Opened 9 years ago

Closed 9 years ago

#6398 closed enhancement (fixed)

webkitgtk-2.8.1

Reported by: Fernando de Oliveira Owned by: Fernando de Oliveira
Priority: normal Milestone: 7.8
Component: BOOK Version: SVN
Severity: normal Keywords:
Cc:

Description

​http://webkitgtk.org/releases/webkitgtk-2.8.1.tar.xz

sha256sum: 7d214459d216ca499ad75bccae7f248655d6ef5b295ffafd299773e7331860ff 

This is the first bug fix release in the stable 2.8 series.

What's new in the WebKitGTK+ 2.8.1 release?
===========================================

  - Handle keep-alive connections in GStreamer HTTP source element.
  - Fix a crash in DOMObjectCache when a wrapped object owned by the cache is
    unreffed by the user.
  - Fix rendering of drag and drop icon.
  - Fix the build with REDIRECTED_XCOMPOSITE_WINDOW disabled in X11 platform.
  - Fix the build with Wayland target enabled.
  - Fix the build for HPPA.

Change History (3)

comment:1 by Fernando de Oliveira, 9 years ago

Owner: changed from blfs-book@… to Fernando de Oliveira
Status: new → assigned

comment:2 by Fernando de Oliveira, 9 years ago

Estimated 79 SBU using 1 core, with the equation from ticket #6362 

speedup = 1 / ( ( 1 - f ) + ( f / n ) )

by calculation in libreoffice calc.

This equation seems to be a linear approximation, because inside a parallel process there can be other parallel processes and again, so the process seems to be, in general, non-linear.

Used

  1. version 2.8.0 value of 19 SBU (using parallelism=12) by Bruce.
  1. version 2.8.1 value of 30 SBU with -j4, by me.

The important value to be determined is f.

Given a guessed value for f, the estimated values for 1 core in cases 1. and 2. are different, unless the guess is perfect.

Use "reach target" in lcalc so that this difference is zero, varying the value of f.

Found the value 

f = 0.8293498245

with relative error less than

4.2 E -012 %

This is interesting, meaning that webkitgtk has about 17% of the code that must be run only in one core, not in parallel.

For comparison:

libreoffice: f = 0.96 (obtained by trial and error, so, not so accurate)

ffmpeg: f = 0.92 (obtained building twice, with -j4 and with -j1, to determine f)

Thus, LO is the champion in parallelism, until now.

If someone else is interested, we can discuss after some time, build a database of values for as many as possible packages, if the equation provides consistent values.

I haven't studied deeply the paper, but this is my interpretation, for the moment.

Fixed at 15835.

comment:3 by Fernando de Oliveira, 9 years ago

Resolution: → fixed
Status: assigned → closed
Note: See TracTickets for help on using tickets.