[FFmpeg-devel] [PATCH] Do not assume long is same width as x86 register in swscale.

[Reimar Döffinger]

On Tue, Jan 27, 2009 at 09:05:08PM +0000, M?ns Rullg?rd wrote:
> Reimar D?ffinger <Reimar.Doeffinger at stud.uni-karlsruhe.de> writes:
> > No, that is wrong. int_fast32_t == int32_t would be best for
> > calculations etc., but as soon as it is used for indexed
> > addressing int_fast32_t == int64_t would be best.
> 
> You'll probably find that it is 32 or 64 bits on most modern machines.

Well, it is actually 64 bit at least for 64 bit linux/gcc, so it would work
here, I just think it is a bad idea since what is wanted here is a type
that can be used directly for indexed addressing, even if it was slow
for calculations.

> > The idea of a int_fast32_t type is plain broken for anything but RISC.
> 
> On any of the 36-bit machines, most of them very much non-RISC, it
> makes sense for int_fast32_t to be 36 bits wide whatever the purpose.

Well, I think the criteria would be "on any machine that only supports
one size >=32 bit natively". My statement was definitely over-broad, but
the same "issue" might arise with those machines if they were extended
to support e.g. 72-bit memory addressing...

Greetings,
Reimar D?ffinger