[Openmp-commits] [clang] [llvm] [mlir] [openmp] [Clang][OpenMP] Canonicalize Intra-tiles in Loop Tiling (PR #191114)

[Michael Kruse via Openmp-commits]

================
@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -verify -triple x86_64-unknown-linux-gnu -fclang-abi-compat=latest -std=c++17 -fopenmp -O2 -vectorize-loops -vectorize-slp -emit-llvm %s -o - | FileCheck %s
+// REQUIRES: x86-registered-target
+// expected-no-diagnostics
+
+// Standalone tile stays min-bounded (no body predicate) and vectorizes.
+
+// CHECK-LABEL: define {{.*}} @vec_tile(
+// CHECK: fadd <4 x float>
+// CHECK-NOT: omp.body.next
+extern "C" void vec_tile(float *a, float *b, int n) {
+#pragma omp tile sizes(64)
----------------
Meinersbur wrote:

This test is independent of the intra-tile loop being canonical. In needs to be affected e.g. by a `collapse(2)` or another tile to make a difference.

https://github.com/llvm/llvm-project/pull/191114