what's wrong with -mcpu=arm946e-s on gcc4

matc

11

« on: 07 / July / 2011, 16:09:40 »

Advertisements

Hi,

the makefile say

Code: [Select]

        # TODO should be -mcpu, but breaks things ATM
        CFLAGS+=-mtune=arm946e-s

But after looking on the forum I am not able to found what's the problem.

Trying it with 4.5.2 seems to works.

What's the problem with this options ?

Logged

reyalp

14080

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #1 on: 07 / July / 2011, 21:41:16 »

I think we had configuration issues with GCC/toolchain which prevented it from building correctly, see http://chdk.setepontos.com/index.php?topic=2500.msg24326#msg24326

Logged

Don't forget what the H stands for.

matc

11

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #2 on: 08 / July / 2011, 16:25:01 »

Quote from: reyalp on 07 / July / 2011, 21:41:16

I think we had configuration issues with GCC/toolchain which prevented it from building correctly, see http://chdk.setepontos.com/index.php?topic=2500.msg24326#msg24326

Thanks for the link. But what's strange is that the fix shouldn't be hard. It was proposed 2 years ago.
Seems nobody care about this ?

That weird because -mcpu=arm946e-s should improve performance.

Logged

Lebeau

187

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #3 on: 08 / July / 2011, 21:23:27 »

Matc ! That's a pertinent and interesting question !

I made some searches because I am interested, but not enough experienced in gcc, arm and all. Here are my modest found.
http://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html
http://gcc.gnu.org/wiki/Software_floating_point
http://gcc.gnu.org/onlinedocs/gccint/Soft-float-library-routines.html
http://www.jhauser.us/arithmetic/SoftFloat.html
http://lists.busybox.net/pipermail/buildroot/2011-May/043192.html

I donT know if they are pertinent but, I am motivated

Logged

Galerie Photobucket, Canal YouTube, Galerie Panoramio

reyalp

14080

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #4 on: 09 / July / 2011, 00:21:55 »

Quote from: matc on 08 / July / 2011, 16:25:01

Thanks for the link. But what's strange is that the fix shouldn't be hard. It was proposed 2 years ago.
Seems nobody care about this ?

The existing setup works. I have better things to do than spend days playing lotto with gcc and binutils configurations for some marginal performance gain. It would be nice but it's not a high priority.

Maybe I'm missing it, but I don't see a clear solution in that thread.

Logged

Don't forget what the H stands for.

waterwingz

12138

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #5 on: 09 / July / 2011, 02:11:55 »

Quote from: reyalp on 09 / July / 2011, 00:21:55

The existing setup works. I have better things to do than spend days playing lotto with gcc and binutils configurations for some marginal performance gain. It would be nice but it's not a high priority.

Maybe I'm missing it, but I don't see a clear solution in that thread.

Yup - if its not broken - don't fix it. We have a lot of other things to worry about before this.

Logged

Ported : A1200 SD940 G10 Powershot N G16

reyalp

14080

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #6 on: 09 / July / 2011, 03:57:06 »

Quote from: waterwingz on 09 / July / 2011, 02:11:55

Yup - if its not broken - don't fix it. We have a lot of other things to worry about before this.

It should give a decent memory saving, which is something we can definitely use. I'm not against someone working on this, it's just not near the top of my personal list

Logged

Don't forget what the H stands for.

Microfunguy

4037

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #7 on: 09 / July / 2011, 05:29:45 »

Quote from: reyalp on 09 / July / 2011, 03:57:06

It should give a decent memory saving, which is something we can definitely use.

I thought extended memory was the solution for that ?

Logged

reyalp

14080

Re: what's wrong with -mcpu=arm946e-s on gcc4

« Reply #8 on: 09 / July / 2011, 15:45:53 »

Quote from: Microfunguy on 09 / July / 2011, 05:29:45

I thought extended memory was the solution for that ?

Maybe. The current implementation is very hacky, not clear it will work everywhere. Plus, no matter where you put it, it's taking away memory from the camera.

Logged

Don't forget what the H stands for.