879309 - B2G: Sunspider regression, ss-cordic 22x slower [2013-05-21 … 2013-05-29]

Status: RESOLVED DUPLICATE of bug 875720

(Core :: JavaScript Engine, defect)

Description

[Nicolas B. Pierron [:nbp]]

AWFY was not running in this range of revisions.  We need to investigate what was wrong there and if there is a way to fix this regression.

First bisecting to get a first range in terms of nightly, and then later try to do that in terms of unstable versions of inbound if this is not enough.

Comment 1

[Nicolas B. Pierron [:nbp]]

So far, I've been able to reduce it by using nightly builds.

http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=00b264c7cced&tochange=fe53a0780c6f

I will see if I can get a better regression range in the next days.  Also, this regression seems to be visible in terms of UI responsiveness.  I noticed extremely long pauses recently, some might be related.

Also, running cordic alone in b2g's browser does not seems to highlight the same regressions (like 10x faster).

Comment 2

[bhavana bajaj [:bajaj]]

(In reply to Nicolas B. Pierron [:nbp] from comment #1)
> So far, I've been able to reduce it by using nightly builds.
> 
> http://hg.mozilla.org/integration/mozilla-inbound/
> pushloghtml?fromchange=00b264c7cced&tochange=fe53a0780c6f
> 
> I will see if I can get a better regression range in the next days.  Also,
> this regression seems to be visible in terms of UI responsiveness.  I
> noticed extremely long pauses recently, some might be related.

Hey nbp, any updates here ? Is this bug still a concern ? 
> 
> Also, running cordic alone in b2g's browser does not seems to highlight the
> same regressions (like 10x faster).

Comment 3

[Nicolas B. Pierron [:nbp]]

(In reply to bhavana bajaj [:bajaj] from comment #2)
> Hey nbp, any updates here ? Is this bug still a concern ? 
> > 
> > Also, running cordic alone in b2g's browser does not seems to highlight the
> > same regressions (like 10x faster).

Yes, it is still a big concern, I am instrumenting B2G to add a bisect feature, because git-repo does not have any due to it way of handling repositories.

Bug 868598 added a reference to a mercurial changeset of gaia (don't ask), so I am trying to locate the map-file to at least avoid making mistakes by basing the rest of the bisect on the timestamp of the commits (which is wrong, but mostly correct)

As we do not keep images for inbound, I was unable to download Unagi's images to test a smaller regression range, so I will have to build the intermediate version my-self.  I will use the idle time of hydra to build the extra images for me …

Comment 4

[Nicolas B. Pierron [:nbp]]

(In reply to Nicolas B. Pierron [:nbp] from comment #3)
> […]  I will use the idle time of hydra to build
> the extra images for me …

Hydra is the server that I use for converting mozilla-central to git, and to run are-we-fast-yet on B2G.

Comment 5

[Nicolas B. Pierron [:nbp]]

I reduced it to [1], so this is a duplicate of Bug 875720.

[1] http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=7c89908125e71b8b65f5ed1f9bfdd37d8476a999&tochange=4370f503d69f321878e11c26d07bf17c4413cc58