|
 |
On 13 Aug 1999 02:39:12 -0400, Nieminen Mika wrote:
>Ron Parker <par### [at] fwi com> wrote:
>:>When?
>
>: Probably never. Portability has always been more important than
>: speed. In any case, both SSE and 3DNow are single-precision only,
>: so it would be hard to make POV benefit from them.
>
> If someone makes a compiler which can optimize code for SSE or 3DNow,
>it will be absolutely possible to compile an optimized version.
> However, such a compiler is hardly possible (for the reason Ron stated;
>if you use doubles, the compiler _can't_ optimize them to floats).
Actually, Intel does make such a compiler (for SSE, obviously) and it's a
drop-in replacement for MSVC. I have access to it. The bad news: It makes
my K6 run 3 seconds *slower* on skyvase than the same code compiled with
the original MSVC compiler. If anyone wants to give me a P///, I can test
it with that...
Post a reply to this message
|
|
