есть прога "Тест эффективности основных операций MPI"
не могу найти где скачать mpi.h чтоб добавить в Microsoft Visual Studio 2010,соответсвенно не могу её откомпелировать.у кого установлены нормально MPI,можете пожалуйста проверить,работает вообще прога,или нет.
| Код |
#include<mpi.h>
#include<stdlib.h>
#include<stdio.h>
#define Wtime MPI_Wtime
#define MEGABYTE (1024*1024)
#define BUFFER_SIZE (16*MEGABYTE)
#define TAG1 17
#define MASTER_RANK 0
#define MAX_REQUESTS 16
#define MILLION 1000000.0
static int NTIMES = 1000;
static int NTIMES_PREC = 10000;
static int REPEAT_ALL = 1;
static int REPEAT_EACH_TEST = 1;
#define NTIMES_TIMING NTIMES_PREC
#define NTIMES_BARRIER NTIMES_PREC
#define NTIMES_LATENCY NTIMES_PREC
#define TEST_SENDRECV 0
#define MASTER if(i_am_the_master)
int nproc,myid,rank,pool_size;
int i_am_the_master = 0;
static int msglen_min = 1;
static int msglen_max = 1000;
static int msglen_multiplier = 10;
static int msglen_step = 0;
int *buf,*obuf;
FILE *out;
/* йНЕОБ РТПГЕУУПТПЧ, ЙМЙ ХЪМПЧ, ОБ ЛПФПТЩИ ТБВПФБАФ РТПГЕУУЩ ЪБДБЮЙ */
static char nodename[32];
static char* nodenames;
int namelen = 32;
void Test_MPI_Routines(MPI_Comm comm,int len);
int checktick();
int main(int argc,char *argv[]) {
int i,k;
int bufsize = BUFFER_SIZE;
int msglen;
double t_start,t_end;
char processor_name[128],output_file_name[128];
int namelen = 128;
MPI_Status status;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&nproc);
MPI_Comm_rank(MPI_COMM_WORLD,&myid);
i_am_the_master = (myid == MASTER_RANK);
rank = myid;
pool_size = nproc;
out = stdout;
t_start = Wtime();
MASTER printf("MPItest/C 2.1: running %d processes...\n",nproc);
if(nproc < 2) {
MASTER printf("MPItest expects at least 2 processes\n");
MPI_Finalize();
exit(0);
}
MASTER /* read command line parameters */
for(i = 1; i < argc; i ++)
switch(argv[i][0]) {
case 'm': msglen_min = atoi(argv[i]+1); break;
case 'M': msglen_max = atoi(argv[i]+1); break;
case 'K': msglen_multiplier = atoi(argv[i]+1); msglen_step = 0; break;
case 'T': NTIMES = atoi(argv[i]+1); break;
case 't': NTIMES_PREC = atoi(argv[i]+1); break;
case 'R': REPEAT_ALL = atoi(argv[i]+1); break;
case 'r': REPEAT_EACH_TEST = atoi(argv[i]+1); break;
case 'b': bufsize = (atoi(argv[i]+1)*MEGABYTE)/sizeof(int); break;
case 'o':
strcpy(output_file_name,argv[i]+1);
out = fopen(output_file_name,"a+");
if(out == NULL)
{
printf("WARNING: Can't open or create %s, writing to stdout\n",output_file_name);
out = stdout;
}
else
printf("Writing output to %s\n",output_file_name);
break;
default: fprintf(stderr,"WARNING: unrecognized option: %s\n",argv[i]);
break;
}
MPI_Bcast(&NTIMES,1,MPI_INT,MASTER_RANK,MPI_COMM_W ORLD);
MPI_Bcast(&NTIMES_PREC,1,MPI_INT,MASTER_RANK,MPI_C OMM_WORLD);
MPI_Bcast(&REPEAT_ALL,1,MPI_INT,MASTER_RANK,MPI_CO MM_WORLD);
MPI_Bcast(&REPEAT_EACH_TEST,1,MPI_INT,MASTER_RANK, MPI_COMM_WORLD);
MPI_Bcast(&msglen_min,1,MPI_INT,MASTER_RANK,MPI_CO MM_WORLD);
MPI_Bcast(&msglen_max,1,MPI_INT,MASTER_RANK,MPI_CO MM_WORLD);
MPI_Bcast(&msglen_multiplier,1,MPI_INT,MASTER_RANK ,MPI_COMM_WORLD);
MPI_Bcast(&bufsize,1,MPI_INT,MASTER_RANK,MPI_COMM_ WORLD);
MPI_Barrier(MPI_COMM_WORLD);
/* рЩФБЕНУС ОБРЕЮБФБФШ ЙНЕОБ ЧУЕИ ХЪМПЧ, ОБ ЛПФПТЩИ ТБВПФБЕФ ФЕУФ */
MPI_Get_processor_name(nodename,&namelen);
if(i_am_the_master)
{
nodenames = (char*) malloc(35*pool_size);
strcpy(nodenames,nodename);
/* зПМПЧОПК РТПГЕУУ РПМХЮБЕФ ЙНЕОБ ПУФБМШОЩИ Й УПВЙТБЕФ ЙИ Ч ПДОХ УФТПЛХ */
for(i = 1; i < pool_size; i ++)
{
strcat(nodenames," ");
MPI_Recv(nodename,32,MPI_CHAR,i,2003,MPI_COMM_WORL D,&status);
strcat(nodenames,nodename);
}
fprintf(out,"\tNodes: %s\n",nodenames);
fflush(out);
}
else
MPI_Send(nodename,32,MPI_CHAR,0,2003,MPI_COMM_WORL D);
fflush(out);
if(bufsize < msglen_max * nproc) {
MASTER printf("Too small buffer size - %d. At least %d words required\n",bufsize,msglen_max * nproc);
MPI_Finalize();
exit(0);
}
buf = (int*) malloc(sizeof(int)*bufsize);
obuf = (int*) malloc(sizeof(int)*bufsize);
if(buf == NULL || obuf == NULL)
{
MASTER printf("Too small buffer size. At least %d words required\n",msglen_max * nproc);
MPI_Finalize();
exit(0);
}
MASTER fprintf(out,"Testing basic MPI routines.\n");
MASTER fprintf(out,"T = %d, t = %d\n",NTIMES,NTIMES_PREC);
for(k = 0; k < REPEAT_ALL; k ++)
{
MASTER fprintf(out,"\n--- Iteration %d ---\n",k);
Test_MPI_Routines(MPI_COMM_WORLD,0);
for(msglen = msglen_min; msglen <= msglen_max; msglen *= msglen_multiplier)
{
MASTER fprintf(out,"\n--- Size %d, Iteration %d ---\n",msglen,k);
Test_MPI_Routines(MPI_COMM_WORLD,msglen);
}
}
t_end = Wtime();
MASTER printf("MPItest/C complete in %g sec.\n",t_end-t_start);
MASTER fclose(out);
MPI_Finalize();
return 0;
}
void Test_MPI_Routines(MPI_Comm comm,int msglen)
{
int i,k;
int nproc,myid,tick;
int i_am_the_master = 0;
double t, t1, t2, tt, dt, lat, tb, best_time;
MPI_Status status, statuses[MAX_REQUESTS];
MPI_Request request, requests[MAX_REQUESTS];
MPI_Comm_size(MPI_COMM_WORLD,&nproc);
MPI_Comm_rank(MPI_COMM_WORLD,&myid);
i_am_the_master = (myid == MASTER_RANK);
MPI_Barrier(comm);
/* Initialization of buffer data */
for(i = 0; i < BUFFER_SIZE; i ++)
buf[i] = myid;
if(msglen == 0)
{
/* ----- Measuring time for a MPI_Wtime() call ------ */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
dt = 0;
t1 = Wtime();
MASTER for(i = 0; i < NTIMES_TIMING; i ++)
{
t = Wtime();
}
t2 = Wtime();
dt = t2 - t1;
dt /= (double) NTIMES_TIMING;
t = dt;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER tick = checktick();
MASTER fprintf(out,"%6.1f\tTIMING\tclock tick = %d\n",t,tick);
MASTER fflush(out);
/* ----------- Testing barrier synchronization ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
tb = 0;
t1 = Wtime();
for(i = 0; i < NTIMES_BARRIER; i ++)
{
MPI_Barrier(comm);
}
t2 = Wtime();
tb = t2 - t1;
tb /= (double) NTIMES_BARRIER;
t = tb;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tBARRIER\n",t);
fflush(out);
/* ----------- Testing signal send/receive ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
if(i_am_the_master)
{
t1 = Wtime();
for(i = 0; i < NTIMES_LATENCY; i ++)
{
MPI_Send(NULL,0,MPI_INT,1,TAG1,comm);
MPI_Recv(NULL,0,MPI_INT,1,TAG1,comm,&status);
}
t2 = Wtime();
t = t2 - t1;
}
else if(myid == 1)
for(i = 0; i < NTIMES_LATENCY; i ++)
{
MPI_Recv(NULL,0,MPI_INT,MASTER_RANK,TAG1,comm,&sta tus);
MPI_Send(NULL,0,MPI_INT,MASTER_RANK,TAG1,comm);
}
t /= (double) NTIMES_LATENCY;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tSIGNAL SEND & RECEIVE (double latency)\n",t);
fflush(out);
} /* ЛПОЕГ ВМПЛБ ДМС msglen = 0 */
else
{
/* ----------- Testing MPI_Reduce ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Reduce(buf,obuf,msglen,MPI_INT,MPI_SUM,MASTER_ RANK,comm);
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tREDUCE (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing MPI_Allreduce ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Allreduce(buf,obuf,msglen,MPI_INT,MPI_SUM,comm );
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tALLREDUCE (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing broadcast routine ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Bcast(buf,msglen,MPI_INT,MASTER_RANK,comm);
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tBROADCAST (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing gather routine ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Gather(buf,msglen,MPI_INT,obuf,msglen,MPI_INT, MASTER_RANK,comm);
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tGATHER (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing Allgather routine ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Allgather(buf,msglen,MPI_INT,obuf,msglen,MPI_I NT,comm);
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tALLGATHER (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing All-to-All routine ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Alltoall(buf,msglen,MPI_INT,obuf,msglen,MPI_IN T,comm);
}
t2 = Wtime();
t = (t2 - t1);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tALL-TO-ALL (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing non-blocking send with wait ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
if(i_am_the_master)
{
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Isend(buf,msglen,MPI_INT,1,TAG1,comm,&request) ;
MPI_Wait(&request,&status);
}
t2 = Wtime();
t = (t2 - t1);
}
else if(myid == 1)
for(i = 0; i < NTIMES; i ++)
{
MPI_Irecv(obuf,msglen,MPI_INT,MASTER_RANK,TAG1,com m,&request);
MPI_Wait(&request,&status);
}
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tISEND & WAIT (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing blocking vector send ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
if(i_am_the_master)
{
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Send(buf,msglen,MPI_INT,1,TAG1,comm);
}
t2 = Wtime();
t = (t2 - t1);
}
else if(myid == 1)
for(i = 0; i < NTIMES; i ++)
MPI_Recv(obuf,msglen,MPI_INT,MASTER_RANK,TAG1,comm ,&status);
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tSEND (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing simultaneous send & receive ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
if(i_am_the_master)
{
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Sendrecv(buf,msglen,MPI_INT,1,TAG1,obuf,msglen ,MPI_INT,1,TAG1,comm,statuses);
}
t2 = Wtime();
t = (t2 - t1);
}
else if(myid == 1)
for(i = 0; i < NTIMES; i ++)
MPI_Sendrecv(buf,msglen,MPI_INT,MASTER_RANK,TAG1,o buf,msglen,MPI_INT,MASTER_RANK,TAG1,comm,statuses) ;
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tSENDRECV (%d)\n",t,msglen);
fflush(out);
/* ----------- Testing sequential send & receive ---------- */
best_time = 0;
for(k = 0; k < REPEAT_EACH_TEST; k ++)
{
t = 0;
MPI_Barrier(comm);
if(i_am_the_master)
{
t1 = Wtime();
for(i = 0; i < NTIMES; i ++)
{
MPI_Send(buf,msglen,MPI_INT,1,TAG1,comm);
MPI_Recv(obuf,msglen,MPI_INT, 1,TAG1,comm,&status);
}
t2 = Wtime();
t = (t2 - t1);
}
else if(myid == 1)
for(i = 0; i < NTIMES; i ++)
{
MPI_Recv(obuf,msglen,MPI_INT,MASTER_RANK,TAG1,comm ,&status);
MPI_Send(buf,msglen,MPI_INT,MASTER_RANK,TAG1,comm) ;
}
t /= (double) NTIMES;
if((t < best_time) || (best_time == 0))
best_time = t;
}
t = best_time * MILLION;
MASTER fprintf(out,"%6.1f\tSEND & RECEIVE (%d)\n",t,msglen);
fflush(out);
}
}
# define M 2000
# ifndef MIN
# define MIN(x,y) ((x)<(y)?(x)y))
# endif
# ifndef MAX
# define MAX(x,y) ((x)>(y)?(x)y))
# endif
int checktick()
{
int i, minDelta, Delta;
double t1, t2, timesfound[M];
/* Collect a sequence of M unique time values from the system. */
for (i = 0; i < M; i++) {
t1 = Wtime();
while( ((t2=Wtime()) - t1) < 1.0E-6 )
;
timesfound[i] = t1 = t2;
}
/*
* Determine the minimum difference between these M values.
* This result will be our estimate (in microseconds) for the
* clock granularity.
*/
minDelta = 1000000;
for (i = 1; i < M; i++) {
Delta = (int)( 1.0E6 * (timesfound[i]-timesfound[i-1]));
minDelta = MIN(minDelta, MAX(Delta,0));
}
return(minDelta);
}
|
M
JackYF | Нам не надо простыней -- пользуйся кнопкой "код". |
|
