Skip to content

Conversation

@jmmartinez
Copy link

We add a --offload-arch=... option, that we can use to switch the target architecture. This can be a string such as 'gfx{version}', or 'amdgcnspirv'.

This option defaults to 'gfx90a' such that if it is not specified, we keep the old behavior ('gfx90a' was hardcoded).

If we pass an incorrect option we get:

$ ./hip-tpl.py --offload-arch=aaa
usage: hip-tpl.py [-h] [--offload-arch OFFLOAD_ARCH]
hip-tpl.py: error: argument --offload-arch: invalid amdgpu_target value: 'aaa'

And for the help:

$ ./hip-tpl.py --help
usage: hip-tpl.py [-h] [--offload-arch OFFLOAD_ARCH]

options:
  -h, --help            show this help message and exit
  --offload-arch OFFLOAD_ARCH
                        Offload architecture to be forwarded to AMDGPU_ARCHS during the test suite configuration. e.g. gfx90a, gfx1010, amdgcnspirv

We add a --offload-arch=... option, that we can use to switch the target
architecture. This can be a string such as 'gfx{version}', or
'amdgcnspirv'.

This option defaults to 'gfx90a' such that if it is not specified, we
keep the old behaviour ('gfx90a' was hardcoded).
@jmmartinez jmmartinez requested a review from jplehr December 12, 2025 14:57
@jmmartinez jmmartinez self-assigned this Dec 12, 2025
@jplehr jplehr requested review from Kewen12 and Meinersbur December 15, 2025 14:37
@jmmartinez jmmartinez requested a review from jplehr December 22, 2025 16:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants