Function processImage

cpp/5_Domain_Specific/postProcessGL/main.cpp:357–389

copy image and process using CUDA

// copy image and process using CUDA
void processImage()
{
    // run the Cuda kernel
    process(image_width, image_height, blur_radius);

// CUDA generated data in cuda memory or in a mapped PBO made of BGRA 8 bits
// 2 solutions, here :
// - use glTexSubImage2D(), there is the potential to loose performance in
// possible hidden conversion
// - map the texture and blit the result thanks to CUDA API
#ifdef USE_TEXSUBIMAGE2D
    glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo_dest);

    glBindTexture(GL_TEXTURE_2D, tex_cudaResult);
    glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
    SDK_CHECK_ERROR_GL();
    glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0);
    glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
#else
    // We want to copy cuda_dest_resource data to the texture
    // map buffer objects to get CUDA device pointers
    cudaArray *texture_ptr;
    checkCudaErrors(cudaGraphicsMapResources(1, &cuda_tex_result_resource, 0));
    checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(&texture_ptr, cuda_tex_result_resource, 0, 0));

    int num_texels    = image_width * image_height;
    int num_values    = num_texels * 4;
    int size_tex_data = sizeof(GLubyte) * num_values;
    checkCudaErrors(cudaMemcpyToArray(texture_ptr, 0, 0, cuda_dest_resource, size_tex_data, cudaMemcpyDeviceToDevice));

    checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_tex_result_resource, 0));
#endif
}

// display image to the screen as textured quad
void displayImage(GLuint texture)