[llvm] [LAA] Increase vectorize memory check threshold to 192 (PR #228382)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-analysis

Author: Igor Kirillov (igogo-x86)

<details>
<summary>Changes</summary>

Together with the proposed switch to
-stencil-runtime-check-merge=auto in #<!-- -->228373, it allows two of the three hot loops in 709.cactus (SPEC2026) to vectorize.

In testing with stencil merging enabled and a threshold above 180, the 709.cactus_r score improved by approximately 16% on Neoverse-V2.

---
Full diff: https://github.com/llvm/llvm-project/pull/228382.diff


1 Files Affected:

- (modified) llvm/lib/Analysis/LoopAccessAnalysis.cpp (+1-1) 


``````````diff
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index 367616e8e1e74..a7b0929ff49bc 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -99,7 +99,7 @@ static cl::opt<unsigned, true> VectorizeMemoryCheckThreshold(
     "vectorize-memory-check-threshold", cl::Hidden,
     cl::desc("The maximum allowed number of runtime memory checks"),
     cl::location(VectorizerParams::VectorizeMemoryCheckThreshold),
-    cl::init(128));
+    cl::init(192));
 unsigned VectorizerParams::VectorizeMemoryCheckThreshold;
 
 /// The maximum iterations used to merge memory checks

``````````

</details>


https://github.com/llvm/llvm-project/pull/228382