ZigZag - Deep Learning Hardware Design Space Exploration
This repository presents the novel version of our tried-and-tested hardware Architecture-Mapping Design Space Exploration (DSE) Framework for Deep Learning (DL) accelerators. ZigZag bridges the gap between algorithmic DL decisions and their acceleration cost on specialized accelerators through a fast and accurate hardware cost estimation.
|
Class that collects all the data transfer rate (periodic) information for each DTL (data transfer link). More...
Public Member Functions | |
def | __init__ (self, int real_cycle, int allowed_cycle, int period, int period_count, LayerOperand layer_op, int mem_lv, DataDirection mov_dir) |
def | __str__ (self) |
def | __repr__ (self) |
def | __hash__ (self) |
Public Attributes | |
real_cycle | |
allowed_cycle | |
period | |
period_count | |
The total number of period across the whole NN layer computation. More... | |
stall_or_slack_per_period | |
stall_or_slack | |
mem_updating_window | |
Class that collects all the data transfer rate (periodic) information for each DTL (data transfer link).
def __init__ | ( | self, | |
int | real_cycle, | ||
int | allowed_cycle, | ||
int | period, | ||
int | period_count, | ||
LayerOperand | layer_op, | ||
int | mem_lv, | ||
DataDirection | mov_dir | ||
) |
real_cycle | Within each period, the actual number of cycles used for transferring the amount of data, depended on the memory bw and the data amount to be transferred at that memory level. |
period | The turnaround cycle at that memory level, which equals to the product of all the temporal loops of current and below memory level. |
period_count | The total number of period across the whole NN layer computation. |
def __hash__ | ( | self | ) |
def __repr__ | ( | self | ) |
def __str__ | ( | self | ) |
allowed_cycle |
mem_updating_window |
period |
period_count |
The total number of period across the whole NN layer computation.
real_cycle |
stall_or_slack |
stall_or_slack_per_period |