int main(int argc,char **argv)
{
PetscErrorCode ierr;
PetscInt nthreads,i;
PetscInt *ranks;
PetscScalar *values;
PetscInitialize(&argc,&argv,(char *)0,help);
ierr = PetscThreadCommView(PETSC_COMM_WORLD,0);CHKERRQ(ierr);
ierr = PetscThreadCommGetNThreads(PETSC_COMM_WORLD,&nthreads);CHKERRQ(ierr);
ierr = PetscMalloc(nthreads*sizeof(PetscInt),&ranks);CHKERRQ(ierr);
ierr = PetscMalloc(nthreads*sizeof(PetscScalar),&values);CHKERRQ(ierr);
for(i=0;i < nthreads;i++) {
ranks[i] = i; values[i] = i;
}
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)kernel_func1,2,ranks,values);CHKERRQ(ierr);
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)kernel_func2,2,ranks,values);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
ierr = PetscFree(ranks);CHKERRQ(ierr);
ierr = PetscFree(values);CHKERRQ(ierr);
PetscFinalize();
return 0;
}
int main(int argc,char *argv[])
{
PetscErrorCode ierr;
int quantum, checktick();
int BytesPerWord;
int j, k;
double scalar=3.0, t, times[4][NTIMES];
PetscInitialize(&argc,&argv,0,help);
/* --- SETUP --- determine precision and check timing --- */
/*printf(HLINE);
printf("STREAM version $Revision: 5.9 $\n");
printf(HLINE); */
BytesPerWord = sizeof(double);
printf("This system uses %d bytes per DOUBLE PRECISION word.\n",BytesPerWord);
printf(HLINE);
#if defined(NO_LONG_LONG)
printf("Array size = %d, Offset = %d\n", N, OFFSET);
#else
printf("Array size = %llu, Offset = %d\n", (unsigned long long) N, OFFSET);
#endif
printf("Total memory required = %.1f MB.\n",(3.0 * BytesPerWord) * ((double)N / 1048576.0));
printf("Each test is run %d times, but only\n", NTIMES);
printf("the *best* time for each is used.\n");
printf(HLINE);
#if !STATIC_ALLOC
a = malloc((N+OFFSET)*sizeof(double));
b = malloc((N+OFFSET)*sizeof(double));
c = malloc((N+OFFSET)*sizeof(double));
#endif
#if WITH_PTHREADS
ierr = PetscThreadCommGetNThreads(PETSC_COMM_WORLD,&nworkThreads);CHKERRQ(ierr);
ierr = PetscMalloc((nworkThreads+1)*sizeof(PetscInt),&trstarts);CHKERRQ(ierr);
PetscInt Q,R,nloc;
PetscBool S;
Q = (N+OFFSET)/nworkThreads;
R = (N+OFFSET) - Q*nworkThreads;
trstarts[0] = 0;
for (j=0; j < nworkThreads; j++) {
S = (PetscBool)(j < R);
nloc = S ? Q+1 : Q;
trstarts[j+1] = trstarts[j]+nloc;
}
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_Initialize_Kernel,1,&scalar);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
# else
for (j=0; j<N; j++) {
a[j] = 1.0;
b[j] = 2.0;
c[j] = 0.0;
}
#endif
/*printf(HLINE);*/
/* Get initial value for system clock. */
if ((quantum = checktick()) >= 1) ;
/* printf("Your clock granularity/precision appears to be %d microseconds.\n", quantum); */
else quantum = 1;
/* printf("Your clock granularity appears to be less than one microsecond.\n"); */
t = mysecond();
#if WITH_PTHREADS
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_2A_Kernel,0);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#else
for (j = 0; j < N; j++) a[j] = 2.0E0 * a[j];
#endif
t = 1.0E6 * (mysecond() - t);
/* printf("Each test below will take on the order of %d microseconds.\n", (int)t);
printf(" (= %d clock ticks)\n", (int) (t/quantum));
printf("Increase the size of the arrays if this shows that\n");
printf("you are not getting at least 20 clock ticks per test.\n");
printf(HLINE);
*/
/* --- MAIN LOOP --- repeat test cases NTIMES times --- */
for (k=0; k<NTIMES; k++) {
times[0][k] = mysecond();
#if WITH_PTHREADS
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_Copy_Kernel,0);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#else
for (j=0; j<N; j++) c[j] = a[j];
#endif
times[0][k] = mysecond() - times[0][k];
times[1][k] = mysecond();
#if WITH_PTHREADS
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_Scale_Kernel,1,&scalar);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#else
for (j=0; j<N; j++) b[j] = scalar*c[j];
#endif
times[1][k] = mysecond() - times[1][k];
times[2][k] = mysecond();
#if WITH_PTHREADS
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_Add_Kernel,0);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#else
for (j=0; j<N; j++) c[j] = a[j]+b[j];
#endif
times[2][k] = mysecond() - times[2][k];
times[3][k] = mysecond();
#if WITH_PTHREADS
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)tuned_STREAM_Triad_Kernel,1,&scalar);CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#else
for (j=0; j<N; j++) a[j] = b[j]+scalar*c[j];
#endif
times[3][k] = mysecond() - times[3][k];
}
/* --- SUMMARY --- */
for (k=1; k<NTIMES; k++) /* note -- skip first iteration */
for (j=0; j<4; j++) {
avgtime[j] = avgtime[j] + times[j][k];
mintime[j] = MIN(mintime[j], times[j][k]);
maxtime[j] = MAX(maxtime[j], times[j][k]);
}
printf("Function Rate (MB/s) \n");
for (j=0; j<4; j++) {
avgtime[j] = avgtime[j]/(double)(NTIMES-1);
printf("%s%11.4f \n", label[j], 1.0E-06 * bytes[j]/mintime[j]);
}
/* printf(HLINE);*/
#if WITH_PTHREADS
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);CHKERRQ(ierr);
#endif
/* --- Check Results --- */
checkSTREAMresults();
/* printf(HLINE);*/
PetscFinalize();
return 0;
}
int main(int argc,char **argv)
{
PetscErrorCode ierr;
PetscInt i,j,k,N=100,**counters,tsize;
PetscInitialize(&argc,&argv,(char *)0,help);
ierr = PetscThreadCommView(PETSC_COMM_WORLD,PETSC_VIEWER_STDOUT_WORLD);
CHKERRQ(ierr);
ierr = PetscOptionsGetInt(PETSC_NULL,"-N",&N,PETSC_NULL);
CHKERRQ(ierr);
ierr = PetscThreadCommGetNThreads(PETSC_COMM_WORLD,&tsize);
CHKERRQ(ierr);
ierr = PetscMalloc(tsize*sizeof(*counters),&counters);
CHKERRQ(ierr);
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)CounterInit_kernel,1,counters);
CHKERRQ(ierr);
for (i=0; i<10; i++) {
PetscReal t0,t1;
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);
CHKERRQ(ierr);
ierr = PetscGetTime(&t0);
CHKERRQ(ierr);
for (j=0; j<N; j++) {
/* ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)CounterIncrement_kernel,1,counters);CHKERRQ(ierr); */
ierr = PetscThreadCommRunKernel1(PETSC_COMM_WORLD,(PetscThreadKernel)CounterIncrement_kernel,counters);
CHKERRQ(ierr);
}
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);
CHKERRQ(ierr);
ierr = PetscGetTime(&t1);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"Time per kernel: %g us\n",1e6*(t1-t0)/N);
CHKERRQ(ierr);
}
for (i=0; i<10; i++) {
PetscReal t0,t1;
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);
CHKERRQ(ierr);
ierr = PetscGetTime(&t0);
CHKERRQ(ierr);
for (j=0; j<N; j++) {
#pragma omp parallel num_threads(tsize)
{
PetscInt trank = omp_get_thread_num();
CounterIncrement_kernel(trank,counters);
}
}
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);
CHKERRQ(ierr);
ierr = PetscGetTime(&t1);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"OpenMP inline time per kernel: %g us\n",1e6*(t1-t0)/N);
CHKERRQ(ierr);
}
for (i=0; i<10; i++) {
PetscReal t0,t1;
ierr = PetscGetTime(&t0);
CHKERRQ(ierr);
for (j=0; j<N; j++) {
CounterIncrement_kernel(0,counters);
}
ierr = PetscGetTime(&t1);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"Serial inline time per single kernel: %g us\n",1e6*(t1-t0)/N);
CHKERRQ(ierr);
}
for (i=0; i<10; i++) {
PetscReal t0,t1;
ierr = PetscGetTime(&t0);
CHKERRQ(ierr);
for (j=0; j<N; j++) {
for (k=0; k<tsize; k++) CounterIncrement_kernel(k,counters);
}
ierr = PetscGetTime(&t1);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"Serial inline time per kernel: %g us\n",1e6*(t1-t0)/N);
CHKERRQ(ierr);
}
ierr = PetscThreadCommRunKernel(PETSC_COMM_WORLD,(PetscThreadKernel)CounterFree_kernel,1,counters);
CHKERRQ(ierr);
ierr = PetscThreadCommBarrier(PETSC_COMM_WORLD);
CHKERRQ(ierr);
ierr = PetscFree(counters);
CHKERRQ(ierr);
PetscFinalize();
return 0;
}