LL with OpenCL - Page 5

sanaris · 2013-08-01, 17:36

It does not see any of my devices...
I should get at least CPU resources, and my Nvidia card has libOpenCL but none of that is working.

Code:

start M6478992 fft length = 393216
Error: Context::Context() failed. Error code : CL_DEVICE_NOT_FOUND
Location : MatrixMulDouble.hpp:354
OPENCL_V< CLFFT_INVALID_CONTEXT > (3191): Getting device array size ( ::clGetContextInfo() )
OPENCL_V< CLFFT_INVALID_CONTEXT > (115): SetEnvelope failed
Failed to create c2c plan. status=-34; failed step=1
Segmentation fault (core dumped)

I did NOT installed Catalyst, because it would obviously overwrite my current OpenCL.so (until I get some switches between them).

msft · 2013-08-02, 04:53

Hi,

Quote:

Originally Posted by sanaris

It does not see any of my devices...
I should get at least CPU resources, and my Nvidia card has libOpenCL but none of that is working.

clAmdFfft not support Nvidia card.

LaurV · 2013-08-02, 06:23

@sanaris:
grrrr... if you have a cuda-enabled nvidia card, then forget about doing LL with openCL on that card, and use cudaLucas. It is a mature product, works very nice, and objectively, cuda is 3 times faster then openCL for nvidia cards (they "emulate" openCL somehow, it is not as natively supported cuda stuff).

kracker · 2013-08-02, 16:51

Quote:

Originally Posted by sanaris

It does not see any of my devices...
I should get at least CPU resources, and my Nvidia card has libOpenCL but none of that is working.

Code:

start M6478992 fft length = 393216
Error: Context::Context() failed. Error code : CL_DEVICE_NOT_FOUND
Location : MatrixMulDouble.hpp:354
OPENCL_V< CLFFT_INVALID_CONTEXT > (3191): Getting device array size ( ::clGetContextInfo() )
OPENCL_V< CLFFT_INVALID_CONTEXT > (115): SetEnvelope failed
Failed to create c2c plan. status=-34; failed step=1
Segmentation fault (core dumped)

I did NOT installed Catalyst, because it would obviously overwrite my current OpenCL.so (until I get some switches between them).

http://sourceforge.net/projects/cudalucas/files/

Also, on another note I think found my problem...

msft · 2013-08-05, 13:23

Hi.
New Version.

kracker · 2013-08-06, 19:17

Hmm. Compiling works now, although I get error on launch "Error with clAmdFftSetPlanPrecision". I tried removing that just for curiosity and it works... actually it doesn't but the gpu is doing something.. Or maybe not.

Code:

~/0.44
$ ./a.exe 1398269

start M1398269 fft length = 73728
Platform :Advanced Micro Devices, Inc.
Device 0 : Capeverde
Device 1 : BeaverCreek

Build Options are : -D KHR_DP_EXTENSION
Iteration 10000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:07 real, 0.7430 ms/iter, ETA 17:05)
Iteration 20000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:08 real, 0.7640 ms/iter, ETA 17:26)
Iteration 30000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:08 real, 0.7800 ms/iter, ETA 17:40)
Iteration 40000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:08 real, 0.8620 ms/iter, ETA 19:23)
Iteration 50000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:09 real, 0.8570 ms/iter, ETA 19:08)
Iteration 60000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:08 real, 0.8390 ms/iter, ETA 18:35)
Iteration 70000 M( 1398269 )C, 0x0000000000000000, n = 73728, CUDALucas v1.66 err = 0.0003744 (0:08 real, 0.7270 ms/iter, ETA 15:59)

msft · 2013-08-06, 23:27

Quote:

Originally Posted by kracker

actually it doesn't but the gpu is doing something.. Or maybe not.

Hi,
You need dump value after all fft or small fft test program.

kracker · 2013-08-07, 00:45

Quote:

Originally Posted by msft

Hi,
You need dump value after all fft or small fft test program.

Sorry, what do you mean by it? Something like a memory dump?

msft · 2013-08-07, 01:14

Quote:

Originally Posted by kracker

Sorry, what do you mean by it? Something like a memory dump?

The value of the array x.
May not sense.

LaurV · 2013-08-07, 10:52

Quote:

Originally Posted by kracker

Sorry, what do you mean by it? Something like a memory dump?

He says that the FFT length selected for that exponent is WAY too big. You need either a lower FFT, or a higher exponent.

BTW, is someone going to build a windows version of this?

kracker · 2013-08-07, 15:24

I have a few "experimental" windows exe's here. Unfortunately, they don't quite work yet...

@msft: strange, setting clAmdFftSetPlanPrecision to CLFFT_SINGLE works but on DOUBLE it crashes with "Error with clAmdFftSetPlanPrecision"... Hmm

2013-08-02, 06:23	#47
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 3×23×149 Posts	@sanaris: grrrr... if you have a cuda-enabled nvidia card, then forget about doing LL with openCL on that card, and use cudaLucas. It is a mature product, works very nice, and objectively, cuda is 3 times faster then openCL for nvidia cards (they "emulate" openCL somehow, it is not as natively supported cuda stuff). Last fiddled with by LaurV on 2013-08-02 at 06:26 Reason: @sanaris (crosspost)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
mfakto: an OpenCL program for Mersenne prefactoring	Bdot	GPU Computing	1719	2023-01-16 15:51
Can't get OpenCL to work on HD7950 Ubuntu 14.04.5 LTS	VictordeHolland	Linux	4	2018-04-11 13:44
OpenCL accellerated lattice siever	pstach	Factoring	1	2014-05-23 01:03
OpenCL for FPGAs	TObject	GPU Computing	2	2013-10-12 21:09
AMD's Graphics Core Next- a reason to accelerate towards OpenCL?	Belteshazzar	GPU Computing	19	2012-03-07 18:58

2013-08-07, 15:24	#55
kracker "Mr. Meeseeks" Jan 2012 California, USA 2⁷×17 Posts	I have a few "experimental" windows exe's here. Unfortunately, they don't quite work yet... @msft: strange, setting clAmdFftSetPlanPrecision to CLFFT_SINGLE works but on DOUBLE it crashes with "Error with clAmdFftSetPlanPrecision"... Hmm