Skip to content

Commit d8e3e1a

Browse files
committed
[3rdparty] Bump cutlass_fpA_intB_gemm to fix SM90 build (#18291)
This PR fixes a SM90 build issue when CUTLASS is enabled. The issue is because a source file indluced a CUTLASS header file that has been removed since CUTLASS 4. Simply removing the header fixes the build issue.
1 parent 3900556 commit d8e3e1a

File tree

2 files changed

+10
-5
lines changed

2 files changed

+10
-5
lines changed

tests/python/codegen/test_gpu_codegen_allreduce.py

Lines changed: 9 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,13 +14,14 @@
1414
# KIND, either express or implied. See the License for the
1515
# specific language governing permissions and limitations
1616
# under the License.
17+
import numpy as np
18+
import pytest
19+
import tvm_ffi
20+
1721
import tvm
1822
import tvm.testing
19-
import numpy as np
2023
from tvm.script import tir as T
2124

22-
import pytest
23-
2425

2526
@T.prim_func
2627
def reduce(a: T.handle, b: T.handle, d1: T.int32, d2: T.int32, d3: T.int32) -> None:
@@ -96,7 +97,11 @@ def optional_metal_compile_callback(define_metal_compile_callback):
9697

9798
@tvm.register_global_func(name, override=True)
9899
def compile_metal(src, target):
99-
return tvm.contrib.xcode.compile_metal(src, sdk="macosx")
100+
from tvm.contrib.xcode import ( # pylint: disable=import-outside-toplevel
101+
compile_metal,
102+
)
103+
104+
return compile_metal(src, sdk="macosx")
100105

101106
yield
102107

0 commit comments

Comments
 (0)