Skip to content
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

Jump to bottom

[SYCL] Add vec<bfloat16> support to math builtins #14002

Merged
merged 16 commits into from
Jun 7, 2024
Merged

[SYCL] Add vec<bfloat16> support to math builtins #14002

merged 16 commits into from
Jun 7, 2024

Conversation

uditagarwal97
Copy link
Contributor

@uditagarwal97 uditagarwal97 commented Jun 1, 2024 •
edited
Loading

This PR

  1. Updates extension to add math builtins for vec<bfloat16> and corresponding swizzles.
  2. Implements unoptimized support for vec<bloat16>/swizzles in math builtins and adds a test case for the same.
  3. Adds a test to check the device code generated for vec<bloat16> math builtins.

I will make a follow-up PR to optimize vec<bfloat16> math builtins. I think we can use elementwise builtins for ext_vector_type (https://clang.llvm.org/docs/LanguageExtensions.html#vectors-and-extended-vectors) to optimize vec<bfloat16> math builtins. The device code test case will help visualizing/reviewing math builtin optimizations.

@uditagarwal97 uditagarwal97 self-assigned this Jun 1, 2024
@uditagarwal97 uditagarwal97 marked this pull request as ready for review June 1, 2024 05:47
@uditagarwal97 uditagarwal97 requested review from a team as code owners June 1, 2024 05:47
@uditagarwal97
Update sycl_ext_oneapi_bfloat16_math_functions.asciidoc
0e0a740 
to clarify vec/swizzle overloads of isnan, fma, and fmax
@uditagarwal97
Update sycl_ext_oneapi_bfloat16_math_functions.asciidoc
7df752a 
by clarifying log/expr/rint/trunc overloads for vec<bfloat16>
cperkinsintel
cperkinsintel reviewed Jun 5, 2024
View reviewed changes
sycl/doc/extensions/experimental/sycl_ext_oneapi_bfloat16_math_functions.asciidoc Outdated
} // namespace sycl::ext::oneapi::experimental
```

===== Description
====== Overloads (1) and (2)
Copy link
Contributor

Choose a reason for hiding this comment

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

thanks for these numbered descriptions.

asudarsa
asudarsa approved these changes Jun 5, 2024
View reviewed changes
Copy link
Contributor

@asudarsa asudarsa left a comment

Choose a reason for hiding this comment

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

Changes to BFloat16 test looks good. Thanks

@uditagarwal97
Update sycl_ext_oneapi_bfloat16_math_functions.asciidoc
5f47ea7 
to align bfloat16 builtins documentation with that of SYCL
@uditagarwal97
Copy link
Contributor Author

Changes to BFloat16 test looks good. Thanks

Thanks. I've augmented the test case to also add the tests for builtin overloads accepting swizzles.

asudarsa
asudarsa approved these changes Jun 6, 2024
View reviewed changes
Copy link
Contributor

@asudarsa asudarsa left a comment

Choose a reason for hiding this comment

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

Looks good from the DPC++ Tools 9offload utilities) perspective

Thanks

@uditagarwal97
Copy link
Contributor Author

Linux Gen12 failures are unrelated:

Failed Tests (2):
  SYCL :: Basic/submit_time.cpp
  SYCL :: syclcompat/memory/memory_management_test3.cpp

Basic/submit_time.cpp seems like a flaky failure. Also, reported in #13591 (comment)
syclcompat/memory/memory_management_test3.cpp is reported in #14086

@intel/llvm-gatekeepers the PR is ready to be merged!

@steffenlarsen steffenlarsen merged commit b925bd8 into intel:sycl Jun 7, 2024
14 of 15 checks passed
ianayl pushed a commit to ianayl/sycl that referenced this pull request Jun 13, 2024
@uditagarwal97 @ianayl
[SYCL] Add vec<bfloat16> support to math builtins (intel#14002)
f7653cb 
This PR
1. Updates extension to add math builtins for `vec<bfloat16>` and
corresponding swizzles.
2. Implements **unoptimized** support for `vec<bloat16>`/swizzles in
math builtins and adds a test case for the same.
3. Adds a test to check the device code generated for `vec<bloat16>`
math builtins.

I will make a follow-up PR to optimize `vec<bfloat16>` math builtins. I
think we can use elementwise builtins for `ext_vector_type`
(https://clang.llvm.org/docs/LanguageExtensions.html#vectors-and-extended-vectors)
to optimize `vec<bfloat16>` math builtins. The device code test case
will help visualizing/reviewing math builtin optimizations.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@cperkinsintel cperkinsintel cperkinsintel approved these changes

@gmlueck gmlueck gmlueck approved these changes

@asudarsa asudarsa asudarsa approved these changes

Assignees

@uditagarwal97 uditagarwal97

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@uditagarwal97 @asudarsa @cperkinsintel @gmlueck @steffenlarsen