[mlir] skip 0-D vector in BubbleUpBitCastForStridedSliceInsert

[python3kgae]

Summary

InsertStridedSliceOp will not get 0-D vector result, so just skip 0-D vector in BubbleUpBitCastForStridedSliceInsert.
Fixes #60214 https://github.com/llvm/llvm-project/issues/60214

[python3kgae]

ping.

[ThomasRaoux]

mlir/test/Dialect/Vector/vector-cast-0D.mlir

6 ↗

(On Diff #491208)

I don't think it is worth adding a new file for this. Can you move this test to vector-transforms.mlir?

[python3kgae]

Move test to vector-transforms.mlir

[python3kgae]

Fix wrong history.

[python3kgae]

mlir/test/Dialect/Vector/vector-cast-0D.mlir

6 ↗

(On Diff #491208)

Fixed.

[ThomasRaoux]

LGTM

mlir/lib/Dialect/Vector/Transforms/VectorTransforms.cpp

struct BubbleUpBitCastForStridedSliceInsert
     : public OpRewritePattern<vector::BitCastOp> {
   using OpRewritePattern::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::BitCastOp bitcastOp,
                                 PatternRewriter &rewriter) const override {
     VectorType castSrcType = bitcastOp.getSourceVectorType();
     VectorType castDstType = bitcastOp.getResultVectorType();
     assert(castSrcType.getRank() == castDstType.getRank());
+    // Skip 0-D vector which will not from InsertStridedSliceOp.
+    if (castSrcType.getRank() == 0)
+      return failure();
 
     int64_t castSrcLastDim = castSrcType.getShape().back();
     int64_t castDstLastDim = castDstType.getShape().back();
     // Require casting to less elements for now; other cases to be implemented.
     if (castSrcLastDim < castDstLastDim)
       return failure();
 
     assert(castSrcLastDim % castDstLastDim == 0);

mlir/test/Dialect/Vector/vector-transforms.mlir

 // CHECK-LABEL: func @bubble_up_bitcast_in_strided_slice_insert_larger_odd_shape
 func.func @bubble_up_bitcast_in_strided_slice_insert_larger_odd_shape(%dst: vector<8xf16>, %src: vector<3xf16>) -> vector<4xf32> {
   // CHECK: vector.insert_strided_slice
   // CHECK-NEXT: vector.bitcast
   %0 = vector.insert_strided_slice %src, %dst {offsets = [0], strides = [1]} : vector<3xf16> into vector<8xf16>
   %cast = vector.bitcast %0: vector<8xf16> to vector<4xf32>
   return %cast: vector<4xf32>
 }
+
+// Make sure not crash on 0-D vector.
+// CHECK-LABEL:func.func @vec_0D
+// CHECK-NEXT:vector.bitcast
+func.func @vec_0D(%arg0: vector<f32>) -> vector<i32> {
+  %0 = vector.bitcast %arg0 : vector<f32> to vector<i32>
+  return %0 : vector<i32>
+}