https://github.com/jundaf2/cutlass-kernel-volta-gemm

volta fp16 gemm kernel
https://github.com/jundaf2/cutlass-kernel-volta-gemm

Last synced: 2 days ago
JSON representation

volta fp16 gemm kernel

• Host: GitHub
• URL: https://github.com/jundaf2/cutlass-kernel-volta-gemm
• Owner: jundaf2
• Created: 2023-09-04T03:44:58.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-10-01T10:13:35.000Z (over 1 year ago)
• Last Synced: 2024-11-15T13:33:06.074Z (2 months ago)
• Language: Cuda
• Homepage:
• Size: 161 KB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# cutlass-kernel-volta-gemm

This is a cutlass-based kernel-level GEMM for Volta architecture.

## Dependencies

- pytorch

- pytest

## Notes

First deploy blocks, then deploy warps.

- Block

    - BM = 64

    - BN = 64

- Warp 

    - WM = 16

    - WN = 16

NUM_WARPS = (BMxBN)/(WMxWN) = 16

NUM_THREADS_PER_CTA = WARP_SIZExNUM_WARPS = 32x16 = 512