LL with OpenCL
 

I try ported LL to OpenCL.

Sometime clAmdFft-1.10.321 have precision problem.
7750:
[QUOTE]
$ sh -x ./run.sh
+ rm *.o a.out
+ g++ -c main.cpp -I /opt/AMDAPP/include/ -I /opt/clAmdFft-1.10.321/include/
+ g++ -c clFFTPlans.cpp -I /opt/AMDAPP/include/ -I /opt/clAmdFft-1.10.321/include/
+ g++ main.o clFFTPlans.o /opt/clAmdFft-1.10.321/lib64/libclAmdFft.Runtime.so -lOpenCL -lfftw3
+ export LD_LIBRARY_PATH=:/opt/clAmdFft-1.10.321/lib64/
+ time ./a.out
Using device: Capeverde
AmdFFT_Z2Z size= 2048 time= 0.080000 msec
Everything went fine!
3.52user 7.27system 0:09.48elapsed 113%CPU (0avgtext+0avgdata 381360maxresident)k
0inputs+1680outputs (0major+38531minor)pagefaults 0swaps
+ diff+ fft_fftw.dat fft_cl.dat
head -n 40
2,16c2,16
< 1 -1.673975300084231e+05 1.663735300084231e+05
< 2 -8.395397960358397e+04 8.292997960358398e+04
< 3 -5.613911373064458e+04 5.511511373064457e+04
< 4 -4.223141898575165e+04 4.120741898575165e+04
< 5 -3.388659268655724e+04 3.286259268655724e+04
< 6 -2.832320060429759e+04 2.729920060429759e+04
< 7 -2.434919949694704e+04 2.332519949694705e+04
< 8 -2.136856774250665e+04 2.034456774250665e+04
< 9 -1.905018221676692e+04 1.802618221676692e+04
< 10 -1.719536904441296e+04 1.617136904441295e+04
< 11 -1.567769939498128e+04 1.465369939498128e+04
< 12 -1.441288738083825e+04 1.338888738083825e+04
< 13 -1.334258123301454e+04 1.231858123301454e+04
< 14 -1.242510111596777e+04 1.140110111596777e+04
< 15 -1.162988181643994e+04 1.060588181643995e+04
---
> 1 -1.673975291656877e+05 1.663735279914048e+05
> 2 -8.395397960358400e+04 8.292997960358398e+04
> 3 -5.613911438886409e+04 5.511511209925302e+04
> 4 -4.223141898575164e+04 4.120741898575164e+04
> 5 -3.388659171768492e+04 3.286259310540232e+04
> 6 -2.832320060429760e+04 2.729920060429759e+04
> 7 -2.434919920591193e+04 2.332519938365545e+04
> 8 -2.136856763176386e+04 2.034456749267493e+04
> 9 -1.905018202758640e+04 1.802618176363859e+04
> 10 -1.719536897123582e+04 1.617136882738273e+04
> 11 -1.567769950455412e+04 1.465369873990205e+04
> 12 -1.441288733405879e+04 1.338888718385923e+04
> 13 -1.334258082835818e+04 1.231858121315468e+04
> 14 -1.242510108971055e+04 1.140110093114974e+04
> 15 -1.162988166206707e+04 1.060588158595162e+04
18c18
< 17 -1.031992388900484e+04 9.295923889004827e+03
---
> 17 -1.031992383364508e+04 9.295923772879578e+03
[/QUOTE]
Compare LINE 3.

I hear the AMD dev forums is a nice place to ask if you get stuck, [URL="http://devgurus.amd.com/welcome"]here[/URL].

[QUOTE=kracker;343516]I hear the AMD dev forums is a nice place to ask if you get stuck, [URL="http://devgurus.amd.com/welcome"]here[/URL].[/QUOTE]
Thank you information.
Problem of the addition order is subtle.
I'll try to investigate a little more.

I just noticed something... The GTX 780 has a DP ratio of 1/24, GCN has a ratio of around 1/4. Not sure if it will mean anything, just something I read. :popcorn:

Mul loop on GPU.
[QUOTE]
$ pwd
/opt/AMDAPP/samples/opencl/cl/0.19/0.19
msft@msft-desktop:/opt/AMDAPP/samp
les/opencl/cl/0.19/0.19$ sh -x ./run.sh
+ export LD_LIBRARY_PATH=:/opt/clAmdFft-1.10.321/lib64/
$ time ./a.out 216091
Platform :Advanced Micro Devices, Inc.
Device 0 : Capeverde

Build Options are : -D KHR_DP_EXTENSION
---
216001 32768
M( 216091 )P, n = 32768, MacLucasFFTW v8.1 Ballester

real 12m43.947s
user 6m19.212s
sys 3m8.580s
[/QUOTE]

All Loop on GPU.:smile:
7750:
[QUOTE]
$ pwd
/opt/AMDAPP/samples/opencl/cl/0.27/0.27
$ sh -x ./run.sh
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/clAmdFft-1.10.321/lib64/
$ time ./a.out 216091
Platform :Advanced Micro Devices, Inc.
Device 0 : Capeverde
----
216001 32768
M( 216091 )P, n = 32768, MacLucasFFTW v8.1 Ballester

real 16m51.839s
user 1m26.733s
sys 3m35.473s
[/QUOTE]
Caution: Very Slow and System crash suddenly.

Is there any way you can give me binaries for linux or even windows for me to tinker around with? :max: I tried compiling, but it just doesn't work for me here...

[QUOTE=kracker;343740]Is there any way you can give me binaries for linux or even windows for me to tinker around with? :max: I tried compiling, but it just doesn't work for me here...[/QUOTE]
Can you compile MatrixMulImage on linux ?
[QUOTE]
desktop:/opt/AMDAPP/samples/opencl/cl/app/MatrixMulImage$ make
mkdir -p depends/x86_64
perl ../../../../../make/fastdep.pl -I. -I../../../../../include -I../../../../
../samples/opencl/SDKUtil/include -I../../../../../samples/bolt/BoltUtil/includ
e -I../../../../../samples/C++Amp/AmpUtil/include --obj-suffix='.o' --obj-prefi
x='build/debug/x86_64//' MatrixMulImage.cpp > depends/x86_64/MatrixMulImage.depe
nd
mkdir -p build/debug/x86_64/
Building build/debug/x86_64//MatrixMulImage.o
g++ -Wpointer-arith -Wfloat-equal -g3 -ffor-scope -I ../../../../../sample
s/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//MatrixMulImage.o -c MatrixMulImage.c
pp
Building build/debug/x86_64/MatrixMulImage
g++ -o build/debug/x86_64/MatrixMulImage build/debug/x86_64//MatrixMulImage.o -l
pthread -ldl -L/usr/X11R6/lib -lSDKUtil -lOpenCL -L../../../../../lib/x86_6
4 -L../../../../../TempSDKUtil/lib/x86_64 -L"/opt/AMDAPP/lib/x86_64"
install -D build/debug/x86_64/MatrixMulImage ../../../../../samples/opencl/bin/x
86_64/MatrixMulImage
for f in MatrixMulImage_Kernels.cl; do \
install -D $f ../../../../../samples/opencl/bin/x86_64/$f; \
done
[/QUOTE]

0.27 a.out on Ubuntu 12.04 LTS
[QUOTE]
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/clAmdFft-1.10.321/lib64/
time ./a.out 607
[/QUOTE]

I tried booting linux after not using it for around a month I think... for some reason wouldn't boot, so I tried reinstalling both Ubuntu and Fedora.. no luck, it boots to a black screen, the fallback graphic mode and nomodeset didn't work either... damn linux :P