[PATCH] D124284: [SLP]Try partial store vectorization if supported by target.

Mon Apr 25 10:56:48 PDT 2022

ABataev added a comment.

In D124284#3472397 <https://reviews.llvm.org/D124284#3472397>, @ABataev wrote:

> In D124284#3472395 <https://reviews.llvm.org/D124284#3472395>, @xbolva00 wrote:
>
>> Looks great. Could you also observe some runtime perf improvements?
>
> My system is pretty busy so the perf numbers would not be correct most probably.

Just an example:

  test-suite :: MultiSource/Benchmarks/llubenchmark/llu.test  10.33     17.52     69.7%

The test is not affected at all.

As to some long run tests:

  test-suite :: External/SPEC/CINT2006/401.bzip2/401.bzip2.test  28.09     29.60      5.4%
  test-suite :: External/SPEC/CINT2017rate/502.gcc_r/502.gcc_r.test  32.65     34.00      4.1%
  test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test  96.71    100.30      3.7%
  test-suite :: External/SPEC/CINT2017speed/605.mcf_s/605.mcf_s.test  49.65     50.82      2.4%
  test-suite :: External/SPEC/CINT2017rate/505.mcf_r/505.mcf_r.test  50.21     50.93      1.4%
  test-suite :: External/SPEC/CINT2017speed/602.gcc_s/602.gcc_s.test  33.50     31.27     -6.6%
  test-suite :: External/SPEC/CINT2017rate/520.omnetpp_r/520.omnetpp_r.test  60.73     55.16     -9.2%
  test-suite :: External/SPEC/CINT2017rate/557.xz_r/557.xz_r.test  36.94     32.76    -11.3%

The less %, the better. Geomean is -100%  but just like I said I would not trust these numbers.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D124284/new/

https://reviews.llvm.org/D124284