Messages in this thread | | | Date | Wed, 24 Jun 1998 21:12:15 -0400 (EDT) | From | "Richard B. Johnson" <> | Subject | Re: BogoMips |
| |
On Wed, 24 Jun 1998, Gordon Oliver wrote:
> ... Richard B. Johnson said ... > >This subject has come up so may times I thought I would take a look > >at it. Most know that BogoMips stands for "bogus" and is just a number > >obtained to calibrate short delay-loops. However, some are disturbed > >that recompiling the kernel will sometimes result in a different > >number. Even though this number means nothing to the end-user, to > >help prevent it from changing, I have provided this patch. It should > >patch cleanly on most recent kernels. > <snip> > > you might try reading the archives as well. This has been hashed over > several times. Different processors have different effects, and in the > end it was determined that pretty much any choice of offset/alignment caused > "bogomips jitter" on some processor. E.g. a pentium will slow down if > the loop is in something like the first 16 bytes of a page (I don't > remember the exact number or if it was first/last...). Thats why > __delay got moved to a routine, instead of being inline. > -gordo
I have been reading about this for well over a year. The __delay() routine got moved to a fixed place rather than some inline code. The problem is the fixed place did not align the actual delay loop to some place stable. Therefore, with a different sized kernel, or with someone playing with -malign on the gcc command-line, you get a different BogoMips reading. The patch fixes this.
Cheers, Dick Johnson ***** FILE SYSTEM MODIFIED ***** Penguin : Linux version 2.1.105 on an i586 machine (66.15 BogoMips). Warning : It's hard to remain at the trailing edge of technology.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |