[AMD][gfx95] DSV4 wo_b (dp-attention): route to tuned bpreshuffle GEMM instead of triton (#38227)

Co-authored-by: lixiufei-leo <lixiufei-leo@users.noreply.github.com>
This commit is contained in:
lixiufei-leo
2026-09-07 00:36:15 -07:00
committed by GitHub
co-authored by lixiufei-leo
parent 644841c50c
commit a8edafff7c
@@ -183,7 +183,6 @@ def use_aiter_triton_gemm_w8a8_tuned_gfx950(n: int, k: int) -> bool:
(512, 7168),
(7168, 2048),
(7168, 2304),
(7168, 16384),
(7168, 256),
(8192, 1024),
(8192, 32768),