[llvm] [Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs (PR #113637)

Tex Riddell via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 6 17:33:03 PST 2024


tex3d wrote:

@farzonl 

> * Add libfunc mapping to intrinsic in `llvm/lib/Analysis/ValueTracking.cpp` `llvm::getIntrinsicForCallSite`

Added.

> * Add `atan2` name check in `llvm/include/llvm/Analysis/TargetTransformInfoImpl.h` `isLoweredToCall`

Added, but it had no test impact, so I'm looking for feedback on how this can be tested.

> * I noticed there were some examples that use a amdlibm vector form below. Should we do a follow on PR that adds atan2 amdlibm and updates `llvm/test/Transforms/LoopVectorize/X86/amdlibm-calls.ll`?

It turns out these example files have all the code under `#if 0`, and there doesn't appear to be any declarations of these vector operations in the header.  I'll leave this for a later update when the operations are added to amdlibm.

https://github.com/llvm/llvm-project/pull/113637


More information about the llvm-commits mailing list