Reduce a little overhead (#871)

* some small overhead improvements

* use result_type in rms_norm

* remove release force

* fix + use non-vector version

* revert compile change

* fix ops

* a little more overhead

* a little more cleanup and overhead
This commit is contained in:
Awni Hannun
2024-03-22 17:29:36 -07:00
committed by GitHub
parent 6ee1112f30
commit be98f4ab6b
13 changed files with 239 additions and 240 deletions

View File

@@ -54,6 +54,12 @@ struct PrintFormatter {
extern PrintFormatter global_formatter;
/** The type from promoting the arrays' types with one another. */
inline Dtype result_type(const array& a, const array& b) {
return promote_types(a.dtype(), b.dtype());
}
inline Dtype result_type(const array& a, const array& b, const array& c) {
return promote_types(result_type(a, b), c.dtype());
}
Dtype result_type(const std::vector<array>& arrays);
std::vector<int> broadcast_shapes(