[Question] FastFPE math emulation

[Question] FastFPE math emulation

[Kristoff Bonne]

Hi,

I am playing around with a "friendlyarm mini2440" development board.
It's a system based on a 400 Mhz samsung s3c2440 CPU, which is itself
based on a ARM920T core.
This system does not have a FPU. (nor integer DSP instruction-set).


I want to experiment to use this board for audio-related application,
e.g. the codec2 vocoder.
The implementation of that codec is now based on floating-point math.

First tests indicate that, using default settings, the board is not fast
enough to do realtime encoding or decoding of codec2 . (encoding a 34
second audio-sample takes 42 seconds).


Yesterday, I noticed the kernel has an option "FastFPE math emulation"
but when I install a kernel with that option enabled, I see no
difference at all.

- does FastFPE work on this kind of CPU.
- Is there something that needs to be done to activate it? (setting,
option during compiling, ...?)


Info:
# uname -a
Linux FriendlyARM 2.6.32.2-FriendlyARM #9 Wed Sep 7 21:38:49 CEST 2011
armv4tl unknown

$ grep -i fpe .config
CONFIG_FPE_NWFPE=y
# CONFIG_FPE_NWFPE_XP is not set
CONFIG_FPE_FASTFPE=y


# cat /proc/cpuinfo
Processor	: ARM920T rev 0 (v4l)
BogoMIPS	: 201.93
Features	: swp half
CPU implementer	: 0x41
CPU architecture: 4T
CPU variant	: 0x1
CPU part	: 0x920
CPU revision	: 0

Hardware	: FriendlyARM Mini2440 development board
Revision	: 0000
Serial		: 0000000000000000



http://friendlyarm.net/products/mini2440?lang=en
http://codec2.org/


Cheerio!
Kr. Bonne.