// Like 'search', but do in parallel across 2 nodes with 2 threads each
int psearch(uint8_t *str, int len, const unsigned char *hash)
{
if (len <= BLOCKLEN) {
search_args a = { {0}, len, 0, len, hash };
strcpy((char*)a.str, (char*)str);
search(&a);
return found;
}
// Iterate over blocks, searching 4 blocks at a time in parallel
int done = 0;
do {
int i;
search_args a[nthreads];
for (i = 0; i < nthreads; i++) {
strcpy((char*)a[i].str, (char*)str);
//cprintf("forking child to check '%s'\n", str);
a[i].len = len;
a[i].lo = 0;
a[i].hi = BLOCKLEN;
a[i].hash = hash;
bench_fork(i, search, &a[i]);
done |= incstr(str, BLOCKLEN, len);
}
for (i = 0; i < nthreads; i++)
bench_join(i); // collect results
if (found)
return 1;
} while (!done);
return 0; // no match at this string length
}
void
matmult(int nbi, int nbj, int dim)
{
assert(dim >= 1 && dim <= MAXDIM);
assert(nbi >= 1 && nbi <= dim); assert(dim % nbi == 0);
assert(nbj >= 1 && nbj <= dim); assert(dim % nbj == 0);
int nth = nbi*nbj;
assert(nth >= 1 && nth <= 256);
int bi,bj;
struct tharg arg[256];
// Fork off a thread to compute each cell in the result matrix
for (bi = 0; bi < nbi; bi++)
for (bj = 0; bj < nbj; bj++) {
int child = bi*nbi + bj;
arg[child].bi = bi;
arg[child].bj = bj;
arg[child].nbi = nbi;
arg[child].nbj = nbj;
arg[child].dim = dim;
bench_fork(child, blkmult, &arg[child]);
}
// Now go back and merge in the results of all our children
for (bi = 0; bi < nbi; bi++)
for (bj = 0; bj < nbj; bj++) {
int child = bi*nbi + bj;
bench_join(child);
}
}
void writetest(struct args *a)
{
int th;
for (th = 0; th < a->nthreads; th++) {
a->thread = th;
bench_fork(th, writefun, &a);
}
for (th = 0; th < a->nthreads; th++) {
bench_join(th);
}
}
int main(int argc, char *argv[])
{
bench_fork();
bench_thread();
bench_context_fork();
bench_context_thread();
return 0;
}
