update rtd build image

.readthedocs.yaml

@@ -1,7 +1,7 @@
 version: 2
 
 build:
-  os: "ubuntu-20.04"
+  os: "ubuntu-24.04"
   tools:
     python: "3.12"
   jobs:

src/kyupy/__init__.py

@@ -12,6 +12,8 @@ import gzip
 
 import numpy as np
 
+type NestedNumericList = list['int|NestedNumericList']
+type NestedStrIntDict = dict[str,'int|NestedStrIntDict']
 
 _pop_count_lut = np.asarray([bin(x).count('1') for x in range(256)])
 
@@ -111,9 +113,9 @@ class Timer:
 
 
 class Timers:
-    def __init__(self, t={}): self.timers = defaultdict(Timer) | t
+    def __init__(self, t={}): self.timers: dict[str, Timer] = defaultdict(Timer) | t
     def __getitem__(self, name): return self.timers[name]
-    def __repr__(self): return '{' + ', '.join([f'{k}: {v}' for k, v in self.timers.items()]) + '}'
+    def __repr__(self): return '{' + ', '.join([f'{k}: {v:.2f}' for k, v in self.dict().items()]) + '}'
     def __add__(self, t):
         tmr = Timers(self.timers)
         for k, v in t.timers.items(): tmr.timers[k] += v
@@ -121,7 +123,7 @@ class Timers:
     def sum(self):
         return sum([v.s for v in self.timers.values()])
     def dict(self):
-        return dict([(k, v.s) for k, v in self.timers.items()])
+        return dict(sorted([(k, v.s) for k, v in self.timers.items()], key=lambda item: -item[1]))
 
 
 class Log:

src/kyupy/bench.py

@@ -14,7 +14,7 @@ Since simulators only support at most four-input-gates, gates with more than fou
 from lark import Lark, Transformer
 
 from .circuit import Circuit, Node, Line
-from . import readtext, batchrange
+from . import readtext, batchrange, log
 
 def treeify(l, max_degree=4):
     if len(l) <= max_degree: return l
@@ -26,12 +26,32 @@ class BenchTransformer(Transformer):
     def __init__(self, name):
         super().__init__()
         self.c = Circuit(name)
+        self.inputs = set()
+        self.outputs = set()
 
     def start(self, _): return self.c
 
     def parameters(self, args): return [self.c.get_or_add_fork(str(name)) for name in args if name is not None]
 
-    def interface(self, args): self.c.io_nodes.extend(args[0])
+    def input(self, args):
+        for input in args[0]:
+            if input in self.outputs:
+                new_input = self.c.get_or_add_fork(input.name + '~i')
+                Line(self.c, new_input, input)
+                log.warn(f'input-output passthrough, renaming input: {input.name} -> {new_input.name}')
+                input = new_input
+            self.inputs.add(input)
+            self.c.io_nodes.append(input)
+
+    def output(self, args):
+        for output in args[0]:
+            if output in self.inputs:
+                new_output = self.c.get_or_add_fork(output.name + '~o')
+                Line(self.c, output, new_output)
+                log.warn(f'input-output passthrough, renaming output: {output.name} -> {new_output.name}')
+                output = new_output
+            self.outputs.add(output)
+            self.c.io_nodes.append(output)
 
     def _cell_tree_inner(self, name, kind, inner_kind, drivers):
         cell = Node(self.c, name, f'{kind}{len(drivers)}')
@@ -69,8 +89,8 @@ class BenchTransformer(Transformer):
 GRAMMAR = r"""
     start: (statement)*
     statement: input | output | assignment
-    input: ("INPUT" | "input") parameters -> interface
-    output: ("OUTPUT" | "output") parameters -> interface
+    input: ("INPUT" | "input") parameters
+    output: ("OUTPUT" | "output") parameters
     assignment: NAME "=" NAME parameters
     parameters: "(" [ NAME ( "," NAME )* ] ")"
     NAME: /[-_a-z0-9]+/i

src/kyupy/circuit.py

@@ -17,8 +17,7 @@ import re
 
 import numpy as np
 
-type NestedNumericList = list['int|NestedNumericList']
-type NestedStrIntDict = dict[str,'int|NestedStrIntDict']
+from . import NestedNumericList, NestedStrIntDict
 
 class GrowingList[T](list[T]):
     def __setitem__(self, index, value):

src/kyupy/def_file.py

@@ -6,7 +6,7 @@ as an intermediate representation (:class:`DefFile` object).
 
 from collections import defaultdict
 
-from lark import Lark, Transformer, Tree
+from lark import Lark, Transformer
 
 from kyupy import readtext
 
@@ -85,6 +85,8 @@ class DefFile:
         self.pins = {}
         self.specialnets = {}
         self.nets = {}
+        self.design = ""
+        self.diearea = 0
 
 
 class DefTransformer(Transformer):
@@ -140,8 +142,10 @@ class DefTransformer(Transformer):
     def comp_stmt(self, args):
         name = args[0].value
         kind = args[1].value
-        point = args[2]
-        orientation = args[3].value
+        for aidx in range(2, len(args)):
+            if args[aidx] in ('PLACED', 'FIXED'):
+                point = args[aidx+1]
+                orientation = args[aidx+2].value
                 self.def_file.components[name] = (kind, point, orientation)
 
     def pins_stmt(self, args):
@@ -207,6 +211,7 @@ GRAMMAR = r"""
                 | /DIEAREA/ point+ ";"
                 | /ROW/ ID ID NUMBER NUMBER ID do_step ";"
                 | /TRACKS/ /[XY]/ NUMBER "DO" NUMBER "STEP" NUMBER "LAYER" ID ";"
+                | /GCELLGRID/ /[XY]/ NUMBER "DO" NUMBER "STEP" NUMBER ";"
                 | propdef | vias | nondef | comp | pins | pinprop | spnets | nets
 
     propdef: "PROPERTYDEFINITIONS" propdef_stmt* "END" "PROPERTYDEFINITIONS"
@@ -228,7 +233,9 @@ GRAMMAR = r"""
                         | "+" /VIA/ ID )* ";"
 
     comp: "COMPONENTS" NUMBER ";" comp_stmt* "END" "COMPONENTS"
-    comp_stmt: "-" ID ID "+" "PLACED" point ID ";"
+    comp_stmt: "-" ID ID ( "+" /PLACED/ point ID
+                         | "+" /FIXED/ point ID
+                         | "+" /SOURCE/ ID )* ";"
 
     pins: "PINS" NUMBER ";" pins_stmt* "END" "PINS"
     pins_stmt: "-" ID pins_opt* ";"
@@ -239,6 +246,8 @@ GRAMMAR = r"""
             | "+" /PORT/
             | "+" /LAYER/ ID point point
             | "+" /PLACED/ point ID
+            | "+" /FIXED/ point ID
+
 
     pinprop: "PINPROPERTIES" NUMBER ";" pinprop_stmt* "END" "PINPROPERTIES"
     pinprop_stmt: "-" "PIN" ID "+" "PROPERTY" ID STRING ";"
@@ -266,10 +275,10 @@ GRAMMAR = r"""
     wire: ID wire_opt points
     wire_opt: ( "TAPER" | "TAPERRULE" ID )? ("STYLE" ID)?
 
-    points: point ( point | points_via )+
-    points_via: ID ORIENTATION?
+    points: point ( point | points_via_or_rect )+
+    points_via_or_rect: ID ( ORIENTATION? | "(" (NUMBER|SIGNED_NUMBER|/\*/) (NUMBER|SIGNED_NUMBER|/\*/) (NUMBER|SIGNED_NUMBER|/\*/) (NUMBER|SIGNED_NUMBER|/\*/) ")" )
 
-    point: "(" (NUMBER|/\*/) (NUMBER|/\*/) NUMBER? ")"
+    point: "(" (NUMBER|SIGNED_NUMBER|/\*/) (NUMBER|SIGNED_NUMBER|/\*/) NUMBER? ")"
 
     do_step: "DO" NUMBER "BY" NUMBER "STEP" (NUMBER|SIGNED_NUMBER) (NUMBER|SIGNED_NUMBER)
 
@@ -284,9 +293,9 @@ GRAMMAR = r"""
     """
 
 
-def parse(text):
+def parse(text) -> DefFile:
     """Parses the given ``text`` and returns a :class:`DefFile` object."""
-    return Lark(GRAMMAR, parser="lalr", transformer=DefTransformer()).parse(text)
+    return Lark(GRAMMAR, parser="lalr", transformer=DefTransformer()).parse(text) # type: ignore
 
 
 def load(file):

src/kyupy/logic.py

@@ -49,7 +49,7 @@ from collections.abc import Iterable
 import numpy as np
 from numpy.typing import DTypeLike
 
-from . import numba, hr_bytes
+from . import numba, hr_bytes, NestedNumericList
 
 
 ZERO = 0b000
@@ -104,7 +104,7 @@ _mv_merge = np.array([[0,0,0,1,1,1,1,1], # 0
                       ], dtype=np.uint8)
 
 
-def interpret(value):
+def interpret(value) -> int|NestedNumericList:
     """Converts characters, strings, and lists of them to lists of logic constants defined above.
 
     :param value: A character (string of length 1), Boolean, Integer, None, or Iterable.
@@ -123,11 +123,12 @@ def interpret(value):
     return UNKNOWN
 
 
-def mvarray(*a):
+def mvarray(*a) -> np.ndarray[tuple[int]|tuple[int,int],np.dtype[np.uint8]]:
     """Converts (lists of) Boolean values or strings into a multi-valued array.
 
-    The given values are interpreted and the axes are arranged as per KyuPy's convention.
-    Use this function to convert strings into multi-valued arrays.
+    Iterable over characters, Booleans, Integers (0,1), None are interpreted as a single pattern.
+    Multi-character strings are Iterables over characters and are interpreted as a single pattern.
+    Lists of said patterns form a pattern set as ndarray with the axes are arranged as per KyuPy's convention.
     """
     mva = np.array(interpret(a), dtype=np.uint8)
     if mva.ndim < 2: return mva

src/kyupy/logic_sim.py

@@ -51,6 +51,7 @@ class LogicSim(sim.SimOps):
         nbytes = cdiv(sims, 8)
 
         self.c = np.zeros((self.c_len, self.mdim, nbytes), dtype=np.uint8)
+        self.c_dirty = np.full(self.c_len, 1, dtype=np.uint8)
         self.s = np.zeros((2, self.s_len, 3, nbytes), dtype=np.uint8)
         """Logic values of the sequential elements (flip-flops) and ports.
 
@@ -97,7 +98,7 @@ class LogicSim(sim.SimOps):
         t1 = self.c_locs[self.tmp2_idx]
         if self.m == 2:
             if inject_cb is None:
-                c_prop_2v_cpu(self.ops, self.c_locs, self.c, int(fault_line), fault_mask, int(fault_model))
+                c_prop_2v_cpu(self.ops, self.c_locs, self.c, self.c_dirty, int(fault_line), fault_mask, int(fault_model))
             else:
                 for op, o0l, i0l, i1l, i2l, i3l in self.ops[:,:6]:
                     o0, i0, i1, i2, i3 = [self.c_locs[x] for x in (o0l, i0l, i1l, i2l, i3l)]
@@ -356,6 +357,19 @@ class LogicSim2V(sim.SimOps):
         Storage locations are indirectly addressed.
         Data for line `l` is in `self.c[self.c_locs[l]]`.
         """
+        self.c_dirty = np.full(self.c_len, 1, dtype=np.uint8)
+        """Marker for logic values that have changed recently for use in incremental simulation.
+
+        A node n will be evaluated by `c_prop()` if and only if at least one of its inputs is marked dirty (`c_dirty[c_locs[n.ins[i]]] == 1`).
+        If a node is evaluated, its output is marked dirty.
+        By default, this array is all-1, therefore every node is evaluated every time by `c_prop()`.
+        Enable incremental simulation by setting this array to all-0, change signal values in `c` and set `c_dirty` to 1 for all changed signals.
+        Next call to `c_prop()` will only evaluate nodes that read from the changed signals.
+        To restore the original simulation state, revert the signal value changes in `c`, and call `c_prop()` again.
+        Caveats:
+        - Injected changes that are located downstream of other changes will have no effect.
+        - Only works with `c_reuse=False` (default).
+        """
         self.s_assign = np.zeros((self.s_len, self.sims), dtype=np.uint8)
         """Logic values assigned to the ports and flip-flops of the circuit.
 
@@ -390,7 +404,7 @@ class LogicSim2V(sim.SimOps):
                 fault_mask2 = np.full(self.c.shape[-1], 0, dtype=np.uint8)
                 fault_mask2[:len(fault_mask)] = fault_mask
                 fault_mask = fault_mask2
-        c_prop_2v_cpu(self.ops, self.c_locs, self.c, int(fault_line), fault_mask, int(fault_model))
+        c_prop_2v_cpu(self.ops, self.c_locs, self.c, self.c_dirty, int(fault_line), fault_mask, int(fault_model))
 
     def c_to_s(self):
         """Captures the results of the combinational portion into ``self.s_result``.
@@ -424,9 +438,11 @@ class LogicSim2V(sim.SimOps):
 
 
 @numba.njit
-def c_prop_2v_cpu(ops, c_locs, c, fault_line, fault_mask, fault_model):
+def c_prop_2v_cpu(ops, c_locs, c, c_dirty, fault_line, fault_mask, fault_model):
     for op, o0l, i0l, i1l, i2l, i3l in ops[:,:6]:
         o0, i0, i1, i2, i3 = [c_locs[x] for x in (o0l, i0l, i1l, i2l, i3l)]
+        if fault_line < 0 or o0l != fault_line:  # fault injection forces node evaluation
+            if not (c_dirty[i0] | c_dirty[i1] | c_dirty[i2] | c_dirty[i3]): continue
         if op == sim.BUF1: c[o0]=c[i0]
         elif op == sim.INV1: c[o0] = ~c[i0]
         elif op == sim.AND2: c[o0] = c[i0] & c[i1]
@@ -466,8 +482,9 @@ def c_prop_2v_cpu(ops, c_locs, c, fault_line, fault_mask, fault_model):
                 c[o0] = c[o0] & ~fault_mask
             elif fault_model == 1:
                 c[o0] = c[o0] | fault_mask
-            else:
+            elif fault_model == 2:
                 c[o0] = c[o0] ^ fault_mask
+        c_dirty[o0] = 1
 
 
 class LogicSim4V(sim.SimOps):

src/kyupy/sim.py

@@ -189,10 +189,10 @@ class SimOps:
         levels = []
 
         ppio2idx = dict((n, i) for i, n in enumerate(circuit.s_nodes(KYUPY)))
-        root_nodes = set([n for n in circuit.s_nodes(KYUPY) if len(n.ins) > 0] + [n for n in circuit.nodes if len(n.outs) == 0])  # start from POs, PPOs, and any dangling nodes
+        root_nodes = set([n for n in circuit.nodes if len(n.outs) == 0 or KYUPY.is_dff(n.kind)])  # start from POs (that are unused by circuit itself), PPOs, and any dangling nodes
         readers = np.array([1 if l.reader in root_nodes else len(l.reader.outs) for l in circuit.lines], dtype=np.int32)  # for ref-counting forks
 
-        level_lines = [n.ins[0] for n in root_nodes if len(n.ins) > 0 ]
+        level_lines = [n.ins[0] for n in sorted(root_nodes, key=lambda n: n.index) if len(n.ins) > 0 ]
         # FIXME: Should probably instanciate buffers for PPOs and attach DFF clocks
 
         while len(level_lines) > 0:  # traverse the circuit level-wise back towards (P)PIs
@@ -201,10 +201,15 @@ class SimOps:
 
             for l in level_lines:
                 n = l.driver
-                assert len(n.ins) <= 4, f'Too many input pins for node {n}. Map circuit to KYUPY techlib first using resolce_tlib_cells().'
+                assert len(n.ins) <= 4, f'Too many input pins for node {n}. Map circuit to KYUPY techlib first using resolve_tlib_cells().'
                 in_idxs = [n.ins[x].index if len(n.ins) > x and n.ins[x] is not None else self.zero_idx for x in [0,1,2,3]]
                 if n in ppio2idx:
+                    in_idxs[1] = self.zero_idx
+                    if 'dff' in n.kind.lower() or in_idxs[0] == self.zero_idx:
                         in_idxs[0] = self.ppi_offset + ppio2idx[n]
+                    else:  # this is a PO that is also used within the circuit. Treat as normal fork with ref counting. TODO: check strip_forks=True case for this.
+                        readers[n.ins[0]] -= 1
+                        if readers[n.ins[0]] == 0: prev_level_lines.append(n.ins[0])
                     if l.driver_pin == 1 and 'dff' in n.kind.lower():  # second output of DFF is inverted
                         level_ops.append((INV1, l.index, *in_idxs, *a_ctrl[l]))
                     else:
@@ -312,6 +317,11 @@ class SimOps:
             if len(n.ins) > 0:
                 self.c_locs[self.ppo_offset + i], self.c_caps[self.ppo_offset + i] = self.c_locs[n.ins[0]], self.c_caps[n.ins[0]]
 
+        # sanity check
+        for oidx, io in zip(*np.nonzero(self.c_locs[self.ops[:, 1:6]] < 0)):
+            log.error(f'unallocated {["o0","i0","i1","i2","i3"][io]} in op {self.ops[oidx]}')
+        assert np.min(self.c_locs[self.ops[:, 1:6]]) >= 0, "Unallocated signal lines."
+
         # line use information
         self.line_use_start = np.full(self.c_locs_len, -1, dtype=np.int32)
         self.line_use_stop = np.full(self.c_locs_len, len(self.levels), dtype=np.int32)
@@ -346,3 +356,12 @@ class SimOps:
 
         self.pippi_c_locs = np.concatenate([self.pi_c_locs, self.ppi_c_locs])
         self.poppo_c_locs = np.concatenate([self.po_c_locs, self.ppo_c_locs])
+
+    def cone_ops(self, origin_lines: set[int]):
+        active_lines = {l for l in origin_lines}
+        ops = []
+        for kind, out0, in0, in1, in2, in3, *params in self.ops:
+            if in0 in active_lines or in1 in active_lines or in2 in active_lines or in3 in active_lines or out0 in origin_lines:
+                ops.append((kind, out0, in0, in1, in2, in3, *params))
+                active_lines.add(out0)
+        return np.asarray(ops, dtype=np.int32)

tests/test_logic_sim.py

@@ -20,6 +20,18 @@ def test_dangling():
     sim.simulate(p)
     assert p[1,0] == logic.ONE
 
+def test_used_output():
+    c = bench.parse('input(i0) output(o0) output(o1) o0=BUF1(i0) o1=BUF(o0)')
+    sim = LogicSim2V(c)
+    expect = mvarray('000', '111')
+    actual = sim.simulate(expect.copy())
+    np.testing.assert_array_equal(
+        expect[c.io_locs('o0')],
+        actual[c.io_locs('o0')])
+    np.testing.assert_array_equal(
+        expect[c.io_locs('o1')],
+        actual[c.io_locs('o1')])
+
 def assert_equal_shape_and_contents(actual, desired):
     desired = np.array(desired, dtype=np.uint8)
     assert actual.shape == desired.shape