[llvm] [AArch64] Don't run loop-idiom-vectorize pass in the O0 pipeline (PR #156802)

[David Green via llvm-commits]

================
@@ -0,0 +1,10 @@
+; RUN: opt -mtriple=aarch64 -S -passes='default<O0>' -print-pipeline-passes < %s | FileCheck --check-prefix=O0 %s
+; RUN: opt -mtriple=aarch64 -S -passes='default<O2>' -print-pipeline-passes < %s | FileCheck %s
+
+; CHECK: loop-idiom-vectorize
+; O0-NOT: loop-idiom-vectorize
----------------
davemgreen wrote:

I was thinking of testing the entire pipeline for -O0 as it should be very small, and would prevent us from adding extra passes to it again. The other optimization levels are a bit long for that.

https://github.com/llvm/llvm-project/pull/156802