Skip to content
Snippets Groups Projects

Add folding example

Merged Oscar Gustafsson requested to merge folding into master
1 file
+ 126
0
Compare changes
  • Side-by-side
  • Inline
+ 126
0
 
"""
 
=======================
 
Comparison with folding
 
=======================
 
 
This is a common example when illustrating folding.
 
 
In general, the main problem with folding is to determine a suitable folding order. This
 
corresponds to scheduling the operations.
 
 
Here, the folding order is the same for the adders as in the standard solution to this
 
problem, but the order of the multipliers is different to keep each memory variable
 
shorter than the scheduling period.
 
 
"""
 
 
from b_asic.architecture import Architecture, Memory, ProcessingElement
 
from b_asic.core_operations import Addition, ConstantMultiplication
 
from b_asic.schedule import Schedule
 
from b_asic.signal_flow_graph import SFG
 
from b_asic.special_operations import Delay, Input, Output
 
 
in1 = Input("IN")
 
T1 = Delay()
 
T2 = Delay(T1)
 
a = ConstantMultiplication(0.2, T1, "a")
 
b = ConstantMultiplication(0.3, T1, "b")
 
c = ConstantMultiplication(0.4, T2, "c")
 
d = ConstantMultiplication(0.6, T2, "d")
 
add2 = a + c
 
add1 = in1 + add2
 
add3 = b + d
 
T1 << add1
 
out1 = Output(add1 + add3, "OUT")
 
 
sfg = SFG(inputs=[in1], outputs=[out1], name="Bi-quad folding example")
 
 
# %%
 
# Set latencies and execution times
 
sfg.set_latency_of_type(ConstantMultiplication.type_name(), 2)
 
sfg.set_latency_of_type(Addition.type_name(), 1)
 
sfg.set_execution_time_of_type(ConstantMultiplication.type_name(), 1)
 
sfg.set_execution_time_of_type(Addition.type_name(), 1)
 
 
# %%
 
# Create schedule
 
schedule = Schedule(sfg, cyclic=True)
 
schedule.show(title='Original schedule')
 
 
# %%
 
# Reschedule to only require one adder and one multiplier
 
schedule.move_operation('out1', 2)
 
schedule.move_operation('add3', 2)
 
schedule.move_operation('cmul3', -3)
 
schedule.move_operation('add4', 3)
 
schedule.move_operation('cmul2', -3)
 
schedule.set_schedule_time(4)
 
schedule.move_operation('cmul2', 1)
 
schedule.move_operation('cmul1', 1)
 
schedule.move_operation('in1', 3)
 
schedule.move_operation('cmul3', -1)
 
schedule.move_operation('cmul1', 1)
 
schedule.show(title='Improved schedule')
 
 
# %%
 
# Extract operations and create processing elements
 
operations = schedule.get_operations()
 
adders = operations.get_by_type_name('add')
 
adders.show(title="Adder executions")
 
mults = operations.get_by_type_name('cmul')
 
mults.show(title="Multiplier executions")
 
inputs = operations.get_by_type_name('in')
 
inputs.show(title="Input executions")
 
outputs = operations.get_by_type_name('out')
 
outputs.show(title="Output executions")
 
 
p1 = ProcessingElement(adders, entity_name="adder")
 
p2 = ProcessingElement(mults, entity_name="cmul")
 
p_in = ProcessingElement(inputs, entity_name='input')
 
p_out = ProcessingElement(outputs, entity_name='output')
 
 
# %%
 
# Extract and assign memory variables
 
mem_vars = schedule.get_memory_variables()
 
mem_vars.show(title="All memory variables")
 
direct, mem_vars = mem_vars.split_on_length()
 
mem_vars.show(title="Non-zero time memory variables")
 
mem_vars_set = mem_vars.split_on_ports(read_ports=1, write_ports=1, total_ports=2)
 
 
memories = []
 
for i, mem in enumerate(mem_vars_set):
 
memory = Memory(mem, memory_type="RAM", entity_name=f"memory{i}")
 
memories.append(memory)
 
mem.show(title=f"{memory.entity_name}")
 
memory.assign("left_edge")
 
memory.show_content(title=f"Assigned {memory.entity_name}")
 
 
direct.show(title="Direct interconnects")
 
 
# %%
 
# Create architecture
 
arch = Architecture({p1, p2, p_in, p_out}, memories, direct_interconnects=direct)
 
 
# %%
 
# The architecture can be rendered in enriched shells.
 
#
 
# .. graphviz::
 
#
 
# digraph {
 
# node [shape=record]
 
# memory0 [label="{{<in0> in0}|memory0|{<out0> out0}}"]
 
# memory1 [label="{{<in0> in0}|memory1|{<out0> out0}}"]
 
# output [label="{{<in0> in0}|output}"]
 
# adder [label="{{<in0> in0|<in1> in1}|adder|{<out0> out0}}"]
 
# input [label="{input|{<out0> out0}}"]
 
# cmul [label="{{<in0> in0}|cmul|{<out0> out0}}"]
 
# memory1:out0 -> adder:in1 [label=2]
 
# cmul:out0 -> adder:in0 [label=2]
 
# memory0:out0 -> cmul:in0 [label=4]
 
# adder:out0 -> output:in0 [label=1]
 
# input:out0 -> adder:in0 [label=1]
 
# adder:out0 -> memory0:in0 [label=1]
 
# adder:out0 -> adder:in1 [label=2]
 
# memory0:out0 -> adder:in0 [label=1]
 
# cmul:out0 -> memory1:in0 [label=2]
 
# }
Loading