[mlir][vector] Check minor identity map in FoldExtractSliceIntoTransferRead

[antiagainst]

Summary

vecotr.transfer_read ops with minor identity indexing map is rank
reducing, with implicit leading unit dimensions. This should be
a natural extension to support in addition to full identity indexing
maps.

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

public:
 
   LogicalResult matchAndRewrite(TransferReadOp xferOp,
                                 PatternRewriter &rewriter) const override {
     // TODO: support 0-d corner case.
     if (xferOp.getTransferRank() == 0)
       return failure();
     if (xferOp.hasOutOfBoundsDim())
       return failure();
-    if (!xferOp.getPermutationMap().isIdentity())
+    if (!xferOp.getPermutationMap().isMinorIdentity())
       return failure();
     if (xferOp.getMask())
       return failure();
     auto extractOp = xferOp.getSource().getDefiningOp<tensor::ExtractSliceOp>();
     if (!extractOp)
       return failure();
     if (!extractOp.hasUnitStride())
       return failure();

mlir/test/Dialect/Vector/canonicalize.mlir

func.func @transfer_read_of_extract_slice(%t : tensor<?x?xf32>, %s1 : index, %s2 : index) -> vector<5x6xf32> {
   %cst = arith.constant 0.0 : f32
   %0 = tensor.extract_slice %t[5, %s1] [10, %s2] [1, 1] : tensor<?x?xf32> to tensor<10x?xf32>
   %1 = vector.transfer_read %0[%c3, %c4], %cst {in_bounds = [true, true]} : tensor<10x?xf32>, vector<5x6xf32>
   return %1 : vector<5x6xf32>
 }
 
 // -----
 
+// CHECK-LABEL: func @transfer_read_of_extract_slice(
+//  CHECK-SAME:     %[[t:.*]]: tensor<?x?xf32>, %[[s1:.*]]: index, %[[s2:.*]]: index
+//   CHECK-DAG:   %[[c4:.*]] = arith.constant 4 : index
+//   CHECK-DAG:   %[[c8:.*]] = arith.constant 8 : index
+//       CHECK:   %[[add:.*]] = arith.addi %[[s1]], %[[c4]]
+//       CHECK:   %[[r:.*]] = vector.transfer_read %[[t]][%[[c8]], %[[add]]], %{{.*}} {in_bounds = [true]} : tensor<?x?xf32>, vector<6xf32>
+//       CHECK:   return %[[r]]
+func.func @transfer_read_of_extract_slice(%t : tensor<?x?xf32>, %s1 : index, %s2 : index) -> vector<6xf32> {
+  %c3 = arith.constant 3 : index
+  %c4 = arith.constant 4 : index
+  %cst = arith.constant 0.0 : f32
+  %0 = tensor.extract_slice %t[5, %s1] [10, %s2] [1, 1] : tensor<?x?xf32> to tensor<10x?xf32>
+  %1 = vector.transfer_read %0[%c3, %c4], %cst {in_bounds = [true]} : tensor<10x?xf32>, vector<6xf32>
+  return %1 : vector<6xf32>
+}
+
+// -----
+
 // CHECK-LABEL: func @transfer_read_of_extract_slice_rank_reducing(
 //  CHECK-SAME:     %[[t:.*]]: tensor<?x?x?xf32>, %[[s1:.*]]: index, %[[s2:.*]]: index
 //   CHECK-DAG:   %[[c3:.*]] = arith.constant 3 : index
 //   CHECK-DAG:   %[[c5:.*]] = arith.constant 5 : index
 //   CHECK-DAG:   %[[c10:.*]] = arith.constant 10 : index
 //       CHECK:   %[[add:.*]] = arith.addi %[[s1]], %[[c3]]
 //       CHECK:   %[[r:.*]] = vector.transfer_read %[[t]][%[[c5]], %[[add]], %[[c10]]], %{{.*}} {in_bounds = [true, true]} : tensor<?x?x?xf32>, vector<5x6xf32>
 //       CHECK:   return %[[r]]