int main(int argc , char **argv)
{
#if PRINT_TIME
double start, end;
#endif
extract_arguments(argc, argv);
#if PRINT_TIME
start = time_so_far();
#endif
if(mr_init())
exit(EXIT_FAILURE);
start_threads();
wait_threads();
if(mr_reduce())
exit(EXIT_FAILURE);
#if PRINT_TIME
/* Done here, to avoid counting the printing... */
end = time_so_far();
#endif
if(mr_print())
exit(EXIT_FAILURE);
if(mr_destroy())
exit(EXIT_FAILURE);
#if PRINT_TIME
printf("Done in %g msec\n", end-start);
#endif
exit(EXIT_SUCCESS);
}
int
main(int argc, char *argv[])
{
final_data_kv_t final_vals;
int fd;
struct defsplitter_state ps;
char *fdata;
char *fname;
struct stat finfo;
int i;
int nprocs = 0, map_tasks = 0, ndisp = 5, reduce_tasks = 0, quiet = 0;
int c;
// Make sure a filename is specified
fname = argv[1];
if (argc < 2) {
lr_usage(argv[0]);
exit(EXIT_FAILURE);
}
while ((c = getopt(argc - 1, argv + 1, "p:l:m:r:q")) != -1) {
switch (c) {
case 'p':
nprocs = atoi(optarg);
break;
case 'l':
ndisp = atoi(optarg);
break;
case 'm':
map_tasks = atoi(optarg);
break;
case 'r':
reduce_tasks = atoi(optarg);
break;
case 'q':
quiet = 1;
break;
default:
lr_usage(argv[0]);
exit(EXIT_FAILURE);
break;
}
}
// Read in the file
assert((fd = open(fname, O_RDONLY)) >= 0);
// Get the file info (for file length)
assert(fstat(fd, &finfo) == 0);
// Memory map the file
assert((fdata = mmap(0, finfo.st_size + 1,
PROT_READ | PROT_WRITE, MAP_PRIVATE, fd,
0)) != MAP_FAILED);
// Setup scheduler args
mr_param_t mr_param;
memset(&mr_param, 0, sizeof(mr_param_t));
mr_param.nr_cpus = nprocs;
mr_param.map_func = linear_regression_map;
mr_param.app_arg.atype = atype_mapreduce;
mr_param.app_arg.mapreduce.reduce_func = linear_regression_reduce;
mr_param.app_arg.mapreduce.combiner = linear_regression_combine;
memset(&final_vals, 0, sizeof(final_vals));
mr_param.app_arg.mapreduce.results = &final_vals;
defsplitter_init(&ps, fdata, finfo.st_size -
(finfo.st_size % sizeof(POINT_T)), map_tasks,
sizeof(POINT_T));
mr_param.split_func = defsplitter; // Array splitter;
mr_param.split_arg = &ps; // Array to regress
mr_param.part_func = linear_regression_partition;
mr_param.key_cmp = intkeycmp;
//#define PREFETCH
#ifdef PREFETCH
int sum = 0;
for (int i = 0; i < finfo.st_size; i += 4096) {
sum += fdata[i];
}
printf("ignore this %d\n", sum);
#endif
mr_print(!quiet, "Linear regression: running...\n");
assert(mr_run_scheduler(&mr_param) == 0);
mr_print_stats();
long long n;
double a, b, xbar, ybar, r2;
long long SX_ll = 0, SY_ll = 0, SXX_ll = 0, SYY_ll = 0, SXY_ll = 0;
// ADD UP RESULTS
for (i = 0; i < final_vals.length; i++) {
keyval_t *curr = &final_vals.data[i];
switch ((long int) curr->key) {
case KEY_SX:
SX_ll = (long long) curr->val;
break;
case KEY_SY:
SY_ll = (long long) curr->val;
break;
case KEY_SXX:
SXX_ll = (long long) curr->val;
break;
case KEY_SYY:
SYY_ll = (long long) curr->val;
break;
case KEY_SXY:
SXY_ll = (long long) curr->val;
break;
default:
// INVALID KEY
assert(0);
break;
}
}
double SX = (double) SX_ll;
double SY = (double) SY_ll;
double SXX = (double) SXX_ll;
double SYY = (double) SYY_ll;
double SXY = (double) SXY_ll;
n = (long long) finfo.st_size / sizeof(POINT_T);
b = (double) (n * SXY - SX * SY) / (n * SXX - SX * SX);
a = (SY_ll - b * SX_ll) / n;
xbar = (double) SX_ll / n;
ybar = (double) SY_ll / n;
r2 = (double) (n * SXY - SX * SY) * (n * SXY -
SX * SY) / ((n * SXX -
SX * SX) * (n * SYY -
SY * SY));
if (!quiet) {
printf("%2d Linear Regression Results:\n", nprocs);
printf("\ta = %lf\n", a);
printf("\tb = %lf\n", b);
printf("\txbar = %lf\n", xbar);
printf("\tybar = %lf\n", ybar);
printf("\tr2 = %lf\n", r2);
printf("\tSX = %lld\n", SX_ll);
printf("\tSY = %lld\n", SY_ll);
printf("\tSXX = %lld\n", SXX_ll);
printf("\tSYY = %lld\n", SYY_ll);
printf("\tSXY = %lld\n", SXY_ll);
}
free(final_vals.data);
assert(munmap(fdata, finfo.st_size + 1) == 0);
assert(close(fd) == 0);
mr_finalize();
return 0;
}
