[PATCH] D120257: [AArch64][SME] Replace +streaming-sve feature with +streaming-compatible-<feature>.
Cullen Rhodes via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 22 12:29:47 PST 2022
c-rhodes added inline comments.
================
Comment at: llvm/lib/Target/AArch64/AArch64.td:441-460
+def FeatureStreamingCompatibleNEON
+ : SubtargetFeature<"streaming-compatible-neon",
+ "HasStreamingCompatibleNEON", "true",
+ "Enable subset of NEON instructions that are valid in "
+ "both streaming SVE mode as well as regular mode",
+ [FeatureFPARMv8, FeatureFullFP16]>;
+
----------------
paulwalker-arm wrote:
> Do we need to split the feature flags based on other feature flag boundaries? Why not just `FeatureStreamingCompatible`? I'm not saying the above is wrong, but it does seem like an extra level of indirection so just trying to understand the reasoning.
> Do we need to split the feature flags based on other feature flag boundaries? Why not just `FeatureStreamingCompatible`? I'm not saying the above is wrong, but it does seem like an extra level of indirection so just trying to understand the reasoning.
I agree, naming aside I don't understand what the limitations of the existing feature flag are and why this extra granularity at the flag level is required.
================
Comment at: llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCTargetDesc.cpp:60-69
- // Most of the NEON instruction set isn't supported in streaming mode on SME
- // targets, disable NEON unless explicitly requested.
- bool RequestedNEON = FS.contains("neon");
- bool RequestedStreamingSVE = FS.contains("streaming-sve");
- MCSubtargetInfo *STI =
- createAArch64MCSubtargetInfoImpl(TT, CPU, /*TuneCPU*/ CPU, FS);
- if (RequestedStreamingSVE && !RequestedNEON &&
----------------
why is this removed?
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D120257/new/
https://reviews.llvm.org/D120257
More information about the llvm-commits
mailing list