List of recommended LAME compiles, Discussion Options

[AgentMil]

Are these the same as the LAME 3.90 Stable binaries?

[Dibrom]

Yes, except they are my own compiles. They should be a bit faster than the "fast" compiles from Mitiok.

[Dibrom]

Updated with even faster compiles.

[Enig]

Dibrom, would you please offer a slow version compiled by MSVC which I and others may be happy to get because it produce a bit smaller MP3s.

Thank you in advance.

[Dibrom]

OK, I'll have to upload these a little later though.

[jkml]

Originally posted by Enig
Dibrom, would you please offer a slow version compiled by MSVC which I and others may be happy to get because it produce a bit smaller MP3s.

Does anyone know why the executables compiled with different compilers produce MP3's of different sizes? Isn't the source code the same except compiler-specific optimizations? :confused:

[superorc]

each uses a different assembler to produce the files, and each one produces different output then another one. but i tried mitiok's and dibroms and noticed little i think mitioks might have been faster on my amd thunderbird 1.4 ghz but even then on --alt-preset insane i get like 7.2x realtime and around 4.5 realtime --alt-preset standard.

[Dibrom]

Originally posted by superorc
but i tried mitiok's and dibroms and noticed little i think mitioks might have been faster on my amd thunderbird 1.4 ghz but even then on --alt-preset insane i get like 7.2x realtime and around 4.5 realtime --alt-preset standard.

Hrmm.. are you sure his was faster? The things I changed should not have slowed down encoding.

Also, did you try the faster version? That one should for sure be faster than his normal compile. If it isn't, then something is wrong..

Make sure that when comparing speed you test on the [b]exact same file with both compiles, some files will encode slower than others, especially with --alt-preset standard.

[john33]

The lame_enc.dll which has been provided by Dibrom is DLL3. The others are still available at rjamorim's site.

BTW, thanks for the 'blessing', Dibrom, I feel rather flattered!!

john33

[jkml]

Originally posted by superorc
each uses a different assembler to produce the files, and each one produces different output then another one. but i tried mitiok's and dibroms and noticed little i think mitioks might have been faster on my amd thunderbird 1.4 ghz but even then on --alt-preset insane i get like 7.2x realtime and around 4.5 realtime --alt-preset standard.

The executables will certainly be different, but, given the same wav file, they should produce identical MP3 files, right? It is the algorithm that affects the output (e.g. size of the encoded MP3 files) and the implementation of that algorithm that affects the executable (speed and size of lame.exe itself). So why does the MSVC-compiled lame.exe produce a smaller .mp3 file than the one produced by the ICL-compiled lame.exe??

:confused:

[MaTTeR]

Any chance this DLL has the new --alt-presets compiled in it? I don't mean to be pushy, just curious.

Happy Holidays!

EDIT-- Argh...just seen the fine print at the top about mapping the --alt-presets. Sorry.

[Dibrom]

Originally posted by jkml
The executables will certainly be different, but, given the same wav file, they should produce identical MP3 files, right?  It is the algorithm that affects the output (e.g. size of the encoded MP3 files) and the implementation of that algorithm that affects the executable (speed and size of lame.exe itself).  So why does the MSVC-compiled lame.exe produce a smaller .mp3 file than the one produced by the ICL-compiled lame.exe??

A lot of the algorithms are purposely based on not [b]absolutely precise math (floating point). Mainly due to this, and issues such as rounding when converting from floating point to integers and the like, when compilers begin to aggressively optimize the code in favor of speed, certain differences can creep in.

In short, the code isn't as "precise" as you think it is, and certain compilers take advantage of this more than others.

[superorc]

yes i used the exact same file. i didnt use the normal version i used the fast version and it was still slower.maybe it has to do with me not having SSE?

[Dibrom]

Originally posted by superorc
yes i used the exact same file. i didnt use the normal version i used the fast version and it was still slower.maybe it has to do with me not having SSE?

LAME doesn't make use of SSE, and the flags I use to compile with do not allow ICL to use SSE during vectorization either.

Can you give me some info on your system? And can you post some histograms of my compile vs the other one? In the testing I have done, my compile is a bit faster and a few other people seem to have confirmed this also.

And to be sure, you are testing on the exact same file?

Unless Mitiok is doing something different with his new compile, I don't see how it is possible that my compile (especially the fast one) is actually slower. If this is the case on your system though I'd be interested in more information.

[Dibrom]

Here are some benchmarks of my "fast" compile vs the compile on Mitiok's page, run on a p2 300 laptop:

http://static.hydrogenaudio.org/extra/p2bench.txt

Tomorrow I'll post some benchmarks from a p3 900, athlon 1.4ghz and a p4 1.7ghz as well.

[superorc]

ok i guess im wrong. i benchmarked both, but instead of using one of those test clips i used Rammstein - Du Hast for the test song.

http://www.mycgiserver.com/~superorc/lame.txt

heres my system info:

athlon 1.4 ghz tbird
1 gig ram
60 gig hd, and 20 gig both ntfs
win2k.

also do you know what mitiok uses for compiling??? the msvc or icl?

[Dibrom]

Thanks, that clears it up some

Mitiok uses ICL 4.5, same as me, I just use some slightly different compile time optimizations. Produced files are bit identical though.

[AgentMil]

On my computer using mitiok's compiles yield a speed of 3.4x (*approx. it goes up and down) and using Dibrom's compiles yielded a speed of 3.4~5x (*approx. it goes up and down).

Dibroms compiles are faster on my account by the 5% - 10% mentioned on an earlier post.

Also FYI "fast" version encoded files are a tiny,weeny, little bit bigger than slower compiles.

AgentMil

[Dibrom]

Originally posted by AgentMil
Also FYI "fast" version encoded files are a tiny,weeny, little bit bigger than slower compiles.

Hrmm.. are you sure? They should be bit identical. Is LAME reporting a different bitrate between the two? Or are you looking purely at filesize? More info would be cool

[superorc]

i will be compiling my own version of lame, but using whats normally used to compile lame in win32 and thats msvc + nasm. ill see how that goes and add it to my list.

[AgentMil]

HEHE I meant ICL encoded binaries when compared with binaries compiled using a different compiler.

I compared 3 different compiles yours (Dibroms), Mitioks and www.mp3-tech.org compile.

Listed below are the speed in order of fastest to slowest:
1.) Dibrom
2.) mitioks
3.) www.mp3-tech.org (I only get 2.7X on this compile)

Listed below are bitrates in order of smallest to biggest:
1.) www.mp3-tech.org
2.) mitioks and dibrom

I hope this clears things up.

AgentMil

[Dibrom]

AgentMil:

Ah, OK, I thought you meant going from my normal ICL compile to my "fast" compile. There should be no bitrate difference there. With MSVC compiles though, there is a slight difference. MSVC is slightly smaller, and usually a fair deal slower. I'll probably be putting an MSVC compile up before too long here..

[AgentMil]

It also seems that there is two different compiles of mitioks binaries.

I downloaded one from mitioks site as soon as it was available, and then today I decided to download it again (because it was not on my laptop, and was too lazy to boot up desktop to copy file), and then went to run, I noticed that the ")" at the end of the (www.mp3-dev.org) had reappeared in the "new" exe as in the first release exe the ) was missing, so I booted up my desktop and did fc/b on the two exe, and found that it was different.

Can anyone else verify this?

AgentMil

[Dibrom]

I updated my benchmarks with results from an MSVC compile (I'll be posting the MSVC compile shortly).

And yes, Mitiok updated the compile to add the ) back in.

EDIT: Oops, forgot to run tests with "fast" standard. I'll do that and add results in a few.

EDIT2: "fast" standard results are now in.

[AgentMil]

HEHE that is very funny, a whole new re compile because of a missing ). Guess it didn't look to good aesthetically with the missing ) .

Waiting for your MSVC version, so I can test it out.
Dibrom is it possible to upload test results as in a text file onto this forum?

AgentMil