Re: Death by pointers.

>to ensure better use of cache lines, but in C, it cannot be, because
>some_array_ptr MIGHT be aliased to global. For large enough values of
>m, the transformed piece of code is at least 3 times faster on a
>high-performance workstation. Fortran rules usually allow this
>transformation.
I think what this really comes down to, is that the hardware
designers are being too clever by half. They should really be making
relatively simple chips, which are easy to couple many of them loosely
together.


Parallel programming can be really hairy if you go at it at fine time
grain, but there is more than enough going on on a modern workstation
to merely paralell process at a very high level.


Every time I look at the absobloominglutely amazing optimizations in
something like the pentium or a p6, I say to my self 'Wrong way
folks, too clever folks, go the other way folks, Keep It Simple
Stupid!'




John Carter
Institute for Water Quality Studies. Department of Water Affairs.
Internet : ece@dwaf-hri.pwv.gov.za Phone : 27-12-808-0374x194
Fax : 27-12-808-0338 [Host for Afwater list server]
[You might be surprised how many people have one big program that they want
to run fast. Multiple processes don't help much. -John]
--

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