Re: [speex-dev] Re: [speex-dev]

Gervase Markham gerv
Wed Jul 14 08:04:48 PDT 2004

In-Reply-To: <1085699564.3286.8.camel at>
References: <6735222D09423F448E2A238E06D260F12F6DBE at inexg.GRAPECITY.NET>	<1085416539.4993.11.camel at>	<40B30D61.3080601 at>
<40B65BD9.2020904 at>
<1085699564.3286.8.camel at>
Message-ID: <40F54B90.10700 at>

Jean-Marc Valin wrote:
> Right now, the code *does* assume that a char is 8 bits.

The decoder works on my [16-bit char] platform fine in floating-point
mode; it just fails in fixed-point mode. Should I be surprised at this,
or is that what you would expect?

> However, this
> assumption is only made in the bit packet (bits.c), so it should be
> fairly easy to make it work for a platform where a char is 16 bits.

So I would have to go through bits.c, work out which bits were 8-bit
dependent, and try and change them to work with 16-bit chars?

Is there any documentation on how bits.c works, at a high level? It
doesn't seem to have any comments in it.

Do you think it might be smart to attempt to get this working by, say,
using the Linux or Windows version and #defining "char" to "short"? That
might be easier to debug than an embedded test board...

> Still regarding fixed-point, right now, spx_word_16 is defined as
> "short" and spx_word_32 is defined as "int", but that's only in arch.h
> and eventually, it should be replaced by something a bit better.

That's probably OK; my shorts are 16 bits too, and I'm fairly sure my
ints are 32.

> Jean-Marc Valin

By the way, this URL is currently a 404...


More information about the Speex-dev mailing list