-
Notifications
You must be signed in to change notification settings - Fork 787
[SYCL] Remove esimd_emulator from filters #12428
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
This PR separates out the removal of esimd_emulator from PR##12384 |
sycl/source/detail/config.cpp
Outdated
{{"host", backend::host}, | ||
{"opencl", backend::opencl}, | ||
{"level_zero", backend::ext_oneapi_level_zero}, | ||
{"cuda", backend::ext_oneapi_cuda}, | ||
{"hip", backend::ext_oneapi_hip}, | ||
{"esimd_emulator", backend::ext_intel_esimd_emulator}, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this might be an ABI break and thus unacceptable.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
esimd_emulator is not functional starting from 2024.0 - we have removed all the code for emulator, maybe except few places like this one.
It is not ABI break IMO, if it was we would have fails in those special tests that verify presence of global symbols.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If return type was part of the mangling, the symbol name would have changed (and that would definitely be an ABI break). I'm not sure what would be the implications of a code compiled expecting 8-element std::array
and only looking into a 7-element array.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@aelovikov-intel I went through all the code usages of getSyclBeMap()
in DPC++ but did not found any usage that implicitly requires the SYCL backend std::array
to have 8-elements. In all the cases, the returned array is traversed using iterators with no hard-coded requirement on its size.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Now that I rethink again, I see your point. Some pre-compiled binaries using libsycl might expect the return value of getSyclBeMap to have 8 elements. So, @aelovikov-intel what do you suggest, should we table this change until the next ABI breaking window?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It depends on the usage of this array. Maybe we can just keep this line and apply the rest of the PR. Still, it's better to discuss that with @v-klochkov and @turinevgeny
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This array is not ESIMD specific. If usage of it creates problems, then we can wait till the next ABI breaking window.
I don't see any urgency in removing the enum now.
@v-klochkov I've reverted the ABI-breaking change for the time being. If rest of the changes looks good, could you please approve this? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If it's good with ESIMD folks, then it's good with me as well.
This PR removes esimd_emulator from device filters and filter selectors. Esimd_emulator is already depreciated.