#include <TensorExecutor.h>
Public Types | |
enum | { value = false } |
Evaluating TensorBroadcastingOp via coefficient of packet path is extremely expensive. If expression has at least one broadcast op in it, and it supports block based evaluation, we always prefer it, even for the small tensors. For all other tileable ops, block evaluation overhead for small tensors (fits into L1) is too large, and we fallback on vectorized evaluation.
Definition at line 42 of file TensorExecutor.h.
anonymous enum |
Enumerator | |
---|---|
value |
Definition at line 43 of file TensorExecutor.h.