API tools faq
paste
Advertisement
manac68974

Untitled

manac68974
Dec 6th, 2019
101
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 28.23 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. int IsNodeInternalCornerOrSide(int current_node_global_offset_x, int current_node_global_offset_y){
  34.  
  35. //corners
  36. //left bottom corner
  37. if (current_node_global_offset_x == 0 && current_node_global_offset_y == 0){
  38. return 2;
  39. }
  40.  
  41. //left top corner
  42. if (current_node_global_offset_x == 0 && current_node_global_offset_y == N){
  43. return 4;
  44. }
  45.  
  46. //right bottom corner
  47. if (current_node_global_offset_x == M && current_node_global_offset_y == 0){
  48. return 6;
  49. }
  50.  
  51. //right top corner
  52. if (current_node_global_offset_x == M && current_node_global_offset_y == N){
  53. return 8;
  54. }
  55.  
  56. //sides
  57. //left side
  58. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  59. current_node_global_offset_x == 0){
  60. return 1;
  61. }
  62.  
  63. //right side
  64. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  65. current_node_global_offset_x == M){
  66. return 3;
  67. }
  68.  
  69. //bottom side
  70. if (current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  71. current_node_global_offset_y == 0){
  72. return 5;
  73. }
  74.  
  75. //top side
  76. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  77. current_node_global_offset_y == N)){
  78. return 7;
  79. }
  80.  
  81. //internal
  82. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1) &&
  83. (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1)){
  84. return 0;
  85. }
  86.  
  87. return -1;
  88. }
  89.  
  90. double k(double x, double y) {
  91. return 1 + pow(x + y, 2);
  92. }
  93.  
  94. double q(double x, double y) {
  95. return 1;
  96. }
  97.  
  98. double u(double x, double y) {
  99. return 2.0 / (1 + pow(x, 2) + pow(y, 2));
  100. }
  101.  
  102. // psi_R(x, y) = k(A2, y) * du/dx(A2, y)
  103. double psi_R(double y) {
  104. return (-12) * (pow((y + 3), 2) + 1) / pow((pow(y, 2) + 10), 2);
  105. }
  106.  
  107. // psi_L(x, y) = -k(A1, y) * du/dx(A1, y)
  108. double psi_L(double y) {
  109. return (-8) * (pow((y - 2), 2) + 1) / pow((pow(y, 2) + 5), 2);
  110. }
  111.  
  112. // psi_T(x, y) = k(x, B2) * du/dy(x, B2)
  113. double psi_T(double x) {
  114. return (-16) * (pow((x + 4), 2) + 1) / pow((pow(x, 2) + 17), 2);
  115. }
  116.  
  117. // psi_B(x, y) = -k(x, B1) * du/dy(x, B1)
  118. double psi_B(double x) {
  119. return (-4) * (pow((x - 1), 2) + 1) / pow((pow(x, 2) + 2), 2);
  120. }
  121.  
  122. // right-part function of Poisson equation
  123. double F(double x, double y) {
  124. return 2 * (pow(x,4) + pow(y,4) + 2 * (pow(x,2) + 3) * pow(y,2) + 6 * pow(x,2) + 16*x*y + 5)
  125. / pow((1 + pow(x, 2) + pow(y, 2)), 3);
  126. }
  127.  
  128. //inner_product(A[i], internal_data)
  129. double ComputeMagicInnerProductA_iw (int current_node_global_offset_x, int current_node_global_offset_y){
  130.  
  131. int glob_x = current_node_global_offset_x;
  132. int glob_y = current_node_global_offset_y;
  133.  
  134. double result = 0.0;
  135.  
  136. map <string,bool> neighbours = {
  137. {"left", true},
  138. {"right", true},
  139. {"bottom", true},
  140. {"top", true}
  141. };
  142.  
  143. double left_neighbour = 0.0, right_neighbour = 0.0, bottom_neighbour = 0.0, top_neighbour = 0.0, this_node = 0.0;
  144. double left_coeff = 1.0, right_coeff = 1.0, bottom_coeff = 1.0, top_coeff = 1.0, this_coeff = 1.0;
  145.  
  146. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  147. case 2:
  148. //left bottom corner
  149. neighbours["left"] = false;
  150. neighbours["bottom"] = false;
  151. break;
  152. case 4:
  153. //left top corner
  154. neighbours["left"] = false;
  155. neighbours["top"] = false;
  156. break;
  157. case 6:
  158. //right bottom corner
  159. neighbours["right"] = false;
  160. neighbours["bottom"] = false;
  161. break;
  162. case 8:
  163. //right top corner
  164. neighbours["right"] = false;
  165. neighbours["top"] = false;
  166. break;
  167. case 1:
  168. //left side
  169. neighbours["left"] = false;
  170. break;
  171. case 3:
  172. //right side
  173. neighbours["right"] = false;
  174. break;
  175. case 5:
  176. //bottom side
  177. neighbours["bottom"] = false;
  178. break;
  179. case 7:
  180. //top side
  181. neighbours["top"] = false;
  182. break;
  183. case 0:
  184. //internal
  185. break;
  186. default:
  187. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y<<endl;
  188. }
  189.  
  190. if (!neighbours["left"]){
  191. right_coeff = 2.0;
  192. left_coeff = 0.0;
  193. }
  194.  
  195. if (!neighbours["right"]){
  196. left_coeff = 2.0;
  197. right_coeff = 0.0;
  198. }
  199.  
  200. if (!neighbours["bottom"]){
  201. top_coeff = 2.0;
  202. bottom_coeff = 0.0;
  203. }
  204.  
  205. if (!neighbours["top"]){
  206. bottom_coeff = 2.0;
  207. top_coeff = 0.0;
  208. }
  209.  
  210.  
  211.  
  212. if (neighbours["left"]){
  213. left_coeff *= -k(x_left + (glob_x - 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  214. left_neighbour = Get(glob_x - 1, glob_y);
  215. }
  216.  
  217. if (neighbours["right"]){
  218. right_coeff *= -k(x_left + (glob_x + 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  219. right_neighbour = Get(glob_x + 1, glob_y);
  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. bottom_neighbour = Get(glob_x, glob_y - 1);
  225. }
  226.  
  227. if (neighbours["top"]){
  228. top_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y + 0.5) * h_y) / pow(h_y, 2);
  229. top_neighbour = Get(glob_x, glob_y + 1);
  230. }
  231.  
  232. this_coeff = q(x_left + glob_x * h_x, y_bottom + glob_y * h_y) - left_coeff - right_coeff - bottom_coeff - top_coeff;
  233. this_node = Get(glob_x, glob_y);
  234.  
  235. result = left_coeff * left_neighbour +
  236. right_coeff * right_neighbour +
  237. bottom_coeff * bottom_neighbour +
  238. top_coeff * top_neighbour +
  239. this_coeff * this_node;
  240.  
  241. return result;
  242. }
  243.  
  244. double GetNodeFromB(int current_node_global_offset_x, int current_node_global_offset_y) {
  245.  
  246. int glob_x = current_node_global_offset_x;
  247. int glob_y = current_node_global_offset_y;
  248.  
  249. double result = 0.0;
  250.  
  251. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  252. case 2:
  253. //left bottom corner
  254. result = F(x_left, y_bottom) + 2.0 / h_x * psi_L(y_bottom) + 2.0 / h_y * psi_B(x_left);
  255. break;
  256. case 4:
  257. //left top corner
  258. result = F(x_left, y_top) + 2.0 / h_x * psi_L(y_top) + 2.0 / h_y * psi_T(x_left);
  259. break;
  260. case 6:
  261. //right bottom corner
  262. result = F(x_right, y_bottom) + 2.0 / h_x * psi_R(y_bottom) + 2.0 / h_y * psi_B(x_right);
  263. break;
  264. case 8:
  265. //right top corner
  266. result = F(x_right, y_top) + 2.0 / h_x * psi_R(y_top) + 2.0 / h_y * psi_T(x_right);
  267. break;
  268. case 1:
  269. //left side
  270. result = F(x_left, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_L(y_bottom + glob_y * h_y);
  271. break;
  272. case 3:
  273. //right side
  274. result = F(x_right, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_R(y_bottom + glob_y * h_y);
  275. break;
  276. case 5:
  277. //bottom side
  278. result = F(x_left + glob_x * h_x, y_bottom) + 2.0 / h_y * psi_B(x_left + glob_x * h_x);
  279. break;
  280. case 7:
  281. //top side
  282. result = F(x_left + glob_x * h_x, y_top) + 2.0 / h_y * psi_T(x_left + glob_x * h_x);
  283. break;
  284. case 0:
  285. //internal
  286. result = F(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  287. break;
  288. default:
  289. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y <<endl;
  290.  
  291. }
  292.  
  293. return result;
  294.  
  295. }
  296.  
  297. double GetNodeFromExact(int current_node_global_offset_x, int current_node_global_offset_y) {
  298.  
  299. int glob_x = current_node_global_offset_x;
  300. int glob_y = current_node_global_offset_y;
  301.  
  302. return u(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  303.  
  304. }
  305.  
  306. void ComputeMatrixR(){
  307. /*if(my_rank == 0)
  308. cout << "[INFO]: Computation of matrix r started"<<endl;
  309. */
  310.  
  311. vector<double> r_tmp_matrix (cur_block_size_x*cur_block_size_y, 0.0);
  312.  
  313. //#pragma omp parallel for
  314. for(int i = 0; i < cur_block_size_x; ++i){
  315. for(int j = 0; j < cur_block_size_y; ++j){
  316.  
  317. int current_node_global_offset_x = GetGlobalX(i),
  318. current_node_global_offset_y = GetGlobalY(j);
  319.  
  320. int glob_x = current_node_global_offset_x,
  321. glob_y = current_node_global_offset_y;
  322.  
  323. r_tmp_matrix [ j + cur_block_size_y*i ] = ComputeMagicInnerProductA_iw(glob_x,glob_y) - GetNodeFromB(glob_x, glob_y);
  324.  
  325. }
  326. }
  327.  
  328. //#pragma omp parallel for
  329. for(int i = 0; i < cur_block_size_x; ++i){
  330. for(int j = 0; j < cur_block_size_y; ++j){
  331.  
  332. old_internal_data[ j + cur_block_size_y*i ] = internal_data[ j + cur_block_size_y*i];
  333. internal_data[ j + cur_block_size_y*i ] = r_tmp_matrix[ j + cur_block_size_y*i];
  334.  
  335. }
  336. }
  337.  
  338. //old_internal_data = internal_data;
  339. //internal_data = r_tmp_matrix;
  340.  
  341. SyncMPI();
  342.  
  343. }
  344.  
  345. double ComputeTauAndStopCase(bool &should_i_stop){
  346.  
  347. double local_Ar_r_inner_product_sum = 0.0;
  348. double local_Ar_Ar_inner_product_sum = 0.0;
  349. double global_Ar_r_inner_product_sum = 0.0;
  350. double global_Ar_Ar_inner_product_sum = 0.0;
  351. double local_r_norm = 0.0;
  352. double global_r_norm = 0.0;
  353.  
  354. //#pragma omp parallel for
  355. for(int i = 0; i < cur_block_size_x; ++i){
  356. for(int j = 0; j < cur_block_size_y; ++j){
  357. double rho = 1.0;
  358.  
  359. int current_node_global_offset_x = GetGlobalX(i),
  360. current_node_global_offset_y = GetGlobalY(j);
  361.  
  362. int glob_x = current_node_global_offset_x,
  363. glob_y = current_node_global_offset_y;
  364.  
  365. double tmp_Ar_i_j = ComputeMagicInnerProductA_iw(glob_x, glob_y);
  366.  
  367. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  368. case 2:
  369. case 4:
  370. case 6:
  371. case 8:
  372. //angle
  373. rho = 0.25;
  374. break;
  375. case 1:
  376. case 3:
  377. case 5:
  378. case 7:
  379. //side
  380. rho = 0.5;
  381. break;
  382. case 0:
  383. //internal
  384. rho = 1.0;
  385. break;
  386. default:
  387. cout << "[ERROR]: Bad global coords compute tau. Global:" << glob_x << " " << glob_y << endl;
  388. }
  389.  
  390. double tmp_cur_node_value = Get(glob_x, glob_y);
  391.  
  392. local_Ar_r_inner_product_sum += rho * tmp_Ar_i_j * tmp_cur_node_value * h_x*h_y;
  393. local_Ar_Ar_inner_product_sum += rho * pow (tmp_Ar_i_j, 2) * h_x*h_y;
  394. local_r_norm += rho * pow(tmp_cur_node_value, 2) * h_x*h_y;
  395.  
  396. }
  397. }
  398.  
  399. //cout << "[DEBUG]: Local"<< local_Ar_r_inner_product_sum << endl;
  400.  
  401. MPI_Allreduce(&local_Ar_r_inner_product_sum, &global_Ar_r_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  402. comm);
  403.  
  404. //cout << "[DEBUG]: "<< global_Ar_r_inner_product_sum << endl;
  405.  
  406. MPI_Allreduce(&local_Ar_Ar_inner_product_sum, &global_Ar_Ar_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  407. MPI_COMM_WORLD);
  408.  
  409. //cout << "[DEBUG]: "<< global_Ar_Ar_inner_product_sum << endl;
  410.  
  411. double global_tau = global_Ar_r_inner_product_sum/ global_Ar_Ar_inner_product_sum;
  412.  
  413. MPI_Allreduce(&local_r_norm, &global_r_norm, 1, MPI_DOUBLE, MPI_SUM,
  414. MPI_COMM_WORLD);
  415.  
  416. double eps = 1e-06;
  417.  
  418. if (global_r_norm < 0){
  419. cout << "[ERROR]: bad global r norm" << endl;
  420. }
  421.  
  422. current_norm = fabs(global_tau)*sqrt(global_r_norm);
  423.  
  424. //if (my_rank == 0)
  425. // cout << "[DEBUG]: solution norm "<< tmp_norm << endl;
  426.  
  427. if (current_norm <= eps){
  428. should_i_stop = true;
  429. }else{
  430. should_i_stop = false;
  431. }
  432.  
  433. return global_tau;
  434.  
  435. }
  436.  
  437.  
  438. void ComputeNewW(double tau){
  439. //#pragma omp parallel for
  440. for(int i = 0; i < cur_block_size_x; ++i){
  441. for(int j = 0; j < cur_block_size_y; ++j){
  442. 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 ];
  443. //old_internal_data[ j + cur_block_size_y*i ] = 0.0;
  444. }
  445. }
  446. }
  447.  
  448. int GetGlobalX(int i){
  449. return cur_block_global_offset_x + i;
  450. }
  451.  
  452. int GetGlobalY(int j){
  453. return cur_block_global_offset_y + j;
  454. }
  455.  
  456. public:
  457. 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){
  458.  
  459. M = inpM;
  460. N = inpN;
  461.  
  462. Gx = inpGx;
  463. Gy = inpGy;
  464.  
  465. x_left = inpx_left;
  466. x_right = inpx_right;
  467. y_bottom = inpy_bottom;
  468. y_top = inpy_top;
  469.  
  470. h_x = double((x_right - x_left)) / M;
  471. h_y = double((y_top - y_bottom)) / N;
  472.  
  473. my_rank = inprank;
  474. comm = inpcomm;
  475.  
  476. ofstream myfile;
  477. stringstream ss;
  478. ss<< "./file" << inp_cur_block_global_coors_x << "_" << inp_cur_block_global_coors_y << ".res";
  479.  
  480. string filename;
  481.  
  482. ss >> filename;
  483.  
  484. my_file.open (filename);
  485.  
  486.  
  487. cur_block_global_coors_x = inp_cur_block_global_coors_x;
  488. cur_block_global_coors_y = inp_cur_block_global_coors_y;
  489.  
  490. global_block_size_x = (M + 1) / Gx;
  491. global_block_size_y = (N + 1) / Gy;
  492.  
  493. cur_block_size_x = global_block_size_x;
  494. cur_block_size_y = global_block_size_y;
  495.  
  496. cur_block_global_offset_x = global_block_size_x * cur_block_global_coors_x;
  497. cur_block_global_offset_y = global_block_size_y * cur_block_global_coors_y;
  498.  
  499. if (cur_block_global_offset_x + 2 * global_block_size_x > M + 1){
  500. cur_block_size_x += (M + 1) % Gx;
  501. }
  502.  
  503. if (cur_block_global_offset_y + 2 * global_block_size_y > N + 1){
  504. cur_block_size_y += (N + 1) % Gy;
  505. }
  506.  
  507. internal_data.resize(cur_block_size_x * cur_block_size_y);
  508. old_internal_data.resize(cur_block_size_x * cur_block_size_y);
  509.  
  510.  
  511. //OX
  512. external_data[0].resize(cur_block_size_y);
  513. external_data[1].resize(cur_block_size_y);
  514.  
  515. //OY
  516. external_data[2].resize(cur_block_size_x);
  517. external_data[3].resize(cur_block_size_x);
  518.  
  519. }
  520.  
  521. double Get(int i, int j) {
  522. return GetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y);
  523. }
  524.  
  525. void Set(int i, int j, double v) {
  526. return SetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y, v);
  527. }
  528.  
  529. void SyncMPI(){
  530.  
  531. /*if(my_rank == 0)
  532. cout << "[INFO]: Sync started"<< endl;
  533. */
  534.  
  535. MPI_Barrier(MPI_COMM_WORLD);
  536.  
  537. //left and right sides
  538. //#pragma omp parallel for
  539. for(int j = 0; j < cur_block_size_y; ++j){
  540.  
  541. external_data[ 0 ][ j ] = GetLocalIndex(0, j);//internal_data[ j ];
  542. external_data[ 1 ][ j ] = GetLocalIndex(cur_block_size_x - 1, j);//internal_data[ j + cur_block_size_y * (cur_block_size_x - 1) ];
  543.  
  544. }
  545.  
  546. //bottom and top sides
  547. //#pragma omp parallel for
  548. for(int i = 0; i < cur_block_size_x; ++i){
  549.  
  550. external_data[ 2 ][ i ] = GetLocalIndex(i, 0);//internal_data[ cur_block_size_y*i ];
  551. external_data[ 3 ][ i ] = GetLocalIndex(i, cur_block_size_y - 1); //internal_data[ (cur_block_size_y - 1) + cur_block_size_y*i ];
  552.  
  553. }
  554.  
  555. int my_coords[2];
  556. int targets_ranks[4];
  557.  
  558. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  559.  
  560. int neighbour_offsets[ 4 ][ 2 ] = {
  561. { -1, 0 },{ 1, 0 },
  562. { 0, -1 },{ 0, 1 }
  563. };
  564.  
  565. //#pragma omp parallel for
  566. for(int i = 0; i < 4; i++){
  567.  
  568. int target_coords[2];
  569.  
  570. target_coords[ 0 ] = my_coords[ 0 ] + neighbour_offsets[ i ][ 0 ];
  571. target_coords[ 1 ] = my_coords[ 1 ] + neighbour_offsets[ i ][ 1 ];
  572.  
  573. if (target_coords[0] >= 0 && target_coords[0] < Gx && target_coords[1] >= 0 && target_coords[1] < Gy){
  574.  
  575. MPI_Cart_rank(comm, target_coords, &targets_ranks[ i ]);
  576.  
  577. }
  578. else{
  579. targets_ranks[i] = -1;
  580. }
  581. }
  582.  
  583. //Now we have rank for all targets
  584.  
  585. for(int axis = 0; axis < 2; axis++){
  586.  
  587. int parity_bit = (my_coords[ axis ]) % 2;
  588.  
  589. //if parity_bit == 0 then
  590. // zuerst mit links, dann mit rechts tauschen
  591. //elif parity_bit == 1:
  592. // zuerst mit rechts, dann mit links tauschen
  593.  
  594. for(int tmp = 0; tmp < 2; tmp++){
  595. parity_bit = 1 - parity_bit;
  596.  
  597. //target id in external_data und targets_ranks
  598. int target_idx = 2 * axis + (1 - parity_bit);
  599.  
  600. if (targets_ranks[target_idx] != -1){
  601.  
  602. // вычисляем теги отправки и приема
  603. // в них зашиты номер ноды, ось, направление
  604. int send_tag = 100000 + my_rank * 100 + axis * 10 + parity_bit;
  605. int recv_tag = 100000 + targets_ranks[ target_idx ] * 100 + axis * 10 + (1-parity_bit);
  606.  
  607. MPI_Status tmp_status;
  608. // если отправка не на себя, то отправляем
  609. if(my_rank != targets_ranks[ target_idx ]){
  610.  
  611. MPI_Sendrecv_replace(&external_data[ target_idx ][ 0 ], external_data[ target_idx ].size(),
  612. MPI_DOUBLE, targets_ranks[ target_idx ], send_tag, targets_ranks[ target_idx ], recv_tag,
  613. comm, &tmp_status);
  614.  
  615. }
  616. }
  617. }
  618. }
  619.  
  620. MPI_Barrier(MPI_COMM_WORLD);
  621.  
  622. }
  623.  
  624. void DoIteration(bool &should_i_stop){
  625.  
  626. ComputeMatrixR();
  627.  
  628. //Now R Matrix is in internal_data
  629.  
  630. double tau = ComputeTauAndStopCase(should_i_stop);
  631.  
  632. //We have in our block jetzt:
  633. //in internal_data: R Matrix
  634. //in old_internal_data: w aus letzte iteration
  635. //and we have tau
  636. //jetzt koennen wir naechste w finden
  637.  
  638. ComputeNewW(tau);
  639.  
  640. SyncMPI();
  641.  
  642. }
  643.  
  644. double GetLocalIndex(int i, int j){
  645. //internal data
  646. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  647. return internal_data[ j + cur_block_size_y*i ];
  648. }
  649.  
  650. //external data
  651. //OX
  652. if((j >= 0) && (j < cur_block_size_y)){
  653.  
  654. if (i == -1)
  655. return external_data[ 0 ][ j ];
  656.  
  657. if (i == cur_block_size_x)
  658. return external_data[ 1 ][ j ];
  659.  
  660. }
  661.  
  662. //OY
  663. if((i >= 0) && (i < cur_block_size_x)){
  664.  
  665. if (j == -1)
  666. return external_data[ 2 ][ i ];
  667. if (j == cur_block_size_y)
  668. return external_data[ 3 ][ i ];
  669.  
  670. }
  671.  
  672. cout << "[ERROR]: bad local index" << endl;
  673.  
  674. return nan("");
  675. }
  676.  
  677. void SetLocalIndex(int i, int j, double v){
  678. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  679. internal_data[ j + cur_block_size_y*i ] = v;
  680. }else{
  681. cout << "[ERROR]: trying to set data outside the local area" << endl;
  682. }
  683. }
  684.  
  685. double CompareWithExact(bool print_data_to_files) {
  686.  
  687. double local_diff_norm = 0.0;
  688. double global_diff_norm = 0.0;
  689. /*
  690. if (my_rank == 0)
  691. cout << "[INFO]: Starting computing compare with exact" << endl;
  692. */
  693. vector <double> tmp_elements(cur_block_size_x * cur_block_size_y, 0.0);
  694.  
  695. //#pragma omp parallel for
  696. for (int i = 0; i < cur_block_size_x; ++i){
  697. for (int j = 0; j < cur_block_size_y; ++j){
  698.  
  699.  
  700. int current_node_global_offset_x = GetGlobalX(i),
  701. current_node_global_offset_y = GetGlobalY(j);
  702.  
  703. int glob_x = current_node_global_offset_x,
  704. glob_y = current_node_global_offset_y;
  705.  
  706. double tmp_elem = Get(glob_x, glob_y) - GetNodeFromExact(glob_x, glob_y);
  707.  
  708. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  709.  
  710. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  711.  
  712. tmp_elements [j + cur_block_size_y*i] = tmp_elem;
  713.  
  714. }
  715. }
  716.  
  717. //cout << "[INFO]: local max diff in " << cur_block_global_offset_x << " " << cur_block_global_offset_y << " " << local_diff_norm << endl;
  718.  
  719.  
  720. for (int i = 0; i < cur_block_size_x; ++i){
  721. for (int j = 0; j < cur_block_size_y; ++j){
  722.  
  723. double rho = 1.0;
  724.  
  725. int current_node_global_offset_x = GetGlobalX(i),
  726. current_node_global_offset_y = GetGlobalY(j);
  727.  
  728. int glob_x = current_node_global_offset_x,
  729. glob_y = current_node_global_offset_y;
  730.  
  731. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  732. case 2:
  733. case 4:
  734. case 6:
  735. case 8:
  736. //angle
  737. rho = 0.25;
  738. break;
  739. case 1:
  740. case 3:
  741. case 5:
  742. case 7:
  743. //side
  744. rho = 0.5;
  745. break;
  746. case 0:
  747. //internal
  748. rho = 1.0;
  749. break;
  750. default:
  751. cout << "[ERROR]: Bad global coords compute exact. Local:" << i << " " << j << ". Global:" << glob_x << " " << glob_y <<endl;
  752. }
  753.  
  754.  
  755. double tmp_elem = tmp_elements [j + cur_block_size_y*i];
  756.  
  757. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  758. local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  759. }
  760. }
  761.  
  762. MPI_Allreduce(&local_diff_norm, &global_diff_norm, 1, MPI_DOUBLE, MPI_SUM,
  763. MPI_COMM_WORLD);
  764.  
  765. global_diff_norm = sqrt(global_diff_norm);
  766.  
  767.  
  768.  
  769.  
  770.  
  771. if (print_data_to_files){
  772. for (int i = 0; i < cur_block_size_x; ++i){
  773. for (int j = 0; j < cur_block_size_y; ++j){
  774.  
  775. int current_node_global_offset_x = GetGlobalX(i),
  776. current_node_global_offset_y = GetGlobalY(j);
  777.  
  778. int glob_x = current_node_global_offset_x,
  779. glob_y = current_node_global_offset_y;
  780.  
  781. my_file << Get(glob_x, glob_y) << ":" << GetNodeFromExact(glob_x, glob_y) << "\t";
  782. }
  783. my_file << endl;
  784. }
  785. }
  786.  
  787. return global_diff_norm;
  788. }
  789.  
  790. double GetCurrentNorm(){
  791. return current_norm;
  792. }
  793. };
  794.  
  795.  
  796.  
  797.  
  798. int main(int argc, char* argv[]){
  799.  
  800. const double x_left = -2, x_right = 3;
  801. const double y_bottom = -1, y_top = 4;
  802.  
  803. double time_start, time_stop;
  804. int N, Gx, Gy;
  805. int dim[2], period[2];
  806.  
  807. MPI_Comm comm;
  808.  
  809. //N - global grid size
  810. N = atoi(argv[1]);
  811.  
  812. //Gx
  813. Gx = dim[0] = atoi(argv[2]);
  814.  
  815. //Gy
  816. Gy = dim[1] = atoi(argv[3]);
  817.  
  818. bool print_data_to_files = false;
  819.  
  820. if (argc == 4){
  821. print_data_to_files = true;
  822. }
  823. period[0]=0;
  824. period[1]=0;
  825.  
  826.  
  827. MPI_Init(&argc, &argv);
  828.  
  829. time_start = MPI_Wtime();
  830.  
  831. int world_size;
  832.  
  833. int my_rank;
  834.  
  835. MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  836. MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  837.  
  838.  
  839. if(my_rank == 0){
  840. cout << "[INFO]: N = " << N << endl;
  841. cout << "[INFO]: Gx = " << dim[0] << endl;
  842. cout << "[INFO]: Gy = " << dim[1] << endl;
  843. }
  844.  
  845.  
  846.  
  847. if(argc <= 3){
  848. if(my_rank == 0)
  849. cout << "[ERROR]: Usage: mpieval <N> <Gx> <Gy> [<print_data_to_files>]" << endl;
  850.  
  851. MPI_Abort(MPI_COMM_WORLD, 1);
  852. }
  853.  
  854. if(Gx * Gy != world_size){
  855. if(my_rank == 0)
  856. cout << "[ERROR]: mpi world size is not equal to "<< Gx << "*" << Gy << endl;
  857.  
  858. MPI_Abort(MPI_COMM_WORLD, 1);
  859. }
  860.  
  861. MPI_Cart_create(MPI_COMM_WORLD, 2, dim, period, 1, &comm);
  862.  
  863. /*if(my_rank == 0)
  864. cout << "[INFO]: Cart created"<<endl;
  865. */
  866.  
  867. MPI_Comm_rank(comm, &my_rank);
  868.  
  869. int my_coords[2];
  870.  
  871. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  872.  
  873.  
  874.  
  875.  
  876. 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);
  877.  
  878.  
  879. int iteration_num = 0;
  880.  
  881. bool should_i_stop = false;
  882.  
  883. while ( should_i_stop != true ){
  884.  
  885. w_func.DoIteration(should_i_stop);
  886.  
  887. if ( (my_rank == 0) && (iteration_num % 10000 == 0) ){
  888. cout << "[INFO]: Iteration " << iteration_num << " difference with previous is: "<< w_func.GetCurrentNorm() << endl;
  889. }
  890.  
  891. iteration_num++;
  892. }
  893.  
  894. MPI_Barrier(MPI_COMM_WORLD);
  895.  
  896. //vector <double> local_elements;
  897.  
  898. double comparing = w_func.CompareWithExact(print_data_to_files);
  899.  
  900. if (my_rank == 0)
  901. cout << "[INFO]: Diff with exact solution: " << comparing << endl;
  902.  
  903.  
  904.  
  905. //myfile.close();
  906.  
  907. time_stop = MPI_Wtime();
  908. if( my_rank == 0 )
  909. cout << "Finished!" << endl
  910. << "Total iterations: " << iteration_num << endl
  911. << "Elapsed time: " << (time_stop - time_start) << endl
  912. << "Time per iteration: " << (time_stop - time_start) / double(iteration_num) << endl;
  913.  
  914. MPI_Finalize();
  915.  
  916. return 0;
  917. }
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!