Closed tgrogers closed 2 months ago
I built, traced, and simulated cutlass 3 with the QV100-SASS config on gpu-app-collection commit 051a445 for the first 50 kernels. We can trace more of the kernels if we clear out space for the traces on the shared drive
Update required for Cutlass version from 2 to 3. After the update, the cutlass version 3 kernels should be traced and executed with accel-sim.
More information on changes in cutlass 3 are found here: https://github.com/NVIDIA/cutlass