The-Homebrew-Cloud/skyline
2
0
Fork 0
mirror of https://github.com/skyline-emu/skyline.git synced 2024-11-14 07:45:05 +01:00
Code Issues Packages Projects Releases Wiki Activity

Finish off transform feedback implementation

Browse Source
This commit is contained in:
Billy Laws 2022-10-09 14:07:00 +01:00
parent 36fd885b49
commit b72720e8db
5 changed files with 25 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/maxwell_3d.cpp
View File

@ -272,9 +272,11 @@ namespace skyline::gpu::interconnect::maxwell3d {
u32 vertexOffset; u32 vertexOffset;
u32 firstInstance; u32 firstInstance;
bool indexed; bool indexed;
bool transformFeedbackEnable;
}; };
auto *drawParams{ctx.executor.allocator->EmplaceUntracked<DrawParams>(DrawParams{stateUpdater, auto *drawParams{ctx.executor.allocator->EmplaceUntracked<DrawParams>(DrawParams{stateUpdater,
count, first, instanceCount, vertexOffset, firstInstance, indexed})}; count, first, instanceCount, vertexOffset, firstInstance, indexed,
ctx.gpu.traits.supportsTransformFeedback ? transformFeedbackEnable : false})};
const auto &surfaceClip{clearEngineRegisters.surfaceClip}; const auto &surfaceClip{clearEngineRegisters.surfaceClip};
vk::Rect2D scissor{ vk::Rect2D scissor{
@ -285,11 +287,16 @@ namespace skyline::gpu::interconnect::maxwell3d {
ctx.executor.AddSubpass([drawParams](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &, GPU &gpu, vk::RenderPass, u32) { ctx.executor.AddSubpass([drawParams](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &, GPU &gpu, vk::RenderPass, u32) {
drawParams->stateUpdater.RecordAll(gpu, commandBuffer); drawParams->stateUpdater.RecordAll(gpu, commandBuffer);
if (drawParams->transformFeedbackEnable)
commandBuffer.beginTransformFeedbackEXT(0, {}, {});
if (drawParams->indexed) if (drawParams->indexed)
commandBuffer.drawIndexed(drawParams->count, drawParams->instanceCount, drawParams->first, static_cast<i32>(drawParams->vertexOffset), drawParams->firstInstance); commandBuffer.drawIndexed(drawParams->count, drawParams->instanceCount, drawParams->first, static_cast<i32>(drawParams->vertexOffset), drawParams->firstInstance);
else else
commandBuffer.draw(drawParams->count, drawParams->instanceCount, drawParams->first, drawParams->firstInstance); commandBuffer.draw(drawParams->count, drawParams->instanceCount, drawParams->first, drawParams->firstInstance);
if (drawParams->transformFeedbackEnable)
commandBuffer.endTransformFeedbackEXT(0, {}, {});
}, scissor, {}, activeState.GetColorAttachments(), activeState.GetDepthAttachment(), !ctx.gpu.traits.quirks.relaxedRenderPassCompatibility); }, scissor, {}, activeState.GetColorAttachments(), activeState.GetDepthAttachment(), !ctx.gpu.traits.quirks.relaxedRenderPassCompatibility);
constantBuffers.ResetQuickBind(); constantBuffers.ResetQuickBind();

2
app/src/main/cpp/skyline/gpu/interconnect/maxwell_3d/maxwell_3d.h
View File

@ -84,6 +84,6 @@ namespace skyline::gpu::interconnect::maxwell3d {
void Clear(engine::ClearSurface &clearSurface); void Clear(engine::ClearSurface &clearSurface);
void Draw(engine::DrawTopology topology, bool indexed, u32 count, u32 first, u32 instanceCount, u32 vertexOffset, u32 firstInstance); void Draw(engine::DrawTopology topology, bool transformFeedbackEnable, bool indexed, u32 count, u32 first, u32 instanceCount, u32 vertexOffset, u32 firstInstance);
}; };
} }

14
app/src/main/cpp/skyline/soc/gm20b/engines/maxwell/types.h
View File

@ -959,15 +959,17 @@ namespace skyline::soc::gm20b::engine::maxwell3d::type {
}; };
static_assert(sizeof(StreamOutBuffer) == (sizeof(u32) * 8)); static_assert(sizeof(StreamOutBuffer) == (sizeof(u32) * 8));
struct TransformFeedbackBufferState { struct StreamOutControl {
u32 bufferIndex; u8 streamSelect : 2;
u32 varyingCount; u32 _pad0_ : 30;
u32 stride; u8 componentCount : 8;
u32 _pad1_ : 24;
u32 strideBytes;
u32 _pad_; u32 _pad_;
}; };
static_assert(sizeof(TransformFeedbackBufferState) == (sizeof(u32) * 4)); static_assert(sizeof(StreamOutControl) == (sizeof(u32) * 4));
constexpr static size_t TransformFeedbackVaryingCount{0x80}; constexpr static size_t StreamOutLayoutSelectAttributeCount{0x80};
struct VertexStream { struct VertexStream {
union { union {

9
app/src/main/cpp/skyline/soc/gm20b/engines/maxwell_3d.cpp
View File

@ -23,6 +23,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
.rasterizationRegisters = {*registers.rasterEnable, *registers.frontPolygonMode, *registers.backPolygonMode, *registers.oglCullEnable, *registers.oglCullFace, *registers.windowOrigin, *registers.oglFrontFace, *registers.viewportClipControl, *registers.polyOffset, *registers.provokingVertex, *registers.pointSize, *registers.zClipRange}, .rasterizationRegisters = {*registers.rasterEnable, *registers.frontPolygonMode, *registers.backPolygonMode, *registers.oglCullEnable, *registers.oglCullFace, *registers.windowOrigin, *registers.oglFrontFace, *registers.viewportClipControl, *registers.polyOffset, *registers.provokingVertex, *registers.pointSize, *registers.zClipRange},
.depthStencilRegisters = {*registers.depthTestEnable, *registers.depthWriteEnable, *registers.depthFunc, *registers.depthBoundsTestEnable, *registers.stencilTestEnable, *registers.twoSidedStencilTestEnable, *registers.stencilOps, *registers.stencilBack}, .depthStencilRegisters = {*registers.depthTestEnable, *registers.depthWriteEnable, *registers.depthFunc, *registers.depthBoundsTestEnable, *registers.stencilTestEnable, *registers.twoSidedStencilTestEnable, *registers.stencilOps, *registers.stencilBack},
.colorBlendRegisters = {*registers.logicOp, *registers.singleCtWriteControl, *registers.ctWrites, *registers.blendStatePerTargetEnable, *registers.blendPerTargets, *registers.blend}, .colorBlendRegisters = {*registers.logicOp, *registers.singleCtWriteControl, *registers.ctWrites, *registers.blendStatePerTargetEnable, *registers.blendPerTargets, *registers.blend},
.transformFeedbackRegisters = {*registers.streamOutputEnable, *registers.streamOutControls, *registers.streamOutLayoutSelect},
.globalShaderConfigRegisters = {*registers.postVtgShaderAttributeSkipMask, *registers.bindlessTexture, *registers.apiMandatedEarlyZEnable}, .globalShaderConfigRegisters = {*registers.postVtgShaderAttributeSkipMask, *registers.bindlessTexture, *registers.apiMandatedEarlyZEnable},
.ctSelect = *registers.ctSelect .ctSelect = *registers.ctSelect
}; };
@ -33,7 +34,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
.pipelineRegisters = MakePipelineStateRegisters(registers), .pipelineRegisters = MakePipelineStateRegisters(registers),
.vertexBuffersRegisters = util::MergeInto<REGTYPE(VertexBufferState), type::VertexStreamCount>(*registers.vertexStreams, *registers.vertexStreamLimits), .vertexBuffersRegisters = util::MergeInto<REGTYPE(VertexBufferState), type::VertexStreamCount>(*registers.vertexStreams, *registers.vertexStreamLimits),
.indexBufferRegisters = {*registers.indexBuffer}, .indexBufferRegisters = {*registers.indexBuffer},
.transformFeedbackBuffersRegisters = util::MergeInto<REGTYPE(TransformFeedbackBufferState), type::StreamOutBufferCount>(*registers.streamOutBuffers, *registers.streamOutEnable), .transformFeedbackBuffersRegisters = util::MergeInto<REGTYPE(TransformFeedbackBufferState), type::StreamOutBufferCount>(*registers.streamOutBuffers, *registers.streamOutputEnable),
.viewportsRegisters = util::MergeInto<REGTYPE(ViewportState), type::ViewportCount>(registers.viewports[0], registers.viewportClips[0], *registers.viewports, *registers.viewportClips, *registers.windowOrigin, *registers.viewportScaleOffsetEnable), .viewportsRegisters = util::MergeInto<REGTYPE(ViewportState), type::ViewportCount>(registers.viewports[0], registers.viewportClips[0], *registers.viewports, *registers.viewportClips, *registers.windowOrigin, *registers.viewportScaleOffsetEnable),
.scissorsRegisters = util::MergeInto<REGTYPE(ScissorState), type::ViewportCount>(*registers.scissors), .scissorsRegisters = util::MergeInto<REGTYPE(ScissorState), type::ViewportCount>(*registers.scissors),
.lineWidthRegisters = {*registers.lineWidth, *registers.lineWidthAliased, *registers.aliasedLineWidthEnable}, .lineWidthRegisters = {*registers.lineWidth, *registers.lineWidthAliased, *registers.aliasedLineWidthEnable},
@ -74,7 +75,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
__attribute__((always_inline)) void Maxwell3D::FlushDeferredDraw() { __attribute__((always_inline)) void Maxwell3D::FlushDeferredDraw() {
if (batchEnableState.drawActive) { if (batchEnableState.drawActive) {
batchEnableState.drawActive = false; batchEnableState.drawActive = false;
interconnect.Draw(deferredDraw.drawTopology, deferredDraw.indexed, deferredDraw.drawCount, deferredDraw.drawFirst, deferredDraw.instanceCount, deferredDraw.drawBaseVertex, deferredDraw.drawBaseInstance); interconnect.Draw(deferredDraw.drawTopology, *registers.streamOutputEnable, deferredDraw.indexed, deferredDraw.drawCount, deferredDraw.drawFirst, deferredDraw.instanceCount, deferredDraw.drawBaseVertex, deferredDraw.drawBaseInstance);
deferredDraw.instanceCount = 1; deferredDraw.instanceCount = 1;
} }
} }
@ -346,7 +347,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
registers.globalBaseInstanceIndex = globalBaseInstanceIndex; registers.globalBaseInstanceIndex = globalBaseInstanceIndex;
} }
interconnect.Draw(topology, false, vertexArrayCount, vertexArrayStart, instanceCount, 0, globalBaseInstanceIndex); interconnect.Draw(topology, *registers.streamOutputEnable, false, vertexArrayCount, vertexArrayStart, instanceCount, 0, globalBaseInstanceIndex);
} }
void Maxwell3D::DrawIndexedInstanced(bool setRegs, u32 drawTopology, u32 indexBufferCount, u32 instanceCount, u32 globalBaseVertexIndex, u32 indexBufferFirst, u32 globalBaseInstanceIndex) { void Maxwell3D::DrawIndexedInstanced(bool setRegs, u32 drawTopology, u32 indexBufferCount, u32 instanceCount, u32 globalBaseVertexIndex, u32 indexBufferFirst, u32 globalBaseInstanceIndex) {
@ -359,6 +360,6 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
registers.globalBaseInstanceIndex = globalBaseInstanceIndex; registers.globalBaseInstanceIndex = globalBaseInstanceIndex;
} }
interconnect.Draw(topology, true, indexBufferCount, indexBufferFirst, instanceCount, globalBaseVertexIndex, globalBaseInstanceIndex); interconnect.Draw(topology, *registers.streamOutputEnable, true, indexBufferCount, indexBufferFirst, instanceCount, globalBaseVertexIndex, globalBaseInstanceIndex);
} }
} }

6
app/src/main/cpp/skyline/soc/gm20b/engines/maxwell_3d.h
View File

@ -119,8 +119,8 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
Register<0xDF, u32> rasterEnable; Register<0xDF, u32> rasterEnable;
Register<0xE0, std::array<type::StreamOutBuffer, type::StreamOutBufferCount>> streamOutBuffers; Register<0xE0, std::array<type::StreamOutBuffer, type::StreamOutBufferCount>> streamOutBuffers;
Register<0x1C0, std::array<type::TransformFeedbackBufferState, type::StreamOutBufferCount>> transformFeedbackBufferStates; Register<0x1C0, std::array<type::StreamOutControl, type::StreamOutBufferCount>> streamOutControls;
Register<0x1D1, u32> streamOutEnable; Register<0x1D1, u32> streamOutputEnable;
Register<0x200, std::array<type::ColorTarget, type::ColorTargetCount>> colorTargets; Register<0x200, std::array<type::ColorTarget, type::ColorTargetCount>> colorTargets;
Register<0x280, std::array<type::Viewport, type::ViewportCount>> viewports; Register<0x280, std::array<type::Viewport, type::ViewportCount>> viewports;
@ -354,7 +354,7 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
Register<0x982, type::BindlessTexture> bindlessTexture; //!< The index of the constant buffer containing bindless texture descriptors Register<0x982, type::BindlessTexture> bindlessTexture; //!< The index of the constant buffer containing bindless texture descriptors
Register<0xA00, std::array<u32, (type::TransformFeedbackVaryingCount / sizeof(u32)) * type::StreamOutBufferCount>> transformFeedbackVaryings; Register<0xA00, std::array<std::array<u8, type::StreamOutLayoutSelectAttributeCount>, type::StreamOutBufferCount>> streamOutLayoutSelect;
}; };
static_assert(sizeof(Registers) == (EngineMethodsEnd * sizeof(u32))); static_assert(sizeof(Registers) == (EngineMethodsEnd * sizeof(u32)));
#pragma pack(pop) #pragma pack(pop)