[llvm] [llvm] Improve TLI for Darwin libsystem_m functions (PR #109479)

Tex Riddell via llvm-commits llvm-commits at lists.llvm.org
Fri Oct 25 16:17:54 PDT 2024


================
@@ -0,0 +1,43 @@
+; RUN: opt -vector-library=Darwin_libsystem_m -passes=inject-tli-mappings,loop-vectorize -S < %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
----------------
tex3d wrote:

I have a PR (#113637) adding atan2 to `isTriviallyVectorizable`, as well as adding missing massv cases for the intrinsic.  This was waiting behind other lowering PRs that have now merged.  I can rebase my PR on top of this one and make it about the massv change only once this merges.

I was trying to investigate what impact adding atan2 to `isTriviallyVectorizable` actually has, since all the supported veclib call vectorization occurs without it.  I can't seem to identify a test to modify or add which demonstrates/verifies the impact of the `isTriviallyVectorizable` change alone.

https://github.com/llvm/llvm-project/pull/109479


More information about the llvm-commits mailing list