You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
During some host-side perf measurements in #3346 we've found out that keeping 2 dynamic vectors for an array of 5 elements is inefficient, since we are extensively creating and coping TensorDescriptor across the library.
To minimize host side overhead we need a simple vector-like container with statically allocate storage of "at most" 5 elements.
It has to be something like template <typename Type, size_t Capacity> class InlineVector;
The text was updated successfully, but these errors were encountered:
During some host-side perf measurements in #3346 we've found out that keeping 2 dynamic vectors for an array of 5 elements is inefficient, since we are extensively creating and coping TensorDescriptor across the library.
To minimize host side overhead we need a simple vector-like container with statically allocate storage of "at most" 5 elements.
It has to be something like
template <typename Type, size_t Capacity> class InlineVector;
The text was updated successfully, but these errors were encountered: