[Mlir-commits] [mlir] 45aaa67 - [mlir][tensor] Fix one-shot bufferization of tensor.reshape.

[Ingo Müller]

Author: Ingo Müller
Date: 2023-05-26T09:39:49Z
New Revision: 45aaa67fceadaeb3bf76bab21d36c1f337d97491

URL: https://github.com/llvm/llvm-project/commit/45aaa67fceadaeb3bf76bab21d36c1f337d97491
DIFF: https://github.com/llvm/llvm-project/commit/45aaa67fceadaeb3bf76bab21d36c1f337d97491.diff

LOG: [mlir][tensor] Fix one-shot bufferization of tensor.reshape.

I believe that the previous implementation did not work on any input. It
called getMemRefType with `layout = {}`, presumably with the intention
to create a MemrefType with identity layout. However, the implementation
of that function returns a MemrefType with *unknown* layout if it is
provided with a default-constructed layout attribute. This patch uses
getMemRefTypeWithStaticIdentityLayout instead, with has identical
behavior except for the case of a default-constructed layout, which it
passes on as-is to the MemrefType.

This problem did not surface in the test because tensor.reshape was not
tested with -one-shot-bufferize. This patch introduces a test copied
from the tests for -tesnor-bufferize adapted in as follows: since the
test is run with "bufferize-function-boundaries", a tensor that is
passed into the function is bufferized into a memref with unknown
layout, which wouldn't be a valid intput for memref.reshape, so the
tests now uses a tensor constructed with arith.constant inside of the
function.

Reviewed By: springerm

Differential Revision: https://reviews.llvm.org/D151544

Added: 
    

Modified: 
    mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
    mlir/test/Dialect/Tensor/one-shot-bufferize.mlir

Removed: 
    


################################################################################
diff  --git a/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp b/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
index 545a9d09c4aba..1a4fc3bb5bbad 100644
--- a/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
+++ b/mlir/lib/Dialect/Tensor/Transforms/BufferizableOpInterfaceImpl.cpp
@@ -992,8 +992,8 @@ struct ReshapeOpInterface
         getBuffer(rewriter, reshapeOp.getShape(), options);
     if (failed(srcBuffer) || failed(shapeBuffer))
       return failure();
-    auto resultMemRefType = getMemRefType(
-        reshapeOp.getResult(), options, /*layout=*/{},
+    auto resultMemRefType = getMemRefTypeWithStaticIdentityLayout(
+        reshapeOp.getResult().getType(),
         cast<BaseMemRefType>(srcBuffer->getType()).getMemorySpace());
     replaceOpWithNewBufferizedOp<memref::ReshapeOp>(
         rewriter, op, resultMemRefType, *srcBuffer, *shapeBuffer);

diff  --git a/mlir/test/Dialect/Tensor/one-shot-bufferize.mlir b/mlir/test/Dialect/Tensor/one-shot-bufferize.mlir
index a4c868cf35d5c..399fd05139a52 100644
--- a/mlir/test/Dialect/Tensor/one-shot-bufferize.mlir
+++ b/mlir/test/Dialect/Tensor/one-shot-bufferize.mlir
@@ -398,3 +398,21 @@ func.func @parallel_insert_slice_source_out_of_place(%in: tensor<1xf32>, %out: t
   // CHECK: }
   return
 }
+
+// -----
+
+// CHECK-LABEL: func @tensor.reshape(
+func.func @tensor.reshape() -> tensor<2x2x5xf32> {
+  // CHECK-DAG: %[[M1:.*]] = memref.cast %{{.*}} : memref<2x10xf32> to memref<?x10xf32>
+  %t1_static = arith.constant dense<0.> : tensor<2x10xf32>
+  %t1 = tensor.cast %t1_static : tensor<2x10xf32> to tensor<?x10xf32>
+
+  // CHECK: %[[SHAPE:.*]] = memref.get_global @{{.*}} : memref<3xi64>
+  %shape = arith.constant dense<[2, 2, 5]> : tensor<3xi64>
+
+  // CHECK: %[[RESHAPED:.*]] = memref.reshape %[[M1]](%[[SHAPE]]) : (memref<?x10xf32>, memref<3xi64>) -> memref<2x2x5xf32>
+  %reshaped = tensor.reshape %t1(%shape) : (tensor<?x10xf32>, tensor<3xi64>) -> tensor<2x2x5xf32>
+
+  // CHECK: return %[[RESHAPED]]
+  return %reshaped : tensor<2x2x5xf32>
+}