vogar should use -Xdexopt:none for non-performance work
GoogleCodeExporter opened this issue · comments
Google Code Exporter commented
vogar should use -Xdexopt:none for non-performance work, for quicker
turn-around times when we aren't going to benefit from dexopt. i suggested that
"are we running a caliper benchmark?" is a good heuristic, but ben pointed out
that we probably want the CB to dexopt even though it's slow, just because the
real system will run dexopted code.
Original issue reported on code.google.com by e...@google.com
on 22 Jun 2010 at 10:12
Google Code Exporter commented
Fantastic idea.
Original comment by jessewil...@google.com
on 23 Jun 2010 at 3:42
Google Code Exporter commented
Maybe use the benchmark heuristic, but then have a --force-dexopt argument for
the CB to use?
Original comment by jsha...@google.com
on 3 Aug 2010 at 7:27
Google Code Exporter commented
personally, i think we _don't_ want it on the CB. i blame it (without proof)
for the ridiculous variation in the number of tests we run. maybe your
early-fail combined with a huge monitor-timeout helps?
(ben's point isn't really relevant as long as the CB's host/x86 rather than
device/arm.)
Original comment by e...@google.com
on 3 Aug 2010 at 7:36
Google Code Exporter commented
this is done now, no?
Original comment by e...@google.com
on 12 Sep 2010 at 10:24
Google Code Exporter commented
Done now.
Original comment by jessewil...@google.com
on 13 Sep 2010 at 1:36
- Changed state: Done