Skip to content

Conversation

steffenlarsen
Copy link
Contributor

Previously, the implementation would prioritize native device code binaries over IR modules indiscriminately. However, for input state and object state, native device code images are unsuitable. As such, this commit makes it so that the native device code images are only selected when the target state is executable.

As an effect of this change, input and object tests will no longer run for CUDA and HIP. This is appropriate given these targets only currently support generating native device code images, which are meant to be as close to executable state as possible. Note that the driver doesn't currently make such limitations of the generated SYCLBIN files, so producing input and object state binaries with CUDA and HIP targets will result in kernel bundles with no binary images.

Previously, the implementation would prioritize native device code
binaries over IR modules indiscriminately. However, for input state and
object state, native device code images are unsuitable. As such, this
commit makes it so that the native device code images are only selected
when the target state is executable.

As an effect of this change, input and object tests will no longer run
for CUDA and HIP. This is appropriate given these targets only currently
support generating native device code images, which are meant to be as
close to executable state as possible. Note that the driver doesn't
currently make such limitations of the generated SYCLBIN files, so
producing input and object state binaries with CUDA and HIP targets will
result in kernel bundles with no binary images.

Signed-off-by: Larsen, Steffen <[email protected]>
@steffenlarsen
Copy link
Contributor Author

Prereq to #20174

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.

2 participants