Single-threaded performance

[jodavies]

At the 2026 workshop, we discussed the possibility of simplifying the code by "removing form", i.e. making tform the only compilation mode, and tform -w0 taking the place of form for single-threaded jobs (and tform -w1 being equivalent to tform -w0, which it is not currently).

To start, here are benchmarks comparing form with tform -w0 and tform -w1. The performance discrepancies in a few of the benchmarks need to be resolved, before we can considering removing form as a supported build.

Parallel speedup w.r.t form-master

Benchmark	tform-master -w0	tform-master -w1
chromatic	0.99 ± 0.00	0.99 ± 0.01
color	1.01 ± 0.01	1.01 ± 0.01
fmft	1.01 ± 0.01	0.99 ± 0.01
forcer	1.01 ± 0.00	0.97 ± 0.00
forcer-exp	1.00 ± 0.00	0.96 ± 0.00
hyperform	1.00 ± 0.01	0.98 ± 0.00
ibp-box2l	0.99 ± 0.00	0.98 ± 0.00
ibp-mbox1l	0.99 ± 0.00	0.99 ± 0.00
ibp-npbox2l	0.99 ± 0.00	0.99 ± 0.00
ibp-nptri3l	0.98 ± 0.01	0.97 ± 0.00
ibp-pent1l	1.00 ± 0.00	0.99 ± 0.00
ibp-tri3l	0.99 ± 0.02	0.98 ± 0.02
minceex	1.00 ± 0.00	0.99 ± 0.00
mincer	0.96 ± 0.00	0.96 ± 0.00
mzv-dm	1.02 ± 0.01	0.93 ± 0.01
sort-disk	1.00 ± 0.02	0.99 ± 0.02
sort-large	0.98 ± 0.01	0.96 ± 0.02
sort-small	0.98 ± 0.01	0.97 ± 0.01
trace	0.93 ± 0.01	0.93 ± 0.01