[Mlir-commits] [mlir] [mlir][vector] Fix crash on untraceable masks in getCompressedMaskOp (PR #207299)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Wed Jul 8 04:28:15 PDT 2026
================
@@ -109,3 +109,31 @@ func.func @vector_maskedstore_2d_i8_negative(%arg0: index, %arg1: index, %arg2:
// CHECK-LABEL: func @vector_maskedstore_2d_i8_negative
// CHECK: memref.alloc() : memref<3x8xi8>
// CHECK-NOT: i32
+
+// -----
+
+// The mask is a block argument, so its defining op cannot be traced back to a
+// supported mask-creation op. This must gracefully bail out instead of crashing
+// (see llvm/llvm-project#206928).
+
+func.func @vector_maskedstore_i8_block_arg_mask_negative(%arg0: memref<?xi8>, %mask: vector<8xi1>, %value: vector<8xi8>) {
----------------
marquisburg wrote:
Done, renamed all the negative tests in the file to use the correct prefix as per the guidelines!
https://github.com/llvm/llvm-project/pull/207299
More information about the Mlir-commits
mailing list