[Mlir-commits] [mlir] [mlir][Vector] Add support for poison indices to `Extract/IndexOp` (PR #123488)

[Jakub Kuderski]

================
@@ -132,6 +132,26 @@ func.func @extract_from_create_mask_dynamic_position(%dim0: index, %index: index
 
 // -----
 
+// CHECK-LABEL: @extract_scalar_poison_idx
+func.func @extract_scalar_poison_idx(%a: vector<4x5xf32>) -> f32 {
+  //  CHECK-NOT: vector.extract
+  // CHECK-NEXT: ub.poison : f32
+  %0 = vector.extract %a[-1, 0] : f32 from vector<4x5xf32>
----------------
kuhar wrote:

Should we have one test with multiple poison indices, e.g.: `[-1, -1, -1]`? This works with the existing implementation, but I wonder if it would make sense to be extra cautious and make sure that nothing in the future makes an accidental assumption of these ops having at most one poison index.

https://github.com/llvm/llvm-project/pull/123488