On Mon, Jul 25 2022, Taylor Blau wrote: > On Wed, Jul 20, 2022 at 06:38:23PM +0000, Abhradeep Chakraborty via GitGitGadget wrote: >> From: Abhradeep Chakraborty <chakrabortyabhradeep79@xxxxxxxxx> >> >> Enable `pack.writeReverseIndex` before running pack-bitmap related >> performance tests. >> >> The performance difference with `pack.writeReverseIndex` enabled and >> with disabled are given below - > > Thanks; this order of changes in the t/perf suite makes sense to me. One > note, this sort of change where we're comparing all of the tests in a > single t/perf file against themselves before and after some change, it > is helpful to do (in t/perf) > > ./run HEAD . p5310-pack-bitmaps.sh > > which compares HEAD to what's in the current tree. You'll get the > results side-by-side, which makes them a little easier to scan. You can > also aggregate results together from multiple runs with the > t/perf/aggregate.perl script. > > One gotcha (that has often bitten me in the past) is that when running > the perf suite with `.` as your build target, it uses whatever git > binary is sitting in your tree. So make sure that it is both (a) > up-to-date, ie., that it is the result of compiling what's currently in > your tree, and (b) that it is compiled with the same settings as what > you built HEAD with. > > I have often scratched my head at why the result of running some perf > suite on '.' seems much slower than it should be, only to realize that > the "git" binary sitting in my tree was built with -O0 or something. Rather than comparing HEAD to your current tree it's generally better to do something like: GIT_PERF_MAKE_OPTS='-j3' ./run HEAD~ HEAD [...]
