Skip to content

[mlir][linalg] Allow pack consumer fusion if the tile size is greater than dimension size. #149438

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

Merged
merged 3 commits into from
Jul 18, 2025
Merged

[mlir][linalg] Allow pack consumer fusion if the tile size is greater than dimension size. #149438

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
a4ed7d8
[mlir][linalg] Allow pack consumer fusion if the tile size is greater…
hanhanW Jul 18, 2025
4c992de
add a comment
hanhanW Jul 18, 2025
6c51a22
correct the comment
hanhanW Jul 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions mlir/lib/Dialect/Linalg/Transforms/TilingInterfaceImpl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -911,14 +911,16 @@ struct PackOpTiling

// If a dimension is not tiled, it is always valid to fuse the pack op,
// even if the op has padding semantics. Because it always generates a
// full slice along the dimension.
// full slice along the dimension. The tile sizes are for unpacked
// domain, i.e., `srcDimSize`, so `tileSize < srcDimSize` means that the
// dimension is tiled.
// TODO: It could be untiled if the `srcDimSize` is dynamic. It is a
// hard check to determine if a dimension is tiled or not.
int64_t srcDimSize = packOp.getSourceType().getDimSize(dim);
int64_t destDimSize = outerShapeWithoutTranspose[dim];
bool isTiled = failed(cstTileSize) ||
Copy link
Contributor

Choose a reason for hiding this comment

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

A comment here explaining the < would be nice, something like what you have in the description.

ShapedType::isDynamic(srcDimSize) ||
cstTileSize.value() != srcDimSize;
cstTileSize.value() < srcDimSize;
if (!isTiled) {
outerDimOffsets.push_back(offsets[dim]);
if (ShapedType::isStatic(destDimSize)) {
Expand Down
50 changes: 50 additions & 0 deletions mlir/test/Interfaces/TilingInterface/tile-and-fuse-consumer.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -451,6 +451,56 @@ module attributes {transform.with_named_sequence} {

// -----

#map = affine_map<(d0) -> (-d0 + 4, 16)>
func.func @fuse_pack_consumer_if_single_iteration(%arg0: tensor<4x4xf32>) -> tensor<1x4x16x1xf32> {
%0 = tensor.empty() : tensor<1x4x16x1xf32>
%1 = tensor.empty() : tensor<4x4xf32>
%2 = scf.forall (%arg1) = (0) to (4) step (16) shared_outs(%arg2 = %1) -> (tensor<4x4xf32>) {
Copy link
Contributor

Choose a reason for hiding this comment

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

Q: wouldn't/shouldn't this (and in general one-iteration loops) be folded away? Probably it should happen at a different point/place, but still just wondering :)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

If this is scf.for, we can fold them away. What I'm not sure is the case that distribution mapping attributes are present. I think the semantic is that you'll need to distribute the computation using the core_id/thread_id. It may be correct to still fold it away though.

%3 = affine.min #map(%arg1)
%extracted_slice = tensor.extract_slice %arg0[%arg1, 0] [%3, 4] [1, 1] : tensor<4x4xf32> to tensor<?x4xf32>
%extracted_slice_0 = tensor.extract_slice %arg2[%arg1, 0] [%3, 4] [1, 1] : tensor<4x4xf32> to tensor<?x4xf32>
%4 = linalg.exp ins(%extracted_slice : tensor<?x4xf32>) outs(%extracted_slice_0 : tensor<?x4xf32>) -> tensor<?x4xf32>
scf.forall.in_parallel {
tensor.parallel_insert_slice %4 into %arg2[%arg1, 0] [%3, 4] [1, 1] : tensor<?x4xf32> into tensor<4x4xf32>
}
}
%cst = arith.constant 0.000000e+00 : f32
%pack = linalg.pack %2 padding_value(%cst : f32) outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [16, 1] into %0 : tensor<4x4xf32> -> tensor<1x4x16x1xf32>
return %pack : tensor<1x4x16x1xf32>
}

module attributes {transform.with_named_sequence} {
transform.named_sequence @__transform_main(%arg0: !transform.any_op {transform.readonly}) {
%0 = transform.structured.match ops{["tensor.parallel_insert_slice"]} in %arg0 : (!transform.any_op) -> !transform.any_op
%1 = transform.structured.match ops{["scf.forall"]} in %arg0 : (!transform.any_op) -> !transform.any_op
%consumer, %fused_consumer = transform.test.fuse_consumer %0 in(%1) : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)
transform.yield
}
}
// CHECK: #[[MAP:.*]] = affine_map<(d0) -> (-d0 + 4, 16)>
// CHECK: func.func @fuse_pack_consumer_if_single_iteration(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-DAG: %[[PACK_INIT:.*]] = tensor.empty() : tensor<1x4x16x1xf32>
// CHECK-DAG: %[[ELEM_INIT:.*]] = tensor.empty() : tensor<4x4xf32>
// CHECK-DAG: %[[PAD_VAL:.*]] = arith.constant 0.000000e+00 : f32
// CHECK: %{{.*}}:2 = scf.forall (%[[IV:.*]]) = (0) to (4) step (16)
// CHECK-SAME: shared_outs(%[[ELEM_OUT_ARG:.*]] = %[[ELEM_INIT]], %[[PACK_OUT_ARG:.*]] = %[[PACK_INIT]])
// CHECK-DAG: %[[SIZE:.+]] = affine.min #[[MAP]](%[[IV]])
// CHECK-DAG: %[[ELEM_SRC:.*]] = tensor.extract_slice %[[ARG0]][%[[IV]], 0] [%[[SIZE]], 4] [1, 1]
// CHECK-DAG: %[[ELEM_DEST:.*]] = tensor.extract_slice %[[ELEM_OUT_ARG]][%[[IV]], 0] [%[[SIZE]], 4] [1, 1]
// CHECK: %[[ELEM:.*]] = linalg.exp
// CHECK-SAME: ins(%[[ELEM_SRC]]
// CHECK-SAME: outs(%[[ELEM_DEST]]
// CHECK-DAG: %[[TILED_PACK_DEST:.*]] = tensor.extract_slice %[[PACK_OUT_ARG]][%[[IV]], 0, 0, 0] [1, 4, 16, 1] [1, 1, 1, 1]
// CHECK: %[[PACK:.*]] = linalg.pack %[[ELEM]]
// CHECK-SAME: padding_value(%[[PAD_VAL]] : f32)
// CHECK-SAME: outer_dims_perm = [0, 1] inner_dims_pos = [0, 1] inner_tiles = [16, 1]
// CHECK-SAME: into %[[TILED_PACK_DEST]]
// CHECK: scf.forall.in_parallel {
// CHECK: tensor.parallel_insert_slice %[[ELEM]] into %[[ELEM_OUT_ARG]][%[[IV]], 0] [%[[SIZE]], 4] [1, 1]
// CHECK: tensor.parallel_insert_slice %[[PACK]] into %[[PACK_OUT_ARG]][%[[IV]], 0, 0, 0] [1, 4, 16, 1] [1, 1, 1, 1]

// -----

func.func @fuse_perfect_tiling_pack_consumer_with_outer_dims_perm(%arg0: tensor<64x32xf32>, %arg1: tensor<64x32xf32>, %arg2: tensor<2x64x16x1xf32>) -> tensor<2x64x16x1xf32> {
%0 = scf.forall (%arg3) = (0) to (32) step (16) shared_outs(%arg4 = %arg1) -> (tensor<64x32xf32>) {
Expand Down