[XPU][OptEW] Define -intel-triton-optimize-elementwise-parallelism pass
#2631
Conversation
This file contains 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
…pass Define pass improving elementwise parallelism by avoiding layout conversions leading to data duplication between threads. See pass documentation for more information. Signed-off-by: victor-eds <victor.perez@codeplay.com>
victor-eds
requested review from
whitneywhtsang,
etiotto,
chengjunlu and
a team
|
First step for #2562. |
victor-eds
commented
Nov 5, 2024
| // CHECK: %[[VAL_4:.*]] = tt.splat %[[VAL_2]] : f32 -> tensor<128xf32, #triton_gpu.slice<{dim = 1, parent = #[[$ATTR_4]]}>> | ||
| %2 = tt.splat %arg2 : f32 -> tensor<128xf32, #triton_gpu.slice<{dim = 1, parent = #blocked1}>> | ||
| %3 = arith.addf %0, %1 : tensor<128xf32, #triton_gpu.slice<{dim = 1, parent = #blocked1}>> | ||
| // CHECK: %[[VAL_5:.*]] = arith.addf %[[VAL_4]], %[[VAL_3]] : tensor<128xf32, #triton_gpu.slice<{dim = 1, parent = #[[$ATTR_4]]}>> |
There was a problem hiding this comment.
The layout conversion that would result from %[[VAL_3]] conversion is erased by the pass as we can see
victor-eds
commented
Nov 5, 2024
Comment on lines
406
to
407
| let dependentDialects = ["mlir::triton::TritonDialect", | ||
| "mlir::triton::gpu::TritonGPUDialect"]; |
There was a problem hiding this comment.
If we find an elementwise operation from dialect X, that means the dialect has been loaded already, so no need to have it as a dependent dialect in order to create this. We want to include these two for future proofing, tho.
etiotto
reviewed
Nov 5, 2024
third_party/intel/include/Dialect/TritonIntelGPU/Transforms/Passes.td
Outdated
Show resolved
Hide resolved
|
@intel/triton-codeplay-reviewers @etiotto @whitneywhtsang @chengjunlu Can I get reviews here? I already have a PR hanging on this one and will add more on Monday, so I'd like to begin getting this work merged. |
|
I may change the approach to this next week. Delay reviews for now. |
etiotto
approved these changes
Nov 11, 2024
chengjunlu
approved these changes
Nov 12, 2024
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.
Define pass improving elementwise parallelism by avoiding layout conversions leading to data duplication between threads.
See pass documentation for more information.