perf(vm): non-allocating finally-handler check on returns [2/4 of #39]

[mparrett]

Part 2 of the #39 split.

findAllExceptionHandlers allocates a []*ExceptionHandler on every OpReturn/OpReturnUndefined, where the common case is an empty exception table and the result is immediately reduced to "first handler that is a finally". findPendingHandler fuses that filter into the scan and returns the handler directly.

It's an exact refactor of the selection logic: same frameCount/closure guards, same exception table, same iteration order, same pc >= TryStart && pc < TryEnd predicate. The includeIterCleanup flag covers the two call sites that also accept IsIteratorCleanup. findAllExceptionHandlers is unchanged and still used by the unwinding and generator paths.

Also reorders vm.frameCount == 1 && function != nil && function.Name == "<script>" so the plain load short-circuits before the string compare, which otherwise ran on every function return.

Where this sits in the #39 split

#39 bundled ~6 changes; per review it's split into four independent PRs, all branched off current main:

	branch	contents
1	perf/vm-numeric-compare-rem	comparison fast path, integer %, drop unreachable per-op guards
2	perf/vm-finally-nonalloc	non-allocating finally-handler check
3	perf/vm-dispatch-deadcode	dead debug blocks, redundant IsNaN cleanup
4	perf/vm-array-index-accessor-atomic	arrayIndexAccessorSeen → atomic.Bool

The other three pieces you flagged as unmentioned — the IsObject() range check, the ToInteger() fast path, and the arrayIndexAccessorSeen latch itself — already landed on main separately, so they aren't repeated here. Together these four are the remainder of #39.

The four merge onto main in sequence with no conflicts; the union passes TestScripts, pkg/vm, pkg/compiler, and go test -race ./pkg/vm/.

🤖 Generated with Claude Code

[mparrett]

Regression on the arithmetic workloads. Measured on a dedicated c7a.2xlarge (EPYC 9R14), go1.26.0, against the shared merge base 3167412e85c2. Targeted A/B: alternating launches, b.N pinned at 8, median of 5–8 launches per arm. Negative is faster.

benchmark	Δ
Arith	+29.3%
Add	+26.2%
MatrixMult	+3.9%
FibPlaceholderRun	−0.5%
SetIndex	−6.0%

Arith was re-run with true per-launch interleaving because the size of it was surprising: +29.3% interleaved against +29.5% grouped, with a 1.0% spread within each arm. It reproduces.

The shape is odd and might point at the cause: Arith and Add regress hard, Fib and MatrixMult do not, and SetIndex improves. If the finally-handler check moved onto a path that tight arithmetic loops take but recursion and matrix work do not, that would fit — worth a look before this lands.

Floors from null controls in the same run: two commits compiling to byte-identical binaries measure up to 3.7% apart on ./tests, and a layout control is no worse — so anything under ~4% is not attributable. These are far outside that. Full write-up and raw data: perf-session-remeasure-results.md.

[mparrett]

Correcting my verdict here: the +29.3% is real, and it is not this change.

The exposure runs the wrong way. Fib executes the OpReturn path this PR
changes about five million times per iteration and measured −0.5%.
Arith and Add, which execute roughly two returns against millions of loop
iterations, measured +29.3% and +26.2%. The benchmarks that regressed are the
ones that barely touch the changed code.

What is left is code layout — this change re-encodes 35.6% of (*VM).run's
65,397 instructions through spill-slot reassignment. Details and the missing
control in #52.

The non-allocating handler check still looks right on its merits, and it has no
demonstrated cost or benefit on this suite. I would leave this open and
unmeasured rather than close it as a regression.

[nooga]

Verified locally: builds clean, merges without conflict alongside #40-43 (checked in sequence), go vet clean, full go test ./... (all packages) green, plus the vm-diff logic checked by hand (NaN comparison semantics, %-fast-path sign/overflow guards, findPendingHandler table-order equivalence to the old findAllExceptionHandlers scan, atomic.Bool race fix). See PR discussion for details.