Cycles: more opencl tweaks, status is:

* kernel has shading nodes / textures disabled, amd/nvidia opencl
  compilers choke on these, need to figure out how to avoid this
* works in cycles_test, not available as option in blender yet
* kernel compiles and runs with opencl 1.1 from intel/amd/nvidia

5 files changed, 44 insertions, 30 deletions

diff --git a/intern/cycles/app/cycles_test.cpp b/intern/cycles/app/cycles_test.cpp
index 2603458f088..698974b6277 100644
--- a/intern/cycles/app/cycles_test.cpp
+++ b/intern/cycles/app/cycles_test.cpp
@@ -284,7 +284,7 @@ using namespace ccl;
 
 int main(int argc, const char **argv)
 {
-	path_init();
+	path_init("../blender/intern/cycles");
 
 	options_parse(argc, argv);
 
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index ef416dfb8dc..34c92986f46 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -45,7 +45,7 @@ CCL_NAMESPACE_BEGIN
 class OpenCLDevice : public Device
 {
 public:
-	cl_context cxGPUContext;
+	cl_context cxContext;
 	cl_command_queue cqCommandQueue;
 	cl_platform_id cpPlatform;
 	cl_device_id cdDevice;
@@ -121,18 +121,27 @@ public:
 	OpenCLDevice(bool background_)
 	{
 		background = background_;
+		vector<cl_platform_id> platform_ids;
+		cl_uint num_platforms;
 
 		/* setup device */
-		ciErr = clGetPlatformIDs(1, &cpPlatform, NULL);
+		ciErr = clGetPlatformIDs(0, NULL, &num_platforms);
 		opencl_assert(ciErr);
+		assert(num_platforms != 0);
 
-		ciErr = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_CPU, 1, &cdDevice, NULL);
+		platform_ids.resize(num_platforms);
+		ciErr = clGetPlatformIDs(num_platforms, &platform_ids[0], NULL);
 		opencl_assert(ciErr);
 
-		cxGPUContext = clCreateContext(0, 1, &cdDevice, NULL, NULL, &ciErr);
+		cpPlatform = platform_ids[0]; /* todo: pick specified platform && device */
+
+		ciErr = clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_ALL, 1, &cdDevice, NULL);
+		opencl_assert(ciErr);
+
+		cxContext = clCreateContext(0, 1, &cdDevice, NULL, NULL, &ciErr);
 		opencl_assert(ciErr);
 
-		cqCommandQueue = clCreateCommandQueue(cxGPUContext, cdDevice, 0, &ciErr);
+		cqCommandQueue = clCreateCommandQueue(cxContext, cdDevice, 0, &ciErr);
 		opencl_assert(ciErr);
 		
 		/* compile kernel */
@@ -141,15 +150,11 @@ public:
 
 		string build_options = "";
 
-		//string csource = "../blender/intern/cycles";
-		//build_options += "-I " + csource + "/kernel -I " + csource + "/util";
-
-		build_options += " -I " + path_get("kernel"); /* todo: escape path */
-
-		build_options += " -Werror";
-		build_options += " -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END=";
+		string csource = "../blender/intern/cycles";
+		build_options += "-I " + path_get("kernel") + " -I " + path_get("util"); /* todo: escape path */
+		build_options += " -Werror -cl-fast-relaxed-math -cl-strict-aliasing";
 
-		cpProgram = clCreateProgramWithSource(cxGPUContext, 1, (const char **)&source, &source_len, &ciErr);
+		cpProgram = clCreateProgramWithSource(cxContext, 1, (const char **)&source, &source_len, &ciErr);
 
 		opencl_assert(ciErr);
 
@@ -178,7 +183,7 @@ public:
 		ckFilmConvertKernel = clCreateKernel(cpProgram, "kernel_ocl_tonemap", &ciErr);
 		opencl_assert(ciErr);
 
-		null_mem = (device_ptr)clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, 1, NULL, &ciErr);
+		null_mem = (device_ptr)clCreateBuffer(cxContext, CL_MEM_READ_ONLY, 1, NULL, &ciErr);
 	}
 
 	~OpenCLDevice()
@@ -196,7 +201,7 @@ public:
 		clReleaseKernel(ckFilmConvertKernel);  
 		clReleaseProgram(cpProgram);
 		clReleaseCommandQueue(cqCommandQueue);
-		clReleaseContext(cxGPUContext);
+		clReleaseContext(cxContext);
 	}
 
 	string description()
@@ -213,11 +218,11 @@ public:
 		size_t size = mem.memory_size();
 
 		if(type == MEM_READ_ONLY)
-			mem.device_pointer = (device_ptr)clCreateBuffer(cxGPUContext, CL_MEM_READ_ONLY, size, NULL, &ciErr);
+			mem.device_pointer = (device_ptr)clCreateBuffer(cxContext, CL_MEM_READ_ONLY, size, NULL, &ciErr);
 		else if(type == MEM_WRITE_ONLY)
-			mem.device_pointer = (device_ptr)clCreateBuffer(cxGPUContext, CL_MEM_WRITE_ONLY, size, NULL, &ciErr);
+			mem.device_pointer = (device_ptr)clCreateBuffer(cxContext, CL_MEM_WRITE_ONLY, size, NULL, &ciErr);
 		else
-			mem.device_pointer = (device_ptr)clCreateBuffer(cxGPUContext, CL_MEM_READ_WRITE, size, NULL, &ciErr);
+			mem.device_pointer = (device_ptr)clCreateBuffer(cxContext, CL_MEM_READ_WRITE, size, NULL, &ciErr);
 
 		opencl_assert(ciErr);
 	}
@@ -321,7 +326,7 @@ public:
 
 		opencl_assert(ciErr);
 
-		size_t local_size[2] = {1, 1};
+		size_t local_size[2] = {8, 8};
 		size_t global_size[2] = {global_size_round_up(local_size[0], d_w), global_size_round_up(local_size[1], d_h)};
 
 		/* run kernel */
@@ -389,7 +394,7 @@ public:
 
 		opencl_assert(ciErr);
 
-		size_t local_size[2] = {1, 1};
+		size_t local_size[2] = {8, 8};
 		size_t global_size[2] = {global_size_round_up(local_size[0], d_w), global_size_round_up(local_size[1], d_h)};
 
 		/* run kernel */
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index bc1f8bd40a5..b6d758369cc 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -72,6 +72,12 @@ SET(svm_headers
 	svm/volume.h
 	)
 
+SET(util_headers
+	../util/util_color.h
+	../util/util_math.h
+	../util/util_transform.h
+	../util/util_types.h)
+
 # CUDA module
 
 IF("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
@@ -118,11 +124,13 @@ ENDIF()
 # OPENCL kernel
 
 IF(WITH_CYCLES_OPENCL)
-	SET(util_headers
-		../util/util_color.h
-		../util/util_math.h
-		../util/util_transform.h
-		../util/util_types.h)
+	#SET(kernel_preprocessed ${CMAKE_CURRENT_BINARY_DIR}/kernel_preprocessed.cl)
+	#ADD_CUSTOM_COMMAND(
+	#	OUTPUT ${kernel_preprocessed}
+	#	COMMAND gcc -x c++ -E ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cl -I ${CMAKE_CURRENT_SOURCE_DIR}/../util/ -DCCL_NAMESPACE_BEGIN= -DCCL_NAMESPACE_END= -DWITH_OPENCL -o ${kernel_preprocessed}
+	#	DEPENDS ${kernel_sources} ${util_headers})
+	#ADD_CUSTOM_TARGET(cycles_kernel_preprocess ALL DEPENDS ${kernel_preprocessed})
+	#INSTALL(FILES ${kernel_preprocessed} DESTINATION ${CYCLES_INSTALL_PATH}/cycles/kernel)
 
 	INSTALL(FILES kernel.cl ${headers} DESTINATION ${CYCLES_INSTALL_PATH}/cycles/kernel)
 	INSTALL(FILES ${svm_headers} DESTINATION ${CYCLES_INSTALL_PATH}/cycles/kernel/svm)
diff --git a/intern/cycles/kernel/kernel_compat_opencl.h b/intern/cycles/kernel/kernel_compat_opencl.h
index 0127093d8ce..3d493c61fe4 100644
--- a/intern/cycles/kernel/kernel_compat_opencl.h
+++ b/intern/cycles/kernel/kernel_compat_opencl.h
@@ -22,7 +22,10 @@
 #define __KERNEL_GPU__
 #define __KERNEL_OPENCL__
 
-CCL_NAMESPACE_BEGIN
+/* no namespaces in opencl */
+#define CCL_NAMESPACE_BEGIN
+#define CCL_NAMESPACE_END
+#define WITH_OPENCL
 
 /* in opencl all functions are device functions, so leave this empty */
 #define __device
@@ -104,7 +107,5 @@ __device float kernel_tex_interp_(__global float *data, int width, float x)
 
 #include "util_types.h"
 
-CCL_NAMESPACE_END
-
 #endif /* __KERNEL_COMPAT_OPENCL_H__ */
 
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 992a4d95b54..c0cb3fc8a09 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -34,8 +34,8 @@ CCL_NAMESPACE_BEGIN
 #define __BACKGROUND__
 #define __EMISSION__
 #define __CAUSTICS_TRICKS__
-#define __SVM__
 #ifndef __KERNEL_OPENCL__
+#define __SVM__
 #define __TEXTURES__
 #endif
 #define __RAY_DIFFERENTIALS__