Ray Tracing Pipeline #820

kevyuu · 2025-01-21T05:58:52Z

Description

Implement Ray Tracing Pipeline

Testing

Using example_tests

TODO list:

Implement Ray Tracing Pipeline Creation
Implement Ray Tracing Pipeline Binding Command
Implement Trace Ray Command

devshgraphicsprogrammingjenkins · 2025-01-21T05:59:03Z

[CI]: Can one of the admins verify this patch?

Signed-off-by: kevyuu <[email protected]>

devshgraphicsprogramming · 2025-01-24T13:54:49Z

include/nbl/video/CVulkanRayTracingPipeline.h

+  class CVulkanRayTracingPipeline final : public IGPURayTracingPipeline
+  {


don't indent just because of a namespace

devshgraphicsprogramming · 2025-01-24T13:55:41Z

include/nbl/video/IGPUCommandPool.h

+class IGPUCommandPool::CTraceRaysCmd final : public IFixedSizeCommand<CTraceRaysCmd>
+{
+    public:
+      CTraceRaysCmd(
+        core::smart_refctd_ptr<const IGPUBuffer>&& raygenGroupBuffer,
+        core::smart_refctd_ptr<const IGPUBuffer>&& hitGroupsBuffer,
+        core::smart_refctd_ptr<const IGPUBuffer>&&  missGroupsBuffer,
+        core::smart_refctd_ptr<const IGPUBuffer>&& callableGroupsBuffer) :
+        m_raygenGroupBuffer(raygenGroupBuffer),
+        m_hitGroupsBuffer(hitGroupsBuffer),
+        m_missGroupsBuffer(missGroupsBuffer),
+        m_callableGroupsBuffer(callableGroupsBuffer) {}
+
+
+    private:
+        core::smart_refctd_ptr<const IGPUBuffer> m_raygenGroupBuffer;
+        core::smart_refctd_ptr<const IGPUBuffer> m_hitGroupsBuffer;
+        core::smart_refctd_ptr<const IGPUBuffer> m_missGroupsBuffer;
+        core::smart_refctd_ptr<const IGPUBuffer> m_callableGroupsBuffer;
+};
+
+
+class IGPUCommandPool::CBindRayTracingPipelineCmd final : public IFixedSizeCommand<CBindRayTracingPipelineCmd>
+{
+    public:
+        CBindRayTracingPipelineCmd(core::smart_refctd_ptr<const IGPURayTracingPipeline>&& pipeline) : m_pipeline(std::move(pipeline)) {}
+
+    private:
+        core::smart_refctd_ptr<const IGPURayTracingPipeline> m_pipeline;
+};


very nice you got the hang of it!

devshgraphicsprogramming · 2025-01-24T13:57:07Z

src/nbl/asset/utils/CHLSLCompiler.cpp

+    case asset::IShader::E_SHADER_STAGE::ESS_RAYGEN: [[fallthrough]];
+    case asset::IShader::E_SHADER_STAGE::ESS_ANY_HIT: [[fallthrough]];
+    case asset::IShader::E_SHADER_STAGE::ESS_CLOSEST_HIT: [[fallthrough]];
+    case asset::IShader::E_SHADER_STAGE::ESS_MISS: [[fallthrough]];
+    case asset::IShader::E_SHADER_STAGE::ESS_INTERSECTION: [[fallthrough]];
+    case asset::IShader::E_SHADER_STAGE::ESS_CALLABLE: 
+        return L"lib";


@alichraghi what were the old problems when making a lib stage shader with our HLSL stuff?

I remember we only had issues with the Compute and Graphics pipelines using it

microsoft/DirectXShaderCompiler#6959 got closed but then DXC was crashing when i updated it but seems like it's actually usable now

ok lets reheat your PR with the lib target and test it on an example that really needs unscrewing of multiple preprocessor and template compilations (maybe arithmetic unit test?)

devshgraphicsprogramming · 2025-01-24T15:28:14Z

src/nbl/asset/interchange/CHLSLLoader.cpp

 		{".comp.hlsl",IShader::E_SHADER_STAGE::ESS_COMPUTE},
 		{".mesh.hlsl",IShader::E_SHADER_STAGE::ESS_MESH},
 		{".task.hlsl",IShader::E_SHADER_STAGE::ESS_TASK},
+		{".rgen.hlsl",IShader::E_SHADER_STAGE::ESS_RAYGEN},
+		{".rahit.hlsl",IShader::E_SHADER_STAGE::ESS_ANY_HIT},
+		{".rchit.hlsl",IShader::E_SHADER_STAGE::ESS_CLOSEST_HIT},
+		{".rmiss.hlsl",IShader::E_SHADER_STAGE::ESS_MISS},
+		{".rint.hlsl",IShader::E_SHADER_STAGE::ESS_INTERSECTION},
+		{".rcall.hlsl",IShader::E_SHADER_STAGE::ESS_CALLABLE},
 	};
 	auto shaderStage = IShader::E_SHADER_STAGE::ESS_UNKNOWN;


ah one thing that #754 was missing was something like this, to let .hlsl only (no stage.hlsl) extension map to library all-or-library stage

Signed-off-by: kevyuu <[email protected]>

devshgraphicsprogramming · 2025-01-26T20:46:23Z

include/nbl/asset/IBuffer.h

+template<typename BufferType>
+struct SStridedBufferRegion
+{
+	static constexpr inline size_t WholeBuffer = ~0ull;
+
+	size_t offset = 0ull;
+	size_t stride = 0;
+	size_t size = WholeBuffer;
+	core::smart_refctd_ptr<BufferType> buffer = nullptr;
+
+
+	inline operator SStridedBufferRegion<const BufferType>&() {return *reinterpret_cast<SStridedBufferRegion<const BufferType>*>(this);}
+	inline operator const SStridedBufferRegion<const BufferType>&() const {return *reinterpret_cast<const SStridedBufferRegion<const BufferType>*>(this);}
+
+	explicit inline operator bool() const {return isValid();}
+
+	inline bool isValid() const
+	{
+		if (!buffer || offset>=buffer->getSize() || size==0ull || stride>buffer->getSize())
+			return false;
+		return actualSize()<=buffer->getSize()-offset;
+	}
+
+	inline size_t actualSize() const
+	{
+		return size!=WholeBuffer ? size:buffer->getSize();
+	}
+	inline bool operator==(const SStridedBufferRegion<const BufferType>& rhs) const { return buffer==rhs.buffer && offset==rhs.offset && actualSize()==rhs.actualSize() && stride==rhs.stride; }
+	inline bool operator!=(const SStridedBufferRegion<const BufferType>& rhs) const { return !operator==(rhs); }
+};
+


I don't want to have this new thing in the API, take an SBufferOffset + stride in the functions that take this

devshgraphicsprogramming · 2025-01-26T20:46:43Z