mersenneforum.org Sr2sieve on PPC/Linux
 Register FAQ Search Today's Posts Mark Forums Read

 2006-12-11, 17:59 #56 BlisteringSheep     Oct 2006 On a Suzuki Boulevard C90 3668 Posts To be fair, I do not have the same gcc versions. On the blades, I'm using RedHat with their version of gcc 4.0.0. On the PowerMac, I'm using a hand-built gcc 4.1.1. But this is the first time where I've seen something cause an improvement on one platform without also improving the other.
2006-12-12, 01:53   #57
geoff

Mar 2003
New Zealand

13·89 Posts

Quote:
 Originally Posted by BlisteringSheep To be fair, I do not have the same gcc versions. On the blades, I'm using RedHat with their version of gcc 4.0.0. On the PowerMac, I'm using a hand-built gcc 4.1.1. But this is the first time where I've seen something cause an improvement on one platform without also improving the other.
It is quite possible that the different GCC versions are the cause of the speed difference, but that would really need to be tested by using different compilers on the same machine.

I'll leave the code as it is for now, I can add a condition to only use the modification for certain GCC versions if that turns out to be the cause.

2006-12-12, 04:34   #58
BlisteringSheep

Oct 2006
On a Suzuki Boulevard C90

3668 Posts

Quote:
 Originally Posted by geoff It is quite possible that the different GCC versions are the cause of the speed difference, but that would really need to be tested by using different compilers on the same machine. I'll leave the code as it is for now, I can add a condition to only use the modification for certain GCC versions if that turns out to be the cause.
Yes, I would certainly agree, though I wouldn't have expected such a dramatic difference between 4.0 & 4.1 (whereas it's a clear-cut win in general going from gcc3 to gcc4). I should get a chance in the next couple of days to build gcc 4.1.1 on the 970MPs and will report back any results. I also have not had a chance to do any of the other register linkage experiments.

One other thing: for Linux when using the inline assembly code, can you add
Code:
-Xassembler -mregnames
to the CFLAGS please? This will cause some harmless warnings (gcc: -mregnames: linker input file unused because linking not done) depending on the compiler version, but is necessary.

The only other source change I am making now is BASE from 5 to 2 and HASH_MAX_DENSITY from 0.65 to 0.32, since that showed an improvement on all of the machines.

Last fiddled with by BlisteringSheep on 2006-12-12 at 04:36 Reason: added other source changes

2006-12-18, 02:28   #59
geoff

Mar 2003
New Zealand

13·89 Posts

Quote:
 Originally Posted by BlisteringSheep One other thing: for Linux when using the inline assembly code, can you add Code: -Xassembler -mregnames to the CFLAGS please? This will cause some harmless warnings (gcc: -mregnames: linker input file unused because linking not done) depending on the compiler version, but is necessary.
That is done in version 1.4.10.

I have also made a change to the inline mulmod code: Change the Makefile line with USE_INLINE_MULMOD=1 to USE_INLINE_MULMOD=3 to try it out. It is the same idea as in version 1.4.9 but done using a global variable, the GCC version that didn't like the 1.4.9 modification might like this one better.

(You can set USE_INLINE_MULMOD=2 to use the 1.4.9 modification).

2006-12-26, 17:45   #60
BlisteringSheep

Oct 2006
On a Suzuki Boulevard C90

3668 Posts

Quote:
 Originally Posted by geoff That is done in version 1.4.10. I have also made a change to the inline mulmod code: Change the Makefile line with USE_INLINE_MULMOD=1 to USE_INLINE_MULMOD=3 to try it out. It is the same idea as in version 1.4.9 but done using a global variable, the GCC version that didn't like the 1.4.9 modification might like this one better. (You can set USE_INLINE_MULMOD=2 to use the 1.4.9 modification).
Geoff, I just wanted to let you know that I have been doing a _lot_ of testing with this. I hope to have the results down to a manageable summary in the next few days.

 2006-12-26, 20:43 #61 jasong     "Jason Goatcher" Mar 2005 350710 Posts How does sr2sieve Linux compare to proth sieve linux? Specifically, in terms of Seventeen or Bust sieving.
2006-12-26, 20:48   #62
BlisteringSheep

Oct 2006
On a Suzuki Boulevard C90

3668 Posts

Quote:
 Originally Posted by jasong How does sr2sieve Linux compare to proth sieve linux? Specifically, in terms of Seventeen or Bust sieving.
Don't I wish I knew. :) I'm running Linux on PowerPC 970s, so there is no proth_sieve for me.

2006-12-26, 21:02   #63
axn

Jun 2003

537710 Posts

Quote:
 Originally Posted by jasong How does sr2sieve Linux compare to proth sieve linux? Specifically, in terms of Seventeen or Bust sieving.
You could always benchmark it yourselves

Anyway, on x86, I think sr2sieve is about 35% slower than proth_sieve for sob. I am quoting this from memory, so I could be wrong. If you can do the benchmark and post it here, that'd be super -- especially with a 64 bit build of sr2sieve.

2006-12-26, 21:10   #64
BlisteringSheep

Oct 2006
On a Suzuki Boulevard C90

2·3·41 Posts

Quote:
 Originally Posted by axn1 Anyway, on x86, I think sr2sieve is about 35% slower than proth_sieve for sob.
That sounds about like what I remember from the couple of x86 sieving tests that I did. I do have an EM64T (3.2 GHz Nocona) that I could do a benchmark on; is there a 64-bit proth_sieve for Linux?

2006-12-26, 23:15   #65
rogue

"Mark"
Apr 2003
Between here and the

11001111110012 Posts

Quote:
 Originally Posted by BlisteringSheep Don't I wish I knew. :) I'm running Linux on PowerPC 970s, so there is no proth_sieve for me.
There is a version of proth_sieve for PPC. Contact Greenbank.

2006-12-26, 23:48   #66
BlisteringSheep

Oct 2006
On a Suzuki Boulevard C90

2·3·41 Posts

Quote:
 Originally Posted by rogue There is a version of proth_sieve for PPC. Contact Greenbank.
I haven't been able to get in touch with him lately. The last time I did, the PPC version was OS/X only (I need Linux).

 Similar Threads Thread Thread Starter Forum Replies Last Post rogue Software 304 2021-11-06 13:51 pepi37 Software 5 2013-08-09 22:31 SaneMur Information & Answers 2 2011-08-21 22:04 mgpower0 Prime Sierpinski Project 54 2008-07-15 16:50 nuggetprime Riesel Prime Search 40 2007-12-03 06:01

All times are UTC. The time now is 16:39.

Fri Jul 1 16:39:41 UTC 2022 up 78 days, 14:41, 0 users, load averages: 2.17, 2.01, 1.73