[Coco] Drivewire mixed results...

[Chuck Youse]

On Mon, 2008-09-29 at 12:33 -0400, Chuck Youse wrote:
> On Mon, 2008-09-29 at 10:47 -0500, David Gacke wrote:

> But then you had to go and bring up SIMD and compilers with aggressive
> global optimizations, and I'll grant that theoretically you have some
> ground there - but aside from the most obvious cases (the use of
> constants, etc.) I don't see how an optimizer could make use of any of
> this information while still guaranteeing standard behavior.

Let me clarify: by "this information" I mean the explicit typing, which
is what we're talking about.  An optimizer smart enough to unroll the
loop into, say, groups of 8, and parallelize a checksum computation will
also be smart enough to know that the sum of 8 unsigned 8-bit integers
will always fit into a short.

When it comes to totalling the groups, that will be a conventional
operation and will probably take place in a full integer anyway; while
it's possible that a REALLY smart optimizer could figure out that the
sum of the sum is also a matrix operation, it wouldn't choose to do so
because that would almost certainly result in a register overflow and
subsequent dump and restore from memory.

Anyway, just thinking out loud now.  Getting my fork ready!

C.