Skip to content

sycl: add ROLL operation support#16665

Merged
NeoZhangJianyu merged 6 commits into
ggml-org:masterfrom
tamarPal:feature/sycl-roll-sync
Oct 27, 2025
Merged

sycl: add ROLL operation support#16665
NeoZhangJianyu merged 6 commits into
ggml-org:masterfrom
tamarPal:feature/sycl-roll-sync

Conversation

@tamarPal

@tamarPal tamarPal commented Oct 19, 2025 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary

Implements the ROLL operator for the SYCL backend, enabling multi-dimensional tensor rolling on SYCL devices (Intel GPUs). Provides efficient circular shifts along tensor axes. The changes are focused and follow SYCL patterns.

Changes

  • Added ROLL kernel in roll.cpp with multi-axis support
  • ROLL dispatch in ggml-sycl.cpp
  • Integrated into the SYCL compute pipeline

Implementation

  • 4D tensor support: per-axis shifts with normalization
  • Parallel processing using range<3> following existing SYCL patterns
  • Direct GPU memory access for optimal performance
  • Zero-shift optimization: fast memcpy path

Testing

  • ROLL operations verified against CPU reference
  • Multi-axis scenarios tested with various shift combinations
  • In-place safety validated

Performance

  • GPU memory-optimized kernel design
  • Efficient modular arithmetic for shift normalization
  • Supports tensors up to 4D with configurable axis shifts
  • Zero-overhead for identity operations (all shifts = 0)

Compatibility

  • F32 tensors, OpenCL and Level Zero backends
  • Follows existing SYCL backend conventions

tamarPal added 2 commits October 19, 2025 15:13
sycl: add ROLL operation support
4b76ed1 
- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results
Merge remote-tracking branch 'upstream/master' into feature/sycl-roll…
c44a4df 
…-sync
@github-actions github-actions Bot added ggml changes relating to the ggml tensor library for machine learning SYCL https://en.wikipedia.org/wiki/SYCL - GPU programming language labels Oct 19, 2025
@tamarPal

tamarPal commented Oct 19, 2025

Copy link
Copy Markdown
Contributor Author

@ggerganov @NeoZhangJianyu,
this PR adds SYCL backend support for GGML_OP_ROLL.
All tests pass locally and the implementation currently supports F32 tensors.

tamarPal added 2 commits October 21, 2025 14:49
fix: remove trailing whitespace from roll.cpp
dea85e1 
- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60
ci: retrigger
386db09
@tamarPal

tamarPal commented Oct 21, 2025

Copy link
Copy Markdown
Contributor Author

Hi @CISC @NeoZhangJianyu
I’ve addressed the requested changes and fixed the formatting issues.
Previously, two tests had failed, but they were unrelated to my code.
The PR is now ready, and I’d really appreciate it if you could approve the workflow run and review it for merge.
Thanks a lot!

NeoZhangJianyu
NeoZhangJianyu reviewed Oct 22, 2025
View reviewed changes
Comment thread ggml/src/ggml-sycl/roll.cpp Outdated
Comment thread ggml/src/ggml-sycl/roll.cpp Outdated
NeoZhangJianyu
NeoZhangJianyu approved these changes Oct 23, 2025
View reviewed changes

@NeoZhangJianyu NeoZhangJianyu left a comment

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.

It's good job!

Thank you!

@tamarPal

tamarPal commented Oct 23, 2025 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

Hi @NeoZhangJianyu @CISC!
I’ve fixed the .editorconfig issue in roll.hpp.
All checks should pass now — please let me know if anything else is needed.
Thanks again for the review and approval!

@CISC

CISC commented Oct 25, 2025

Copy link
Copy Markdown
Member

The last commit changed the whole implementation and warrants a re-review.

@NeoZhangJianyu NeoZhangJianyu merged commit 2b9bd9b into ggml-org:master Oct 27, 2025
72 checks passed
Anico2 added a commit to Anico2/llama.cpp that referenced this pull request Jan 15, 2026
@Anico2
sycl: add ROLL operation support (ggml-org#16665)
7f39c0b 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
blime4 referenced this pull request in blime4/llama.cpp Feb 5, 2026
@tamarPal
sycl: add ROLL operation support (#16665)
9bc2622 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
Seunghhon pushed a commit to Seunghhon/llama.cpp that referenced this pull request Apr 26, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
e73116f 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
ljubomirj pushed a commit to ljubomirj/llama.cpp that referenced this pull request May 6, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
06e172f 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
my-other-github-account pushed a commit to my-other-github-account/llama.cpp that referenced this pull request May 15, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
437cdb2 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
my-other-github-account pushed a commit to my-other-github-account/llama.cpp that referenced this pull request May 15, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
cc95fa8 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
phibya pushed a commit to ziee-ai/llama.cpp that referenced this pull request May 29, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
7ce419e 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
fewtarius pushed a commit to fewtarius/CachyLLama that referenced this pull request May 30, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
1006c86 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
fukuro-kun pushed a commit to fukuro-kun/fukuro-llama-cpp-turboquant that referenced this pull request Jul 5, 2026
@tamarPal
sycl: add ROLL operation support (ggml-org#16665)
9317717 
* sycl: add ROLL operation support

- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results

* fix: remove trailing whitespace from roll.cpp

- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60

* ci: retrigger

* sycl: remove wait() calls from ROLL operation

* fix: editorconfig — LF endings + final newline for roll.hpp

---------

Co-authored-by: tamarPal <tamarPal@example.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer
@NeoZhangJianyu NeoZhangJianyu NeoZhangJianyu approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

ggml changes relating to the ggml tensor library for machine learning SYCL https://en.wikipedia.org/wiki/SYCL - GPU programming language

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@tamarPal @CISC @NeoZhangJianyu