The semantics of AffineMap in MemRef

[QiJune]

Here is an example:

#imapA = (d0, d1) [s0] -> (d0 + s0, d1)
%o = 1
%i = 0
%j = 0
%A = alloc()[%o] : memref<16x32xf32, #imapA, memspace0>
%v = load %A[%i, %j] : memref<16x32xf32, #imapA, memspace0>

It allocates a two dimension array, A[16][32].
The index of load is set to (0, 0).
Since there is an affinemap, %o is set to 1, the load operation will return A[1][0] actually.

Is my understanding correct? Thanks!

[joker-eph]

Unless we should track fixing the documentation with this bug, I think the best place to ask question and/or discuss such things is the mailing: https://groups.google.com/a/tensorflow.org/forum/#!forum/mlir

To answer your question, imapA is more exactly a semi-affine map and used here as a layout-map according to https://github.com/tensorflow/mlir/blob/master/g3doc/LangRef.md#layout-map
(your understanding of the indexing seems correct to me otherwise).

Your example in a form that would parse today would be:

#imapA = (d0, d1) [s0] -> (d0 + s0, d1)
func @test() {
  %o = constant 1 : index
  %i = constant 0 : index
  %j = constant 0 : index
  %A = alloc()[%o] : memref<16x32xf32, #imapA, 0>
  %v = load %A[%i, %j] : memref<16x32xf32, #imapA, 0>
  return
}