Fix starpu_malloc_on_node for OpenCL when not called from the GPU worker

include/starpu_opencl.h

@@ -1,6 +1,6 @@
 /* StarPU --- Runtime system for heterogeneous multicore architectures.
  *
- * Copyright (C) 2010-2013  Université de Bordeaux 1
+ * Copyright (C) 2010-2014  Université de Bordeaux 1
  * Copyright (C) 2010, 2011, 2012, 2013  Centre National de la Recherche Scientifique
  *
  * StarPU is free software; you can redistribute it and/or modify
@@ -76,7 +76,7 @@ int starpu_opencl_collect_stats(cl_event event);
 
 int starpu_opencl_set_kernel_args(cl_int *err, cl_kernel *kernel, ...);
 
-cl_int starpu_opencl_allocate_memory(cl_mem *addr, size_t size, cl_mem_flags flags);
+cl_int starpu_opencl_allocate_memory(int devid, cl_mem *addr, size_t size, cl_mem_flags flags);
 
 cl_int starpu_opencl_copy_ram_to_opencl(void *ptr, unsigned src_node, cl_mem buffer, unsigned dst_node, size_t size, size_t offset, cl_event *event, int *ret);
 

src/datawizard/malloc.c

@@ -50,7 +50,7 @@ struct malloc_pinned_codelet_struct
 //{
 //	struct malloc_pinned_codelet_struct *s = arg;
 //        //        *(s->ptr) = malloc(s->dim);
-//        starpu_opencl_allocate_memory((void **)(s->ptr), s->dim, CL_MEM_READ_WRITE|CL_MEM_ALLOC_HOST_PTR);
+//        starpu_opencl_allocate_memory(devid, (void **)(s->ptr), s->dim, CL_MEM_READ_WRITE|CL_MEM_ALLOC_HOST_PTR);
 //}
 //#endif
 
@@ -434,7 +434,7 @@ _starpu_malloc_on_node(unsigned dst_node, size_t size)
                                 int ret;
 				cl_mem ptr;
 
-				ret = starpu_opencl_allocate_memory(&ptr, size, CL_MEM_READ_WRITE);
+				ret = starpu_opencl_allocate_memory(_starpu_memory_node_get_devid(dst_node), &ptr, size, CL_MEM_READ_WRITE);
 				if (ret)
 				{
 					addr = 0;

src/drivers/opencl/driver_opencl.c

@@ -214,16 +214,15 @@ cl_int _starpu_opencl_deinit_context(int devid)
 }
 #endif
 
-cl_int starpu_opencl_allocate_memory(cl_mem *mem STARPU_ATTRIBUTE_UNUSED, size_t size STARPU_ATTRIBUTE_UNUSED, cl_mem_flags flags STARPU_ATTRIBUTE_UNUSED)
+cl_int starpu_opencl_allocate_memory(int devid, cl_mem *mem STARPU_ATTRIBUTE_UNUSED, size_t size STARPU_ATTRIBUTE_UNUSED, cl_mem_flags flags STARPU_ATTRIBUTE_UNUSED)
 {
 #ifdef STARPU_SIMGRID
 	STARPU_ABORT();
 #else
 	cl_int err;
         cl_mem memory;
-        struct _starpu_worker *worker = _starpu_get_local_worker_key();
 
-	memory = clCreateBuffer(contexts[worker->devid], flags, size, NULL, &err);
+	memory = clCreateBuffer(contexts[devid], flags, size, NULL, &err);
 	if (err == CL_OUT_OF_HOST_MEMORY) return err;
         if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
 
@@ -234,7 +233,7 @@ cl_int starpu_opencl_allocate_memory(cl_mem *mem STARPU_ATTRIBUTE_UNUSED, size_t
 	 */
 	char dummy = 0;
 	cl_event ev;
-	err = clEnqueueWriteBuffer(alloc_queues[worker->devid], memory, CL_TRUE,
+	err = clEnqueueWriteBuffer(alloc_queues[devid], memory, CL_TRUE,
 				   0, sizeof(dummy), &dummy,
 				   0, NULL, &ev);
 	if (err == CL_MEM_OBJECT_ALLOCATION_FAILURE)