mppenc 1.16 slower? maybe it's my compiler

21 January 2007, 07:41 pm

hey folks. i was glad to hear there's a new mppenc out. i was hoping for a binary, like previous releases, but i went ahead and compiled it.

i finally got around to using it today, and it seems the encoding speed is pretty slow. I'm currently running at around 7x, where normally with 1.15 i'd be getting 22x. not that it makes much difference, since i'm ripping and encoding at the same time, but it still worries me.

i noticed 1.15 uses nasm to compile and 1.16 doesn't.

i'm running Arch Linux, which is a bleeding edge i686 distro. i have cmake 2.4.5, gcc 4.1.2, automake 1.10, autoconf 2.61, nasm 0.98.39 ... that's all i could think of that is relevant.

anyway, so i'm hoping to see people reporting their 1.15 vs 1.16 speeds (i'm assuming no slowdown, since somebody else would have complained by now), and hopefully a statically linked binary. or maybe instructions on how to extract a deb package, since they have binaries.

thanks!

21 January 2007, 07:53 pm

don't you hate when people post without doing some research first?

well, i extracted the rarewares deb package, and yes, it encodes faster.

so then something about my compiler is producing slower builds than what you guys are using, that wasn't there in 1.15.

xmixahlx · 21 January 2007, 11:12 pm

the difference is compiler flags... try using -O3 -ffast-math

later

26 January 2007, 08:24 pm

Quote:

Originally Posted by xmixahlx

the difference is compiler flags... try using -O3 -ffast-math

hmm that hasn't improved things. am i doing it wrong?

make clean
CFLAGS="-O3 -ffast-math" make

i also tried exporting CFLAGS first, then running make. same results.

Lefungus · 27 January 2007, 12:50 pm

I only got those awful speeds when using a debug binary. Anyway, it's entirely dependent on your CFLAGS. You can gain a bit by tweaking it, or crippling it.
mppenc-1.16 *should* use correct cflags by default (they're no the best for your arch but should work ok on most cpu).
Since you're using arch linux, please post the PKGBUILD you're using.

About nasm, it's not used in 1.16 because as far as I'm aware:
- It's 3dnow code only
- It has been disabled for ages in previous mppenc versions

29 January 2007, 08:01 am

actually, i don't have a PKGBUILD as i simply tried building mppenc to see how it performed (i wanted to see how the new fast seeking compared between 1.15 and 1.16 on a 30-minute recording)

i used the default cflags at first, then tried specifying the ones xmix recommended.
lefungus, you're an arch user, right? i think i've seen you on the forums there (i'm known as paranoos usually). you've compiled mppenc on an arch system, and it ran full speed? i haven't done anything strange (to my knowledge!), so i can't understand it. especially considering i can compile 1.15 and it works fine.

Lefungus · 29 January 2007, 06:18 pm

I used arch a few years ago, I tried to reinstall it this week end, but all their iso refuse to boot on my computer, probably due too an incompatibility with my motherboard.

What you can try is setting
"set(CMAKE_VERBOSE_MAKEFILE false)"
to
"set(CMAKE_VERBOSE_MAKEFILE true)"
in CMakeLists.txt
and attach the log somewhere here

30 January 2007, 05:40 am

alright, here's the log.
(hmm... the file doesn't appear on the list below. the file is stored here. hopefully that works.)

Lefungus · 30 January 2007, 07:12 pm

It's not the log I want

Try, with the modified CMakeLists.txt

"make clean && cmake . && make > output.log 2>&1"

and post output.log somewhere

01 February 2007, 06:29 am

whoops

here's a paste of the output

01 February 2007, 06:36 am

umm....

it appears that trying to run make inside the mppenc-1.16 directory yields good results, while running make while in the src subdirectory results in the slow build i was experiencing earlier.

strange! well, thanks anyway!

21 January 2007, 07:41 pm	#1
nospaulatu Guest Posts: n/a	mppenc 1.16 slower? maybe it's my compiler hey folks. i was glad to hear there's a new mppenc out. i was hoping for a binary, like previous releases, but i went ahead and compiled it. i finally got around to using it today, and it seems the encoding speed is pretty slow. I'm currently running at around 7x, where normally with 1.15 i'd be getting 22x. not that it makes much difference, since i'm ripping and encoding at the same time, but it still worries me. i noticed 1.15 uses nasm to compile and 1.16 doesn't. i'm running Arch Linux, which is a bleeding edge i686 distro. i have cmake 2.4.5, gcc 4.1.2, automake 1.10, autoconf 2.61, nasm 0.98.39 ... that's all i could think of that is relevant. anyway, so i'm hoping to see people reporting their 1.15 vs 1.16 speeds (i'm assuming no slowdown, since somebody else would have complained by now), and hopefully a statically linked binary. or maybe instructions on how to extract a deb package, since they have binaries. thanks!

21 January 2007, 11:12 pm	#3
xmixahlx Musepack developer Join Date: Nov 2004 Location: seattle, washington usa Posts: 111	the difference is compiler flags... try using -O3 -ffast-math later __________________ -xmixahlx, the one they call "mike" http://xmixahlx.com -\|- http://rarewares.org

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
mppenc 1.16/libmpcdec 1.2.3	Seed	General	8	11 December 2006 03:03 am
mppenc 1.16 installation on kubuntu?	monkman	MPC for UNIX	2	02 December 2006 05:56 pm
mppenc 1.15v	Seed	General	5	05 September 2005 01:03 am
mppenc 1.15u	Seed	General	7	03 March 2005 04:37 am
mppenc 1.15t	Seed	General	1	25 February 2005 12:17 pm

21 January 2007, 07:53 pm	#2
nospaulatu Guest Posts: n/a	don't you hate when people post without doing some research first? well, i extracted the rarewares deb package, and yes, it encodes faster. so then something about my compiler is producing slower builds than what you guys are using, that wasn't there in 1.15.

27 January 2007, 12:50 pm	#5
Lefungus Procrastinator Join Date: Jul 2004 Posts: 131	I only got those awful speeds when using a debug binary. Anyway, it's entirely dependent on your CFLAGS. You can gain a bit by tweaking it, or crippling it. mppenc-1.16 should use correct cflags by default (they're no the best for your arch but should work ok on most cpu). Since you're using arch linux, please post the PKGBUILD you're using. About nasm, it's not used in 1.16 because as far as I'm aware: - It's 3dnow code only - It has been disabled for ages in previous mppenc versions

29 January 2007, 08:01 am	#6
nospaulatu Guest Posts: n/a	actually, i don't have a PKGBUILD as i simply tried building mppenc to see how it performed (i wanted to see how the new fast seeking compared between 1.15 and 1.16 on a 30-minute recording) i used the default cflags at first, then tried specifying the ones xmix recommended. lefungus, you're an arch user, right? i think i've seen you on the forums there (i'm known as paranoos usually). you've compiled mppenc on an arch system, and it ran full speed? i haven't done anything strange (to my knowledge!), so i can't understand it. especially considering i can compile 1.15 and it works fine.

29 January 2007, 06:18 pm	#7
Lefungus Procrastinator Join Date: Jul 2004 Posts: 131	I used arch a few years ago, I tried to reinstall it this week end, but all their iso refuse to boot on my computer, probably due too an incompatibility with my motherboard. What you can try is setting "set(CMAKE_VERBOSE_MAKEFILE false)" to "set(CMAKE_VERBOSE_MAKEFILE true)" in CMakeLists.txt and attach the log somewhere here

30 January 2007, 05:40 am	#8
nospaulatu Guest Posts: n/a	alright, here's the log. (hmm... the file doesn't appear on the list below. the file is stored here. hopefully that works.)

30 January 2007, 07:12 pm	#9
Lefungus Procrastinator Join Date: Jul 2004 Posts: 131	It's not the log I want Try, with the modified CMakeLists.txt "make clean && cmake . && make > output.log 2>&1" and post output.log somewhere

01 February 2007, 06:29 am	#10
nospaulatu Guest Posts: n/a	whoops here's a paste of the output

01 February 2007, 06:36 am	#11
nospaulatu Guest Posts: n/a	umm.... it appears that trying to run make inside the mppenc-1.16 directory yields good results, while running make while in the src subdirectory results in the slow build i was experiencing earlier. strange! well, thanks anyway!