[Mlir-commits] [mlir] [mlir][vector] Add mask elimination transform (PR #99314)

[Benjamin Maxwell]

================
@@ -0,0 +1,138 @@
+// RUN: mlir-opt %s -split-input-file -test-eliminate-vector-masks  | FileCheck %s
+
+// This tests a general pattern the vectorizer tends to emit.
+
+// CHECK-LABEL: @eliminate_redundant_masks_through_insert_and_extracts
+// CHECK: %[[ALL_TRUE_MASK:.*]] = arith.constant dense<true> : vector<[4]xi1>
+// CHECK: vector.transfer_read {{.*}} %[[ALL_TRUE_MASK]]
+// CHECK: vector.transfer_write {{.*}} %[[ALL_TRUE_MASK]]
+func.func @eliminate_redundant_masks_through_insert_and_extracts(%tensor: tensor<1x1000xf32>) {
+  %c0 = arith.constant 0 : index
+  %c4 = arith.constant 4 : index
+  %c1000 = arith.constant 1000 : index
+  %c0_f32 = arith.constant 0.0 : f32
+  %vscale = vector.vscale
+  %c4_vscale = arith.muli %vscale, %c4 : index
+  %extracted_slice_0 = tensor.extract_slice %tensor[0, 0] [1, %c4_vscale] [1, 1] : tensor<1x1000xf32> to tensor<1x?xf32>
+  %output_tensor = scf.for %i = %c0 to %c1000 step %c4_vscale iter_args(%arg = %extracted_slice_0) -> tensor<1x?xf32> {
----------------
MacDue wrote:

not really :+1: 

https://github.com/llvm/llvm-project/pull/99314