fix sim scheduler for circuits with POs that are also used internally

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)]

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 +201,14 @@ 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[0] = self.ppi_offset + ppio2idx[n]
+                    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 +316,11 @@ 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 +355,12 @@ 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