, std::size_t offset) const; @@ -1263,6 +1269,18 @@ public: return execute_on; } + enum class MMMEDrawMode : u32 { + Undefined, + Array, + Indexed, + }; + + struct MMEDrawState { + MMMEDrawMode current_mode{MMMEDrawMode::Undefined}; + u32 current_count; + u32 instance_count; + } mme_draw; + private: void InitializeRegisterDefaults(); @@ -1275,6 +1293,8 @@ private: /// Start offsets of each macro in macro_memory std::array macro_positions = {}; + std::array mme_inline{}; + /// Memory for macro code MacroMemory macro_memory; @@ -1402,6 +1422,7 @@ ASSERT_REG_POSITION(stencil_front_mask, 0x4E7); ASSERT_REG_POSITION(frag_color_clamp, 0x4EA); ASSERT_REG_POSITION(screen_y_control, 0x4EB); ASSERT_REG_POSITION(vb_element_base, 0x50D); +ASSERT_REG_POSITION(vb_base_instance, 0x50E); ASSERT_REG_POSITION(clip_distance_enabled, 0x544); ASSERT_REG_POSITION(point_size, 0x546); ASSERT_REG_POSITION(zeta_enable, 0x54E); -- cgit v1.2.3 From 7606da5611b5626790e99b4387e033eaea20c2cb Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 15 Sep 2019 14:25:07 -0400 Subject: VideoCore: Corrections to the MME Inliner and removal of hacky instance management. --- src/video_core/engines/maxwell_3d.cpp | 35 +++++++++++++++----- src/video_core/engines/maxwell_3d.h | 7 ++-- src/video_core/renderer_opengl/gl_rasterizer.cpp | 38 ++++++++++------------ .../renderer_opengl/gl_shader_decompiler.cpp | 10 +++++- src/video_core/shader/shader_ir.cpp | 12 +++++++ src/video_core/shader/shader_ir.h | 10 ++++++ 6 files changed, 81 insertions(+), 31 deletions(-) (limited to 'src/video_core/engines') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 48fc1a9e1..d1f63a5eb 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -260,6 +260,9 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3 // Execute the current macro. macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters); + if (mme_draw.current_mode != MMMEDrawMode::Undefined) { + FlushMMEInlineDraw(); + } } void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) { @@ -426,25 +429,37 @@ void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) { regs.reg_array[method] = method_call.argument; if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) || method == MAXWELL3D_REG_INDEX(index_array.count)) { - MMMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count) - ? MMMEDrawMode::Array - : MMMEDrawMode::Indexed; - u32 count = method_call.argument; + const MMMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count) + ? MMMEDrawMode::Array + : MMMEDrawMode::Indexed; + const u32 count = method_call.argument; while (true) { if (mme_draw.current_mode == MMMEDrawMode::Undefined) { - mme_draw.current_mode = expected_mode; - mme_draw.current_count = count; - mme_draw.instance_count = 1; + if (mme_draw.gl_begin_consume) { + mme_draw.current_mode = expected_mode; + mme_draw.current_count = count; + mme_draw.instance_count = 1; + mme_draw.gl_begin_consume = false; + mme_draw.gl_end_count = 0; + } break; } else { - if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count) { + if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count && + mme_draw.instance_mode && mme_draw.gl_begin_consume) { mme_draw.instance_count++; + mme_draw.gl_begin_consume = false; break; } else { FlushMMEInlineDraw(); } } } + } else if (method == MAXWELL3D_REG_INDEX(draw.vertex_begin_gl)) { + mme_draw.instance_mode = + (regs.draw.instance_next != 0) || (regs.draw.instance_cont != 0); + mme_draw.gl_begin_consume = true; + } else { + mme_draw.gl_end_count++; } } else { if (mme_draw.current_mode != MMMEDrawMode::Undefined) { @@ -458,6 +473,7 @@ void Maxwell3D::FlushMMEInlineDraw() { LOG_DEBUG(HW_GPU, "called, topology={}, count={}", static_cast(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); auto debug_context = system.GetGPUDebugContext(); @@ -488,6 +504,9 @@ void Maxwell3D::FlushMMEInlineDraw() { mme_draw.current_mode = MMMEDrawMode::Undefined; mme_draw.current_count = 0; mme_draw.instance_count = 0; + mme_draw.instance_mode = false; + mme_draw.gl_begin_consume = false; + mme_draw.gl_end_count = 0; } void Maxwell3D::ProcessMacroUpload(u32 data) { diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 1547d930e..8fd3ec85c 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1277,8 +1277,11 @@ public: struct MMEDrawState { MMMEDrawMode current_mode{MMMEDrawMode::Undefined}; - u32 current_count; - u32 instance_count; + u32 current_count{}; + u32 instance_count{}; + bool instance_mode{}; + bool gl_begin_consume{}; + u32 gl_end_count{}; } mme_draw; private: diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 5df7f3f56..f71a22738 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -788,13 +788,13 @@ void RasterizerOpenGL::DrawArrays() { DrawPrelude(); auto& maxwell3d = system.GPU().Maxwell3D(); - auto& regs = maxwell3d.regs; - auto current_instance = maxwell3d.state.current_instance; - auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); + const auto& regs = maxwell3d.regs; + const auto current_instance = maxwell3d.state.current_instance; + const auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); if (accelerate_draw == AccelDraw::Indexed) { - auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); - auto count = regs.index_array.count; - auto base_vertex = static_cast(regs.vb_element_base); + const auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); + const auto count = regs.index_array.count; + const auto base_vertex = static_cast(regs.vb_element_base); const auto index_buffer_ptr = reinterpret_cast(index_buffer_offset); if (current_instance > 0) { glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format, @@ -805,8 +805,8 @@ void RasterizerOpenGL::DrawArrays() { base_vertex); } } else { - auto count = regs.vertex_buffer.count; - auto vertex_first = regs.vertex_buffer.first; + const auto count = regs.vertex_buffer.count; + const auto vertex_first = regs.vertex_buffer.first; if (current_instance > 0) { glDrawArraysInstancedBaseInstance(primitive_mode, vertex_first, count, 1, current_instance); @@ -819,21 +819,19 @@ void RasterizerOpenGL::DrawArrays() { maxwell3d.dirty.memory_general = false; } -#pragma optimize("", off) - void RasterizerOpenGL::DrawMultiArrays() { DrawPrelude(); auto& maxwell3d = system.GPU().Maxwell3D(); - auto& regs = maxwell3d.regs; - auto& draw_setup = maxwell3d.mme_draw; - auto num_instances = draw_setup.instance_count; - auto base_instance = static_cast(regs.vb_base_instance); - auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); + const auto& regs = maxwell3d.regs; + const auto& draw_setup = maxwell3d.mme_draw; + const auto num_instances = draw_setup.instance_count; + const auto base_instance = static_cast(regs.vb_base_instance); + const auto primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology); if (draw_setup.current_mode == Tegra::Engines::Maxwell3D::MMMEDrawMode::Indexed) { - auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); - auto count = regs.index_array.count; - auto base_vertex = static_cast(regs.vb_element_base); + const auto index_format = MaxwellToGL::IndexFormat(regs.index_array.format); + const auto count = regs.index_array.count; + const auto base_vertex = static_cast(regs.vb_element_base); const auto index_buffer_ptr = reinterpret_cast(index_buffer_offset); if (num_instances > 1) { glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format, @@ -844,8 +842,8 @@ void RasterizerOpenGL::DrawMultiArrays() { base_vertex); } } else { - auto count = regs.vertex_buffer.count; - auto vertex_first = regs.vertex_buffer.first; + const auto count = regs.vertex_buffer.count; + const auto vertex_first = regs.vertex_buffer.first; if (num_instances > 1) { glDrawArraysInstancedBaseInstance(primitive_mode, vertex_first, count, num_instances, base_instance); diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 62e32697e..f7e86ab26 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -462,6 +462,14 @@ private: code.AddLine("float gl_PointSize;"); } + if (ir.UsesInstanceId()) { + code.AddLine("int gl_InstanceID;"); + } + + if (ir.UsesVertexId()) { + code.AddLine("int gl_VertexID;"); + } + --code.scope; code.AddLine("}};"); code.AddNewLine(); @@ -964,7 +972,7 @@ private: switch (element) { case 2: // Config pack's first value is instance_id. - return {"gl_InstanceID", Type::Uint}; + return {"gl_InstanceID", Type::Int}; case 3: return {"gl_VertexID", Type::Int}; } diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index bbbab0bca..2c357f310 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -114,6 +114,18 @@ Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buff break; } } + if (index == Attribute::Index::TessCoordInstanceIDVertexID) { + switch (element) { + case 2: + uses_instance_id = true; + break; + case 3: + uses_vertex_id = true; + break; + default: + break; + } + } if (index == Attribute::Index::ClipDistances0123 || index == Attribute::Index::ClipDistances4567) { const auto clip_index = diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 6aed9bb84..2f03d83ba 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -124,6 +124,14 @@ public: return uses_point_size; } + bool UsesInstanceId() const { + return uses_instance_id; + } + + bool UsesVertexId() const { + return uses_vertex_id; + } + bool HasPhysicalAttributes() const { return uses_physical_attributes; } @@ -373,6 +381,8 @@ private: bool uses_viewport_index{}; bool uses_point_size{}; bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes + bool uses_instance_id{}; + bool uses_vertex_id{}; Tegra::Shader::Header header; }; -- cgit v1.2.3 From 7761e44d186deff278e51814bdec0c25e8e1a09c Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Thu, 19 Sep 2019 11:41:07 -0400 Subject: Rasterizer: Refactor and simplify DrawBatch Interface. --- src/video_core/engines/maxwell_3d.cpp | 4 ++-- src/video_core/rasterizer_interface.h | 12 ++-------- src/video_core/renderer_opengl/gl_rasterizer.cpp | 29 ++++++++---------------- src/video_core/renderer_opengl/gl_rasterizer.h | 6 ++--- 4 files changed, 16 insertions(+), 35 deletions(-) (limited to 'src/video_core/engines') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index d1f63a5eb..b8f071bc0 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -486,7 +486,7 @@ void Maxwell3D::FlushMMEInlineDraw() { "Illegal combination of instancing parameters"); const bool is_indexed = mme_draw.current_mode == MMMEDrawMode::Indexed; - rasterizer.AccelerateDrawMultiBatch(is_indexed); + rasterizer.DrawMultiBatch(is_indexed); if (debug_context) { debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); @@ -657,7 +657,7 @@ void Maxwell3D::DrawArrays() { } const bool is_indexed{regs.index_array.count && !regs.vertex_buffer.count}; - rasterizer.AccelerateDrawBatch(is_indexed); + rasterizer.DrawBatch(is_indexed); if (debug_context) { debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index fe0d82255..deadd70ec 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -29,10 +29,10 @@ public: virtual ~RasterizerInterface() {} /// Draw the current batch of vertex arrays - virtual void DrawArrays() = 0; + virtual bool DrawBatch(bool is_indexed) = 0; /// Draw the current batch of multiple instasnces of vertex arrays - virtual void DrawMultiArrays() = 0; + virtual bool DrawMultiBatch(bool is_indexed) = 0; /// Clear the current framebuffer virtual void Clear() = 0; @@ -72,14 +72,6 @@ public: return false; } - virtual bool AccelerateDrawBatch(bool is_indexed) { - return false; - } - - virtual bool AccelerateDrawMultiBatch(bool is_indexed) { - return false; - } - /// Increase/decrease the number of object in pages touching the specified region virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {} diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index b86aa49f3..b5c55482f 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -342,18 +342,6 @@ std::size_t RasterizerOpenGL::CalculateIndexBufferSize() const { static_cast(regs.index_array.FormatSizeInBytes()); } -bool RasterizerOpenGL::AccelerateDrawBatch(bool is_indexed) { - accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; - DrawArrays(); - return true; -} - -bool RasterizerOpenGL::AccelerateDrawMultiBatch(bool is_indexed) { - accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; - DrawMultiArrays(); - return true; -} - template static constexpr auto RangeFromInterval(Map& map, const Interval& interval) { return boost::make_iterator_range(map.equal_range(interval)); @@ -764,14 +752,15 @@ struct DrawParams { } }; -void RasterizerOpenGL::DrawArrays() { +bool RasterizerOpenGL::DrawBatch(bool is_indexed) { + accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; 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 = accelerate_draw == AccelDraw::Indexed; + draw_call.is_indexed = is_indexed; draw_call.num_instances = static_cast(1); draw_call.base_instance = static_cast(current_instance); draw_call.is_instanced = current_instance > 0; @@ -787,19 +776,20 @@ void RasterizerOpenGL::DrawArrays() { } draw_call.DispatchDraw(); - accelerate_draw = AccelDraw::Disabled; maxwell3d.dirty.memory_general = false; + accelerate_draw = AccelDraw::Disabled; + return true; } -void RasterizerOpenGL::DrawMultiArrays() { +bool RasterizerOpenGL::DrawMultiBatch(bool is_indexed) { + accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays; 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 = - draw_setup.current_mode == Tegra::Engines::Maxwell3D::MMMEDrawMode::Indexed; + draw_call.is_indexed = is_indexed; draw_call.num_instances = static_cast(draw_setup.instance_count); draw_call.base_instance = static_cast(regs.vb_base_instance); draw_call.is_instanced = draw_setup.instance_count > 1; @@ -815,8 +805,9 @@ void RasterizerOpenGL::DrawMultiArrays() { } draw_call.DispatchDraw(); - accelerate_draw = AccelDraw::Disabled; maxwell3d.dirty.memory_general = false; + accelerate_draw = AccelDraw::Disabled; + return true; } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index b4e21b9de..682f0becc 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -57,8 +57,8 @@ public: ScreenInfo& info); ~RasterizerOpenGL() override; - void DrawArrays() override; - void DrawMultiArrays() override; + bool DrawBatch(bool is_indexed) override; + bool DrawMultiBatch(bool is_indexed) override; void Clear() override; void DispatchCompute(GPUVAddr code_addr) override; void FlushAll() override; @@ -72,8 +72,6 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; - bool AccelerateDrawBatch(bool is_indexed) override; - bool AccelerateDrawMultiBatch(bool is_indexed) override; void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override; void LoadDiskResources(const std::atomic_bool& stop_loading, const VideoCore::DiskResourceLoadCallback& callback) override; -- cgit v1.2.3 From 68f5aff64f231e77fed726b91619061e308b7b2d Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 22 Sep 2019 07:23:13 -0400 Subject: Maxwell3D: Corrections and refactors to MME instance refactor --- src/video_core/engines/maxwell_3d.cpp | 69 +++++++++++++----------- src/video_core/engines/maxwell_3d.h | 7 ++- src/video_core/rasterizer_interface.h | 2 +- src/video_core/renderer_opengl/gl_rasterizer.cpp | 12 +---- 4 files changed, 46 insertions(+), 44 deletions(-) (limited to 'src/video_core/engines') diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index b8f071bc0..e9c15beff 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -260,7 +260,7 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3 // Execute the current macro. macro_interpreter.Execute(macro_positions[entry], num_parameters, parameters); - if (mme_draw.current_mode != MMMEDrawMode::Undefined) { + if (mme_draw.current_mode != MMEDrawMode::Undefined) { FlushMMEInlineDraw(); } } @@ -423,37 +423,40 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) { } } +void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) { + if (mme_draw.current_mode == MMEDrawMode::Undefined) { + if (mme_draw.gl_begin_consume) { + mme_draw.current_mode = expected_mode; + mme_draw.current_count = count; + mme_draw.instance_count = 1; + mme_draw.gl_begin_consume = false; + mme_draw.gl_end_count = 0; + } + return; + } else { + if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count && + mme_draw.instance_mode && mme_draw.gl_begin_consume) { + mme_draw.instance_count++; + mme_draw.gl_begin_consume = false; + return; + } else { + FlushMMEInlineDraw(); + } + } + // Tail call in case it needs to retry. + StepInstance(expected_mode, count); +} + void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) { const u32 method = method_call.method; if (mme_inline[method]) { regs.reg_array[method] = method_call.argument; if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) || method == MAXWELL3D_REG_INDEX(index_array.count)) { - const MMMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count) - ? MMMEDrawMode::Array - : MMMEDrawMode::Indexed; - const u32 count = method_call.argument; - while (true) { - if (mme_draw.current_mode == MMMEDrawMode::Undefined) { - if (mme_draw.gl_begin_consume) { - mme_draw.current_mode = expected_mode; - mme_draw.current_count = count; - mme_draw.instance_count = 1; - mme_draw.gl_begin_consume = false; - mme_draw.gl_end_count = 0; - } - break; - } else { - if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count && - mme_draw.instance_mode && mme_draw.gl_begin_consume) { - mme_draw.instance_count++; - mme_draw.gl_begin_consume = false; - break; - } else { - FlushMMEInlineDraw(); - } - } - } + const MMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count) + ? MMEDrawMode::Array + : MMEDrawMode::Indexed; + StepInstance(expected_mode, method_call.argument); } else if (method == MAXWELL3D_REG_INDEX(draw.vertex_begin_gl)) { mme_draw.instance_mode = (regs.draw.instance_next != 0) || (regs.draw.instance_cont != 0); @@ -462,7 +465,7 @@ void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) { mme_draw.gl_end_count++; } } else { - if (mme_draw.current_mode != MMMEDrawMode::Undefined) { + if (mme_draw.current_mode != MMEDrawMode::Undefined) { FlushMMEInlineDraw(); } CallMethod(method_call); @@ -485,8 +488,10 @@ void Maxwell3D::FlushMMEInlineDraw() { ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont, "Illegal combination of instancing parameters"); - const bool is_indexed = mme_draw.current_mode == MMMEDrawMode::Indexed; - rasterizer.DrawMultiBatch(is_indexed); + const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed; + if (ShouldExecute()) { + rasterizer.DrawMultiBatch(is_indexed); + } if (debug_context) { debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); @@ -501,7 +506,7 @@ void Maxwell3D::FlushMMEInlineDraw() { } else { regs.vertex_buffer.count = 0; } - mme_draw.current_mode = MMMEDrawMode::Undefined; + mme_draw.current_mode = MMEDrawMode::Undefined; mme_draw.current_count = 0; mme_draw.instance_count = 0; mme_draw.instance_mode = false; @@ -657,7 +662,9 @@ void Maxwell3D::DrawArrays() { } const bool is_indexed{regs.index_array.count && !regs.vertex_buffer.count}; - rasterizer.DrawBatch(is_indexed); + if (ShouldExecute()) { + rasterizer.DrawBatch(is_indexed); + } if (debug_context) { debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr); diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 8fd3ec85c..4c97759ed 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1269,14 +1269,14 @@ public: return execute_on; } - enum class MMMEDrawMode : u32 { + enum class MMEDrawMode : u32 { Undefined, Array, Indexed, }; struct MMEDrawState { - MMMEDrawMode current_mode{MMMEDrawMode::Undefined}; + MMEDrawMode current_mode{MMEDrawMode::Undefined}; u32 current_count{}; u32 instance_count{}; bool instance_mode{}; @@ -1369,6 +1369,9 @@ private: /// Handles a write to the VERTEX_END_GL register, triggering a draw. void DrawArrays(); + + // Handles a instance drawcall from MME + void StepInstance(MMEDrawMode expected_mode, u32 count); }; #define ASSERT_REG_POSITION(field_name, position) \ diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index deadd70ec..5b0eca9e2 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -31,7 +31,7 @@ public: /// Draw the current batch of vertex arrays virtual bool DrawBatch(bool is_indexed) = 0; - /// Draw the current batch of multiple instasnces of vertex arrays + /// Draw the current batch of multiple instances of vertex arrays virtual bool DrawMultiBatch(bool is_indexed) = 0; /// Clear the current framebuffer diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index a6fe7dd71..246b892c5 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -749,13 +749,9 @@ bool RasterizerOpenGL::DrawBatch(bool is_indexed) { MICROPROFILE_SCOPE(OpenGL_Drawing); - auto& maxwell3d = system.GPU().Maxwell3D(); - if (!maxwell3d.ShouldExecute()) { - return false; - } - DrawPrelude(); + auto& maxwell3d = system.GPU().Maxwell3D(); const auto& regs = maxwell3d.regs; const auto current_instance = maxwell3d.state.current_instance; DrawParams draw_call{}; @@ -785,13 +781,9 @@ bool RasterizerOpenGL::DrawMultiBatch(bool is_indexed) { MICROPROFILE_SCOPE(OpenGL_Drawing); - auto& maxwell3d = system.GPU().Maxwell3D(); - if (!maxwell3d.ShouldExecute()) { - return false; - } - DrawPrelude(); + auto& maxwell3d = system.GPU().Maxwell3D(); const auto& regs = maxwell3d.regs; const auto& draw_setup = maxwell3d.mme_draw; DrawParams draw_call{}; -- cgit v1.2.3