Skip to content

[MLIR][TilingInterface] Extend consumer fusion for multi-use of producer shared by terminator ops #110105

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 5 commits into from
Sep 30, 2024
Merged

[MLIR][TilingInterface] Extend consumer fusion for multi-use of producer shared by terminator ops #110105

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
eae4114
[MLIR][TilingInterface] Extend consumer fusion for multi-use of producer
Abhishek-Varma Sep 26, 2024
85af1e3
Address review comments
Abhishek-Varma Sep 30, 2024
bba3e9c
Address final review comment
Abhishek-Varma Sep 30, 2024
b5f1689
Add early exit check too
Abhishek-Varma Sep 30, 2024
482cf73
Comment
Abhishek-Varma Sep 30, 2024
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
38 changes: 23 additions & 15 deletions mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1481,21 +1481,29 @@ checkAssumptionForFusingConsumer(tensor::InsertSliceOp candidateSliceOp) {
/// failure otherwise.
static FailureOr<OpOperand *> getConsumerFromUses(Value val,
Block *containingOpBlock) {
// Step 1. Check that the value has exactly one use.
if (!llvm::hasSingleElement(val.getUses()))
return failure();
// Step 2. Get uses.
OpOperand &operand = (*val.getUses().begin());
Operation *consumerOp = operand.getOwner();
// TODO: We have to init result of consumer before scf.for, use
// DestinationStyleOpInterface to get result shape from init for now.
// Add support for other op such as op has InferTypeOpInterface.
if (!isa<TilingInterface>(consumerOp) ||
!isa<DestinationStyleOpInterface>(consumerOp))
return failure();
if (containingOpBlock != consumerOp->getBlock())
return failure();
return &operand;
// Check that the value has exactly one use which isn't a scf.yield or a
// tensor.parallel_insert_slice op.
OpOperand *operand = nullptr;
for (OpOperand &opOperand : val.getUses()) {
Operation *consumerOp = opOperand.getOwner();
if (isa<scf::YieldOp, tensor::ParallelInsertSliceOp>(consumerOp))
continue;
if (operand)
return failure();
// TODO: We have to init result of consumer before scf.for, use
// DestinationStyleOpInterface to get result shape from init for now.
// Add support for other op such as op has InferTypeOpInterface.
if (!isa<TilingInterface>(consumerOp) ||
!isa<DestinationStyleOpInterface>(consumerOp))
return failure();
if (containingOpBlock != consumerOp->getBlock())
return failure();
operand = &opOperand;
}

if (operand)
return operand;
return failure();
}

/// Find the perfectly nested loops outside of given loop(included) sorted from
Expand Down
71 changes: 71 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 @@ -437,3 +437,74 @@ module attributes {transform.with_named_sequence} {
// CHECK: scf.yield %[[LOOP_RESULT2]]#0, %[[LOOP_RESULT2]]#1 :
// CHECK: }
// CHECK: return %[[LOOP_RESULT1]]#1 :

// -----

// This test case checks fusion of consumer even if the producer has multiple uses.
// The multiple uses of the producer essentially means that besides the consumer
// op in concern, the only other uses of the producer are allowed in :-
// 1. scf.yield
// 2. tensor.parallel_insert_slice

module {
module {
func.func @fuse_consumer_for_multi_use_producer(%arg0: tensor<256x512xf32>, %arg1: tensor<512x256xf32>, %arg2: tensor<256x256xf32>) -> (tensor<256x256xf32>, tensor<256x256xf32>) {
%c0 = arith.constant 0 : index
%c64 = arith.constant 64 : index
%c256 = arith.constant 256 : index
%cst = arith.constant 0.000000e+00 : f32
%0 = tensor.empty() : tensor<256x256xf32>
%1 = linalg.fill ins(%cst : f32) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
%2:2 = scf.for %arg3 = %c0 to %c256 step %c64 iter_args(%arg4 = %1, %arg5 = %arg2) -> (tensor<256x256xf32>, tensor<256x256xf32>) {
%3 = scf.for %arg6 = %c0 to %c256 step %c64 iter_args(%arg7 = %arg4) -> (tensor<256x256xf32>) {
%extracted_slice = tensor.extract_slice %arg7[%arg3, %arg6] [64, 64] [1, 1] : tensor<256x256xf32> to tensor<64x64xf32>
%extracted_slice_0 = tensor.extract_slice %arg0[%arg3, 0] [64, 512] [1, 1] : tensor<256x512xf32> to tensor<64x512xf32>
%extracted_slice_1 = tensor.extract_slice %arg1[0, %arg6] [512, 64] [1, 1] : tensor<512x256xf32> to tensor<512x64xf32>
%5 = linalg.matmul ins(%extracted_slice_0, %extracted_slice_1 : tensor<64x512xf32>, tensor<512x64xf32>) outs(%extracted_slice : tensor<64x64xf32>) -> tensor<64x64xf32>
%inserted_slice = tensor.insert_slice %5 into %arg7[%arg3, %arg6] [64, 64] [1, 1] : tensor<64x64xf32> into tensor<256x256xf32>
scf.yield %inserted_slice : tensor<256x256xf32>
}
%4 = linalg.add ins(%3, %arg5 : tensor<256x256xf32>, tensor<256x256xf32>) outs(%0 : tensor<256x256xf32>) -> tensor<256x256xf32>
scf.yield %3, %4 : tensor<256x256xf32>, tensor<256x256xf32>
}
return %2#0, %2#1 : tensor<256x256xf32>, tensor<256x256xf32>
}
}
module attributes {transform.with_named_sequence} {
transform.named_sequence @__transform_main(%arg0: !transform.any_op {transform.readonly}) {
%0 = transform.structured.match ops{["tensor.insert_slice"]} in %arg0 : (!transform.any_op) -> !transform.any_op
%consumer, %fused_consumer = transform.test.fuse_consumer %0 : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
transform.yield
}
}
}
// CHECK: func.func @fuse_consumer_for_multi_use_producer(
// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]: tensor<256x512xf32>
// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]: tensor<512x256xf32>
// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]: tensor<256x256xf32>
// CHECK: %[[dest0:.*]] = tensor.empty() : tensor<256x256xf32>
// CHECK: %[[dest1:.*]] = linalg.fill
// CHECK-SAME: outs(%[[dest0]] :
// CHECK: %[[LOOP_RESULT1:.*]]:2 = scf.for %[[IV1:.*]] = %[[C0]]
// CHECK-SAME: iter_args(%[[FIRST_OUT_ARG1:.*]] = %[[dest1]], %[[SECOND_OUT_ARG1:.*]] = %[[ARG2]])
// CHECK-SAME: {
// CHECK: %[[LOOP_RESULT2:.*]]:2 = scf.for %[[IV2:.*]] = %[[C0]]
// CHECK-SAME: iter_args(%[[FIRST_OUT_ARG2:.*]] = %[[FIRST_OUT_ARG1]], %[[SECOND_OUT_ARG2:.*]] = %[[dest0]])
// CHECK-SAME: {
// CHECK: %[[MAT_OUT_SLICE:.*]] = tensor.extract_slice %[[FIRST_OUT_ARG2]][%[[IV1]], %[[IV2]]] [64, 64] [1, 1]
// CHECK: %[[INPUT_SLICE:.*]] = tensor.extract_slice %[[ARG0]][%[[IV1]], 0] [64, 512] [1, 1]
// CHECK: %[[WEIGHT_SLICE:.*]] = tensor.extract_slice %[[ARG1]][0, %[[IV2]]] [512, 64] [1, 1]
// CHECK: %[[TILED_MAT_OUT:.*]] = linalg.matmul
// CHECK-SAME: outs(%[[MAT_OUT_SLICE]] :
// CHECK: %[[INSERT_MAT:.*]] = tensor.insert_slice %[[TILED_MAT_OUT]] into %[[FIRST_OUT_ARG2]][%[[IV1]], %[[IV2]]] [64, 64] [1, 1]
// CHECK: %[[ADD_OPERAND2_SLICE:.*]] = tensor.extract_slice %[[SECOND_OUT_ARG1]][%[[IV1]], %[[IV2]]] [64, 64] [1, 1]
// CHECK: %[[ADD_OUT_SLICE:.*]] = tensor.extract_slice %[[SECOND_OUT_ARG2]][%[[IV1]], %[[IV2]]] [64, 64] [1, 1]
// CHECK: %[[TILED_ADD_OUT:.*]] = linalg.add
// CHECK-SAME: ins(%[[TILED_MAT_OUT]], %[[ADD_OPERAND2_SLICE]] :
// CHECK-SAME: outs(%[[ADD_OUT_SLICE]] :
// CHECK: %[[INSERT_ADD:.*]] = tensor.insert_slice %[[TILED_ADD_OUT]] into %[[SECOND_OUT_ARG2]][%[[IV1]], %[[IV2]]] [64, 64] [1, 1]
// CHECK: scf.yield %[[INSERT_MAT]], %[[INSERT_ADD]] :
// CHECK: }
// CHECK: scf.yield %[[LOOP_RESULT2]]#0, %[[LOOP_RESULT2]]#1 :
// CHECK: }
// CHECK: return %[[LOOP_RESULT1]]#0, %[[LOOP_RESULT1]]#1 :
Loading