int main() {
time_t start = time(NULL);
int dim = L * (nmax + 1);
const real epsg = EPSG;
const real epsf = EPSF;
const real epsx = EPSX;
const int maxits = MAXITS;
stpscal = 0.5;
int info;
real* x;
int* nbd;
real* l;
real* u;
memAlloc<real>(&x, dim);
memAlloc<int>(&nbd, dim);
memAlloc<real>(&l, dim);
memAlloc<real>(&u, dim);
memAllocHost<real>(&f_tb_host, &f_tb_dev, 1);
cudaSetDeviceFlags(cudaDeviceMapHost);
cublasCreate_v2(&cublasHd);
U = 1;
J = 0.1;
mu = 0.5;
initProb(x, nbd, l, u, dim);
lbfgsbminimize(dim, 4, x, epsg, epsf, epsx, maxits, nbd, l, u, info);
printf("info: %d\n", info);
printf("f: %e\n", *f_tb_host);
real* x_host = new real[dim];
memCopy(x_host, x, dim * sizeof(real), cudaMemcpyDeviceToHost);
printf("x: ");
for (int i = 0; i < dim; i++) {
printf("%f, ", x_host[i]);
}
printf("\n");
memFreeHost(f_tb_host);
memFree(x);
memFree(nbd);
memFree(l);
memFree(u);
cublasDestroy_v2(cublasHd);
cudaDeviceReset();
time_t end = time(NULL);
printf("Runtime: %ld", end-start);
}
real BFGSOptimization()
{
printf("Before init sites(point_num=%d).\n", point_num);
// Use L-BFGS method to compute new sites
const real epsg = EPSG;
const real epsf = EPSF;
const real epsx = EPSX;
const int maxits = MAXITS;
stpscal = 2.0f; //Set for different problems!
int info;
total_time = 0;
total_time_func = 0;
real* x;
int* nbd;
real* l;
real* u;
memAlloc<real>(&x, point_num * 2);
memAlloc<int>(&nbd, point_num * 2);
memAlloc<real>(&l, point_num * 2);
memAlloc<real>(&u, point_num * 2);
memAllocHost<real>(&f_tb_host, &f_tb_dev, 1);
// Kernel이 처리할 수 있도록 site_list를 매핑하는 site_list_dev를 전달.
// site_list는 InitializeSites()에서 지정
// Device에 할당된 x에 site_list가 복사
InitSites(x, (float*)site_list_dev, sizeof(SiteType) / sizeof(float), nbd, l, u, point_num * 2, screenwidth, screenheight);
printf("Start optimization...");
get_timestamp(start_time);
int m = 8;
if (point_num * 2 < m)
m = point_num * 2;
// 내부적으로 funcgrad()를 호출
lbfgsbminimize(point_num*2, m, x, epsg, epsf, epsx, maxits, nbd, l, u, info);
//printf("Ending code:%d\n", info);
get_timestamp(end_time);
elapsed_time = (end_time-start_time);
total_time += elapsed_time;
printf("Done.\n JFA Time: %lf\tBFGS Time: %lf\tTotal time: %lf\t", total_time_func, elapsed_time - total_time_func, elapsed_time);
bReCompute = false;
real f = DrawVoronoi(x);
// Device에 저장된 x가 실제 이동된 site 정보인 듯
// 이를 Host로 복사한 후, site_list에 할당
real* x_host = new real[point_num * 2];
memCopy(x_host, x, point_num * 2 * sizeof(real), cudaMemcpyDeviceToHost);
FILE* fp = fopen("Result.txt", "w");
for(int i = 0; i < point_num; i++) {
real ix = x_host[i * 2];
real iy = x_host[i * 2 + 1];
real ox = (ix + 1) * (screenwidth - 1) / 2.0 + 1.0;
real oy = (iy + 1) * (screenheight - 1) / 2.0 + 1.0;
if(1.0f > ox || ox > screenwidth - 1)
continue;
if(1.0f > oy || oy > screenheight - 1)
continue;
fprintf(fp, "%f, %f\n", ox, oy);
}
fclose(fp);
delete[] x_host;
memFreeHost(f_tb_host);
memFree(x);
memFree(nbd);
memFree(l);
memFree(u);
return f;
}