decimal-scaled 0.5.0

Const-generic base-10 fixed-point decimals (D18/D38/D76/D153/D307 and the half-width tiers up to D1232) with integer-only transcendentals correctly rounded to within 0.5 ULP — exact at the type's last representable place. Deterministic across every platform; no_std-friendly.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
// SPDX-FileCopyrightText: 2026 John Moxley
// SPDX-License-Identifier: MIT OR Apache-2.0

//! Exact integer-power pin for `powf` — one generic kernel for every width.
//!
//! When the exponent is an exact integer `n` and the base is an exact
//! positive integer `b`, `b^n` is an **exact rational** — `b^n` for `n >= 0`,
//! `1 / b^|n|` for `n < 0`. Its correctly-rounded storage value is fixed by
//! integer arithmetic alone, so emitting it directly stops the
//! `exp(n · ln b)` composition's to-nearest round-off from landing one LSB
//! off under a directed mode (`Floor` / `Trunc` / `Ceiling`) at these
//! algebraic-exact points (e.g. `10^-2 = 0.01`, `25^-3 = 0.000064`). The
//! composition is correct to nearest, but a deferred reciprocal whose scaled
//! integer power overflows storage (`powf_series_2limb::powi_raw_checked`
//! returns `None`) was being computed through it, ~1 ULP low — so `Floor` /
//! `Trunc` rounded to `0.00999…9` instead of the exact `0.01`.
//!
//! This is the power analogue of `exp::exp_series_2limb::exp2_exact_pin` +
//! `round_pow2_fraction`: a single `St: BigInt`-generic pin called by both
//! the narrow `Fixed` kernel (`powf_series_2limb`, `St = Int<2>`) and the
//! wide schoolbook kernel (`pow_schoolbook`, `St = C::Storage`).
//!
//! The pin DEFERS to the composition ([`ExactPin::Defer`]) ONLY when it
//! cannot prove anything: a non-integer exponent, a non-integer base, or
//! `|n|` above the fast-path threshold. A positive integer power that
//! overflows storage is a PROOF of out-of-range ([`ExactPin::OutOfRange`]):
//! the seamed narrow kernel propagates it as its `None` (the policy dispatch
//! wrapper applies the overflow contract's panic; the `checked_` surface
//! propagates), and the default [`powi_exact_pin`] wrapper panics directly
//! for the wide shells. Deferring instead would be wrong — the composition's
//! to-nearest approximation could be directed-rounded (Floor / Trunc) back
//! inside the range at an out-by-one boundary.

use crate::int::types::traits::BigInt;
use crate::support::rounding::{should_bump, RoundingMode};

/// Verdict of an exact-power pin: the pin either DECIDES the cell — an
/// exact value, or proof the exact result exceeds the decimal range —
/// or declines and defers to the `exp(n · ln b)` composition.
pub(crate) enum ExactPin<V> {
    /// The correctly-rounded exact storage value.
    Value(V),
    /// Exact integer arithmetic proved the result exceeds the decimal
    /// range. The kernel seam returns `None` for it (the default policy
    /// dispatch wrapper applies the overflow contract's panic); the
    /// unseamed default wrappers panic directly.
    OutOfRange,
    /// The pin does not apply — defer to the composition.
    Defer,
}

/// Default (panicking) form of [`powi_exact_pin_checked`] for the unseamed
/// wide shells: `Value` → `Some`, `Defer` → `None`, and a proven
/// out-of-range applies the overflow contract's panic directly.
#[inline]
pub(crate) fn powi_exact_pin<St: BigInt, const SCALE: u32>(
    base: St,
    exp: St,
    storage_max: St,
    mode: RoundingMode,
) -> Option<St> {
    match powi_exact_pin_checked::<St, SCALE>(base, exp, storage_max, mode) {
        ExactPin::Value(v) => Some(v),
        ExactPin::OutOfRange => {
            crate::support::diagnostics::overflow_panic_with_scale("powf kernel", SCALE)
        }
        ExactPin::Defer => None,
    }
}

/// Correctly-rounded storage value of `b^n` at scale `SCALE`, when `b` is an
/// exact positive integer and `n` an exact integer with
/// `|n| <= INT_FAST_PATH_THRESHOLD`. `base` / `exp` are raw storage
/// (`value · 10^SCALE`); `storage_max` is the tier's representable maximum
/// (`Int<N>::MAX` for the narrow path, `C::storage_max()` for the wide path),
/// used only to reject a positive power that has left the decimal range.
///
/// [`ExactPin::Defer`] signals "this pin does not apply — defer to the
/// `exp(n · ln b)` composition": a fractional base or exponent (the genuinely
/// transcendental case) or `|n|` past the threshold. A positive power out of
/// range is [`ExactPin::OutOfRange`] (the overflow is proof) — detected once
/// here; each wrapper applies its policy (seam `None` / contract panic). For
/// `n < 0` the result is always in `(0, 1]` and so always representable.
#[inline]
pub(crate) fn powi_exact_pin_checked<St: BigInt, const SCALE: u32>(
    base: St,
    exp: St,
    storage_max: St,
    mode: RoundingMode,
) -> ExactPin<St> {
    // `10^SCALE` — the raw value `1.0`, sourced from the baked table so it is
    // exact at every tier (`St::TEN.pow` would wrap; `pow10::dispatch` does
    // not).
    let one_s = crate::consts::pow10::dispatch::<St>(SCALE);

    // The exponent must be an exact integer `n` (`exp` an exact multiple of
    // `10^SCALE`) with `|n|` inside the integer fast-path threshold.
    let (n_big, e_rem) = exp.div_rem(one_s);
    if e_rem != St::ZERO {
        return ExactPin::Defer;
    }
    let thresh =
        St::from_i128(crate::algos::pow::powf_series_2limb::INT_FAST_PATH_THRESHOLD as i128);
    if n_big > thresh || n_big < St::ZERO - thresh {
        return ExactPin::Defer;
    }
    let n = n_big.to_i128() as i32;

    // The base must be an exact positive integer `b >= 1`.
    if base <= St::ZERO {
        return ExactPin::Defer;
    }
    let (bv, b_rem) = base.div_rem(one_s);
    if b_rem != St::ZERO {
        return ExactPin::Defer; // fractional base — defer to the composition
    }

    // `b^0 = 1` and `1^n = 1` are exactly `1.0` for every mode.
    if n == 0 || bv == St::ONE {
        return ExactPin::Value(one_s);
    }
    let k = n.unsigned_abs();

    if n > 0 {
        // `b^n · 10^SCALE` — an exact integer when it fits the decimal range.
        // A `checked_pow` / `checked_mul` overflow, or a value past
        // `storage_max`, is PROOF the exact `b^n` exceeds the decimal range
        // (integer `b >= 2`, `n > 0`: the power is monotone): report
        // `OutOfRange` per the overflow contract (the default wrappers panic
        // in debug AND release; the `checked_` seam propagates `None`) rather
        // than deferring to the `exp(n·ln b)` composition, whose to-nearest
        // approximation can directed-round (Floor / Trunc) back INSIDE the
        // range at an out-by-one boundary (the `exp2(127)` hair case:
        // `2^127` at scale 0 is `i128::MAX + 1`).
        match bv
            .checked_pow(k)
            .and_then(|p| p.checked_mul(one_s))
            .filter(|v| *v <= storage_max)
        {
            Some(v) => ExactPin::Value(v),
            None => ExactPin::OutOfRange,
        }
    } else {
        // `b^-k = 1 / b^k`, stored as `round(10^SCALE / b^k)` — a strictly
        // positive value in `(0, 1]`.
        match bv.checked_pow(k) {
            Some(d) => {
                // `d = b^k` fits storage; round `10^SCALE / d` under `mode`.
                let (q, r) = one_s.div_rem(d);
                if r == St::ZERO {
                    return ExactPin::Value(q); // exact — no rounding
                }
                // Half-comparison `2r vs d`, formed as `r vs d − r` to avoid
                // overflowing `2r` (`r < d <= MAX`). `q` is the truncated
                // quotient; the result is positive.
                let cmp = r.cmp(&(d - r));
                let q_is_odd = q.bit(0);
                let bump = should_bump(mode, cmp, q_is_odd, true);
                ExactPin::Value(if bump { q + St::ONE } else { q })
            }
            None => {
                // `b^k` overflowed storage ⟹ `b^k > MAX >= 2·10^SCALE`
                // (every tier keeps at least one integer digit, so
                // `10^SCALE <= MAX / 2`), hence `0 < 1/b^k < ½` LSB:
                // a sub-resolution positive, strictly below the half
                // boundary. Only `Ceiling` rounds it up to one LSB.
                let bump = should_bump(mode, core::cmp::Ordering::Less, false, true);
                ExactPin::Value(if bump { St::ONE } else { St::ZERO })
            }
        }
    }
}

/// `Some(n)` when `exp` (raw storage at `SCALE`) is an exact integer with
/// `|n|` inside the integer fast-path threshold — the shared exponent gate of
/// [`powi_exact_pin`] and the fractional-base chain.
pub(crate) fn exp_as_small_int_raw<St: BigInt, const SCALE: u32>(exp: St) -> Option<i32> {
    let one_s = crate::consts::pow10::dispatch::<St>(SCALE);
    let (n_big, e_rem) = exp.div_rem(one_s);
    if e_rem != St::ZERO {
        return None;
    }
    let thresh =
        St::from_i128(crate::algos::pow::powf_series_2limb::INT_FAST_PATH_THRESHOLD as i128);
    if n_big > thresh || n_big < St::ZERO - thresh {
        return None;
    }
    Some(n_big.to_i128() as i32)
}

/// `10^d` in `St`, or `None` past the width (checked ×10 loop — this is a
/// cold pin path; `d` is bounded by `SCALE·(k+1)` in practice).
fn checked_pow10<St: BigInt>(d: u32) -> Option<St> {
    let ten = St::from_i128(10);
    let mut v = St::ONE;
    for _ in 0..d {
        v = v.checked_mul(ten)?;
    }
    Some(v)
}

/// `floor(10^e / d)` with remainder, by STREAMING long division — the
/// power-of-ten dividend is never materialised, so the reciprocal pin works
/// at MAX_SCALE where `10^(SCALE+f·k)` itself exceeds the width.
enum Pow10Div<St> {
    /// The quotient and remainder; quotient verified `<= storage_max`.
    Q(St, St),
    /// The quotient exceeds the decimal range — PROOF of overflow (the
    /// quotient only grows as digits stream in).
    OutOfRange,
    /// `d` is too close to the width for the digit step (`r·10` overflowed) —
    /// not a proof of anything; the caller defers.
    Wide,
}

fn div_pow10_small<St: BigInt>(e: u32, d: St, storage_max: St) -> Pow10Div<St> {
    let ten = St::from_i128(10);
    let (mut q, mut r) = St::ONE.div_rem(d);
    for _ in 0..e {
        let r10 = match r.checked_mul(ten) {
            Some(v) => v,
            None => return Pow10Div::Wide,
        };
        let (digit, rr) = r10.div_rem(d); // digit <= 9: r < d ⇒ r·10 < 10·d
        q = match q.checked_mul(ten).and_then(|v| v.checked_add(digit)) {
            Some(v) => v,
            None => return Pow10Div::OutOfRange,
        };
        if q > storage_max {
            return Pow10Div::OutOfRange;
        }
        r = rr;
    }
    Pow10Div::Q(q, r)
}

/// Correctly-rounded `base^n` for a TERMINATING-DECIMAL base — the
/// fractional-base sibling of [`powi_exact_pin`], reached when that pin
/// declines (non-integer base). The base is REDUCED to its significant
/// digits first: `base = m / 10^f` with the trailing zeros of the raw
/// stripped, so `m` is small for real literals (`2.5 -> m = 25, f = 1`) and
/// `m^k` never touches a `2·SCALE` product. Then `base^n = m^k / 10^(f·k)`
/// (or its reciprocal) is placed on the `SCALE` grid by EXACT integer
/// arithmetic — a shift when it terminates, a single half-compared rounding
/// otherwise (ties included), at every scale.
///
/// `None` (defer to the guarded composition) only when `m^k` or a needed
/// power of ten exceeds the width — never a proof of anything. A POSITIVE
/// exact result beyond the decimal range panics per the overflow contract
/// (the proof is in hand), mirroring [`powi_exact_pin`].
pub(crate) fn powi_terminating_pin<St: BigInt, const SCALE: u32>(
    base: St,
    n: i32,
    storage_max: St,
    mode: RoundingMode,
) -> Option<St> {
    debug_assert!(n != 0);
    if base <= St::ZERO {
        return None;
    }
    // Reduce: base raw = m · 10^z, m not divisible by 10; f = significant
    // fraction length of the VALUE (f <= SCALE; f == 0 means integer base,
    // which powi_exact_pin already owns but is handled here for totality).
    let ten = St::from_i128(10);
    let mut m = base;
    let mut z = 0u32;
    loop {
        let (q, r) = m.div_rem(ten);
        if r != St::ZERO || z >= SCALE {
            break;
        }
        m = q;
        z += 1;
    }
    let f = SCALE - z;
    let k = n.unsigned_abs();
    let mk = m.checked_pow(k)?; // base's significant digits ^ k — small for real literals
    let fk = f.checked_mul(k)?;

    // Round the exact rational `num_pow10 / den` (or `mk · 10^shift`) onto the
    // SCALE grid; `positive` results only (base > 0).
    let place = |q: St, r: St, d: St| -> St {
        if r == St::ZERO {
            return q;
        }
        let cmp = r.cmp(&(d - r));
        let bump = should_bump(mode, cmp, q.bit(0), true);
        if bump {
            q + St::ONE
        } else {
            q
        }
    };

    let v = if n > 0 {
        // base^k = mk / 10^fk; at SCALE the raw is mk · 10^(SCALE - fk) when
        // that shift is non-negative (exact), else a single rounded division.
        if fk <= SCALE {
            mk.checked_mul(checked_pow10::<St>(SCALE - fk)?)?
        } else {
            match checked_pow10::<St>(fk - SCALE) {
                Some(d) => {
                    let (q, r) = mk.div_rem(d);
                    place(q, r, d)
                }
                None => {
                    // The divisor exceeds the width while mk fits: the result
                    // is at least one order below ½ LSB — a sub-resolution
                    // positive; only Ceiling rounds it up.
                    let bump = should_bump(mode, core::cmp::Ordering::Less, false, true);
                    if bump {
                        St::ONE
                    } else {
                        St::ZERO
                    }
                }
            }
        }
    } else {
        // base^-k = 10^fk / mk; the raw at SCALE is 10^(SCALE + fk) / mk,
        // single-rounded — exact directed/tie handling for terminating AND
        // non-terminating reciprocals alike (1.5^-1 = 0.666...). When the
        // power-of-ten numerator exceeds the width (MAX_SCALE cells), the
        // division streams digit-by-digit instead.
        let e = SCALE.checked_add(fk)?;
        match checked_pow10::<St>(e) {
            Some(num) => {
                let (q, r) = num.div_rem(mk);
                place(q, r, mk)
            }
            None => match div_pow10_small::<St>(e, mk, storage_max) {
                Pow10Div::Q(q, r) => place(q, r, mk),
                Pow10Div::OutOfRange => crate::support::diagnostics::overflow_panic_with_scale(
                    "powf kernel",
                    SCALE,
                ),
                Pow10Div::Wide => return None,
            },
        }
    };
    if v > storage_max {
        // Exact arithmetic put the correctly-rounded result beyond the
        // decimal range — proof of overflow; the contract panics.
        crate::support::diagnostics::overflow_panic_with_scale("powf kernel", SCALE)
    }
    Some(v)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::int::types::Int;

    #[test]
    fn terminating_pin_exact_ties_and_reciprocals() {
        let pin = |base: i128, n: i32, mode: RoundingMode| {
            powi_terminating_pin::<Int<2>, 2>(Int::<2>::from_i128(base), n, Int::<2>::MAX, mode)
                .map(|v| v.as_i128())
        };
        for mode in MODES {
            // 2.5^2 = 6.25 at scale 2 — exact at every mode.
            assert_eq!(pin(250, 2, mode), Some(625), "{mode:?} 2.5^2");
            // 0.5^-2 = 4 — exact reciprocal.
            assert_eq!(pin(50, -2, mode), Some(400), "{mode:?} 0.5^-2");
        }
        // 1.5^3 = 3.375 at scale 2 — an exact TIE the single rounding decides.
        assert_eq!(pin(150, 3, RoundingMode::HalfToEven), Some(338)); // 337 odd -> up
        assert_eq!(pin(150, 3, RoundingMode::HalfTowardZero), Some(337));
        assert_eq!(pin(150, 3, RoundingMode::Trunc), Some(337));
        assert_eq!(pin(150, 3, RoundingMode::Ceiling), Some(338));
        // 0.5^-2 = 4 at MAX-scale-like cells: 10^(SCALE+2) exceeds i128, so
        // the reciprocal streams its long division (the gate4 residue).
        let pin37 = |base: i128, n: i32, mode: RoundingMode| {
            powi_terminating_pin::<Int<2>, 37>(Int::<2>::from_i128(base), n, Int::<2>::MAX, mode)
                .map(|v| v.as_i128())
        };
        let half_raw = 5 * 10_i128.pow(36);
        for mode in MODES {
            assert_eq!(pin37(half_raw, -2, mode), Some(4 * 10_i128.pow(37)), "{mode:?} 0.5^-2 @37");
        }
        // 1.5^-1 = 0.666... — non-terminating reciprocal, residual above half.
        assert_eq!(pin(150, -1, RoundingMode::Floor), Some(66));
        assert_eq!(pin(150, -1, RoundingMode::HalfToEven), Some(67));
        assert_eq!(pin(150, -1, RoundingMode::Ceiling), Some(67));
    }

    // D38<S>: storage MAX ≈ 1.7·10^38 (full i128). Each base below is
    // representable at its chosen scale (`base · 10^S <= MAX`), but its
    // scaled power `base^|n| · 10^S` overflows i128 — exactly the case the
    // narrow integer fast path defers on.
    const MODES: [RoundingMode; 6] = [
        RoundingMode::HalfToEven,
        RoundingMode::HalfAwayFromZero,
        RoundingMode::HalfTowardZero,
        RoundingMode::Trunc,
        RoundingMode::Floor,
        RoundingMode::Ceiling,
    ];

    /// Pin called on already-scaled raw storage at scale `SC`.
    fn pin_raw<const SC: u32>(base_raw: i128, exp_raw: i128, mode: RoundingMode) -> Option<i128> {
        powi_exact_pin::<Int<2>, SC>(
            Int::<2>::from_i128(base_raw),
            Int::<2>::from_i128(exp_raw),
            Int::<2>::MAX,
            mode,
        )
        .map(|v| v.as_i128())
    }

    /// Pin on integer `base`/`exp` values, scaled to `SC` internally.
    fn pin<const SC: u32>(base: i128, exp: i128, mode: RoundingMode) -> Option<i128> {
        let one = 10_i128.pow(SC);
        pin_raw::<SC>(base * one, exp * one, mode)
    }

    #[track_caller]
    fn check_exact<const SC: u32>(base: i128, exp: i128, divisor: i128) {
        let one = 10_i128.pow(SC);
        for mode in MODES {
            assert_eq!(
                pin::<SC>(base, exp, mode),
                Some(one / divisor),
                "base={base} exp={exp} scale={SC} mode={mode:?}"
            );
        }
    }

    #[test]
    fn exact_reciprocals_are_mode_independent() {
        // Exact powers: the rational lands exactly on a grid line, so every
        // mode returns the same value (no round-off to misdirect). Bases ≤ 17
        // at scale 37; the larger bases at scale 36 (where they remain ≤ MAX).
        check_exact::<37>(10, -2, 100); // 0.01
        check_exact::<37>(16, -2, 256); // 0.00390625
        check_exact::<37>(4, -3, 64); // 0.015625
        check_exact::<37>(5, -3, 125); // 0.008
        check_exact::<36>(20, -2, 400); // 0.0025
        check_exact::<36>(25, -2, 625); // 0.0016
        check_exact::<36>(25, -3, 15_625); // 0.000064
    }

    #[test]
    fn inexact_reciprocal_rounds_each_direction() {
        // 1/3 = 0.333…3̅ at scale 37: q = floor(10^37 / 3), remainder 1 (< half),
        // so the directed/nearest split is the LSB.
        let q = 10_i128.pow(37) / 3; // 333…3 (37 threes)
        // remainder 10^37 mod 3 == 1, strictly below half → round down except Ceiling.
        assert_eq!(pin::<37>(3, -1, RoundingMode::Floor), Some(q));
        assert_eq!(pin::<37>(3, -1, RoundingMode::Trunc), Some(q));
        assert_eq!(pin::<37>(3, -1, RoundingMode::HalfToEven), Some(q));
        assert_eq!(pin::<37>(3, -1, RoundingMode::Ceiling), Some(q + 1));
    }

    #[test]
    fn positive_powers_are_exact_integers() {
        let one = 10_i128.pow(37);
        for mode in MODES {
            assert_eq!(pin::<37>(2, 3, mode), Some(8 * one));
            assert_eq!(pin::<37>(17, 1, mode), Some(17 * one)); // in-range edge
        }
    }

    #[test]
    #[should_panic(expected = "powf kernel")]
    fn positive_overflow_panics() {
        // 10² = 100 exceeds the D38<37> decimal range; the exact arithmetic is
        // PROOF of the overflow, so the pin panics per the contract (deferring
        // let the directed-down composition round the approximation back into
        // range — the exp2(127) hair case).
        let _ = pin::<37>(10, 2, RoundingMode::HalfToEven);
    }

    #[test]
    fn non_integer_base_or_exp_defers() {
        // 2.5 (fractional base) and 0.5 (fractional exponent) are not this
        // pin's job — defer to the composition.
        let one = 10_i128.pow(37);
        assert_eq!(pin_raw::<37>(one * 5 / 2, -2 * one, RoundingMode::Floor), None);
        assert_eq!(pin_raw::<37>(2 * one, one / 2, RoundingMode::Floor), None);
    }
}