[ROCm] Add MLA dual RMS norm fusion (Q, KV) pass for DeepSeek/Kimi-K2 (#39242)

Signed-off-by: Rita Brugarolas Brufau <rita.brugarolasbrufau@amd.com>
This commit is contained in:
Rita Brugarolas
2026-04-20 14:56:27 +00:00
committed by GitHub
parent b42e878ec0
commit fb5635d3f9
8 changed files with 361 additions and 1 deletions
+1
View File
@@ -56,6 +56,7 @@ Fusions:
- `-cc.pass_config.fuse_norm_quant=True`*
- `-cc.pass_config.fuse_act_quant=True`*
- `-cc.pass_config.fuse_act_padding=True`†
- `-cc.pass_config.fuse_mla_dual_rms_norm=True`†
\* These fusions are only enabled when either op is using a custom kernel, otherwise Inductor fusion is better.</br>
† These fusions are ROCm-only and require AITER.