Re: Calculating cycles/byte
From: Mok-Kong Shen (mok-kong.shen_at_t-online.de)
Date: 07/29/03
- Next message: Jim Steuert: "Re: Magic Flight: A Public Key Algorithm Based on a Lossy Commutative Rng Mixer"
- Previous message: Mark Wooding: "Re: random data"
- In reply to: J. Campbell: "Calculating cycles/byte"
- Next in thread: Douglas A. Gwyn: "Re: Calculating cycles/byte"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Date: Tue, 29 Jul 2003 19:59:38 +0200
"J. Campbell" wrote:
>
> I've been toying with some hash-functions and wanted to know how
> people measure "cycles/byte". On my 1.5 GHz, win xp machine (on-line
> with several programs running), I can push an 18 MB zip file through
> my alg in about 0.1 sec. Back of envelope, I get ((1.5 bil cyc/sec) *
> (0.1 sec) / (18 mil bytes) = 8.3 cyc/byte.)
>
> Does anyone know how to accurately calculate the cyc/byte?
The same instruction may need different cycles for
different memebers of e.g. the Intel family. So I wonder
why do you want that quantity (since it can't be translated
to different hardware). If you do want that, you could use
the C standard function clock() to time the processing of
n bytes, with n sufficiently large in order to get a
significant value, say of the order of 100 sec. Then knowing
the MHz of your machine, I suppose you could get a reasonably
good result (though not 'very' accurate in my experience).
(I would prefer not to let concurrent user jobs be run during
the measurement, though I am not sure whether that's really
necessary.)
M. K. Shen
- Next message: Jim Steuert: "Re: Magic Flight: A Public Key Algorithm Based on a Lossy Commutative Rng Mixer"
- Previous message: Mark Wooding: "Re: random data"
- In reply to: J. Campbell: "Calculating cycles/byte"
- Next in thread: Douglas A. Gwyn: "Re: Calculating cycles/byte"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Relevant Pages
|
|
