[SYCL][Fusion] Enable fusion of kernels with different ND-ranges #8209
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
All kernels with the same (or unspecified) local size and offset can be fused. In order to make this work, some builtins getting index space information must be remapped and the resulting ND-range of the fused kernel, calculated. The ND-range of the fused kernel will have: 1. The same number of dimensions as the input ND-range with the higher number of dimensions; 2. The same local size as the shared local size (or unspecified) 3. The same offset as the shared offset 4. The global size will be the **greatest** input global size as per the following ordering: i. Number of work items (enforces correctness); ii. Number of occurrences (less remappings needed); iii. Lexical order of the dimensions (introduces determinism). Builtins obtaining the local/global size/id, work-group id, number of work-groups or offset are remapped introducing as per an alwaysinline function that can be reused along the fusion pass. More information can be found in the Builtins.cpp file, where the remapping logic is implemented. Signed-off-by: Victor Perez <[email protected]>
victor-eds
force-pushed
the
kernel-fusion/different-nd-ranges
branch
from
5e2927d to
eff2d2f
Compare
|
Reviewers guide: |
victor-eds
added a commit
to victor-eds/llvm-test-suite
that referenced
this pull request
Feb 6, 2023
Check that kernel fusion works when fusing kernels with different ND-ranges (different global sizes and dimensions). Implementation: intel/llvm#8209 Signed-off-by: Victor Perez <[email protected]>
|
/verify with intel/llvm-test-suite#1575 |
sommerlukas
requested changes
Feb 7, 2023
sommerlukas
requested changes
Feb 8, 2023
sommerlukas
approved these changes
Feb 8, 2023
victor-eds
added
sycl-mlir
|
I have no rights to merge this commit. Can I get it merged, please? @intel/llvm-gatekeepers |
|
Pre-commit is failing to build it. Looks like you need an |
@steffenlarsen Thanks for pointing out. I've added the header. Failures are now due to missing tests updates (PR in same state). |
|
/verify with intel/llvm-test-suite#1575 |
It appears there are two unrelated tests failing. Fusion should not affect those whatsoever. |
|
Failure in ESIMD/imulh_umulh.cpp addressed in intel/llvm-test-suite#1593. |
steffenlarsen
approved these changes
Feb 14, 2023
steffenlarsen
pushed a commit
to intel/llvm-test-suite
that referenced
this pull request
Feb 14, 2023
Check that kernel fusion works when fusing kernels with different ND-ranges (different global sizes and dimensions). Implementation: intel/llvm#8209 --------- Signed-off-by: Victor Perez <[email protected]> Co-authored-by: Lukas Sommer <[email protected]>
aelovikov-intel
pushed a commit
to aelovikov-intel/llvm
that referenced
this pull request
Mar 27, 2023
…/llvm-test-suite#1575) Check that kernel fusion works when fusing kernels with different ND-ranges (different global sizes and dimensions). Implementation: intel#8209 --------- Signed-off-by: Victor Perez <[email protected]> Co-authored-by: Lukas Sommer <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
All kernels with the same (or unspecified) local size and offset can be fused. In order to make this work, some builtins getting index space information must be remapped and the resulting ND-range of the fused kernel, calculated.
The ND-range of the fused kernel will have:
i. Number of work items (enforces correctness);
ii. Number of occurrences (less remappings needed);
iii. Lexical order of the dimensions (introduces determinism).
Builtins obtaining the local/global size/id, work-group id, number of work-groups or offset are remapped introducing as per an alwaysinline function that can be reused along the fusion pass. More information can be found in the Builtins.cpp file, where the remapping logic is implemented.
Signed-off-by: Victor Perez [email protected]