A multiple k/c sieve for Sierpinski/Riesel problems

[masser]

3.4 Ghz Xeon

length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
: 96.104 million mulmods per second.
CMOV: 84.809 million mulmods per second.
SSE2: 90.849 million mulmods per second.

[Cruelty]

C2D E6600 @ 3GHz

length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
    : 171.582 million mulmods per second.
CMOV: 162.850 million mulmods per second.
SSE2: 270.042 million mulmods per second.

A64 3400+ @ 2.4GHz

length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
    : 151.300 million mulmods per second.
CMOV: 136.461 million mulmods per second.
SSE2: 114.695 million mulmods per second.

[ltd]

I have a very strange result on a P4 HT.

The first dataset was created when the machine was runnig idle.

D:\download>mulmod8.exe
length = 1000, iterations = 10000, b = 2, p = 4611686018427387817:
: 80.000 million mulmods per second.
CMOV: 71.111 million mulmods per second.
SSE2: 80.000 million mulmods per second.

The second set was generated with the eulernet client running

D:\download>mulmod8.exe
length = 1000, iterations = 10000, b = 2, p = 4611686018427387817:
: 64.000 million mulmods per second.
CMOV: 45.714 million mulmods per second.
SSE2: 91.429 million mulmods per second.

The third set was created with llr.exe running.

D:\download>mulmod8.exe
length = 1000, iterations = 10000, b = 2, p = 4611686018427387817:
: 49.231 million mulmods per second.
CMOV: 40.000 million mulmods per second.
SSE2: 71.111 million mulmods per second.

Lars

[Cruelty]

Originally Posted by ltd

I have a very strange result on a P4 HT.

Try running mulmod8.exe as suggested by Geoff and see if that helps:

mulmod8 1000 1000000

[ltd]

Did a new test with the extra parameters for mulmod.
Results are closer together now but the tendency stays that mulmod runs
faster if a second process(eulernet) runs in parallel. ( SSE2 Results)

System otherwise idle
D:\download>mulmod8.exe 1000 1000000
length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
: 80.706 million mulmods per second.
CMOV: 65.507 million mulmods per second.
SSE2: 82.157 million mulmods per second.

eulernet running
D:\download>mulmod8.exe 1000 1000000
length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
: 65.776 million mulmods per second.
CMOV: 57.399 million mulmods per second.
SSE2: 88.398 million mulmods per second.

llr.exe running
D:\download>mulmod8.exe 1000 1000000
length = 1000, iterations = 1000000, b = 2, p = 4611686018427387817:
: 49.155 million mulmods per second.
CMOV: 42.638 million mulmods per second.
SSE2: 66.667 million mulmods per second.

[Cruelty]

Strange indeed - I don't know any valid explanation for this...

[Flatlander]

Not some speedstep/power saving thing is it?

[ltd]

Originally Posted by Flatlander

Not some speedstep/power saving thing is it?

Not that i would know. Normaly that machine run llr.exe and eulernet in parallel with idle priority and gives the expected performance for such a
combination all the time.

Lars

[ET_]

Originally Posted by ltd

Not that i would know. Normaly that machine run llr.exe and eulernet in parallel with idle priority and gives the expected performance for such a
combination all the time.

Lars

Does Eulernet make use of SSE2 registers?

Luigi

[ltd]

As far as i know it does not use any SSE2 registers.

Lars

[tnerual]

Originally Posted by ET_

Does Eulernet make use of SSE2 registers?

Luigi

not at all, eulernet is not updated since 2001 and is based on a 1999 code ... there wasn't any SSE at this time i think