[PATCH] D41599: [X86] Lowering X86 avx512 sqrt intrinsics to IR - LLVM

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: test/CodeGen/X86/avx512-intrinsics-upgrade.ll:4
 
+declare <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone
+
----------------
tkrupa wrote:
> craig.topper wrote:
> > This patch doesn't appear to have removed the scalar intrinsics. I dont' see any AutoUpgrade code or removal from X86InstrinsicsInfo.h
> You're right, they only get lowered in clang.  I gave the reasoning in the comment to the last upload.
> Is it enough to just move these 4 tests back to test/CodeGen/X86/avx512-intrinsics.ll or is it crucial to also lower them in LLVM part?
You can just move them back. But if clang isn't using them, make sure the GCCBuiltin is removed from IntrinsicsX86.td and leave a FIXME saying that they can be removed. There are lot of FIXMEs like that in that file already.


Repository:
  rL LLVM

https://reviews.llvm.org/D41599