faster mv xor and new merge fct

src/kyupy/bench.py

@@ -5,6 +5,10 @@ Historically it was first used in the
 `ISCAS89 benchmark set <https://people.engr.ncsu.edu/brglez/CBL/benchmarks/ISCAS89/>`_.
 Besides loading these benchmarks, this module is also useful for easily constructing simple circuits:
 ``c = bench.parse('input(x, y) output(a, o, n) a=and(x,y) o=or(x,y) n=not(x)')``.
+
+This parser automatically translates gate to simulation primitives supported by kyupy.
+In most cases this means just adding the number of inputs to the name: e.g. NAND -> NAND2, OR -> OR4, INV -> INV1.
+Since simulators only support at most four-input-gates, gates with more than four inputs are implemented as tree.
 """
 
 from lark import Lark, Transformer

src/kyupy/circuit.py

@@ -614,19 +614,23 @@ class Circuit:
             if marks[n]:
                 yield n
 
-    def fanout(self, origin_nodes):
+    def fanout(self, origin_nodes: list[Node], node_filter = lambda n: 'dff' not in n.name.lower()):
         """Generator function to iterate over the fan-out cone of a given list of origin nodes.
 
-        Nodes are yielded in topological order.
+        origin_nodes are yielded first, followed by nodes driven by them in a breadth-first manner.
+        The search stops at nodes for which node_filter returns False.
+        Only origin_nodes and nodes for which node_filter returned True are yielded.
+        By default, search stops at flip-flops.
         """
-        marks = [False] * len(self.nodes)
-        for n in origin_nodes:
-            marks[n] = True
-        for n in self.topological_order():
-            if not marks[n]:
-                for line in n.ins.without_nones():
-                    marks[n] |= marks[line.driver]
-            if marks[n]:
+        queue = deque(origin_nodes)
+        yielded = set()
+        while len(queue) > 0:
+            n = queue.popleft()
+            for line in n.outs.without_nones():
+                succ = line.reader
+                if succ not in yielded and node_filter(succ):
+                    yielded.add(succ)
+                    queue.append(succ)
             yield n
 
     def fanout_free_regions(self):

src/kyupy/logic.py

@@ -81,6 +81,29 @@ on a signal. ``'N'``, ``'n'``, and ``'v'`` are interpreted as ``NPULSE``.
 """
 
 
+#                      0 X - 1 P R F N
+_mv_xor   = np.array([[0,1,1,3,4,5,6,7], # 0
+                      [1,1,1,1,1,1,1,1], # X
+                      [1,1,1,1,1,1,1,1], # -
+                      [3,1,1,0,7,6,5,4], # 1
+                      [4,1,1,7,4,5,6,7], # P
+                      [5,1,1,6,5,4,7,6], # R
+                      [6,1,1,5,6,7,4,5], # F
+                      [7,1,1,4,7,6,5,4], # N
+                      ], dtype=np.uint8)
+
+#                      0 X - 1 P R F N
+_mv_merge = np.array([[0,0,0,1,1,1,1,1], # 0
+                      [0,1,1,3,4,5,6,7], # X
+                      [0,1,2,3,4,5,6,7], # -
+                      [1,3,3,3,1,1,1,1], # 1
+                      [1,4,4,1,4,1,1,1], # P
+                      [1,5,5,1,1,5,1,1], # R
+                      [1,6,6,1,1,1,6,1], # F
+                      [1,7,7,1,1,1,1,7], # N
+                      ], dtype=np.uint8)
+
+
 def interpret(value):
     """Converts characters, strings, and lists of them to lists of logic constants defined above.
 
@@ -191,17 +214,6 @@ def mv_and(x1, x2, out=None):
     return out
 
 
-def _mv_xor(out, *ins):
-    any_unknown = (ins[0] == UNKNOWN) | (ins[0] == UNASSIGNED)
-    for inp in ins[1:]: any_unknown |= (inp == UNKNOWN) | (inp == UNASSIGNED)
-
-    out[...] = ZERO
-    for inp in ins:
-        np.bitwise_xor(out, inp & 0b011, out=out)
-        np.bitwise_or(out, inp & 0b100, out=out)
-    np.putmask(out, any_unknown, UNKNOWN)
-
-
 def mv_xor(x1, x2, out=None):
     """A multi-valued XOR operator.
 
@@ -211,7 +223,25 @@ def mv_xor(x1, x2, out=None):
     :return: A multi-valued array with the result.
     """
     out = out or np.empty(np.broadcast(x1, x2).shape, dtype=np.uint8)
-    _mv_xor(out, x1, x2)
+    out[...] = _mv_xor[x1, x2]
+    return out
+
+
+def mv_merge(x1, x2, out=None):
+    """A multi-valued merge operator.
+
+    Merging identical values ``a`` and ``a`` yields ``a``.
+    Merging ``X`` and ``-`` yields ``X``.
+    Merging any value ``a`` with ``X`` or ``-`` yields ``a``.
+    Merging any other values ``a != b`` yields ``X``.
+
+    :param x1: A multi-valued array.
+    :param x2: A multi-valued array.
+    :param out: An optional storage destination. If None, a new multi-valued array is returned.
+    :return: A multi-valued array with the result.
+    """
+    out = out or np.empty(np.broadcast(x1, x2).shape, dtype=np.uint8)
+    out[...] = _mv_merge[x1, x2]
     return out
 
 

src/kyupy/logic_sim.py

@@ -492,7 +492,7 @@ class LogicSim4V(sim.SimOps):
         self._full_mask = np.full(self.c.shape[-1], 255, dtype=np.uint8)
 
     def __repr__(self):
-        return f'{{name: "{self.circuit.name}", sims: {self.sims}, c_bytes: {eng(self.c.nbytes)}}}'
+        return f'{{name: "{self.circuit.name}", sims: {self.sims}, s_len: {self.s_len}, c_bytes: {eng(self.c.nbytes)}}}'
 
     def s_to_c(self):
         """Assigns the values from ``self.s_assign`` to the inputs of the combinational portion.

src/kyupy/sim.py

@@ -191,7 +191,7 @@ class SimOps:
         root_nodes = set([n for n in circuit.s_nodes if len(n.ins) > 0] + [n for n in circuit.nodes if len(n.outs) == 0])  # start from POs, 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]
+        level_lines = [n.ins[0] for n in root_nodes 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
@@ -330,11 +330,16 @@ class SimOps:
         self.prim_counts = dict(d)
 
         self.pi_s_locs = np.flatnonzero(self.c_locs[self.ppi_offset+np.arange(len(self.circuit.io_nodes))] >= 0)
+        """Locations in pattern array (axis 0) of all primary inputs (excluding flip-flops)."""
         self.po_s_locs = np.flatnonzero(self.c_locs[self.ppo_offset+np.arange(len(self.circuit.io_nodes))] >= 0)
+        """Locations in pattern array (axis 0) of all primary outputs (excluding flip-flops)."""
         self.ppio_s_locs = np.arange(len(self.circuit.io_nodes), self.s_len)
+        """Locations in pattern array (axis 0) of all flip-flops. Pseudo-primary inputs (PPI) and pseudo-primary outputs (PPO) share the same location."""
 
         self.pippi_s_locs = np.concatenate([self.pi_s_locs, self.ppio_s_locs])
+        """Locations in pattern array (axis 0) of all primary and pseudo-primary inputs (input ports and flip-flops)."""
         self.poppo_s_locs = np.concatenate([self.po_s_locs, self.ppio_s_locs])
+        """Locations in pattern array (axis 0) of all primary and pseudo-primary outputs (output ports and flip-flops)."""
 
         self.pi_c_locs = self.c_locs[self.ppi_offset+self.pi_s_locs]
         self.po_c_locs = self.c_locs[self.ppo_offset+self.po_s_locs]

src/kyupy/techlib.py

@@ -128,7 +128,7 @@ OAI21X1 input(A0,A1,B0)          output(Y)    Y=OAI21(A0,A1,B0)    ;
 OAI22X1 input(A0,A1,B0,B1)       output(Y)    Y=OAI22(A0,A1,B0,B1) ;
 OAI33X1 input(A0,A1,A2,B0,B1,B2) output(Y)    AA=OR2(A0,A1) BB=OR2(B0,B1) Y=OAI22(AA,A2,BB,B2) ;
 ADDFX1  input(A,B,CI)            output(CO,S) AB=XOR2(A,B) CO=XOR2(AB,CI) S=AO22(AB,CI,A,B)    ;
-ADDHX1  input(A,B)               output(CO,S) CO=XOR2(A,B) S=AND2(A,B)                         ;
+ADDHX1  input(A,B)               output(CO,S) CO=AND2(A,B) S=XOR2(A,B)                         ;
 
 DFFX1    input(CK,D)             output(Q,QN) Q=DFF(D,CK) QN=INV1(Q) ;
 DFFSRX1  input(CK,D,RN,SN)       output(Q,QN) DR=AND2(D,RN) SET=INV1(SN) DRS=OR2(DR,SET) Q=DFF(DRS,CK) QN=INV1(Q) ;
@@ -172,7 +172,7 @@ XNOR2_X1     input(A1,A2)       output(ZN) ZN=XNOR2(A1,A2)       ;
 
 MUX2_X1 input(I0,I1,S) output(Z) Z=MUX21(I0,I1,S) ;
 
-HA_X1 input(A,B) output(CO,S) CO=XOR2(A,B) S=AND2(A,B) ;
+HA_X1 input(A,B) output(CO,S) CO=AND2(A,B) S=XOR2(A,B) ;
 FA_X1 input(A,B,CI) output(CO,S) AB=XOR2(A,B) CO=XOR2(AB,CI) S=AO22(CI,A,B) ;
 
 AOI21_X{1,2} input(A1,A2,B)     output(ZN) ZN=AOI21(A1,A2,B)     ;
@@ -225,7 +225,7 @@ XNOR2_X{1,2}   input(A,B)         output(ZN) ZN=XNOR2(A,B)         ;
 
 MUX2_X{1,2} input(A,B,S) output(Z) Z=MUX21(A,B,S) ;
 
-HA_X1 input(A,B) output(CO,S) CO=XOR2(A,B) S=AND2(A,B) ;
+HA_X1 input(A,B) output(CO,S) CO=AND2(A,B) S=XOR2(A,B) ;
 FA_X1 input(A,B,CI) output(CO,S) AB=XOR2(A,B) CO=XOR2(AB,CI) S=AO22(CI,A,B) ;
 
 AOI21_X{1,2,4} input(A,B1,B2)     output(ZN) ZN=AOI21(B1,B2,A)     ;
@@ -340,7 +340,7 @@ MUX41X{1,2}$ input(A1,A2,A3,A4,S0,S1) output(Y) A=MUX21(A1,A2,S0) B=MUX21(A3,A4,
 
 DEC24X{1,2}$ input(A0,A1) output(Y0,Y1,Y2,Y3) A0B=INV1(A0) A1B=INV1(A1) Y0=NOR2(A0,A1) Y1=AND(A0,A1B) Y2=AND(A0B,A1) Y3=AND(A0,A1) ;
 FADDX{1,2}$ input(A,B,CI) output(S,CO) AB=XOR2(A,B) CO=XOR2(AB,CI) S=AO22(AB,CI,A,B) ;
-HADDX{1,2}$ input(A0,B0) output(SO,C1) C1=XOR2(A0,B0) SO=AND2(A0,B0) ;
+HADDX{1,2}$ input(A0,B0) output(SO,C1) C1=AND2(A0,B0) SO=XOR2(A0,B0) ;
 
 {,AO}DFFARX{1,2}$ input(D,CLK,RSTB)      output(Q,QN) DR=AND2(D,RSTB) Q=DFF(DR,CLK) QN=INV1(Q) ;
 DFFASRX{1,2}$     input(D,CLK,RSTB,SETB) output(Q,QN) DR=AND2(D,RSTB) SET=INV1(SETB) DRS=OR2(DR,SET) Q=DFF(DRS,CLK) QN=INV1(Q) ;
@@ -430,7 +430,7 @@ MUX41X{1,2}$ input(IN1,IN2,IN3,IN4,S0,S1) output(Q) A=MUX21(IN1,IN2,S0) B=MUX21(
 
 DEC24X{1,2}$ input(IN1,IN2) output(Q0,Q1,Q2,Q3) IN1B=INV1(IN1) IN2B=INV1(IN2) Q0=NOR2(IN1,IN2) Q1=AND(IN1,IN2B) Q2=AND(IN1B,IN2) Q3=AND(IN1,IN2) ;
 FADDX{1,2}$ input(A,B,CI) output(S,CO) AB=XOR2(A,B) CO=XOR2(AB,CI) S=AO22(AB,CI,A,B) ;
-HADDX{1,2}$ input(A0,B0) output(SO,C1) C1=XOR2(A0,B0) SO=AND2(A0,B0) ;
+HADDX{1,2}$ input(A0,B0) output(SO,C1) C1=AND2(A0,B0) SO=XOR2(A0,B0) ;
 
 {,AO}DFFARX{1,2}$ input(D,CLK,RSTB)      output(Q,QN) DR=AND2(D,RSTB) Q=DFF(DR,CLK) QN=INV1(Q) ;
 DFFASRX{1,2}$     input(D,CLK,RSTB,SETB) output(Q,QN) DR=AND2(D,RSTB) SET=INV1(SETB) DRS=OR2(DR,SET) Q=DFF(DRS,CLK) QN=INV1(Q) ;

src/kyupy/vcd.py

@@ -41,6 +41,9 @@ class Scope:
             s = s.parent_scope
         return '/'.join(reversed(parts))
 
+    def __repr__(self) -> str:
+        return self.path
+
 
 class VcdHeader:
     def __init__(self) -> None:
@@ -99,10 +102,12 @@ class VcdVarMap:
 
 
 class VcdData:
-    def __init__(self, var_map, steps, data):
+    def __init__(self, var_map: VcdVarMap, steps, data):
         self.var_map = var_map
         self.steps = steps
         self.data = data
+    def __repr__(self) -> str:
+        return f'{{variables: {len(self.var_map.var_list)}, bits: {self.data.shape[0]}, steps: {self.data.shape[1]}}}'
 
 
 class VcdHeaderTransformer(Transformer):
@@ -124,9 +129,9 @@ class VcdHeaderTransformer(Transformer):
 
     def var(self, args):
         type, width, idcode, reference = args
-        reference = reference.strip()
+        reference = reference.value.strip()
         width = int(width)
-        self.header.add_var(type, width, idcode, reference)
+        self.header.add_var(type.value, width, idcode.value, reference)
 
     def start(self, args):
         return self.header
@@ -149,12 +154,13 @@ GRAMMAR = r"""
     """
 
 
-def load(file, var_locs = lambda _: [], step_filter = lambda *_: True):
+def load(file, var_locs = lambda _: [], step_filter = lambda *_: True, s_len = None) -> VcdData:
     """Parses the contents of ``file`` as Verilog Change Dump (VCD).
 
     :param file: A file name or a file handle. Files with `.gz`-suffix are decompressed on-the-fly.
     :param var_locs: A callback ``(var) -> list[int]`` mapping each variable to ndarray column indices. Empty list drops the variable.
     :param step_filter: A callback ``(time, values, var_map) -> bool`` to select which timesteps to include.
+    :param s_len: length of the first axis of returned pattern data.
     :return: A VcdData object with metadata and an ndarray with all values.
     """
     vcd = readtext(file)
@@ -164,10 +170,11 @@ def load(file, var_locs = lambda _: [], step_filter = lambda *_: True):
     vcd_header : VcdHeader = Lark(GRAMMAR, parser="lalr", lexer='contextual', transformer=VcdHeaderTransformer()).parse(vcd_header_str) # type: ignore
     vcd_data = vcd[header_size:].splitlines()
     var_map = VcdVarMap(vcd_header, var_locs)
+    width = s_len if s_len is not None else var_map.total_width
 
     chunk_size = 10240
     chunks = []
-    chunk = np.full((chunk_size, var_map.total_width), logic.UNASSIGNED, dtype=np.uint8)
+    chunk = np.full((chunk_size, width), logic.UNASSIGNED, dtype=np.uint8)
 
     _val_map = {'0': logic.ZERO, '1': logic.ONE,
                 'x': logic.UNKNOWN, 'X': logic.UNKNOWN,
@@ -186,7 +193,7 @@ def load(file, var_locs = lambda _: [], step_filter = lambda *_: True):
                 steps.append(current_time)
                 if step_idx >= chunk_size:
                     chunks.append(chunk)
-                    chunk = np.empty((chunk_size, var_map.total_width), dtype=np.uint8)
+                    chunk = np.empty((chunk_size, width), dtype=np.uint8)
                     chunk[0] = chunks[-1][-1]
                     step_idx = 0
                 else: