mirror of
https://gitlab.gnome.org/GNOME/glib.git
synced 2024-12-24 22:46:15 +01:00
4d5047e0e7
The main use of the performance test is to run it for two (or more) commits
and compare the results. Doing that manually, is cumbersome.
Add a (very hacky) script to help with that. For usage, see the comment
on top of the script.
Example:
# first:
meson build -Dprefix=/tmp/glib/ -Db_lto=true --buildtype release -Ddebug=true
# then:
GLIB_PERFORMANCE_FACTOR=17.06 \
PERF='perf stat -r 4 -B' \
PATCH="2.80.0..th/performance" \
COMMITS="2.79.3 2.80.0" \
/tmp/performance-run.sh -s 1 property-get property-set
This will build the requested $COMMITS and print something like:
...
>>> combined result > /tmp/glib-performance-output.all
Running test property-get
property-get: Property get per second: 35742719 37208288 (+4.1%)
Running test property-set
property-set: Property set per second: 32341232 36942399 (+14.2%)
Running test property-get
property-get: Property get per second: 36934401 37143479 (+0.566%)
Running test property-set
property-set: Property set per second: 38046387 38165548 (+0.313%)
Running test property-get
property-get: Property get per second: 34759576 36359761 (+4.6%)
Running test property-set
property-set: Property set per second: 35262505 37651733 (+6.78%)
Running test property-get
property-get: Property get per second: 37014537 32870906 (-11.2%)
Running test property-set
property-set: Property set per second: 36633026 38216846 (+4.32%)
Performance counter stats for './build/gobject/tests/performance/performance -s 1 property-get property-set' (4 runs):
1,312.18 msec task-clock:u # 1.000 CPUs utilized ( +- 4.82% )
0 context-switches:u # 0.000 /sec
0 cpu-migrations:u # 0.000 /sec
121 page-faults:u # 92.213 /sec ( +- 0.24% )
5,221,701,009 cycles:u # 3.979 GHz ( +- 2.61% )
19,035,814,175 instructions:u # 3.65 insn per cycle ( +- 0.00% )
4,335,306,010 branches:u # 3.304 G/sec ( +- 0.00% )
13,031 branch-misses:u # 0.00% of all branches ( +- 4.17% )
TopdownL1 # 10.3 % tma_backend_bound
# 5.3 % tma_bad_speculation
# 11.4 % tma_frontend_bound
# 73.1 % tma_retiring ( +- 2.15% )
[1] 1.3127 +- 0.0634 seconds time elapsed ( +- 4.83% )
[2] 1.2631 +- 0.0253 seconds time elapsed ( +- 2.00% )
property-get: Property get per second: 35742719 , 36934401 , 34759576 , 37014537 ; 37208288 , 37143479 , 36359761 , 32870906 ;
property-set: Property set per second: 32341232 , 38046387 , 35262505 , 36633026 ; 36942399 , 38165548 , 37651733 , 38216846 ;
|
||
|---|---|---|
| .. | ||
| performance | ||
| accumulator.c | ||
| autoptr.c | ||
| basic-signals.c | ||
| basics-gobject.c | ||
| binding.c | ||
| bindinggroup.c | ||
| boxed.c | ||
| closure-refcount.c | ||
| closure.c | ||
| custom-dispatch.c | ||
| cxx.cpp | ||
| defaultiface.c | ||
| deftype.c | ||
| deprecated-properties.c | ||
| dynamictests.c | ||
| dynamictype.c | ||
| enums.c | ||
| flags.c | ||
| genmarshal.py | ||
| gobject-query.py | ||
| ifaceproperties.c | ||
| marshalers.list | ||
| max-version.c | ||
| meson.build | ||
| mkenums.py | ||
| notify-init2.c | ||
| notify-init.c | ||
| object.c | ||
| objects-refcount1.c | ||
| objects-refcount2.c | ||
| override.c | ||
| param.c | ||
| private.c | ||
| properties-introspection.c | ||
| properties-refcount1.c | ||
| properties-refcount2.c | ||
| properties-refcount3.c | ||
| properties-refcount4.c | ||
| properties.c | ||
| qdata.c | ||
| reference.c | ||
| references.c | ||
| signal-handler.c | ||
| signalgroup.c | ||
| signals-refcount.c | ||
| signals.c | ||
| singleton.c | ||
| taptestrunner.py | ||
| testcommon.h | ||
| testing.c | ||
| testmodule.c | ||
| testmodule.h | ||
| threadtests.c | ||
| type-flags.c | ||
| type.c | ||
| value.c | ||