System just halts for simple CUDA program

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi there, the following code brings my Dell Precision 7500 Box ( with a Tesla C2050 ) to a complete halt. The only thing left I can do is to press to power button to turn off the system. I cannot see what is wrong with this code. Despite the fact it’s the worst implementation of a simple convolution algorithm. It only runs one thread total. Here is the code:

[codebox]

#include

#include

#include

#include <cuda.h>

const unsigned int SRC_N = 10000;

const unsigned int DST_N = 2 * SRC_N - 1;

global void convolve_gpu_simple( float* src_1

                               , float* src_2

                               , float* dst

                               )

{

for( int i = 0; i < SRC_N; ++i )

{

    for( int j = 0; j < SRC_N; ++j )

    {

        dst[ i + j ] += src_1[i] * src_2[j];

    }

}

}

void check_for_error( cudaError_t err )

{

if( err != cudaSuccess )

{

    const char* error_msg = cudaGetErrorString( err );

    exit( 1 );

}

}

void convolve_cpu( float* a, float* b, float* c )

{

for( int i = 0; i < SRC_N; ++i )

{

    for( int j = 0; j < SRC_N; ++j )

    {

        c[ i + j ] += a[i] * b[j];

    }

}

}

bool compare_results( float* a, float* b )

{

unsigned int m = 0;

for( int i = 0; i < DST_N; ++i )

{

    int v_a = *(( unsigned int* )&a[i]);

    int v_b = *(( unsigned int* )&b[i]);

int ulp = abs( v_a - v_b );

    m = max( m, ulp );

if( ulp > 5 )

        return false;

}

return true;

}

using namespace std;

int main(int argc, char** argv)

{

srand( 0 );

float *src_1, *src_2, *dst_cpu, *dst_gpu;

float *dev_src_1, *dev_src_2, *dev_dst;

src_1 = new float[ SRC_N ];

src_2   = new float[ SRC_N ];

dst_cpu = new float[ DST_N ];

dst_gpu = new float[ DST_N ];

check_for_error( cudaMalloc( &dev_src_1, SRC_N * sizeof( float )));

check_for_error( cudaMalloc( &dev_src_2, SRC_N * sizeof( float )));

check_for_error( cudaMalloc( &dev_dst  , DST_N * sizeof( float )));

fill( src_1, src_1 + SRC_N, 1.0f );

fill( src_2, src_2 + SRC_N, 1.0f );

fill( dst_cpu, dst_cpu + DST_N, 0.0f );

fill( dst_gpu, dst_gpu + DST_N, 0.0f );

convolve_cpu( src_1, src_2, dst_cpu );

check_for_error( cudaMemcpy( dev_src_1, src_1, SRC_N * sizeof( float ), cudaMemcpyHostToDevice ));

check_for_error( cudaMemcpy( dev_src_2, src_2, SRC_N * sizeof( float ), cudaMemcpyHostToDevice ));

convolve_gpu_simple<<< 1, 1 >>>( dev_src_1, dev_src_2, dev_dst );

check_for_error( cudaGetLastError() );

check_for_error( cudaMemcpy( dst_gpu, dev_dst, DST_N * sizeof( float ), cudaMemcpyDeviceToHost ));

assert( compare_results( dst_cpu, dst_gpu ));

cudaFree( dev_src_1 );

cudaFree( dev_src_2 );

cudaFree( dev_dst );

delete src_1;

delete[] src_2;

delete[] dst_cpu;

delete[] dst_gpu;

return 0;

}[/codebox]

Any help is very much appreciated. I’m using VS2008 on a Windows 7 x64 box. The NVidia driver is 258.96 and I use CUDA 3.1.

Thanks,

Christian

2

Hi there,

is there anyone who can tell me what the problem might be? This is kinda important for me and my company.

Thanks,

Christian

3

Hi there,

is there anyone who can tell me what the problem might be? This is kinda important for me and my company.

Thanks,

Christian