github/ROCm
1
1
Fork 0
mirror of https://github.com/ROCm/ROCm.git synced 2026-04-05 03:01:17 -04:00
Code Issues Packages Projects Releases Wiki Activity

[HOPPER][OPTIMIZER] remove divOp and remOp from gemm math loop (#2402)

Browse Source 
This is just for Warp Specialization kernels on Hopper. Replace DivOp
and RemOp with SelectOp and AndOp/XorOp.
This commit is contained in:
Beal Wang
2023-10-09 14:42:06 +08:00
committed by GitHub
parent 424e67e727
commit 5812d970a8
6 changed files with 550 additions and 431 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
python/triton/compiler/compiler.py
View File

@@ -130,6 +130,7 @@ def optimize_ttgir(mod, num_stages, num_warps, num_ctas, target,
pm.add_tritongpu_wspipeline_pass(num_stages, num_warps, capability)
pm.add_tritongpu_wsmutex_pass(capability)
pm.add_tritongpu_wsmaterialization_pass(capability)
pm.add_licm_pass()
pm.add_cse_pass()
else:
pm.add_tritongpu_pipeline_pass(num_stages, num_warps, num_ctas, capability)