Closed LeiWang1999 closed 3 days ago
Thank you for all the incredible work on this amazing project!
I was surprised to discover that Triton GEMM on ROCm is also conflict-free. Did we use padding or XOR-based swizzling to avoid bank conflicts?
I’m super curious how this solution is implemented! Where can I find more details on this? Thanks in advance!
Ubuntu 22.04
AMD
MI300X
ROCm 6.2.2
No response
I've found related codes, thanks for your awesome solution, and also some discussions at #635.
thanks @zhanglx13 for your patient response :)
Problem Description
Thank you for all the incredible work on this amazing project!
I was surprised to discover that Triton GEMM on ROCm is also conflict-free. Did we use padding or XOR-based swizzling to avoid bank conflicts?
I’m super curious how this solution is implemented! Where can I find more details on this? Thanks in advance!
Operating System
Ubuntu 22.04
CPU
AMD
GPU
MI300X
ROCm Version
ROCm 6.2.2
ROCm Component
No response
Steps to Reproduce
No response
(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
No response
Additional Information
No response