Re: Addition vs Multiplication on Pentium IV Processors
From: Christian (cstaudinger_at_hotmail.com)
Date: 07/31/03
- Next message: JS: "Re: Assembler versus ANSI C"
- Previous message: Bryan Olson: "Re: Into the Fire"
- In reply to: Paul Rubin: "Re: Addition vs Multiplication on Pentium IV Processors"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Date: 31 Jul 2003 07:11:23 -0700
Intel released VTune for Linux a couple of months ago. See
http://www.intel.com/software/products/vtune/vlin/. You may want to
ask your question in the VTune user-to-user forum at
www.intel.com/IDS/forums
Paul Rubin <http://phr.cx@NOSPAM.invalid> wrote in message news:<7x8yqg2shl.fsf@ruckus.brouhaha.com>...
> "George Joseph" <george.joseph@student.up.ac.za> writes:
> > I would like to know whether or not the Pentium IV processor uses the same
> > number of cycles to compute a 32-bit * 32-bit multiplication as it would do
> > to calculate 32-bit + 32-bit addition?
>
> I think on the p4, multiplication is 3 cycles. It's come up here
> before. It's 2 cycles on the p3. You also have to be concerned about
> the pipeline characteristics. Basically it's very hard to figure out
> these timings without experimentation. Intel has a tool called VTune
> (Windows only) which helps optimize code like that.
>
> The fastest way to do high-precision multiplication on the P4 is to
> use the SSE2 instructions. Eric Young has posted about that and
> there's some sample code on the Intel web site. I think the breakeven
> point for Karatsuba would be much higher than the number sizes
> normally used in cryptography.
- Next message: JS: "Re: Assembler versus ANSI C"
- Previous message: Bryan Olson: "Re: Into the Fire"
- In reply to: Paul Rubin: "Re: Addition vs Multiplication on Pentium IV Processors"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Relevant Pages
|
