This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/main by this push:
new 2e56421dda [DLight] Update Adreno GEMV Rules (#17016)
2e56421dda is described below
commit 2e56421dda32755a0b9c41cd1515ec4f8e4d598e
Author: Siyuan Feng <[email protected]>
AuthorDate: Tue May 21 22:59:36 2024 +0800
[DLight] Update Adreno GEMV Rules (#17016)
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 da6a4ef834..b8a2c6a15f 100644
--- a/python/tvm/dlight/gpu/gemv.py
+++ b/python/tvm/dlight/gpu/gemv.py
@@ -711,7 +711,7 @@ class GEMV(GPUScheduleRule):
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 not isinstance(len_s, int):
