Repository navigation
CUDA Graph phase 3 - memcpy nodes #1314
Description
Cover cuMemcpyAsync and cover cuMemsetAsync.
cuMemsetAsync has multiple bit-width APIs. We can wrap them nicely from the python side so the user doesn't need to worry about what bit-width variant they need to call.
Activity
[-]CUDA Graph - memcpy nodes[/-][+]CUDA Graph phase 3 - memcpy nodes[/+]Andy-Jost commentedon Dec 4, 2025
Is
cuMempyAsynccovered byBuffer.copy_toandBuffer.copy_fromor is something special needed for graphs?Could we similarly cover
cuMemsetAsyncby adding a method toBuffer?rparolin commentedon Dec 4, 2025
Andy-Jost commentedon Dec 4, 2025
Ok, follow-up question:
How do we envision this working? If someone says
buf.fill(1)we still don't have information about the intended bit-width. Are we thinking something likebuf.fill(value=1, width=4)?If the type is a specific NumPy dtype we can do it automatically. Is there a precedent for anything like that, or do we have a list of frameworks/dtypes that we support?
Add Buffer.fill() method for cuMemsetAsync support
Andy-Jost commentedon Dec 4, 2025
Please see the attached draft PR. Is that what we have in mind for
Buffer.fill?Add graph capture tests for Buffer.fill()
leofang commentedon Dec 8, 2025
Sorry for late reply. Continuing the discussion in your PR 🙂 #1318 (comment)
Add Buffer.fill() method for cuMemsetAsync support (#1314) (#1318)
leofang commentedon Dec 15, 2025
@Andy-Jost can correct me if I am wrong, but I think this is done.