papi_command_line fails with CUDA devices having compute capability >= 7.0

Encountering the following error when trying to use the ‘papi_command_line’ tool with any available CUDA counter:

./papi_command_line cuda:::smsp__warps_active.avg:device=0

…

INTERNAL:papi_internal.c:_papi_hwi_map_events_to_native:1162:81141 EXIT:

API:papi.c:PAPI_add_event:2242:81141 EXIT: return: 0

API:papi.c:PAPI_add_named_event:2444:81141 EXIT: return: 0

API:papi.c:PAPI_start:2653:81141 Entry: EventSet: 0

API:papi.c:PAPI_start:2668:81141 EventSet: 0x4930190

INTERNAL:papi_internal.c:_papi_hwi_get_context:2698:81141 Entry: ESI: 0x4930190, is_dirty: 0x7fffffff8028

SUBSTRATE:components/cuda/linux-cuda.c:_cuda11_start:5161:81141 Entering

THREADS:./threads.h:_papi_hwi_lock:76:81141 Skipped lock 7

SUBSTRATE:components/cuda/linux-cuda.c:_cuda11_build_profiling_structures:4664:81141 error: function (*cuCtxPushCurrentPtr) (mydevice->sessionCtx) failed with error 201.

THREADS:./threads.h:_papi_hwi_unlock:90:81141 Skipped unlock 7

components/cuda/linux-cuda.c:_cuda11_start:5205 _cuda11_build_profiling_structures() failed; err=-14.

SUBSTRATE:components/cuda/linux-cuda.c:_cuda11_stop:5633:81141 Entering

THREADS:./threads.h:_papi_hwi_lock:76:81141 Skipped lock 7

THREADS:./threads.h:_papi_hwi_unlock:90:81141 Skipped unlock 7

API:papi.c:PAPI_start:2810:81141 EXIT: return: -14

Build used in above is based on the latest version of the PAPI repository; Issue can be replicated on devices with compute capability 7.0 and higher (tested with 6.0 [No Error], 7.0 [Error], 8.0 [Error]).

Comments (1)