-
Notifications
You must be signed in to change notification settings - Fork 734
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
[SYCL] Stop emitting kernel arguments metadata #2348
Conversation
OpenCL kernel arguments metadata - kernel_arg_addr_space, kernel_arg_access_qual, kernel_arg_type, kernel_arg_base_type, kernel_arg_type_qual and kernel_arg_name - are not needed for SYCL. But this causes optimizations to unnecessarily handle them, so stop emitting them. Signed-off-by: Premanand M Rao <premanand.m.rao@intel.com>
If this could be reviewed asap, I would appreciate it as there is some urgency to fix this very soon. Thanks! |
Looks ok to me but it looks like there are several lit fails with check-sycl |
Seems to me this metadata can be important for underlying OpenCL backends. We probably need to update it in DAE pass instead. |
/summary:run |
@elizabethandrews, @Fznamznon, check-sycl failures are fixed. |
For some reason ESIMD mode uses OpenCL metadata for CM back-end. @kbobrovs, @dendibakh, could you clarify what these metadata are used for? |
Thanks @bader, @Fznamznon and @elizabethandrews ! |
I'm not sure. @cmc-rep, do you know why we need kernel_arg* metadata in ESIMD mode? |
I am not sure either. @kbobrovs, please comment on this |
@premanandrao With this change we stopped emitting buffer_location metadata for ESIMD mode. Was it intended? |
Yes. AFAIK, ESIMD is Intel GPU specific feature, whereas |
Got it, thanks. |
Small fix but yields around 30% speedup for translation SPIR-V to IR. Original commit: KhronosGroup/SPIRV-LLVM-Translator@513b9578d310282
OpenCL kernel arguments metadata - kernel_arg_addr_space,
kernel_arg_access_qual, kernel_arg_type, kernel_arg_base_type,
kernel_arg_type_qual and kernel_arg_name - are not needed
for SYCL. But this causes optimizations to unnecessarily
handle them, so stop emitting them.
Signed-off-by: Premanand M Rao premanand.m.rao@intel.com