NVIDIA/cutlass
1
0
Fork 0
mirror of https://github.com/NVIDIA/cutlass.git synced 2026-04-20 14:59:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Cute-DSL] Add option for issue_clc_query without multicast

Browse Source
This commit is contained in:
Tri Dao
2026-02-11 14:19:29 -05:00
parent 8dbce01473
commit 5b83b34afd
1 changed files with 15 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
python/CuTeDSL/cutlass/cute/arch/clc.py
View File

@@ -23,6 +23,7 @@ from ..typing import Int32, Pointer, Int128
def issue_clc_query(
mbar_ptr: Pointer,
clc_response_ptr: Pointer,
multicast: bool = True,
loc=None,
ip=None,
) -> None:
@@ -39,12 +40,20 @@ def issue_clc_query(
"""
mbar_llvm_ptr = mbar_ptr.llvm_ptr
clc_response_llvm_ptr = clc_response_ptr.llvm_ptr
nvvm.clusterlaunchcontrol_try_cancel_multicast(
clc_response_llvm_ptr,
mbar_llvm_ptr,
loc=loc,
ip=ip,
)
if multicast:
nvvm.clusterlaunchcontrol_try_cancel_multicast(
clc_response_llvm_ptr,
mbar_llvm_ptr,
loc=loc,
ip=ip,
)
else:
nvvm.clusterlaunchcontrol_try_cancel(
clc_response_llvm_ptr,
mbar_llvm_ptr,
loc=loc,
ip=ip,
)
@dsl_user_op