Jochen Voss <[EMAIL PROTECTED]> wrote: > On Sat, Mar 19, 2005 at 10:29:28PM +0000, Joerg Sommer wrote: >> Benjamin Herrenschmidt <[EMAIL PROTECTED]> wrote: >> > On Mon, 2005-03-14 at 16:01 +0000, Jochen Voss wrote: >> > >> >> From other people's posts I conclude that bogomips should approximately >> >> equal clock? >> > >> > It depends on the CPU, but on a 7447A, yes. >>=20 >> Do you know why it is for i386 nearly the double of cpu clock? Or reverse, >> why is it only equal the clock on ppc? > > I am no expert on this, but the delay loop in question is > > _GLOBAL(__delay) > cmpwi 0,r3,0 > mtctr r3 > beqlr > 1: bdnz 1b > blr > > in file "arch/ppc/kernel/misc.S". I do not yet speak powerpc > assembler, but I read bdnz as decrease (register r3?) and branch > backwards to label 1 if the result is not zero. Probably a 7447A CPU
Yes, this is it. > can do this within 1 CPU clock cycle? I didn't check the loop for > i386 but maybe it consists of two instructions or of one which takes > two CPU clock cycles? I found a statement to this in /usr/share/doc/HOWTO/en-txt/BogoMips.gz. What did change in kernel version 2.2.14 is the CPU state setting just before the BogoMips calculation. This affects the BogoMips rating for all Intel and AMD Pentium variations, resulting in approximately 2*clock, where these were not 2*clock. But I found there a PPC should have the double too: PowerPC G4/500 992.87 K. Gustilo <[EMAIL PROTECTED]> PowerPC G4/500 996.64 Heinz Nabielek <[EMAIL PROTECTED]> PowerPC G4/500 999.42 Stephen D. Scotti <[EMAIL PROTECTED]> PowerPC G4/999 1992.29 Paula <[EMAIL PROTECTED]> Are these wrong values or did something changed in the calculation? Jörg. -- Mathematiker beim Kuchenessen (aus dem wahren Leben): J: Du überlegst wohl, wie du das Stück am optimalsten teilst? K: Ja, ich wende gerade den Simplex-Algorithmus darauf an. C: Schau mal, da hast du schon vier Ecken. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
