Repository navigation
analog: bench.t27 -- ngspice wall time and a fixed-step backward-Euler reference for the M0/M1 circuits (Closes #10505) - #10517
Merged
Conversation
…r reference for the M0/M1 circuits (Closes #10505) specs/analog/bench.t27 records, for five circuits (the M0/M1 RC, the series RLC, 3-, 5- and 10-node RC ladders), ngspice-47 wall time (median of 10 runs, process start included), ngspice's own .tran analysis time, internal time points, and two deviations. It also encodes the claim rule: may_claim_speedup is true only with a measured board time and wall(ngspice) / wall(board incl. readout) >= 1. Fixed-step finding: ngspice reproduces backward Euler on the exact grid to float64 rounding (largest deviation 1e-15 to 9.4e-15 V against an f64 LU recurrence) with method=gear maxord=1 trtol=1e7, a B source (no breakpoints) and a pre-roll in which the circuit rests at 0 V. A PWL source cannot: its corners are breakpoints and ngspice restarts its step ramp after each, leaving 4 of 217 points on the grid and v(out) at 5 us off by 1.7e-5 V. specs/analog/decks/ holds a fixed and an adaptive deck per circuit and rc_pwl_naive.cir as the negative control. Measured on Apple M1, macOS 26.3.1, sandboxed session: ngspice wall is 14 to 15 ms for every row, of which .tran is 0.8 to 1.0 ms; it is flat from 1 to 10 unknowns. M0's board time is 12 us compute + 9375 us UART line = 9387 us, so the wall ratio is 1.52; that compares two fixed overheads. Only M0 has a board number; the RLC and the 5-node ladder wait for M1b (#10496); the bitstream build time is not in any receipt, so it is recorded as not measured and no break-even count is stated. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
gHashTag
enabled auto-merge (squash)
October 18, 2026 16:07
Contributor
gHashTag
disabled auto-merge
October 18, 2026 16:08
…aim for any row yet (Closes #10505) Review on PR #10517: may_claim_speedup allowed M0 (wall ratio 1.52), but that ratio compares ngspice process start-up with the UART line, measured inside a sandboxed session, and the bitstream build time is not measured, so the break-even run count FR-001 asks for does not exist. The rule now also requires build_us > 0; with BUILD_US = 0 no row is claimable, and a test shows the same M0 numbers would pass once a build is measured. 11/11 tests, 0 vacuous; seal re-saved and verified. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
PR DashboardGenerated at: 2026-10-18 16:10:40 UTC
Summary
Seal Status
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #10505
Part of #10486 (W7 and W10 of
docs/analog/competitors.md).specs/analog/bench.t27records ngspice-47 wall time for five circuits, the board side where a receipt exists, the claim rule, and the tests that keep them honest.specs/analog/decks/holds the netlists (plain SPICE data, no.controlblock): a fixed-step backward-Euler deck and a default-adaptive deck per circuit, andrc_pwl_naive.ciras the negative control.Measurements
Apple M1, 8 GB, macOS 26.3.1 arm64, sandboxed session (Seatbelt), 2026-10-18, ngspice-47 from Homebrew (SPARSE 1.3). Command:
ngspice -b -r <name>.raw specs/analog/decks/<name>.cir. Wall = whole process, median of 10 timed runs after one warm-up. Deviation columns are volts.Board (only M0 has a receipt): 1200 clocks at 100 MHz = 12 us compute, plus one 108-character line at 115200 baud = 9375 us, so 9387 us including readout (computed from the receipt, not a stopwatch reading). The RLC and the 5-node ladder are pending (M1b, #10496); the 3-node and 10-node ladders have no board top. The bitstream build time is in no receipt in this repository and was not measured: it is recorded as not measured and no break-even count is stated.
Reading: for every row ngspice's wall is 14 to 15 ms and more than 90 percent of it is start-up and file work (
.tranis 0.8 to 1.0 ms); it is flat from 1 to 10 unknowns. On M0 the wall ratio is 14228 / 9387 = 1.52, a ratio of two fixed overheads (process start-up against a UART line), not of two simulators. The compute-only ratio (796 us over 12 us = 66) is a different quantity and is not a speed claim about a run, because the user of a run pays the readout.may_claim_speedupis true only with a measured board time and a ratio of at least 1, and the test over the table finds exactly one claimable row (M0), robust to the fastest of the 10 ngspice runs.Fixed-step finding (W10): ngspice as an independent backward-Euler reference
ngspice reproduces backward Euler on the exact grid to float64 rounding. Target was 1e-9 V; achieved 1e-15 to 9.4e-15 V at every one of the 200 grid points, every node, for all five circuits, against an independent f64 LU recurrence; the 200 steps equal dt to 2.2e-14 relative.
method=gear maxord=1is backward Euler.trtol=1e7stops the truncation test from shrinking a step (without it the step collapses to 2.1e-13 s at the source edge; withoutmaxord=1the steps stay at 50 ns but the method is trapezoidal, 0.024 V from backward Euler on the RC)..tranwith a ramp of steps that double up to tmax and restarts the ramp after every breakpoint; a PWL corner or PULSE edge is a breakpoint. With the natural PWL source (rc_pwl_naive.cir) there are 217 internal points, 4 of them on the 50 ns grid, v(out) at 5 us is off the recurrence by 1.7e-5 V (0.992412 against 0.992396), at 10 us by 1.2e-4 V, and 0.016 V at the first grid point.-r) holds float64 values. The price is 20 extra internal points (220, not 201) and a source that is a function of time.Not covered
linear.t27; it is there to read the ngspice time against size.Gate
t27c test-report specs/analog/bench.t27: 11 tests, 11 pass, 0 vacuous passes.tri test specs/analog/bench.t27: 11 pass, 0 fail.t27c seal --savethen--verify: all hashes MATCH.>=flipped to<=, FIX_RC_A changed, BOARD_WALL_US off by one microsecond, a pending row marked measured) each fail the tests.🤖 Generated with Claude Code