Msieve 1.53 feedback

xilman · 2014-07-23, 17:09

Hitting problems with polynomial finding on a Tesla C1060 (cc=1.3) with msieve 1.53.SVN 967. The card has 2GB of memory and appears to work reliably on other tasks.

The search crashes after a period somewhere between seconds and days with the message "error (line 734): unexpected error". It is not clear which line 734 is appropriate but in stage1_sieve_gpu.c it reads "CUDA_TRY(cuEventSynchronize(t->end_event))"

This is a 64-bit Gentoo machine with driver 337.25. All other installed software is current as of a couple of days ago.

Another system with a GTX 460 (cc=2.1) but otherwise identical appears to work reliably as long as it is given the gpu_mem_mb=384 parameter because it only has 512M memory and msieve seems to want to grab all of it if given the chance.

A more elderly 64-bit Windoze version has run for many weeks (over several sessions of course) without any issues whatsoever.

Paul

wombatman · 2014-08-04, 14:25

Having a small issue as well--compiling in VS2012 without GMP-ECM and with CUDA. I get an unresolved external symbol mul_mod_64. The only reference to it I can find is from an asm file. What am I forgetting?

Compiling with MPIR 2.7.0 and CUDA 6.0 (CC 2.0).

Thanks!

Brian Gladman · 2014-08-04, 21:21

Quote:

Originally Posted by wombatman

Having a small issue as well--compiling in VS2012 without GMP-ECM and with CUDA. I get an unresolved external symbol mul_mod_64. The only reference to it I can find is from an asm file. What am I forgetting?

Compiling with MPIR 2.7.0 and CUDA 6.0 (CC 2.0).

Thanks!

This symbol comes from the file x64_support_masm.asm and is a part of the Windows x64 build. It is included in the Visual Studio as a part of the gnfs/poly sub-project. It requires MASM, the Microsoft assembler.

Brian

wombatman · 2014-08-04, 21:45

I'm leaving GMP-ECM out completely by removing the .props file. The mul_mod_64 thing (per a quick and dirty googling) used to be able to be found in an asm: https://github.com/radii/msieve/blob...64_support.asm

I'm guessing that's not where it is anymore? The error is from gnfs.lib (relation.obj), but I can't find mul_mod_64 in it.

Brian Gladman · 2014-08-04, 22:22

Quote:

Originally Posted by wombatman

I'm leaving GMP-ECM out completely by removing the .props file. The mul_mod_64 thing (per a quick and dirty googling) used to be able to be found in an asm: https://github.com/radii/msieve/blob...64_support.asm

I'm guessing that's not where it is anymore? The error is from gnfs.lib (relation.obj), but I can't find mul_mod_64 in it.

Sorry - I updated my earlier comment, which was wrong.

wombatman · 2014-08-04, 23:27

Thanks. That did get it to compile without issue. On running it, however, I get the following (running a quick test to make I could do the poly search):

Code:

[c:\dorkdom\svn_checkout\msieve-code\trunk\b40c\b40c\radix_sort\../radix_sort/tex_ref.cuh, 71] cudaBindTexture failed (CUDA error 20: invalid channel descriptor)
[C:/Dorkdom/SVN_Checkout/msieve-code/trunk/b40c/sort_engine.cu, 95] sort engine: (CUDA error 20: invalid channel descriptor)

Compilation is with CUDA 6.0.

wombatman · 2014-08-05, 01:08

How odd. Rebuilding with CUDA 5.5 and it works fine now, at least for poly selection. I know MSieve isn't really being actively worked on, but this seems to be a bug of some kind?

wombatman · 2014-08-05, 04:47

This is me with a CUDA-enabled 64-bit version of MSieve for Windows that can handle the large data sets. Thanks again for your help!

jasonp · 2014-08-06, 00:25

I don't have 6.0 right now; the 5.0 toolkit required some makefile fixes (its windows environment variables did not have an ending slash like previous versions did) but otherwise it worked out of the box.

Wombatman, am I correct that your post above means you got it working?

wombatman · 2014-08-06, 00:49

I did indeed (they were celebratory emoticons). Only with CUDA 5.5 though. Compiling with 6.0 gave the error I posted previously. If I can provide any more information to try and nail down what causes it, please let me know.

BudgieJane · 2014-08-06, 09:52

Quote:

Originally Posted by wombatman

This is me with a CUDA-enabled 64-bit version of MSieve for Windows that can handle the large data sets. Thanks again for your help!

Are you going to make it available generally, so that we can all enjoy the fruits of your labours?

2014-07-23, 17:09	#1
xilman Bamboozled! "𒉺𒌌𒇷𒆷𒀭" May 2003 Down not across 10,753 Posts	Msieve 1.53 feedback Hitting problems with polynomial finding on a Tesla C1060 (cc=1.3) with msieve 1.53.SVN 967. The card has 2GB of memory and appears to work reliably on other tasks. The search crashes after a period somewhere between seconds and days with the message "error (line 734): unexpected error". It is not clear which line 734 is appropriate but in stage1_sieve_gpu.c it reads "CUDA_TRY(cuEventSynchronize(t->end_event))" This is a 64-bit Gentoo machine with driver 337.25. All other installed software is current as of a couple of days ago. Another system with a GTX 460 (cc=2.1) but otherwise identical appears to work reliably as long as it is given the gpu_mem_mb=384 parameter because it only has 512M memory and msieve seems to want to grab all of it if given the chance. A more elderly 64-bit Windoze version has run for many weeks (over several sessions of course) without any issues whatsoever. Paul

2014-08-04, 21:45	#4
wombatman I moo ablest echo power! May 2013 11011101001₂ Posts	I'm leaving GMP-ECM out completely by removing the .props file. The mul_mod_64 thing (per a quick and dirty googling) used to be able to be found in an asm: https://github.com/radii/msieve/blob...64_support.asm I'm guessing that's not where it is anymore? The error is from gnfs.lib (relation.obj), but I can't find mul_mod_64 in it. Last fiddled with by wombatman on 2014-08-04 at 21:46

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Msieve 1.50 feedback	firejuggler	Msieve	99	2013-02-17 11:53
Msieve v1.48 feedback	Jeff Gilchrist	Msieve	48	2011-06-10 18:18
Msieve 1.43 feedback	Jeff Gilchrist	Msieve	47	2009-11-24 15:53
Msieve 1.42 feedback	Andi47	Msieve	167	2009-10-18 19:37
Msieve 1.41 Feedback	Batalov	Msieve	130	2009-06-09 16:01

2014-08-04, 14:25	#2
wombatman I moo ablest echo power! May 2013 3351₈ Posts	Having a small issue as well--compiling in VS2012 without GMP-ECM and with CUDA. I get an unresolved external symbol mul_mod_64. The only reference to it I can find is from an asm file. What am I forgetting? Compiling with MPIR 2.7.0 and CUDA 6.0 (CC 2.0). Thanks!

2014-08-05, 01:08	#7
wombatman I moo ablest echo power! May 2013 29×61 Posts	How odd. Rebuilding with CUDA 5.5 and it works fine now, at least for poly selection. I know MSieve isn't really being actively worked on, but this seems to be a bug of some kind?

2014-08-05, 04:47	#8
wombatman I moo ablest echo power! May 2013 29×61 Posts	This is me with a CUDA-enabled 64-bit version of MSieve for Windows that can handle the large data sets. Thanks again for your help!

2014-08-06, 00:25	#9
jasonp Tribal Bullet Oct 2004 DD5₁₆ Posts	I don't have 6.0 right now; the 5.0 toolkit required some makefile fixes (its windows environment variables did not have an ending slash like previous versions did) but otherwise it worked out of the box. Wombatman, am I correct that your post above means you got it working?

2014-08-06, 00:49	#10
wombatman I moo ablest echo power! May 2013 29·61 Posts	I did indeed (they were celebratory emoticons). Only with CUDA 5.5 though. Compiling with 6.0 gave the error I posted previously. If I can provide any more information to try and nail down what causes it, please let me know.