I’m trying to get metrics on PolyBench-ACC cuda codes (GitHub - cavazos-lab/PolyBench-ACC) for a start I tried 2mm.cu file to get metrics, I get an error:
seafortune@precison3561:~/Documents/PolyBench-ACC-master/CUDA/linear-algebra/kernels/2mm$ ncu --set default --target-processes all --metrics sol_sm,sol_mem,sol_L1TexCacheSOL,sol_L2Cache,sol_Dram,duration,elapsed_cycle,ipc,sm_busy,mem_thput,lTexRate,l2HitRate,memBusy,maxBand,activeWarpSch,warpCycInst,execInst,regPerThread,achievedOccupancy ./2mm
==PROF== Connected to process 92769 (/home/seafortune/Documents/PolyBench-ACC-master/CUDA/linear-algebra/kernels/2mm/2mm)
setting device 0 with name NVIDIA T600 Laptop GPU
==ERROR== Failed to find metric achievedOccupancy
==ERROR== Failed to profile “mm2_kernel1” in process 92769
==PROF== Trying to shutdown target application
==ERROR== The application returned an error code (9).
==ERROR== An error occurred while trying to profile.
==WARNING== No kernels were profiled.
Deleting the achivedOcupucy doesn’t work, this time it gives the error that it could not find the metric before it.
I tried diffrent cuda file from cuda samples and it erros again:
seafortune@precison3561:~/Documents/cuda-samples-master/Samples/0_Introduction/matrixMul$ ncu --target-processes all --metrics sm_efficiency --csv ./matrixMul
[Matrix Multiply Using CUDA] - Starting…
==PROF== Connected to process 97694 (/home/seafortune/Documents/cuda-samples-master/Samples/0_Introduction/matrixMul/matrixMul)
GPU Device 0: “Turing” with compute capability 7.5
MatrixA(320,320), MatrixB(640,320)
Computing result using CUDA Kernel…
==ERROR== Failed to find metric sm_efficiency
==ERROR== Failed to profile “MatrixMulCUDA” in process 97694
==PROF== Trying to shutdown target application
==ERROR== The application returned an error code (9).
==ERROR== An error occurred while trying to profile.
==WARNING== No kernels were profiled.
I use latest cuda toolkit and nsight compute