The Game of Life in CUDA

pootle1 · December 12, 2009, 7:50pm

OOh, nice one - even runs quite well on my laptop ( with 9300m )

jacquesm · December 21, 2009, 1:59am

I’ve spent some time optimizing the code and to my surprise it is mostly the display updates that consume a lot of time.

(gtx 260)

erdooom · December 21, 2009, 6:48am

I don’t know what optimizations you did, but i got the kernel down to half a milisecond. for a 1024x1024 on a quadro 5800. but just coping the data back takes 3 miliseconds, and rendering … of course using gl interop and vbo rendering will speed stuff up. Oh and in my version the boarders loop around so that what ever goes out the bottem come back from the top and so forth.

Snowball_Two · December 22, 2009, 10:43am

down to half a millisecond? how much time was it before? :D

go fourth with the optimisation, and give us some code :D

i tryed to make it looping too…but…damn…for any reason it does not work…

erdooom · December 22, 2009, 12:33pm

ill post the new code soon, any ways one thing is that you need to switch the buffers for each run, the last out is now the in and the last in is now the out … hope this helps

biebo · December 24, 2009, 5:47am

erdooom

with your optimisation are you guaranteeing coalesced memory accesses?

erdooom · December 24, 2009, 6:45am

nope not really, even if we find a way to do that (which is very tricky on the border cells of a block since they need to read data that belongs to the next block) it would completely screw over the memory reads for the gl draw. but half a millisecond is pretty good. using gl inter-op would give it a much bigger performance boost. I did move to using the shared memory which improved the time from 0.8 ms to 0.5 …

mitchde · December 25, 2009, 11:29am

How to get this compiled on Mac OS X ?

EDIT: DONE :)

I get 6 FPS CPU (C2D 2.8 GHz) and 28 FPS GPU (8800)

Hmmytheos · December 28, 2009, 4:10am

(Noob in CUDA programming, new here…)

Well i have a similar problem

I have a project to make :

Write a program in CUDA simulating GAME OF LIFE with the following requirements:

Inputs: 1) number of iterations, t
2) file which contains the starting board (nXn)

Actions: 1) Read the starting board
2) do t iterations
3) Write the final board (nxn) in a new file (to be able to use it as input later)
The calculation-development must use the GPU

Notes: In calculations you must use cyclic boundary conditions (the next cell of n-1 cell is 0 cell, and reverse)
Your program must be able to run successfully using boards (nxn) with n<= 2000 and use Shared memory

Any kind of help appreciated

At the moment trying to work on the given code(attached by erdooom and Snowball_Two) so solutions on it will be preferred

erdooom · December 28, 2009, 7:49am

so you preaty much have it all done, all u need is to read the file and write it.

Stelman · December 28, 2009, 2:07pm

Can someone help me out a little bit? I couldn’t understand whether the automaton’s boundaries were cyclical or not. The Neighbors function got me a little confused. In a previous implementation of Game of Life I had written in C# I had used a a 2d array.

NOTE: I am referring to the code that SNOWBALL TWO posted

biebo · December 28, 2009, 9:08pm

I think in snowball two’s code there is no cyclic boundary thing … and in the Dev_Neighbors() , function from conditions like (x+i<SCREENX) && (y+k<SCREENY) i think when it coming to boundaries it doesn’t bother calculating …

Hmmytheos · December 30, 2009, 5:56am

Well i’m a little bit confused…

I have to run my project in a UI (linux-based).

In this UI all the appropriate files and libs in order to run CUDA progs are preinstalled.

so i uploaded my project GoL_perfect.cu(FYI is the exact code uploaded here from erdooom and Snowball_Two)

and tried to compile it but there were lots of errors (concerning mainly OpenGL in some cases)

the command i used was

nvcc -I/usr/local/NVIDIA_CUDA_SDK_2.3/C/common/inc -o test GoL_perfect.cu -L/usr/local/NVIDIA_CUDA_SDK_2.3/C/common/lib -lglut

but i got

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_menu.o): In function `fgDisplayMenu':

(.text+0x12ad): undefined reference to `glMatrixMode'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_menu.o): In function `fgDisplayMenu':

(.text+0x12b2): undefined reference to `glPopMatrix'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_state.o): In function `fghGetConfig':

(.text+0x2db): undefined reference to `glXGetConfig'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_state.o): In function `glutGet':

(.text+0x3dd): undefined reference to `XGetWindowAttributes'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_state.o): In function `glutGet':

(.text+0x480): undefined reference to `XTranslateCoordinates'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_state.o): In function `glutGet':

(.text+0x4c7): undefined reference to `XTranslateCoordinates'

/usr/lib/gcc/x86_64-redhat-linux/4.1.2/../../../../lib64/libglut.a(libglut_la-freeglut_window.o): In function `glutFullScreen':

… and More

trying with glut32

$ nvcc -I/usr/local/NVIDIA_CUDA_SDK_2.3/C/common/inc -o test GoL_perfect.cu -L/usr/local/NVIDIA_CUDA_SDK_2.3/C/common/lib/glut32

/tmp/tmpxft_00000867_00000000-11_GoL_perfect.o: In function `computeFPS()':

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x422): undefined reference to `cutGetTimerValue'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x46f): undefined reference to `glutSetWindowTitle'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x484): undefined reference to `cutResetTimer'

/tmp/tmpxft_00000867_00000000-11_GoL_perfect.o: In function `display()':

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x6a9): undefined reference to `cutStartTimer'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x6c4): undefined reference to `glMatrixMode'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x6c9): undefined reference to `glLoadIdentity'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x6fc): undefined reference to `gluOrtho2D'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x706): undefined reference to `glClear'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x710): undefined reference to `glMatrixMode'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x715): undefined reference to `glLoadIdentity'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x827): undefined reference to `glFlush'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x869): undefined reference to `cutStopTimer'

/tmp/tmpxft_00000867_00000000-11_GoL_perfect.o: In function `main':

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0x9a1): undefined reference to `cutGetCmdLineArgumentstr'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xa28): undefined reference to `cutCreateTimer'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xa4c): undefined reference to `glutInit'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xa56): undefined reference to `glutInitDisplayMode'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xa65): undefined reference to `glutInitWindowSize'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xa6f): undefined reference to `glutCreateWindow'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xafe): undefined reference to `glutMainLoop'

tmpxft_00000867_00000000-10_GoL_perfect.ii:(.text+0xb09): undefined reference to `cutDeleteTimer'

collect2: ld returned 1 exit status

(the above are just small parts of the actual error messages)

so i am wondering :

Which is the exact command to compile my project properly?

It is required to compile and run it on this UI

connecting using PuttY

PS:i am assuming that the problem is with the code line #include <GL/glut.h>

but when i include the folder “inc” and the folder “GL” is its subfolder doesnt mean that all should be ok?(and compiler should recongnize glut.h)

Sarnath · December 30, 2009, 9:55am

-L path for glut must be added to command line?

biebo · December 30, 2009, 9:12pm

This is my makefile which I mostly use this might help you. :rolleyes:

[codebox]SDK_ROOT = /home/bibrak/NVIDIA_GPU_Computing_SDK

CUDA_LIB_PATH = /usr/local/cuda/lib

INLCUDE = -I$(SDK_ROOT)/C/common/inc -I/usr/local/cuda/include

LIBS = -L$(SDK_ROOT)/C/lib -lcutil -lglut

LIBS += -L$(CUDA_LIB_PATH) -lcudart -lcuda

SRC_CU = GoL.cu

SRC_CXX =

all:

nvcc $(INLCUDE) $(LIBS) $(SRC_CU) $(SRC_CXX)[/codebox]

[b]

remember there is a tab before nvcc[/b]

BTW, are you running the GUI through PUTTY ?

I run GUI of a remote linux machine through

ssh -X user@address

i am not sure, if there is a way to doing it on PUTTY.

Hmmytheos · December 30, 2009, 11:39pm

there is…

My professor who owns the linux server told us to run the project there.

so i am connecting to my UI through Putty (which is ussing ssh btw)

anyway

the only info that he gave to me is to install from /root directory the CUDA SDK 2.3 installer into my home directory…

but in the /root directory i only found NVIDIA_CUDA_SDK_2.02.0807.1535_linux.run

so i installed it in ~/NVIDIA_CUDA_SDK folder. Installation seems to ended ok but when i tried to modify the .bash_profile following

nvidia’s README about linux(and taken as granted that drivers and toolkit were already installed by my professor)

i change it and it looks like

# .bash_profile

# Get the aliases and functions

if [ -f ~/.bashrc ]; then

		. ~/.bashrc

fi

# User specific environment and startup programs

PATH=$PATH:$HOME/bin:/usr/local/cuda/bin:~/NVIDIA_CUDA_SDK/bin

export PATH

LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib:~/NVIDIA_CUDA_SDK/lib

export LD_LIBRARY_PATH

but when i go further to building the project examples i get errors

/usr/bin/ld: cannot find -lcudart

collect2: ld returned 1 exit status

make[1]: *** [../../bin/linux/release/quasirandomGenerator] Error 1

make[1]: Leaving directory `/export/home/username/NVIDIA_CUDA_SDK/projects/quasirandomGenerator'

make: *** [projects/quasirandomGenerator/Makefile.ph_build] Error 2

so smthn is wrong…

biebo · December 31, 2009, 7:02pm

bibrak@bibrak-laptop:/usr/local/cuda/lib$ ls
libcublasemu.so libcublas.so.2 libcudart.so.2.3 libcufft.so
libcublasemu.so.2 libcublas.so.2.3 libcufftemu.so libcufft.so.2
libcublasemu.so.2.3 libcudart.so libcufftemu.so.2 libcufft.so.2.3
libcublas.so libcudart.so.2 libcufftemu.so.2.3

check for the libcudart.so
if its there then check your compilation command does it include -lcudart

some thing like

nvcc program.cu -lcudart

biebo · December 31, 2009, 7:06pm

hi all…

I have implemented a different kernel of the Game of Life.

please look at it, and comment on, "Does it make COALESCED memory accesses ? "

biebo

Hmmytheos · December 31, 2009, 7:48pm

path /usr/local/cuda/lib

libcublasemu.so libcublas.so.2 libcudart.so.2.3 libcufft.so

libcublasemu.so.2 libcublas.so.2.3 libcufftemu.so libcufft.so.2

libcublasemu.so.2.3 libcudart.so libcufftemu.so.2 libcufft.so.2.3

libcublas.so libcudart.so.2 libcufftemu.so.2.3

path /usr/local/cuda/lib64

libcublasemu.so libcublas.so.2 libcudart.so.2.3 libcufft.so

libcublasemu.so.2 libcublas.so.2.3 libcufftemu.so libcufft.so.2

libcublasemu.so.2.3 libcudart.so libcufftemu.so.2 libcufft.so.2.3

libcublas.so libcudart.so.2 libcufftemu.so.2.3

path /usr/local/cuda/bin

bin2c cudafe++ filehash nvcc.profile ptxvars.cu

cudafe fatbin nvcc ptxas

with the command nvcc -o test1 GoL_perfect.cu -lGL -lglut -lcudart

i get

/tmp/tmpxft_00000332_00000000-11_GoL_perfect.o: In function `computeFPS()':

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x422): undefined reference to `cutGetTimerValue'

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x484): undefined reference to `cutResetTimer'

/tmp/tmpxft_00000332_00000000-11_GoL_perfect.o: In function `display()':

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x6a9): undefined reference to `cutStartTimer'

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x6fc): undefined reference to `gluOrtho2D'

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x869): undefined reference to `cutStopTimer'

/tmp/tmpxft_00000332_00000000-11_GoL_perfect.o: In function `main':

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0x9a1): undefined reference to `cutGetCmdLineArgumentstr'

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0xa28): undefined reference to `cutCreateTimer'

tmpxft_00000332_00000000-10_GoL_perfect.ii:(.text+0xb09): undefined reference to `cutDeleteTimer'

collect2: ld returned 1 exit status

biebo · December 31, 2009, 7:54pm

try adding -lcutil

Topic		Replies	Views
Cannot run any CUDA kernels CUDA runtime doesn't recognize NVIDIA GPU CUDA Programming and Performance	26	12345	August 24, 2010
Problem with performance running parallel blocks CUDA Programming and Performance	3	582	March 26, 2018
my speedy FFT 3x faster than CUFFT CUDA Programming and Performance	139	241051	November 16, 2011
Integration with XCODE? Anyone created a CUDA template project? CUDA Programming and Performance	101	577748	August 10, 2014
HPLinpack for CUDA Any interest? CUDA Programming and Performance	27	11953	May 10, 2012
CUDA very slow performance CUDA Programming and Performance	21	16702	March 6, 2020
New Features in CUDA 7.5 Technical Blog	66	1086	August 10, 2016
Help me understand "-Xptxas -dlcm=cg" (take 2) CUDA Programming and Performance	1	6975	November 24, 2010
Instructions for making CUDA 2.3 work on Snow Leopard How to fix permissions and nvcc linker errors CUDA Programming and Performance	33	98603	April 30, 2010
[SOLVED] Run CUDA on dedicated NVIDIA GPU while connecting monitors to Intel HD graphics, is this possible? CUDA Setup and Installation	15	71791	December 9, 2018

The Game of Life in CUDA

Related topics