CUDA1

#include <stdio.h>
#include <cuda_runtime.h>
#include "..\include\opencv\highgui.h"

void run_grayscale( uchar4 *color_pic, uchar4 *bw_pic, int sizex, int
sizey );

int main( int numarg, char **arg )
{
if ( numarg < 2 )
{
printf( "Enter picture filename!\n" );
return 1;
}

// nacteni obrazku
IplImage *bgr_img = cvLoadImage( arg[ 1 ] );
int sizex = bgr_img->width;
int sizey = bgr_img->height;

// alokace poli uchar4 pro body obrazku a jeho naplneni
uchar4 *bgr_pole = new uchar4[ sizex * sizey ];
uchar4 *bw_pole = new uchar4[ sizex/2 * sizey/2 ];
for ( int y = 0; y < sizey; y++ )
for ( int x = 0; x < sizex; x++ )
{
CvScalar s = cvGet2D( bgr_img, y, x );
uchar4 bgr = { (char) s.val[ 0 ], (char) s.val[ 1 ], (char) s.val[ 2
] };
bgr_pole[ y * sizex + x ] = bgr;

}

// volani funkce ze souboru .cu
run_grayscale( bgr_pole, bw_pole, sizex, sizey );

IplImage *bw_img = cvCreateImage( cvSize( sizex/2, sizey/2 ),
IPL_DEPTH_8U, 3 );

// ziskana data ulozime do noveho obrazku
for ( int y = 0; y < sizey/2; y++ )
for ( int x = 0; x < sizex/2; x++ )
{
uchar4 bgr = bw_pole[ y * sizex/2 + x ];
CvScalar s = {bgr.x, bgr.y,bgr.z };
cvSet2D( bw_img, y, x, s );

}

// zobrazeni puvodniho obrazku a vysledku
cvShowImage( "Color", bgr_img );
cvShowImage( "GrayScale", bw_img );
cvWaitKey( 0 );
}





#include <cuda_runtime.h>
#include <stdio.h>

// Demo kernel pro prevod barevneho bodu na odstin sede.
__global__ void kernel_resize( uchar4 *original, uchar4* resizedpic, int
sizex, int sizey )
{
// souradnice vlakna, kontrola rozmeru obrazku
int y = blockDim.y * blockIdx.y + threadIdx.y;
if ( y >= sizey ) return;
int x = blockDim.x * blockIdx.x + threadIdx.x;
if ( x >= sizex ) return;

uchar4 bgr = original[ y * sizex + x ];


// ulozeni bodu do obrazku
if(y>(sizey/2-50) && y<(sizey/2+50) && x>(sizex/2-50) &&
x<(sizex/2+50))
{
bgr.x = bgr.y = bgr.z = 255;
}


if(y%2==0 && x%2==0)
resizedpic[ y/2 * sizex/2 + x/2 ] = bgr;

}

void run_grayscale( uchar4 *color_pic, uchar4* bw_pic, int sizex, int
sizey )
{
cudaError_t cerr;
// alokace pameti ve videokarte
uchar4 *cudaOriginal;
uchar4 *cudaResized;
cerr = cudaMalloc( &cudaOriginal, sizex * sizey * sizeof( uchar4 ) );
if ( cerr != cudaSuccess )
printf( "CUDA Error [%d] - '%s'\n", __LINE__, cudaGetErrorString( cerr
) );

cerr = cudaMalloc( &cudaResized, (sizex/2) * (sizey/2) * sizeof( uchar4
) );
if ( cerr != cudaSuccess )
printf( "CUDA Error [%d] - '%s'\n", __LINE__, cudaGetErrorString( cerr
) );

// prenos barevneho obrazku do videokarty
cerr = cudaMemcpy( cudaOriginal, color_pic, sizex * sizey * sizeof(
uchar4 ), cudaMemcpyHostToDevice );
if ( cerr != cudaSuccess )
printf( "CUDA Error [%d] - '%s'\n", __LINE__, cudaGetErrorString( cerr
) );

int block = 16;
dim3 blocks( ( sizex + block - 1 ) / block, ( sizey + block - 1 ) /
block );
dim3 threads( block, block );

// vytvoreni bloku s vlakny, matice vlaken muze byt vetsi, nez samotny
obrazek!
kernel_resize<<< blocks, threads >>>(cudaOriginal, cudaResized , sizex,
sizey );

if ( ( cerr = cudaGetLastError() ) != cudaSuccess )
printf( "CUDA Error [%d] - '%s'\n", __LINE__, cudaGetErrorString( cerr
) );

// prenos dat z videokarty
cerr = cudaMemcpy( bw_pic, cudaResized, (sizex/2) * (sizey/2) * sizeof(
uchar4 ), cudaMemcpyDeviceToHost );
if ( cerr != cudaSuccess )
printf( "CUDA Error [%d] - '%s'\n", __LINE__, cudaGetErrorString( cerr
) );

// uvolneni alokovane pameti ve videokarte
cudaFree( cudaOriginal );
cudaFree( cudaResized );
}