Use --num-thread=10 when dexing Android code (saves 5 seconds)

Use --num-thread=10 when dexing Android code (saves 5 seconds)

Timing for creating chrome_apk/classes.dex:

No flag:
real    0m10.767s
user    0m16.480s
sys     0m0.444s

--num-threads=4
real    0m7.450s
user    0m11.837s
sys     0m0.606s

--num-threads=10 (or 20, or 80)
real    0m5.399s
user    0m9.266s
sys     0m0.420s

Timings performed on a z620 workstation.

BUG=

Committed: https://crrev.com/0f9ad3a20275b77d228f67c2edd89dbea0e3e9ea
Cr-Commit-Position: refs/heads/master@{#327358}

[agrieve, 2015-04-28 18:26:50]

agrieve@chromium.org changed reviewers:
+ cjhopman@chromium.org

[cjhopman, 2015-04-28 18:34:29]

lgtm

[agrieve, 2015-04-28 19:28:43]

The CQ bit was checked by agrieve@chromium.org

[commit-bot: I haz the power, 2015-04-28 19:29:01]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1109193002/1

[commit-bot: I haz the power, 2015-04-28 20:34:26]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-04-28 20:35:19]

Patchset 1 (id:??) landed as
https://crrev.com/0f9ad3a20275b77d228f67c2edd89dbea0e3e9ea
Cr-Commit-Position: refs/heads/master@{#327358}

[Lei Zhang, 2015-05-05 02:03:49]

thestig@chromium.org changed reviewers:
+ thestig@chromium.org

[Lei Zhang, 2015-05-05 02:03:50]

(drive by) Why not just use multiprocessing.cpu_count() ? Using 10 threads may
not make sense on a quad core machine, or on a machine in the future with a
zillion cores.

[agrieve1, 2015-05-05 02:11:52]

On 2015/05/05 02:03:50, Lei Zhang wrote:
> (drive by) Why not just use multiprocessing.cpu_count() ? Using 10 threads may
> not make sense on a quad core machine, or on a machine in the future with a
> zillion cores.

Thought about using # of cores, but my machine has 40 cores, and going past 10
didn't speed things up at all. On the flip side, if you have fewer than 10
cores, it might make it a bit slower, but if it's IO bound, then the speedup is
still realized. Would be fine if someone with a quadcore machine wanted to
further tune, but didn't think it worth looking at myself (really, dx itself
should be choosing # of threads)

[gsennton, 2015-05-06 13:08:07]

A revert of this CL (patchset #1 id:1) has been created in
https://codereview.chromium.org/1129793002/ by gsennton@chromium.org.

The reason for reverting is: This patch makes the creation of dex files
non-deterministic -- dex files from different runs of the same build might not
be exactly similar.

This is a problem because for 64-bit webview APKs we need to merge a 32-bit APK
into the 64-bit APK (to make 32-bit apps work). When we do this we check that
the files in the different APks are similar and this check doesn't work if the
dex files in the two APKs are not exactly the same.
https://code.google.com/p/chromium/issues/detail?id=483665.