min_data.cpp

Go to the documentation of this file.

 
 
#include <cstddef>
#include <cstdio>
#include <cstdlib>
#include <ctime>
#include <vector>
#include <algorithm>
#include <chrono>
 
#include "window.h"
 
#include "vec.h"
#include "program.h"
#include "uniforms.h"
 
 
GLuint program= 0;
GLuint input_buffer= 0;
GLuint tmp_buffer= 0;
 
std::vector<int> input;
 
 
const int logNmax= 26;
 
int init( )
{
    srand(time(nullptr));
    
    int o= rand() % 10000;
    for(int i= 0; i < (1<<logNmax); i++)
        input.push_back( o + rand() % 10000);
    
    glGenBuffers(1, &input_buffer);
    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, input_buffer);
    glBufferData(GL_SHADER_STORAGE_BUFFER, sizeof(int) * input.size(), input.data(), GL_STATIC_READ);
    
    glGenBuffers(1, &tmp_buffer);
    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, tmp_buffer);
    glBufferData(GL_SHADER_STORAGE_BUFFER, sizeof(int) * input.size(), input.data(), GL_STREAM_READ);
 
#define ATOMIC
    
#ifndef ATOMIC
    // version reduction globale
    program= read_program("tutos/min_data.glsl");
#else
    // version atomique
    program= read_program("tutos/min_atomic.glsl");
    program_print_errors(program);
#endif
    
    return 0;
}
 
int quit( )
{
    release_program(program);
    glDeleteBuffers(1, &input_buffer);
    glDeleteBuffers(1, &tmp_buffer);
    return 0;
}
 
 
int main( int argc, char **argv )
{
    // etape 1 : creer la fenetre
    Window window= create_window(1024, 640,  4,3);       // openGL version 4.3
    if(window == NULL)
        return 1;
 
    // etape 2 : creer un contexte opengl pour pouvoir dessiner
    Context context= create_context(window);
    if(context == NULL)
        return 1;
 
    // etape 3 : creation des objets
    if(init() < 0)
    {
        printf("[error] init failed.\n");
        return 1;
    }
 
    // taille max des groupes de shaders
    GLint threads_max= 0;
    glGetIntegerv(GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS, &threads_max);
    printf("threads max %d\n", threads_max);
    
    // dimension max des groupes de shaders
    GLint size_max[3]= { };
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 0, &size_max[0]);
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 1, &size_max[1]);
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 2, &size_max[2]);
    printf("size max %d %d %d\n", size_max[0], size_max[1], size_max[2]);
    
    // nombre max de groupes de shaders
    GLint groups_max[3]= { };
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, &groups_max[0]);
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 1, &groups_max[1]);
    glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 2, &groups_max[2]);
    printf("groups max %d %d %d\n", groups_max[0], groups_max[1], groups_max[2]);
    
    // taille de la memoire partagee
    GLint size= 0;
    glGetIntegerv(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE, &size);
    printf("shared memory %dKb\n", size / 1024);
    printf("\n");
    
#ifndef ATOMIC
    FILE *out= fopen("min_data.txt", "wt");
#else
    FILE *out= fopen("min_atomic.txt", "wt");
    //~ FILE *out= fopen("min_atomic_shared.txt", "wt");
    //~ FILE *out= fopen("min_copy.txt", "wt");
#endif
    
    // pour N= 2, 4, 8, ... 2^n
    for(int logN= 1; logN < logNmax; logN++)
    {
        int N= (1<<logN);
        
        // run
        glUseProgram(program);
        
        // mesure le temps d'execution du dispatch 
        std::chrono::high_resolution_clock::time_point start= std::chrono::high_resolution_clock::now();
 
    #ifndef ATOMIC
        // version reduction avec synchronisation externe
        // 1ere etape
        program_uniform(program, "N", (unsigned int) N);
        program_uniform(program, "n", (unsigned int) 0);
        glDispatchCompute((N+1023) / 1024, 1, 1);
        
        // etapes suivantes
        program_uniform(program, "N", (unsigned int) 0);
        for(int n= N/2; n > 1; n= n/2)
        {
            program_uniform(program, "n", (unsigned int) n);
            
            // attendre que les resultats temporaires sont dispos
            glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
            
            //~ for(int i=0 ; i < 100; i++)
            glDispatchCompute((n+1023) / 1024, 1, 1);
        }
        
    #else
        // version avec synchronisation "atomique"
        program_uniform(program, "N", (unsigned int) N);
        // un seul dispatch
        for(int i=0 ; i < 100; i++)
        {
            glDispatchCompute((N+1023) / 1024, 1, 1);
            //~ glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
        }
    #endif
        
        // recupere les resultats tmp
        glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
        
        std::chrono::high_resolution_clock::time_point stop= std::chrono::high_resolution_clock::now();
        long int gpu_time= std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
        
        std::vector<int> tmp(input.size(), 0);
        glBindBuffer(GL_SHADER_STORAGE_BUFFER, tmp_buffer);
        glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, 0, sizeof(int) * tmp.size(), tmp.data());
        
        printf("gpu min %d:", tmp[1]);
        //~ printf(" %02dms %03dus\n", (int) (gpu_time / 1000000), (int) ((gpu_time / 1000) % 1000));
        printf(" %03dus\n", (int) (gpu_time / 1000));
        
        {
            // verification
            std::chrono::high_resolution_clock::time_point start= std::chrono::high_resolution_clock::now();
            
            // version sequentielle cpu
            int m= input[0];
            for(int i= 1; i < N; i++)
                    m= std::min(m, input[i]);
            
            std::chrono::high_resolution_clock::time_point stop= std::chrono::high_resolution_clock::now();
            long int cpu_time= std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
            
            printf("cpu min %d:", m);
            printf(" %02dms %03dus\n", (int) (cpu_time / 1000000), (int) ((cpu_time / 1000) % 1000));
            
            // ecrire les temps d'execution dans le fichier
            fprintf(out, "%d %f %f\n", N, gpu_time / 1000.f, cpu_time / 1000.f);
            
            printf("\n");
        }
    }
    fclose(out);
 
    // etape 5 : nettoyage
    quit();
    release_context(context);
    release_window(window);
    return 0;
}