CostModelEvaluation Class Reference

Class that stores inputs and runs them through the zigzag cost model. More...

Inheritance diagram for CostModelEvaluation:

Collaboration diagram for CostModelEvaluation:

Public Member Functions
def	__init__ (self, *Accelerator accelerator, LayerNode layer, SpatialMappingInternal spatial_mapping, SpatialMappingInternal spatial_mapping_int, TemporalMapping temporal_mapping, bool access_same_data_considered_as_no_access=True, float cycles_per_op=1.0)
	After initialization, the cost model evaluation is run. More...
None	run (self)
	Run the cost model evaluation. More...
None	calc_memory_utilization (self)
None	calc_memory_word_access (self)
None	calc_energy (self)
	Calculates the energy cost of this cost model evaluation by calculating the memory reading/writing energy. More...
None	calc_mac_energy_cost (self)
	Calculate the dynamic MAC energy. More...
def	calc_memory_energy_cost (self)
	Computes the memories reading/writing energy by converting the access patterns in self.mapping to energy breakdown using the memory hierarchy of the core on which the layer is mapped. More...
None	calc_latency (self)
	Calculate latency in 4 steps. More...
None	calc_double_buffer_flag (self)
	This function checks the double-buffer possibility for each operand at each memory level (minimal memory BW requirement case) by comparing the physical memory size with the effective data size, taking into account the memory sharing between operands. More...
def	calc_allowed_and_real_data_transfer_cycle_per_data_transfer_link (self)
	Construct a 4-way data transfer pattern for each unit mem, calculate {allowed_mem_updating_cycle, real_data_trans_cycle, DTL_SS_cycle} per period. More...
MemoryAccesses	calculate_real_data_transfer_cycles (self, LayerOperand layer_op, MemoryOperand mem_op, int mem_lvl_id)
MemoryAccesses	calculate_allowed_transfer_cycles (self, LayerOperand layer_op, int mem_lv)
None	combine_data_transfer_rate_per_physical_port (self)
	Consider memory sharing and port sharing, combine the data transfer activity Step 1: collect port activity per memory instance per physical memory port Step 2: calculate SS combine and MUW union parameters per physical memory port. More...
list[PortBeginOrEndActivity]	calc_loading_single_port_period_count_1 (self, MemoryPort port, list[tuple[MemoryOperand, int, DataDirection]] mem_op_level_direction_combs, float total_req_bw_aver_computation)
	Calculate the loading activities for the memory port movement directions with period count of 1. More...
list[PortBeginOrEndActivity]	calc_loading_single_port_period_count_greater_than_1 (self, MemoryPort port, list[tuple[MemoryOperand, int, DataDirection]] mem_op_level_direction_combs)
	Calculate the loading and offloading activities with mem port movement directions period > 1. More...
def	calc_loading_single_port (self, MemoryPort port)
	Calculate the loading and offloading activities for a single memory port. More...
def	calc_onloading_combined (self)
def	calc_offloading_combined (self)
tuple[float, float]	calc_borrowed_loading_cycles_and_bandwidth (self)
	Calculate the amount of cycles and bandwidth that are borrowed from the computation phase. More...
def	calc_data_loading_latency (self)
	Calculate the number of data onloading/offloading cycles. More...
None	calc_overall_latency (self)
	This function integrates the previous calculated SScomb, data loading and off-loading cycle to get the overall latency. More...
FourWayDataMoving[int]	get_inst_bandwidth (self, MemoryLevel memory_level, MemoryOperand memory_operand, float scaling=1)
	Given a memory level and a memory operand, compute the memory's instantaneous bandwidth required for the memory operand during computation. More...
FourWayDataMoving[int]	get_total_inst_bandwidth (self, MemoryLevel memory_level, float scaling=1)
	Given a cost model evaluation and a memory level, compute the memory's total instantaneous bandwidth required throughout the execution of the layer that corresponds to this CME. More...
def	__str__ (self)
def	__repr__ (self)
Public Member Functions inherited from CostModelEvaluationABC
None	__init__ (self)
def	core (self)
"CumulativeCME"	__add__ (self, "CostModelEvaluationABC" other)
def	__mul__ (self, int number)
dict[str, float]	__simplejsonrepr__ (self)
	Simple JSON representation used for saving this object to a simple json file. More...
def	__jsonrepr__ (self)
	JSON representation used for saving this object to a json file. More...

Static Public Member Functions
tuple[list[float], list[float]]	reduce_balanced (list[float] c_list, list[float] m_list, float s)
	Balance c_list towards minimums m_list with a total maximum reduction of s. More...

Public Attributes
	accelerator
	spatial_mapping
	spatial_mapping_int
	temporal_mapping
	access_same_data_considered_as_no_access
	mem_level_list
	mem_hierarchy_dict
	mem_size_dict
	mem_sharing_tuple
	memory_operand_links
	spatial_mapping_dict_int
	mapping
	mapping_int
	active_mem_level
	cycles_per_op
	mem_utili_individual
	mem_utili_shared
	effective_mem_utili_individual
	effective_mem_utili_shared
	memory_word_access
	mac_energy
	mem_energy_breakdown
	mem_energy_breakdown_further
	mem_energy
	double_buffer_true
	allowed_mem_update_cycle
	real_data_trans_cycle
	port_activity_collect
	mem_updating_window_union_collect
	stall_slack_comb_collect
	stall_slack_comb
	data_loading_cc_pair_combined_per_op
	data_loading_individual_part
	data_loading_half_shared_part
	data_loading_shared_part
	data_offloading_cc_pair_combined
	data_onloading_cycle
	data_offloading_cycle
	loading_offloading_cycles_borrowed_from_computation
	loading_offloading_bandwidth_borrowed_from_computation
	ideal_cycle
	ideal_temporal_cycle
	mac_spatial_utilization
	latency_total0
	latency_total1
	latency_total2
	mac_utilization0
	mac_utilization1
	mac_utilization2

Detailed Description

Class that stores inputs and runs them through the zigzag cost model.

Initialize the cost model evaluation with the following inputs:

• accelerator: the accelerator that includes the core on which to run the layer
• layer: the layer to run
• spatial_mapping: the spatial mapping
• temporal_mapping: the temporal mapping

From these parameters, the following attributes are computed:

• core: The core on which the layer is ran. This should be specified in the LayerNode attributes.
• mapping: The combined spatial and temporal mapping object where access patterns are computed.

The following cost model attributes are also initialized:

• mem_energy_breakdown: The energy breakdown for all operands
• energy: The total energy

After initialization, the cost model evaluation is run.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
		*Accelerator	accelerator,
		LayerNode	layer,
		SpatialMappingInternal	spatial_mapping,
		SpatialMappingInternal	spatial_mapping_int,
		TemporalMapping	temporal_mapping,
		bool	access_same_data_considered_as_no_access = True,
		float	cycles_per_op = 1.0
	)

After initialization, the cost model evaluation is run.

Parameters

accelerator	the accelerator that includes the core on which to run the
layer	the layer to run
access_same_data_considered_as_no_access	(optional)

Member Function Documentation

__repr__()

def __repr__

(

self

)

__str__()

def __str__

(

self

)

calc_allowed_and_real_data_transfer_cycle_per_data_transfer_link()

def calc_allowed_and_real_data_transfer_cycle_per_data_transfer_link

(

self

)

Construct a 4-way data transfer pattern for each unit mem, calculate {allowed_mem_updating_cycle, real_data_trans_cycle, DTL_SS_cycle} per period.

TODO cleanup

Here is the call graph for this function:

Here is the caller graph for this function:

calc_borrowed_loading_cycles_and_bandwidth()

tuple[float, float] calc_borrowed_loading_cycles_and_bandwidth

(

self

)

Calculate the amount of cycles and bandwidth that are borrowed from the computation phase.

calc_data_loading_latency()

def calc_data_loading_latency

(

self

)

Calculate the number of data onloading/offloading cycles.

This function first gathers the correct cycles per port, then combines them.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_double_buffer_flag()

None calc_double_buffer_flag

(

self

)

This function checks the double-buffer possibility for each operand at each memory level (minimal memory BW requirement case) by comparing the physical memory size with the effective data size, taking into account the memory sharing between operands.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_energy()

None calc_energy

(

self

)

Calculates the energy cost of this cost model evaluation by calculating the memory reading/writing energy.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_latency()

None calc_latency

(

self

)

Calculate latency in 4 steps.

  1) As we already calculated the ideal data transfer rate in combined_mapping.py (in the Mapping class),
  here we start with calculating the required (or allowed) memory updating window by comparing the effective
  data size with the physical memory size at each level. If the effective data size is smaller than 50%
  of the physical memory size, then we take the whole period as the allowed memory updating window
  (double buffer effect);
  otherwise we take the the period divided by the top_ir_loop as the allowed memory updating window.

  2) Then, we compute the real data transfer rate given the actual memory bw per functional port pair,
  assuming we have enough memory ports.

  3) In reality, there is no infinite memory port to use. So, as the second step, we combine the real
  data transfer attributes per physical memory port.

  4) Finally, we combine the stall/slack of each memory port to get the final latency.

Reimplemented in CostModelEvaluationForIMC.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_loading_single_port()

def calc_loading_single_port	(		self,
		MemoryPort	port
	)

Calculate the loading and offloading activities for a single memory port.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_loading_single_port_period_count_1()

list[PortBeginOrEndActivity] calc_loading_single_port_period_count_1	(		self,
		MemoryPort	port,
		list[tuple[MemoryOperand, int, DataDirection]]	mem_op_level_direction_combs,
		float	total_req_bw_aver_computation
	)

Calculate the loading activities for the memory port movement directions with period count of 1.

Here is the call graph for this function:

Here is the caller graph for this function:

calc_loading_single_port_period_count_greater_than_1()

list[PortBeginOrEndActivity] calc_loading_single_port_period_count_greater_than_1	(		self,
		MemoryPort	port,
		list[tuple[MemoryOperand, int, DataDirection]]	mem_op_level_direction_combs
	)

Calculate the loading and offloading activities with mem port movement directions period > 1.

Here is the caller graph for this function:

calc_mac_energy_cost()

None calc_mac_energy_cost

(

self

)

Calculate the dynamic MAC energy.

Reimplemented in CostModelEvaluationForIMC.

Here is the caller graph for this function:

calc_memory_energy_cost()

def calc_memory_energy_cost

(

self

)

Computes the memories reading/writing energy by converting the access patterns in self.mapping to energy breakdown using the memory hierarchy of the core on which the layer is mapped.

The energy breakdown is saved in self.mem_energy_breakdown.

The energy total consumption is saved in self.energy_total.

Here is the caller graph for this function:

calc_memory_utilization()

None calc_memory_utilization

(

self

)

Here is the call graph for this function:

Here is the caller graph for this function:

calc_memory_word_access()

None calc_memory_word_access

(

self

)

Here is the call graph for this function:

Here is the caller graph for this function:

calc_offloading_combined()

def calc_offloading_combined

(

self

)

calc_onloading_combined()

def calc_onloading_combined

(

self

)

Here is the caller graph for this function:

calc_overall_latency()

None calc_overall_latency

(

self

)

This function integrates the previous calculated SScomb, data loading and off-loading cycle to get the overall latency.

Parameters

cycles_per_op

cycle counts per mac operand (>1 for bit-serial computation)

Here is the caller graph for this function:

calculate_allowed_transfer_cycles()

MemoryAccesses calculate_allowed_transfer_cycles	(		self,
		LayerOperand	layer_op,
		int	mem_lv
	)

Here is the caller graph for this function:

calculate_real_data_transfer_cycles()

MemoryAccesses calculate_real_data_transfer_cycles	(		self,
		LayerOperand	layer_op,
		MemoryOperand	mem_op,
		int	mem_lvl_id
	)

Here is the caller graph for this function:

combine_data_transfer_rate_per_physical_port()

None combine_data_transfer_rate_per_physical_port

(

self

)

Consider memory sharing and port sharing, combine the data transfer activity Step 1: collect port activity per memory instance per physical memory port Step 2: calculate SS combine and MUW union parameters per physical memory port.

Here is the caller graph for this function:

get_inst_bandwidth()

FourWayDataMoving[int] get_inst_bandwidth	(		self,
		MemoryLevel	memory_level,
		MemoryOperand	memory_operand,
		float	scaling = 1
	)

Given a memory level and a memory operand, compute the memory's instantaneous bandwidth required for the memory operand during computation.

An additonal scaling factor can be applied to the returned bandwidth NOTE: This function is used in Stream

Here is the caller graph for this function:

get_total_inst_bandwidth()

FourWayDataMoving[int] get_total_inst_bandwidth	(		self,
		MemoryLevel	memory_level,
		float	scaling = 1
	)

Given a cost model evaluation and a memory level, compute the memory's total instantaneous bandwidth required throughout the execution of the layer that corresponds to this CME.

Raises an assertion error if the given memory level is not included in this CME's memory hierarchy. NOTE: this function is used in Stream

Here is the call graph for this function:

reduce_balanced()

tuple[list[float], list[float]] reduce_balanced ( list[float]  c_list, list[float]  m_list, float  s  )

static

Balance c_list towards minimums m_list with a total maximum reduction of s.

Here is the caller graph for this function:

run()

None run

(

self

)

Run the cost model evaluation.

Reimplemented in CostModelEvaluationForIMC.

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

accelerator

accelerator

access_same_data_considered_as_no_access

access_same_data_considered_as_no_access

active_mem_level

active_mem_level

allowed_mem_update_cycle

allowed_mem_update_cycle

cycles_per_op

cycles_per_op

data_loading_cc_pair_combined_per_op

data_loading_cc_pair_combined_per_op

data_loading_half_shared_part

data_loading_half_shared_part

data_loading_individual_part

data_loading_individual_part

data_loading_shared_part

data_loading_shared_part

data_offloading_cc_pair_combined

data_offloading_cc_pair_combined

data_offloading_cycle

data_offloading_cycle

data_onloading_cycle

data_onloading_cycle

double_buffer_true

double_buffer_true

effective_mem_utili_individual

effective_mem_utili_individual

effective_mem_utili_shared

effective_mem_utili_shared

ideal_cycle

ideal_cycle

ideal_temporal_cycle

ideal_temporal_cycle

latency_total0

latency_total0

latency_total1

latency_total1

latency_total2

latency_total2

loading_offloading_bandwidth_borrowed_from_computation

loading_offloading_bandwidth_borrowed_from_computation

loading_offloading_cycles_borrowed_from_computation

loading_offloading_cycles_borrowed_from_computation

mac_energy

mac_energy

mac_spatial_utilization

mac_spatial_utilization

mac_utilization0

mac_utilization0

mac_utilization1

mac_utilization1

mac_utilization2

mac_utilization2

mapping

mapping

mapping_int

mapping_int

mem_energy

mem_energy

mem_energy_breakdown

mem_energy_breakdown

mem_energy_breakdown_further

mem_energy_breakdown_further

mem_hierarchy_dict

mem_hierarchy_dict

mem_level_list

mem_level_list

mem_sharing_tuple

mem_sharing_tuple

mem_size_dict

mem_size_dict

mem_updating_window_union_collect

mem_updating_window_union_collect

mem_utili_individual

mem_utili_individual

mem_utili_shared

mem_utili_shared

memory_operand_links

memory_operand_links

memory_word_access

memory_word_access

port_activity_collect

port_activity_collect

real_data_trans_cycle

real_data_trans_cycle

spatial_mapping

spatial_mapping

spatial_mapping_dict_int

spatial_mapping_dict_int

spatial_mapping_int

spatial_mapping_int

stall_slack_comb

stall_slack_comb

stall_slack_comb_collect

stall_slack_comb_collect

temporal_mapping

temporal_mapping

---

The documentation for this class was generated from the following file:

• /home/runner/work/zigzag/zigzag/zigzag/cost_model/cost_model.py