[all-commits] [llvm/llvm-project] 68469a: [LV] Disable runtime unrolling for vectorized loops.

[Florian Hahn via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 68469a80cb7468c577701d725a56a86454b8f0cb
      https://github.com/llvm/llvm-project/commit/68469a80cb7468c577701d725a56a86454b8f0cb
  Author: Florian Hahn <flo at fhahn.com>
  Date:   2023-01-06 (Fri, 06 Jan 2023)

  Changed paths:
    M llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
    M llvm/test/Transforms/LoopVectorize/ARM/pointer_iv.ll
    M llvm/test/Transforms/LoopVectorize/ARM/tail-folding-loop-hint.ll
    M llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll
    M llvm/test/Transforms/LoopVectorize/X86/already-vectorized.ll
    M llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
    M llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
    M llvm/test/Transforms/LoopVectorize/X86/invariant-load-gather.ll
    M llvm/test/Transforms/LoopVectorize/X86/invariant-store-vectorization.ll
    M llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll
    M llvm/test/Transforms/LoopVectorize/X86/metadata-enable.ll
    M llvm/test/Transforms/LoopVectorize/X86/tail_loop_folding.ll
    M llvm/test/Transforms/LoopVectorize/X86/uniform_mem_op.ll
    M llvm/test/Transforms/LoopVectorize/followup.ll
    M llvm/test/Transforms/LoopVectorize/if-pred-non-void.ll
    M llvm/test/Transforms/LoopVectorize/induction.ll
    M llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll
    M llvm/test/Transforms/LoopVectorize/invariant-store-vectorization-2.ll
    M llvm/test/Transforms/LoopVectorize/invariant-store-vectorization.ll
    M llvm/test/Transforms/LoopVectorize/memdep-fold-tail.ll
    M llvm/test/Transforms/LoopVectorize/optsize.ll
    M llvm/test/Transforms/LoopVectorize/pointer-select-runtime-checks.ll
    M llvm/test/Transforms/LoopVectorize/reduction-with-invariant-store.ll
    M llvm/test/Transforms/LoopVectorize/runtime-check.ll
    M llvm/test/Transforms/LoopVectorize/vectorize-once.ll
    M llvm/test/Transforms/PhaseOrdering/AArch64/hoisting-sinking-required-for-vectorization.ll
    M llvm/test/Transforms/PhaseOrdering/X86/excessive-unrolling.ll
    M llvm/test/Transforms/PhaseOrdering/X86/vdiv.ll

  Log Message:
  -----------
  [LV] Disable runtime unrolling for vectorized loops.

This patch adds metadata to disable runtime unrolling to the vectorized
loop. If runtime unrolling/interleaving is considered profitable, LV
will interleave the loop directly. There should be no need to perform
runtime unrolling at a later stage.

Note that we already add metadata to disable runtime unrolling to the
scalar loop after vectorization.

The additional unrolling unnecessarily increases code size and compile
time. In addition to that we have several bug reports of unncessary
runtime unrolling for vectorized loops, e.g. PR40961

Compile-time improvements:

  NewPM-O3: -1.04%
  NewPM-ReleaseThinLTO: -0.59%
  NewPM-ReleaseLTO-g: -0.97%

https://llvm-compile-time-tracker.com/compare.php?from=ce1be13a868d0f8afa367975558c1a6175cce33a&to=78bc2e67f22e9e10e61cdb6cdac4bb857d95eb1b&stat=instructions:u

Fixes #40306.

Reviewed By: lebedev.ri, nikic

Differential Revision: https://reviews.llvm.org/D115261