[all-commits] [llvm/llvm-project] 81b2c2: AMDGPU: Use kill instruction to hint soft clause l...

[Matt Arsenault via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 81b2c23b77716e1074f3e84259321cb9b8c9b2eb
      https://github.com/llvm/llvm-project/commit/81b2c23b77716e1074f3e84259321cb9b8c9b2eb
  Author: Matt Arsenault <Matthew.Arsenault at amd.com>
  Date:   2021-02-26 (Fri, 26 Feb 2021)

  Changed paths:
    M llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
    M llvm/lib/Target/AMDGPU/SIPostRABundler.cpp
    M llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i128.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement-stack-lower.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i16.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.large.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.div.fmas.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.update.dpp.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/load-constant.96.ll
    M llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll
    M llvm/test/CodeGen/AMDGPU/disable_form_clauses.ll
    M llvm/test/CodeGen/AMDGPU/ds_write2.ll
    M llvm/test/CodeGen/AMDGPU/gfx-callable-argument-types.ll
    M llvm/test/CodeGen/AMDGPU/limit-soft-clause-reg-pressure.mir
    M llvm/test/CodeGen/AMDGPU/memory-legalizer-global-agent.ll
    M llvm/test/CodeGen/AMDGPU/memory-legalizer-global-singlethread.ll
    M llvm/test/CodeGen/AMDGPU/memory-legalizer-global-system.ll
    M llvm/test/CodeGen/AMDGPU/memory-legalizer-global-wavefront.ll
    M llvm/test/CodeGen/AMDGPU/memory-legalizer-global-workgroup.ll
    M llvm/test/CodeGen/AMDGPU/memory_clause.ll
    M llvm/test/CodeGen/AMDGPU/memory_clause.mir
    M llvm/test/CodeGen/AMDGPU/postra-bundle-memops.mir
    M llvm/test/CodeGen/AMDGPU/reserved-reg-in-clause.mir
    M llvm/test/CodeGen/AMDGPU/soft-clause-dbg-value.mir
    M llvm/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll
    M llvm/test/CodeGen/AMDGPU/splitkit-getsubrangeformask.ll
    M llvm/test/CodeGen/AMDGPU/unallocatable-bundle-regression.mir
    M llvm/test/CodeGen/AMDGPU/vector_shuffle.packed.ll
    M llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll

  Log Message:
  -----------
  AMDGPU: Use kill instruction to hint soft clause live ranges

Previously we would use a bundle to hint the register allocator to not
overwrite the pointers in a sequence of loads to avoid breaking soft
clauses. This bundling was based on a fuzzy register pressure
heuristic, so we could not guarantee using more registers than are
really available. This would result in register allocator failing on
unsatisfiable bundles. Use a kill to artificially extend the live
ranges, so we can always succeed at register allocation even if it
means extra spills in the worst case.

This seems to capture most of the benefit of the bundle while avoiding
most of the risk presented by the bundle. However the lit tests do
show a handful of regressions. In some cases with sequences of
volatile loads, unused load components end up getting reallocated to
the next load which forces a wait between. There are also a few small
scheduling regressions where a hazard used to be avoided, and one
spill torture test which for some reason nearly doubles the stack
usage. There is also a bit of noise from leftover kills (it may make
sense for post-RA pseudos to strip all of these out).