[ODE] Some SSE in Quick step

[Adam D. Moss]

James Arthur wrote:
> I'm guessing that FORTRAN95 would be of no use to you. *sigh*

Haha, nice idea. :)

> VectorC? www.codeplay.com - not tried it, mind.

Right, I'd heard about this.  It's not really what I'm looking
for, though.

I was hoping that something had bubbled up through the C
standard by now (ANSI or.. de facto) that meant that it was
possible to use SIMD pretty portably, abstracted at a low-level
with compiler-level support for arch-aware scheduling and register
allocation (that is to say, giving a lot more expressiveness
and optimization opportunities even than sets of header macros with
various flavours of inline asm) while still letting the programmer
be pretty explicit about vectorisation instead of waiting for
the compiler's auto-SIMD-ing heuristics to get better (I'm not
sure they could ever be THAT good without a lot of manual
guidance, since SIMD instruction sets seem universally
alignment-sensitive and a compiler doesn't have enough implicit
information to guess the alignment of a pointer ahead of time
in most cases).

But I guess nothing like that is showing up yet.

--Adam
-- 
Adam D. Moss   . ,,^^   adam at gimp.org   http://www.foxbox.org/   co:3