[llvm-dev] TriviallyVectorizable Attribute in TD file
Craig Topper via llvm-dev
llvm-dev at lists.llvm.org
Tue Nov 24 13:44:11 PST 2020
For the second part, I don't think most targets use type overloaded
intrinsics. For sure X86 doesn't. So in general they can't be vectorized by
just changing the type from scalar to vector. They would need to be
remapped to a different intrinsic name. The vectorize would also need to
avoid creating a type that the backend isn't prepared to see on a target
specific intrinsic.
~Craig
On Tue, Nov 24, 2020 at 1:34 PM Mark Schimmel via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> It seems wrong to me that VectorUtils.cpp contains this switch statement:
>
>
>
> /// Return true if all of the intrinsic's arguments and return type are
> scalars
>
> /// for the scalar form of the intrinsic, and vectors for the vector form
> of the
>
> /// intrinsic (except operands that are marked as always being scalar by
>
> /// hasVectorInstrinsicScalarOpd).
>
> bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
>
> switch (ID) {
>
> case Intrinsic::bswap: // Begin integer bit-manipulation.
>
> case Intrinsic::bitreverse:
>
>
>
> First it would be easy to create an attribute for this that is defined in
> the TD file similar to others (IntrNoMem, IntrSpeculatable, IntrWillReturn,
> Commutative).
>
> Second, it should apply to target intrinsics and allow them to vectorize
> also.
>
>
>
> // Intrinsic isTriviallyVectorizable
>
> def TriviallyVectorizable : IntrinsicProperty;
>
>
>
> …
>
>
>
> Would you be open to a patch for this functionality?
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20201124/7d48f98c/attachment.html>
More information about the llvm-dev
mailing list