RISC-V LLVM sync-up call 12th Sep 2019

A group of us have been having weekly sync-up calls on RISC-V
backend development for a while. It's not so much about decision-making, just
general status update on WIP patches and discussing some wider ISA related
issues that are relevant to LLVM and where we'd like more feedback. I think
we've invited most active RISC-V backend contributors previously. If you've
been left off the list, then I apologise. To ensure anyone with an
interest knows of its existence and can drop in drop in, we thought it
made sense to advertise on llvm-dev.

Reminder: the purpose is to co-ordinate between active contributors. If you
have support questions etc then it's best to post to llvm-dev.

We have a call each Thursday at 4pm BST, via
<https://meet.google.com/ske-zcog-spp>. I appreciate if you haven't joined
previously that's no notice - but we look forward to talking next week
or beyond!

Today issues to discuss include:
* GCC torture suite support in the LLVM test suite
  * Still stuck on reviews and waiting on feedback re whether we need to
    support Makefile as well as CMake - see https://reviews.llvm.org/D66887
* Status of psABI sanity check
  * Now in separate repo
    <https://github.com/lowRISC/riscv-elf-psabi-sanity-checker>
* Use of register arithmetic
  * See <https://reviews.llvm.org/D67423>. Luís is updating the approach and
    will refresh the patch soon.
* Verifying MachineInstr immediates
  * See <https://reviews.llvm.org/D67397>. This approach seems worthwhile to
    protect against programming errors, and it might be worth discussing how
    it could be applied to other backends (or indeed, if there's a better way
    of achieving the same goal).
* Sanitizers patch
  * Needs a reviewer from a sanitizer expert rather than a RISC-V expert
    realy, but at first look seems sensible. Can anyone help out?
    <https://reviews.llvm.org/D66870>
* Statistics tracking in the backend
  * Sam posted a patch to track the number of compressed instructions.
    <https://reviews.llvm.org/D67495>. We'd be interested in ideas others
    might have on stats it might be useful to track in addition to this.

Best,

Alex