[FFmpeg-devel] [PATCH] AAC: type puns for 16 bit floating point rounding

Robert Swain robert.swain
Tue Dec 9 18:22:25 CET 2008

2008/12/9 M?ns Rullg?rd <mans at mansr.com>:
> Robert Swain wrote:
>> 2008/12/5 M?ns Rullg?rd <mans at mansr.com>:
>>> Uoti Urpala <uoti.urpala at pp1.inet.fi> writes:
>> [lots of arguing]
>> So, where are we with this?
> The type punning using unions is defined by the latest version of the C
> spec, and it seems to work with most compilers.
> Two questions remain to be answered for each platform:
> 1. Does it use IEEE754 floating-point format?
> 2. Is the hack faster than using normal float operations?
> We know 1 to be true for everything we're ever likely to encounter.
> 2 needs to be tested.

Alex says he's tested it and confirmed it to be faster on x86 and
x86_64 (core 2 and core 2 mobile, not necessarily respectively).

M?ns - do you want to test it on ARM?

Can someone test it on PPC? Can someone test it on an older x86/x86_64
CPU? I have a Core Duo MacBook so I'll test it on that but I was
thinking more along the lines of someone with an ancient K6 or P2/P3
or something. :) What else do we care about?


More information about the ffmpeg-devel mailing list