Pattern to hoist pack unpack ops from scf.for op #21431
Draft
+284
−0
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
compiler/src/iree/compiler/Codegen/Common/GPU/GPUPackToIntrinsics.cpp
Outdated
Show resolved
Hide resolved
compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_pack_to_instrinsics.mlir
Outdated
Show resolved
Hide resolved
compiler/src/iree/compiler/Codegen/Common/GPU/test/gpu_pack_to_instrinsics.mlir
Outdated
Show resolved
Hide resolved
bfd1ee6 to
9b2603a
Compare
compiler/src/iree/compiler/Codegen/Common/GPU/GPUPackToIntrinsics.cpp
Outdated
Show resolved
Hide resolved
compiler/src/iree/compiler/Codegen/Common/GPU/GPUPackToIntrinsics.cpp
Outdated
Show resolved
Hide resolved
compiler/src/iree/compiler/Codegen/Common/GPU/GPUPackToIntrinsics.cpp
Outdated
Show resolved
Hide resolved
compiler/src/iree/compiler/Codegen/Common/GPU/GPUPackToIntrinsics.cpp
Outdated
Show resolved
Hide resolved
8d43595 to
144417a
Compare
Signed-off-by: Yash Deshpande <ydeshpan@amd.com>
144417a to
19a1783
Compare
| for (auto operand : yieldOp.getOperands()) { | ||
| unpackOp = operand.getDefiningOp<linalg::UnPackOp>(); | ||
| if (!unpackOp) { | ||
| tiedResultIdx++; |
There was a problem hiding this comment.
you can use for for (auto [tiedResultIdx, operand] = llvm::enumerate(yieldOp.getOperands()), this way you dont need to increment tiedResultIdx at different points.
| #map = affine_map<(d0, d1, d2) -> (d0, d2)> | ||
| #map1 = affine_map<(d0, d1, d2) -> (d2, d1)> | ||
| #map2 = affine_map<(d0, d1, d2) -> (d0, d1)> | ||
| func.func @hoist_pack_unpack(%arg0 : tensor<4x8x16x4xf32>, %arg1 : tensor<8x8x16x4xf32>, %arg2 : tensor<64x127xf32>) -> tensor<64x127xf32>{ |
There was a problem hiding this comment.
is this test, testing something that the later test with two pack/unpack pairs is not? If it is not, you can remove this one.
| #map = affine_map<(d0, d1, d2) -> (d0, d2)> | ||
| #map1 = affine_map<(d0, d1, d2) -> (d2, d1)> | ||
| #map2 = affine_map<(d0, d1, d2) -> (d0, d1)> | ||
| func.func @no_hoist_pack_unpack(%arg0 : tensor<4x8x16x4xf32>, %arg1 : tensor<8x4x16x4xf32>, %arg2 : tensor<64x64xf32>) -> tensor<64x64xf32>{ |
There was a problem hiding this comment.
can you make the test names more descriptive, e.g no_hoist_pack_unpack_mismatched no_hoist_pack_unpack_mismatched_otherusers
| %c0 = arith.constant 0 : index | ||
| %c32 = arith.constant 32 : index | ||
| %padding_value = arith.constant 0.000000e+00 : f32 | ||
| %1, %result0 = scf.for %arg3 = %c0 to %c512 step %c32 iter_args(%arg5 = %input, %arg4 = %arg2) -> (tensor<64x128xf32>, tensor<64x127xf32>) { |
There was a problem hiding this comment.
maybe do another argument in the middle that doesn't have a pack/unpack on it?
| linalg::PackOp packOp; | ||
| int64_t tiedResultIdx = 0; | ||
|
|
||
| // Iterate through all operands of yieldOp & hoist each available |
There was a problem hiding this comment.
I think this comment is misleading, you hoist the first legal pack/unpack pair?
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.
This pattern looks for a
scf.yieldop within anscf.for, whose producer is alinalg.unpackop & hoist both the unpack & the corresponding pack ops outside thescf.forwithin GPUPackToIntrinsicsPass.This pattern is helpful if we want to do reduction tiling before packToIntrinsics.