I have just uploaded two new compiles of oggenc 1.0 to Mirror 1.

Both of these are optimised for P4, but one is compiled with ICL 6.0 and the other with ICL 7.0. It would be interesting to know which is the faster!!

I regret that I am unable to test these myself as I run an AthlonXP and not a P4 system.

cool...just had a play around they seem to be around 33% quicker than the standard compiles, theres not really much difference between the icl6 and icl7 compiles, the 7 is marginally quicker

-J

Thanks for the feedback. 33% is quite an increase!

Where do you find the 'standard' (non-P4 optimized) compiles?

Same place you got the P4 versions, John!!

hello,

in a quick test ( 2 samples) the output of the P4-icl6 compile is about 3-4 kbps smaller than the all the others's

Oggenc 1.0, oggenc 1.0 icl6, oggenc 1.0 gcc 3.2, oggenc 1.0 icl6 (31 Oct 02) are almost identical.

P4-icl7 is clearly faster than P4-icl6 (5%).

Both P4 optimized versions are 40%-50 faster than the official oggenc 1.0 (unziped from oggtools)

System : P4 1.6Mhz, 512Mb sdram, win2000.

Thanks for the feedback. It seems certain, then, that ICL 7 produces faster code for the P4. Just for the record, the limited testing I've done on my own Athlon system suggests the opposite!

You mentioned that the P4-ICL 6 compile generated slightly smaller encodes, how about the ICL 7 version? Both compiles were produced using identical optimisation settings.

I would expect any of the ICL compiles to be faster than the vorbis-tools download as that was compiled using MSVC 6 with very basic optimisations.

hello,

John 33, the P4-icl7 compile output is almost identical to all the non P4 optimized compiles. There is a very slight difference from the official, smaller than the one that exists between the official and the gcc 3.2 compile.

Take a look

Sample 1 (King Crimson mix , length 16:37)

oggenc 1.0 (oggtools), q6 --> 22.894.652 bytes -- enc time 147 sec

oggenc 1.0 icl6(27 Jul 02), q6 --> 22.894.586 bytes -- 136 sec

oggenc 1.0 gcc 3.2, q6 --> 22.894.442 bytes -- 128 sec

-//- P4-icl6, q6 --> 22.542.507 bytes -- 107 sec

-//- P4-icl7, q6 --> 22.894.735 bytes -- 101 sec

Interesting. On that basis, I think one would have to suggest that the ICL 7 compile should be the one to use, both from a speed and bit similarity standpoint. I did try compiling with ICL 4.5 sometime back, just as an academic exercise, and the output from that was of a significantly different size (smaller) compared to the other compilers.

Anyway, thanks again. All this is of academic interest to me with my Athlon but is certainly relevant to the P4 ogg community!

How much faster do you think an ICL compile is vs. MSVC?

According to dethis, above, 40-50%. The offiical binaries were compiled by me using MSVC 6, as was requested.

is there a reason why those binaries don't run at all on my P3 ?

If you mean the P4 optimised ones, it's because they are optimised specifically for P4 only, there is no generic code in there at all. In other words, if you don't have a P4, don't try them, they won't run!

ok

Great work john33, thanks a lot

I did a brief test with my Celeron 1.7 GHz (Willamette P4 core), 640 MB ram (only PC 100 SDRAM )

track was a 6:00 min Heavy Metal wav, several encodes to make sure the wav is in cache, no command switches

-normal .exe from vorbis.com command line tools (2002-7-19):
elapsed time 1:05, rate 5.53 x realtime, bitrate 111,0

-John33's standard ICL 6 compile (2002-10-31):
elapsed time 1:01, rate 5.90 x realtime, bitrate 111,0

-John33's new ICL 6 compile for Pentium 4 (2002-12-06):
elapsed time 0:48, rate 7.49 x realtime, bitrate 109,1

-John33's new ICL 7 compile for Pentium 4 (2002-12-06):
elapsed time 0:46, rate 7.82 x realtime, bitrate 111,0
(25% savings against John33's standard, 30% saved against vorbis.com reference)


Looks like the P4 ICL 6 compile has different output, and ICL 7 is faster so I would not use the ICL 6 one

Great work, thanks again

Well, unless anyone has any objections, I'll take the ICL 6 P4 compile down tomorrow. It seems that the ICL 7 version is definitely the compile that should be used.

for a 3:46min file i got the following results on my p4 2.0 1024mb rdram i850

P4 ICL7 Compile 19.61secs
Compile 30.10.2002 29.79secs


thats about 30% faster.

According to my calculations, that's actually 51.9% faster.

btw John33, can you make a vorbisgain.exe (and maybe decoder, probably in one zip package) with the same ICL 7 settings?

~30% time savings is immense

thx and happy weekend

hehe, depends from what point of view
I used 100% - (100 * faster time / standard time)
so for Xenion's numbers that would be 34,17% savings

john33 is truly one of the more remarkable people I know.

@sony

Nah, it's not point of view. It's semantics. If you're going to talk about how much faster something is, then it's 51.9% faster, from his statistics. If you're going to talk about how much time it saves you, -- that's a different matter and yes, his/your figures are correct.

Seems like a reasonable request!!

I was thinking of posting P4 compiles of most of the ogg vorbis stuff. I'll do it over the next 24 hours.

does this ICL7 compile have SSE2 support or is SSE2 the reason why it is so much faster ?

Definitely yes to both.