diff options
author | Akim Demaille <akim.demaille@gmail.com> | 2020-05-10 11:46:40 +0200 |
---|---|---|
committer | Akim Demaille <akim.demaille@gmail.com> | 2020-05-10 11:51:46 +0200 |
commit | 9c21da3363054cccf7618d0b3a2ec91f68e67662 (patch) | |
tree | e2a680e41dcae8449f5def6990b4423dec2b8079 /README-alpha | |
parent | febf6115a0d7799b8d4b80e1bb5e89236182d7ca (diff) | |
download | bison-9c21da3363054cccf7618d0b3a2ec91f68e67662.tar.gz |
bench: add support to randomize the order of execution
It's amazing how much the order matters. To a point that many of
these benches are meaningless. For instance (some of the benches
where run with `make -C benches/latest rand
BENCHFLAGS=--benchmark_min_time=3`):
compiler: g++ -std=c++11 -O2
0. %define nofinal
1.
--------------------------------------------------
Benchmark Time CPU Iterations
--------------------------------------------------
BM_y0 1543 ns 1541 ns 441660
BM_y1 1521 ns 1520 ns 456535
--------------------------------------------------
BM_y0 1531 ns 1530 ns 440584
BM_y1 1512 ns 1511 ns 457591
--------------------------------------------------
BM_y0 1539 ns 1538 ns 2749330
BM_y1 1516 ns 1515 ns 2771500
--------------------------------------------------
BM_y0 1571 ns 1570 ns 2600782
BM_y1 1542 ns 1541 ns 2708349
--------------------------------------------------
BM_y0 1530 ns 1529 ns 2670363
BM_y1 1519 ns 1518 ns 2764096
--------------------------------------------------
Benchmark Time CPU Iterations
--------------------------------------------------
BM_y1 1529 ns 1528 ns 451937
BM_y0 1508 ns 1507 ns 453944
--------------------------------------------------
BM_y1 1525 ns 1524 ns 2750684
BM_y0 1516 ns 1515 ns 2794034
--------------------------------------------------
BM_y1 1526 ns 1525 ns 2749620
BM_y0 1515 ns 1514 ns 2808112
--------------------------------------------------
BM_y1 1524 ns 1523 ns 4475844
BM_y0 1502 ns 1501 ns 4611665
* etc/bench.pl.in: here.
Diffstat (limited to 'README-alpha')
0 files changed, 0 insertions, 0 deletions