Speex dev - Jan 2006 - Compiler warnings with GCC 3.4.4 and perfomance question on a Arm 7

> OK, I forgot to update the ARM4 code. That should definitely be fixed
> (otherwise it's not going to work).
If this is not going to work then why does it work on my Arm?  Any
quick fix I can do for now?
> > The second warning is in ltp_arm4.h in "pitch_xcorr("
getting a warning of:
> > "matching constraint does not allow a register" on the  
" __asm__
> > __volatile__ " line.  I have no idea how to fix this warning.
>
> can you give me more info on this?
Not sure what info will help but here it goes...
I am using GCC 3.4.4 compiling ltp.c I get 8 warnings and all of them
seem to be when
"#define OVERRIDE_PITCH_XCORR
static void pitch_xcorr(const spx_word16_t *_x, const spx_word16_t
*_y, spx_word32_t *corr, int len, int nb_pitch, char *stack) "

is included from ltp_arm4.h
The error places the cursor on the          __asm__ __volatile__ (    
 line in the function.   I am compiling with Optimization Level 1.

I did a quick google search and found this where someone else had a
similar problem when switching from 3.2.3 to GCC 3.4.0.
http://gcc.gnu.org/ml/gcc/2004-04/msg01455.html

So what version of GCC are you using?
>
> I have found the performance/MHz of ARM CPUs is extremely variable.
> Given that, the results you're getting may make sense. Just be aware
> that there are still a few float operations left in wideband, so you
> might be hurt a bit by that.
>
Any suggestions as to which functions I should look at for the float
operations to change to Integer?  (assuming it is a simple task to
change the code from float to int or if I can copy how you change the
same kind of statements in the narrowband float to int conversion
otherwise I will probability will not figure it out and will just live
with the performance I have now.)

Thanks
Tom

On Tue, 2006-01-24 at 10:33 -0500, tom abcd wrote:
> > OK, I forgot to update the ARM4 code. That should definitely be fixed
> > (otherwise it's not going to work).
> 
> If this is not going to work then why does it work on my Arm?  Any
> quick fix I can do for now?
Didn't you mention you were only using the decoder? That function is
used in the encoder only.
> is included from ltp_arm4.h
> The error places the cursor on the          __asm__ __volatile__ (    
>  line in the function.   I am compiling with Optimization Level 1.
> 
> I did a quick google search and found this where someone else had a
> similar problem when switching from 3.2.3 to GCC 3.4.0.
> http://gcc.gnu.org/ml/gcc/2004-04/msg01455.html
I'll need to have a closer look at this to understand what's really
going on. I suspect the generated code is still right, though.
> So what version of GCC are you using?
I was using gcc 3.3 at the time I wrote this code.
> Any suggestions as to which functions I should look at for the float
> operations to change to Integer?  (assuming it is a simple task to
> change the code from float to int or if I can copy how you change the
> same kind of statements in the narrowband float to int conversion
> otherwise I will probability will not figure it out and will just live
> with the performance I have now.)
I think most of the float ops left are in sb_celp.c but float->fixed
conversion isn't trivial, since you always need to make sure that the
quality isn't degraded by the loss of accuracy (for each change made).

	Jean-Marc

> Didn't you mention you were only using the decoder? That function is
> used in the encoder only.
Yes I am only using the decoder - now it make sense why it works (I
did not check how the function was used only that I got the warning.)
> I'll need to have a closer look at this to understand what's really
> going on. I suspect the generated code is still right, though.
>
Ok - will just use as is for now then
>
> I think most of the float ops left are in sb_celp.c but float->fixed
> conversion isn't trivial, since you always need to make sure that the
> quality isn't degraded by the loss of accuracy (for each change made).
Sounds like this is beyond what I want to try to do and I have a
working solution now so I just use this as is.

Thanks
Tom