Patch file for PoCL 1.7 with CUDA 6.5 backend on ARMv7 (Jetson TK1)

pocl-cuda6.5.patch

diff --git a/lib/CL/devices/cuda/pocl-cuda.c b/lib/CL/devices/cuda/pocl-cuda.c
index cd9e044..06a9cde 100644
--- a/lib/CL/devices/cuda/pocl-cuda.c
+++ b/lib/CL/devices/cuda/pocl-cuda.c
@@ -1082,7 +1082,7 @@ pocl_cuda_submit_kernel (CUstream stream, _cl_command_node *cmd,
     }
 
   if (constantMemBytes > constant_mem_size)
-    POCL_ABORT ("[CUDA] Total constant buffer size %u exceeds %lu allocated\n",
+    POCL_ABORT ("[CUDA] Total constant buffer size %u exceeds %u allocated\n",
                 constantMemBytes, constant_mem_size);
 
   unsigned arg_index = meta->num_args;
@@ -1191,9 +1191,9 @@ pocl_cuda_submit_node (_cl_command_node *node, cl_command_queue cq, int locked)
       result = cuMemHostGetDevicePointer (&dev_ext_event_flag,
                                            event_data->ext_event_flag, 0);
       CUDA_CHECK (result, "cuMemHostGetDevicePointer");
-      result = cuStreamWaitValue32 (stream, dev_ext_event_flag, 1,
-                                    CU_STREAM_WAIT_VALUE_GEQ);
-      CUDA_CHECK (result, "cuStreamWaitValue32");
+      //result = cuStreamWaitValue32 (stream, dev_ext_event_flag, 1,
+      //                              CU_STREAM_WAIT_VALUE_GEQ);
+      //CUDA_CHECK (result, "cuStreamWaitValue32");
     }
 
   /* Create and record event for command start if profiling enabled */
@@ -1428,6 +1428,7 @@ pocl_cuda_finalize_command (cl_device_id device, cl_event event)
   cuCtxSetCurrent (((pocl_cuda_device_data_t *)device->data)->context);
   result = cuEventSynchronize (event_data->end);
   CUDA_CHECK (result, "cuEventSynchronize");
+  
 
   /* Clean up mapped memory allocations */
   if (event->command_type == CL_COMMAND_UNMAP_MEM_OBJECT)
@@ -1450,8 +1451,11 @@ pocl_cuda_finalize_command (cl_device_id device, cl_event event)
     {
 #if defined __arm__
       /* On ARM with USE_HOST_PTR, perform explict copies back from device */
-      cl_kernel kernel = event->command.run.kernel;
-      pocl_argument *arguments = event->command.run.arguments;
+      //cl_kernel kernel = event->command.run.kernel;
+      //pocl_argument *arguments = event->command.run.arguments;
+      cl_kernel kernel = event->command->command.run.kernel;
+      pocl_argument *arguments = event->command->command.run.arguments;
+      pocl_kernel_metadata_t *meta = kernel->meta;
       unsigned i;
       for (i = 0; i < meta->num_args; i++)
         {

Idea : pocl/pocl#600