[PATCH] D145392: [AArch64] Don't map llvm sqrt intrinsics to veclib functions

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm accepted this revision.
paulwalker-arm added a comment.

I suspect we'll want to replace `replace-intrinsics-with-veclib-sqrt.ll` with more exhaustive testing for all the intrinsics but that's a job for another day.  I've a couple of observations but otherwise the patch looks good.



================
Comment at: llvm/test/CodeGen/AArch64/replace-intrinsics-with-veclib-sqrt.ll:1
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-attributes
+; RUN: opt -vector-library=sleefgnuabi -replace-with-veclib -S < %s | FileCheck %s
----------------
Given the simplicity of the tests I doubt you need these extra arguments.


================
Comment at: llvm/test/CodeGen/AArch64/replace-intrinsics-with-veclib-sqrt.ll:4
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+
----------------
Is this required?


================
Comment at: llvm/test/CodeGen/AArch64/replace-intrinsics-with-veclib-sqrt.ll:53-55
+declare double @llvm.sqrt.f64(double) #0
+declare float @llvm.sqrt.f32(float) #0
+attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
----------------
These three are not used within the test and can be removed?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D145392/new/

https://reviews.llvm.org/D145392