[PATCH] [MachineSink] Use the real post dominator tree
Jingyue Wu
jingyue at google.com
Mon Oct 6 15:19:26 PDT 2014
Compilation time from llvm-test-suite. Three columns are benchmark name, original compilation time (in seconds), and the slowdown/speedup caused by this change (negative means the new version is faster). I only picked the ones with compilation time longer than one second, because other data points are very noisy.
I don't see a significant regression in compilation time based on these numbers.
```
clamscan 9.6843 -0.7650%
ldecod 4.5576 0.6604%
lencod 9.8791 1.5637%
SPASS 9.5813 2.0035%
make_dparser 2.0540 1.2723%
kc 10.0371 -0.1296%
lua 2.6484 2.4356%
oggenc 3.1777 1.0269%
siod 1.4711 1.3638%
sqlite3 12.4904 0.4944%
treecc 1.3868 -1.7237%
7zip-benchmark 27.6821 -0.2909%
smg2000 4.7685 2.4668%
bullet 20.0863 0.6708%
espresso 3.6563 1.7755%
gs 2.9518 -0.8027%
consumer-jpeg 3.4321 -0.4087%
consumer-lame 2.8166 2.2606%
consumer-typeset 7.7711 0.4109%
office-ispell 1.1021 0.0816%
paq8p 1.3805 0.0290%
timberwolfmc 6.3654 -1.0692%
mybison 1.1789 1.8489%
ControlFlow-dbl 1.0986 0.1001%
ControlFlow-flt 1.0994 -1.3461%
ControlLoops-dbl 1.0301 -1.3881%
ControlLoops-flt 1.0259 -0.1170%
CrossingThresholds-dbl 1.0041 -0.6896%
Equivalencing-flt 1.0074 -2.5841%
Expansion-dbl 1.0220 -0.4315%
Expansion-flt 1.0275 0.0097%
GlobalDataFlow-dbl 1.0226 -0.2546%
GlobalDataFlow-flt 1.0147 -0.2072%
InductionVariable-dbl 1.0076 -0.3878%
LinearDependence-dbl 1.0397 -0.0866%
LinearDependence-flt 1.0477 -1.8301%
LoopRestructuring-flt 1.0237 -0.9422%
Reductions-dbl 1.0482 0.1239%
Reductions-flt 1.0262 2.6349%
pairlocalalign 6.6571 -1.3261%
cjpeg 3.3882 2.9228%
tramp3d-v4 15.5355 0.0097%
loop_unroll 2.6732 0.4218%
simple_types_constant_folding 2.1412 0.5031%
simple_types_loop_invariant 1.7099 -0.3104%
spirit 3.0955 0.3676%
```
http://reviews.llvm.org/D5633
More information about the llvm-commits
mailing list