timer_examples

timer() example (version9)

Self-contained correctness example for OpenVAF's new @(timer(t0, period)) event-control support (Enhancement-8, Feature A), covering DC, AC, and transient analysis. Uses the version9 toolchain:

• compiler : ../OpenVAF-master-20260610/target/opt/openvaf-r (built with --features openvaf-driver/llvm18)
• simulator: ../ngspice-46/build/src/ngspice (locally built, OSDI-capable — not the system-wide ngspice)

See ../Enhancement-8.md for the full implementation writeup, and ../cross_examples/README.md for the shared "known limitations" that also apply to timer.

The model

timer_demo.va: @(timer(t0, period)) count = count + 1.0; — fires once at t0, then periodically every period seconds after, incrementing a persistent tick counter exposed on V(out). Also reports each firing via $strobe (observable in ngspice's console output).

Running

../OpenVAF-master-20260610/target/opt/openvaf-r timer_demo.va -o timer_demo.osdi
../ngspice-46/build/src/ngspice -b dc_sim_timer.cir
../ngspice-46/build/src/ngspice -b ac_sim_timer.cir
../ngspice-46/build/src/ngspice -b tran_timer.cir
python3 plot_timer.py   # -> timer_dc.png, timer_ac.png, timer_tran.png

DC / AC: documented non-behavior

timer_demo.va has no voltage input port (it's a purely time-triggered generator, V(out) <+ count;) — so unlike above()/cross()'s demos, there's no natural node to sweep:

• DC (dc_sim_timer.cir) sweeps TEMP instead of a voltage. A DC operating-point solve has no notion of elapsed time, so t0 is never reached — timer_dc.png shows V(out) = count pinned at exactly 0 from -40°C to 125°C.
• AC (ac_sim_timer.cir) injects a probe current directly into out and measures its response. The AC operating point is likewise a single fixed (t=0) bias, so count stays 0 and V(out) <+ count behaves as a stiff (zero small-signal impedance) contribution — timer_ac.png shows the response pinned at (numerically) zero from 1Hz to 1MHz.

Both are "documented non-behavior" checks, same convention as cross_examples/README.md's DC/AC. timer()'s actual (real, periodic, persistent-counter) behavior only shows up in the transient plot.

Transient: the real event-detection behavior

timer_tran.png shows V(out) = count as a staircase, stepping up by 1 exactly at t0 and every period after.

Results (raw firing data)

tran_timer.cir sets t0 = 2ms, period = 1ms, over a 7ms transient sweep. Expected: first firing at t=2ms, then every 1ms after (2,3,4,5,6ms), plus whatever the last accepted timepoint near t=7ms happens to be:

timer fired at t=0.0020028 count=1
timer fired at t=0.0030028 count=2
timer fired at t=0.0040028 count=3
timer fired at t=0.0050028 count=4
timer fired at t=0.0060028 count=5
timer fired at t=0.007 count=6

Matches exactly (first firing at t0, then period-spaced, count incrementing by exactly 1 per firing — persistent state working correctly).

Compile-time verification (--dump-unopt-mir)

--dump-unopt-mir confirms timer's scheduling logic compiles to a real, non-dead-code conditional branch:

v21 = fge v20, v19    // fired = Abstime >= next_fire_time
v22 = fadd v19, v16   // rescheduled = next_fire_time + period
br v21, block5, block6

Known limitations

See ../cross_examples/README.md — the "no persistent state inside the event body" compiler bug that used to block this example's tick counter has been fixed (see ../Enhancement-8.md §2, limitation 1); the "eval-granularity, not exact-time-forced" detection design still applies. The "first .model-card override ignored" ngspice bug that used to affect t0 here has also been fixed (see ../Enhancement-8.md §2, limitation 3) — no workaround needed anymore.