Re: Optimizations in C/C++ code doing matrix multiplications

marjan.sterk@ijs.si (Marjan Sterk) wrote in message news:02-03-194...
> On 22 Mar 2002 21:03:51 -0500, "H. Ellenberger" <ele1@gmx.ch> wrote:
>
> >Which compilers are smart enough to automagically optimize this kind
> >of data acess by introducing hidden pointer variables (in registers)
> >incremented and thus avoiding most multiplications?


I think you'll find that most modern optimizing compilers will do
strength reduction, linear function test replacement, and inducation
variable optimizations to produce fairly good code for matrix
multiplications.


However, you'll probably also find very few compilers that will do
loop interchange & blocking (which can have a dramatic effect on
performance) as well as ones that have support for things like the
restrict keyword of C99 (which would allow the optimizer to assume
that the destination matrix didn't overlap either of the source
matrices if they are passed in as pointer parameters).


Mark Lacey
mark@marklacey.com
[My impression is that Fortran compilers do that kind of stuff, other
languages don't. -John]

Return to the comp.compilers page.
Search the comp.compilers archives again.