On 01.06.2012 19:24, Philip Martin wrote: > Philip Martin <philip.mar...@wandisco.com> writes: > >> Is COPY_TWO_BYTES a significant optimisation? > r1325896 implies this is most significant for ra_svn. So I compared the > testsuite over ra_svn built with the "fast" code and the "slow" code. > The CPU used by the testsuite and svnserve was: > > "fast" code > > 749.93user 404.98system 9:09.25elapsed testsuite > 0:32.43 svnserve > > 742.84user 400.00system 8:59.75elapsed testsuite > 0:32.70 svnserve > > 739.51user 400.87system 8:58.91elapsed testsuite > 0:32.66 svnserve > > "slow" code > > 745.06user 402.36system 9:04.21elapsed testsuite > 0:32.73 svnserve > > 739.94user 401.86system 8:57.13elapsed testsuite > 0:32.58 svnserve > > On my Linux box this optimisation isn't significant. Perhaps it makes a > difference on some other platform or with some other test scenario, but > at the moment it's not clear why we would want with this code.
I can't see how any compiler these days, except possibly one that targets really tiny embedded platforms, would do worse with memcpy than with the "optimized" code. I suspect someone has been a bit too eager with optimizations without actually measuring their impact. :) -- Brane
