On Sun, 2004-10-10 at 20:50, Robert Snyder wrote:
> Slack is fast and nimble. But noticed with some applications the
> extra optimaztion that are targeted for my Pentium 4 processor Seem
> to make a difference. With near identical setups between gentoo and
> slackware I noticing better Frames per sec in games. Large emulation
> products such as Basilisk II , SheepShaver and PearPC are snappier in
> there functions. All I was saying that thing felt snappier, saying
> something felt is far for sciencific benchmarks
Ah, yes, the gcc -O3 option must be used then.
That will liberally use SSE instead of the FPU.
I'd take a look at the "accuracy" of those optimizations.
Because the P3/P4 have dedicated SSE pipes.
Those pipes do not offer the accuracy and precision of the FPU.
In a nutshell, they weren't designed with accuracy or precision in mind,
but performance.
That means they were purposely engineered with "lossy math."
Tom's hardware guide was the first one to discover this years ago.
Comparing SSE to FPU MPEG-4 image quality, it was signficiant.
_All_ floating point math in a CPU is error-prone.
But the difference between x87 FPU and P3+ SSE are radically different.
The former, we're talking 16+ digits down.
With the latter, we could be talking very significant errors.
All done on _purpose_.
-- Bryan
P.S. Athlons/Opterons use their used FPU pipes for SSE operations --
i.e., no loss of accuracy or precision.
-- Bryan J. Smith b.j.smith@ieee.org ------------------------------------------------------------------ "Communities don't have rights. Only individuals in the community have rights. ... That idea of community rights is firmly rooted in the 'Communist Manifesto.'" -- Michael Badnarik----------------------------------------------------------------------- This list is provided as an unmoderated internet service by Networked Knowledge Systems (NKS). Views and opinions expressed in messages posted are those of the author and do not necessarily reflect the official policy or position of NKS or any of its employees.
This archive was generated by hypermail 2.1.3 : Fri Aug 01 2014 - 19:33:41 EDT