forked from OSchip/llvm-project
21 lines
803 B
Plaintext
21 lines
803 B
Plaintext
|
These are some quick and dirty tools for measuring the performance impact
|
||
|
|
of a change to llgo by sampling the results of running the libgo benchmark
|
||
|
|
suite. They can be used to calculate the geo-mean and 95% confidence interval
|
||
|
|
using the Student's t-test. The benchcomp program massages the output of the
|
||
|
|
Go benchmark tools into a form that can be read by the R program analyze.R
|
||
|
|
which runs the statistics.
|
||
|
|
|
||
|
|
To use, clpatch this into gofrontend:
|
||
|
|
https://codereview.appspot.com/103550047/
|
||
|
|
|
||
|
|
then run:
|
||
|
|
|
||
|
|
make
|
||
|
|
make -C workdir/gofrontend_build/libgo-stage1 bench 2>&1 | tee before.out
|
||
|
|
# make changes
|
||
|
|
make
|
||
|
|
make -C workdir/gofrontend_build/libgo-stage1 bench 2>&1 | tee after.out
|
||
|
|
utils/benchcomp/benchcomp benchns before.out after.out | R -f utils/benchcomp/analyze.R
|
||
|
|
|
||
|
|
The results should be displayed on stdout.
|