[Mlir-commits] [mlir] [MLIR][SCFToGPU] Guard operands before AffineApplyOp::create to avoid crash (PR #167959)

Mehdi Amini llvmlistbot at llvm.org
Mon Nov 17 06:30:33 PST 2025


================
@@ -673,3 +673,55 @@ func.func @nested_parallel_with_side_effect() {
 
 // CHECK: gpu.launch
 // CHECK-NOT: scf.parallel
+
+// -----
+
+// RUN: mlir-opt %s --convert-parallel-loops-to-gpu | FileCheck %s --check-prefix=CRASH-REG
----------------
joker-eph wrote:

Why this extra RUN here? Can't you just use CHECK as a prefix and rely on the existing execution? (same below)

https://github.com/llvm/llvm-project/pull/167959


More information about the Mlir-commits mailing list