RHI: Add Compute Pipeline abstraction with D3D12 and OpenGL support

- Add SetComputeShader/GetComputeShader/HasComputeShader to RHIPipelineState
- Add m_computePipelineState for D3D12 compute PSO
- Add m_computeProgram/m_computeShader for OpenGL
- Fix OpenGLCommandList::DispatchCompute bug (was ignoring x,y,z params)
- Fix OpenGLShader::GetID usage in OpenGLPipelineState
- Mark Priority 8 as completed in RHI_Design_Issues.md

RHI_Design_Issues.md

@@ -563,7 +563,7 @@ class RHITexture : public RHIResource { ... };
 | 5 | TransitionBarrier 针对 View 而非 Resource | 🟡 中 | 中 | ✅ 已完成 |
 | 6 | SetGlobal* 空操作 | 🟡 中 | 低 | ✅ 已完成 |
 | 7 | OpenGL 特有方法暴露 | 🟡 中 | 高 | ❌ 未完成 |
-| 8 | 缺少 Compute Pipeline 抽象 | 🟡 中 | 中 | ❌ 未完成 |
+| 8 | 缺少 Compute Pipeline 抽象 | 🟡 中 | 中 | ✅ 已完成 |
 
 ---
 

engine/include/XCEngine/RHI/D3D12/D3D12PipelineState.h

@@ -31,6 +31,7 @@ public:
     void SetTopology(uint32_t topologyType) override;
     void SetRenderTargetFormats(uint32_t count, const uint32_t* formats, uint32_t depthFormat) override;
     void SetSampleCount(uint32_t count) override;
+    void SetComputeShader(RHIShader* shader) override;
 
     // State query
     const RasterizerDesc& GetRasterizerState() const override { return m_rasterizerDesc; }
@@ -38,6 +39,8 @@ public:
     const DepthStencilStateDesc& GetDepthStencilState() const override { return m_depthStencilDesc; }
     const InputLayoutDesc& GetInputLayout() const override { return m_inputLayoutDesc; }
     PipelineStateHash GetHash() const override;
+    RHIShader* GetComputeShader() const override { return m_computeShader; }
+    bool HasComputeShader() const override { return m_csBytecode.pShaderBytecode != nullptr && m_csBytecode.BytecodeLength > 0; }
 
     // Finalization
     bool IsFinalized() const override { return m_finalized; }
@@ -45,12 +48,14 @@ public:
 
     // Shader Bytecode (set by CommandList when binding)
     void SetShaderBytecodes(const D3D12_SHADER_BYTECODE& vs, const D3D12_SHADER_BYTECODE& ps, const D3D12_SHADER_BYTECODE& gs = {});
+    void SetComputeShaderBytecodes(const D3D12_SHADER_BYTECODE& cs);
 
     // Lifecycle
     void Shutdown() override;
     ID3D12PipelineState* GetPipelineState() const { return m_pipelineState.Get(); }
+    ID3D12PipelineState* GetComputePipelineState() const { return m_computePipelineState.Get(); }
     void* GetNativeHandle() override { return m_pipelineState.Get(); }
-    PipelineType GetType() const override { return PipelineType::Graphics; }
+    PipelineType GetType() const override { return HasComputeShader() ? PipelineType::Compute : PipelineType::Graphics; }
 
     void Bind() override;
     void Unbind() override;
@@ -71,6 +76,7 @@ public:
 
 private:
     bool CreateD3D12PSO();
+    bool CreateD3D12ComputePSO();
 
     ID3D12Device* m_device;
     bool m_finalized = false;
@@ -91,10 +97,13 @@ private:
     D3D12_SHADER_BYTECODE m_vsBytecode = {};
     D3D12_SHADER_BYTECODE m_psBytecode = {};
     D3D12_SHADER_BYTECODE m_gsBytecode = {};
+    D3D12_SHADER_BYTECODE m_csBytecode = {};
+    class RHIShader* m_computeShader = nullptr;
     ID3D12RootSignature* m_rootSignature = nullptr;
 
     // D3D12 resources
     ComPtr<ID3D12PipelineState> m_pipelineState;
+    ComPtr<ID3D12PipelineState> m_computePipelineState;
     std::vector<D3D12_INPUT_ELEMENT_DESC> m_inputElements;
 };
 

engine/include/XCEngine/RHI/OpenGL/OpenGLCommandList.h

@@ -123,7 +123,7 @@ public:
 
     void Dispatch(uint32_t x, uint32_t y, uint32_t z) override;
     void DispatchIndirect(unsigned int buffer, size_t offset);
-    void DispatchCompute(unsigned int x, unsigned int y, unsigned int z, unsigned int groupX, unsigned int groupY, unsigned int groupZ);
+    void DispatchCompute(unsigned int x, unsigned int y, unsigned int z);
 
     void MemoryBarrier(unsigned int barriers);
     void TextureBarrier();

engine/include/XCEngine/RHI/OpenGL/OpenGLPipelineState.h

@@ -79,6 +79,7 @@ public:
     void SetTopology(uint32_t topologyType) override;
     void SetRenderTargetFormats(uint32_t count, const uint32_t* formats, uint32_t depthFormat) override;
     void SetSampleCount(uint32_t count) override;
+    void SetComputeShader(RHIShader* shader) override;
 
     // State query
     const RasterizerDesc& GetRasterizerState() const override { return m_rasterizerDesc; }
@@ -86,6 +87,8 @@ public:
     const DepthStencilStateDesc& GetDepthStencilState() const override { return m_depthStencilDesc; }
     const InputLayoutDesc& GetInputLayout() const override { return m_inputLayoutDesc; }
     PipelineStateHash GetHash() const override;
+    RHIShader* GetComputeShader() const override { return m_computeShader; }
+    bool HasComputeShader() const override { return m_computeProgram != 0; }
 
     // Finalization (OpenGL doesn't need it)
     bool IsFinalized() const override { return true; }
@@ -94,7 +97,7 @@ public:
     // Lifecycle
     void Shutdown() override;
     void* GetNativeHandle() override { return reinterpret_cast<void*>(static_cast<uintptr_t>(m_program)); }
-    PipelineType GetType() const override { return PipelineType::Graphics; }
+    PipelineType GetType() const override { return HasComputeShader() ? PipelineType::Compute : PipelineType::Graphics; }
 
     void Bind() override;
     void Unbind() override;
@@ -130,6 +133,8 @@ private:
     DepthStencilStateDesc m_depthStencilDesc;
     uint32_t m_topologyType = 0;
     unsigned int m_program = 0;
+    unsigned int m_computeProgram = 0;
+    class RHIShader* m_computeShader = nullptr;
     bool m_programAttached = false;
 
     // OpenGL specific state

engine/include/XCEngine/RHI/RHIPipelineState.h

@@ -6,6 +6,8 @@
 namespace XCEngine {
 namespace RHI {
 
+class RHIShader;
+
 class RHIPipelineState {
 public:
     virtual ~RHIPipelineState() = default;
@@ -18,6 +20,7 @@ public:
     virtual void SetTopology(uint32_t topologyType) = 0;  // PrimitiveTopologyType
     virtual void SetRenderTargetFormats(uint32_t count, const uint32_t* formats, uint32_t depthFormat) = 0;
     virtual void SetSampleCount(uint32_t count) = 0;
+    virtual void SetComputeShader(RHIShader* shader) = 0;
 
     // State query
     virtual const RasterizerDesc& GetRasterizerState() const = 0;
@@ -25,6 +28,8 @@ public:
     virtual const DepthStencilStateDesc& GetDepthStencilState() const = 0;
     virtual const InputLayoutDesc& GetInputLayout() const = 0;
     virtual PipelineStateHash GetHash() const = 0;
+    virtual RHIShader* GetComputeShader() const = 0;
+    virtual bool HasComputeShader() const = 0;
 
     // Finalization (D3D12/Vulkan creates real PSO)
     virtual bool IsFinalized() const = 0;

engine/src/RHI/D3D12/D3D12PipelineState.cpp

@@ -1,4 +1,5 @@
 #include "XCEngine/RHI/D3D12/D3D12PipelineState.h"
+#include "XCEngine/RHI/D3D12/D3D12Shader.h"
 #include <cstring>
 
 namespace XCEngine {
@@ -121,8 +122,19 @@ void D3D12PipelineState::SetShaderBytecodes(const D3D12_SHADER_BYTECODE& vs, con
     m_gsBytecode = gs;
 }
 
+void D3D12PipelineState::SetComputeShader(RHIShader* shader) {
+    m_computeShader = shader;
+}
+
+void D3D12PipelineState::SetComputeShaderBytecodes(const D3D12_SHADER_BYTECODE& cs) {
+    m_csBytecode = cs;
+}
+
 bool D3D12PipelineState::Finalize() {
     if (m_finalized) return true;
+    if (HasComputeShader()) {
+        return CreateD3D12ComputePSO();
+    }
     return CreateD3D12PSO();
 }
 
@@ -190,8 +202,27 @@ bool D3D12PipelineState::CreateD3D12PSO() {
     return true;
 }
 
+bool D3D12PipelineState::CreateD3D12ComputePSO() {
+    if (!m_csBytecode.pShaderBytecode || !m_csBytecode.BytecodeLength) {
+        return false;
+    }
+
+    D3D12_COMPUTE_PIPELINE_STATE_DESC desc = {};
+    desc.pRootSignature = m_rootSignature;
+    desc.CS = m_csBytecode;
+
+    HRESULT hr = m_device->CreateComputePipelineState(&desc, IID_PPV_ARGS(&m_computePipelineState));
+    if (FAILED(hr)) {
+        return false;
+    }
+
+    m_finalized = true;
+    return true;
+}
+
 void D3D12PipelineState::Shutdown() {
     m_pipelineState.Reset();
+    m_computePipelineState.Reset();
     m_finalized = false;
 }
 

engine/src/RHI/OpenGL/OpenGLCommandList.cpp

@@ -271,11 +271,8 @@ void OpenGLCommandList::DispatchIndirect(unsigned int buffer, size_t offset) {
     glBindBuffer(GL_DISPATCH_INDIRECT_BUFFER, 0);
 }
 
-void OpenGLCommandList::DispatchCompute(unsigned int x, unsigned int y, unsigned int z, unsigned int groupX, unsigned int groupY, unsigned int groupZ) {
+void OpenGLCommandList::DispatchCompute(unsigned int x, unsigned int y, unsigned int z) {
-    glDispatchCompute(groupX, groupY, groupZ);
+    glDispatchCompute(x, y, z);
-    (void)x;
-    (void)y;
-    (void)z;
 }
 
 void OpenGLCommandList::MemoryBarrier(unsigned int barriers) {

engine/src/RHI/OpenGL/OpenGLPipelineState.cpp

@@ -1,4 +1,5 @@
 #include "XCEngine/RHI/OpenGL/OpenGLPipelineState.h"
+#include "XCEngine/RHI/OpenGL/OpenGLShader.h"
 #include <glad/glad.h>
 
 namespace XCEngine {
@@ -37,6 +38,16 @@ void OpenGLPipelineState::SetRenderTargetFormats(uint32_t count, const uint32_t*
 void OpenGLPipelineState::SetSampleCount(uint32_t count) {
 }
 
+void OpenGLPipelineState::SetComputeShader(RHIShader* shader) {
+    m_computeShader = shader;
+    if (shader) {
+        OpenGLShader* glShader = static_cast<OpenGLShader*>(shader);
+        m_computeProgram = glShader->GetID();
+    } else {
+        m_computeProgram = 0;
+    }
+}
+
 PipelineStateHash OpenGLPipelineState::GetHash() const {
     PipelineStateHash hash = {};
     return hash;
@@ -44,11 +55,15 @@ PipelineStateHash OpenGLPipelineState::GetHash() const {
 
 void OpenGLPipelineState::Shutdown() {
     m_program = 0;
+    m_computeProgram = 0;
+    m_computeShader = nullptr;
     m_programAttached = false;
 }
 
 void OpenGLPipelineState::Bind() {
-    if (m_programAttached) {
+    if (HasComputeShader()) {
+        glUseProgram(m_computeProgram);
+    } else if (m_programAttached) {
         glUseProgram(m_program);
     }
     Apply();