1 solaris
2.5.1 usparc
167mhz
?? - SC4
.0 cc
-fast
-Xa
-xO5
3 For the ultra sparc
, SunC
4.0 cc
-fast
-Xa
-xO5
, running
'des_opts'
4 gives a speed of
475,000 des
/s
while 'speed' gives
417,000 des
/s
.
5 I belive the difference is tied up in optimisation that the compiler
6 is able to perform when the code is
'inlined'. For
'speed', the DES
7 routines are being linked from a library
. I
'll record the higher
8 speed since if performance is everything, you can always inline
11 [ 16-Jan-06 - I've been playing with the
12 '-xtarget=ultra -xarch=v8plus -Xa -xO5 -Xa'
13 and while it makes the des_opts numbers much slower
, it makes the
14 actual
'speed' numbers look better which is a realistic version of
15 using the libraries
. ]
18 16 r1 p
475516.90 100.0%
19 16 r2 p
439388.10 92.4%
20 16 c i
427001.40 89.8%
21 16 c p
419516.50 88.2%
22 4 r2 p
409491.70 86.1%
23 4 r1 p
404266.90 85.0%
26 4 r1 i
362742.20 76.3%
27 16 r2 i
331275.50 69.7%
28 16 r1 i
324730.60 68.3%
29 4 r2 i
63535.10 13.4% <-- very very weird
, must be cache problems
.
30 -DDES_UNROLL
-DDES_RISC1
-DDES_PTR