Re: [fluid-dev] Crashing bug in fluidsynth (CVS)

[Josh Green]

On Fri, 2005-09-16 at 17:18 -0700, Mo DeJong wrote:
> On Thu, 15 Sep 2005 01:44:37 -0700
> Mo DeJong <address@hidden> wrote:
> 
> ...
> > This is running in a 32 bit Win32 machine. As far as I can tell, this has 
> > nothing
> > to do with the decl type of fluid_phase_t, there is something very wrong 
> > with
> > the pointer voice->sample->data, it seems that when the dsp code in
> > fluid_voice.c is entered, the voice->sample->data has already gone bad
> 
> The attached patch adds an array access before entering the dsp loop to
> make it easier to see this problem. It also fixes up the code so that it is
> valid C code (decls before use) when the SSE ifdef is enabled. The patch
> also moves dsp_buf onto the heap instead of keeping it on the stack since
> that makes it easier to debug with a malloc debugger in case something
> were to go wrong in the processing loop. The patch adds two locals
> dsp_data_start and dsp_data_end that track the dsp_data bounds and
> adds some #if commented code to check the bounds in the dsp loop if
> a developer wanted to enable those extra checks.
> 
> cheers
> Mo DeJong
> 

Do you have SSE enabled on your setup?  This option is rather
experimental and doesn't really provide any performance increase right
now AFAIK.
Your patch looks useful for debugging this issue, but do you think it
should be something included in the source?  I suppose it could all be
#ifdef'd with a define like "FLUID_CHECK_DSP_BOUNDARY" or something.  In
retrospect though, it SHOULD be working :)  When this problem is fixed
it may not be needed anymore.  I do see the issue with the SSE code and
decls being defined after instructions, which should probably be fixed
(too bad the SSE code isn't actually useful currently).  Cheers!
        Josh Green

signature.asc
Description: This is a digitally signed message part