[Mlir-commits] [mlir] [mlir][memref] Use array notation instead of permutation map for memref.transpose (PR #67880)

[Mehdi Amini]

================
@@ -2125,30 +2125,41 @@ def TensorStoreOp : MemRef_Op<"tensor_store",
 def MemRef_TransposeOp : MemRef_Op<"transpose", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
     Pure]>,
-    Arguments<(ins AnyStridedMemRef:$in, AffineMapAttr:$permutation)>,
+    Arguments<(ins AnyStridedMemRef:$in, DenseI64ArrayAttr:$permutation)>,
     Results<(outs AnyStridedMemRef)> {
   let summary = "`transpose` produces a new strided memref (metadata-only)";
   let description = [{
     The `transpose` op produces a strided memref whose sizes and strides
     are a permutation of the original `in` memref. This is purely a metadata
     transformation.
 
+    The permutation is given in the form of an array of indices following the rule:
+    `dim(result, i) = dim(input, permutation[i])`
+
     Example:
 
     ```mlir
-    %1 = memref.transpose %0 (i, j) -> (j, i) : memref<?x?xf32> to memref<?x?xf32, affine_map<(d0, d1)[s0] -> (d1 * s0 + d0)>>
+    %1 = memref.transpose %0 [1, 0] : memref<?x?xf32> to memref<?x?xf32, affine_map<(d0, d1)[s0] -> (d1 * s0 + d0)>>
     ```
----------------
joker-eph wrote:

We should give an example to be sure, and with 3 dimensions and static shapes, otherwise it's ambiguous.

What about replacing the example with:

```
    %1 = memref.transpose %0 [1, 2, 0] : memref<7x8x9xf32> to memref<8x9x7xf32>
```



https://github.com/llvm/llvm-project/pull/67880