Hi, Cuda driver api function cuGetErrorString is available in version 6.0 and higher.
Currently, when the driver that is used does not contain this function, the libgomp nvptx plugin will not build (PLUGIN_NVPTX_DYNAMIC == 0) or run (PLUGIN_NVPTX_DYNAMIC == 1). This patch fixes this problem by testing for the presence of the function, and handling absence. Build on x86_64 with nvptx accelerator and reg-tested libgomp, both with and without --without-cuda-driver. Committed to trunk. Thanks, - Tom [libgomp, nvptx] Allow cuGetErrorString to be NULL 2018-08-08 Tom de Vries <tdevr...@suse.de> * plugin/cuda-lib.def (cuGetErrorString): Use CUDA_ONE_CALL_MAYBE_NULL. * plugin/plugin-nvptx.c (cuda_error): Handle if cuGetErrorString is not present. --- libgomp/plugin/cuda-lib.def | 2 +- libgomp/plugin/plugin-nvptx.c | 10 +++++++--- 2 files changed, 8 insertions(+), 4 deletions(-) diff --git a/libgomp/plugin/cuda-lib.def b/libgomp/plugin/cuda-lib.def index be8e3b3ec4d6..6365cdbfcbe4 100644 --- a/libgomp/plugin/cuda-lib.def +++ b/libgomp/plugin/cuda-lib.def @@ -15,7 +15,7 @@ CUDA_ONE_CALL (cuEventQuery) CUDA_ONE_CALL (cuEventRecord) CUDA_ONE_CALL (cuEventSynchronize) CUDA_ONE_CALL (cuFuncGetAttribute) -CUDA_ONE_CALL (cuGetErrorString) +CUDA_ONE_CALL_MAYBE_NULL (cuGetErrorString) CUDA_ONE_CALL (cuInit) CUDA_ONE_CALL (cuLaunchKernel) CUDA_ONE_CALL (cuLinkAddData) diff --git a/libgomp/plugin/plugin-nvptx.c b/libgomp/plugin/plugin-nvptx.c index 589d6596cc2f..b549b7740039 100644 --- a/libgomp/plugin/plugin-nvptx.c +++ b/libgomp/plugin/plugin-nvptx.c @@ -161,13 +161,17 @@ init_cuda_lib (void) static const char * cuda_error (CUresult r) { + const char *fallback = "unknown cuda error"; const char *desc; + if (!CUDA_CALL_EXISTS (cuGetErrorString)) + return fallback; + r = CUDA_CALL_NOCHECK (cuGetErrorString, r, &desc); - if (r != CUDA_SUCCESS) - desc = "unknown cuda error"; + if (r == CUDA_SUCCESS) + return desc; - return desc; + return fallback; } static unsigned int instantiated_devices = 0;