Enable neon float hardware support

Ricardo Crudo ricardo.crudo at gmail.com
Fri Mar 14 11:57:28 EDT 2014


Hello,

I'm working on an embedded linux audio device with Allwinner A10/A20
cpu. For this device is very important use the neon cpu resource.
I've some questions about kernel and toolchains.

1. To use the neon I need to build the kernel using any particular
option or apply any patch?
2. I need to build the linux kernel using arm-linux-gnueabi-xxx or
arm-linux-gnueabihf-xxx toolchain? What's the difference?
3. the gcc option -mfloat-abi=hard is applicable to my situation?
4. There is any difference use -mfpu='neon' with arm-linux-gnueabi-xxx
or arm-linux-gnueabihf-xxx?

Currently, to build my application, I'm using the following gcc flags:
-O3 -ffast-math -march=armv7 -mtune=cortex-a8 -mfpu='neon'
-funsafe-math-optimizations

I'd like to hear some opinions about this, Is this a good setup?
What's the risk using -funsafe-math-optimizations?

Remarks:
* application here is a program that process audio algorithm
* currently I'm using a cubieboard 1/2 running a arch linux distro
from archlinuxarm.org

Regards,
Ricardo



More information about the linux-arm-kernel mailing list