int main(){
const char* rows[5] = {"SUM", "SUMSTR", "SUMCONSTSTR",
"COPY", "COPYCONSTSTR"};
const char* cols[1] = {"n=1e9 doubles"};
const int n = 1000*1000*1000;
double *a, *b;
init_mem(a, b, n);
double bw[5];
bw[0] = time(a, b, n, SUM);
bw[1] = time(a, b, n, SUMSTR);
bw[2] = time(a, b, n, SUMCONSTSTR);
bw[3] = time(a, b, n, COPY);
bw[4] = time(a, b, n, COPYCONSTSTR);
verify_dir("DBG/");
char fname[200];
sprintf(fname, "DBG/time_sc_stride_%d.txt", STR);
link_cout(fname);
Table tbl;
tbl.dim(5, 1);
tbl.rows(rows);
tbl.cols(cols);
tbl.data(bw);
char banner[200];
sprintf(banner, "memory bw in bytes/cycle, stride = %d\n", STR);
tbl.print(banner);
unlink_cout();
release_mem(a, b);
}
int main(){
enum leib_enum flag[7] = {LEIB, PLL, PLLX,
FOR, FORCHUNK, PLLFOR,
SCTN};
const char* rows[7] = {"serial", "parallel", "parallelx",
"for", "forchunk", "parallelfor",
"section"};
const char* cols[2] = {"pi", "cycles/term"};
const int nthreads = atoi(getenv("OMP_NUM_THREADS"));
std::cout<<"num of threads = "<<nthreads<<std::endl;
const long n = 1l*1000*1000*1000*10;
double data[14];
struct leib_struct ans;
for(int i=0; i < 7; i++){
ans = time_leibniz(n, nthreads, flag[i]);
data[i] = ans.pi;
data[i+7] = ans.cycles/n;
}
Table tbl;
tbl.dim(7, 2);
tbl.rows(rows);
tbl.cols(cols);
tbl.data(data);
verify_dir("DBG");
link_cout("DBG/time_leibniz.txt");
char banner[200];
sprintf(banner, "sum of %ld terms of Leibniz", n);
tbl.print(banner);
unlink_cout();
}
int main(){
int rank, nprocs;
mpi_initialize(rank, nprocs);
mpi_print_name(rank);
const char *rows[6] = {"1e3", "1e4", "1e5",
"1e6", "1e7", "1e8"};
int nlist[6] = {1000, 1000*10, 1000*100,
1000*1000, 1000*1000*10, 1000*1000*100};
const char* cols[2] = {"cycles", "b/w"};
double data[12];
for(int i=0; i < 6; i++){
data[i] = time(rank, nprocs, nlist[i]);
data[i+6] = 32.0*nlist[i]/data[i];
}
if(rank == 0){//output table
char fname[200];
verify_dir("DBG");
sprintf(fname, "DBG/time_cycle_NP%d.txt", nprocs);
link_cout(fname);
Table tbl;
tbl.dim(6, 2);
tbl.rows(rows);
tbl.cols(cols);
tbl.data(data);
tbl.print();
unlink_cout();
}
mpi_finalize();
}
int main(){
int dim[4] = {100, 1000, 2000, 6000};
int count[4] = {4000, 40, 10, 1};
verify_dir("DBG/");
const char* fname = "DBG/cpp-mult.txt";
link_cout(fname);
for(int i=0; i < 4; i++){
timecppmult(dim[i], count[i]);
}
unlink_cout();
}
int main(){
const int nrows = 12;
int nlist[nrows] = {1, 2, 4,
8, 16, 24,
32, 40, 60,
1000, 10*1000, 100*1000};
const char *rows[nrows] = {"1", "2", "4",
"8", "16", "24",
"32", "40", "60",
"1000", "10*1000", "100*1000"};
const int ncols = 3;
const char *cols[ncols] = {"min", "median", "max"};
double cyc2dram[nrows*ncols];
StatVector stats(NTRIALS);
for(int i=0; i < nrows; i++){
measure_latency(nlist[i], stats);
cyc2dram[i+0*nrows] = stats.min();
cyc2dram[i+1*nrows] = stats.median();
cyc2dram[i+2*nrows] = stats.max();
}
verify_dir("DBG");
#ifdef MEMWALK
const char* fname = "DBG/latency_memwalk.txt";
#else
const char* fname = "DBG/latency_clflush.txt";
#endif
link_cout(fname);
Table tbl;
tbl.dim(nrows, ncols);
tbl.rows(rows);
tbl.cols(cols);
tbl.data(cyc2dram);
char banner[200];
sprintf(banner, "cycles to dram, ntrials = %d, when N pages accessed",
NTRIALS);
tbl.print(banner);
unlink_cout();
}
