"""
B-ASIC Schedule Module.
Contains the schedule class for scheduling operations in an SFG.
"""
import io
import sys
from collections import defaultdict
from typing import Dict, List, Optional, Tuple, cast
import matplotlib.pyplot as plt
import numpy as np
from matplotlib.axes import Axes
from matplotlib.figure import Figure
from matplotlib.lines import Line2D
from matplotlib.patches import PathPatch, Polygon
from matplotlib.path import Path
from matplotlib.ticker import MaxNLocator
from b_asic import Signal
from b_asic._preferences import (
EXECUTION_TIME_COLOR,
LATENCY_COLOR,
OPERATION_GAP,
SCHEDULE_OFFSET,
SIGNAL_COLOR,
SIGNAL_LINEWIDTH,
SPLINE_OFFSET,
)
from b_asic.graph_component import GraphID
from b_asic.operation import Operation
from b_asic.port import InputPort, OutputPort
from b_asic.process import MemoryVariable, Process
from b_asic.resources import ProcessCollection
from b_asic.signal_flow_graph import SFG
from b_asic.special_operations import Delay, Input, Output
# Need RGB from 0 to 1
_EXECUTION_TIME_COLOR = tuple(c / 255 for c in EXECUTION_TIME_COLOR)
_LATENCY_COLOR = tuple(c / 255 for c in LATENCY_COLOR)
_SIGNAL_COLOR = tuple(c / 255 for c in SIGNAL_COLOR)
class Schedule:
"""
Schedule of an SFG with scheduled Operations.
Parameters
----------
sfg : SFG
The signal flow graph to schedule.
schedule_time : int, optional
The schedule time. If not provided, it will be determined by the scheduling
algorithm.
cyclic : bool, default: False
If the schedule is cyclic.
scheduling_algorithm : {'ASAP'}, optional
The scheduling algorithm to use. Currently, only "ASAP" is supported.
"""
_sfg: SFG
_start_times: Dict[GraphID, int]
_laps: Dict[GraphID, int]
_schedule_time: int
_cyclic: bool
_y_locations: Dict[GraphID, Optional[int]]
def __init__(
self,
sfg: SFG,
schedule_time: Optional[int] = None,
cyclic: bool = False,
scheduling_algorithm: str = "ASAP",
):
"""Construct a Schedule from an SFG."""
self._sfg = sfg
self._start_times = {}
self._laps = defaultdict(lambda: 0)
self._cyclic = cyclic
self._y_locations = defaultdict(lambda: None)
if scheduling_algorithm == "ASAP":
self._schedule_asap()
else:
raise NotImplementedError(
f"No algorithm with name: {scheduling_algorithm} defined."
)
max_end_time = self.get_max_end_time()
if schedule_time is None:
self._schedule_time = max_end_time
elif schedule_time < max_end_time:
raise ValueError(f"Too short schedule time. Minimum is {max_end_time}.")
else:
self._schedule_time = schedule_time
def start_time_of_operation(self, graph_id: GraphID) -> int:
"""
Return the start time of the operation with the specified by *graph_id*.
"""
if graph_id not in self._start_times:
raise ValueError(f"No operation with graph_id {graph_id} in schedule")
return self._start_times[graph_id]
def get_max_end_time(self) -> int:
"""Return the current maximum end time among all operations."""
max_end_time = 0
for graph_id, op_start_time in self._start_times.items():
op = cast(Operation, self._sfg.find_by_id(graph_id))
for outport in op.outputs:
max_end_time = max(
max_end_time,
op_start_time + cast(int, outport.latency_offset),
)
return max_end_time
def forward_slack(self, graph_id: GraphID) -> int:
"""
Return how much an operation can be moved forward in time.
Parameters
----------
graph_id : GraphID
The graph id of the operation.
Returns
-------
The number of time steps the operation with *graph_id* can ba moved
forward in time.
See also
--------
backward_slack
slacks
"""
if graph_id not in self._start_times:
raise ValueError(f"No operation with graph_id {graph_id} in schedule")
slack = sys.maxsize
output_slacks = self._forward_slacks(graph_id)
# Make more pythonic
for signal_slacks in output_slacks.values():
for signal_slack in signal_slacks.values():
slack = min(slack, signal_slack)
return slack
def _forward_slacks(
self, graph_id: GraphID
) -> Dict["OutputPort", Dict["Signal", int]]:
ret = {}
start_time = self._start_times[graph_id]
op = cast(Operation, self._sfg.find_by_id(graph_id))
for output_port in op.outputs:
output_slacks = {}
available_time = start_time + cast(int, output_port.latency_offset)
for signal in output_port.signals:
destination = cast(InputPort, signal.destination)
usage_time = (
cast(int, destination.latency_offset)
+ self._start_times[destination.operation.graph_id]
+ self._schedule_time * self._laps[signal.graph_id]
)
output_slacks[signal] = usage_time - available_time
ret[output_port] = output_slacks
return ret
def backward_slack(self, graph_id: GraphID) -> int:
"""
Return how much an operation can be moved backward in time.
Parameters
----------
graph_id : GraphID
The graph id of the operation.
Returns
-------
The number of time steps the operation with *graph_id* can ba moved
backward in time.
.. note:: The backward slack is positive, but a call to :func:`move_operation`
should be negative to move the operation backward.
See also
--------
forward_slack
slacks
"""
if graph_id not in self._start_times:
raise ValueError(f"No operation with graph_id {graph_id} in schedule")
slack = sys.maxsize
input_slacks = self._backward_slacks(graph_id)
# Make more pythonic
for signal_slacks in input_slacks.values():
for signal_slack in signal_slacks.values():
slack = min(slack, signal_slack)
return slack
def _backward_slacks(self, graph_id: GraphID) -> Dict[InputPort, Dict[Signal, int]]:
ret = {}
start_time = self._start_times[graph_id]
op = cast(Operation, self._sfg.find_by_id(graph_id))
for input_port in op.inputs:
input_slacks = {}
usage_time = start_time + cast(int, input_port.latency_offset)
for signal in input_port.signals:
source = cast(OutputPort, signal.source)
available_time = (
cast(int, source.latency_offset)
+ self._start_times[source.operation.graph_id]
- self._schedule_time * self._laps[signal.graph_id]
)
input_slacks[signal] = usage_time - available_time
ret[input_port] = input_slacks
return ret
def slacks(self, graph_id: GraphID) -> Tuple[int, int]:
"""
Return the backward and forward slacks of operation *graph_id*. That is, how
much the operation can be moved backward and forward in time.
Parameters
----------
graph_id : GraphID
The graph id of the operation.
Returns
-------
A tuple as ``(backward_slack, forward_slack)``.
.. note:: The backward slack is positive, but a call to :func:`move_operation`
should be negative to move the operation backward.
See also
--------
backward_slack
forward_slack
"""
if graph_id not in self._start_times:
raise ValueError(f"No operation with graph_id {graph_id} in schedule")
return self.backward_slack(graph_id), self.forward_slack(graph_id)
def print_slacks(self) -> None:
raise NotImplementedError
def set_schedule_time(self, time: int) -> "Schedule":
"""
Set a new schedule time.
Parameters
----------
time : int
The new schedule time. If it is too short, a ValueError will be raised.
See also
--------
get_max_time
"""
if time < self.get_max_end_time():
raise ValueError(
f"New schedule time ({time}) too short, minimum:"
f" {self.get_max_end_time()}."
)
self._schedule_time = time
return self
@property
def sfg(self) -> SFG:
return self._sfg
@property
def start_times(self) -> Dict[GraphID, int]:
return self._start_times
@property
def laps(self) -> Dict[GraphID, int]:
return self._laps
@property
def schedule_time(self) -> int:
return self._schedule_time
@property
def cyclic(self) -> bool:
return self._cyclic
def increase_time_resolution(self, factor: int) -> "Schedule":
"""
Increase time resolution for a schedule.
Parameters
==========
factor : int
The time resolution increment.
"""
self._start_times = {k: factor * v for k, v in self._start_times.items()}
for graph_id in self._start_times:
cast(Operation, self._sfg.find_by_id(graph_id))._increase_time_resolution(
factor
)
self._schedule_time *= factor
return self
def _get_all_times(self) -> List[int]:
"""
Return a list of all times for the schedule. Used to check how the
resolution can be modified.
"""
# Local values
ret = [self._schedule_time, *self._start_times.values()]
# Loop over operations
for graph_id in self._start_times:
op = cast(Operation, self._sfg.find_by_id(graph_id))
ret += [cast(int, op.execution_time), *op.latency_offsets.values()]
# Remove not set values (None)
ret = [v for v in ret if v is not None]
return ret
def get_possible_time_resolution_decrements(self) -> List[int]:
"""Return a list with possible factors to reduce time resolution."""
vals = self._get_all_times()
maxloop = min(val for val in vals if val)
if maxloop <= 1:
return [1]
ret = [1]
for candidate in range(2, maxloop + 1):
if not any(val % candidate for val in vals):
ret.append(candidate)
return ret
def decrease_time_resolution(self, factor: int) -> "Schedule":
"""
Decrease time resolution for a schedule.
Parameters
==========
factor : int
The time resolution decrement.
See also
========
get_possible_time_resolution_decrements
"""
possible_values = self.get_possible_time_resolution_decrements()
if factor not in possible_values:
raise ValueError(
f"Not possible to decrease resolution with {factor}. Possible"
f" values are {possible_values}"
)
self._start_times = {k: v // factor for k, v in self._start_times.items()}
for graph_id in self._start_times:
cast(Operation, self._sfg.find_by_id(graph_id))._decrease_time_resolution(
factor
)
self._schedule_time = self._schedule_time // factor
return self
def move_operation(self, graph_id: GraphID, time: int) -> "Schedule":
"""
Move an operation in the schedule.
Parameters
----------
graph_id : GraphID
The graph id of the operation to move.
time : int
The time to move. If positive move forward, if negative move backward.
"""
if graph_id not in self._start_times:
raise ValueError(f"No operation with graph_id {graph_id} in schedule")
(backward_slack, forward_slack) = self.slacks(graph_id)
if not -backward_slack <= time <= forward_slack:
raise ValueError(
f"Operation {graph_id} got incorrect move: {time}. Must be"
f" between {-backward_slack} and {forward_slack}."
)
tmp_start = self._start_times[graph_id] + time
new_start = tmp_start % self._schedule_time
# Update input laps
input_slacks = self._backward_slacks(graph_id)
for in_port, signal_slacks in input_slacks.items():
tmp_usage = tmp_start + cast(int, in_port.latency_offset)
new_usage = tmp_usage % self._schedule_time
for signal, signal_slack in signal_slacks.items():
new_slack = signal_slack + time
old_laps = self._laps[signal.graph_id]
tmp_prev_available = tmp_usage - new_slack
prev_available = tmp_prev_available % self._schedule_time
laps = new_slack // self._schedule_time
source_op = signal.source.operation
if new_usage < prev_available:
print("Incrementing input laps 1")
laps += 1
if (
prev_available == 0
and new_usage == 0
and (
tmp_prev_available > 0
or tmp_prev_available == 0
and not isinstance(source_op, Input)
)
):
print("Incrementing input laps 2")
laps += 1
print(
[
"Input",
signal.source.operation,
time,
tmp_start,
signal_slack,
new_slack,
old_laps,
laps,
new_usage,
prev_available,
tmp_usage,
tmp_prev_available,
]
)
self._laps[signal.graph_id] = laps
# Update output laps
output_slacks = self._forward_slacks(graph_id)
for out_port, signal_slacks in output_slacks.items():
tmp_available = tmp_start + cast(int, out_port.latency_offset)
new_available = tmp_available % self._schedule_time
for signal, signal_slack in signal_slacks.items():
new_slack = signal_slack - time
tmp_next_usage = tmp_available + new_slack
next_usage = tmp_next_usage % self._schedule_time
laps = new_slack // self._schedule_time
if next_usage < new_available:
laps += 1
print("Incrementing output laps 1")
if new_available == 0 and (new_slack > 0 or next_usage == 0):
print("Incrementing output laps 2")
laps += 1
print(
[
"Output",
signal_slack,
new_slack,
old_laps,
laps,
new_available,
next_usage,
tmp_available,
tmp_next_usage,
]
)
self._laps[signal.graph_id] = laps
# Set new start time
self._start_times[graph_id] = new_start
return self
def _remove_delays(self) -> None:
delay_list = self._sfg.find_by_type_name(Delay.type_name())
while delay_list:
delay_op = cast(Delay, delay_list[0])
delay_input_id = delay_op.input(0).signals[0].graph_id
delay_output_ids = [sig.graph_id for sig in delay_op.output(0).signals]
self._sfg = cast(SFG, self._sfg.remove_operation(delay_op.graph_id))
for output_id in delay_output_ids:
self._laps[output_id] += 1 + self._laps[delay_input_id]
del self._laps[delay_input_id]
delay_list = self._sfg.find_by_type_name(Delay.type_name())
def _schedule_asap(self) -> None:
"""Schedule the operations using as-soon-as-possible scheduling."""
pl = self._sfg.get_precedence_list()
if len(pl) < 2:
print("Empty signal flow graph cannot be scheduled.")
return
non_schedulable_ops = set()
for outport in pl[0]:
op = outport.operation
if op.type_name() not in [Delay.type_name()]:
if op.graph_id not in self._start_times:
# Set start time of all operations in the first iter to 0
self._start_times[op.graph_id] = 0
else:
non_schedulable_ops.add(op.graph_id)
for outport in pl[1]:
op = outport.operation
if op.graph_id not in self._start_times:
# Set start time of all operations in the first iter to 0
self._start_times[op.graph_id] = 0
for outports in pl[2:]:
for outport in outports:
op = outport.operation
if op.graph_id not in self._start_times:
# Schedule the operation if it does not have a start time yet.
op_start_time = 0
for inport in op.inputs:
if len(inport.signals) != 1:
raise ValueError(
"Error in scheduling, dangling input port detected."
)
if inport.signals[0].source is None:
raise ValueError(
"Error in scheduling, signal with no source detected."
)
source_port = inport.signals[0].source
source_end_time = None
if source_port.operation.graph_id in non_schedulable_ops:
source_end_time = 0
else:
source_op_time = self._start_times[
source_port.operation.graph_id
]
if source_port.latency_offset is None:
raise ValueError(
f"Output port {source_port.index} of"
" operation"
f" {source_port.operation.graph_id} has no"
" latency-offset."
)
source_end_time = (
source_op_time + source_port.latency_offset
)
if inport.latency_offset is None:
raise ValueError(
f"Input port {inport.index} of operation"
f" {inport.operation.graph_id} has no"
" latency-offset."
)
op_start_time_from_in = source_end_time - inport.latency_offset
op_start_time = max(op_start_time, op_start_time_from_in)
self._start_times[op.graph_id] = op_start_time
for output in self._sfg.find_by_type_name(Output.type_name()):
output = cast(Output, output)
source_port = cast(OutputPort, output.inputs[0].signals[0].source)
if source_port.operation.graph_id in non_schedulable_ops:
self._start_times[output.graph_id] = 0
else:
if source_port.latency_offset is None:
raise ValueError(
f"Output port {source_port.index} of operation"
f" {source_port.operation.graph_id} has no"
" latency-offset."
)
self._start_times[output.graph_id] = self._start_times[
source_port.operation.graph_id
] + cast(int, source_port.latency_offset)
self._remove_delays()
def _get_memory_variables_list(self) -> List['MemoryVariable']:
ret: List['MemoryVariable'] = []
for graph_id, start_time in self._start_times.items():
slacks = self._forward_slacks(graph_id)
for outport, signals in slacks.items():
reads = {
cast(InputPort, signal.destination): slack
for signal, slack in signals.items()
}
ret.append(
MemoryVariable(
start_time + cast(int, outport.latency_offset),
outport,
reads,
outport.operation.graph_id,
)
)
return ret
def get_memory_variables(self) -> ProcessCollection:
"""
Return a :class:`~b_asic.resources.ProcessCollection` containing all
memory variables.
Returns
-------
ProcessCollection
"""
return ProcessCollection(
set(self._get_memory_variables_list()), self.schedule_time
)
def _get_y_position(
self, graph_id, operation_height=1.0, operation_gap=None
) -> float:
if operation_gap is None:
operation_gap = OPERATION_GAP
y_location = self._y_locations[graph_id]
if y_location is None:
# Assign the lowest row number not yet in use
used = set(loc for loc in self._y_locations.values() if loc is not None)
possible = set(range(len(self._start_times))) - used
y_location = min(possible)
self._y_locations[graph_id] = y_location
return operation_gap + y_location * (operation_height + operation_gap)
def _plot_schedule(self, ax: Axes, operation_gap: Optional[float] = None) -> None:
"""Draw the schedule."""
line_cache = []
def _draw_arrow(
start: List[float], end: List[float], name: str = "", laps: int = 0
):
"""Draw an arrow from *start* to *end*."""
if end[0] < start[0] or laps > 0: # Wrap around
if start not in line_cache:
line = Line2D(
[start[0], self._schedule_time + SCHEDULE_OFFSET],
[start[1], start[1]],
color=_SIGNAL_COLOR,
lw=SIGNAL_LINEWIDTH,
)
ax.add_line(line)
ax.text(
self._schedule_time + SCHEDULE_OFFSET,
start[1],
name,
verticalalignment="center",
)
line = Line2D(
[-SCHEDULE_OFFSET, end[0]],
[end[1], end[1]],
color=_SIGNAL_COLOR,
lw=SIGNAL_LINEWIDTH,
)
ax.add_line(line)
ax.text(
-SCHEDULE_OFFSET,
end[1],
f"{name}: {laps}",
verticalalignment="center",
horizontalalignment="right",
)
line_cache.append(start)
elif end[0] == start[0]:
p = Path(
[
start,
[start[0] + SPLINE_OFFSET, start[1]],
[start[0] + SPLINE_OFFSET, (start[1] + end[1]) / 2],
[start[0], (start[1] + end[1]) / 2],
[start[0] - SPLINE_OFFSET, (start[1] + end[1]) / 2],
[start[0] - SPLINE_OFFSET, end[1]],
end,
],
[
Path.MOVETO,
Path.CURVE4,
Path.CURVE4,
Path.CURVE4,
Path.CURVE4,
Path.CURVE4,
Path.CURVE4,
],
)
pp = PathPatch(
p,
fc='none',
ec=_SIGNAL_COLOR,
lw=SIGNAL_LINEWIDTH,
zorder=10,
)
ax.add_patch(pp)
else:
p = Path(
[
start,
[(start[0] + end[0]) / 2, start[1]],
[(start[0] + end[0]) / 2, end[1]],
end,
],
[Path.MOVETO, Path.CURVE4, Path.CURVE4, Path.CURVE4],
)
pp = PathPatch(
p,
fc='none',
ec=_SIGNAL_COLOR,
lw=SIGNAL_LINEWIDTH,
zorder=10,
)
ax.add_patch(pp)
def _draw_offset_arrow(start, end, start_offset, end_offset, name="", laps=0):
"""Draw an arrow from *start* to *end*, but with an offset."""
_draw_arrow(
[start[0] + start_offset[0], start[1] + start_offset[1]],
[end[0] + end_offset[0], end[1] + end_offset[1]],
name=name,
laps=laps,
)
ytickpositions = []
yticklabels = []
ax.set_axisbelow(True)
ax.grid()
for graph_id, op_start_time in self._start_times.items():
y_pos = self._get_y_position(graph_id, operation_gap=operation_gap)
op = cast(Operation, self._sfg.find_by_id(graph_id))
# Rewrite to make better use of NumPy
(
latency_coordinates,
execution_time_coordinates,
) = op.get_plot_coordinates()
_x, _y = zip(*latency_coordinates)
x = np.array(_x)
y = np.array(_y)
xy = np.stack((x + op_start_time, y + y_pos))
ax.add_patch(Polygon(xy.T, fc=_LATENCY_COLOR))
if execution_time_coordinates:
_x, _y = zip(*execution_time_coordinates)
x = np.array(_x)
y = np.array(_y)
ax.plot(
x + op_start_time,
y + y_pos,
color=_EXECUTION_TIME_COLOR,
linewidth=3,
)
ytickpositions.append(y_pos + 0.5)
yticklabels.append(cast(Operation, self._sfg.find_by_id(graph_id)).name)
for graph_id, op_start_time in self._start_times.items():
op = cast(Operation, self._sfg.find_by_id(graph_id))
out_coordinates = op.get_output_coordinates()
source_y_pos = self._get_y_position(graph_id, operation_gap=operation_gap)
for output_port in op.outputs:
for output_signal in output_port.signals:
destination = cast(InputPort, output_signal.destination)
destination_op = destination.operation
destination_start_time = self._start_times[destination_op.graph_id]
destination_y_pos = self._get_y_position(
destination_op.graph_id, operation_gap=operation_gap
)
destination_in_coordinates = (
destination.operation.get_input_coordinates()
)
_draw_offset_arrow(
out_coordinates[output_port.index],
destination_in_coordinates[destination.index],
[op_start_time, source_y_pos],
[destination_start_time, destination_y_pos],
name=graph_id,
laps=self._laps[output_signal.graph_id],
)
ax.set_yticks(ytickpositions)
ax.set_yticklabels(yticklabels)
# Get operation with maximum position
max_pos_graph_id = max(self._y_locations, key=self._y_locations.get)
y_position_max = (
self._get_y_position(max_pos_graph_id, operation_gap=operation_gap)
+ 1
+ (OPERATION_GAP if operation_gap is None else operation_gap)
)
ax.axis([-1, self._schedule_time + 1, y_position_max, 0]) # Inverted y-axis
ax.xaxis.set_major_locator(MaxNLocator(integer=True))
ax.axvline(
0,
linestyle="--",
color="black",
)
ax.axvline(
self._schedule_time,
linestyle="--",
color="black",
)
def _reset_y_locations(self) -> None:
"""Reset all the y-locations in the schedule to None"""
self._y_locations = self._y_locations = defaultdict(lambda: None)
def plot_in_axes(self, ax: Axes, operation_gap: Optional[float] = None) -> None:
"""
Plot the schedule in a :class:`matplotlib.axes.Axes` or subclass.
Parameters
----------
ax : matplotlib.axes.Axes
The :class:`matplotlib.axes.Axes` to plot in.
operation_gap : float, optional
The vertical distance between operations in the schedule. The height of
the operation is always 1.
"""
def plot(self, operation_gap: Optional[float] = None) -> None:
"""
Plot the schedule. Will display based on the current Matplotlib backend.
Parameters
----------
operation_gap : float, optional
The vertical distance between operations in the schedule. The height of
the operation is always 1.
"""
self._get_figure(operation_gap=operation_gap).show()
def _get_figure(self, operation_gap: Optional[float] = None) -> Figure:
"""
Create a Figure and an Axes and plot schedule in the Axes.
Parameters
----------
operation_gap : float, optional
The vertical distance between operations in the schedule. The height of
the operation is always 1.
Returns
-------
The Matplotlib Figure.
"""
fig, ax = plt.subplots()
self._plot_schedule(ax, operation_gap=operation_gap)
return fig
def _repr_svg_(self) -> str:
"""
Generate an SVG of the schedule. This is automatically displayed in e.g.
Jupyter Qt console.
"""
fig, ax = plt.subplots()
self._plot_schedule(ax)
f = io.StringIO()
fig.savefig(f, format="svg")
return f.getvalue()