[all-commits] [llvm/llvm-project] 15e20d: [Utils][x86] add an option to reduce scrubbing of ...

[RotateRight via All-commits]

  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: 15e20dcb8f9decf1928871d562a3724e8cc1e343
      https://github.com/llvm/llvm-project/commit/15e20dcb8f9decf1928871d562a3724e8cc1e343
  Author: Sanjay Patel <spatel at rotateright.com>
  Date:   2020-02-20 (Thu, 20 Feb 2020)

  Changed paths:
    M llvm/test/CodeGen/X86/avx-splat.ll
    M llvm/test/CodeGen/X86/extractelement-load.ll
    M llvm/utils/UpdateTestChecks/asm.py
    M llvm/utils/update_llc_test_checks.py

  Log Message:
  -----------
  [Utils][x86] add an option to reduce scrubbing of shuffles with memops

I was drafting a patch that would increase broadcast load usage,
but our shuffle scrubbing makes it impossible to see if the memory
operand offset was getting created correctly. I'm proposing to make
that an option (defaulted to 'off' for now to reduce regression
test churn).

The updated files provide examples of tests where we can now verify
that the pointer offset for a loaded memory operand is correct. We
still have stack and constant scrubbing that can obscure the operand
even if we don't scrub the entire instruction.

Differential Revision: https://reviews.llvm.org/D74775