API tools faq
paste
Advertisement
manac68974

Untitled

manac68974
Dec 18th, 2019
93
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 27.14 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. int IsNodeInternalCornerOrSide(int current_node_global_offset_x, int current_node_global_offset_y){
  33.  
  34. //corners
  35. //left bottom corner
  36. if (current_node_global_offset_x == 0 && current_node_global_offset_y == 0){
  37. return 2;
  38. }
  39.  
  40. //left top corner
  41. if (current_node_global_offset_x == 0 && current_node_global_offset_y == N){
  42. return 4;
  43. }
  44.  
  45. //right bottom corner
  46. if (current_node_global_offset_x == M && current_node_global_offset_y == 0){
  47. return 6;
  48. }
  49.  
  50. //right top corner
  51. if (current_node_global_offset_x == M && current_node_global_offset_y == N){
  52. return 8;
  53. }
  54.  
  55. //sides
  56. //left side
  57. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  58. current_node_global_offset_x == 0){
  59. return 1;
  60. }
  61.  
  62. //right side
  63. if (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1 &&
  64. current_node_global_offset_x == M){
  65. return 3;
  66. }
  67.  
  68. //bottom side
  69. if (current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  70. current_node_global_offset_y == 0){
  71. return 5;
  72. }
  73.  
  74. //top side
  75. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1 &&
  76. current_node_global_offset_y == N)){
  77. return 7;
  78. }
  79.  
  80. //internal
  81. if ((current_node_global_offset_x >= 1 && current_node_global_offset_x <= M - 1) &&
  82. (current_node_global_offset_y >= 1 && current_node_global_offset_y <= N - 1)){
  83. return 0;
  84. }
  85.  
  86. return -1;
  87. }
  88.  
  89. double k(double x, double y) {
  90. return 1 + pow(x + y, 2);
  91. }
  92.  
  93. double q(double x, double y) {
  94. return 1;
  95. }
  96.  
  97. double u(double x, double y) {
  98. return 2.0 / (1 + pow(x, 2) + pow(y, 2));
  99. }
  100.  
  101. // psi_R(x, y) = k(A2, y) * du/dx(A2, y)
  102. double psi_R(double y) {
  103. return (-12) * (pow((y + 3), 2) + 1) / pow((pow(y, 2) + 10), 2);
  104. }
  105.  
  106. // psi_L(x, y) = -k(A1, y) * du/dx(A1, y)
  107. double psi_L(double y) {
  108. return (-8) * (pow((y - 2), 2) + 1) / pow((pow(y, 2) + 5), 2);
  109. }
  110.  
  111. // psi_T(x, y) = k(x, B2) * du/dy(x, B2)
  112. double psi_T(double x) {
  113. return (-16) * (pow((x + 4), 2) + 1) / pow((pow(x, 2) + 17), 2);
  114. }
  115.  
  116. // psi_B(x, y) = -k(x, B1) * du/dy(x, B1)
  117. double psi_B(double x) {
  118. return (-4) * (pow((x - 1), 2) + 1) / pow((pow(x, 2) + 2), 2);
  119. }
  120.  
  121. // right-part function of Poisson equation
  122. double F(double x, double y) {
  123. return 2 * (pow(x,4) + pow(y,4) + 2 * (pow(x,2) + 3) * pow(y,2) + 6 * pow(x,2) + 16*x*y + 5)
  124. / pow((1 + pow(x, 2) + pow(y, 2)), 3);
  125. }
  126.  
  127. //inner_product(A[i], internal_data)
  128. double ComputeMagicInnerProductA_iw (int current_node_global_offset_x, int current_node_global_offset_y){
  129.  
  130. int glob_x = current_node_global_offset_x;
  131. int glob_y = current_node_global_offset_y;
  132.  
  133. double result = 0.0;
  134.  
  135. map <string,bool> neighbours = {
  136. {"left", true},
  137. {"right", true},
  138. {"bottom", true},
  139. {"top", true}
  140. };
  141.  
  142. double left_neighbour = 0.0, right_neighbour = 0.0, bottom_neighbour = 0.0, top_neighbour = 0.0, this_node = 0.0;
  143. double left_coeff = 1.0, right_coeff = 1.0, bottom_coeff = 1.0, top_coeff = 1.0, this_coeff = 1.0;
  144.  
  145. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  146. case 2:
  147. //left bottom corner
  148. neighbours["left"] = false;
  149. neighbours["bottom"] = false;
  150. break;
  151. case 4:
  152. //left top corner
  153. neighbours["left"] = false;
  154. neighbours["top"] = false;
  155. break;
  156. case 6:
  157. //right bottom corner
  158. neighbours["right"] = false;
  159. neighbours["bottom"] = false;
  160. break;
  161. case 8:
  162. //right top corner
  163. neighbours["right"] = false;
  164. neighbours["top"] = false;
  165. break;
  166. case 1:
  167. //left side
  168. neighbours["left"] = false;
  169. break;
  170. case 3:
  171. //right side
  172. neighbours["right"] = false;
  173. break;
  174. case 5:
  175. //bottom side
  176. neighbours["bottom"] = false;
  177. break;
  178. case 7:
  179. //top side
  180. neighbours["top"] = false;
  181. break;
  182. case 0:
  183. //internal
  184. break;
  185. default:
  186. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y<<endl;
  187. }
  188.  
  189. if (!neighbours["left"]){
  190. right_coeff = 2.0;
  191. left_coeff = 0.0;
  192. }
  193.  
  194. if (!neighbours["right"]){
  195. left_coeff = 2.0;
  196. right_coeff = 0.0;
  197. }
  198.  
  199. if (!neighbours["bottom"]){
  200. top_coeff = 2.0;
  201. bottom_coeff = 0.0;
  202. }
  203.  
  204. if (!neighbours["top"]){
  205. bottom_coeff = 2.0;
  206. top_coeff = 0.0;
  207. }
  208.  
  209.  
  210.  
  211. if (neighbours["left"]){
  212. left_coeff *= -k(x_left + (glob_x - 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  213. left_neighbour = Get(glob_x - 1, glob_y);
  214. }
  215.  
  216. if (neighbours["right"]){
  217. right_coeff *= -k(x_left + (glob_x + 0.5) * h_x, y_bottom + glob_y * h_y) / pow(h_x, 2);
  218. right_neighbour = Get(glob_x + 1, glob_y);
  219. }
  220.  
  221. if (neighbours["bottom"]){
  222. bottom_coeff *= -k(x_left + glob_x * h_x, y_bottom + (glob_y - 0.5) * h_y) / pow(h_y, 2);
  223. bottom_neighbour = Get(glob_x, glob_y - 1);
  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. top_neighbour = Get(glob_x, glob_y + 1);
  229. }
  230.  
  231. this_coeff = q(x_left + glob_x * h_x, y_bottom + glob_y * h_y) - left_coeff - right_coeff - bottom_coeff - top_coeff;
  232. this_node = Get(glob_x, glob_y);
  233.  
  234. result = left_coeff * left_neighbour +
  235. right_coeff * right_neighbour +
  236. bottom_coeff * bottom_neighbour +
  237. top_coeff * top_neighbour +
  238. this_coeff * this_node;
  239.  
  240. return result;
  241. }
  242.  
  243. double GetNodeFromB(int current_node_global_offset_x, int current_node_global_offset_y) {
  244.  
  245. int glob_x = current_node_global_offset_x;
  246. int glob_y = current_node_global_offset_y;
  247.  
  248. double result = 0.0;
  249.  
  250. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  251. case 2:
  252. //left bottom corner
  253. result = F(x_left, y_bottom) + 2.0 / h_x * psi_L(y_bottom) + 2.0 / h_y * psi_B(x_left);
  254. break;
  255. case 4:
  256. //left top corner
  257. result = F(x_left, y_top) + 2.0 / h_x * psi_L(y_top) + 2.0 / h_y * psi_T(x_left);
  258. break;
  259. case 6:
  260. //right bottom corner
  261. result = F(x_right, y_bottom) + 2.0 / h_x * psi_R(y_bottom) + 2.0 / h_y * psi_B(x_right);
  262. break;
  263. case 8:
  264. //right top corner
  265. result = F(x_right, y_top) + 2.0 / h_x * psi_R(y_top) + 2.0 / h_y * psi_T(x_right);
  266. break;
  267. case 1:
  268. //left side
  269. result = F(x_left, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_L(y_bottom + glob_y * h_y);
  270. break;
  271. case 3:
  272. //right side
  273. result = F(x_right, y_bottom + glob_y * h_y) + 2.0 / h_x * psi_R(y_bottom + glob_y * h_y);
  274. break;
  275. case 5:
  276. //bottom side
  277. result = F(x_left + glob_x * h_x, y_bottom) + 2.0 / h_y * psi_B(x_left + glob_x * h_x);
  278. break;
  279. case 7:
  280. //top side
  281. result = F(x_left + glob_x * h_x, y_top) + 2.0 / h_y * psi_T(x_left + glob_x * h_x);
  282. break;
  283. case 0:
  284. //internal
  285. result = F(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  286. break;
  287. default:
  288. cout << "[ERROR]: Bad global coords compute matrix. Global:" << glob_x << " " << glob_y <<endl;
  289.  
  290. }
  291.  
  292. return result;
  293.  
  294. }
  295.  
  296. double GetNodeFromExact(int current_node_global_offset_x, int current_node_global_offset_y) {
  297.  
  298. int glob_x = current_node_global_offset_x;
  299. int glob_y = current_node_global_offset_y;
  300.  
  301. return u(x_left + glob_x * h_x, y_bottom + glob_y * h_y);
  302.  
  303. }
  304.  
  305. void ComputeMatrixR(){
  306. /*if(my_rank == 0)
  307. cout << "[INFO]: Computation of matrix r started"<<endl;
  308. */
  309.  
  310. vector<double> r_tmp_matrix (cur_block_size_x*cur_block_size_y, 0.0);
  311.  
  312. //#pragma omp parallel for
  313. for(int i = 0; i < cur_block_size_x; ++i){
  314. for(int j = 0; j < cur_block_size_y; ++j){
  315.  
  316. int current_node_global_offset_x = GetGlobalX(i),
  317. current_node_global_offset_y = GetGlobalY(j);
  318.  
  319. int glob_x = current_node_global_offset_x,
  320. glob_y = current_node_global_offset_y;
  321.  
  322. r_tmp_matrix [ j + cur_block_size_y*i ] = ComputeMagicInnerProductA_iw(glob_x,glob_y) - GetNodeFromB(glob_x, glob_y);
  323.  
  324. }
  325. }
  326.  
  327. //#pragma omp parallel for
  328. for(int i = 0; i < cur_block_size_x; ++i){
  329. for(int j = 0; j < cur_block_size_y; ++j){
  330.  
  331. old_internal_data[ j + cur_block_size_y*i ] = internal_data[ j + cur_block_size_y*i];
  332. internal_data[ j + cur_block_size_y*i ] = r_tmp_matrix[ j + cur_block_size_y*i];
  333.  
  334. }
  335. }
  336.  
  337. //old_internal_data = internal_data;
  338. //internal_data = r_tmp_matrix;
  339.  
  340. SyncMPI();
  341.  
  342. }
  343.  
  344. double ComputeTauAndStopCase(bool &should_i_stop){
  345.  
  346. double local_Ar_r_inner_product_sum = 0.0;
  347. double local_Ar_Ar_inner_product_sum = 0.0;
  348. double global_Ar_r_inner_product_sum = 0.0;
  349. double global_Ar_Ar_inner_product_sum = 0.0;
  350. double local_r_norm = 0.0;
  351. double global_r_norm = 0.0;
  352.  
  353. //#pragma omp parallel for
  354. for(int i = 0; i < cur_block_size_x; ++i){
  355. for(int j = 0; j < cur_block_size_y; ++j){
  356. double rho = 1.0;
  357.  
  358. int current_node_global_offset_x = GetGlobalX(i),
  359. current_node_global_offset_y = GetGlobalY(j);
  360.  
  361. int glob_x = current_node_global_offset_x,
  362. glob_y = current_node_global_offset_y;
  363.  
  364. double tmp_Ar_i_j = ComputeMagicInnerProductA_iw(glob_x, glob_y);
  365.  
  366. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  367. case 2:
  368. case 4:
  369. case 6:
  370. case 8:
  371. //angle
  372. rho = 0.25;
  373. break;
  374. case 1:
  375. case 3:
  376. case 5:
  377. case 7:
  378. //side
  379. rho = 0.5;
  380. break;
  381. case 0:
  382. //internal
  383. rho = 1.0;
  384. break;
  385. default:
  386. cout << "[ERROR]: Bad global coords compute tau. Global:" << glob_x << " " << glob_y << endl;
  387. }
  388.  
  389. double tmp_cur_node_value = Get(glob_x, glob_y);
  390.  
  391. local_Ar_r_inner_product_sum += rho * tmp_Ar_i_j * tmp_cur_node_value * h_x*h_y;
  392. local_Ar_Ar_inner_product_sum += rho * pow (tmp_Ar_i_j, 2) * h_x*h_y;
  393. local_r_norm += rho * pow(tmp_cur_node_value, 2) * h_x*h_y;
  394.  
  395. }
  396. }
  397.  
  398. //cout << "[DEBUG]: Local"<< local_Ar_r_inner_product_sum << endl;
  399.  
  400. MPI_Allreduce(&local_Ar_r_inner_product_sum, &global_Ar_r_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  401. comm);
  402.  
  403. //cout << "[DEBUG]: "<< global_Ar_r_inner_product_sum << endl;
  404.  
  405. MPI_Allreduce(&local_Ar_Ar_inner_product_sum, &global_Ar_Ar_inner_product_sum, 1, MPI_DOUBLE, MPI_SUM,
  406. MPI_COMM_WORLD);
  407.  
  408. //cout << "[DEBUG]: "<< global_Ar_Ar_inner_product_sum << endl;
  409.  
  410. double global_tau = global_Ar_r_inner_product_sum/ global_Ar_Ar_inner_product_sum;
  411.  
  412. MPI_Allreduce(&local_r_norm, &global_r_norm, 1, MPI_DOUBLE, MPI_SUM,
  413. MPI_COMM_WORLD);
  414.  
  415. double eps = 1e-06;
  416.  
  417. if (global_r_norm < 0){
  418. cout << "[ERROR]: bad global r norm" << endl;
  419. }
  420.  
  421. current_norm = fabs(global_tau)*sqrt(global_r_norm);
  422.  
  423. //if (my_rank == 0)
  424. // cout << "[DEBUG]: solution norm "<< tmp_norm << endl;
  425.  
  426. if (current_norm <= eps){
  427. if (my_rank == 0)
  428. cout << "[INFO]: r norm is " << sqrt(global_r_norm) << endl;
  429. should_i_stop = true;
  430. }else{
  431. should_i_stop = false;
  432. }
  433.  
  434. return global_tau;
  435.  
  436. }
  437.  
  438.  
  439. void ComputeNewW(double tau){
  440. //#pragma omp parallel for
  441. for(int i = 0; i < cur_block_size_x; ++i){
  442. for(int j = 0; j < cur_block_size_y; ++j){
  443. 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 ];
  444. //old_internal_data[ j + cur_block_size_y*i ] = 0.0;
  445. }
  446. }
  447. }
  448.  
  449. int GetGlobalX(int i){
  450. return cur_block_global_offset_x + i;
  451. }
  452.  
  453. int GetGlobalY(int j){
  454. return cur_block_global_offset_y + j;
  455. }
  456.  
  457. public:
  458. 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){
  459.  
  460. M = inpM;
  461. N = inpN;
  462.  
  463. Gx = inpGx;
  464. Gy = inpGy;
  465.  
  466. x_left = inpx_left;
  467. x_right = inpx_right;
  468. y_bottom = inpy_bottom;
  469. y_top = inpy_top;
  470.  
  471. h_x = double((x_right - x_left)) / M;
  472. h_y = double((y_top - y_bottom)) / N;
  473.  
  474. my_rank = inprank;
  475. comm = inpcomm;
  476.  
  477. cur_block_global_coors_x = inp_cur_block_global_coors_x;
  478. cur_block_global_coors_y = inp_cur_block_global_coors_y;
  479.  
  480. global_block_size_x = (M + 1) / Gx;
  481. global_block_size_y = (N + 1) / Gy;
  482.  
  483. cur_block_size_x = global_block_size_x;
  484. cur_block_size_y = global_block_size_y;
  485.  
  486. cur_block_global_offset_x = global_block_size_x * cur_block_global_coors_x;
  487. cur_block_global_offset_y = global_block_size_y * cur_block_global_coors_y;
  488.  
  489. //if ((cur_block_global_offset_x + global_block_size_x + Gx) > (M + 1)){
  490. if(cur_block_global_coors_x == Gx - 1 ){
  491. cur_block_size_x += (M + 1) % Gx;
  492. }
  493.  
  494. //if ((cur_block_global_offset_y + global_block_size_y + Gy) > (N + 1)){
  495. if(cur_block_global_coors_y == Gy - 1){
  496. cur_block_size_y += (N + 1) % Gy;
  497. }
  498.  
  499. internal_data.resize(cur_block_size_x * cur_block_size_y);
  500. old_internal_data.resize(cur_block_size_x * cur_block_size_y);
  501.  
  502.  
  503. //OX
  504. external_data[0].resize(cur_block_size_y);
  505. external_data[1].resize(cur_block_size_y);
  506.  
  507. //OY
  508. external_data[2].resize(cur_block_size_x);
  509. external_data[3].resize(cur_block_size_x);
  510.  
  511. }
  512.  
  513. double Get(int i, int j) {
  514. return GetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y);
  515. }
  516.  
  517. void Set(int i, int j, double v) {
  518. return SetLocalIndex(i - cur_block_global_offset_x, j - cur_block_global_offset_y, v);
  519. }
  520.  
  521. void SyncMPI(){
  522.  
  523. /*if(my_rank == 0)
  524. cout << "[INFO]: Sync started"<< endl;
  525. */
  526.  
  527. //MPI_Barrier(MPI_COMM_WORLD);
  528.  
  529. //left and right sides
  530. //#pragma omp parallel for
  531. for(int j = 0; j < cur_block_size_y; ++j){
  532.  
  533. external_data[ 0 ][ j ] = GetLocalIndex(0, j);//internal_data[ j ];
  534. external_data[ 1 ][ j ] = GetLocalIndex(cur_block_size_x - 1, j);//internal_data[ j + cur_block_size_y * (cur_block_size_x - 1) ];
  535.  
  536. }
  537.  
  538. //bottom and top sides
  539. //#pragma omp parallel for
  540. for(int i = 0; i < cur_block_size_x; ++i){
  541.  
  542. external_data[ 2 ][ i ] = GetLocalIndex(i, 0);//internal_data[ cur_block_size_y*i ];
  543. external_data[ 3 ][ i ] = GetLocalIndex(i, cur_block_size_y - 1); //internal_data[ (cur_block_size_y - 1) + cur_block_size_y*i ];
  544.  
  545. }
  546.  
  547. int my_coords[2];
  548. int targets_ranks[4];
  549.  
  550. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  551.  
  552. int neighbour_offsets[ 4 ][ 2 ] = {
  553. { -1, 0 },{ 1, 0 },
  554. { 0, -1 },{ 0, 1 }
  555. };
  556.  
  557. //#pragma omp parallel for
  558. for(int i = 0; i < 4; i++){
  559.  
  560. int target_coords[2];
  561.  
  562. target_coords[ 0 ] = my_coords[ 0 ] + neighbour_offsets[ i ][ 0 ];
  563. target_coords[ 1 ] = my_coords[ 1 ] + neighbour_offsets[ i ][ 1 ];
  564.  
  565. if (target_coords[0] >= 0 && target_coords[0] < Gx && target_coords[1] >= 0 && target_coords[1] < Gy){
  566.  
  567. MPI_Cart_rank(comm, target_coords, &targets_ranks[ i ]);
  568.  
  569. }
  570. else{
  571. targets_ranks[i] = -1;
  572. }
  573. }
  574.  
  575. //Now we have rank for all targets
  576.  
  577. for(int axis = 0; axis < 2; axis++){
  578.  
  579. int parity_bit = (my_coords[ axis ]) % 2;
  580.  
  581. //if parity_bit == 0 then
  582. // zuerst mit links, dann mit rechts tauschen
  583. //elif parity_bit == 1:
  584. // zuerst mit rechts, dann mit links tauschen
  585.  
  586. for(int tmp = 0; tmp < 2; tmp++){
  587. parity_bit = 1 - parity_bit;
  588.  
  589. //target id in external_data und targets_ranks
  590. int target_idx = 2 * axis + (1 - parity_bit);
  591.  
  592. if (targets_ranks[target_idx] != -1){
  593.  
  594. int send_tag = 100000 + my_rank * 100 + axis * 10 + parity_bit;
  595. int recv_tag = 100000 + targets_ranks[ target_idx ] * 100 + axis * 10 + (1-parity_bit);
  596.  
  597. MPI_Status tmp_status;
  598.  
  599. if(my_rank != targets_ranks[ target_idx ]){
  600.  
  601. MPI_Sendrecv_replace(&external_data[ target_idx ][ 0 ], external_data[ target_idx ].size(),
  602. MPI_DOUBLE, targets_ranks[ target_idx ], send_tag, targets_ranks[ target_idx ], recv_tag,
  603. comm, &tmp_status);
  604.  
  605. }
  606. }
  607. }
  608. }
  609.  
  610. //MPI_Barrier(MPI_COMM_WORLD);
  611.  
  612. }
  613.  
  614. void DoIteration(bool &should_i_stop){
  615.  
  616. ComputeMatrixR();
  617.  
  618. //Now R Matrix is in internal_data
  619.  
  620. double tau = ComputeTauAndStopCase(should_i_stop);
  621.  
  622. //We have in our block jetzt:
  623. //in internal_data: R Matrix
  624. //in old_internal_data: w aus letzte iteration
  625. //and we have tau
  626. //jetzt koennen wir naechste w finden
  627.  
  628. ComputeNewW(tau);
  629.  
  630. SyncMPI();
  631.  
  632. }
  633.  
  634. double GetLocalIndex(int i, int j){
  635. //internal data
  636. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  637. return internal_data[ j + cur_block_size_y*i ];
  638. }
  639.  
  640. //external data
  641. //OX
  642. if((j >= 0) && (j < cur_block_size_y)){
  643.  
  644. if (i == -1)
  645. return external_data[ 0 ][ j ];
  646.  
  647. if (i == cur_block_size_x)
  648. return external_data[ 1 ][ j ];
  649.  
  650. }
  651.  
  652. //OY
  653. if((i >= 0) && (i < cur_block_size_x)){
  654.  
  655. if (j == -1)
  656. return external_data[ 2 ][ i ];
  657. if (j == cur_block_size_y)
  658. return external_data[ 3 ][ i ];
  659.  
  660. }
  661.  
  662. cout << "[ERROR]: bad local index" << endl;
  663.  
  664. return nan("");
  665. }
  666.  
  667. void SetLocalIndex(int i, int j, double v){
  668. if ((j >= 0) && (j < cur_block_size_y) && (i >= 0) && (i < cur_block_size_x)){
  669. internal_data[ j + cur_block_size_y*i ] = v;
  670. }else{
  671. cout << "[ERROR]: trying to set data outside the local area" << endl;
  672. }
  673. }
  674.  
  675. double CompareWithExact() {
  676.  
  677. double local_diff_norm = 0.0;
  678. double global_diff_norm = 0.0;
  679. /*
  680. if (my_rank == 0)
  681. cout << "[INFO]: Starting computing compare with exact" << endl;
  682. */
  683. vector <double> tmp_elements(cur_block_size_x * cur_block_size_y, 0.0);
  684.  
  685. //#pragma omp parallel for
  686. for (int i = 0; i < cur_block_size_x; ++i){
  687. for (int j = 0; j < cur_block_size_y; ++j){
  688.  
  689.  
  690. int current_node_global_offset_x = GetGlobalX(i),
  691. current_node_global_offset_y = GetGlobalY(j);
  692.  
  693. int glob_x = current_node_global_offset_x,
  694. glob_y = current_node_global_offset_y;
  695.  
  696. double tmp_elem = Get(glob_x, glob_y) - GetNodeFromExact(glob_x, glob_y);
  697.  
  698. //local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  699.  
  700. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  701.  
  702. tmp_elements [j + cur_block_size_y*i] = tmp_elem;
  703.  
  704. }
  705. }
  706.  
  707. //cout << "[INFO]: local max diff in " << cur_block_global_offset_x << " " << cur_block_global_offset_y << " " << local_diff_norm << endl;
  708.  
  709.  
  710. for (int i = 0; i < cur_block_size_x; ++i){
  711. for (int j = 0; j < cur_block_size_y; ++j){
  712.  
  713. double rho = 1.0;
  714.  
  715. int current_node_global_offset_x = GetGlobalX(i),
  716. current_node_global_offset_y = GetGlobalY(j);
  717.  
  718. int glob_x = current_node_global_offset_x,
  719. glob_y = current_node_global_offset_y;
  720.  
  721. switch (IsNodeInternalCornerOrSide(glob_x, glob_y)){
  722. case 2:
  723. case 4:
  724. case 6:
  725. case 8:
  726. //angle
  727. rho = 0.25;
  728. break;
  729. case 1:
  730. case 3:
  731. case 5:
  732. case 7:
  733. //side
  734. rho = 0.5;
  735. break;
  736. case 0:
  737. //internal
  738. rho = 1.0;
  739. break;
  740. default:
  741. cout << "[ERROR]: Bad global coords compute exact. Local:" << i << " " << j << ". Global:" << glob_x << " " << glob_y <<endl;
  742. }
  743.  
  744.  
  745. double tmp_elem = tmp_elements [j + cur_block_size_y*i];
  746.  
  747. local_diff_norm = max( fabs(tmp_elem), local_diff_norm);
  748. //local_diff_norm += rho * pow(tmp_elem, 2) * h_x * h_y;
  749. }
  750. }
  751.  
  752. MPI_Allreduce(&local_diff_norm, &global_diff_norm, 1, MPI_DOUBLE, MPI_MAX,
  753. MPI_COMM_WORLD);
  754.  
  755. //global_diff_norm = sqrt(global_diff_norm);
  756.  
  757.  
  758. return global_diff_norm;
  759. }
  760.  
  761. double GetCurrentNorm(){
  762. return current_norm;
  763. }
  764. };
  765.  
  766.  
  767.  
  768.  
  769. int main(int argc, char* argv[]){
  770.  
  771. const double x_left = -2, x_right = 3;
  772. const double y_bottom = -1, y_top = 4;
  773.  
  774. double time_start, time_stop;
  775. int N, Gx, Gy;
  776. int dim[2], period[2];
  777.  
  778. MPI_Comm comm;
  779.  
  780. //N - global grid size
  781. N = atoi(argv[1]);
  782.  
  783. //Gx
  784. Gx = dim[0] = atoi(argv[2]);
  785.  
  786. //Gy
  787. Gy = dim[1] = atoi(argv[3]);
  788.  
  789. period[0]=0;
  790. period[1]=0;
  791.  
  792.  
  793. MPI_Init(&argc, &argv);
  794.  
  795. time_start = MPI_Wtime();
  796.  
  797. int world_size;
  798.  
  799. int my_rank;
  800.  
  801. MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  802. MPI_Comm_size(MPI_COMM_WORLD, &world_size);
  803.  
  804.  
  805. if(my_rank == 0){
  806. cout << "[INFO]: N = " << N << endl;
  807. cout << "[INFO]: Gx = " << dim[0] << endl;
  808. cout << "[INFO]: Gy = " << dim[1] << endl;
  809. }
  810.  
  811.  
  812.  
  813. if(argc <= 3){
  814. if(my_rank == 0)
  815. cout << "[ERROR]: Usage: mpieval <N> <Gx> <Gy>" << endl;
  816.  
  817. MPI_Abort(MPI_COMM_WORLD, 1);
  818. }
  819.  
  820. if(Gx * Gy != world_size){
  821. if(my_rank == 0)
  822. cout << "[ERROR]: mpi world size is not equal to "<< Gx << "*" << Gy << endl;
  823.  
  824. MPI_Abort(MPI_COMM_WORLD, 1);
  825. }
  826.  
  827. MPI_Cart_create(MPI_COMM_WORLD, 2, dim, period, 1, &comm);
  828.  
  829. /*if(my_rank == 0)
  830. cout << "[INFO]: Cart created"<<endl;
  831. */
  832.  
  833. MPI_Comm_rank(comm, &my_rank);
  834.  
  835. int my_coords[2];
  836.  
  837. MPI_Cart_coords(comm, my_rank, 2, my_coords);
  838.  
  839.  
  840.  
  841.  
  842. 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);
  843.  
  844.  
  845. int iteration_num = 0;
  846.  
  847. bool should_i_stop = false;
  848.  
  849. while ( should_i_stop != true ){
  850.  
  851. w_func.DoIteration(should_i_stop);
  852.  
  853. if ( (my_rank == 0) && (iteration_num % 10000 == 0) ){
  854. cout << "[INFO]: Iteration " << iteration_num << " difference with previous is: "<< w_func.GetCurrentNorm() << endl;
  855. }
  856.  
  857. iteration_num++;
  858. }
  859.  
  860. //MPI_Barrier(MPI_COMM_WORLD);
  861.  
  862. //vector <double> local_elements;
  863.  
  864. double comparing = w_func.CompareWithExact();
  865.  
  866. if (my_rank == 0)
  867. cout << "[INFO]: Diff with exact solution: " << comparing << endl;
  868.  
  869.  
  870.  
  871. //myfile.close();
  872.  
  873. time_stop = MPI_Wtime();
  874. if( my_rank == 0 )
  875. cout << "Finished!" << endl
  876. << "Total iterations: " << iteration_num << endl
  877. << "Elapsed time: " << (time_stop - time_start) << endl
  878. << "Time per iteration: " << (time_stop - time_start) / double(iteration_num) << endl;
  879.  
  880. MPI_Finalize();
  881.  
  882. return 0;
  883. }
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!