API tools faq
paste
Advertisement
manac68974

Untitled

manac68974
Dec 22nd, 2019
249
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 33.85 KB | None | 0 0
raw download clone embed print report
  1. #include <mpi.h>
  2. #include <omp.h>
  3.  
  4. #include <iomanip>
  5. #include <iostream>
  6. #include <vector>
  7. #include <map>
  8. #include <cmath>
  9. #include <cstdlib>
  10. #include <algorithm>
  11. #include <fstream>
  12. #include<sstream>
  13.  
  14. using namespace std;
  15.  
  16. class MPIComputations{
  17. int M, N;
  18. int cur_block_size_x, cur_block_size_y, cur_block_global_offset_x, cur_block_global_offset_y;
  19. int Gx, Gy;
  20. int cur_block_global_coors_x, cur_block_global_coors_y;
  21. double h_x, h_y;
  22. int global_block_size_x, global_block_size_y;
  23. int my_rank;
  24. double current_norm;
  25. MPI_Comm comm;
  26. double x_left, x_right, y_bottom, y_top;
  27. ofstream my_file;
  28.  
  29. vector<double> internal_data;
  30. vector<double> old_internal_data;
  31. vector<double> external_data[4];
  32.  
  33. vector <vector <double> > coefs;
  34.  
  35. int IsNodeInternalCornerOrSide(int current_node_global_offset_x, int current_node_global_offset_y){
  36.  
  37. if ((cur_block_global_coors_x > 0 ) && (cur_block_global_coors_x < Gx - 1) && (cur_block_global_coors_y > 0) && (cur_block_global_coors_y < Gy - 1))
  38. return 0;
  39.  
  40. //corners
  41. //left bottom corner
  42. if (current_node_global_offset_x == 0 && current_node_global_offset_y == 0){
  43. return 2;
  44. }
  45.  
  46. //left top corner
  47. if (current_node_global_offset_x == 0 && current_node_global_offset_y == N){
  48. return 4;
  49. }
  50.  
  51. //right bottom corner
  52. if (current_node_global_offset_x == M && current_node_global_offset_y == 0){
  53. return 6;
  54. }
  55.  
  56. //right top corner
  57. if (current_node_global_offset_x == M && current_node_global_offset_y == N){
  58. return 8;
  59. }
  60.  
  61. //sides
  62. //left side
  63. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  64. current_node_global_offset_x == 0){
  65. return 1;
  66. }
  67.  
  68. //right side
  69. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  70. current_node_global_offset_x == M){
  71. return 3;
  72. }
  73.  
  74. //bottom side
  75. if (current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  76. current_node_global_offset_y == 0){
  77. return 5;
  78. }
  79.  
  80. //top side
  81. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  82. current_node_global_offset_y == N)){
  83. return 7;
  84. }
  85.  
  86. //internal
  87. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1) &&
  88. (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1)){
  89. return 0;
  90. }
  91.  
  92. return -1;
  93. }
  94.  
  95. double k(double x, double y) {
  96. return 1 + pow(x + y, 2);
  97. }
  98.  
  99. double q(double x, double y) {
  100. return 1;
  101. }
  102.  
  103. double u(double x, double y) {
  104. return 2.0 / (1 + pow(x, 2) + pow(y, 2));
  105. }
  106.  
  107. // psi_R(x, y) = k(A2, y) * du/dx(A2, y)
  108. double psi_R(double y) {
  109. return (-12) * (pow((y + 3), 2) + 1) / pow((pow(y, 2) + 10), 2);
  110. }
  111.  
  112. // psi_L(x, y) = -k(A1, y) * du/dx(A1, y)
  113. double psi_L(double y) {
  114. return (-8) * (pow((y - 2), 2) + 1) / pow((pow(y, 2) + 5), 2);
  115. }
  116.  
  117. // psi_T(x, y) = k(x, B2) * du/dy(x, B2)
  118. double psi_T(double x) {
  119. return (-16) * (pow((x + 4), 2) + 1) / pow((pow(x, 2) + 17), 2);
  120. }
  121.  
  122. // psi_B(x, y) = -k(x, B1) * du/dy(x, B1)
  123. double psi_B(double x) {
  124. return (-4) * (pow((x - 1), 2) + 1) / pow((pow(x, 2) + 2), 2);
  125. }
  126.  
  127. // right-part function of Poisson equation
  128. double F(double x, double y) {
  129. return 2 * (pow(x,4) + pow(y,4) + 2 * (pow(x,2) + 3) * pow(y,2) + 6 * pow(x,2) + 16*x*y + 5)
  130. / pow((1 + pow(x, 2) + pow(y, 2)), 3);
  131. }
  132.  
  133. //inner_product(A[i], internal_data)
  134.  
  135. //inner_product(A[i], internal_data)
  136. void ComputeCoefsForNode (double &left_coeff, double &right_coeff, double &bottom_coeff, double &top_coeff, double &this_coeff, int i, int j){
  137.  
  138. int glob_x = GetGlobalX(i);
  139. int glob_y = GetGlobalY(j);
  140.  
  141.  
  142. map <string,bool> neighbours;
  143. neighbours["left"] = true;
  144. neighbours["right"] = true;
  145. neighbours["bottom"] = true;
  146. neighbours["top"] = true;
  147.  
  148. left_coeff = 1.0;
  149. right_coeff = 1.0;
  150. bottom_coeff = 1.0;
  151. top_coeff = 1.0;
  152. this_coeff = 1.0;
  153. if (i > 0 && i < cur_block_size_x - 1 && j > 0 && j < cur_block_size_y - 1)
  154. ;
  155. else{
  156. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  157. case 2:
  158. //left bottom corner
  159. neighbours["left"] = false;
  160. neighbours["bottom"] = false;
  161. break;
  162. case 4:
  163. //left top corner
  164. neighbours["left"] = false;
  165. neighbours["top"] = false;
  166. break;
  167. case 6:
  168. //right bottom corner
  169. neighbours["right"] = false;
  170. neighbours["bottom"] = false;
  171. break;
  172. case 8:
  173. //right top corner
  174. neighbours["right"] = false;
  175. neighbours["top"] = false;
  176. break;
  177. case 1:
  178. //left side
  179. neighbours["left"] = false;
  180. break;
  181. case 3:
  182. //right side
  183. neighbours["right"] = false;
  184. break;
  185. case 5:
  186. //bottom side
  187. neighbours["bottom"] = false;
  188. break;
  189. case 7:
  190. //top side
  191. neighbours["top"] = false;
  192. break;
  193. case 0:
  194. //internal
  195. break;
  196. default:
  197. cout << "[ERROR]: Bad global coords compute coef. Global:" << glob_x << " " << glob_y<<endl;
  198. }
  199.  
  200.  
  201. if (!neighbours["left"]){
  202. right_coeff = 2.0;
  203. left_coeff = 0.0;
  204. }
  205.  
  206. if (!neighbours["right"]){
  207. left_coeff = 2.0;
  208. right_coeff = 0.0;
  209. }
  210.  
  211. if (!neighbours["bottom"]){
  212. top_coeff = 2.0;
  213. bottom_coeff = 0.0;
  214. }
  215.  
  216. if (!neighbours["top"]){
  217. bottom_coeff = 2.0;
  218. top_coeff = 0.0;
  219. }
  220. }
  221.  
  222. if (neighbours["left"]){
  223. left_coeff *= -k(x_left + (glob_x - 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  224. }
  225.  
  226. if (neighbours["right"]){
  227. right_coeff *= -k(x_left + (glob_x + 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  228. }
  229.  
  230. if (neighbours["bottom"]){
  231. bottom_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y - 0.5) * h_y) / pow(h_y, 2);
  232. }
  233.  
  234. if (neighbours["top"]){
  235. top_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y + 0.5) * h_y) / pow(h_y, 2);
  236. }
  237.  
  238. this_coeff = q(x_left + glob_x * h_x, y_bottom + glob_y * h_y) - left_coeff - right_coeff - bottom_coeff - top_coeff;
  239.  
  240. /*coefs[0] = left_coeff;
  241. coefs[1] = right_coeff;
  242. coefs[2] = bottom_coeff;
  243. coefs[3] = top_coeff;
  244. coefs[4] = this_coeff;*/
  245. }
  246.  
  247. double ComputeMagicInnerProductA_iw (int i, int j){
  248.  
  249. int glob_x = GetGlobalX(i);
  250. int glob_y = GetGlobalY(j);
  251.  
  252. double result = 0.0;
  253.  
  254. map <string,bool> neighbours;
  255. neighbours["left"] = true;
  256. neighbours["right"] = true;
  257. neighbours["bottom"] = true;
  258. neighbours["top"] = true;
  259.  
  260.  
  261. vector <double> my_coefs = coefs[j + cur_block_size_y*i];
  262.  
  263. double left_neighbour = 0.0, right_neighbour = 0.0, bottom_neighbour = 0.0, top_neighbour = 0.0, this_node = 0.0;
  264. double left_coeff = my_coefs[0], right_coeff = my_coefs[1], bottom_coeff = my_coefs[2], top_coeff = my_coefs[3], this_coeff = my_coefs[4];
  265.  
  266.  
  267. if (i > 0 && i < cur_block_size_x - 1 && j > 0 && j < cur_block_size_y - 1)
  268. ;
  269. else{
  270. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  271. case 2:
  272. //left bottom corner
  273. neighbours["left"] = false;
  274. neighbours["bottom"] = false;
  275. break;
  276. case 4:
  277. //left top corner
  278. neighbours["left"] = false;
  279. neighbours["top"] = false;
  280. break;
  281. case 6:
  282. //right bottom corner
  283. neighbours["right"] = false;
  284. neighbours["bottom"] = false;
  285. break;
  286. case 8:
  287. //right top corner
  288. neighbours["right"] = false;
  289. neighbours["top"] = false;
  290. break;
  291. case 1:
  292. //left side
  293. neighbours["left"] = false;
  294. break;
  295. case 3:
  296. //right side
  297. neighbours["right"] = false;
  298. break;
  299. case 5:
  300. //bottom side
  301. neighbours["bottom"] = false;
  302. break;
  303. case 7:
  304. //top side
  305. neighbours["top"] = false;
  306. break;
  307. case 0:
  308. //internal
  309. break;
  310. default:
  311. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y<<endl;
  312. }
  313. }
  314.  
  315.  
  316.  
  317. if (neighbours["left"]){
  318. left_neighbour = Get(glob_x - 1, glob_y);
  319. }
  320.  
  321. if (neighbours["right"]){
  322. right_neighbour = Get(glob_x + 1, glob_y);
  323. }
  324.  
  325. if (neighbours["bottom"]){
  326. bottom_neighbour = Get(glob_x, glob_y - 1);
  327. }
  328.  
  329. if (neighbours["top"]){
  330. top_neighbour = Get(glob_x, glob_y + 1);
  331. }
  332.  
  333. this_node = Get(glob_x, glob_y);
  334.  
  335. result = left_coeff * left_neighbour +
  336. right_coeff * right_neighbour +
  337. bottom_coeff * bottom_neighbour +
  338. top_coeff * top_neighbour +
  339. this_coeff * this_node;
  340.  
  341. return result;
  342. }
  343.  
  344. double GetNodeFromB(int current_node_global_offset_x, int current_node_global_offset_y) {
  345.  
  346. int glob_x = current_node_global_offset_x;
  347. int glob_y = current_node_global_offset_y;
  348.  
  349. double result = 0.0;
  350.  
  351. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  352. case 2:
  353. //left bottom corner
  354. result = F(x_left, y_bottom) + 2.0 / h_x * psi_L(y_bottom) + 2.0 / h_y * psi_B(x_left);
  355. break;
  356. case 4:
  357. //left top corner
  358. result = F(x_left, y_top) + 2.0 / h_x * psi_L(y_top) + 2.0 / h_y * psi_T(x_left);
  359. break;
  360. case 6:
  361. //right bottom corner
  362. result = F(x_right, y_bottom) + 2.0 / h_x * psi_R(y_bottom) + 2.0 / h_y * psi_B(x_right);
  363. break;
  364. case 8:
  365. //right top corner
  366. result = F(x_right, y_top) + 2.0 / h_x * psi_R(y_top) + 2.0 / h_y * psi_T(x_right);
  367. break;
  368. case 1:
  369. //left side
  370. result = F(x_left, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_L(y_bottom + glob_y * h_y);
  371. break;
  372. case 3:
  373. //right side
  374. result = F(x_right, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_R(y_bottom + glob_y * h_y);
  375. break;
  376. case 5:
  377. //bottom side
  378. result = F(x_left + glob_x * h_x, y_bottom) + 2.0 / h_y * psi_B(x_left + glob_x * h_x);
  379. break;
  380. case 7:
  381. //top side
  382. result = F(x_left + glob_x * h_x, y_top) + 2.0 / h_y * psi_T(x_left + glob_x * h_x);
  383. break;
  384. case 0:
  385. //internal
  386. result = F(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  387. break;
  388. default:
  389. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y <<endl;
  390.  
  391. }
  392.  
  393. return result;
  394.  
  395. }
  396.  
  397. double GetNodeFromExact(int current_node_global_offset_x, int current_node_global_offset_y) {
  398.  
  399. int glob_x = current_node_global_offset_x;
  400. int glob_y = current_node_global_offset_y;
  401.  
  402. return u(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  403.  
  404. }
  405.  
  406. void ComputeMatrixR(){
  407. /*if(my_rank == 0)
  408. cout << "[INFO]: Computation of matrix r started"<<endl;
  409. */
  410.  
  411. vector<double> r_tmp_matrix (cur_block_size_x*cur_block_size_y, 0.0);
  412.  
  413. #pragma omp parallel for
  414. for(int i = 0; i < cur_block_size_x; ++i){
  415. for(int j = 0; j < cur_block_size_y; ++j){
  416.  
  417. int current_node_global_offset_x = GetGlobalX(i),
  418. current_node_global_offset_y = GetGlobalY(j);
  419.  
  420. int glob_x = current_node_global_offset_x,
  421. glob_y = current_node_global_offset_y;
  422.  
  423. r_tmp_matrix [ j + cur_block_size_y*i ] = ComputeMagicInnerProductA_iw(i,j) - GetNodeFromB(glob_x, glob_y);
  424.  
  425. }
  426. }
  427.  
  428. #pragma omp parallel for
  429. for(int i = 0; i < cur_block_size_x; ++i){
  430. for(int j = 0; j < cur_block_size_y; ++j){
  431.  
  432. old_internal_data[ j + cur_block_size_y*i ] = internal_data[ j + cur_block_size_y*i];
  433. internal_data[ j + cur_block_size_y*i ] = r_tmp_matrix[ j + cur_block_size_y*i];
  434.  
  435. }
  436. }
  437.  
  438. //old_internal_data = internal_data;
  439. //internal_data = r_tmp_matrix;
  440.  
  441. SyncMPI();
  442.  
  443. }
  444.  
  445. double ComputeTauAndStopCase(bool &should_i_stop){
  446.  
  447. double local_Ar_r_inner_product_sum = 0.0;
  448. double local_Ar_Ar_inner_product_sum = 0.0;
  449. double global_Ar_r_inner_product_sum = 0.0;
  450. double global_Ar_Ar_inner_product_sum = 0.0;
  451. double local_r_norm = 0.0;
  452. double global_r_norm = 0.0;
  453.  
  454. vector <double> tmp_Ar_i_js(cur_block_size_x * cur_block_size_y, 0.0);
  455. vector <double> tmp_elements(cur_block_size_x * cur_block_size_y, 0.0);
  456.  
  457. #pragma omp parallel for
  458. for(int i = 0; i < cur_block_size_x; ++i){
  459. for(int j = 0; j < cur_block_size_y; ++j){
  460.  
  461. int current_node_global_offset_x = GetGlobalX(i),
  462. current_node_global_offset_y = GetGlobalY(j);
  463.  
  464. int glob_x = current_node_global_offset_x,
  465. glob_y = current_node_global_offset_y;
  466.  
  467. double tmp_Ar_i_j = ComputeMagicInnerProductA_iw(i, j);
  468.  
  469. double tmp_elem = Get(glob_x, glob_y);
  470.  
  471. tmp_elements [j + cur_block_size_y*i] = tmp_elem;
  472.  
  473. tmp_Ar_i_js [j + cur_block_size_y*i] = tmp_Ar_i_j;
  474. }
  475. }
  476.  
  477.  
  478.  
  479.  
  480. for (int i = 0; i < cur_block_size_x; ++i){
  481. for (int j = 0; j < cur_block_size_y; ++j){
  482.  
  483. double rho = 1.0;
  484.  
  485. int current_node_global_offset_x = GetGlobalX(i),
  486. current_node_global_offset_y = GetGlobalY(j);
  487.  
  488. int glob_x = current_node_global_offset_x,
  489. glob_y = current_node_global_offset_y;
  490.  
  491.  
  492. if ((i > 0) && (i < cur_block_size_x - 1) && (j > 0) && (j < cur_block_size_y - 1))
  493. rho = 1.0;
  494. else{
  495. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  496. case 2:
  497. case 4:
  498. case 6:
  499. case 8:
  500. //angle
  501. rho = 0.25;
  502. break;
  503. case 1:
  504. case 3:
  505. case 5:
  506. case 7:
  507. //side
  508. rho = 0.5;
  509. break;
  510. case 0:
  511. //internal
  512. rho = 1.0;
  513. break;
  514. default:
  515. cout << "[ERROR]: Bad global coords compute exact. Local:" << i << " " << j << ". Global:" << glob_x << " " << glob_y <<endl;
  516. }
  517. }
  518.  
  519. double tmp_cur_node_value = tmp_elements [j + cur_block_size_y*i];
  520. double tmp_Ar_i_j = tmp_Ar_i_js [j + cur_block_size_y*i];
  521. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  522. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  523.  
  524.  
  525. local_Ar_r_inner_product_sum += rho * tmp_Ar_i_j * tmp_cur_node_value * h_x*h_y;
  526. local_Ar_Ar_inner_product_sum += rho * pow (tmp_Ar_i_j, 2) * h_x*h_y;
  527. local_r_norm += rho * pow(tmp_cur_node_value, 2) * h_x*h_y;
  528.  
  529. }
  530. }
  531.  
  532.  
  533.  
  534.  
  535. //cout << "[DEBUG]: Local"<< local_Ar_r_inner_product_sum << endl;
  536.  
  537. MPI_Allreduce(&local_Ar_r_inner_product_sum, &global_Ar_r_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  538. comm);
  539.  
  540. //cout << "[DEBUG]: "<< global_Ar_r_inner_product_sum << endl;
  541.  
  542. MPI_Allreduce(&local_Ar_Ar_inner_product_sum, &global_Ar_Ar_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  543. MPI_COMM_WORLD);
  544.  
  545. //cout << "[DEBUG]: "<< global_Ar_Ar_inner_product_sum << endl;
  546.  
  547. double global_tau = global_Ar_r_inner_product_sum/ global_Ar_Ar_inner_product_sum;
  548.  
  549. MPI_Allreduce(&local_r_norm, &global_r_norm, 1, MPI_DOUBLE, MPI_SUM,
  550. MPI_COMM_WORLD);
  551.  
  552. double eps = 1e-06;
  553.  
  554. if (global_r_norm < 0){
  555. cout << "[ERROR]: bad global r norm" << endl;
  556. }
  557.  
  558. current_norm = fabs(global_tau)*sqrt(global_r_norm);
  559.  
  560. //if (my_rank == 0)
  561. // cout << "[DEBUG]: solution norm "<< tmp_norm << endl;
  562.  
  563. if (current_norm <= eps){
  564. if (my_rank == 0)
  565. cout << "[INFO]: r norm is " << sqrt(global_r_norm) << endl;
  566. should_i_stop = true;
  567. }else{
  568. should_i_stop = false;
  569. }
  570.  
  571. return global_tau;
  572.  
  573. }
  574.  
  575.  
  576. void ComputeNewW(double tau){
  577. #pragma omp parallel for
  578. for(int i = 0; i < cur_block_size_x; ++i){
  579. for(int j = 0; j < cur_block_size_y; ++j){
  580. internal_data[ j + cur_block_size_y*i ] = old_internal_data[ j + cur_block_size_y*i ] - tau * internal_data[ j + cur_block_size_y*i ];
  581. //old_internal_data[ j + cur_block_size_y*i ] = 0.0;
  582. }
  583. }
  584. }
  585.  
  586. int GetGlobalX(int i){
  587. return cur_block_global_offset_x + i;
  588. }
  589.  
  590. int GetGlobalY(int j){
  591. return cur_block_global_offset_y + j;
  592. }
  593.  
  594. public:
  595. MPIComputations(int inpM, int inpN, int inpGx, int inpGy, int inpx_left, int inpx_right, int inpy_bottom, int inpy_top, int inp_cur_block_global_coors_x, int inp_cur_block_global_coors_y, int inprank, MPI_Comm inpcomm){
  596.  
  597. M = inpM;
  598. N = inpN;
  599.  
  600. Gx = inpGx;
  601. Gy = inpGy;
  602.  
  603. x_left = inpx_left;
  604. x_right = inpx_right;
  605. y_bottom = inpy_bottom;
  606. y_top = inpy_top;
  607.  
  608. h_x = double((x_right - x_left)) / M;
  609. h_y = double((y_top - y_bottom)) / N;
  610.  
  611. my_rank = inprank;
  612. comm = inpcomm;
  613.  
  614.  
  615. ofstream myfile;
  616. stringstream ss;
  617. ss<< "./res" << N << Gx*Gy << "_"<< << inp_cur_block_global_coors_x << "_" << inp_cur_block_global_coors_y << ".res";
  618.  
  619. string filename;
  620.  
  621. ss >> filename;
  622.  
  623. my_file.open (filename);
  624.  
  625. cur_block_global_coors_x = inp_cur_block_global_coors_x;
  626. cur_block_global_coors_y = inp_cur_block_global_coors_y;
  627.  
  628. global_block_size_x = (M + 1) / Gx;
  629. global_block_size_y = (N + 1) / Gy;
  630.  
  631. cur_block_size_x = global_block_size_x;
  632. cur_block_size_y = global_block_size_y;
  633.  
  634. cur_block_global_offset_x = global_block_size_x * cur_block_global_coors_x;
  635. cur_block_global_offset_y = global_block_size_y * cur_block_global_coors_y;
  636.  
  637. //if ((cur_block_global_offset_x + global_block_size_x + Gx) > (M + 1)){
  638. if(cur_block_global_coors_x == Gx - 1 ){
  639. cur_block_size_x += (M + 1) % Gx;
  640. }
  641.  
  642. //if ((cur_block_global_offset_y + global_block_size_y + Gy) > (N + 1)){
  643. if(cur_block_global_coors_y == Gy - 1){
  644. cur_block_size_y += (N + 1) % Gy;
  645. }
  646.  
  647. internal_data.resize(cur_block_size_x * cur_block_size_y);
  648. old_internal_data.resize(cur_block_size_x * cur_block_size_y);
  649.  
  650.  
  651.  
  652. vector <vector< double> > tmp_coefs(cur_block_size_x * cur_block_size_y, vector<double>(5, 0.0));
  653.  
  654. coefs = tmp_coefs;
  655.  
  656. for(int i = 0; i < cur_block_size_x; i++){
  657. for (int j = 0; j < cur_block_size_y; j++){
  658. vector <double> &my_coefs = coefs[j + cur_block_size_y*i];
  659. ComputeCoefsForNode(my_coefs[0], my_coefs[1], my_coefs[2], my_coefs[3], my_coefs[4], i, j);
  660. }
  661. }
  662.  
  663. //OX
  664. external_data[0].resize(cur_block_size_y);
  665. external_data[1].resize(cur_block_size_y);
  666.  
  667. //OY
  668. external_data[2].resize(cur_block_size_x);
  669. external_data[3].resize(cur_block_size_x);
  670.  
  671. }
  672.  
  673. double Get(int i, int j) {
  674. return GetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y);
  675. }
  676.  
  677. void Set(int i, int j, double v) {
  678. return SetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y, v);
  679. }
  680.  
  681. void SyncMPI(){
  682.  
  683. /*if(my_rank == 0)
  684. cout << "[INFO]: Sync started"<< endl;
  685. */
  686.  
  687. //MPI_Barrier(MPI_COMM_WORLD);
  688.  
  689. //left and right sides
  690. #pragma omp parallel for
  691. for(int j = 0; j < cur_block_size_y; ++j){
  692.  
  693. external_data[ 0 ][ j ] = GetLocalIndex(0, j);//internal_data[ j ];
  694. external_data[ 1 ][ j ] = GetLocalIndex(cur_block_size_x - 1, j);//internal_data[ j + cur_block_size_y * (cur_block_size_x - 1) ];
  695.  
  696. }
  697.  
  698. //bottom and top sides
  699. #pragma omp parallel for
  700. for(int i = 0; i < cur_block_size_x; ++i){
  701.  
  702. external_data[ 2 ][ i ] = GetLocalIndex(i, 0);//internal_data[ cur_block_size_y*i ];
  703. external_data[ 3 ][ i ] = GetLocalIndex(i, cur_block_size_y - 1); //internal_data[ (cur_block_size_y - 1) + cur_block_size_y*i ];
  704.  
  705. }
  706.  
  707. int my_coords[2];
  708. int targets_ranks[4];
  709.  
  710. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  711.  
  712. int neighbour_offsets[ 4 ][ 2 ] = {
  713. { -1, 0 },{ 1, 0 },
  714. { 0, -1 },{ 0, 1 }
  715. };
  716.  
  717. #pragma omp parallel for
  718. for(int i = 0; i < 4; i++){
  719.  
  720. int target_coords[2];
  721.  
  722. target_coords[ 0 ] = my_coords[ 0 ] + neighbour_offsets[ i ][ 0 ];
  723. target_coords[ 1 ] = my_coords[ 1 ] + neighbour_offsets[ i ][ 1 ];
  724.  
  725. if (target_coords[0] >= 0 && target_coords[0] < Gx && target_coords[1] >= 0 && target_coords[1] < Gy){
  726.  
  727. MPI_Cart_rank(comm, target_coords, &targets_ranks[ i ]);
  728.  
  729. }
  730. else{
  731. targets_ranks[i] = -1;
  732. }
  733. }
  734.  
  735. //Now we have rank for all targets
  736.  
  737. for(int axis = 0; axis < 2; axis++){
  738.  
  739. int parity_bit = (my_coords[ axis ]) % 2;
  740.  
  741. //if parity_bit == 0 then
  742. // zuerst mit links, dann mit rechts tauschen
  743. //elif parity_bit == 1:
  744. // zuerst mit rechts, dann mit links tauschen
  745.  
  746. for(int tmp = 0; tmp < 2; tmp++){
  747. parity_bit = 1 - parity_bit;
  748.  
  749. //target id in external_data und targets_ranks
  750. int target_idx = 2 * axis + (1 - parity_bit);
  751.  
  752. if (targets_ranks[target_idx] != -1){
  753.  
  754. int send_tag = 100000 + my_rank * 100 + axis * 10 + parity_bit;
  755. int recv_tag = 100000 + targets_ranks[ target_idx ] * 100 + axis * 10 + (1-parity_bit);
  756.  
  757. MPI_Status tmp_status;
  758.  
  759. if(my_rank != targets_ranks[ target_idx ]){
  760.  
  761. MPI_Sendrecv_replace(&external_data[ target_idx ][ 0 ], external_data[ target_idx ].size(),
  762. MPI_DOUBLE, targets_ranks[ target_idx ], send_tag, targets_ranks[ target_idx ], recv_tag,
  763. comm, &tmp_status);
  764.  
  765. }
  766. }
  767. }
  768. }
  769.  
  770. //MPI_Barrier(MPI_COMM_WORLD);
  771.  
  772. }
  773.  
  774. void DoIteration(bool &should_i_stop){
  775.  
  776. ComputeMatrixR();
  777.  
  778. //Now R Matrix is in internal_data
  779.  
  780. double tau = ComputeTauAndStopCase(should_i_stop);
  781.  
  782. //We have in our block jetzt:
  783. //in internal_data: R Matrix
  784. //in old_internal_data: w aus letzte iteration
  785. //and we have tau
  786. //jetzt koennen wir naechste w finden
  787.  
  788. ComputeNewW(tau);
  789.  
  790. SyncMPI();
  791.  
  792. }
  793.  
  794. double GetLocalIndex(int i, int j){
  795. //internal data
  796. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  797. return internal_data[ j + cur_block_size_y*i ];
  798. }
  799.  
  800. //external data
  801. //OX
  802. if((j >= 0) && (j < cur_block_size_y)){
  803.  
  804. if (i == -1)
  805. return external_data[ 0 ][ j ];
  806.  
  807. if (i == cur_block_size_x)
  808. return external_data[ 1 ][ j ];
  809.  
  810. }
  811.  
  812. //OY
  813. if((i >= 0) && (i < cur_block_size_x)){
  814.  
  815. if (j == -1)
  816. return external_data[ 2 ][ i ];
  817. if (j == cur_block_size_y)
  818. return external_data[ 3 ][ i ];
  819.  
  820. }
  821.  
  822. cout << "[ERROR]: bad local index" << endl;
  823.  
  824. return nan("");
  825. }
  826.  
  827. void SetLocalIndex(int i, int j, double v){
  828. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  829. internal_data[ j + cur_block_size_y*i ] = v;
  830. }else{
  831. cout << "[ERROR]: trying to set data outside the local area" << endl;
  832. }
  833. }
  834.  
  835. double CompareWithExact() {
  836.  
  837. double local_diff_norm = 0.0;
  838. double global_diff_norm = 0.0;
  839. /*
  840. if (my_rank == 0)
  841. cout << "[INFO]: Starting computing compare with exact" << endl;
  842. */
  843. vector <double> tmp_elements(cur_block_size_x * cur_block_size_y, 0.0);
  844.  
  845. #pragma omp parallel for
  846. for (int i = 0; i < cur_block_size_x; ++i){
  847. for (int j = 0; j < cur_block_size_y; ++j){
  848.  
  849.  
  850. int current_node_global_offset_x = GetGlobalX(i),
  851. current_node_global_offset_y = GetGlobalY(j);
  852.  
  853. int glob_x = current_node_global_offset_x,
  854. glob_y = current_node_global_offset_y;
  855.  
  856. double tmp_elem = Get(glob_x, glob_y) - GetNodeFromExact(glob_x, glob_y);
  857.  
  858. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  859.  
  860. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  861.  
  862. tmp_elements [j + cur_block_size_y*i] = tmp_elem;
  863.  
  864. }
  865. }
  866.  
  867. //cout << "[INFO]: local max diff in " << cur_block_global_offset_x << " " << cur_block_global_offset_y << " " << local_diff_norm << endl;
  868.  
  869.  
  870. for (int i = 0; i < cur_block_size_x; ++i){
  871. for (int j = 0; j < cur_block_size_y; ++j){
  872.  
  873. double rho = 1.0;
  874.  
  875. int current_node_global_offset_x = GetGlobalX(i),
  876. current_node_global_offset_y = GetGlobalY(j);
  877.  
  878. int glob_x = current_node_global_offset_x,
  879. glob_y = current_node_global_offset_y;
  880.  
  881.  
  882. if ((i > 0) && (i < cur_block_size_x - 1) && (j > 0) && (j < cur_block_size_y - 1))
  883. rho = 1.0;
  884. else{
  885. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  886. case 2:
  887. case 4:
  888. case 6:
  889. case 8:
  890. //angle
  891. rho = 0.25;
  892. break;
  893. case 1:
  894. case 3:
  895. case 5:
  896. case 7:
  897. //side
  898. rho = 0.5;
  899. break;
  900. case 0:
  901. //internal
  902. rho = 1.0;
  903. break;
  904. default:
  905. cout << "[ERROR]: Bad global coords compute exact. Local:" << i << " " << j << ". Global:" << glob_x << " " << glob_y <<endl;
  906. }
  907. }
  908.  
  909. double tmp_elem = tmp_elements [j + cur_block_size_y*i];
  910.  
  911. local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  912. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  913. }
  914. }
  915.  
  916. MPI_Allreduce(&local_diff_norm, &global_diff_norm, 1, MPI_DOUBLE, MPI_MAX,
  917. MPI_COMM_WORLD);
  918.  
  919. //global_diff_norm = sqrt(global_diff_norm);
  920.  
  921.  
  922.  
  923. for (int i = 0; i < cur_block_size_x; ++i){
  924. for (int j = 0; j < cur_block_size_y; ++j){
  925.  
  926. int current_node_global_offset_x = GetGlobalX(i),
  927. current_node_global_offset_y = GetGlobalY(j);
  928.  
  929. int glob_x = current_node_global_offset_x,
  930. glob_y = current_node_global_offset_y;
  931.  
  932. my_file << glob_x << "," << glob_y << "," << Get(glob_x, glob_y) << "," << GetNodeFromExact(glob_x, glob_y) << endl;
  933. }
  934. }
  935.  
  936.  
  937. return global_diff_norm;
  938. }
  939.  
  940. double GetCurrentNorm(){
  941. return current_norm;
  942. }
  943. };
  944.  
  945.  
  946.  
  947.  
  948. int main(int argc, char* argv[]){
  949.  
  950.  
  951. const double x_left = -2, x_right = 3;
  952. const double y_bottom = -1, y_top = 4;
  953.  
  954. double time_start, time_stop;
  955. int N, Gx, Gy;
  956. int dim[2], period[2];
  957.  
  958. MPI_Comm comm;
  959.  
  960. //N - global grid size
  961. N = atoi(argv[1]);
  962.  
  963. //Gx
  964. Gx = dim[0] = atoi(argv[2]);
  965.  
  966. //Gy
  967. Gy = dim[1] = atoi(argv[3]);
  968.  
  969.  
  970. stringstream ss;
  971. ss<< "./logs/out" << N << Gx*Gy << ".res";
  972.  
  973. string filename;
  974.  
  975. ss >> filename;
  976.  
  977. ofstream out(filename.c_str());
  978. streambuf *coutbuf = cout.rdbuf(); //save old buf
  979. cout.rdbuf(out.rdbuf());
  980.  
  981.  
  982. stringstream sserr;
  983. sserr<< "./logs/out" << N << Gx*Gy << ".err";
  984.  
  985. string filenameerr;
  986.  
  987. sserr >> filenameerr;
  988.  
  989. ofstream err(filenameerr.c_str());
  990. streambuf *cerrbuf = cerr.rdbuf(); //save old buf
  991. cerr.rdbuf(err.rdbuf());
  992.  
  993. period[0]=0;
  994. period[1]=0;
  995.  
  996.  
  997. MPI_Init(&argc, &argv);
  998.  
  999. time_start = MPI_Wtime();
  1000.  
  1001. int world_size;
  1002.  
  1003. int my_rank;
  1004.  
  1005. MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  1006. MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  1007.  
  1008.  
  1009. if(my_rank == 0){
  1010. cout << "[INFO]: N = " << N << endl;
  1011. cout << "[INFO]: Gx = " << dim[0] << endl;
  1012. cout << "[INFO]: Gy = " << dim[1] << endl;
  1013. }
  1014.  
  1015.  
  1016.  
  1017. if(argc <= 3){
  1018. if(my_rank == 0)
  1019. cout << "[ERROR]: Usage: mpieval <N> <Gx> <Gy>" << endl;
  1020.  
  1021. MPI_Abort(MPI_COMM_WORLD, 1);
  1022. }
  1023.  
  1024. if(Gx * Gy != world_size){
  1025. if(my_rank == 0)
  1026. cout << "[ERROR]: mpi world size is not equal to "<< Gx << "*" << Gy << endl;
  1027.  
  1028. MPI_Abort(MPI_COMM_WORLD, 1);
  1029. }
  1030.  
  1031. MPI_Cart_create(MPI_COMM_WORLD, 2, dim, period, 1, &comm);
  1032.  
  1033. /*if(my_rank == 0)
  1034. cout << "[INFO]: Cart created"<<endl;
  1035. */
  1036.  
  1037. MPI_Comm_rank(comm, &my_rank);
  1038.  
  1039. int my_coords[2];
  1040.  
  1041. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  1042.  
  1043.  
  1044.  
  1045.  
  1046. class MPIComputations w_func(N, N, Gx, Gy, x_left, x_right, y_bottom, y_top, my_coords[0], my_coords[1], my_rank, comm);
  1047.  
  1048.  
  1049. int iteration_num = 0;
  1050.  
  1051. bool should_i_stop = false;
  1052.  
  1053. while ( should_i_stop != true ){
  1054.  
  1055. w_func.DoIteration(should_i_stop);
  1056.  
  1057. if ( (my_rank == 0) && (iteration_num % 10000 == 0)){
  1058. cout << "[INFO]: Iteration " << iteration_num << " difference with previous is: "<< w_func.GetCurrentNorm() << endl;
  1059. }
  1060.  
  1061. iteration_num++;
  1062. }
  1063.  
  1064. //MPI_Barrier(MPI_COMM_WORLD);
  1065.  
  1066. //vector <double> local_elements;
  1067.  
  1068. double comparing = w_func.CompareWithExact();
  1069.  
  1070. if (my_rank == 0)
  1071. cout << "[INFO]: Diff with exact solution: " << comparing << endl;
  1072.  
  1073.  
  1074.  
  1075. //myfile.close();
  1076.  
  1077. time_stop = MPI_Wtime();
  1078. if( my_rank == 0 )
  1079. cout << "Finished!" << endl
  1080. << "Total iterations: " << iteration_num << endl
  1081. << "Elapsed time: " << (time_stop - time_start) << endl
  1082. << "Time per iteration: " << (time_stop - time_start) / double(iteration_num) << endl;
  1083.  
  1084. MPI_Finalize();
  1085.  
  1086. cout.rdbuf(coutbuf); //reset to standard input again
  1087. cerr.rdbuf(cerrbuf); //reset to standard output again
  1088.  
  1089.  
  1090. return 0;
  1091. }
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!