PIOpenCL::Buffer::copyTo() with Buffer

libs/main/opencl/piopencl.h

@@ -228,9 +228,11 @@ public:
 		void copyToContainer();
 		void copyTo(void * data);
 		void copyTo(void * data, int elements_count, int elements_offset = 0);
+		void copyTo(Buffer * buffer, int elements_count = -1, int elements_from_offset = 0, int elements_to_offset = 0);
 		void copyFromContainer();
 		void copyFrom(void * data);
 		void copyFrom(void * data, int elements_count, int elements_offset = 0);
+		void copyFrom(Buffer * buffer, int elements_count = -1, int elements_from_offset = 0, int elements_to_offset = 0);
 		uint elementsCount() const { return elements; }
 
 	private:
@@ -244,6 +246,7 @@ public:
 		void zero();
 		bool init();
 		void * containerData();
+		static void copy(Buffer * buffer_from, Buffer * buffer_to, int elements_count, int elements_from_offset, int elements_to_offset);
 		Context * context_;
 		Direction dir;
 		Container type;

libs/opencl/piopencl.cpp

@@ -240,11 +240,17 @@ PIOpenCL::Program * PIOpenCL::Context::createProgram(const PIString & source, co
 		if (error) (*error) = "Empty program!";
 		return 0;
 	}
-	PIString src_text = PRIVATE->complex_src + source;
-	const char * csrc = src_text.dataAscii();
-	size_t src_size   = src_text.size();
-	cl_int ret        = 0;
-	cl_program prog   = clCreateProgramWithSource(PRIVATE->context, 1, &csrc, &src_size, &ret);
+	static PIString double_ext = "#ifdef cl_khr_fp64\n\
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n\
+#elif defined(cl_amd_fp64)\n\
+#pragma OPENCL EXTENSION cl_amd_fp64 : enable\n\
+#else\n\
+#endif\n";
+	PIString src_text          = PRIVATE->complex_src + double_ext + source;
+	const char * csrc          = src_text.dataAscii();
+	size_t src_size            = src_text.size();
+	cl_int ret                 = 0;
+	cl_program prog            = clCreateProgramWithSource(PRIVATE->context, 1, &csrc, &src_size, &ret);
 	if (ret != 0) {
 		piCout << "[PIOpenCL::Context]"
 			   << "clCreateProgramWithSource error" << ret;
@@ -429,6 +435,11 @@ void PIOpenCL::Buffer::copyTo(void * data, int elements_count, int elements_offs
 }
 
 
+void PIOpenCL::Buffer::copyTo(Buffer * buffer, int elements_count, int elements_from_offset, int elements_to_offset) {
+	copy(this, buffer, elements_count, elements_from_offset, elements_to_offset);
+}
+
+
 void PIOpenCL::Buffer::copyFromContainer() {
 	if (!PRIVATE->buffer || !container) return;
 	copyFrom(containerData());
@@ -445,12 +456,12 @@ void PIOpenCL::Buffer::copyFrom(void * data) {
 }
 
 
-void PIOpenCL::Buffer::copyFrom(void * data, int elements_count, int elements_offset) {
+void PIOpenCL::Buffer::copyFrom(void * data, int elements_count, int elements_from_offset) {
 	if (!PRIVATE->buffer) return;
 	cl_int ret = clEnqueueWriteBuffer(context_->PRIVATEWB->queue,
 	                                  PRIVATE->buffer,
 	                                  CL_TRUE,
-	                                  elements_offset * def.size(),
+									  elements_from_offset * def.size(),
 	                                  elements_count * def.size(),
 	                                  data,
 	                                  0,
@@ -463,6 +474,35 @@ void PIOpenCL::Buffer::copyFrom(void * data, int elements_count, int elements_of
 }
 
 
+void PIOpenCL::Buffer::copyFrom(Buffer * buffer, int elements_count, int elements_from_offset, int elements_to_offset) {
+	copy(buffer, this, elements_count, elements_to_offset, elements_from_offset);
+}
+
+
+void PIOpenCL::Buffer::copy(Buffer * buffer_from,
+                            Buffer * buffer_to,
+                            int elements_count,
+                            int elements_from_offset,
+                            int elements_to_offset) {
+    if (!buffer_from || !buffer_to) return;
+    if (!buffer_from->PRIVATEWB->buffer || !buffer_to->PRIVATEWB->buffer) return;
+    if (elements_count < 0) elements_count = piMini(buffer_from->elements, buffer_to->elements);
+    cl_int ret = clEnqueueCopyBuffer(buffer_from->context_->PRIVATEWB->queue,
+                                     buffer_from->PRIVATEWB->buffer,
+                                     buffer_to->PRIVATEWB->buffer,
+                                     elements_from_offset * buffer_from->def.size(),
+                                     elements_to_offset * buffer_to->def.size(),
+                                     elements_count * buffer_from->def.size(),
+                                     0,
+                                     nullptr,
+                                     nullptr);
+    if (ret != 0) {
+        piCout << "[PIOpenCL::Buffer]"
+               << "clEnqueueCopyBuffer error" << ret;
+    }
+}
+
+
 PIOpenCL::Program::Program() {
 	// piCout << "new program" << this;
 	zero();