[PATCH] D107902: [AArch64][SME] Disable NEON in streaming mode
Cullen Rhodes via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Aug 13 04:53:11 PDT 2021
c-rhodes added inline comments.
================
Comment at: llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp:60
- return createAArch64MCSubtargetInfoImpl(TT, CPU, /*TuneCPU*/ CPU, FS);
+ // Most of the NEON instruction set isn't supported in streaming mode on SME
+ // targets, but if no CPU is specified the default is "generic", which
----------------
c-rhodes wrote:
> paulwalker-arm wrote:
> > Does this need to be dependent on `CPU="Generic"`? I'm thinking you may still need to disable NEON even when targeting a known cpu.
> >
> > I guess I expected some explicit opting in to the fact NEON will be disabled. That's to say I figured the user would explicitly ask for `+streaming-sve` and by not also explicitly asking for `+neon` then they are making a conscious choice to disable NEON.
> > Does this need to be dependent on `CPU="Generic"`? I'm thinking you may still need to disable NEON even when targeting a known cpu.
>
> CPUs with SME support will be defined without `FeatureNEON` in the features, so this won't be an issue?
>
> > I guess I expected some explicit opting in to the fact NEON will be disabled. That's to say I figured the user would explicitly ask for `+streaming-sve` and by not also explicitly asking for `+neon` then they are making a conscious choice to disable NEON.
>
> That's what this change implements?
> > Does this need to be dependent on `CPU="Generic"`? I'm thinking you may still need to disable NEON even when targeting a known cpu.
>
> CPUs with SME support will be defined without `FeatureNEON` in the features, so this won't be an issue?
This is incorrect, misunderstanding on my part. CPUs with SME support will also have NEON and as such FeatureNEON will be implied.
> > I guess I expected some explicit opting in to the fact NEON will be disabled. That's to say I figured the user would explicitly ask for `+streaming-sve` and by not also explicitly asking for `+neon` then they are making a conscious choice to disable NEON.
>
> That's what this change implements?
I've updated the patch to disable NEON regardless of CPU if `+streaming-sve` is specified, unless NEON is explicitly requested.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D107902/new/
https://reviews.llvm.org/D107902
More information about the llvm-commits
mailing list