Factor5 source code thread

[ET_]

Here is my Factor5 source code.

It's far from optimized, and I'm studying a couple cases where it segfaulted.

Apart from that it is reliable as for the results it gives, and is multi-thread ready.

I'd appreciate some tests on quad-core environments.

Luigi

P.S. Would it be possible to make this thread sticky?

[em99010pepe]

Luigi,

Got your email. You need to rename factor5.exe to factor5.pdf to cheat gmail control.

Carlos

[em99010pepe]

Tested on my quad-core and it doesn't work with 4 threads only with 2. In the latter case the processor is always below 48%.

Of course it works running 4 clients with one thread turned on with setting the affinity to each core.

Carlos

[ET_]

Originally Posted by em99010pepe

Tested on my quad-core and it doesn't work with 4 threads only with 2. In the latter case the processor is always below 48%.

Of course it works running 4 clients with one thread turned on with setting the affinity to each core.

Carlos

Thank you Carlos.

It seems that I have to find out a quad-core to make some testing work.
"Doesn't work" means that only two threads keep on working, or that the whole program burst into pieces?

Strange enough, it works fine on dual-core CPUs... I'll investigate.

Luigi.

[em99010pepe]

Originally Posted by ET_

"Doesn't work" means that only two threads keep on working, or that the whole program burst into pieces?

Two threads keep working but not pushing the cores has it should be.
When starting with 4 the CPU goes up to 99% and then reduces to 48%...
The quad-core isn't online so I can't give you remote access to it...maybe this week....

Carlos

[lfm]

Originally Posted by em99010pepe

Two threads keep working but not pushing the cores has it should be.
When starting with 4 the CPU goes up to 99% and then reduces to 48%...
The quad-core isn't online so I can't give you remote access to it...maybe this week....

Carlos

Look like it is working well on my Intel quad Q9450 2.66ghz (Linux 64 bit OS). Near 4 x 100% most of the time, occasionally it looks like the threads exit one at a time and may leave a core idle for a few seconds but then they start up new threads again just fine. Running:

time ./factor5 2147483647 1 68 4

Factor5 v. 5.01 - December 27th, 2007 - AMD64/Linux64 compile - GMP 4.2.2


finds the same 3 factors and takes ...

real 35m36.482s
user 126m16.486s
sys 0m0.480s

with some background processing also going on.

[ET_]

Originally Posted by lfm

Look like it is working well on my Intel quad Q9450 2.66ghz (Linux 64 bit OS). Near 4 x 100% most of the time, occasionally it looks like the threads exit one at a time and may leave a core idle for a few seconds but then they start up new threads again just fine. Running:

time ./factor5 2147483647 1 68 4

Factor5 v. 5.01 - December 27th, 2007 - AMD64/Linux64 compile - GMP 4.2.2


finds the same 3 factors and takes ...

real 35m36.482s
user 126m16.486s
sys 0m0.480s

with some background processing also going on.

Nice to hear it

You are right, threads may stop with a few seconds of time difference because of the implementation. BTW, 88.9% of efficiency sounds acceptable

Luigi

[em99010pepe]

I tried once again (windows version, Q6600@2.9GHz) and I still don't get 100%, only 40%.

Edit:

Running factor5.exe 2147483647 1 68 1, I get 25% CPU usage
Running factor5.exe 2147483647 1 68 2, I get 47% CPU usage
Running factor5.exe 2147483647 1 68 3, doesn't work
Running factor5.exe 2147483647 1 68 4, I get (28-40)% CPU usage

Carlos

[ET_]

Originally Posted by em99010pepe

I tried once again (windows version) and I still don't get 100%, only 40%.

Carlos

Very strange...
Maybe something on the Cygwin environment or the pthreads lib doesn't work correctly. Anyway, it does for 2 cores...

One possibility is to start your machine with some Linux CD distro and launch the Linux version to see if the problem goes away.

If it does, I will delve into the Cygwin environment, if not, it could be some BIOS settings that prevent use of full 4-core power and memory addressing.

Luigi

[em99010pepe]

Originally Posted by ET_

If it does, I will delve into the Cygwin environment, if not, it could be some BIOS settings that prevent use of full 4-core power and memory addressing.

Luigi

dnet works well, it detects the 4 cores, it runs at 100 %.

[ET_]

Originally Posted by em99010pepe

dnet works well, it detects the 4 cores, it runs at 100 %.

Thank you Carlos, it's "my" fault then I will see what I can do to give full power to Windows users, though it will be hard without a quad at home.

If I get Factor5 compile under VisualStudio, it will be easier to address the thread system issue... and to get a working GUI already on the to-do list!

Luigi

P.S. There have been thread issues with Cygwin DLL vers. 1.5.10. Did you try to get last DLL from Cygwin site? Meanwhile I will recheck my development environment.