Which Build Works Best on Your PII?

[realgeorge]

I have tried a lot of builds on my Win98 300MHz PII and I always go back to jayfromtaiwan`s "Fx-040426-Trunk-Oxp-GALFs6-crc32-DOMi" build which is the fastest on my PII, it has never crashed, and it works very well for all the things I normally do. I would just like to know what other PII owners are using, and the performance they are getting so that I can try them, if I haven`t already, in hope of finding an even better build for my PII, if possible.

[sasquatch]

I have tried a lot of builds on my Win98 300MHz PII and I always go back to jayfromtaiwan`s "Fx-040426-Trunk-Oxp-GALFs6-crc32-DOMi" build which is the fastest on my PII, it has never crashed, and it works very well for all the things I normally do. I would just like to know what other PII owners are using, and the performance they are getting so that I can try them, if I haven`t already, in hope of finding an even better build for my PII, if possible.

Thanks for that "review", and I'll treat it as a "tip".

Now, for a newer version of the same, without DOMi.

[realgeorge]

Now, for a newer version of the same, without DOMi.

Yes, and also without SVG.

[sasquatch]

Now, for a newer version of the same, without DOMi.

Yes, and also without SVG.

Yes, that would be great, since there are no working sites which use it, only test sites, and in a VERY limity capacity.

[jayfromtaiwan]

after today's build

[mmoy]

A few people at work are using my 1A build which uses MMX JPEG rendering (I hacked some code to get that to work) and they like it. There's a 23% improvement in JPEG rendering time compared to the official build. The build is fairly old though from sometime
in April.

[jayfromtaiwan]

A few people at work are using my 1A build which uses MMX JPEG rendering (I hacked some code to get that to work) and they like it. There's a 23% improvement in JPEG rendering time compared to the official build. The build is fairly old though from sometime
in April.

yeah seriously your optimizations works like a charm the devs should approve your patch ASAP

btw i was searching for your SSE and MMX patch and I couldn't find them .... did you file them yet?

[mmoy]

The SSE patch is in the Pentium3G kit in the source directory. I haven't made the MMX code available as it was a hack and I've been trying to think of a way to do it so that it could coexist with the SSE and SSE2 code so that I wouldn't have to keep an eye on multiple versions of the code. I think that it could be done with maybe 5 lines of code. One other thing is that the MMX code is not as accurate as the SSE and SSE2 code though I haven't heard of any rendering complaints. I could port my SSE code to MMX which would get rid of the accuracy concern.

The JPEG code comes in a variety of flavors: SLOW Integer, FAST Integer and floating point. My tests with Floating Point weren't good. The MMX code is written using the FAST Integer approach. The SSE2 and SSE code use the SLOW Integer approach. I think that it could be a challenge to get the Mozilla devs to adopt the IFAST code which is why I haven't done anything with it. Still, those willing to use the current MMX IFAST code should have the choice as to whther or not they want to use it.

If I had a PII, I think that I'd want all the speed that I could get.

I'll see if I can get to that in the next week.

[jayfromtaiwan]

can't wait

even though I have said this for like a dozen times, but you should really be one of the devs

[mmoy]

The first step is getting a piece of code in. I'm working on that one. I just need a code review and I've been waiting for two weeks with not a peep from the module owner. At some point, I'll send an email to drivers@mozilla.org asking if they want a 15% performance increase in JPEG on P4s. And hopefully a much bigger improvement in GIF handling. And maybe Linux ports.

[jayfromtaiwan]

At some point, I'll send an email to drivers@mozilla.org asking if they want a 15% performance increase in JPEG on P4s.

lol if they don't want it (which is unlikely) we users + unofficial builders do

ehhhh it only works on Windows? how come? (lol i'm a total newb to C++ programming)

[mmoy]

ehhhh it only works on Windows? how come? (lol i'm a total newb to C++ programming)

Nobody with a Linux box has ported the code to GCC. I'm pretty sure that a port would be just grabbing the Inline Assembler and putting it in whatever format GCC requires. I had a look at it but don't have a Linux build machine though I have a Linux machine available.

[jayfromtaiwan]

ahhhh GCC , nearly forgot about the tiny + powerful stuff

yeah hope someone who knows programming can port it to Linux cuz the patch really make JPEG rendering a lot faster

[mmoy]

ahhhh GCC , nearly forgot about the tiny + powerful stuff

yeah hope someone who knows programming can port it to Linux cuz the patch really make JPEG rendering a lot faster

I think that someone could grab the GCC manual and get the format, then do the appropriate cut and paste on my code and give it to a Linux builder to try out. I'm going to have to port it to Intrinsics anyways as Windows 64 for MSVC++ doesn't support inline assembler.

[realgeorge]

A few people at work are using my 1A build which uses MMX JPEG rendering (I hacked some code to get that to work) and they like it. There's a 23% improvement in JPEG rendering time compared to the official build. The build is fairly old though from sometime
in April.

I did try the mmoy "FireFox-Experimental-2004-04-13-O2-GL-G6-1A" build and it is exceptionally fast and a great build for PII`s. My tests show it to be just as fast as jayfromtaiwan`s "Fx-040426-Trunk-Oxp-GALFs6-crc32-DOMi" on my PII, and I alternate using it . On 5-6-04 I had made a post about this build stating, "Wow, this build (FireFox-Experimental-2004-04-13-O2-GL-G6-1A) is really Hot, beating the fastest Firefox browser that I had ever tested on my PII, and it`s smooth in operation.----- I sure hope you`ll continue to develope this experimental configuration, and not drop it from your builds list." mmoy replied, "The 1A build uses existing MMX code for IDCT though it's a less accurate implementation than the C code stuff. Though I didn't notice any rendering differences when I tried it out. Unfortunately, the code isn't documented." I certainly agree with mmoy`s last statement that "If I had a PII, I think that I'd want all the speed that I could get. I'll see if I can get to that in the next week." Thank you mmoy!