chore(deps): update decimal-scaled requirement from 0.3.3 to 0.4.4 in /bench-history

[dependabot]

Updates the requirements on decimal-scaled to permit the latest version.

Release notes

Sourced from decimal-scaled's releases.

decimal-scaled v0.4.4 — fully correctly-rounded transcendentals

A correctness release. Every *_strict transcendental is now provably correctly rounded — within 0.5 ULP, i.e. 0 LSB of error at the storage scale — under all six rounding modes, at all thirteen widths, across the full 22-function surface. 0.4.3 secured the primary functions under nearest rounding; 0.4.4 extends the guarantee to the directed rounding modes and the derived functions, with no exceptions and no ignored cells.

Fixed

• Directed rounding (Trunc / Floor / Ceiling / HalfTowardZero / HalfAwayFromZero) for the wide-tier transcendentals: the working-scale approximation could previously round one LSB the wrong way near a storage grid line. Resolved with residual-sign Ziv escalation; the nearest-mode fast path is unchanged.
• Derived transcendentals log / log2 / log10 / exp2 / sinh / cosh / tanh / asinh / acosh / atanh / powf are now correctly rounded under every mode (they previously applied their final rounding without the directed-rounding escalation, and several carried cancellation or overflow-edge error at extreme inputs).
• acosh near 1 and atanh near ±1: removed catastrophic cancellation by reformulating through log1p, computing 1∓x and v²−1 as exact working-scale gaps.
• sinh / cosh / tanh at large arguments: corrected a sign-direction error that turned the relative error of e^−|x| into a large absolute error in its reciprocal; the dominant term is now always evaluated at |x|. Wide tiers near the storage-overflow edge now compute in a wider working integer.
• log_b(b^k) and exact-power powf (e.g. powf(4, 0.5)) return the exact result under directed rounding.
• tanh tiny arguments: directed rounding of the compressing linear band.
• D38 log2 / log10 at maximum scale: removed an i128 overflow in the exact-power detection.

Changed

• The correct-rounding guarantee — previously documented for the primary functions under nearest rounding — now holds for the entire transcendental surface under all six rounding modes and all widths.

Internal

• Added correctly-rounded integer log1p / expm1 kernels (internal; a public API is planned).
• Expanded the strict-golden suite to the full 22-function × 13-width × 6-mode matrix, asserting delta == 0 against an external mpmath oracle (286 cells, none ignored).

Changelog

Sourced from decimal-scaled's changelog.

[0.4.4] — 2026-05-21

A correctness release. Every *_strict transcendental is now provably correctly rounded — within 0.5 ULP, i.e. 0 LSB of error at the storage scale — under all six rounding modes, at all thirteen widths, across the full 22-function surface. 0.4.3 secured the primary functions under nearest rounding; 0.4.4 extends the guarantee to the directed rounding modes and the derived functions, with no exceptions and no ignored cells.

Fixed

• Directed rounding (Trunc / Floor / Ceiling / HalfTowardZero / HalfAwayFromZero) for the wide-tier transcendentals: the working-scale approximation could previously round one LSB the wrong way near a storage grid line. Resolved with residual-sign Ziv escalation; the nearest-mode fast path is unchanged.
• Derived transcendentals log / log2 / log10 / exp2 / sinh / cosh / tanh / asinh / acosh / atanh / powf are now correctly rounded under every mode (they previously applied their final rounding without the directed-rounding escalation, and several carried cancellation or overflow-edge error at extreme inputs).
• acosh near 1 and atanh near ±1: removed catastrophic cancellation by reformulating through log1p, computing 1∓x and v²−1 as exact working-scale gaps.
• sinh / cosh / tanh at large arguments: corrected a sign-direction error that turned the relative error of e^−|x| into a large absolute error in its reciprocal; the dominant term is now always evaluated at |x|. Wide tiers near the storage-overflow edge now compute in a wider working integer.
• log_b(b^k) and exact-power powf (e.g. powf(4, 0.5)) return the exact result under directed rounding.
• tanh tiny arguments: directed rounding of the compressing linear band.
• D38 log2 / log10 at maximum scale: removed an i128 overflow in the exact-power detection.

Changed

• The correct-rounding guarantee — previously documented for the primary functions under nearest rounding — now holds for the entire transcendental surface under all six rounding modes and all widths.

Internal

• Added correctly-rounded integer log1p / expm1 kernels (internal; a public API is planned).
• Expanded the strict-golden suite to the full 22-function × 13-width × 6-mode matrix, asserting delta == 0 against an external mpmath oracle (286 cells, none ignored).

... (truncated)

Commits

• 0710fcf docs(figures): regenerate cross-version history charts (add v0.4.3, HEAD=v0.4.4)
• 42c3bbd bench(history): add v0.4.3 to the cross-version matrix; label HEAD as v0.4.4
• 98bdf16 ci(scorecard): pass SCORECARD_TOKEN PAT so Branch-Protection can read protect...
• 7d57048 Release 0.4.4 — fully correctly-rounded transcendentals (#12)
• a1afe91 release: 0.4.3 — precision-hole closure, D115/D1232 overflow fixes, width-ada...
• 2251f5e docs(benchmarks): refresh tables and figures from latest sweep
• 1c3da5b merge: width-adaptive mg_divide buffer — restores wide mul perf (D76/D307 bac...
• 885ffbc merge: fix D115 transcendental work-int Int1024->Int2048 (high-scale exp/ln o...
• 6e2a5d3 perf(mg_divide): size the magnitude buffer to the work-integer width
• 578da4b fix: widen D115 strict transcendental work integer to Int2048
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[dependabot]

Labels

The following labels could not be found: benchmarks, dependabot, dependencies. Please create them before Dependabot can add them to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.

[codspeed-hq]

Merging this PR will not alter performance

⚡ 5 improved benchmarks
❌ 4 regressed benchmarks
✅ 201 untouched benchmarks
⏩ 9 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	next_power_of_two	62.2 ns	91.4 ns	-31.91%
⚡	ceil	318.1 ns	288.9 ns	+10.1%
⚡	clamp	156.7 ns	127.5 ns	+22.88%
⚡	to_num_f64	220.3 ns	191.1 ns	+15.26%
⚡	to_f64	220.3 ns	191.1 ns	+15.26%
⚡	TryFrom_u128	125.6 ns	96.4 ns	+30.26%
❌	A_naive[bound_1e18]	382.2 ns	440.6 ns	-13.24%
❌	A_naive[mid_1e15]	380.6 ns	438.9 ns	-13.29%
❌	A_naive[small_1e9]	380.6 ns	438.9 ns	-13.29%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing dependabot/cargo/bench-history/decimal-scaled-0.4.4 (cc399c4) with main (a9c0e9b)}

Footnotes

1. 9 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[jackmoxley]

@dependabot ignore this dependency

[dependabot]

OK, I won't notify you about decimal-scaled again, unless you re-open this PR.