CUDA component: multiple reads results in erroneous values

Issue #105 resolved
john.rodgers created an issue

The CUDA11 portion of the cuda component is returning erroneous values when multiple PAPI_read operations are called.

Sample expected output:

PAPI Started!
Launching CUDA code
After PAPI stop
    cuda:::sm__warps_launched.sum:device=0,1568

Sample erroneous outputs:

PAPI Started!
After first read, no kernel launched
    cuda:::sm__warps_launched.sum:device=0,0
Launching CUDA code
After PAPI stop
    cuda:::sm__warps_launched.sum:device=0,784
PAPI Started!
Launching CUDA code
After first read, kernel launched
    cuda:::sm__warps_launched.sum:device=0,1568
After PAPI stop
    cuda:::sm__warps_launched.sum:device=0,2352
PAPI Started!
After first read, no kernel launched
    cuda:::sm__warps_launched.sum:device=0,0
Launching CUDA code
After second read, kernel launched
    cuda:::sm__warps_launched.sum:device=0,784
After PAPI stop
    cuda:::sm__warps_launched.sum:device=0,1306
PAPI Started!
After first read, no kernel launched
    cuda:::sm__warps_launched.sum:device=0,0
Launching CUDA code
After second read, kernel launched
    cuda:::sm__warps_launched.sum:device=0,784
After third read, no kernel launched
    cuda:::sm__warps_launched.sum:device=0,1306
After forth read, no kernel launched
    cuda:::sm__warps_launched.sum:device=0,1698
After PAPI stop
    cuda:::sm__warps_launched.sum:device=0,2012

Comments (2)

  1. Log in to comment