implement an incremental fsim

README.md

@@ -40,7 +40,7 @@ Choose `uv-env` as kernel in Jupyter Lab.
 
 Numbers are given in `gfs/s` = `gates * faults * patterns / second`.
 
-Stuck-at fault sim on 1024 patterns, baseline:
+`SAFSimSimple` on 1024 patterns (baseline):
 
 | OS, CPU, RAM | `tests/c6288.bench` | `polito-itc99-b15-sky130` |
 |--------------|---------------------|---------------------------|

main.py

@@ -10,7 +10,8 @@ from kyupy import verilog, bench, log, logic, batchrange, atalanta, stil
 from kyupy.techlib import techlib_by_name, KYUPY
 
 from fsim.static import LineRoles, FaultSet
-from fsim.baseline import SAFSimSimple
+from fsim.simple import SAFSimSimple
+from fsim.incremental import SAFSimIncremental
 
 
 def main():
@@ -80,8 +81,8 @@ def main():
 
     fclasses = safsim.classify_faults(saf_collapsed, patterns)
 
-    log.info(f'fsim time: {safsim.sim_time:.2f}s')
+    log.info(f'{safsim.timers=}')
-    sim_performance = stats['comb'] * len(saf_collapsed) * patterns.shape[1] / safsim.sim_time
+    sim_performance = stats['comb'] * len(saf_collapsed) * patterns.shape[1] / safsim.timers['sim'].s
     log.info(f'fsim performance: {sim_performance:.2e} gfp/s')
     log.info(f'detected by simulation (collapsed): {len(fclasses["DS"])}/{len(saf_collapsed)}  -  {len(fclasses["DS"])/len(saf_collapsed)*100:.2f}%')
 

src/fsim/baseline.py

@@ -1,47 +0,0 @@
-import time
-from collections.abc import Collection
-
-import numpy as np
-
-from kyupy import log, batchrange
-from kyupy.circuit import Circuit
-from kyupy.logic_sim import LogicSim2V
-
-class SAFSimSimple:
-
-    def __init__(self, circuit_resolved: Circuit, batch_size: int):
-        self.sim = LogicSim2V(circuit_resolved, sims=batch_size)
-        self.sim_time = 0
-        pass
-
-    def classify_faults(self, faults: Collection[int], patterns: np.ndarray):
-
-        golden = np.zeros_like(patterns)
-        self.sim.simulate(patterns, golden)
-        log.info(f'golden sim finished.')
-
-        syndrome = np.zeros_like(patterns)
-
-        fclass_NO = set()
-        fclass_DS = set()
-
-        start_time = time.perf_counter()
-
-        with log.progress() as p:
-            for fidx, fault in enumerate(faults):
-                fault_site = fault//2
-                fault_polarity = fault&1
-                p.update((fidx+1) / len(faults), f'DS:{len(fclass_DS)} NO:{len(fclass_NO)}')
-                for bo, bs in batchrange(patterns.shape[1], self.sim.sims):
-                    self.sim.s_assign[:, :bs] = patterns[:, bo:bo+bs]
-                    self.sim.s_to_c()
-                    self.sim.c_prop(fault_line=fault_site, fault_model=fault_polarity)
-                    self.sim.c_to_s()
-                    syndrome[:, bo:bo+bs] = self.sim.s_result[:,:bs]
-                if np.allclose(golden, syndrome):
-                    fclass_NO.add(fault)
-                else:
-                    fclass_DS.add(fault)
-
-        self.sim_time += time.perf_counter() - start_time
-        return {'DS': fclass_DS, 'NO': fclass_NO}

src/fsim/incremental.py

@@ -0,0 +1,47 @@
+import time
+from collections.abc import Collection
+
+import numpy as np
+
+from kyupy import log, batchrange, cdiv, Timers
+from kyupy.circuit import Circuit
+from kyupy.logic_sim import LogicSim2V
+
+class SAFSimIncremental:
+
+    def __init__(self, circuit_resolved: Circuit, batch_size: int):
+        self.sim = LogicSim2V(circuit_resolved, sims=batch_size)
+        self.timers = Timers()
+
+    def classify_faults(self, faults: Collection[int], patterns: np.ndarray):
+
+        with self.timers['startup']:
+            self.sim.c_prop()  # trigger jit
+            c_golden = np.zeros_like(self.sim.c)
+            fclass_DS = set()
+            nbatches = cdiv(patterns.shape[1], self.sim.sims)
+            nfaults = len(faults)
+
+        with self.timers['sim'], log.progress() as p:
+            for bidx, (bo, bs) in enumerate(batchrange(patterns.shape[1], self.sim.sims)):
+                self.sim.s_assign[:, :bs] = patterns[:, bo:bo+bs]
+                self.sim.s_to_c()
+                self.sim.c_dirty[...] = 1
+                with self.timers['sim_full_prop']:
+                    self.sim.c_prop()
+                c_golden[...] = self.sim.c
+                c_golden_poppo = c_golden[self.sim.poppo_c_locs]
+                for fidx, fault in enumerate(faults):
+                    fault_site = fault//2
+                    fault_polarity = fault&1
+                    self.sim.c_dirty[...] = 0
+                    with self.timers['sim_incr_prop']:
+                        self.sim.c_prop(fault_line=fault_site, fault_model=fault_polarity)
+                    with self.timers['sim_incr_eval']:
+                        if not np.all(c_golden_poppo == self.sim.c[self.sim.poppo_c_locs]):
+                            fclass_DS.add(fault)
+                    with self.timers['sim_incr_reset']:
+                        self.sim.c[...] = c_golden # clear fault
+                    p.update(((fidx+1) / nfaults) * ((bidx+1) / nbatches))
+
+        return {'DS': fclass_DS, 'NO': set(faults) - fclass_DS}

src/fsim/simple.py

@@ -0,0 +1,44 @@
+import time
+from collections.abc import Collection
+
+import numpy as np
+
+from kyupy import log, batchrange, Timers
+from kyupy.circuit import Circuit
+from kyupy.logic_sim import LogicSim2V
+
+class SAFSimSimple:
+
+    def __init__(self, circuit_resolved: Circuit, batch_size: int):
+        self.sim = LogicSim2V(circuit_resolved, sims=batch_size)
+        self.timers = Timers()
+
+    def classify_faults(self, faults: Collection[int], patterns: np.ndarray):
+
+        with self.timers['startup']:
+            golden = np.zeros_like(patterns)
+            self.sim.simulate(patterns, golden)
+            syndrome = np.zeros_like(patterns)
+            fclass_NO = set()
+            fclass_DS = set()
+
+        with self.timers['sim'], log.progress() as p:
+            for fidx, fault in enumerate(faults):
+                fault_site = fault//2
+                fault_polarity = fault&1
+                p.update((fidx+1) / len(faults), f'DS:{len(fclass_DS)} NO:{len(fclass_NO)}')
+                for bo, bs in batchrange(patterns.shape[1], self.sim.sims):
+                    self.sim.s_assign[:, :bs] = patterns[:, bo:bo+bs]
+                    self.sim.s_to_c()
+                    with self.timers['sim_prop']:
+                        self.sim.c_prop(fault_line=fault_site, fault_model=fault_polarity)
+                    with self.timers['sim_eval']:
+                        self.sim.c_to_s()
+                        syndrome[:, bo:bo+bs] = self.sim.s_result[:,:bs]
+                with self.timers['sim_eval2']:
+                    if np.allclose(golden, syndrome):
+                        fclass_NO.add(fault)
+                    else:
+                        fclass_DS.add(fault)
+
+        return {'DS': fclass_DS, 'NO': fclass_NO}