[tvm] branch main updated: [bugfix] Fix the write buffer scope of `mma_store_impl` (#14174)

[ma...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

masahi 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 bd8e7d3bd3 [bugfix] Fix the write buffer scope of `mma_store_impl` (#14174)
bd8e7d3bd3 is described below

commit bd8e7d3bd3b03185a537513ea94157c10f42ab82
Author: Zihao Ye <ex...@outlook.com>
AuthorDate: Fri Mar 3 03:59:14 2023 +0800

    [bugfix] Fix the write buffer scope of `mma_store_impl` (#14174)
    
    fix
---
 python/tvm/tir/tensor_intrin/cuda.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/tvm/tir/tensor_intrin/cuda.py b/python/tvm/tir/tensor_intrin/cuda.py
index 6483b99454..da194f885d 100644
--- a/python/tvm/tir/tensor_intrin/cuda.py
+++ b/python/tvm/tir/tensor_intrin/cuda.py
@@ -392,7 +392,7 @@ def get_mma_store_intrin(dtype, local_size, scope="global"):
             a, [WARP_SIZE, local_size], dtype=dtype, scope="warp", offset_factor=1
         )
         C = T.match_buffer(
-            c, [M_DIM, N_DIM], dtype=dtype, scope="global", offset_factor=1, strides=[s0, s1]
+            c, [M_DIM, N_DIM], dtype=dtype, scope=scope, offset_factor=1, strides=[s0, s1]
         )
 
         with T.block("root"):