port r11429 from 1.1: Enable gpu-gpu direct transfer. This does not seem to harm performance any more

doc/doxygen/chapters/environment_variables.doxy

@@ -257,7 +257,8 @@ Disable asynchronous copies between CPU and MIC devices.
 <dd>
 \anchor STARPU_ENABLE_CUDA_GPU_GPU_DIRECT
 \addindex __env__STARPU_ENABLE_CUDA_GPU_GPU_DIRECT
-Enable direct CUDA transfers from GPU to GPU, without copying through RAM.
+Enable (1) or Disable (0) direct CUDA transfers from GPU to GPU, without copying
+through RAM. The default is Enabled.
 This permits to test the performance effect of GPU-Direct.
 </dd>
 

src/core/perfmodel/perfmodel_bus.c

@@ -238,7 +238,7 @@ static void measure_bandwidth_between_dev_and_dev_cuda(int src, int dst)
 	 * since we cleanly shutdown CUDA before returning. */
 	cudaSetDevice(src);
 
-	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") > 0)
+	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") != 0)
 	{
 		cures = cudaDeviceCanAccessPeer(&can, src, dst);
 		if (!cures && can)
@@ -260,7 +260,7 @@ static void measure_bandwidth_between_dev_and_dev_cuda(int src, int dst)
 	 * since we cleanly shutdown CUDA before returning. */
 	cudaSetDevice(dst);
 
-	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") > 0)
+	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") != 0)
 	{
 		cures = cudaDeviceCanAccessPeer(&can, dst, src);
 		if (!cures && can)

src/drivers/cuda/driver_cuda.c

@@ -201,7 +201,7 @@ static void init_context(unsigned devid)
 	starpu_cuda_set_device(devid);
 
 #ifdef HAVE_CUDA_MEMCPY_PEER
-	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") > 0)
+	if (starpu_get_env_number("STARPU_ENABLE_CUDA_GPU_GPU_DIRECT") != 0)
 	{
 		int nworkers = starpu_worker_get_count();
 		for (workerid = 0; workerid < nworkers; workerid++)