46 lines
1.2 KiB
Plaintext
46 lines
1.2 KiB
Plaintext
|
Method:
|
||
|
|
* "pdynload"
|
||
|
|
comes from pdynload/small
|
||
|
|
* "load + ghc"
|
||
|
|
comes from pdynload/null, with lines 13-14
|
||
|
|
uncommented from prog/Main.hs
|
||
|
|
* "dynload"
|
||
|
|
from dynload/simple
|
||
|
|
* "load, no check"
|
||
|
|
from pdynload/null, with lines 13-14 of prog/Main.hs
|
||
|
|
commented out
|
||
|
|
|
||
|
|
For example, to run the "pdynload" test:
|
||
|
|
$ cd pdynload/small
|
||
|
|
$ make
|
||
|
|
$ make check # to prime caches, etc.
|
||
|
|
$ time make check
|
||
|
|
$ time make check
|
||
|
|
$ time make check # run 'time make check' until value converges
|
||
|
|
|
||
|
|
The converged value is entered into the "Raw" timings, and then the
|
||
|
|
scaled timing is calculated for each machine. These scaled values were
|
||
|
|
then averaged over the number of machines, yielding the final
|
||
|
|
"Average" scores -- the average over a number of machines and os.
|
||
|
|
|
||
|
|
Raw timing:
|
||
|
|
pdynload load+ghc dynload load, no check
|
||
|
|
|
||
|
|
0.33 0.25 0.22 0.21 -- P4 2.6 , OpenBSD
|
||
|
|
0.38 0.31 0.29 0.27 -- P4 2.66, Linux
|
||
|
|
0.84 0.77 0.64 0.55 -- Quad P4 2.4, Linux
|
||
|
|
0.76 0.60 0.52 0.50 -- AMD 1.1G, Linux
|
||
|
|
0.95 0.83 0.75 0.72 -- G5 2.0G, Mac OS X
|
||
|
|
-- Quad Itanium 1,Linux
|
||
|
|
|
||
|
|
Scaled:
|
||
|
|
1.57 1.19 1.05 1
|
||
|
|
1.40 1.15 1.07
|
||
|
|
1.52 1.4 1.16
|
||
|
|
1.52 1.2 1.04
|
||
|
|
1.32 1.15 1.04
|
||
|
|
|
||
|
|
Average:
|
||
|
|
=1.46 = 1.218 = 1.07
|
||
|
|
|