From a84f2c9722f4ab1d647df6e4d67121070281e66d Mon Sep 17 00:00:00 2001 From: Siyuan Feng Date: Tue, 21 May 2024 16:34:09 +0800 Subject: [PATCH] [DLight] Update Adreno GEMV Rules When reduction axis is small, it's not necessary to use rfactor. This PR updates the gemv rule to use rfactor only when the reduction axis is large enough. --- python/tvm/dlight/gpu/gemv.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/tvm/dlight/gpu/gemv.py b/python/tvm/dlight/gpu/gemv.py index cbef6235c098..c2de31965afa 100644 --- a/python/tvm/dlight/gpu/gemv.py +++ b/python/tvm/dlight/gpu/gemv.py @@ -706,7 +706,7 @@ def apply( if LOAD_V_SHARED is False: LOAD_V_TILE = 1 - if not isinstance(len_r, int): + if not isinstance(len_r, int) or len_r < LOAD_V_TILE * TR * SCALE_PACK * DEC_PACK: return None if isinstance(len_s, int) and len_s > 32000: