[all-commits] [llvm/llvm-project] 54dafd: [AArch64] Move FeatureSpecRestrict into core 8.0-R...

[Simon Tatham via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 54dafd38c5c6f36e37b266f5f1a588ac11e6db5d
      https://github.com/llvm/llvm-project/commit/54dafd38c5c6f36e37b266f5f1a588ac11e6db5d
  Author: Simon Tatham <simon.tatham at arm.com>
  Date:   2022-03-07 (Mon, 07 Mar 2022)

  Changed paths:
    M llvm/lib/Target/AArch64/AArch64.td
    M llvm/test/MC/Disassembler/AArch64/armv8.5a-specrestrict.txt

  Log Message:
  -----------
  [AArch64] Move FeatureSpecRestrict into core 8.0-R architecture.

It was included in HasV8_0rOps when D88660 first introduced that
architecture definition. In D118045 I moved it out of there and into
ProcessorFeatures.R82, so that -mcpu=cortex-r82 would continue to
behave the same as before but -march=armv8-r would include only the
mandatory parts of the architecture.

In fact, that was a mistake. Firstly, Cortex-R82 _doesn't_ implement
that feature, so it makes no sense to deliberately enable it for that
CPU in particular. But also, it's an extension that only adds system
registers, and we're generally more relaxed about where we enable
those (because kernel developers find it useful to write sysreg-access
instructions after runtime checking, and because sysreg accesses
aren't manufactured during code generation so the risk is small).

So, in line with that usual AArch64 policy, FeatureSpecRestrict ought
to be considered part of 8.0-R for LLVM purposes. So I'm moving it
back into HasV8_0rOps, where it started out.

Reviewed By: lenary

Differential Revision: https://reviews.llvm.org/D120830