Add use of Psyco to GYP on Windows. On my z600 with 12 GB of RAM,

Add use of the Psyco JIT compiler to GYP on Windows.  On my z600 with 12 GB of RAM, this shortens the time taken for a warm run of build/chromium_gyp from approximately 90 seconds down to approximately 70 seconds.  On the other hand, it increases maximum memory usage for the GYP process from ~132 MB to ~158 MB on the same test system.

At the moment it is unknown whether using Psyco on Mac and Linux would pay off; follow-up changes may address this if it is.

BUG=none
TEST=things build correctly

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=79871

[bradn, 2011-03-30 16:31:36]

LGTM with a few comments:

1. Since the library in windows specific (and I assume the pyd is of non-trivial
size). Could we land the third_party portion in trunk/deps/third_party and then
add a deps_os entry to pull it in only for windows?

2. Awesome you're looking at speeding it up. jam and I recently profiled it
(historically it creeps up in runtime and can be beating back). We came up with
a list of potential speedups. Interesting one of them was to look into why where
spending ~30sec in grit at gyp gen time. As I recall you're the orignal grit
author. Haven't had time to follow up on it, but my guess was the the --inputs /
--outputs option pays the full generation cost?

http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium
File build/gyp_chromium (right):

http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium#newcode32
build/gyp_chromium:32: if os.name == 'nt':
sys.platform == 'win32' be more idiomatic with other chromium code.

[Jói, 2011-03-30 16:50:21]

Thanks Brad.

#1 - The entire directory is only about 20 KB, so pretty trivial in size, but I
could still move it if you think it makes sense, let me know.

#2 - You're right I'm the author of grit.  I can probably take a look at why so
much time is spent in grit, I'll put it on my list.  You pay parsing cost for
--inputs and --outputs, but not the full generation cost.  Hopefully some of
this can be cached (more likely on the GYP side); one issue is that Chrome has
an awful lot of .grd files which might mean caching will help less.

http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium
File build/gyp_chromium (right):

http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium#newcode32
build/gyp_chromium:32: if os.name == 'nt':
On 2011/03/30 16:31:36, bradn wrote:
> sys.platform == 'win32' be more idiomatic with other chromium code.

Done.

[bradn, 2011-03-30 17:35:47]

On Wed, Mar 30, 2011 at 9:50 AM, <joi@chromium.org> wrote:

> Thanks Brad.
>
> #1 - The entire directory is only about 20 KB, so pretty trivial in size,
> but I
> could still move it if you think it makes sense, let me know.
>
> I'm somewhat torn. But I polled chase and we concluded that
there's precedent to deps in even small stuff like this.
Makes me a little sad (one more checkout to sync).
Go ahead and deps it in.





> #2 - You're right I'm the author of grit.  I can probably take a look at
> why so
> much time is spent in grit, I'll put it on my list.  You pay parsing cost
> for
> --inputs and --outputs, but not the full generation cost.  Hopefully some
> of
> this can be cached (more likely on the GYP side); one issue is that Chrome
> has
> an awful lot of .grd files which might mean caching will help less.
>
>
>
>
> http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium
> File build/gyp_chromium (right):
>
>
> http://codereview.chromium.org/6778017/diff/2002/build/gyp_chromium#newcode32
> build/gyp_chromium:32: if os.name == 'nt':
> On 2011/03/30 16:31:36, bradn wrote:
>
>> sys.platform == 'win32' be more idiomatic with other chromium code.
>>
>
> Done.
>
>
> http://codereview.chromium.org/6778017/
>

[Jói, 2011-03-30 17:56:22]

> > #1 - The entire directory is only about 20 KB, so pretty trivial in size,

Ehm, that was supposed to be 230 KB, looks like I dropped a keystroke :)

I agree, depping it in seems like the right thing to do.  Will follow up soon
with a separate CL to add it to trunk/deps/third_party and then an update to
this CL.

Cheers,
Jói

[Jói, 2011-03-30 18:18:13]

PTAL

[bradn, 2011-03-30 18:20:25]

LGTM

[jam, 2011-03-31 16:36:41]

THIS IS AWESOME, THANK YOU!

i tried doing this a few weeks ago but wasn't able to get the package to
work :)

On Wed, Mar 30, 2011 at 11:20 AM, <bradnelson@google.com> wrote:

> LGTM
>
>
> http://codereview.chromium.org/6778017/
>

[M-A Ruel, 2011-04-04 15:55:44]

When adding new dependencies, please update svn:ignore accordingly.

I'll fix.

[Jói, 2011-04-04 16:38:36]

Thanks Marc-Antoine, will make sure to do that next time.

Cheers,
Jói


On Mon, Apr 4, 2011 at 11:55 AM,  <maruel@chromium.org> wrote:
> When adding new dependencies, please update svn:ignore accordingly.
>
> I'll fix.
>
> http://codereview.chromium.org/6778017/
>