From 944d2e41546aa6ca179337a645687406f1e8ec5f Mon Sep 17 00:00:00 2001 From: Feng Chen Date: Thu, 29 Dec 2022 15:51:16 +0800 Subject: [PATCH] video_core: Implement maxwell3d draw texture method --- src/video_core/engines/draw_manager.cpp | 31 +++++++++++++++++ src/video_core/engines/draw_manager.h | 20 +++++++++++ src/video_core/engines/maxwell_3d.h | 16 ++++++++- src/video_core/rasterizer_interface.h | 3 ++ .../renderer_null/null_rasterizer.cpp | 1 + .../renderer_null/null_rasterizer.h | 1 + src/video_core/renderer_opengl/gl_device.cpp | 1 + src/video_core/renderer_opengl/gl_device.h | 5 +++ .../renderer_opengl/gl_rasterizer.cpp | 34 +++++++++++++++++++ .../renderer_opengl/gl_rasterizer.h | 1 + .../renderer_vulkan/vk_rasterizer.cpp | 4 +++ .../renderer_vulkan/vk_rasterizer.h | 1 + src/video_core/texture_cache/texture_cache.h | 7 ++++ .../texture_cache/texture_cache_base.h | 3 ++ 14 files changed, 127 insertions(+), 1 deletion(-) diff --git a/src/video_core/engines/draw_manager.cpp b/src/video_core/engines/draw_manager.cpp index 2437121ce8..2685481f95 100644 --- a/src/video_core/engines/draw_manager.cpp +++ b/src/video_core/engines/draw_manager.cpp @@ -51,6 +51,10 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) { LOG_WARNING(HW_GPU, "(STUBBED) called"); break; } + case MAXWELL3D_REG_INDEX(draw_texture.src_y0): { + DrawTexture(); + break; + } default: break; } @@ -179,6 +183,33 @@ void DrawManager::DrawIndexSmall(u32 argument) { ProcessDraw(true, 1); } +void DrawManager::DrawTexture() { + const auto& regs{maxwell3d->regs}; + draw_texture_state.dst_x0 = static_cast(regs.draw_texture.dst_x0) / 4096.f; + draw_texture_state.dst_y0 = static_cast(regs.draw_texture.dst_y0) / 4096.f; + const auto dst_width = static_cast(regs.draw_texture.dst_width) / 4096.f; + const auto dst_height = static_cast(regs.draw_texture.dst_height) / 4096.f; + const bool lower_left{regs.window_origin.mode != + Maxwell3D::Regs::WindowOrigin::Mode::UpperLeft}; + if (lower_left) { + draw_texture_state.dst_y0 -= dst_height; + } + draw_texture_state.dst_x1 = draw_texture_state.dst_x0 + dst_width; + draw_texture_state.dst_y1 = draw_texture_state.dst_y0 + dst_height; + draw_texture_state.src_x0 = static_cast(regs.draw_texture.src_x0) / 4096.f; + draw_texture_state.src_y0 = static_cast(regs.draw_texture.src_y0) / 4096.f; + draw_texture_state.src_x1 = + (static_cast(regs.draw_texture.dx_du) / 4294967295.f) * dst_width + + draw_texture_state.src_x0; + draw_texture_state.src_y1 = + (static_cast(regs.draw_texture.dy_dv) / 4294967295.f) * dst_height + + draw_texture_state.src_y0; + draw_texture_state.src_sampler = regs.draw_texture.src_sampler; + draw_texture_state.src_texture = regs.draw_texture.src_texture; + + maxwell3d->rasterizer->DrawTexture(); +} + void DrawManager::UpdateTopology() { const auto& regs{maxwell3d->regs}; switch (regs.primitive_topology_control) { diff --git a/src/video_core/engines/draw_manager.h b/src/video_core/engines/draw_manager.h index 58d1b2d59f..7c22c49f12 100644 --- a/src/video_core/engines/draw_manager.h +++ b/src/video_core/engines/draw_manager.h @@ -32,6 +32,19 @@ public: std::vector inline_index_draw_indexes; }; + struct DrawTextureState { + f32 dst_x0; + f32 dst_y0; + f32 dst_x1; + f32 dst_y1; + f32 src_x0; + f32 src_y0; + f32 src_x1; + f32 src_y1; + u32 src_sampler; + u32 src_texture; + }; + struct IndirectParams { bool is_indexed; bool include_count; @@ -64,6 +77,10 @@ public: return draw_state; } + const DrawTextureState& GetDrawTextureState() const { + return draw_texture_state; + } + IndirectParams& GetIndirectParams() { return indirect_state; } @@ -81,6 +98,8 @@ private: void DrawIndexSmall(u32 argument); + void DrawTexture(); + void UpdateTopology(); void ProcessDraw(bool draw_indexed, u32 instance_count); @@ -89,6 +108,7 @@ private: Maxwell3D* maxwell3d{}; State draw_state{}; + DrawTextureState draw_texture_state{}; IndirectParams indirect_state{}; }; } // namespace Tegra::Engines diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 0b2fd29289..c89969bb45 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1599,6 +1599,20 @@ public: }; static_assert(sizeof(TIRModulationCoeff) == 0x4); + struct DrawTexture { + s32 dst_x0; + s32 dst_y0; + s32 dst_width; + s32 dst_height; + s64 dx_du; + s64 dy_dv; + u32 src_sampler; + u32 src_texture; + s32 src_x0; + s32 src_y0; + }; + static_assert(sizeof(DrawTexture) == 0x30); + struct ReduceColorThreshold { union { BitField<0, 8, u32> all_hit_once; @@ -2751,7 +2765,7 @@ public: u32 reserved_sw_method2; ///< 0x102C std::array tir_modulation_coeff; ///< 0x1030 std::array spare_nop; ///< 0x1044 - INSERT_PADDING_BYTES_NOINIT(0x30); + DrawTexture draw_texture; ///< 0x1080 std::array reserved_sw_method3_to_7; ///< 0x10B0 ReduceColorThreshold reduce_color_thresholds_unorm8; ///< 0x10CC std::array reserved_sw_method10_to_13; ///< 0x10D0 diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index f44c7df506..f980b12c61 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -46,6 +46,9 @@ public: /// Dispatches an indirect draw invocation virtual void DrawIndirect() {} + /// Dispatches an draw texture invocation + virtual void DrawTexture() = 0; + /// Clear the current framebuffer virtual void Clear(u32 layer_count) = 0; diff --git a/src/video_core/renderer_null/null_rasterizer.cpp b/src/video_core/renderer_null/null_rasterizer.cpp index 2c11345d7b..2b5c7defa6 100644 --- a/src/video_core/renderer_null/null_rasterizer.cpp +++ b/src/video_core/renderer_null/null_rasterizer.cpp @@ -21,6 +21,7 @@ RasterizerNull::RasterizerNull(Core::Memory::Memory& cpu_memory_, Tegra::GPU& gp RasterizerNull::~RasterizerNull() = default; void RasterizerNull::Draw(bool is_indexed, u32 instance_count) {} +void RasterizerNull::DrawTexture() {} void RasterizerNull::Clear(u32 layer_count) {} void RasterizerNull::DispatchCompute() {} void RasterizerNull::ResetCounter(VideoCore::QueryType type) {} diff --git a/src/video_core/renderer_null/null_rasterizer.h b/src/video_core/renderer_null/null_rasterizer.h index 2112aa70e1..51f896e43c 100644 --- a/src/video_core/renderer_null/null_rasterizer.h +++ b/src/video_core/renderer_null/null_rasterizer.h @@ -31,6 +31,7 @@ public: ~RasterizerNull() override; void Draw(bool is_indexed, u32 instance_count) override; + void DrawTexture() override; void Clear(u32 layer_count) override; void DispatchCompute() override; void ResetCounter(VideoCore::QueryType type) override; diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index cee5c32470..22ed16ebfb 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -166,6 +166,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) { has_shader_int64 = HasExtension(extensions, "GL_ARB_gpu_shader_int64"); has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float; has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2; + has_draw_texture = GLAD_GL_NV_draw_texture; warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel; need_fastmath_off = is_nvidia; can_report_memory = GLAD_GL_NVX_gpu_memory_info; diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 2a72d84be0..eaff364260 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -146,6 +146,10 @@ public: return has_sparse_texture_2; } + bool HasDrawTexture() const { + return has_draw_texture; + } + bool IsWarpSizePotentiallyLargerThanGuest() const { return warp_size_potentially_larger_than_guest; } @@ -216,6 +220,7 @@ private: bool has_shader_int64{}; bool has_amd_shader_half_float{}; bool has_sparse_texture_2{}; + bool has_draw_texture{}; bool warp_size_potentially_larger_than_guest{}; bool need_fastmath_off{}; bool has_cbuf_ftou_bug{}; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 7d48af8e17..86199b656d 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -318,6 +318,40 @@ void RasterizerOpenGL::DrawIndirect() { buffer_cache.SetDrawIndirect(nullptr); } +void RasterizerOpenGL::DrawTexture() { + MICROPROFILE_SCOPE(OpenGL_Drawing); + + SCOPE_EXIT({ gpu.TickWork(); }); + query_cache.UpdateCounters(); + + texture_cache.SynchronizeGraphicsDescriptors(); + texture_cache.UpdateRenderTargets(false); + + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); + + SyncState(); + + const auto& draw_texture_state = maxwell3d->draw_manager->GetDrawTextureState(); + const auto& sampler = texture_cache.GetGraphicsSampler(draw_texture_state.src_sampler); + const auto& texture = texture_cache.GetImageView(draw_texture_state.src_texture); + + if (device.HasDrawTexture()) { + glDrawTextureNV(texture.DefaultHandle(), sampler->Handle(), + static_cast(draw_texture_state.dst_x0), + static_cast(draw_texture_state.dst_y0), + static_cast(draw_texture_state.dst_x1), + static_cast(draw_texture_state.dst_y1), 0, + static_cast(draw_texture_state.src_x0) / texture.size.width, + static_cast(draw_texture_state.src_y0) / texture.size.height, + static_cast(draw_texture_state.src_x1) / texture.size.width, + static_cast(draw_texture_state.src_y1) / texture.size.height); + } else { + UNIMPLEMENTED(); + } + + ++num_queued_commands; +} + void RasterizerOpenGL::DispatchCompute() { ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()}; if (!pipeline) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index be4f76c18d..1b5348868b 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -70,6 +70,7 @@ public: void Draw(bool is_indexed, u32 instance_count) override; void DrawIndirect() override; + void DrawTexture() override; void Clear(u32 layer_count) override; void DispatchCompute() override; void ResetCounter(VideoCore::QueryType type) override; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 242bf9602a..0645a4460c 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -265,6 +265,10 @@ void RasterizerVulkan::DrawIndirect() { buffer_cache.SetDrawIndirect(nullptr); } +void RasterizerVulkan::DrawTexture() { + +} + void RasterizerVulkan::Clear(u32 layer_count) { MICROPROFILE_SCOPE(Vulkan_Clearing); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index c661e5b197..deb44dcaa5 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -66,6 +66,7 @@ public: void Draw(bool is_indexed, u32 instance_count) override; void DrawIndirect() override; + void DrawTexture() override; void Clear(u32 layer_count) override; void DispatchCompute() override; void ResetCounter(VideoCore::QueryType type) override; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 87152c8e99..1b01990a44 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -148,6 +148,13 @@ typename P::ImageView& TextureCache

::GetImageView(ImageViewId id) noexcept { return slot_image_views[id]; } +template +typename P::ImageView& TextureCache

::GetImageView(u32 index) noexcept { + const auto image_view_id = VisitImageView(channel_state->graphics_image_table, + channel_state->graphics_image_view_ids, index); + return slot_image_views[image_view_id]; +} + template void TextureCache

::MarkModification(ImageId id) noexcept { MarkModification(slot_images[id]); diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 4eea1f609e..485eaabaa8 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -129,6 +129,9 @@ public: /// Return a reference to the given image view id [[nodiscard]] ImageView& GetImageView(ImageViewId id) noexcept; + /// Get the imageview from the graphics descriptor table in the specified index + [[nodiscard]] ImageView& GetImageView(u32 index) noexcept; + /// Mark an image as modified from the GPU void MarkModification(ImageId id) noexcept;