[Mlir-commits] [mlir] fc64a16 - [mlir] Use rewriter in linalg Detensorize

[Tres Popp]

Author: Tres Popp
Date: 2021-12-15T11:28:18+01:00
New Revision: fc64a164ec5830cb668008d2e3c62e422a07dfae

URL: https://github.com/llvm/llvm-project/commit/fc64a164ec5830cb668008d2e3c62e422a07dfae
DIFF: https://github.com/llvm/llvm-project/commit/fc64a164ec5830cb668008d2e3c62e422a07dfae.diff

LOG: [mlir] Use rewriter in linalg Detensorize

This is to allow rollbacks on failures of dialect lowering to succeed.

Differential Revision: https://reviews.llvm.org/D115789

Added: 
    

Modified: 
    mlir/lib/Dialect/Linalg/Transforms/Detensorize.cpp

Removed: 
    


################################################################################
diff  --git a/mlir/lib/Dialect/Linalg/Transforms/Detensorize.cpp b/mlir/lib/Dialect/Linalg/Transforms/Detensorize.cpp
index 276d4b728017..0257014bfbcd 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Detensorize.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Detensorize.cpp
@@ -73,7 +73,7 @@ class DetensorizeGenericOp : public OpConversionPattern<GenericOp> {
 
     // Split the op's region before the op. This way, we have a clear insertion
     // point in which the op can be inlined.
-    Block *newBlock = originalBlock->splitBlock(op);
+    Block *newBlock = rewriter.splitBlock(originalBlock, Block::iterator(op));
     rewriter.inlineRegionBefore(op.region(), newBlock);
     // Now that op's region is inlined, the operands of its YieldOp are mapped
     // to the materialized target values. Therefore, we can replace the op's