[tvm] branch unity updated: [Unity][Dlight] Matmul tensorization for SM70 (#15503)

[sanirudh]

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

sanirudh pushed a commit to branch unity
in repository https://gitbox.apache.org/repos/asf/tvm.git


The following commit(s) were added to refs/heads/unity by this push:
     new c97753ea7c [Unity][Dlight] Matmul tensorization for SM70 (#15503)
c97753ea7c is described below

commit c97753ea7c999f21df45eff138b053794b06adf6
Author: Ruihang Lai <ruiha...@cs.cmu.edu>
AuthorDate: Tue Aug 8 01:37:22 2023 -0700

    [Unity][Dlight] Matmul tensorization for SM70 (#15503)
    
    Prior to this PR, TensorCore tensorization is not enabled for SM70
    while SM70 devices have TensorCore support. This PR enables the
    tensorization.
---
 python/tvm/dlight/gpu/matmul.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/tvm/dlight/gpu/matmul.py b/python/tvm/dlight/gpu/matmul.py
index 5a179c485c..b219762fc2 100644
--- a/python/tvm/dlight/gpu/matmul.py
+++ b/python/tvm/dlight/gpu/matmul.py
@@ -544,7 +544,7 @@ class Matmul(ScheduleRule):
         sch.transform_block_layout(main_block, matmul_index_map)
 
         block_stmt = sch.get(main_block)
-        if target.kind.name == "cuda" and check_sm_version(target.arch) > 70:
+        if target.kind.name == "cuda" and check_sm_version(target.arch) >= 70:
             apply_tensorization: bool = True
             # the batch dimension is not taken into consideration.
             for item_var in block_stmt.iter_vars[1:]: