第4講計算劃分和任務分配

2、附件是熱傳導問題的串行程序示例。請根據(jù)該實例編寫pthread并行程序。serial: : atomic: :可以看出mtxserial: : atomic: : serial: : atomic: : serial: : atomic: : serial: : atomic: : serial: : atomic: : serial:found :foundatomic: :primescost=13.primescost=1.primescost=1.primesserial: : atomic: : 25—35的范圍時，并行性能略有提高。計算劃分的任務粒度越小，并行程序中任NXPROB*NYPROB*ntsu2。NXPROB，NYPROB，nts，thread_num。FILE:SerialHEAT2DExample-CThisexampleisbasedonatwo-dimensionalheatequation Thetemperatureiscomputedtobehighinthemiddleofthezeroatthe Theboundariesareheldatzerothe Duringthetime-step,anarraycontainingsisused;thesesalternatebetweenolddataandnew#include<stdio.h>#include<stdlib.h>#include<time.h>#include<unistd.h>//#include<math.h>#include<stdlib.h>#include#defineNXPROB#defineNYPROB#defineCLOCK_PER_SEC((clock_t)1000)#defineMax_Thread_Num256#define struct{floatcx;floatcy;intnts;}parms={0.1,0.1,intfloatinttotalThread;intiz,it;intposX,intcolumn, mtx;{ *pMyStatus=threads;intloc_posX,loc_posY;intix,while((*pMyStatus)!=pthread_self())pMyStatus++;loc_posX=posX;loc_posY=posY+=if(posY>={posX+=row;posY=1;}{for(ix=loc_posX;ix<loc_posX+row&&ix<NXPROB-1;{for(iy=loc_posY;iy<loc_posY+column&&iy<NYPROB-1;{u[1-iz][ix][iy]=u[iz][ix][iy]+parms.cx*(u[iz][ix+1][iy]+u[iz][ix-1][iy]-2.0*+parms.cy*(u[iz][ix][iy+1]+u[iz][ix][iy-1]-2.0*}}if(it>=parms.nts)}return(void*)0;}void{intix,voidinidat(),prtdat(),inti,for(ix=0;ix<=NXPROB-1;{u[1][ix][0]=}for(iy=0;iy<=NYPROB-1;{u[1][0][iy]=}totalThread=pthread_cond_init(&cMaster,NULL);pthread_mutex_init(&mtx,NULL);posX=1;posY=if(thread_num>Max_Thread_Num)thread_num=Max_Thread_Num;if(thread_num%2==1)thread_num-=1;column=(NYPROB-2+2)/row=(NXPROB-2+thread_num/2-1)/(thread_num/for(i=0;i<thread_num;while(totalThread!=thread_num)pthread_cond_wait(&cWorker,&mtx);iz=0;for(it=1;it<=parms.nts;{totalThread=while(totalThread!=thread_num)pthread_cond_wait(&cWorker,&mtx);iz=1-iz;}printf("%dthreadnum:%d\n",totalThread,thread_num);prtdat(NXPROB,NYPROB,&u[iz][0][0],"final_p.dat");for(i=0;i<thread_num;i++)pthread_join(threads[i],NULL);}int{structtimespects,te;clock_tbegin,end;doubleserial_cost,//floatintix,iy,iz,voidinidat(),prtdat(), inidat(NXPROB,NYPROB,u);{{{v[i][j][k]=}}}prtdat(NXPROB,NYPROB,u,"initial.dat");begin=clock();doublea=//printf("a:cost=%-15.10f\n",a/CLOCK_PER_SEC);clock_gettime(CLOCK_REALTIME,&ts);for(ix=0;ix<=NXPROB-1;{u[1][ix][0]=}for(iy=0;iy<=NYPROB-1;{u[1][0][iy]=} Iterateoveralliz=0;for(it=1;it<=parms.nts;{iz=1-}end=doubleb=clock_gettime(CLOCK_REALTIME,&te);doubleserial_costs=te.tv_sec-ts.tv_sec+(double)(te.tv_nsec-ts.tv_nsec)/NANO;printf("serial:cost=%-15.10f\n",serial_costs);//doubles_cost=b-//printf("serial:cost=%-15.10f\n",serial_cost);{{{u[i][j][k]=}}}begin=a=//printf("a:cost=%-15.10f\n",a/CLOCK_PER_SEC);doublesyns_cost=te.tv_sec-ts.tv_sec+(double)(te.tv_nsec-ts.tv_nsec)/NANO;printf("syn:cost=%-15.10f\n",syns_cost);printf("speedup=%-15.10f\n",serial_costs/syns_cost);end=clock();b=//printf("b:cost=%-15.10f\n",b/CLOCK_PER_SEC);syn_cost=b-a;} update(nx,ny,u1,u2)intnx,ny;float*u1,{intix,for(ix=1;ix<=nx-2;{for(iy=1;iy<=ny-2;{*(u2+ix*ny+iy)= parms.cx*(*(u1+(ix+1)*ny+iy)+*(u1+(ix-1)*ny+iy)2.0* )parms.cy*(*(u1+ix*ny+iy+1)+*(u1+ix*ny+iy-1)2.0* }}} inidat(nx,ny,u1)intnx,ny;/*floatu1[nx][ny];*/float*u1;{intix,for(ix=0;ix<=nx-1;{for(iy=0;iy<=ny-1;{/*u1[ix][iy]=(float)(ix*(nx-ix-1)*iy*(ny-iy-1));*(u1+ix*ny+iy)=(float)(ix*(nx-ix-1)*iy*(ny-iy-}}}*subroutineprtdat(nx,ny,u1,fnam)int