[FFmpeg-cvslog] r12165 - trunk/libavcodec/dsputil.c

[Måns Rullgård]

Rich Felker <dalias at aerifal.cx> writes:

> On Thu, Feb 21, 2008 at 09:20:00PM +0000, M?ns Rullg?rd wrote:
>> Uoti Urpala <uoti.urpala at pp1.inet.fi> writes:
>> 
>> > On Thu, 2008-02-21 at 20:28 +0200, Uoti Urpala wrote:
>> >> On Thu, 2008-02-21 at 08:54 +0100, lorenm wrote:
>> >> > +static const unsigned long pb_7f = 0x7f7f7f7f7f7f7f7fUL;
>> >> > +static const unsigned long pb_80 = 0x8080808080808080UL;
>> >> 
>> >> I think these should have a comment about truncation on 32-bit platforms
>> >> being intentional. And the code will fail on a 128-bit platform :)
>> >
>> > I forgot to add that these could be written in a way that is independent
>> > of sizeof(long) and doesn't cause compiler warnings about truncation on
>> > platforms with shorter longs:
>> > static const unsigned long pb_7f = -1UL/255 * 0x7f;
>> 
>> This assumes two's complement signed numbers.  Using ~0UL avoids this
>> assumption.
>
> Yes, a completely reasonable assumption. There are plenty of things
> that will go to hell without twos complement and absolutely no
> motivations for other integer representations except for historical
> systems from the 1960s whose creators failed algebra. =)

I don't like relying on assumptions, however reasonable, when a
simple, correct solution exists.

-- 
M?ns Rullg?rd
mans at mansr.com