[Mlir-commits] [mlir] 660f281 - [mlir][Linalg] Make codegen strategy late transformations opt-in

[Nicolas Vasilache]

Author: Nicolas Vasilache
Date: 2021-09-15T11:02:14Z
New Revision: 660f281b5e755663b2d8ad061075eb5f5abdf9fb

URL: https://github.com/llvm/llvm-project/commit/660f281b5e755663b2d8ad061075eb5f5abdf9fb
DIFF: https://github.com/llvm/llvm-project/commit/660f281b5e755663b2d8ad061075eb5f5abdf9fb.diff

LOG: [mlir][Linalg] Make codegen strategy late transformations opt-in

Summary: Making the late transformations opt-in results in less surprising behavior when composing multiple calls to the codegen strategy.

Reviewers:

Subscribers:

Differential revision: https://reviews.llvm.org/D109820

Added: 
    

Modified: 
    mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h
    mlir/test/lib/Dialect/Linalg/TestLinalgCodegenStrategy.cpp

Removed: 
    


################################################################################
diff  --git a/mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h b/mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h
index bee0d5a12800f..d33dd81fa323a 100644
--- a/mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h
+++ b/mlir/include/mlir/Dialect/Linalg/Transforms/CodegenStrategy.h
@@ -152,12 +152,16 @@ struct Vectorize : public Transformation {
 
 /// Options to control the application of late transformations.
 struct LateCodegenStrategyOptions {
+  /// Hoisting transformations are always deemed beneficial and must disabled
+  /// explicitly.
   bool enableLICM = true;
   bool enableHoistRedundantVectorTransfers = true;
   bool enableHoistRedundantVectorTransfersOnTensor = true;
-  bool enableVectorTransferPartialRewrite = true;
-  bool enableVectorContractLowering = true;
-  bool enableVectorToSCFConversion = true;
+  /// Vector lowering operations may result in surprising behavior when
+  /// composing multiple codegen strategies and must be enabled explicitly.
+  bool enableVectorTransferPartialRewrite = false;
+  bool enableVectorContractLowering = false;
+  bool enableVectorToSCFConversion = false;
 };
 
 /// Codegen strategy controls how a Linalg op is progressively lowered.

diff  --git a/mlir/test/lib/Dialect/Linalg/TestLinalgCodegenStrategy.cpp b/mlir/test/lib/Dialect/Linalg/TestLinalgCodegenStrategy.cpp
index 658eb8a61ac98..8f2cd6c689a0d 100644
--- a/mlir/test/lib/Dialect/Linalg/TestLinalgCodegenStrategy.cpp
+++ b/mlir/test/lib/Dialect/Linalg/TestLinalgCodegenStrategy.cpp
@@ -148,6 +148,9 @@ void TestLinalgCodegenStrategy::runStrategy<LinalgOp>(
               .setAlignment(16)
               .setUseFullTileBuffersByDefault(registerPromoteFullTile))
       .vectorizeIf(vectorize, anchorOpName)
+      .setEnableVectorTransferPartialRewrite(true)
+      .setEnableVectorContractLowering(true)
+      .setEnableVectorToSCFConversion(true)
       .setVectorTransformsOptions(
           vector::VectorTransformsOptions()
               .setVectorTransformsOptions(vectorContractLowering)
@@ -177,6 +180,9 @@ void TestLinalgCodegenStrategy::runStrategy(
               .setAlignment(16)
               .setUseFullTileBuffersByDefault(registerPromoteFullTile))
       .template vectorizeIf<OpType>(vectorize)
+      .setEnableVectorTransferPartialRewrite(true)
+      .setEnableVectorContractLowering(true)
+      .setEnableVectorToSCFConversion(true)
       .setVectorTransformsOptions(
           vector::VectorTransformsOptions()
               .setVectorTransformsOptions(vectorContractLowering)