Rewrite hash and == function for hb_array_t as a plain loop.
Profling showed that the current implementation were accounting for nearly all processing time in some cases. These implementations look to be about 10x faster.
This commit is contained in:
parent
831daf4c76
commit
eab191fd9c
|
@ -332,20 +332,21 @@ hb_sorted_array (T (&array_)[length_])
|
|||
template <typename T>
|
||||
bool hb_array_t<T>::operator == (const hb_array_t<T> &o) const
|
||||
{
|
||||
return length == o.length &&
|
||||
+ hb_zip (*this, o)
|
||||
| hb_map ([] (hb_pair_t<T&, T&> &&_) { return _.first == _.second; })
|
||||
| hb_all
|
||||
;
|
||||
if (o.length != this->length) return false;
|
||||
for (unsigned int i = 0; i < this->length; i++) {
|
||||
if (this->arrayZ[i] != o.arrayZ[i]) return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
uint32_t hb_array_t<T>::hash () const
|
||||
{
|
||||
return
|
||||
+ hb_iter (*this)
|
||||
| hb_map (hb_hash)
|
||||
| hb_reduce ([] (uint32_t a, uint32_t b) { return a * 31 + b; }, 0)
|
||||
;
|
||||
uint32_t current = 0;
|
||||
for (unsigned int i = 0; i < this->length; i++) {
|
||||
current = current * 31 + hb_hash (this->arrayZ[i]);
|
||||
}
|
||||
return current;
|
||||
}
|
||||
|
||||
typedef hb_array_t<const char> hb_bytes_t;
|
||||
|
|
Loading…
Reference in New Issue