lkml.org 
[lkml]   [1998]   [Feb]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: BogoMIPS?
    On Wed, 4 Feb 1998, Chuck Carson wrote:

    > 1) What is a BogoMIP?

    It's the value determined to calibrate a delay() loop.
    More specifically:
    On the iX86 architecture, the following loop is executed for short delays:
    1: decl %eax
    jns 1b
    ret
    whichs equivalents the instruction "while (eax) eax--;" in C (eax is a
    32bit register which is set to a calculated value before the loop is
    called).
    The BogoMIPS value indicates the number of Megaloops executed per second.
    On some machines (6x86 for example), the CPU takes exactly one cycle to
    execute the loop, so you find a value matching the frequency. The pentium
    takes 2.5 cycles. I don't know about the MMX.

    > After numerous kernel recompilings I have noticed that this number changes
    > constantly and in many times will be much more or much less than the
    > previous kernel. (I believe I have seen this as high as 720.xx)

    On Pentium machines, the alignment of the branch target (i.e.the 1:) has
    some influence on the execution speed. That's why the loop is no longer
    inlined in recent kernels, so it's possible to have the loop always
    execute with the same speed (and to control alignment?). This change was
    done in 2.0.31. As the assymbly code in arch/i386/lib/delay.S does not
    specifiy any alignment, there is still a chance to find different values
    after recompiling the kernel, I believe.

    > 3) /proc/cpuinfo shows twp cpu's listed but the system does not behave as if
    > it was scheduling
    > both cpu's (when compiling software mainly). Before, under 4.2 I compiled
    > the same code on
    > the same machine _much_ faster. How can I verify both cpu's are being
    > scheduled without
    > relying on some text file created at boot up?

    Be sure, that your system load is low. Compile some software package
    twice:
    1) make clean; time make
    2) make clean; time make -j2

    The second one should be much faster than the first one (in theory: double
    as fast) if both CPUs are used.

    > btw: This is a Tyan board with 2 matched Intel 200mmx chips. Any known
    > problems with these boards?

    No idea.

    Kurt Garloff, Dortmund
    <K.Garloff@ping.de>
    PGP key on http://www.student.physik.uni-dortmund.de/homepages/garloff


    \
     
     \ /
      Last update: 2005-03-22 13:41    [W:4.181 / U:0.012 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site