void hex_volume_internal(Ioss::ElementBlock *block, const std::vector<double> &coordinates,
const std::vector<T> &connectivity, size_t nelem)
{
const double one12th = 1.0 / 12.0;
double gradop12x[24];
double x[8], y[8], z[8];
std::vector<double> volume(nelem);
size_t t1 = timer();
size_t count = 0;
for (size_t ielem = 0; ielem < nelem; ++ielem) {
if (count++ >= nelem / 100) {
OUTPUT << ".";
count = 0;
}
for (size_t j = 0; j < 8; j++) {
size_t node = connectivity[ielem * 8 + j] - 1;
x[j] = coordinates[node * 3 + 0];
y[j] = coordinates[node * 3 + 1];
z[j] = coordinates[node * 3 + 2];
}
comp_grad12x(&gradop12x[0], x, y, z);
const double volume12x = dot8(x, &gradop12x[0]);
volume[ielem] = volume12x * one12th;
}
size_t t2 = timer();
if (nelem > 0) {
OUTPUT << "\n"
<< std::setw(12) << block->name() << "\tMin volume = " << std::setw(12)
<< *std::min_element(volume.begin(), volume.end()) << " Max volume = " << std::setw(12)
<< *std::max_element(volume.begin(), volume.end()) << " Elements = " << std::setw(12)
<< nelem << " Time/Elem = " << double(t2 - t1) / nelem << " micro-sec\n";
}
}
int main(int argc, char *argv[])
{
int i, j;
unsigned char *z;
unsigned char r[100*1000];
double drand48();
unsigned char **threaded_topics, **threaded_docs;
float t0, t1;
float tmp[280];
init_compand();
threaded_docs = calloc(100, sizeof(threaded_docs[0]));
for (i=0;i<100;i++) {
threaded_docs[i] = calloc(280, sizeof(threaded_docs[0][0]));
for (j=0;j<280;j++) {
tmp[j] = gauss_dev(0, 1);
}
scv_normalize(tmp);
compress_vector(threaded_docs[i], tmp, 280);
j = drand48() * (i+1);
z = threaded_docs[i];
threaded_docs[i] = threaded_docs[j];
threaded_docs[j] = z;
}
threaded_topics = calloc(1000, sizeof(threaded_topics[0]));
for (i=0;i<1000;i++) {
threaded_topics[i] = calloc(280, sizeof(threaded_topics[0][0]));
for (j=0;j<280;j++) {
tmp[j] = gauss_dev(0, 1);
}
scv_normalize(tmp);
compress_vector(threaded_topics[i], tmp, 280);
j = drand48() * (i+1);
z = threaded_topics[i];
threaded_topics[i] = threaded_topics[j];
threaded_topics[j] = z;
}
/* convectis style threaded matrix
by vector
by matrix
*/
t0 = millitime();
for (i=0;i<100;i++) {
for (j=0;j<1000;j++) {
r[i*1000+j] = dot8(threaded_topics[j], threaded_docs[i], 280);
}
}
t1 = millitime();
printf("%.3f seconds for %d convectis style categorizations\n",
t1-t0, 100);
t0 = millitime();
for (j=0;j<1000;j++) {
for (i=0;i<100;i++) {
r[i*1000+j] = dot8(threaded_topics[j], threaded_docs[i], 280);
}
}
t1 = millitime();
printf("%.3f seconds for %d convectis matrix categorizations\n",
t1-t0, 100);
return 0;
}