seam_carver_functions.c

/*
LICENSE - Public Domain (www.unlicense.org)

This is free and unencumbered software released into the public domain.
Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
software, either in source code form or as a compiled binary, for any purpose,
commercial or non-commercial, and by any means.
In jurisdictions that recognize copyright laws, the author or authors of this
software dedicate any and all copyright interest in the software to the public
domain. We make this dedication for the benefit of the public at large and to
the detriment of our heirs and successors. We intend this dedication to be an
overt act of relinquishment in perpetuity of all present and future rights to
this software under copyright law.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/


#include <stdlib.h>
#include <stdio.h>

#include "seam_carver.h"
#include "cuda_kernels.h"

uchar4 *build_pixels(const unsigned char *imgv, int w, int h){
    uchar4 *pixels = (uchar4*)malloc(w*h*sizeof(uchar4));
    int i, j;
    uchar4 pix;
    for(i = 0; i < h; i++){
        for(j = 0; j < w; j++){
            pix.x = imgv[i*3*w + 3*j];
            pix.y = imgv[i*3*w + 3*j + 1];
            pix.z = imgv[i*3*w + 3*j + 2]; 
            pixels[i*w + j] = pix;            
        }
    }
    return pixels;
}

unsigned char *flatten_pixels(uchar4 *pixels, int w, int h, int new_w){
    unsigned char *flattened = (unsigned char*)malloc(3*new_w*h*sizeof(unsigned char));
    int i, j;
    uchar4 pix;
    for(i = 0; i < h; i++){
        for(j = 0; j < new_w; j++){ 
            pix = pixels[i*w + j];
            flattened[3*i*new_w + 3*j] = pix.x;
            flattened[3*i*new_w + 3*j + 1] = pix.y;
            flattened[3*i*new_w + 3*j + 2] = pix.z;
        }
    }
    return flattened;
}

void seam_carver_init(seam_carver *sc, seam_carver_mode mode, unsigned char* imgv, int w, int h){
    sc->w = w;
    sc->h = h;
    sc->current_w = w;
    sc->mode = mode;
    
    sc->h_pixels = build_pixels(imgv, w, h);
   
    cudaMalloc((void**)&sc->d_pixels, w*h*sizeof(uchar4)); 
    cudaMalloc((void**)&sc->d_pixels_swap, w*h*sizeof(uchar4)); 
    
    if(sc->mode != SEAM_CARVER_APPROX_MODE){
        cudaMalloc((void**)&(sc->d_costs.left), w*h*sizeof(short)); 
        cudaMalloc((void**)&(sc->d_costs.up), w*h*sizeof(short)); 
        cudaMalloc((void**)&(sc->d_costs.right), w*h*sizeof(short)); 
    }
    //printf("%lu \n",sizeof(short));
    
    if(sc->mode == SEAM_CARVER_UPDATE_MODE){
        cudaMalloc((void**)&(sc->d_costs_swap.left), w*h*sizeof(short)); 
        cudaMalloc((void**)&(sc->d_costs_swap.up), w*h*sizeof(short)); 
        cudaMalloc((void**)&(sc->d_costs_swap.right), w*h*sizeof(short));
    }
    
    cudaMalloc((void**)&sc->d_M, w*h*sizeof(int)); 
    
    if(sc->mode == SEAM_CARVER_APPROX_MODE)
        sc->reduce_row = &(sc->d_M[0]); //first row
    else
        sc->reduce_row = &(sc->d_M[w*(h-1)]); //last row
        
    if(sc->mode == SEAM_CARVER_APPROX_MODE){
        cudaMalloc((void**)&sc->d_index_map, w*h*sizeof(int));
        cudaMalloc((void**)&sc->d_offset_map, w*h*sizeof(int));
        //cudaMallocHost((void**)&sc->h_index_map, w*h*sizeof(int));
        //cudaMallocHost((void**)&sc->h_seam, h*sizeof(int));
        //cudaStreamCreate(&sc->kernel_stream);
        //cudaStreamCreate(&sc->copy_stream);
    }
    
    //alloc on device for indices
    cudaMalloc((void**)&sc->d_indices, w*sizeof(int)); 
    cudaMalloc((void**)&sc->d_indices_ref, w*sizeof(int)); 
    cudaMalloc((void**)&sc->d_seam, h*sizeof(int));    
}

void seam_carver_resize(seam_carver *sc, int seams_to_remove){
    uchar4* pixels_tmp;
    cost_data costs_tmp;
    int num_iterations;
    
    //copy image pixels from host to device 
    cudaMemcpy(sc->d_pixels, sc->h_pixels, sc->w*sc->h*sizeof(uchar4), cudaMemcpyHostToDevice);       
    int* indices = (int*)malloc(sc->w*sizeof(int));
    for(int i = 0; i < sc->w; i++){
        indices[i] = i;
    }     
    //copy indices reference to device
    cudaMemcpy(sc->d_indices_ref, indices, sc->w*sizeof(int), cudaMemcpyHostToDevice);   
    
    if(sc->mode == SEAM_CARVER_UPDATE_MODE){
        compute_costs(*sc);
    }
    
    num_iterations = 0;
    while(num_iterations < seams_to_remove){
        
        if(sc->mode == SEAM_CARVER_STANDARD_MODE){
            compute_costs(*sc);
        }
        
        if(sc->mode != SEAM_CARVER_APPROX_MODE){
            compute_M(*sc);
            find_min_index(*sc); 
            find_seam(*sc);
        }
        else{
            approx_setup(*sc);       
            approx_M(*sc);
            find_min_index(*sc);
            approx_seam(*sc);               
        }
        
        remove_seam(*sc);   
        //swap pixels
        pixels_tmp = sc->d_pixels;
        sc->d_pixels = sc->d_pixels_swap;
        sc->d_pixels_swap = pixels_tmp;
        
        if(sc->mode == SEAM_CARVER_UPDATE_MODE){
            update_costs(*sc);
            //swap costs
            costs_tmp = sc->d_costs;
            sc->d_costs = sc->d_costs_swap;
            sc->d_costs_swap = costs_tmp;
        }
             
        sc->current_w = sc->current_w - 1;
        num_iterations = num_iterations + 1;
    }
    
    cudaMemcpy(sc->h_pixels, sc->d_pixels, sc->w*sc->h*sizeof(uchar4), cudaMemcpyDeviceToHost);
    sc->output = flatten_pixels(sc->h_pixels, sc->w, sc->h, sc->current_w); 
    free(indices);
}

void seam_carver_destroy(seam_carver *sc){
    cudaFree(sc->d_pixels);
    cudaFree(sc->d_pixels_swap);
    if(sc->mode != SEAM_CARVER_APPROX_MODE){
        cudaFree(sc->d_costs.left);
        cudaFree(sc->d_costs.up);
        cudaFree(sc->d_costs.right);
    }
    if(sc->mode == SEAM_CARVER_UPDATE_MODE){
        cudaFree(sc->d_costs_swap.left);
        cudaFree(sc->d_costs_swap.up);
        cudaFree(sc->d_costs_swap.right);
    }
    cudaFree(sc->d_M); 
    cudaFree(sc->d_indices); 
    cudaFree(sc->d_indices_ref); 
    cudaFree(sc->d_seam);
    if(sc->mode == SEAM_CARVER_APPROX_MODE){
        cudaFree(sc->d_index_map);
        cudaFree(sc->d_offset_map);
    }
    free(sc->h_pixels);
    free(sc->output);   
}