SpatialMappingConversionStage Class Reference

Pipeline stage that converts the spatial mapping from a user-provided spatial mapping across operational array dimensions to the internal spatial mapping representation used in the cost model. More...

Inheritance diagram for SpatialMappingConversionStage:

Collaboration diagram for SpatialMappingConversionStage:

Public Member Functions
def	__init__ (self, list[StageCallable] list_of_callables, *Accelerator accelerator, LayerNode layer, **Any kwargs)
	Initialize the accelerator and layer attributes. More...
def	run (self)
tuple[SpatialMappingInternal, SpatialMappingInternal]	convert_user_spatial_mapping (self, SpatialMapping user_spatial_mapping)
	Convert the SpatialMapping instance in user-defined format (spatial mapping across operational array dimensions) to the SpatialMappingInternal representation. More...
None|tuple[LayerDim, UnrollFactor]	generate_limited_user_spatial_mapping (self, LayerDimSizes layer_dim_sizes, OADimension oa_dim, tuple[LayerDim, UnrollFactor] spatial_loop, SpatialMapping user_spatial_mapping, SpatialMapping limited_user_spatial_mapping, bool allow_decimal_sm_loop_size=True)
SpatialMappingPerMemLvl	generate_mapping_per_mem_lvl (self, SpatialMapping user_spatial_mapping)
	This function is to convert spatial mapping to mapping_per_mem_lvl, which attaches spatial mapping to different memory levels. More...
def	check_if_oa_dim_mapping_is_first_max (self, OADimension oa_dim, LayerDim loop_dim_unrolled, SpatialMapping user_spatial_mapping)
	For the case when there is layer dimension that is mapped on multiple oa dimensions. More...
UnrollFactor	calc_unrolled_loop_size_on_early_oa_dims (self, OADimension oa_dim, LayerDim loop_dim_unrolled, SpatialMapping user_spatial_mapping)
Public Member Functions inherited from Stage
def	__init__ (self, list["StageCallable"] list_of_callables, **Any kwargs)
def	__iter__ (self)
bool	is_leaf (self)

Public Attributes
	layer
	accelerator
	memory_operand_links
	user_spatial_mapping
Public Attributes inherited from Stage
	kwargs
	list_of_callables

Detailed Description

Pipeline stage that converts the spatial mapping from a user-provided spatial mapping across operational array dimensions to the internal spatial mapping representation used in the cost model.

Constructor & Destructor Documentation

__init__()

def __init__	(		self,
		list[StageCallable]	list_of_callables,
		*Accelerator	accelerator,
		LayerNode	layer,
		**Any	kwargs
	)

Initialize the accelerator and layer attributes.

Member Function Documentation

calc_unrolled_loop_size_on_early_oa_dims()

UnrollFactor calc_unrolled_loop_size_on_early_oa_dims	(		self,
		OADimension	oa_dim,
		LayerDim	loop_dim_unrolled,
		SpatialMapping	user_spatial_mapping
	)

Here is the caller graph for this function:

check_if_oa_dim_mapping_is_first_max()

def check_if_oa_dim_mapping_is_first_max	(		self,
		OADimension	oa_dim,
		LayerDim	loop_dim_unrolled,
		SpatialMapping	user_spatial_mapping
	)

For the case when there is layer dimension that is mapped on multiple oa dimensions.

We need to decide on which oa dimension to adjust the unrolling if the total unrolling size is not a multiple of the layer dimension size. In this case, we decide to only adjust the unrolling size of the first oa dimension with the largest unrolling. This function is to check if the given oa_dim has the largest unrolling for the given loop_dim_unrolled.

Here is the caller graph for this function:

convert_user_spatial_mapping()

tuple[SpatialMappingInternal, SpatialMappingInternal] convert_user_spatial_mapping	(		self,
		SpatialMapping	user_spatial_mapping
	)

Convert the SpatialMapping instance in user-defined format (spatial mapping across operational array dimensions) to the SpatialMappingInternal representation.

For this conversion we need to know:

• the user defined spatial mapping
• the core (i.e. operational array) on which the unrolling happens,and the memory hierarchy that is connected to that operational array.

  Parameters

  user_spatial_mapping

  The SpatialMapping in user-defined format to be converted.

  Returns

  : A SpatialMappingInternal object with the converted spatial mapping.

Here is the call graph for this function:

Here is the caller graph for this function:

generate_limited_user_spatial_mapping()

None | tuple[LayerDim, UnrollFactor] generate_limited_user_spatial_mapping	(		self,
		LayerDimSizes	layer_dim_sizes,
		OADimension	oa_dim,
		tuple[LayerDim, UnrollFactor]	spatial_loop,
		SpatialMapping	user_spatial_mapping,
		SpatialMapping	limited_user_spatial_mapping,
		bool	allow_decimal_sm_loop_size = True
	)

Here is the call graph for this function:

Here is the caller graph for this function:

generate_mapping_per_mem_lvl()

SpatialMappingPerMemLvl generate_mapping_per_mem_lvl	(		self,
		SpatialMapping	user_spatial_mapping
	)

This function is to convert spatial mapping to mapping_per_mem_lvl, which attaches spatial mapping to different memory levels.

TODO This should be a class

Here is the caller graph for this function:

run()

def run

(

self

)

Reimplemented from Stage.

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

accelerator

accelerator

layer

layer

memory_operand_links

memory_operand_links

user_spatial_mapping

user_spatial_mapping

---

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

• /home/runner/work/zigzag/zigzag/zigzag/stages/mapping/spatial_mapping_conversion.py