Skip to content

[SYCL][DOC] Document -foffload-fp32-prec-div/sqrt options #17033

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.

Sign up for GitHub

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

Merged
aelovikov-intel merged 5 commits into from
Feb 26, 2025
Merged

[SYCL][DOC] Document -foffload-fp32-prec-div/sqrt options #17033

Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
95b399c
[SYCL][DOC] Document -foffload-fp32-prec-div/sqrt options
Feb 17, 2025
18c7cc9
[SYCL] Add E2E test for -foffload-fp32-prec-div/sqrt
Feb 17, 2025
3388045
Revert "[SYCL] Add E2E test for -foffload-fp32-prec-div/sqrt"
Feb 17, 2025
97248eb
add deprecation note
Feb 26, 2025
065d2e3
fix deprecation note
Feb 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions sycl/doc/UsersManual.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -350,6 +350,18 @@ and not recommended to use in production environment.

## Other options

**`-f[no-]offload-fp32-prec-sqrt`**

@MrSidims MrSidims Feb 17, 2025

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@gmlueck during writing this document I've realized, that there is a similar option already, that works only for CUDA and HIP. I believe, that these options should be merged under sycl-fp32-prec-sqrt name and div option should be renamed to sycl-fp32-prec-div (unless we want to deprecate sycl naming for them, which also might make sense in a certain cases). WDYT?

@npmiller npmiller Feb 17, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have no strong feelings on which option should remain (offload or sycl), but I do agree that we should merge these.


Enable use of correctly rounded `sycl::sqrt` function as defined by IEE754.
Without this flag, the default precision requirement for `sycl::sqrt` is 3
ULP.

**`-f[no-]offload-fp32-prec-div`**

Enable use of correctly rounded divide operation as defined by IEE754.
Without this flag, the default precision requirement for divide in SYCL is
2.5 ULP.

**`-fsycl-device-only`**

Compile only device part of the code and ignore host part.
Expand Down