Pointer Prefetch - Intel PXA255 User Manual

Optimization Guide
The last two iterations will prefetch superfluous data. The problem can be avoid by unrolling the
end of the loop.
for(i=0; i<NMAX-2; i++)
{
prefetch(data[i+2]);
sum += data[i];
}
sum += data[NMAX-2];
sum += data[NMAX-1];
Unfortunately, prefetch loop unrolling does not work on loops with indeterminate iterations.
A.4.4.9.

Pointer Prefetch

Not all looping constructs contain induction variables. However, prefetching techniques can still be
applied. Consider the following linked list traversal example:
while(p) {
do_something(p->data);
p = p->next;
}
The pointer variable p becomes a pseudo induction variable and the data pointed to by p->next can
be prefetched to reduce data transfer latency for the next iteration of the loop. Linked lists should
be converted to arrays as much as possible.
while(p) {
prefetch(p->next);
do_something(p->data);
p = p->next;
}
Recursive data structure traversal is another construct where prefetching can be applied. This is
similar to linked list traversal. Consider the following pre-order traversal of a binary tree:
preorder(treeNode *t) {
if(t) {
process(t->data);
preorder(t->left);
preorder(t->right);
}
}
The pointer variable t becomes the pseudo induction variable in a recursive loop. The data
structures pointed to by the values t->left and t->right can be prefetched for the next iteration of
the loop.
preorder(treeNode *t) {
if(t) {
prefetch(t->right);
prefetch(t->left);
process(t->data);
preorder(t->left);
preorder(t->right);
}
}
Note the order reversal of the prefetches in relationship to the usage. If there is a cache conflict and
data is evicted from the cache then only the data from the first prefetch is lost.
A-22 Intel® XScale™ Microarchitecture User's Manual