Re: loop optimizations, mark 12[a]

Re: loop optimizations, mark 12[a]

[Richard Henderson]

On Mon, Jul 06, 1998 at 06:22:57PM -0600, Jeffrey A Law wrote:
> Odd as I thought I'd done the test correctly -- ie test without the
> patch, apply the patch and retest.

Know that there is an enormous variance in the results from one run
to the next with identical binaries.  This may be attributable to 
cache lossage from the Linux kernel not doing page colouring.

What we need in order to verify this hypothesis is for someone with
a PC164 (or close equivalent -- alphaserver 1000) running Digital Unix
to run 145.fpppp and report the results along with the variance across
three to five runs.

Volunteers?


r~

Re: loop optimizations, mark 12[a]

[Jeffrey A Law]

  In message < 19980706174246.D2419@dot.cygnus.com >you write:
  > On Mon, Jul 06, 1998 at 06:22:57PM -0600, Jeffrey A Law wrote:
  > > Odd as I thought I'd done the test correctly -- ie test without the
  > > patch, apply the patch and retest.
  > 
  > Know that there is an enormous variance in the results from one run
  > to the next with identical binaries.  This may be attributable to 
  > cache lossage from the Linux kernel not doing page colouring.
I haven't seen that on my x86 boxes.  Results from one run to the next
are usually pretty similar, except for trivially small benchmarks.

jeff