[Mlir-commits] [mlir] [MLIR][Vector] Fix transferOps optimization inside maskOp (PR #90835)

[Andrzej Warzyński]

================
@@ -59,6 +59,37 @@ func.func @permutation_with_mask_transfer_write_scalable(%arg0: vector<4x[8]xi16
 
     return
 }
+
+
+#map = affine_map<(d0)[s0] -> (-d0 + s0, 4)>
+#map1 = affine_map<(d0, d1) -> (d0, 0, d1)>
+// CHECK-LABEL: func @masked_permutation_transfer_read
+//  CHECK-SAME:        %[[ARG_0:.*]]: tensor<?x1xf32>,
+//  CHECK-SAME:        %[[ARG_1:.*]]: vector<4x1xi1>
+//       CHECK: vector.transfer_read %[[ARG_0]]{{.*}}: tensor<?x1xf32>, vector<4x4x1xf32> } : vector<4x1xi1> -> vector<4x4x1xf32>
+func.func @masked_permutation_transfer_read(%arg0: tensor<?x1xf32>, %mask : vector<4x1xi1>) {
+  %cst = arith.constant 0.000000e+00 : f32
+  %c0 = arith.constant 0 : index
+  %3 = vector.mask %mask { vector.transfer_read %arg0[%c0, %c0], %cst {permutation_map = #map1} : tensor<?x1xf32>, vector<4x4x1xf32> } : vector<4x1xi1> -> vector<4x4x1xf32>
+  call @dostuff(%3) : (vector<4x4x1xf32>) -> ()
----------------
banach-space wrote:

Is this to prevent DCE? You can use `  "test.some_use"(%3) : (vector<4x4x1xf32>) -> ()` instead ;-)

https://github.com/llvm/llvm-project/pull/90835