youngkingdom/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v4.2.1 update. (#2667)

Browse Source
This commit is contained in:
Junkai-Wu
2025-09-24 02:25:14 +08:00
committed by GitHub
parent 59b61c606f
commit ee914c3cec
12 changed files with 163 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/CuTeDSL/base_dsl/runtime/cuda.py
View File

@ -255,7 +255,13 @@ def initialize_cuda_context(device_id: int = 0, flags: int = 0):
_log().info(f"{cuDevice} <-- cuDeviceGet")
# Create context
_log().info(f"cuCtxCreate {0} {cuDevice}")
context = checkCudaErrors(cuda.cuCtxCreate(0, cuDevice))
if cuda.CUDA_VERSION >= 13000:
# Use cuCtxCreate_v4 API with explicit CUctxCreateParams None, since v2
# and v3 API has been removed from CTK 13.
# See https://github.com/NVIDIA/cuda-python/pull/792
context = checkCudaErrors(cuda.cuCtxCreate(None, 0, cuDevice))
else:
context = checkCudaErrors(cuda.cuCtxCreate(0, cuDevice))
_log().info(f"{context} <-- cuCtxCreate")
return context