[mlir][linalg] Add bufferization for linalg.softmax
#97019
Merged
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
Implement the `BufferizableOpInterface` for `linalg.softmax`. The op is not a `LinalgOp`, so it is not covered by the "catch all" `LinalgOp` interface implementation.
matthias-springer
requested review from
dcaballe,
nicolasvasilache and
rengolin
as code owners
|
@llvm/pr-subscribers-mlir Author: Matthias Springer (matthias-springer) ChangesImplement the Full diff: https://github.com/llvm/llvm-project/pull/97019.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp b/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
index 81a5398dabcb7..be158af09d398 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
@@ -162,6 +162,35 @@ struct LinalgOpInterfaceHelper {
(Ops::template attachInterface<LinalgOpInterface<Ops>>(*ctx), ...);
}
};
+
+struct SoftmaxOpInterface
+ : public DstBufferizableOpInterfaceExternalModel<SoftmaxOpInterface,
+ linalg::SoftmaxOp> {
+ bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
+ const AnalysisState &state) const {
+ // Output operand is not read.
+ auto softmaxOp = cast<linalg::SoftmaxOp>(op);
+ return &opOperand == &softmaxOp.getInputMutable();
+ }
+
+ LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
+ const BufferizationOptions &options) const {
+ auto softmaxOp = cast<linalg::SoftmaxOp>(op);
+ FailureOr<Value> inputBuffer =
+ getBuffer(rewriter, softmaxOp.getInput(), options);
+ if (failed(inputBuffer))
+ return failure();
+ FailureOr<Value> outputBuffer =
+ getBuffer(rewriter, softmaxOp.getOutput(), options);
+ if (failed(outputBuffer))
+ return failure();
+ rewriter.create<linalg::SoftmaxOp>(softmaxOp.getLoc(),
+ /*result=*/TypeRange(), *inputBuffer,
+ *outputBuffer, softmaxOp.getDimension());
+ replaceOpWithBufferizedValues(rewriter, op, *outputBuffer);
+ return success();
+ }
+};
} // namespace
void mlir::linalg::registerBufferizableOpInterfaceExternalModels(
@@ -174,5 +203,7 @@ void mlir::linalg::registerBufferizableOpInterfaceExternalModels(
#define GET_OP_LIST
#include "mlir/Dialect/Linalg/IR/LinalgStructuredOps.cpp.inc"
>::registerOpInterface(ctx);
+
+ SoftmaxOp::attachInterface<SoftmaxOpInterface>(*ctx);
});
}
diff --git a/mlir/test/Dialect/Linalg/bufferize.mlir b/mlir/test/Dialect/Linalg/bufferize.mlir
index e8ab1184b1fd2..f416cd9fcf0b2 100644
--- a/mlir/test/Dialect/Linalg/bufferize.mlir
+++ b/mlir/test/Dialect/Linalg/bufferize.mlir
@@ -189,3 +189,20 @@ func.func @bufferize_dot(%in: tensor<4xf32>, %out: tensor<f32>) -> tensor<f32> {
// CHECK: %[[OUT_TENSOR:.*]] = bufferization.to_tensor %[[ALLOC]] : memref<f32>
// CHECK: return %[[OUT_TENSOR]]
}
+
+// -----
+
+// CHECK-LABEL: func @bufferize_softmax(
+// CHECK-SAME: %[[arg0:.*]]: tensor<2x16x32xf32>, %[[arg1:.*]]: tensor<2x16x32xf32>
+// CHECK: %[[m0:.*]] = bufferization.to_memref %[[arg0]]
+// CHECK: %[[alloc:.*]] = memref.alloc()
+// CHECK-NOT: memref.copy
+// CHECK: linalg.softmax dimension(2) ins(%[[m0]] : {{.*}}) outs(%[[alloc:.*]] : {{.*}})
+// CHECK: %[[result:.*]] = bufferization.to_tensor %[[alloc]]
+// CHECK: return %[[result]]
+func.func @bufferize_softmax(%arg0: tensor<2x16x32xf32>, %arg1: tensor<2x16x32xf32>) -> tensor<2x16x32xf32> {
+ %1 = linalg.softmax dimension(2)
+ ins(%arg0 : tensor<2x16x32xf32>)
+ outs(%arg1: tensor<2x16x32xf32>) -> tensor<2x16x32xf32>
+ return %1 : tensor<2x16x32xf32>
+}
|
|
@llvm/pr-subscribers-mlir-linalg Author: Matthias Springer (matthias-springer) ChangesImplement the Full diff: https://github.com/llvm/llvm-project/pull/97019.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp b/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
index 81a5398dabcb7..be158af09d398 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/BufferizableOpInterfaceImpl.cpp
@@ -162,6 +162,35 @@ struct LinalgOpInterfaceHelper {
(Ops::template attachInterface<LinalgOpInterface<Ops>>(*ctx), ...);
}
};
+
+struct SoftmaxOpInterface
+ : public DstBufferizableOpInterfaceExternalModel<SoftmaxOpInterface,
+ linalg::SoftmaxOp> {
+ bool bufferizesToMemoryRead(Operation *op, OpOperand &opOperand,
+ const AnalysisState &state) const {
+ // Output operand is not read.
+ auto softmaxOp = cast<linalg::SoftmaxOp>(op);
+ return &opOperand == &softmaxOp.getInputMutable();
+ }
+
+ LogicalResult bufferize(Operation *op, RewriterBase &rewriter,
+ const BufferizationOptions &options) const {
+ auto softmaxOp = cast<linalg::SoftmaxOp>(op);
+ FailureOr<Value> inputBuffer =
+ getBuffer(rewriter, softmaxOp.getInput(), options);
+ if (failed(inputBuffer))
+ return failure();
+ FailureOr<Value> outputBuffer =
+ getBuffer(rewriter, softmaxOp.getOutput(), options);
+ if (failed(outputBuffer))
+ return failure();
+ rewriter.create<linalg::SoftmaxOp>(softmaxOp.getLoc(),
+ /*result=*/TypeRange(), *inputBuffer,
+ *outputBuffer, softmaxOp.getDimension());
+ replaceOpWithBufferizedValues(rewriter, op, *outputBuffer);
+ return success();
+ }
+};
} // namespace
void mlir::linalg::registerBufferizableOpInterfaceExternalModels(
@@ -174,5 +203,7 @@ void mlir::linalg::registerBufferizableOpInterfaceExternalModels(
#define GET_OP_LIST
#include "mlir/Dialect/Linalg/IR/LinalgStructuredOps.cpp.inc"
>::registerOpInterface(ctx);
+
+ SoftmaxOp::attachInterface<SoftmaxOpInterface>(*ctx);
});
}
diff --git a/mlir/test/Dialect/Linalg/bufferize.mlir b/mlir/test/Dialect/Linalg/bufferize.mlir
index e8ab1184b1fd2..f416cd9fcf0b2 100644
--- a/mlir/test/Dialect/Linalg/bufferize.mlir
+++ b/mlir/test/Dialect/Linalg/bufferize.mlir
@@ -189,3 +189,20 @@ func.func @bufferize_dot(%in: tensor<4xf32>, %out: tensor<f32>) -> tensor<f32> {
// CHECK: %[[OUT_TENSOR:.*]] = bufferization.to_tensor %[[ALLOC]] : memref<f32>
// CHECK: return %[[OUT_TENSOR]]
}
+
+// -----
+
+// CHECK-LABEL: func @bufferize_softmax(
+// CHECK-SAME: %[[arg0:.*]]: tensor<2x16x32xf32>, %[[arg1:.*]]: tensor<2x16x32xf32>
+// CHECK: %[[m0:.*]] = bufferization.to_memref %[[arg0]]
+// CHECK: %[[alloc:.*]] = memref.alloc()
+// CHECK-NOT: memref.copy
+// CHECK: linalg.softmax dimension(2) ins(%[[m0]] : {{.*}}) outs(%[[alloc:.*]] : {{.*}})
+// CHECK: %[[result:.*]] = bufferization.to_tensor %[[alloc]]
+// CHECK: return %[[result]]
+func.func @bufferize_softmax(%arg0: tensor<2x16x32xf32>, %arg1: tensor<2x16x32xf32>) -> tensor<2x16x32xf32> {
+ %1 = linalg.softmax dimension(2)
+ ins(%arg0 : tensor<2x16x32xf32>)
+ outs(%arg1: tensor<2x16x32xf32>) -> tensor<2x16x32xf32>
+ return %1 : tensor<2x16x32xf32>
+}
|
matthias-springer
deleted the
users/matthias-springer/linalg_softmax_bufferize
branch
lravenclaw
pushed a commit
to lravenclaw/llvm-project
that referenced
this pull request
Jul 3, 2024
Implement the `BufferizableOpInterface` for `linalg.softmax`. The op is not a `LinalgOp`, so it is not covered by the "catch all" `LinalgOp` interface implementation.
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.
Implement the
BufferizableOpInterfaceforlinalg.softmax. The op is not aLinalgOp, so it is not covered by the "catch all"LinalgOpinterface implementation.