Merged cuda_rendering branch

CMakeLists.txt

@@ -105,18 +105,16 @@ if (NOT OPENIMAGEIO_FOUND)
 	MESSAGE(FATAL_ERROR "--> Could not locate required OpenImageIO files - Please check ${OPENIMAGEIO_SEARCH_PATH}")
 endif()
 
-if (NOT OPENCL_FOUND AND NOT LUXRAYS_DISABLE_OPENCL)
-	MESSAGE(ERROR "--> Could not locate required OpenCL files, disabling OpenCL support - Please check ${OPENCL_SEARCH_PATH}")
-	SET(LUXRAYS_DISABLE_OPENCL 1)
+if (NOT OPENCL_FOUND AND LUXRAYS_ENABLE_OPENCL)
+	MESSAGE(FATAL_ERROR "--> Could not locate required OpenCL files, disabling OpenCL support - Please check ${OPENCL_SEARCH_PATH}")
 endif()
 
-if (NOT CUDA_FOUND AND LUXRAYS_ENABLE_CUDA)
-	MESSAGE(FATAL_ERROR "--> Could not locate required CUDA files - Please check ${CUDA_TOOLKIT_ROOT_DIR}")
+if (NOT OPENGL_FOUND AND LUXRAYS_ENABLE_OPENCL)
+	MESSAGE(FATAL_ERROR "--> Could not locate required OpenGL files, disabling OpenCL support, disabling samples build")
 endif()
 
-if (NOT OPENGL_FOUND AND NOT LUXRAYS_DISABLE_OPENCL)
-	MESSAGE(ERROR "--> Could not locate required OpenGL files, disabling OpenCL support, disabling samples build")
-	SET(LUXRAYS_DISABLE_OPENCL 1)
+if ((NOT CUDA_FOUND AND LUXRAYS_ENABLE_CUDA) OR (LUXRAYS_ENABLE_CUDA AND NOT LUXRAYS_ENABLE_OPENCL))
+	MESSAGE(FATAL_ERROR "--> CUDA support requires to enable OpenCL support too")
 endif()
 
 if (NOT EMBREE_FOUND)
@@ -141,8 +139,8 @@ endif()
 #
 ################################################################################
 
-if (LUXRAYS_DISABLE_OPENCL)
-	ADD_DEFINITIONS("-DLUXRAYS_DISABLE_OPENCL")
+if (LUXRAYS_ENABLE_OPENCL)
+	ADD_DEFINITIONS("-DLUXRAYS_ENABLE_OPENCL")
 endif()
 
 if (LUXRAYS_ENABLE_CUDA)

cmake/Dependencies.cmake

@@ -99,7 +99,7 @@ if (OPENGL_FOUND)
 endif ()
 
 # OpenCL
-if (NOT LUXRAYS_DISABLE_OPENCL)
+if (LUXRAYS_ENABLE_OPENCL)
 	set(OPENCL_ROOT                "${OPENCL_SEARCH_PATH}")
 	find_package(OpenCL)
 

cmake/SpecializedConfig/Config_Dade.cmake

@@ -21,8 +21,8 @@ SET(CMAKE_LIBRARY_PATH "/home/david/projects/luxcorerender/LinuxCompile/target-6
 SET(Blosc_USE_STATIC_LIBS   "ON")
 
 # Note: CUDA requires to have enabled OpenCL support too
-#SET(LUXRAYS_ENABLE_CUDA TRUE)
-#SET(LUXRAYS_DISABLE_OPENCL TRUE)
+SET(LUXRAYS_ENABLE_CUDA TRUE)
+SET(LUXRAYS_ENABLE_OPENCL TRUE)
 #SET(BUILD_LUXCORE_DLL TRUE)
 
 #SET(CMAKE_BUILD_TYPE "Debug")

cmake/SpecializedConfig/Config_Linux-Static.cmake

@@ -12,7 +12,7 @@ SET(CMAKE_INCLUDE_PATH "${LuxRays_SOURCE_DIR}/../target-64-sse2/include;${LuxRay
 SET(CMAKE_LIBRARY_PATH "${LuxRays_SOURCE_DIR}/../target-64-sse2/lib;${LuxRays_SOURCE_DIR}/../target-64-sse2")
 SET(Blosc_USE_STATIC_LIBS   "ON")
 
-#SET(LUXRAYS_DISABLE_OPENCL TRUE)
+#SET(LUXRAYS_ENABLE_OPENCL TRUE)
 #SET(BUILD_LUXCORE_DLL TRUE)
 
 #SET(CMAKE_BUILD_TYPE "Debug")

include/luxcore/luxcore.h

@@ -111,9 +111,12 @@ CPP_EXPORT CPP_API void KernelCacheFill(const luxrays::Properties &config, void
  * properties currently supported are:
  * - version.number\n
  *		The LuxCore version. As string with "major.minor" version format.
- * - compile.LUXRAYS_DISABLE_OPENCL\n
- *		true or false if the sources has been compiled with LUXRAYS_DISABLE_OPENCL and
+ * - compile.LUXRAYS_ENABLE_OPENCL and compile.LUXRAYS_DISABLE_OPENCL\n
+ *		true or false if the sources has been compiled with LUXRAYS_ENABLE_OPENCL and
  *		OpenCL rendering engines are available or not.
+ * - compile.LUXRAYS_ENABLE_CUDA\n
+ *		true or false if the sources has been compiled with LUXRAYS_ENABLE_CUDA and
+ *		CUDA support is available or not.
  * - compile.LUXCORE_DISABLE_EMBREE_BVH_BUILDER\n
  *		true or false if the sources has been compiled with LUXCORE_DISABLE_EMBREE_BVH_BUILDER and
  *		Embree BVH builder is used for OpenCL or not. This is now always false.

include/luxrays/accelerators/bvh.cl

@@ -1,4 +1,4 @@
-#line 2 "bvh_kernel.cl"
+#line 2 "bvh.cl"
 
 /***************************************************************************
  * Copyright 1998-2020 by authors (see AUTHORS.txt)                        *
@@ -120,8 +120,8 @@ void Accelerator_Intersect(
 	const uint stopNode = BVHNodeData_GetSkipIndex(accelNodePage0[0].nodeData); // Non-existent
 #endif
 
-	const float3 rayOrig = (float3)(ray->o.x, ray->o.y, ray->o.z);
-	const float3 rayDir = (float3)(ray->d.x, ray->d.y, ray->d.z);
+	const float3 rayOrig = MAKE_FLOAT3(ray->o.x, ray->o.y, ray->o.z);
+	const float3 rayDir = MAKE_FLOAT3(ray->d.x, ray->d.y, ray->d.z);
 	const float mint = ray->mint;
 	float maxt = ray->maxt;
 
@@ -146,16 +146,16 @@ void Accelerator_Intersect(
 		const float4 data1 = *data;
 
 		//const uint nodeData = node->nodeData;
-		const uint nodeData = as_uint(data1.s2);
+		const uint nodeData = as_uint(data1.z);
 		if (BVHNodeData_IsLeaf(nodeData)) {
 			// It is a leaf, check the triangle
 
 			//const uint i0 = node->triangleLeaf.v[0];
 			//const uint i1 = node->triangleLeaf.v[1];
 			//const uint i2 = node->triangleLeaf.v[2];
-			const uint v0 = as_uint(data0.s0);
-			const uint v1 = as_uint(data0.s1);
-			const uint v2 = as_uint(data0.s2);
+			const uint v0 = as_uint(data0.x);
+			const uint v1 = as_uint(data0.y);
+			const uint v2 = as_uint(data0.z);
 
 #if (BVH_VERTS_PAGE_COUNT == 1)
 			// Fast path for when there is only one memory page
@@ -180,8 +180,8 @@ void Accelerator_Intersect(
 #endif
 
 			//const uint triangleIndex = node->triangleLeaf.triangleIndex;
-			const uint meshIndex = as_uint(data0.s3);
-			const uint triangleIndex = as_uint(data1.s0);
+			const uint meshIndex = as_uint(data0.w);
+			const uint triangleIndex = as_uint(data1.x);
 
 			Triangle_Intersect(rayOrig, rayDir, mint, &maxt, &hitMeshIndex, &hitTriangleIndex,
 					&b1, &b2, meshIndex, triangleIndex, p0, p1, p2);
@@ -194,8 +194,8 @@ void Accelerator_Intersect(
 			// It is a node, check the bounding box
 			//const float3 pMin = VLOAD3F(&node->bvhNode.bboxMin[0]);
 			//const float3 pMax = VLOAD3F(&node->bvhNode.bboxMax[0]);
-			const float3 pMin = (float3)(data0.s0, data0.s1, data0.s2);
-			const float3 pMax = (float3)(data0.s3, data1.s0, data1.s1);
+			const float3 pMin = MAKE_FLOAT3(data0.x, data0.y, data0.z);
+			const float3 pMax = MAKE_FLOAT3(data0.w, data1.x, data1.y);
 
 			if (BBox_IntersectP(pMin, pMax, rayOrig, invRayDir, mint, maxt)) {
 #if (BVH_NODES_PAGE_COUNT == 1)

include/luxrays/accelerators/bvhaccel.h

@@ -28,6 +28,8 @@
 
 namespace luxrays {
 
+class HardwareIntersectionDevice;
+
 // BVHAccel Declarations
 class BVHAccel : public Accelerator {
 public:
@@ -36,7 +38,10 @@ class BVHAccel : public Accelerator {
 	virtual ~BVHAccel();
 
 	virtual AcceleratorType GetType() const { return ACCEL_BVH; }
-	virtual OpenCLKernel *NewOpenCLKernel(OpenCLIntersectionDevice *device) const;
+
+	virtual bool HasDataParallelSupport(const IntersectionDevice &device) const;
+	virtual HardwareIntersectionKernel *NewHardwareIntersectionKernel(HardwareIntersectionDevice &device) const;
+
 	virtual void Init(const std::deque<const Mesh *> &meshes,
 		const u_longlong totalVertexCount,
 		const u_longlong totalTriangleCount);
@@ -45,11 +50,9 @@ class BVHAccel : public Accelerator {
 
 	static BVHParams ToBVHParams(const Properties &props);
 
+	friend class BVHKernel;
+	friend class MBVHKernel;
 	friend class MBVHAccel;
-#if !defined(LUXRAYS_DISABLE_OPENCL)
-	friend class OpenCLBVHKernel;
-	friend class OpenCLMBVHKernel;
-#endif
 
 private:
 	BVHParams params;

include/luxrays/accelerators/embreeaccel.h

@@ -35,10 +35,12 @@ class EmbreeAccel : public Accelerator {
 
 	virtual AcceleratorType GetType() const { return ACCEL_EMBREE; }
 
-	virtual OpenCLKernel *NewOpenCLKernel(OpenCLIntersectionDevice *device) const { return NULL; }
-	virtual bool CanRunOnOpenCLDevice(OpenCLIntersectionDevice *device) const {
+	virtual bool HasDataParallelSupport(const IntersectionDevice &device) const {
 		return false;
 	}
+	virtual HardwareIntersectionKernel *NewHardwareIntersectionKernel(HardwareIntersectionDevice &device) const {
+		return nullptr;
+	}
 
 	virtual void Init(const std::deque<const Mesh *> &meshes,
 		const u_longlong totalVertexCount,

include/luxrays/accelerators/mbvh.cl

@@ -1,4 +1,4 @@
-#line 2 "mbvh_kernel.cl"
+#line 2 "mbvh.cl"
 
 /***************************************************************************
  * Copyright 1998-2020 by authors (see AUTHORS.txt)                        *
@@ -152,8 +152,8 @@ void Accelerator_Intersect(
 
 	uint currentMeshOffset = 0;
 
-	const float3 rootRayOrig = (float3)(ray->o.x, ray->o.y, ray->o.z);
-	const float3 rootRayDir = (float3)(ray->d.x, ray->d.y, ray->d.z);
+	const float3 rootRayOrig = MAKE_FLOAT3(ray->o.x, ray->o.y, ray->o.z);
+	const float3 rootRayDir = MAKE_FLOAT3(ray->d.x, ray->d.y, ray->d.z);
 	const float mint = ray->mint;
 	float maxt = ray->maxt;
 #if defined(MBVH_HAS_MOTIONSYSTEMS)
@@ -213,17 +213,17 @@ void Accelerator_Intersect(
 		const float4 data1 = *data;
 
 		//const uint nodeData = node->nodeData;
-		const uint nodeData = as_uint(data1.s2);
+		const uint nodeData = as_uint(data1.z);
 		if (BVHNodeData_IsLeaf(nodeData)) {
 			if (insideLeafTree) {
 				// I'm inside a leaf tree, I have to check the triangle
 
 				//const uint i0 = node->triangleLeaf.v[0];
 				//const uint i1 = node->triangleLeaf.v[1];
 				//const uint i2 = node->triangleLeaf.v[2];
-				const uint v0 = as_uint(data0.s0);
-				const uint v1 = as_uint(data0.s1);
-				const uint v2 = as_uint(data0.s2);
+				const uint v0 = as_uint(data0.x);
+				const uint v1 = as_uint(data0.y);
+				const uint v2 = as_uint(data0.z);
 
 #if (MBVH_VERTS_PAGE_COUNT == 1)
 				// Fast path for when there is only one memory page
@@ -249,8 +249,8 @@ void Accelerator_Intersect(
 
 				//const uint meshIndex = node->triangleLeaf.meshIndex + currentMeshOffset;
 				//const uint triangleIndex = node->triangleLeaf.triangleIndex;
-				const uint meshIndex = as_uint(data0.s3) + currentMeshOffset;
-				const uint triangleIndex = as_uint(data1.s0);
+				const uint meshIndex = as_uint(data0.w) + currentMeshOffset;
+				const uint triangleIndex = as_uint(data1.x);
 				Triangle_Intersect(currentRayOrig, currentRayDir, mint, &maxt, &hitMeshIndex, &hitTriangleIndex, &b1, &b2,
 					meshIndex, triangleIndex, p0, p1, p2);
 #if (MBVH_NODES_PAGE_COUNT == 1)
@@ -265,7 +265,7 @@ void Accelerator_Intersect(
 
 #if defined(MBVH_HAS_TRANSFORMATIONS)				
 				//const uint transformIndex = node->bvhLeaf.transformIndex;
-				const uint transformIndex = as_int(data0.s1);
+				const uint transformIndex = as_int(data0.y);
 				if (transformIndex != NULL_INDEX) {
 					// Transform ray origin and direction
 					__global const Matrix4x4* restrict m = &leafTransformations[transformIndex];
@@ -276,7 +276,7 @@ void Accelerator_Intersect(
 
 #if defined(MBVH_HAS_MOTIONSYSTEMS)
 				//const uint motionIndex = node->bvhLeaf.motionIndex;
-				const uint motionIndex = as_int(data0.s2);
+				const uint motionIndex = as_int(data0.z);
 				if (motionIndex != NULL_INDEX) {
 					// Transform ray origin and direction
 					Matrix4x4 m;
@@ -287,10 +287,10 @@ void Accelerator_Intersect(
 #endif
 
 				//currentMeshOffset = node->bvhLeaf.meshOffsetIndex;
-				currentMeshOffset = as_int(data0.s3);
+				currentMeshOffset = as_int(data0.w);
 
 				//const uint leafIndex = node->bvhLeaf.leafIndex;
-				const uint leafIndex = as_int(data0.s0);
+				const uint leafIndex = as_int(data0.x);
 #if (MBVH_NODES_PAGE_COUNT == 1)
 				currentRootNode = currentNode + 1;
 				currentNode = leafIndex;
@@ -317,8 +317,8 @@ void Accelerator_Intersect(
 			// It is a node, check the bounding box
 			//const float3 pMin = VLOAD3F(&node->bvhNode.bboxMin[0]);
 			//const float3 pMax = VLOAD3F(&node->bvhNode.bboxMax[0]);
-			const float3 pMin = (float3)(data0.s0, data0.s1, data0.s2);
-			const float3 pMax = (float3)(data0.s3, data1.s0, data1.s1);
+			const float3 pMin = MAKE_FLOAT3(data0.x, data0.y, data0.z);
+			const float3 pMax = MAKE_FLOAT3(data0.w, data1.x, data1.y);
 
 			if (BBox_IntersectP(pMin, pMax, currentRayOrig, 1.f / currentRayDir, mint, maxt)) {
 #if (MBVH_NODES_PAGE_COUNT == 1)

include/luxrays/accelerators/mbvhaccel.h

@@ -28,6 +28,8 @@
 
 namespace luxrays {
 
+class HardwareIntersectionDevice;
+
 // MBVHAccel Declarations
 class MBVHAccel : public Accelerator {
 public:
@@ -36,7 +38,10 @@ class MBVHAccel : public Accelerator {
 	virtual ~MBVHAccel();
 
 	virtual AcceleratorType GetType() const { return ACCEL_MBVH; }
-	virtual OpenCLKernel *NewOpenCLKernel(OpenCLIntersectionDevice *device) const;
+
+	virtual bool HasDataParallelSupport(const IntersectionDevice &device) const;
+	virtual HardwareIntersectionKernel *NewHardwareIntersectionKernel(HardwareIntersectionDevice &device) const;
+
 	virtual void Init(const std::deque<const Mesh *> &meshes,
 		const u_longlong totalVertexCount,
 		const u_longlong totalTriangleCount);
@@ -46,9 +51,7 @@ class MBVHAccel : public Accelerator {
 
 	virtual bool Intersect(const Ray *ray, RayHit *hit) const;
 
-#if !defined(LUXRAYS_DISABLE_OPENCL)
-	friend class OpenCLMBVHKernel;
-#endif
+	friend class MBVHKernel;
 
 private:
 	static bool MeshPtrCompare(const Mesh *, const Mesh *);

include/luxrays/core/accelerator.h

@@ -32,8 +32,9 @@ typedef enum {
 	ACCEL_AUTO, ACCEL_BVH, ACCEL_MBVH, ACCEL_EMBREE
 } AcceleratorType;
 
-class OpenCLKernel;
-class OpenCLIntersectionDevice;
+class IntersectionDevice;
+class HardwareIntersectionDevice;
+class HardwareIntersectionKernel;
 
 class Accelerator {
 public:
@@ -42,8 +43,8 @@ class Accelerator {
 
 	virtual AcceleratorType GetType() const = 0;
 
-	virtual OpenCLKernel *NewOpenCLKernel(OpenCLIntersectionDevice *device) const = 0;
-	virtual bool CanRunOnOpenCLDevice(OpenCLIntersectionDevice *device) const { return true; }
+	virtual bool HasDataParallelSupport(const IntersectionDevice &device) const = 0;
+	virtual HardwareIntersectionKernel *NewHardwareIntersectionKernel(HardwareIntersectionDevice &device) const = 0;
 
 	virtual void Init(const std::deque<const Mesh *> &meshes, const u_longlong totalVertexCount, const u_longlong totalTriangleCount) = 0;
 	virtual bool DoesSupportUpdate() const { return false; }

include/luxrays/core/context.h

@@ -168,7 +168,7 @@ class Context {
 	void SetVerbose(const bool v) { verbose = v; }
 	bool IsVerbose() const { return verbose; }
 
-#if !defined(LUXRAYS_DISABLE_OPENCL)
+#if defined(LUXRAYS_ENABLE_OPENCL)
 	friend class OpenCLIntersectionDevice;
 #endif