[ElimLoadStore] created -eliminate-affine-load-store

[kumasento]

func @foo(%A: memref<?xf32>, %i: index, %a: f32) -> f32 {
  affine.store %a, %A[%i] : memref<?xf32>
  %v = affine.load %A[%i] : memref<?xf32>
  affine.store %a, %A[%i] : memref<?xf32>
  return %v : f32
}

will be squashed into

func @foo(%A: memref<?xf32>, %i: index, %a: f32) -> f32 {
  affine.store %a, %A[%i] : memref<?xf32>
  return %a : f32
}

---

func @foo(%A: memref<?xi32>) {
  %fst = affine.load %A[0] : memref<?xi32>
  %fst2 = affine.load %A[0] : memref<?xi32>
  // Should only leave one.

  affine.for %i = 1 to 10 {
    %cur1 = affine.load %A[%i] : memref<?xi32>
    %cur2 = affine.load %A[%i] : memref<?xi32>
    %next = affine.load %A[%i + 1] : memref<?xi32> // should not remove

    %doubled = arith.addi %cur1, %cur2 : i32
    %sum = arith.addi %next, %doubled : i32
    %final = arith.addi %fst2, %sum : i32

    affine.store %final, %A[100 + %i] : memref<?xi32>
  }

  return
}

will be simplified to

func @foo(%A: memref<?xi32>) {
  %fst = affine.load %A[0] : memref<?xi32>

  affine.for %i = 1 to 10 {
    %cur1 = affine.load %A[%i] : memref<?xi32>
    %next = affine.load %A[%i + 1] : memref<?xi32> // should not remove

    %doubled = arith.addi %cur1, %cur1 : i32
    %sum = arith.addi %next, %doubled : i32
    %final = arith.addi %fst, %sum : i32

    affine.store %final, %A[100 + %i] : memref<?xi32>
  }

  return
}