OS/2, eCS & ArcaOS - Technical > Internet

New SeaMonkey builds

(1/6) > >>

Dave Yeo:
So I've built 3 SeaMonkey packages based on the latest (and probably last) code from Bitwise. Each is built with different optimizations, curious what works best for people.
All target i686 for now. i686 is the minimum requirement as Mozilla uses some atomic instructions that are not available in earlier CPU's.
This one is built with straight -O3 optimization, should be the fastest, also the most memory use <url>https://bitbucket.org/dryeo/dry-comm-esr31/downloads/seamonkey-2.42.9esr_r3.en-US.os2.zip</url>
This one is built the same as the latest Firefox, with the JavaScript engine using -O3 and the rest using -Os for minimal size. This is also what other platforms such as Linux use, <url>https://bitbucket.org/dryeo/dry-comm-esr31/downloads/seamonkey-2.42.9esr_r3_s.en-US.os2.zip</url> xul.dll is almost 10 MB smaller.
Rich pointed out that using -O3 is likely to expose GCC bugs, so I also built one using straight -O2, which is what we traditionally used until late in the 10ESR cycle. May be the most stable but worst performing. Firefox built with this often had a timing issue and failed to play Youtube videos, getting stuck at the loading phase. <url>https://bitbucket.org/dryeo/dry-comm-esr31/downloads/seamonkey-2.42.9esr_r3_O2.en-US.os2.zip</url>

There's also a build of Thunderbird with straight -O3 optimization that I'm using, <url>https://bitbucket.org/dryeo/dry-comm-esr31/downloads/thunderbird-45.8.0_r3.en-US.os2.zip</url> If requested, I can build with different optimizations.

I'll probably delete the exp builds so if anyone wants to save them... and will build for other CPU's.


OS4User:

--- Quote from: Dave Yeo on June 09, 2018, 07:59:33 am ---and will build for other CPU's.

--- End quote ---

is it possible to build with AVX optimization ?

Rich Walsh:

--- Quote from: Dave Yeo on June 09, 2018, 07:59:33 am ---I've built 3 SeaMonkey packages based on the latest (and probably last) code from Bitwise. Each is built with different optimizations, curious what works best for people.

--- End quote ---

First off, let me thank you for your responsiveness. :)

I d/l'd all 3 as soon as I saw this post and put icons for each on my Desktop so I could test them side-by-side. My first stop for each was YouTube - a site I seldom use because I can only take so much frustration. Let me tell you, regardless of which build I tried, this was the best YouTube experience I've ever had. Certainly there were some dropped frames with both HTML5/H264 and WEBM/VPX, but only once did the browser become momentarily unresponsive (I think it was with -Os and VPX). Using the -O3 version, I was surprised to find that VPX put less of a strain on my quad-core than h264 - but both performed well.

Of course, there's only so much that mere optimization tweaks can fix. I'm posting this using the -O3 build, and after watching several Sam Smith videos, one core keeps spiking to 100% every few seconds. I'll keep testing but my initial impression is that -Os is a noticeably weaker performer than the others. (Update: that -O3 build failed to terminate when I closed it and left the WPS unresponsive - had to kill it.)

BTW... which compiler did you use?

Dave Yeo:

--- Quote from: Rich Walsh on June 09, 2018, 06:05:04 pm ---
Of course, there's only so much that mere optimization tweaks can fix. I'm posting this using the -O3 build, and after watching several Sam Smith videos, one core keeps spiking to 100% every few seconds. I'll keep testing but my initial impression is that -Os is a noticeably weaker performer than the others. (Update: that -O3 build failed to terminate when I closed it and left the WPS unresponsive - had to kill it.)

--- End quote ---

Yes, this -O3 build seems to have problems terminating, a new problem with SM but one that Firefox had until switched to the -Os.

--- Quote ---BTW... which compiler did you use?

--- End quote ---

I'm currently using Paul's build of 5.10. Not sure what, but something seems to have broken in my environment with 4.92, the browser built with it seemed fine until going to YouTube, where VP9 videos would just display static with a message that something went wrong.
Newer compilers fail due to changes in how they parse the C++ code, which is a shame as Paul says he fixed the alignment issues at some point.

xynixme:

--- Quote ---curious what works best for people.
--- End quote ---

Testing... Will take a while.

First results: seamonkey-2.42.9esr_r3.en-US.os2.zip (65945436 bytes) works with a (single core) Pentium 4 2.8 GHz. No adjusted settings, no disabled extras. The CPU load still is 100%, BUT ... it's no longer a honest 100%. Almost as if it's executing NOPs until the CPU load is 100%. So it's usable. Repsonsive. I'll try this with a Pentium III and a T42p too, and all other SM releases based on the same FF release.

Navigation

[0] Message Index

[#] Next page

Go to full version