API tools faq
paste
Advertisement
manac68974

Untitled

manac68974
Dec 18th, 2019
102
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 29.92 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.  
  28. vector<double> internal_data;
  29. vector<double> old_internal_data;
  30. vector<double> external_data[4];
  31.  
  32. vector <vector <double>> coefs;
  33.  
  34. int IsNodeInternalCornerOrSide(int current_node_global_offset_x, int current_node_global_offset_y){
  35.  
  36. //corners
  37. //left bottom corner
  38. if (current_node_global_offset_x == 0 && current_node_global_offset_y == 0){
  39. return 2;
  40. }
  41.  
  42. //left top corner
  43. if (current_node_global_offset_x == 0 && current_node_global_offset_y == N){
  44. return 4;
  45. }
  46.  
  47. //right bottom corner
  48. if (current_node_global_offset_x == M && current_node_global_offset_y == 0){
  49. return 6;
  50. }
  51.  
  52. //right top corner
  53. if (current_node_global_offset_x == M && current_node_global_offset_y == N){
  54. return 8;
  55. }
  56.  
  57. //sides
  58. //left side
  59. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  60. current_node_global_offset_x == 0){
  61. return 1;
  62. }
  63.  
  64. //right side
  65. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  66. current_node_global_offset_x == M){
  67. return 3;
  68. }
  69.  
  70. //bottom side
  71. if (current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  72. current_node_global_offset_y == 0){
  73. return 5;
  74. }
  75.  
  76. //top side
  77. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  78. current_node_global_offset_y == N)){
  79. return 7;
  80. }
  81.  
  82. //internal
  83. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1) &&
  84. (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1)){
  85. return 0;
  86. }
  87.  
  88. return -1;
  89. }
  90.  
  91. double k(double x, double y) {
  92. return 1 + pow(x + y, 2);
  93. }
  94.  
  95. double q(double x, double y) {
  96. return 1;
  97. }
  98.  
  99. double u(double x, double y) {
  100. return 2.0 / (1 + pow(x, 2) + pow(y, 2));
  101. }
  102.  
  103. // psi_R(x, y) = k(A2, y) * du/dx(A2, y)
  104. double psi_R(double y) {
  105. return (-12) * (pow((y + 3), 2) + 1) / pow((pow(y, 2) + 10), 2);
  106. }
  107.  
  108. // psi_L(x, y) = -k(A1, y) * du/dx(A1, y)
  109. double psi_L(double y) {
  110. return (-8) * (pow((y - 2), 2) + 1) / pow((pow(y, 2) + 5), 2);
  111. }
  112.  
  113. // psi_T(x, y) = k(x, B2) * du/dy(x, B2)
  114. double psi_T(double x) {
  115. return (-16) * (pow((x + 4), 2) + 1) / pow((pow(x, 2) + 17), 2);
  116. }
  117.  
  118. // psi_B(x, y) = -k(x, B1) * du/dy(x, B1)
  119. double psi_B(double x) {
  120. return (-4) * (pow((x - 1), 2) + 1) / pow((pow(x, 2) + 2), 2);
  121. }
  122.  
  123. // right-part function of Poisson equation
  124. double F(double x, double y) {
  125. return 2 * (pow(x,4) + pow(y,4) + 2 * (pow(x,2) + 3) * pow(y,2) + 6 * pow(x,2) + 16*x*y + 5)
  126. / pow((1 + pow(x, 2) + pow(y, 2)), 3);
  127. }
  128.  
  129. //inner_product(A[i], internal_data)
  130.  
  131. //inner_product(A[i], internal_data)
  132. void ComputeCoefsForNode (double &left_coeff, double &right_coeff, double &bottom_coeff, double &top_coeff, double &this_coeff, int i, int j){
  133.  
  134. int glob_x = GetGlobalX(i);
  135. int glob_y = GetGlobalY(j);
  136.  
  137. map <string,bool> neighbours = {
  138. {"left", true},
  139. {"right", true},
  140. {"bottom", true},
  141. {"top", true}
  142. };
  143.  
  144. left_coeff = 1.0;
  145. right_coeff = 1.0;
  146. bottom_coeff = 1.0;
  147. top_coeff = 1.0;
  148. this_coeff = 1.0;
  149.  
  150. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  151. case 2:
  152. //left bottom corner
  153. neighbours["left"] = false;
  154. neighbours["bottom"] = false;
  155. break;
  156. case 4:
  157. //left top corner
  158. neighbours["left"] = false;
  159. neighbours["top"] = false;
  160. break;
  161. case 6:
  162. //right bottom corner
  163. neighbours["right"] = false;
  164. neighbours["bottom"] = false;
  165. break;
  166. case 8:
  167. //right top corner
  168. neighbours["right"] = false;
  169. neighbours["top"] = false;
  170. break;
  171. case 1:
  172. //left side
  173. neighbours["left"] = false;
  174. break;
  175. case 3:
  176. //right side
  177. neighbours["right"] = false;
  178. break;
  179. case 5:
  180. //bottom side
  181. neighbours["bottom"] = false;
  182. break;
  183. case 7:
  184. //top side
  185. neighbours["top"] = false;
  186. break;
  187. case 0:
  188. //internal
  189. break;
  190. default:
  191. cout << "[ERROR]: Bad global coords compute coef. Global:" << glob_x << " " << glob_y<<endl;
  192. }
  193.  
  194. if (!neighbours["left"]){
  195. right_coeff = 2.0;
  196. left_coeff = 0.0;
  197. }
  198.  
  199. if (!neighbours["right"]){
  200. left_coeff = 2.0;
  201. right_coeff = 0.0;
  202. }
  203.  
  204. if (!neighbours["bottom"]){
  205. top_coeff = 2.0;
  206. bottom_coeff = 0.0;
  207. }
  208.  
  209. if (!neighbours["top"]){
  210. bottom_coeff = 2.0;
  211. top_coeff = 0.0;
  212. }
  213.  
  214. if (neighbours["left"]){
  215. left_coeff *= -k(x_left + (glob_x - 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  216. }
  217.  
  218. if (neighbours["right"]){
  219. right_coeff *= -k(x_left + (glob_x + 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  220. }
  221.  
  222. if (neighbours["bottom"]){
  223. bottom_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y - 0.5) * h_y) / pow(h_y, 2);
  224. }
  225.  
  226. if (neighbours["top"]){
  227. top_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y + 0.5) * h_y) / pow(h_y, 2);
  228. }
  229.  
  230. this_coeff = q(x_left + glob_x * h_x, y_bottom + glob_y * h_y) - left_coeff - right_coeff - bottom_coeff - top_coeff;
  231.  
  232. /*coefs[0] = left_coeff;
  233. coefs[1] = right_coeff;
  234. coefs[2] = bottom_coeff;
  235. coefs[3] = top_coeff;
  236. coefs[4] = this_coeff;*/
  237. }
  238.  
  239. double ComputeMagicInnerProductA_iw (int i, int j){
  240.  
  241. int glob_x = GetGlobalX(i);
  242. int glob_y = GetGlobalY(j);
  243.  
  244. double result = 0.0;
  245.  
  246. map <string,bool> neighbours = {
  247. {"left", true},
  248. {"right", true},
  249. {"bottom", true},
  250. {"top", true}
  251. };
  252.  
  253. vector <double> my_coefs = coefs[j + cur_block_size_y*i];
  254.  
  255. double left_neighbour = 0.0, right_neighbour = 0.0, bottom_neighbour = 0.0, top_neighbour = 0.0, this_node = 0.0;
  256. 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];
  257.  
  258. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  259. case 2:
  260. //left bottom corner
  261. neighbours["left"] = false;
  262. neighbours["bottom"] = false;
  263. break;
  264. case 4:
  265. //left top corner
  266. neighbours["left"] = false;
  267. neighbours["top"] = false;
  268. break;
  269. case 6:
  270. //right bottom corner
  271. neighbours["right"] = false;
  272. neighbours["bottom"] = false;
  273. break;
  274. case 8:
  275. //right top corner
  276. neighbours["right"] = false;
  277. neighbours["top"] = false;
  278. break;
  279. case 1:
  280. //left side
  281. neighbours["left"] = false;
  282. break;
  283. case 3:
  284. //right side
  285. neighbours["right"] = false;
  286. break;
  287. case 5:
  288. //bottom side
  289. neighbours["bottom"] = false;
  290. break;
  291. case 7:
  292. //top side
  293. neighbours["top"] = false;
  294. break;
  295. case 0:
  296. //internal
  297. break;
  298. default:
  299. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y<<endl;
  300. }
  301.  
  302.  
  303.  
  304. if (neighbours["left"]){
  305. left_neighbour = Get(glob_x - 1, glob_y);
  306. }
  307.  
  308. if (neighbours["right"]){
  309. right_neighbour = Get(glob_x + 1, glob_y);
  310. }
  311.  
  312. if (neighbours["bottom"]){
  313. bottom_neighbour = Get(glob_x, glob_y - 1);
  314. }
  315.  
  316. if (neighbours["top"]){
  317. top_neighbour = Get(glob_x, glob_y + 1);
  318. }
  319.  
  320. this_node = Get(glob_x, glob_y);
  321.  
  322. result = left_coeff * left_neighbour +
  323. right_coeff * right_neighbour +
  324. bottom_coeff * bottom_neighbour +
  325. top_coeff * top_neighbour +
  326. this_coeff * this_node;
  327.  
  328. return result;
  329. }
  330.  
  331. double GetNodeFromB(int current_node_global_offset_x, int current_node_global_offset_y) {
  332.  
  333. int glob_x = current_node_global_offset_x;
  334. int glob_y = current_node_global_offset_y;
  335.  
  336. double result = 0.0;
  337.  
  338. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  339. case 2:
  340. //left bottom corner
  341. result = F(x_left, y_bottom) + 2.0 / h_x * psi_L(y_bottom) + 2.0 / h_y * psi_B(x_left);
  342. break;
  343. case 4:
  344. //left top corner
  345. result = F(x_left, y_top) + 2.0 / h_x * psi_L(y_top) + 2.0 / h_y * psi_T(x_left);
  346. break;
  347. case 6:
  348. //right bottom corner
  349. result = F(x_right, y_bottom) + 2.0 / h_x * psi_R(y_bottom) + 2.0 / h_y * psi_B(x_right);
  350. break;
  351. case 8:
  352. //right top corner
  353. result = F(x_right, y_top) + 2.0 / h_x * psi_R(y_top) + 2.0 / h_y * psi_T(x_right);
  354. break;
  355. case 1:
  356. //left side
  357. result = F(x_left, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_L(y_bottom + glob_y * h_y);
  358. break;
  359. case 3:
  360. //right side
  361. result = F(x_right, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_R(y_bottom + glob_y * h_y);
  362. break;
  363. case 5:
  364. //bottom side
  365. result = F(x_left + glob_x * h_x, y_bottom) + 2.0 / h_y * psi_B(x_left + glob_x * h_x);
  366. break;
  367. case 7:
  368. //top side
  369. result = F(x_left + glob_x * h_x, y_top) + 2.0 / h_y * psi_T(x_left + glob_x * h_x);
  370. break;
  371. case 0:
  372. //internal
  373. result = F(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  374. break;
  375. default:
  376. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y <<endl;
  377.  
  378. }
  379.  
  380. return result;
  381.  
  382. }
  383.  
  384. double GetNodeFromExact(int current_node_global_offset_x, int current_node_global_offset_y) {
  385.  
  386. int glob_x = current_node_global_offset_x;
  387. int glob_y = current_node_global_offset_y;
  388.  
  389. return u(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  390.  
  391. }
  392.  
  393. void ComputeMatrixR(){
  394. /*if(my_rank == 0)
  395. cout << "[INFO]: Computation of matrix r started"<<endl;
  396. */
  397.  
  398. vector<double> r_tmp_matrix (cur_block_size_x*cur_block_size_y, 0.0);
  399.  
  400. //#pragma omp parallel for
  401. for(int i = 0; i < cur_block_size_x; ++i){
  402. for(int j = 0; j < cur_block_size_y; ++j){
  403.  
  404. int current_node_global_offset_x = GetGlobalX(i),
  405. current_node_global_offset_y = GetGlobalY(j);
  406.  
  407. int glob_x = current_node_global_offset_x,
  408. glob_y = current_node_global_offset_y;
  409.  
  410. r_tmp_matrix [ j + cur_block_size_y*i ] = ComputeMagicInnerProductA_iw(i,j) - GetNodeFromB(glob_x, glob_y);
  411.  
  412. }
  413. }
  414.  
  415. //#pragma omp parallel for
  416. for(int i = 0; i < cur_block_size_x; ++i){
  417. for(int j = 0; j < cur_block_size_y; ++j){
  418.  
  419. old_internal_data[ j + cur_block_size_y*i ] = internal_data[ j + cur_block_size_y*i];
  420. internal_data[ j + cur_block_size_y*i ] = r_tmp_matrix[ j + cur_block_size_y*i];
  421.  
  422. }
  423. }
  424.  
  425. //old_internal_data = internal_data;
  426. //internal_data = r_tmp_matrix;
  427.  
  428. SyncMPI();
  429.  
  430. }
  431.  
  432. double ComputeTauAndStopCase(bool &should_i_stop){
  433.  
  434. double local_Ar_r_inner_product_sum = 0.0;
  435. double local_Ar_Ar_inner_product_sum = 0.0;
  436. double global_Ar_r_inner_product_sum = 0.0;
  437. double global_Ar_Ar_inner_product_sum = 0.0;
  438. double local_r_norm = 0.0;
  439. double global_r_norm = 0.0;
  440.  
  441. //#pragma omp parallel for
  442. for(int i = 0; i < cur_block_size_x; ++i){
  443. for(int j = 0; j < cur_block_size_y; ++j){
  444. double rho = 1.0;
  445.  
  446. int current_node_global_offset_x = GetGlobalX(i),
  447. current_node_global_offset_y = GetGlobalY(j);
  448.  
  449. int glob_x = current_node_global_offset_x,
  450. glob_y = current_node_global_offset_y;
  451.  
  452. double tmp_Ar_i_j = ComputeMagicInnerProductA_iw(i, j);
  453.  
  454. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  455. case 2:
  456. case 4:
  457. case 6:
  458. case 8:
  459. //angle
  460. rho = 0.25;
  461. break;
  462. case 1:
  463. case 3:
  464. case 5:
  465. case 7:
  466. //side
  467. rho = 0.5;
  468. break;
  469. case 0:
  470. //internal
  471. rho = 1.0;
  472. break;
  473. default:
  474. cout << "[ERROR]: Bad global coords compute tau. Global:" << glob_x << " " << glob_y << endl;
  475. }
  476.  
  477. double tmp_cur_node_value = Get(glob_x, glob_y);
  478.  
  479. local_Ar_r_inner_product_sum += rho * tmp_Ar_i_j * tmp_cur_node_value * h_x*h_y;
  480. local_Ar_Ar_inner_product_sum += rho * pow (tmp_Ar_i_j, 2) * h_x*h_y;
  481. local_r_norm += rho * pow(tmp_cur_node_value, 2) * h_x*h_y;
  482.  
  483. }
  484. }
  485.  
  486. //cout << "[DEBUG]: Local"<< local_Ar_r_inner_product_sum << endl;
  487.  
  488. MPI_Allreduce(&local_Ar_r_inner_product_sum, &global_Ar_r_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  489. comm);
  490.  
  491. //cout << "[DEBUG]: "<< global_Ar_r_inner_product_sum << endl;
  492.  
  493. MPI_Allreduce(&local_Ar_Ar_inner_product_sum, &global_Ar_Ar_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  494. MPI_COMM_WORLD);
  495.  
  496. //cout << "[DEBUG]: "<< global_Ar_Ar_inner_product_sum << endl;
  497.  
  498. double global_tau = global_Ar_r_inner_product_sum/ global_Ar_Ar_inner_product_sum;
  499.  
  500. MPI_Allreduce(&local_r_norm, &global_r_norm, 1, MPI_DOUBLE, MPI_SUM,
  501. MPI_COMM_WORLD);
  502.  
  503. double eps = 1e-06;
  504.  
  505. if (global_r_norm < 0){
  506. cout << "[ERROR]: bad global r norm" << endl;
  507. }
  508.  
  509. current_norm = fabs(global_tau)*sqrt(global_r_norm);
  510.  
  511. //if (my_rank == 0)
  512. // cout << "[DEBUG]: solution norm "<< tmp_norm << endl;
  513.  
  514. if (current_norm <= eps){
  515. if (my_rank == 0)
  516. cout << "[INFO]: r norm is " << sqrt(global_r_norm) << endl;
  517. should_i_stop = true;
  518. }else{
  519. should_i_stop = false;
  520. }
  521.  
  522. return global_tau;
  523.  
  524. }
  525.  
  526.  
  527. void ComputeNewW(double tau){
  528. //#pragma omp parallel for
  529. for(int i = 0; i < cur_block_size_x; ++i){
  530. for(int j = 0; j < cur_block_size_y; ++j){
  531. 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 ];
  532. //old_internal_data[ j + cur_block_size_y*i ] = 0.0;
  533. }
  534. }
  535. }
  536.  
  537. int GetGlobalX(int i){
  538. return cur_block_global_offset_x + i;
  539. }
  540.  
  541. int GetGlobalY(int j){
  542. return cur_block_global_offset_y + j;
  543. }
  544.  
  545. public:
  546. 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){
  547.  
  548. M = inpM;
  549. N = inpN;
  550.  
  551. Gx = inpGx;
  552. Gy = inpGy;
  553.  
  554. x_left = inpx_left;
  555. x_right = inpx_right;
  556. y_bottom = inpy_bottom;
  557. y_top = inpy_top;
  558.  
  559. h_x = double((x_right - x_left)) / M;
  560. h_y = double((y_top - y_bottom)) / N;
  561.  
  562. my_rank = inprank;
  563. comm = inpcomm;
  564.  
  565. cur_block_global_coors_x = inp_cur_block_global_coors_x;
  566. cur_block_global_coors_y = inp_cur_block_global_coors_y;
  567.  
  568. global_block_size_x = (M + 1) / Gx;
  569. global_block_size_y = (N + 1) / Gy;
  570.  
  571. cur_block_size_x = global_block_size_x;
  572. cur_block_size_y = global_block_size_y;
  573.  
  574. cur_block_global_offset_x = global_block_size_x * cur_block_global_coors_x;
  575. cur_block_global_offset_y = global_block_size_y * cur_block_global_coors_y;
  576.  
  577. //if ((cur_block_global_offset_x + global_block_size_x + Gx) > (M + 1)){
  578. if(cur_block_global_coors_x == Gx - 1 ){
  579. cur_block_size_x += (M + 1) % Gx;
  580. }
  581.  
  582. //if ((cur_block_global_offset_y + global_block_size_y + Gy) > (N + 1)){
  583. if(cur_block_global_coors_y == Gy - 1){
  584. cur_block_size_y += (N + 1) % Gy;
  585. }
  586.  
  587. internal_data.resize(cur_block_size_x * cur_block_size_y);
  588. old_internal_data.resize(cur_block_size_x * cur_block_size_y);
  589.  
  590.  
  591.  
  592. coefs(cur_block_size_x * cur_block_size_y, vector<double>(5, 0.0));
  593.  
  594. for(int i = 0; i < cur_block_size_x; i++){
  595. for (int j = 0; j < cur_block_size_y; j++){
  596. vector <double> my_coefs = coefs[j + cur_block_size_y*i];
  597. ComputeCoefsForNode(my_coefs[0], my_coefs[1], my_coefs[2], my_coefs[3], my_coefs[4], i, j)
  598. }
  599. }
  600.  
  601. //OX
  602. external_data[0].resize(cur_block_size_y);
  603. external_data[1].resize(cur_block_size_y);
  604.  
  605. //OY
  606. external_data[2].resize(cur_block_size_x);
  607. external_data[3].resize(cur_block_size_x);
  608.  
  609. }
  610.  
  611. double Get(int i, int j) {
  612. return GetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y);
  613. }
  614.  
  615. void Set(int i, int j, double v) {
  616. return SetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y, v);
  617. }
  618.  
  619. void SyncMPI(){
  620.  
  621. /*if(my_rank == 0)
  622. cout << "[INFO]: Sync started"<< endl;
  623. */
  624.  
  625. //MPI_Barrier(MPI_COMM_WORLD);
  626.  
  627. //left and right sides
  628. //#pragma omp parallel for
  629. for(int j = 0; j < cur_block_size_y; ++j){
  630.  
  631. external_data[ 0 ][ j ] = GetLocalIndex(0, j);//internal_data[ j ];
  632. external_data[ 1 ][ j ] = GetLocalIndex(cur_block_size_x - 1, j);//internal_data[ j + cur_block_size_y * (cur_block_size_x - 1) ];
  633.  
  634. }
  635.  
  636. //bottom and top sides
  637. //#pragma omp parallel for
  638. for(int i = 0; i < cur_block_size_x; ++i){
  639.  
  640. external_data[ 2 ][ i ] = GetLocalIndex(i, 0);//internal_data[ cur_block_size_y*i ];
  641. external_data[ 3 ][ i ] = GetLocalIndex(i, cur_block_size_y - 1); //internal_data[ (cur_block_size_y - 1) + cur_block_size_y*i ];
  642.  
  643. }
  644.  
  645. int my_coords[2];
  646. int targets_ranks[4];
  647.  
  648. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  649.  
  650. int neighbour_offsets[ 4 ][ 2 ] = {
  651. { -1, 0 },{ 1, 0 },
  652. { 0, -1 },{ 0, 1 }
  653. };
  654.  
  655. //#pragma omp parallel for
  656. for(int i = 0; i < 4; i++){
  657.  
  658. int target_coords[2];
  659.  
  660. target_coords[ 0 ] = my_coords[ 0 ] + neighbour_offsets[ i ][ 0 ];
  661. target_coords[ 1 ] = my_coords[ 1 ] + neighbour_offsets[ i ][ 1 ];
  662.  
  663. if (target_coords[0] >= 0 && target_coords[0] < Gx && target_coords[1] >= 0 && target_coords[1] < Gy){
  664.  
  665. MPI_Cart_rank(comm, target_coords, &targets_ranks[ i ]);
  666.  
  667. }
  668. else{
  669. targets_ranks[i] = -1;
  670. }
  671. }
  672.  
  673. //Now we have rank for all targets
  674.  
  675. for(int axis = 0; axis < 2; axis++){
  676.  
  677. int parity_bit = (my_coords[ axis ]) % 2;
  678.  
  679. //if parity_bit == 0 then
  680. // zuerst mit links, dann mit rechts tauschen
  681. //elif parity_bit == 1:
  682. // zuerst mit rechts, dann mit links tauschen
  683.  
  684. for(int tmp = 0; tmp < 2; tmp++){
  685. parity_bit = 1 - parity_bit;
  686.  
  687. //target id in external_data und targets_ranks
  688. int target_idx = 2 * axis + (1 - parity_bit);
  689.  
  690. if (targets_ranks[target_idx] != -1){
  691.  
  692. int send_tag = 100000 + my_rank * 100 + axis * 10 + parity_bit;
  693. int recv_tag = 100000 + targets_ranks[ target_idx ] * 100 + axis * 10 + (1-parity_bit);
  694.  
  695. MPI_Status tmp_status;
  696.  
  697. if(my_rank != targets_ranks[ target_idx ]){
  698.  
  699. MPI_Sendrecv_replace(&external_data[ target_idx ][ 0 ], external_data[ target_idx ].size(),
  700. MPI_DOUBLE, targets_ranks[ target_idx ], send_tag, targets_ranks[ target_idx ], recv_tag,
  701. comm, &tmp_status);
  702.  
  703. }
  704. }
  705. }
  706. }
  707.  
  708. //MPI_Barrier(MPI_COMM_WORLD);
  709.  
  710. }
  711.  
  712. void DoIteration(bool &should_i_stop){
  713.  
  714. ComputeMatrixR();
  715.  
  716. //Now R Matrix is in internal_data
  717.  
  718. double tau = ComputeTauAndStopCase(should_i_stop);
  719.  
  720. //We have in our block jetzt:
  721. //in internal_data: R Matrix
  722. //in old_internal_data: w aus letzte iteration
  723. //and we have tau
  724. //jetzt koennen wir naechste w finden
  725.  
  726. ComputeNewW(tau);
  727.  
  728. SyncMPI();
  729.  
  730. }
  731.  
  732. double GetLocalIndex(int i, int j){
  733. //internal data
  734. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  735. return internal_data[ j + cur_block_size_y*i ];
  736. }
  737.  
  738. //external data
  739. //OX
  740. if((j >= 0) && (j < cur_block_size_y)){
  741.  
  742. if (i == -1)
  743. return external_data[ 0 ][ j ];
  744.  
  745. if (i == cur_block_size_x)
  746. return external_data[ 1 ][ j ];
  747.  
  748. }
  749.  
  750. //OY
  751. if((i >= 0) && (i < cur_block_size_x)){
  752.  
  753. if (j == -1)
  754. return external_data[ 2 ][ i ];
  755. if (j == cur_block_size_y)
  756. return external_data[ 3 ][ i ];
  757.  
  758. }
  759.  
  760. cout << "[ERROR]: bad local index" << endl;
  761.  
  762. return nan("");
  763. }
  764.  
  765. void SetLocalIndex(int i, int j, double v){
  766. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  767. internal_data[ j + cur_block_size_y*i ] = v;
  768. }else{
  769. cout << "[ERROR]: trying to set data outside the local area" << endl;
  770. }
  771. }
  772.  
  773. double CompareWithExact() {
  774.  
  775. double local_diff_norm = 0.0;
  776. double global_diff_norm = 0.0;
  777. /*
  778. if (my_rank == 0)
  779. cout << "[INFO]: Starting computing compare with exact" << endl;
  780. */
  781. vector <double> tmp_elements(cur_block_size_x * cur_block_size_y, 0.0);
  782.  
  783. //#pragma omp parallel for
  784. for (int i = 0; i < cur_block_size_x; ++i){
  785. for (int j = 0; j < cur_block_size_y; ++j){
  786.  
  787.  
  788. int current_node_global_offset_x = GetGlobalX(i),
  789. current_node_global_offset_y = GetGlobalY(j);
  790.  
  791. int glob_x = current_node_global_offset_x,
  792. glob_y = current_node_global_offset_y;
  793.  
  794. double tmp_elem = Get(glob_x, glob_y) - GetNodeFromExact(glob_x, glob_y);
  795.  
  796. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  797.  
  798. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  799.  
  800. tmp_elements [j + cur_block_size_y*i] = tmp_elem;
  801.  
  802. }
  803. }
  804.  
  805. //cout << "[INFO]: local max diff in " << cur_block_global_offset_x << " " << cur_block_global_offset_y << " " << local_diff_norm << endl;
  806.  
  807.  
  808. for (int i = 0; i < cur_block_size_x; ++i){
  809. for (int j = 0; j < cur_block_size_y; ++j){
  810.  
  811. double rho = 1.0;
  812.  
  813. int current_node_global_offset_x = GetGlobalX(i),
  814. current_node_global_offset_y = GetGlobalY(j);
  815.  
  816. int glob_x = current_node_global_offset_x,
  817. glob_y = current_node_global_offset_y;
  818.  
  819. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  820. case 2:
  821. case 4:
  822. case 6:
  823. case 8:
  824. //angle
  825. rho = 0.25;
  826. break;
  827. case 1:
  828. case 3:
  829. case 5:
  830. case 7:
  831. //side
  832. rho = 0.5;
  833. break;
  834. case 0:
  835. //internal
  836. rho = 1.0;
  837. break;
  838. default:
  839. cout << "[ERROR]: Bad global coords compute exact. Local:" << i << " " << j << ". Global:" << glob_x << " " << glob_y <<endl;
  840. }
  841.  
  842.  
  843. double tmp_elem = tmp_elements [j + cur_block_size_y*i];
  844.  
  845. local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  846. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  847. }
  848. }
  849.  
  850. MPI_Allreduce(&local_diff_norm, &global_diff_norm, 1, MPI_DOUBLE, MPI_MAX,
  851. MPI_COMM_WORLD);
  852.  
  853. //global_diff_norm = sqrt(global_diff_norm);
  854.  
  855.  
  856. return global_diff_norm;
  857. }
  858.  
  859. double GetCurrentNorm(){
  860. return current_norm;
  861. }
  862. };
  863.  
  864.  
  865.  
  866.  
  867. int main(int argc, char* argv[]){
  868.  
  869. const double x_left = -2, x_right = 3;
  870. const double y_bottom = -1, y_top = 4;
  871.  
  872. double time_start, time_stop;
  873. int N, Gx, Gy;
  874. int dim[2], period[2];
  875.  
  876. MPI_Comm comm;
  877.  
  878. //N - global grid size
  879. N = atoi(argv[1]);
  880.  
  881. //Gx
  882. Gx = dim[0] = atoi(argv[2]);
  883.  
  884. //Gy
  885. Gy = dim[1] = atoi(argv[3]);
  886.  
  887. period[0]=0;
  888. period[1]=0;
  889.  
  890.  
  891. MPI_Init(&argc, &argv);
  892.  
  893. time_start = MPI_Wtime();
  894.  
  895. int world_size;
  896.  
  897. int my_rank;
  898.  
  899. MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  900. MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  901.  
  902.  
  903. if(my_rank == 0){
  904. cout << "[INFO]: N = " << N << endl;
  905. cout << "[INFO]: Gx = " << dim[0] << endl;
  906. cout << "[INFO]: Gy = " << dim[1] << endl;
  907. }
  908.  
  909.  
  910.  
  911. if(argc <= 3){
  912. if(my_rank == 0)
  913. cout << "[ERROR]: Usage: mpieval <N> <Gx> <Gy>" << endl;
  914.  
  915. MPI_Abort(MPI_COMM_WORLD, 1);
  916. }
  917.  
  918. if(Gx * Gy != world_size){
  919. if(my_rank == 0)
  920. cout << "[ERROR]: mpi world size is not equal to "<< Gx << "*" << Gy << endl;
  921.  
  922. MPI_Abort(MPI_COMM_WORLD, 1);
  923. }
  924.  
  925. MPI_Cart_create(MPI_COMM_WORLD, 2, dim, period, 1, &comm);
  926.  
  927. /*if(my_rank == 0)
  928. cout << "[INFO]: Cart created"<<endl;
  929. */
  930.  
  931. MPI_Comm_rank(comm, &my_rank);
  932.  
  933. int my_coords[2];
  934.  
  935. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  936.  
  937.  
  938.  
  939.  
  940. 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);
  941.  
  942.  
  943. int iteration_num = 0;
  944.  
  945. bool should_i_stop = false;
  946.  
  947. while ( should_i_stop != true ){
  948.  
  949. w_func.DoIteration(should_i_stop);
  950.  
  951. if ( (my_rank == 0) && (iteration_num % 10000 == 0) ){
  952. cout << "[INFO]: Iteration " << iteration_num << " difference with previous is: "<< w_func.GetCurrentNorm() << endl;
  953. }
  954.  
  955. iteration_num++;
  956. }
  957.  
  958. //MPI_Barrier(MPI_COMM_WORLD);
  959.  
  960. //vector <double> local_elements;
  961.  
  962. double comparing = w_func.CompareWithExact();
  963.  
  964. if (my_rank == 0)
  965. cout << "[INFO]: Diff with exact solution: " << comparing << endl;
  966.  
  967.  
  968.  
  969. //myfile.close();
  970.  
  971. time_stop = MPI_Wtime();
  972. if( my_rank == 0 )
  973. cout << "Finished!" << endl
  974. << "Total iterations: " << iteration_num << endl
  975. << "Elapsed time: " << (time_stop - time_start) << endl
  976. << "Time per iteration: " << (time_stop - time_start) / double(iteration_num) << endl;
  977.  
  978. MPI_Finalize();
  979.  
  980. return 0;
  981. }
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!