[BUG]: PTXAS emits advisory regarding cp.async.bulk.*.multicast use on sm_90

[ahendriksen]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Something else

Component

libcu++

Describe the bug

nvbug 4598481

Tests:
cuda/ptx/ptx.cp.async.bulk.tensor.compile.pass.cpp
cuda/ptx/ptx.cp.async.bulk.compile.pass.cpp

1. PTXAS emits an advisory about multicast being advised to be used on sm90a when compiled for sm90.
2. Lit and/or nvcc considers this a warning
3. we compile with warnings as errors
4. tests fail

When compiling with NVRTC, 1, 2, 3, or 4 doesn't trigger so the test doesn't fail.

So we have a test that we cannot make behave consistently on all platforms.

How to Reproduce

Run cuda/ptx/ptx.cp.async.bulk.tensor.compile.pass.cpp with -arch sm_90 flags

Expected behavior

Should either fail on all platforms or not fail.

Best behavior would be to not consider any output from ptxas as a warning or error, but that is not something we can fix here..

Reproduction link

No response

Operating System

No response

nvidia-smi output

No response

NVCC version

No response