mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   GPU Computing (https://www.mersenneforum.org/forumdisplay.php?f=92)
-   -   LL with OpenCL (https://www.mersenneforum.org/showthread.php?t=18297)

kracker 2013-09-10 18:15

[QUOTE=TeknoHog;352625]I have followed this project with great interest, and I look forward to a usable release. So far, it has been a little hard to piece together all the information necessary for compilation. For example, when trying to compile 0.59, I was left wondering what openclsdkdefs.mk is supposed to be, and where I should find SDKApplication.hpp. I use Linux and I'm not exactly a beginner with these things, but for some reason I feel completely lost here, as if I was missing some essential package (besides AMD APP SDK and clAmdFft, of course).

One reason why I mention this now is that I recently released an automatic work assignment/submission [URL="https://github.com/teknohog/primetools"]tool[/URL] for mfakto, and I'd like to extend it for LL testing. Of course, I would try CudaLucas if I had any Nvidia hardware.[/QUOTE]

[URL="http://mersenneforum.org/showpost.php?p=346635&postcount=43"]This[/URL] may help.
Also, openclsdkdefs.mk and SDKApplication.hpp is located in the AMD APP SDK. Although, you may want to wait till the switch to clLucas/clFFT is done. Assuming that you have installed AMD APP SDK in the default location on linux, it should be at /opt/AMDAPP/samples/opencl/SDKUtil/include

msft 2013-09-10 19:14

1 Attachment(s)
Hi,
New Version.
[code]
msft@Fujic:/opt/AMDAPP/samples/opencl/cpp_cl/app$ tar -xvf clLucas.1.0.tar.bz2
clLucas.1.0/
clLucas.1.0/Makefile
clLucas.1.0/timeval.c
clLucas.1.0/run.sh
clLucas.1.0/clLucas.cpp
clLucas.1.0/Kernels.cl
clLucas.1.0/Kernels.hpp
clLucas.1.0/Kernels.cpp
msft@Fujic:/opt/AMDAPP/samples/opencl/cpp_cl/app/clLucas.1.0$ make
mkdir -p depends/x86_64
perl ../../../../../make/fastdep.pl -I. -I../../../../../include -I../../../../../samples/opencl/SDKUtil/include -I../../../../../samples/bolt/BoltUtil/include -I../../../../../samples/C++Amp/AmpUtil/include --obj-suffix='.o' --obj-prefix='build/debug/x86_64//' clLucas.cpp > depends/x86_64/clLucas.depend
mkdir -p build/debug/x86_64/
Building build/debug/x86_64//clLucas.o
g++ -Wpointer-arith -Wfloat-equal -O3 -Wno-conversion-null -Wno-write-strings -Wno-pointer-arith -I /opt/AMDAPP/include/ -I /opt/clFFT-2.0/include -I /opt/clFFT-2.0/src/include -g3 -ffor-scope -I ../../../../../samples/opencl/SDKUtil/include -I ../../../../../samples/bolt/BoltUtil/include -I ../../../../../samples/C++Amp/AmpUtil/include -I "/opt/AMDAPP/include" -I ../../../../../include -o build/debug/x86_64//clLucas.o -c clLucas.cpp
Building build/debug/x86_64/clLucas
g++ -o build/debug/x86_64/clLucas build/debug/x86_64//clLucas.o -lpthread -ldl -L/usr/X11R6/lib -O3 /opt/clFFT-2.0/library/libclFFT.so -lOpenCL -lSDKUtil -lOpenCL -L../../../../../lib/x86_64 -L../../../../../TempSDKUtil/lib/x86_64 -L"/opt/AMDAPP/lib/x86_64"
install -D build/debug/x86_64/clLucas ../../../../../samples/opencl/bin/x86_64/clLucas
for f in Kernels.cl; do \
install -D $f ../../../../../samples/opencl/bin/x86_64/$f; \
done
msft@Fujic:/opt/AMDAPP/samples/opencl/cpp_cl/app/clLucas.1.0$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/clFFT-2.0/library/
msft@Fujic:/opt/AMDAPP/samples/opencl/cpp_cl/app/clLucas.1.0$ ../../../../../samples/opencl/bin/x86_64/clLucas 1398269
Platform :Advanced Micro Devices, Inc.
Device 0 : Capeverde


start M1398269 fft length = 73728
Iteration 10000 M( 1398269 )C, 0xa4a6d2f0e34629db, n = 73728, clLucas v1.00 err = 0.09375 (0:18 real, 1.7608 ms/iter, ETA 40:29)
Iteration 20000 M( 1398269 )C, 0x4483211b6f1ce250, n = 73728, clLucas v1.00 err = 0.09375 (0:16 real, 1.6361 ms/iter, ETA 37:21)
Iteration 30000 M( 1398269 )C, 0x32e7a8b19576df92, n = 73728, clLucas v1.00 err = 0.09375 (0:16 real, 1.6365 ms/iter, ETA 37:05)
[/code]

kracker 2013-09-10 19:40

1 Attachment(s)
A few notes.
Does -aggressive work? it seems turning that on uses a lot of cpu.
Also, might want to consider switching from clamdfft to clfft, (clfft.h)

:smile:

msft 2013-09-10 20:59

[QUOTE=kracker;352643]A few notes.
Does -aggressive work? it seems turning that on uses a lot of cpu.
Also, might want to consider switching from clamdfft to clfft, (clfft.h)

:smile:[/QUOTE]

Please check large exponent,like 36666666.
clfft.h is right.My mistake. :smile:

kracker 2013-09-10 21:42

Ok, so if anyone has a 58xx+,68xx+,77xx+ and want to test out a few things, PM me. :smile: (Windows x64, 32 bit if needed)

TeknoHog 2013-09-11 14:18

[QUOTE=kracker;352629][URL="http://mersenneforum.org/showpost.php?p=346635&postcount=43"]This[/URL] may help.
Also, openclsdkdefs.mk and SDKApplication.hpp is located in the AMD APP SDK. Although, you may want to wait till the switch to clLucas/clFFT is done. Assuming that you have installed AMD APP SDK in the default location on linux, it should be at /opt/AMDAPP/samples/opencl/SDKUtil/include[/QUOTE]

Thanks! It seems my APP SDK 2.5 is too old; Gentoo doesn't have packages for the newest versions, and I have some software where 2.5 is recommended - I guess I'll have to install by hand.

VictordeHolland 2013-09-11 18:27

Kracker,
What are the driver and SDK requirements for the test you send?
Catalyst 13.4 gives me a missing libgcc_s_seh-1.dll error, so I'm downloading Catalyst 13.6Beta2 and APP SDK v2.8.1.0 now and will try again :).

kracker 2013-09-11 18:45

[QUOTE=VictordeHolland;352754]Kracker,
What are the driver and SDK requirements for the test you send?
Catalyst 13.4 gives me a missing libgcc_s_seh-1.dll error, so I'm downloading Catalyst 13.6Beta2 and APP SDK v2.8.1.0 now and will try again :).[/QUOTE]

:facepalm: Sorry, forgot to link it statically! you don't need the appsdk and catalyst, I'll give you the dll's...

kracker 2013-09-11 18:48

Here they are. also to @LaurV. Sorry again.
[URL]https://www.dropbox.com/s/f24mrkl7aid2haa/clLucas_dll.zip[/URL]

VictordeHolland 2013-09-11 19:41

1 Attachment(s)
Thanks, now it's working!
Windows 7 x64
Catalyst 13.4
AMD HD7950 @800MHz (downclocked it to reference clocks)

Nice GPU load on 36,666,666!
See attached screenshot

VictordeHolland 2013-09-11 19:44

clLucas_x64_test-1.0 M1,398,269 with HD7950
 
1 Attachment(s)
And also the test on the much smaller expo (lower GPU load).


All times are UTC. The time now is 13:00.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.