CUDA compile trouble

Hi!

I have some trouble with compiling simple cuda progrsmm from SDK on my system and don’t now what i shud do with this. Nvidia device - GeForce 9500GT

My system satisfies Minimum Software Requirements from README for dev driver:

Software Element             Installed on system

    ---------------------    ---------------------    

    Linux kernel             2.4.32                                                                                  

    Kernel modutils          2.4.25                  

    binutils                 2.17.50.0.6-14                  

    GNU make                 3.81                     

    gcc                      2.95.4

All needed paths exported in ~/.bashrc:

[user@MCBC ~]# echo $LD_LIBRARY_PATH

:/usr/local/cula/lib:/usr/local/cuda/lib\

[user@MCBC ~]# echo $PATH

...:/usr/local/cuda/bin

Information about installed dev driver:

NVRM version: NVIDIA UNIX x86 Kernel Module  260.19.12  Fri Oct  8 13:22:24 PDT 2010

I’m trying o compile exemple from sdk called VectorAdd.cu

[user@MCBC ~]# nvcc vectorAdd.cu

and i catch

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

Please help. Where to dig?

p.s.: sorry for my english

Hi!

I have some trouble with compiling simple cuda progrsmm from SDK on my system and don’t now what i shud do with this. Nvidia device - GeForce 9500GT

My system satisfies Minimum Software Requirements from README for dev driver:

Software Element             Installed on system

    ---------------------    ---------------------    

    Linux kernel             2.4.32                                                                                  

    Kernel modutils          2.4.25                  

    binutils                 2.17.50.0.6-14                  

    GNU make                 3.81                     

    gcc                      2.95.4

All needed paths exported in ~/.bashrc:

[user@MCBC ~]# echo $LD_LIBRARY_PATH

:/usr/local/cula/lib:/usr/local/cuda/lib\

[user@MCBC ~]# echo $PATH

...:/usr/local/cuda/bin

Information about installed dev driver:

NVRM version: NVIDIA UNIX x86 Kernel Module  260.19.12  Fri Oct  8 13:22:24 PDT 2010

I’m trying o compile exemple from sdk called VectorAdd.cu

[user@MCBC ~]# nvcc vectorAdd.cu

and i catch

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

Please help. Where to dig?

p.s.: sorry for my english

vectorAdd.cu uses

#include <cutil_inline.h>

you need to include path /NVIDIA_GPU_Computing_SDK/C/common/inc

suppose you install SDK at your home directory, then

nvcc -c -I~/NVIDIA_GPU_Computing_SDK/C/common/inc vectorAdd.cu

then link with /NVIDIA_GPU_Computing_SDK/C/lib/libcutil.a

gcc vectorAdd.o -L~/NVIDIA_GPU_Computing_SDK/C/lib -lcutil

vectorAdd.cu uses

#include <cutil_inline.h>

you need to include path /NVIDIA_GPU_Computing_SDK/C/common/inc

suppose you install SDK at your home directory, then

nvcc -c -I~/NVIDIA_GPU_Computing_SDK/C/common/inc vectorAdd.cu

then link with /NVIDIA_GPU_Computing_SDK/C/lib/libcutil.a

gcc vectorAdd.o -L~/NVIDIA_GPU_Computing_SDK/C/lib -lcutil

thanks, but did not help… similarly i have:

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

thanks, but did not help… similarly i have:

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????
  1. can you compile SDK?

  2. I apologize that you need full path, try

nvcc -c -I/usr/local/cuda/include -I[your home directory]/NVIDIA_GPU_Computing_SDK/C/common/inc -arch=sm_10 vectorAdd.cu

  1. can you compile SDK?

  2. I apologize that you need full path, try

nvcc -c -I/usr/local/cuda/include -I[your home directory]/NVIDIA_GPU_Computing_SDK/C/common/inc -arch=sm_10 vectorAdd.cu

  1. did not compiles:
[user@MCBC ~]# cd NVIDIA_GPU_Computing_SDK/C/

[user@MCBC C]# make

make[1]: Entering directory `/user/NVIDIA_GPU_Computing_SDK/C/common'

In file included from src/bank_checker.cpp:15:

../common/inc/bank_checker.h:22: limits: ??? ?????? ????? ??? ????????

make[1]: *** [obj/i386/release/bank_checker.cpp.o] Error 1

make[1]: Leaving directory `/user/NVIDIA_GPU_Computing_SDK/C/common'

make: *** [lib/libcutil.so] Error 2

2

[user@MCBC C]# nvcc -c -I/usr/local/cuda/include -I/user/NVIDIA_GPU_Computing_SDK/C/common/inc -arch=sm_10 /user/NVIDIA_GPU_Computing_SDK/C/src/vectorAdd/vectorAdd.cu

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????
  1. did not compiles:
[user@MCBC ~]# cd NVIDIA_GPU_Computing_SDK/C/

[user@MCBC C]# make

make[1]: Entering directory `/user/NVIDIA_GPU_Computing_SDK/C/common'

In file included from src/bank_checker.cpp:15:

../common/inc/bank_checker.h:22: limits: ??? ?????? ????? ??? ????????

make[1]: *** [obj/i386/release/bank_checker.cpp.o] Error 1

make[1]: Leaving directory `/user/NVIDIA_GPU_Computing_SDK/C/common'

make: *** [lib/libcutil.so] Error 2

2

[user@MCBC C]# nvcc -c -I/usr/local/cuda/include -I/user/NVIDIA_GPU_Computing_SDK/C/common/inc -arch=sm_10 /user/NVIDIA_GPU_Computing_SDK/C/src/vectorAdd/vectorAdd.cu

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

It seems that gcc 2.95.4 does not work when compiling SDK.

I modify vectorAdd.cu as attached file. I remove all macro, so you don’t need to include NVIDIA_GPU_Computing_SDK/C/common/inc

try this vectorAdd.cu by

nvcc -arch=sm_11 -c vectorAdd.cu

g++  vectorAdd.o -L/usr/local/cuda/lib64  -lcudart

If this works, then you may consider upgrading your gcc

It seems that gcc 2.95.4 does not work when compiling SDK.

I modify vectorAdd.cu as attached file. I remove all macro, so you don’t need to include NVIDIA_GPU_Computing_SDK/C/common/inc

try this vectorAdd.cu by

nvcc -arch=sm_11 -c vectorAdd.cu

g++  vectorAdd.o -L/usr/local/cuda/lib64  -lcudart

If this works, then you may consider upgrading your gcc

I tried this, but problem still here

[user@MCBC ~]# nvcc -arch=sm_11 -c vectorAdd.cu

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

I tried this, but problem still here

[user@MCBC ~]# nvcc -arch=sm_11 -c vectorAdd.cu

cpp0: cuda_runtime.h: ??? ?????? ????? ??? ????????

something wrong with nvcc.

we can do simple test, edit a file named main.cu

// main.cu

__global__  void foo( int *x, int *y )

{

        y[threadIdx.x] = x[threadIdx.x] ;

}

nvcc -ptx -arch=sm_11 main.cu

it should work because foo is a simple copy kernel.

If it does not work, then you should check version of nvcc

nvcc --version

According to backward compatible, new driver would work on old toolkit, so you don’t need to update toolkit.

However I would suggest that you install cuda 3.2

something wrong with nvcc.

we can do simple test, edit a file named main.cu

// main.cu

__global__  void foo( int *x, int *y )

{

        y[threadIdx.x] = x[threadIdx.x] ;

}

nvcc -ptx -arch=sm_11 main.cu

it should work because foo is a simple copy kernel.

If it does not work, then you should check version of nvcc

nvcc --version

According to backward compatible, new driver would work on old toolkit, so you don’t need to update toolkit.

However I would suggest that you install cuda 3.2

Same problem(((

[user@MCBC ~]# nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver

Copyright (c) 2005-2010 NVIDIA Corporation

Built on Thu_Sep__9_22:17:49_PDT_2010

Cuda compilation tools, release 3.2, V0.2.1221

I installed the latest version of driver from Nvidia devzone… somfing like 260.19.12

Same problem(((

[user@MCBC ~]# nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver

Copyright (c) 2005-2010 NVIDIA Corporation

Built on Thu_Sep__9_22:17:49_PDT_2010

Cuda compilation tools, release 3.2, V0.2.1221

I installed the latest version of driver from Nvidia devzone… somfing like 260.19.12

Now I see that laid out a new driver version 260.19.14. Try it and tell later…

Now I see that laid out a new driver version 260.19.14. Try it and tell later…