[PATCH] D148905: [Vectorize] Fix vectorization and folding of llvm.is.fpclass

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/lib/Analysis/VectorUtils.cpp:131
   default:
-    return false;
+    return OpdIdx == -1;
   }
----------------
fhahn wrote:
> I'm not sure about the change in the default here, IICU this is now considering any other intrinsic to have the return type overloaded?
Yes. All callers of `isVectorIntrinsicWithOverloadTypeAtArg` were already making this assumption implicitly. I have changed them to make an explicit check for `isVectorIntrinsicWithOverloadTypeAtArg(ID, -1)`.


================
Comment at: llvm/test/Transforms/LoopVectorize/ARM/is_fpclass.ll:4
+
+target triple = "aarch64-unknown-linux-gnu"
+
----------------
fhahn wrote:
> The triple doesn't match the sub-directory, `aarch64-` needs to be in the `AArch64` subdirectory.
> 
> But does the test need to be target specific in the first place? Could you just use `-force-vector-width=4 -force-vector-interleave=1` or something, without a triple?
> 
> Also, would be good to add the test separately.
Thanks. I didn't know how to remove the triple. I will try the -force-vector-* options.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D148905/new/

https://reviews.llvm.org/D148905