[FFmpeg-devel] Fixed point implementations of AAC decoder, AC3 decoder and 32bit FFT

[Zeljko Lukac]

Here are some benchmark scores for all three profiles:  

	            MIPS			                           ARM					
	  FLOAT	FIXED       GAIN [%]	      FLOAT	      FIXED	      GAIN [%]	
LC	  12.42	11.10	10.57	      56.31	      22.12	       60.71
HE	  40.47	37.93	  6.26	    177.12	      45.98	       74.04
HEv2	  77.44	71.52	  7.64	    246.20	      75.68	       69.26

Measured on 

MIPS Malta board (MIPS 74Kf)
mips-linux-gnu-gcc (Sourcery G++ Lite 4.4-303) 4.4.1

and 

ARM BeagleBoard-xM
arm-none-linux-gnueabi-gcc (Sourcery CodeBench Lite 2011.09-70) 4.6.1

Fixed point implementation is faster than floating point on MIPS and ARM.

Best regards,
Zeljko



________________________________________
From: Michael Niedermayer [michaelni at gmx.at]
Sent: Tuesday, June 18, 2013 6:18 PM
To: FFmpeg development discussions and patches
Cc: Mirjana Vulin; Zeljko Lukac; Bojan Zivkovic
Subject: Re: [FFmpeg-devel] Fixed point implementations of AAC decoder, AC3 decoder and 32bit FFT

On Tue, Jun 18, 2013 at 02:36:54PM +0000, Nedeljko Babic wrote:
> Hello,
>
> Has anyone had a chance to start reviewing any of the patches from this
> set (FFT, AC3 decoder or AAC decoder patches)?

Do you have some benchmark scores that show what speed effect these
have for MIPS ?

(also if these are faster on more than just MIPS then this could
 raise interrest in people and also bring forth more people who might
 want to improve the code further)
so benchmarks on other arhcitectures are welcome too, if anyone wants
to do such benchmarks

[...]
--
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Awnsering whenever a program halts or runs forever is
On a turing machine, in general impossible (turings halting problem).
On any real computer, always possible as a real computer has a finite number
of states N, and will either halt in less than N cycles or never halt.