[mlir][Linalg] Allow propagation of pack through multi use pad #98039
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
This allows bubbling `tensor.pack` through `tensor.pad` when the pad has multiple uses. A new pad is created and a `tensor.unpack` is inserted to connect the packed pad with the new users. To keep the previous behavior, the layout propagation control function can be modified to disallow multi-use propagation.
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-linalg Author: Quinn Dawkins (qedawkins) ChangesThis allows bubbling To keep the previous behavior, the layout propagation control function can be modified to disallow multi-use propagation. Full diff: https://github.com/llvm/llvm-project/pull/98039.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp b/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp
index 6984bc2dff498..5f7cf30335e99 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp
@@ -491,9 +491,6 @@ class BubbleUpPackThroughPadOp final : public OpRewritePattern<tensor::PackOp> {
if (!controlFn(padOp))
return failure();
- if (!padOp.getResult().hasOneUse())
- return failure();
-
// TODO: Enable padding when the padding values are the same.
if (packOp.getPaddingValue())
return failure();
@@ -510,7 +507,6 @@ class BubbleUpPackThroughPadOp final : public OpRewritePattern<tensor::PackOp> {
return failure();
ArrayRef<int64_t> innerDimsPos = packOp.getInnerDimsPos();
- ArrayRef<int64_t> outerDimsPerm = packOp.getOuterDimsPerm();
// Bail out if one of the padded dimension is a tiled one.
llvm::SmallBitVector paddedDims = padOp.getPaddedDims();
@@ -524,11 +520,13 @@ class BubbleUpPackThroughPadOp final : public OpRewritePattern<tensor::PackOp> {
OpBuilder::InsertionGuard guard(rewriter);
rewriter.setInsertionPoint(padOp);
+ ArrayRef<int64_t> outerDimsPerm = packOp.getOuterDimsPerm();
+ SmallVector<OpFoldResult> mixedTiles = packOp.getMixedTiles();
auto empty = tensor::PackOp::createDestinationTensor(
- rewriter, loc, padOp.getSource(), packOp.getMixedTiles(), innerDimsPos,
+ rewriter, loc, padOp.getSource(), mixedTiles, innerDimsPos,
outerDimsPerm);
- Value packedSource = rewriter.create<tensor::PackOp>(
- loc, padOp.getSource(), empty, innerDimsPos, packOp.getMixedTiles(),
+ auto sourcePack = rewriter.create<tensor::PackOp>(
+ loc, padOp.getSource(), empty, innerDimsPos, mixedTiles,
/*padding=*/std::nullopt, outerDimsPerm);
// If we have `outer_dims_perms` we need to adjust the padded dimensions.
@@ -545,9 +543,22 @@ class BubbleUpPackThroughPadOp final : public OpRewritePattern<tensor::PackOp> {
highPad.append(pointLoopsSize, rewriter.getIndexAttr(0));
auto newPadOp = rewriter.create<tensor::PadOp>(
- loc, /*result=*/Type(), packedSource, lowPad, highPad, paddingVal,
+ loc, /*result=*/Type(), sourcePack, lowPad, highPad, paddingVal,
padOp.getNofold());
+
+ // If the pad has more than one user, create an unpack on the new pad to
+ // replace the other uses.
+ if (!padOp->hasOneUse()) {
+ auto unpackEmpty = tensor::UnPackOp::createDestinationTensor(
+ rewriter, loc, newPadOp, mixedTiles, innerDimsPos, outerDimsPerm);
+ Value unpackedPad = rewriter.create<tensor::UnPackOp>(
+ loc, newPadOp, unpackEmpty, innerDimsPos, mixedTiles, outerDimsPerm);
+ rewriter.replaceAllUsesExcept(padOp, unpackedPad, sourcePack);
+ }
+
+ // Replace the pack with the new pad.
rewriter.replaceOp(packOp, newPadOp.getResult());
+
return success();
}
diff --git a/mlir/test/Dialect/Linalg/data-layout-propagation.mlir b/mlir/test/Dialect/Linalg/data-layout-propagation.mlir
index 626dd8b697e59..d9206432379fb 100644
--- a/mlir/test/Dialect/Linalg/data-layout-propagation.mlir
+++ b/mlir/test/Dialect/Linalg/data-layout-propagation.mlir
@@ -458,23 +458,23 @@ func.func @unpack_on_input(%arg0: tensor<12x2x56x56x32xf32>, %init: tensor<12x56
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_UNPACK_EMPTY:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
-// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// CHECK: %[[ARG1_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
-// CHECK: %[[ARG1_PACK:.+]] = tensor.pack %[[ARG1]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[ARG1_PACK:.+]] = tensor.pack %[[ARG1]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG1_PACK_EMPTY]]
// CHECK: %[[ARG0_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
-// CHECK: %[[ARG0_PACK:.+]] = tensor.pack %[[UNPACKED_ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[ARG0_PACK:.+]] = tensor.pack %[[UNPACKED_ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_PACK_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[ARG0_PACK]]
// CHECK-SAME: outs(%[[ARG1_PACK]]
-// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[RES]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[RES]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// -----
@@ -537,20 +537,20 @@ func.func @forward_tensor_empty(%arg0: tensor<12x2x56x56x32xf32>) -> tensor<12x5
// CHECK-LABEL: func.func @forward_tensor_empty
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]
// CHECK: %[[ARG0_UNPACK_EMPTY:.+]] = tensor.empty() : tensor<12x56x56x64xf32>
-// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack %[[ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// CHECK: %[[DEST:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
// CHECK: %[[ARG0_PACK_EMPTY:.+]] = tensor.empty() : tensor<12x2x56x56x32xf32>
-// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_PACK_EMPTY]]
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]]]
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK-SAME: outs(%[[DEST]]
// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[RES]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[ARG0_UNPACK_EMPTY]]
// -----
@@ -571,8 +571,8 @@ func.func @pad_valid_unpack_propagation(%arg0: tensor<1x2x56x56x32xf32>) -> tens
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[PADDED:.+]] = tensor.pad %[[ARG0]] low[0, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x58x58x64xf32>
-// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[PADDED]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[PADDED]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x2x58x58x32xf32> -> tensor<1x58x58x64xf32>
// -----
@@ -614,8 +614,8 @@ func.func @pad_along_unpacked_dim(%arg0: tensor<1x2x56x56x32xf32>) -> tensor<1x5
// CHECK: %[[ARG0:.+]]: tensor<1x2x56x56x32xf32>)
// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x56x56x64xf32>
-// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
+// CHECK: %[[UNPACK:.+]] = tensor.unpack %[[ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [3] inner_tiles = [32]
// CHECK-SAME: into %[[EMPTY]] : tensor<1x2x56x56x32xf32> -> tensor<1x56x56x64xf32>
// CHECK: %[[PADDED:.+]] = tensor.pad %[[UNPACK]] low[0, 1, 1, 1] high[0, 1, 1, 1]
@@ -687,6 +687,29 @@ func.func @pad_along_packed_dim(%arg0: tensor<1x60x56x56xf32>) -> tensor<1x2x58x
// -----
+func.func @multi_use_pad_pack_propagation(%arg0: tensor<1x64x56x56xf32>) -> (tensor<1x64x58x58xf32>, tensor<1x2x58x58x32xf32>) {
+ %cst = arith.constant 0.000000e+00 : f32
+ %padded = tensor.pad %arg0 low[0, 0, 1, 1] high[0, 0, 1, 1] {
+ ^bb0(%arg3: index, %arg4: index, %arg5: index, %arg6: index):
+ tensor.yield %cst : f32
+ } : tensor<1x64x56x56xf32> to tensor<1x64x58x58xf32>
+ %0 = tensor.empty() : tensor<1x2x58x58x32xf32>
+ %1 = tensor.pack %padded inner_dims_pos = [1] inner_tiles = [32] into %0 : tensor<1x64x58x58xf32> -> tensor<1x2x58x58x32xf32>
+ return %padded, %1 : tensor<1x64x58x58xf32>, tensor<1x2x58x58x32xf32>
+}
+
+// CHECK-LABEL: func.func @multi_use_pad_pack_propagation(
+// CHECK-SAME: %[[ARG0:.+]]: tensor<1x64x56x56xf32>)
+// CHECK: %[[CST:.+]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[EMPTY:.+]] = tensor.empty() : tensor<1x2x56x56x32xf32>
+// CHECK: %[[PACKED:.+]] = tensor.pack %[[ARG0]] inner_dims_pos = [1] inner_tiles = [32]
+// CHECK-SAME: into %[[EMPTY]] : tensor<1x64x56x56xf32> -> tensor<1x2x56x56x32xf32>
+// CHECK: %[[PADDED:.+]] = tensor.pad %[[PACKED]] low[0, 0, 1, 1, 0] high[0, 0, 1, 1, 0]
+// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[PADDED]] inner_dims_pos = [1] inner_tiles = [32]
+// CHECK: return %[[UNPACKED]], %[[PADDED]]
+
+// -----
+
#map0 = affine_map<(d0, d1) -> (d0, d1)>
func.func @would_break_dominance(%arg0: tensor<128x256xi32>) -> tensor<4x16x16x32xi32>{
%init = tensor.empty() : tensor<128x256xi32>
@@ -713,7 +736,7 @@ func.func @would_break_dominance(%arg0: tensor<128x256xi32>) -> tensor<4x16x16x3
// CHECK-SAME: outs(%[[EMPTY]]
// CHECK: %[[ALLOC:.+]] = bufferization.alloc_tensor() : tensor<4x16x16x32xi32>
// CHECK-NEXT: %{{.+}} = tensor.pack %[[GEN]]
-// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
+// CHECK-SAME: inner_dims_pos = [1, 0] inner_tiles = [16, 32]
// CHECK-SAME: into %[[ALLOC]]
// -----
@@ -760,19 +783,19 @@ func.func @unpack_empty_inner_dims(%arg0: tensor<12x64x56x56xf32>) -> tensor<12x
// CHECK-LABEL: func.func @unpack_empty_inner_dims
// CHECK: %[[UNPACKED_ARG0:.+]] = tensor.unpack
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
-// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
+// CHECK: %[[PACKED_ARG0:.+]] = tensor.pack %[[UNPACKED_ARG0]]
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
// CHECK: %[[RES:.+]] = linalg.generic
// CHECK-SAME: ins(%[[PACKED_ARG0]]
// CHECK: %[[UNPACKED:.+]] = tensor.unpack %[[RES]]
-// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
+// CHECK-SAME: outer_dims_perm = [0, 3, 1, 2] inner_dims_pos = [] inner_tiles = []
// -----
#map0 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
#map1 = affine_map<(d0, d1, d2) -> (d0, d1)>
-func.func @reduction_pack_transpose_inner_dims(%arg0: tensor<128x256x32xi32>,
+func.func @reduction_pack_transpose_inner_dims(%arg0: tensor<128x256x32xi32>,
%arg1: tensor<128x256xi32>) -> tensor<4x16x16x32xi32>{
%elem = linalg.generic {indexing_maps = [#map0, #map1], iterator_types = ["parallel", "parallel", "reduction"]}
ins(%arg0 : tensor<128x256x32xi32>)
@@ -810,7 +833,7 @@ func.func @reduction_pack_transpose_inner_dims(%arg0: tensor<128x256x32xi32>,
// -----
-func.func @reduction_pack_with_outer_dims(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100xi32>,
+func.func @reduction_pack_with_outer_dims(%arg0: tensor<100x128x200x256xi32>, %arg1: tensor<100xi32>,
%arg2: tensor<128xi32>, %init_reduction: tensor<100x128x256xi32>) -> tensor<4x16x100x16x32xi32>
{
%reduction = linalg.generic {
@@ -867,7 +890,7 @@ func.func @reduction_pack_with_outer_dims(%arg0: tensor<100x128x200x256xi32>, %a
#map0 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2 * 2 + d4, d3 * 2 + d5)>
#map1 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d4, d5)>
#map2 = affine_map<(d0, d1, d2, d3, d4, d5) -> (d1, d2, d3)>
-func.func @unpack_different_destination_shape(%arg0: tensor<1x1x1080x1920x16xi32>,
+func.func @unpack_different_destination_shape(%arg0: tensor<1x1x1080x1920x16xi32>,
%filter: tensor<2x2xi32>) -> tensor<16x540x960xi32>{
%init = tensor.empty() : tensor<16x540x960xi32>
%empty = tensor.empty() : tensor<1x16x1080x1920xi32>
|
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 allows bubbling
tensor.packthroughtensor.padwhen the pad has multiple uses. A new pad is created and atensor.unpackis inserted to connect the packed pad with the new users.To keep the previous behavior, the layout propagation control function can be modified to disallow multi-use propagation.