@scriptc/runtime 0.0.0 → 0.0.2

package/vendor/ryu/d2s_intrinsics.h

@@ -0,0 +1,357 @@
+// Copyright 2018 Ulf Adams
+//
+// The contents of this file may be used under the terms of the Apache License,
+// Version 2.0.
+//
+//    (See accompanying file LICENSE-Apache or copy at
+//     http://www.apache.org/licenses/LICENSE-2.0)
+//
+// Alternatively, the contents of this file may be used under the terms of
+// the Boost Software License, Version 1.0.
+//    (See accompanying file LICENSE-Boost or copy at
+//     https://www.boost.org/LICENSE_1_0.txt)
+//
+// Unless required by applicable law or agreed to in writing, this software
+// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.
+#ifndef RYU_D2S_INTRINSICS_H
+#define RYU_D2S_INTRINSICS_H
+
+#include <assert.h>
+#include <stdint.h>
+
+// Defines RYU_32_BIT_PLATFORM if applicable.
+#include "common.h"
+
+// ABSL avoids uint128_t on Win32 even if __SIZEOF_INT128__ is defined.
+// Let's do the same for now.
+#if defined(__SIZEOF_INT128__) && !defined(_MSC_VER) && !defined(RYU_ONLY_64_BIT_OPS)
+#define HAS_UINT128
+#elif defined(_MSC_VER) && !defined(RYU_ONLY_64_BIT_OPS) && defined(_M_X64)
+#define HAS_64_BIT_INTRINSICS
+#endif
+
+#if defined(HAS_UINT128)
+typedef __uint128_t uint128_t;
+#endif
+
+#if defined(HAS_64_BIT_INTRINSICS)
+
+#include <intrin.h>
+
+static inline uint64_t umul128(const uint64_t a, const uint64_t b, uint64_t* const productHi) {
+  return _umul128(a, b, productHi);
+}
+
+// Returns the lower 64 bits of (hi*2^64 + lo) >> dist, with 0 < dist < 64.
+static inline uint64_t shiftright128(const uint64_t lo, const uint64_t hi, const uint32_t dist) {
+  // For the __shiftright128 intrinsic, the shift value is always
+  // modulo 64.
+  // In the current implementation of the double-precision version
+  // of Ryu, the shift value is always < 64. (In the case
+  // RYU_OPTIMIZE_SIZE == 0, the shift value is in the range [49, 58].
+  // Otherwise in the range [2, 59].)
+  // However, this function is now also called by s2d, which requires supporting
+  // the larger shift range (TODO: what is the actual range?).
+  // Check this here in case a future change requires larger shift
+  // values. In this case this function needs to be adjusted.
+  assert(dist < 64);
+  return __shiftright128(lo, hi, (unsigned char) dist);
+}
+
+#else // defined(HAS_64_BIT_INTRINSICS)
+
+static inline uint64_t umul128(const uint64_t a, const uint64_t b, uint64_t* const productHi) {
+  // The casts here help MSVC to avoid calls to the __allmul library function.
+  const uint32_t aLo = (uint32_t)a;
+  const uint32_t aHi = (uint32_t)(a >> 32);
+  const uint32_t bLo = (uint32_t)b;
+  const uint32_t bHi = (uint32_t)(b >> 32);
+
+  const uint64_t b00 = (uint64_t)aLo * bLo;
+  const uint64_t b01 = (uint64_t)aLo * bHi;
+  const uint64_t b10 = (uint64_t)aHi * bLo;
+  const uint64_t b11 = (uint64_t)aHi * bHi;
+
+  const uint32_t b00Lo = (uint32_t)b00;
+  const uint32_t b00Hi = (uint32_t)(b00 >> 32);
+
+  const uint64_t mid1 = b10 + b00Hi;
+  const uint32_t mid1Lo = (uint32_t)(mid1);
+  const uint32_t mid1Hi = (uint32_t)(mid1 >> 32);
+
+  const uint64_t mid2 = b01 + mid1Lo;
+  const uint32_t mid2Lo = (uint32_t)(mid2);
+  const uint32_t mid2Hi = (uint32_t)(mid2 >> 32);
+
+  const uint64_t pHi = b11 + mid1Hi + mid2Hi;
+  const uint64_t pLo = ((uint64_t)mid2Lo << 32) | b00Lo;
+
+  *productHi = pHi;
+  return pLo;
+}
+
+static inline uint64_t shiftright128(const uint64_t lo, const uint64_t hi, const uint32_t dist) {
+  // We don't need to handle the case dist >= 64 here (see above).
+  assert(dist < 64);
+  assert(dist > 0);
+  return (hi << (64 - dist)) | (lo >> dist);
+}
+
+#endif // defined(HAS_64_BIT_INTRINSICS)
+
+#if defined(RYU_32_BIT_PLATFORM)
+
+// Returns the high 64 bits of the 128-bit product of a and b.
+static inline uint64_t umulh(const uint64_t a, const uint64_t b) {
+  // Reuse the umul128 implementation.
+  // Optimizers will likely eliminate the instructions used to compute the
+  // low part of the product.
+  uint64_t hi;
+  umul128(a, b, &hi);
+  return hi;
+}
+
+// On 32-bit platforms, compilers typically generate calls to library
+// functions for 64-bit divisions, even if the divisor is a constant.
+//
+// E.g.:
+// https://bugs.llvm.org/show_bug.cgi?id=37932
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=17958
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=37443
+//
+// The functions here perform division-by-constant using multiplications
+// in the same way as 64-bit compilers would do.
+//
+// NB:
+// The multipliers and shift values are the ones generated by clang x64
+// for expressions like x/5, x/10, etc.
+
+static inline uint64_t div5(const uint64_t x) {
+  return umulh(x, 0xCCCCCCCCCCCCCCCDu) >> 2;
+}
+
+static inline uint64_t div10(const uint64_t x) {
+  return umulh(x, 0xCCCCCCCCCCCCCCCDu) >> 3;
+}
+
+static inline uint64_t div100(const uint64_t x) {
+  return umulh(x >> 2, 0x28F5C28F5C28F5C3u) >> 2;
+}
+
+static inline uint64_t div1e8(const uint64_t x) {
+  return umulh(x, 0xABCC77118461CEFDu) >> 26;
+}
+
+static inline uint64_t div1e9(const uint64_t x) {
+  return umulh(x >> 9, 0x44B82FA09B5A53u) >> 11;
+}
+
+static inline uint32_t mod1e9(const uint64_t x) {
+  // Avoid 64-bit math as much as possible.
+  // Returning (uint32_t) (x - 1000000000 * div1e9(x)) would
+  // perform 32x64-bit multiplication and 64-bit subtraction.
+  // x and 1000000000 * div1e9(x) are guaranteed to differ by
+  // less than 10^9, so their highest 32 bits must be identical,
+  // so we can truncate both sides to uint32_t before subtracting.
+  // We can also simplify (uint32_t) (1000000000 * div1e9(x)).
+  // We can truncate before multiplying instead of after, as multiplying
+  // the highest 32 bits of div1e9(x) can't affect the lowest 32 bits.
+  return ((uint32_t) x) - 1000000000 * ((uint32_t) div1e9(x));
+}
+
+#else // defined(RYU_32_BIT_PLATFORM)
+
+static inline uint64_t div5(const uint64_t x) {
+  return x / 5;
+}
+
+static inline uint64_t div10(const uint64_t x) {
+  return x / 10;
+}
+
+static inline uint64_t div100(const uint64_t x) {
+  return x / 100;
+}
+
+static inline uint64_t div1e8(const uint64_t x) {
+  return x / 100000000;
+}
+
+static inline uint64_t div1e9(const uint64_t x) {
+  return x / 1000000000;
+}
+
+static inline uint32_t mod1e9(const uint64_t x) {
+  return (uint32_t) (x - 1000000000 * div1e9(x));
+}
+
+#endif // defined(RYU_32_BIT_PLATFORM)
+
+static inline uint32_t pow5Factor(uint64_t value) {
+  const uint64_t m_inv_5 = 14757395258967641293u; // 5 * m_inv_5 = 1 (mod 2^64)
+  const uint64_t n_div_5 = 3689348814741910323u;  // #{ n | n = 0 (mod 2^64) } = 2^64 / 5
+  uint32_t count = 0;
+  for (;;) {
+    assert(value != 0);
+    value *= m_inv_5;
+    if (value > n_div_5)
+      break;
+    ++count;
+  }
+  return count;
+}
+
+// Returns true if value is divisible by 5^p.
+static inline bool multipleOfPowerOf5(const uint64_t value, const uint32_t p) {
+  // I tried a case distinction on p, but there was no performance difference.
+  return pow5Factor(value) >= p;
+}
+
+// Returns true if value is divisible by 2^p.
+static inline bool multipleOfPowerOf2(const uint64_t value, const uint32_t p) {
+  assert(value != 0);
+  assert(p < 64);
+  // __builtin_ctzll doesn't appear to be faster here.
+  return (value & ((1ull << p) - 1)) == 0;
+}
+
+// We need a 64x128-bit multiplication and a subsequent 128-bit shift.
+// Multiplication:
+//   The 64-bit factor is variable and passed in, the 128-bit factor comes
+//   from a lookup table. We know that the 64-bit factor only has 55
+//   significant bits (i.e., the 9 topmost bits are zeros). The 128-bit
+//   factor only has 124 significant bits (i.e., the 4 topmost bits are
+//   zeros).
+// Shift:
+//   In principle, the multiplication result requires 55 + 124 = 179 bits to
+//   represent. However, we then shift this value to the right by j, which is
+//   at least j >= 115, so the result is guaranteed to fit into 179 - 115 = 64
+//   bits. This means that we only need the topmost 64 significant bits of
+//   the 64x128-bit multiplication.
+//
+// There are several ways to do this:
+// 1. Best case: the compiler exposes a 128-bit type.
+//    We perform two 64x64-bit multiplications, add the higher 64 bits of the
+//    lower result to the higher result, and shift by j - 64 bits.
+//
+//    We explicitly cast from 64-bit to 128-bit, so the compiler can tell
+//    that these are only 64-bit inputs, and can map these to the best
+//    possible sequence of assembly instructions.
+//    x64 machines happen to have matching assembly instructions for
+//    64x64-bit multiplications and 128-bit shifts.
+//
+// 2. Second best case: the compiler exposes intrinsics for the x64 assembly
+//    instructions mentioned in 1.
+//
+// 3. We only have 64x64 bit instructions that return the lower 64 bits of
+//    the result, i.e., we have to use plain C.
+//    Our inputs are less than the full width, so we have three options:
+//    a. Ignore this fact and just implement the intrinsics manually.
+//    b. Split both into 31-bit pieces, which guarantees no internal overflow,
+//       but requires extra work upfront (unless we change the lookup table).
+//    c. Split only the first factor into 31-bit pieces, which also guarantees
+//       no internal overflow, but requires extra work since the intermediate
+//       results are not perfectly aligned.
+#if defined(HAS_UINT128)
+
+// Best case: use 128-bit type.
+static inline uint64_t mulShift64(const uint64_t m, const uint64_t* const mul, const int32_t j) {
+  const uint128_t b0 = ((uint128_t) m) * mul[0];
+  const uint128_t b2 = ((uint128_t) m) * mul[1];
+  return (uint64_t) (((b0 >> 64) + b2) >> (j - 64));
+}
+
+static inline uint64_t mulShiftAll64(const uint64_t m, const uint64_t* const mul, const int32_t j,
+  uint64_t* const vp, uint64_t* const vm, const uint32_t mmShift) {
+//  m <<= 2;
+//  uint128_t b0 = ((uint128_t) m) * mul[0]; // 0
+//  uint128_t b2 = ((uint128_t) m) * mul[1]; // 64
+//
+//  uint128_t hi = (b0 >> 64) + b2;
+//  uint128_t lo = b0 & 0xffffffffffffffffull;
+//  uint128_t factor = (((uint128_t) mul[1]) << 64) + mul[0];
+//  uint128_t vpLo = lo + (factor << 1);
+//  *vp = (uint64_t) ((hi + (vpLo >> 64)) >> (j - 64));
+//  uint128_t vmLo = lo - (factor << mmShift);
+//  *vm = (uint64_t) ((hi + (vmLo >> 64) - (((uint128_t) 1ull) << 64)) >> (j - 64));
+//  return (uint64_t) (hi >> (j - 64));
+  *vp = mulShift64(4 * m + 2, mul, j);
+  *vm = mulShift64(4 * m - 1 - mmShift, mul, j);
+  return mulShift64(4 * m, mul, j);
+}
+
+#elif defined(HAS_64_BIT_INTRINSICS)
+
+static inline uint64_t mulShift64(const uint64_t m, const uint64_t* const mul, const int32_t j) {
+  // m is maximum 55 bits
+  uint64_t high1;                                   // 128
+  const uint64_t low1 = umul128(m, mul[1], &high1); // 64
+  uint64_t high0;                                   // 64
+  umul128(m, mul[0], &high0);                       // 0
+  const uint64_t sum = high0 + low1;
+  if (sum < high0) {
+    ++high1; // overflow into high1
+  }
+  return shiftright128(sum, high1, j - 64);
+}
+
+static inline uint64_t mulShiftAll64(const uint64_t m, const uint64_t* const mul, const int32_t j,
+  uint64_t* const vp, uint64_t* const vm, const uint32_t mmShift) {
+  *vp = mulShift64(4 * m + 2, mul, j);
+  *vm = mulShift64(4 * m - 1 - mmShift, mul, j);
+  return mulShift64(4 * m, mul, j);
+}
+
+#else // !defined(HAS_UINT128) && !defined(HAS_64_BIT_INTRINSICS)
+
+static inline uint64_t mulShift64(const uint64_t m, const uint64_t* const mul, const int32_t j) {
+  // m is maximum 55 bits
+  uint64_t high1;                                   // 128
+  const uint64_t low1 = umul128(m, mul[1], &high1); // 64
+  uint64_t high0;                                   // 64
+  umul128(m, mul[0], &high0);                       // 0
+  const uint64_t sum = high0 + low1;
+  if (sum < high0) {
+    ++high1; // overflow into high1
+  }
+  return shiftright128(sum, high1, j - 64);
+}
+
+// This is faster if we don't have a 64x64->128-bit multiplication.
+static inline uint64_t mulShiftAll64(uint64_t m, const uint64_t* const mul, const int32_t j,
+  uint64_t* const vp, uint64_t* const vm, const uint32_t mmShift) {
+  m <<= 1;
+  // m is maximum 55 bits
+  uint64_t tmp;
+  const uint64_t lo = umul128(m, mul[0], &tmp);
+  uint64_t hi;
+  const uint64_t mid = tmp + umul128(m, mul[1], &hi);
+  hi += mid < tmp; // overflow into hi
+
+  const uint64_t lo2 = lo + mul[0];
+  const uint64_t mid2 = mid + mul[1] + (lo2 < lo);
+  const uint64_t hi2 = hi + (mid2 < mid);
+  *vp = shiftright128(mid2, hi2, (uint32_t) (j - 64 - 1));
+
+  if (mmShift == 1) {
+    const uint64_t lo3 = lo - mul[0];
+    const uint64_t mid3 = mid - mul[1] - (lo3 > lo);
+    const uint64_t hi3 = hi - (mid3 > mid);
+    *vm = shiftright128(mid3, hi3, (uint32_t) (j - 64 - 1));
+  } else {
+    const uint64_t lo3 = lo + lo;
+    const uint64_t mid3 = mid + mid + (lo3 < lo);
+    const uint64_t hi3 = hi + hi + (mid3 < mid);
+    const uint64_t lo4 = lo3 - mul[0];
+    const uint64_t mid4 = mid3 - mul[1] - (lo4 > lo3);
+    const uint64_t hi4 = hi3 - (mid4 > mid3);
+    *vm = shiftright128(mid4, hi4, (uint32_t) (j - 64));
+  }
+
+  return shiftright128(mid, hi, (uint32_t) (j - 64 - 1));
+}
+
+#endif // HAS_64_BIT_INTRINSICS
+
+#endif // RYU_D2S_INTRINSICS_H

package/vendor/ryu/d2s_small_table.h

@@ -0,0 +1,186 @@
+// Copyright 2018 Ulf Adams
+//
+// The contents of this file may be used under the terms of the Apache License,
+// Version 2.0.
+//
+//    (See accompanying file LICENSE-Apache or copy at
+//     http://www.apache.org/licenses/LICENSE-2.0)
+//
+// Alternatively, the contents of this file may be used under the terms of
+// the Boost Software License, Version 1.0.
+//    (See accompanying file LICENSE-Boost or copy at
+//     https://www.boost.org/LICENSE_1_0.txt)
+//
+// Unless required by applicable law or agreed to in writing, this software
+// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.
+#ifndef RYU_D2S_SMALL_TABLE_H
+#define RYU_D2S_SMALL_TABLE_H
+
+#include <assert.h>
+
+// Defines HAS_UINT128 and uint128_t if applicable.
+#include "ryu/d2s_intrinsics.h"
+
+// These tables are generated by PrintDoubleLookupTable.
+#define DOUBLE_POW5_INV_BITCOUNT 125
+#define DOUBLE_POW5_BITCOUNT 125
+
+static const uint64_t DOUBLE_POW5_INV_SPLIT2[15][2] = {
+  {                    1u, 2305843009213693952u },
+  {  5955668970331000884u, 1784059615882449851u },
+  {  8982663654677661702u, 1380349269358112757u },
+  {  7286864317269821294u, 2135987035920910082u },
+  {  7005857020398200553u, 1652639921975621497u },
+  { 17965325103354776697u, 1278668206209430417u },
+  {  8928596168509315048u, 1978643211784836272u },
+  { 10075671573058298858u, 1530901034580419511u },
+  {   597001226353042382u, 1184477304306571148u },
+  {  1527430471115325346u, 1832889850782397517u },
+  { 12533209867169019542u, 1418129833677084982u },
+  {  5577825024675947042u, 2194449627517475473u },
+  { 11006974540203867551u, 1697873161311732311u },
+  { 10313493231639821582u, 1313665730009899186u },
+  { 12701016819766672773u, 2032799256770390445u }
+};
+static const uint32_t POW5_INV_OFFSETS[22] = {
+  0x54544554, 0x04055545, 0x10041000, 0x00400414, 0x40010000, 0x41155555,
+  0x00000454, 0x00010044, 0x40000000, 0x44000041, 0x50454450, 0x55550054,
+  0x51655554, 0x40004000, 0x01000001, 0x00010500, 0x51515411, 0x05555554,
+  0x50411500, 0x40040000, 0x05040110, 0x00000000
+};
+
+static const uint64_t DOUBLE_POW5_SPLIT2[13][2] = {
+  {                    0u, 1152921504606846976u },
+  {                    0u, 1490116119384765625u },
+  {  1032610780636961552u, 1925929944387235853u },
+  {  7910200175544436838u, 1244603055572228341u },
+  { 16941905809032713930u, 1608611746708759036u },
+  { 13024893955298202172u, 2079081953128979843u },
+  {  6607496772837067824u, 1343575221513417750u },
+  { 17332926989895652603u, 1736530273035216783u },
+  { 13037379183483547984u, 2244412773384604712u },
+  {  1605989338741628675u, 1450417759929778918u },
+  {  9630225068416591280u, 1874621017369538693u },
+  {   665883850346957067u, 1211445438634777304u },
+  { 14931890668723713708u, 1565756531257009982u }
+};
+static const uint32_t POW5_OFFSETS[21] = {
+  0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x40000000, 0x59695995,
+  0x55545555, 0x56555515, 0x41150504, 0x40555410, 0x44555145, 0x44504540,
+  0x45555550, 0x40004000, 0x96440440, 0x55565565, 0x54454045, 0x40154151,
+  0x55559155, 0x51405555, 0x00000105
+};
+
+#define POW5_TABLE_SIZE 26
+static const uint64_t DOUBLE_POW5_TABLE[POW5_TABLE_SIZE] = {
+1ull, 5ull, 25ull, 125ull, 625ull, 3125ull, 15625ull, 78125ull, 390625ull,
+1953125ull, 9765625ull, 48828125ull, 244140625ull, 1220703125ull, 6103515625ull,
+30517578125ull, 152587890625ull, 762939453125ull, 3814697265625ull,
+19073486328125ull, 95367431640625ull, 476837158203125ull,
+2384185791015625ull, 11920928955078125ull, 59604644775390625ull,
+298023223876953125ull //, 1490116119384765625ull
+};
+
+#if defined(HAS_UINT128)
+
+// Computes 5^i in the form required by Ryu, and stores it in the given pointer.
+static inline void double_computePow5(const uint32_t i, uint64_t* const result) {
+  const uint32_t base = i / POW5_TABLE_SIZE;
+  const uint32_t base2 = base * POW5_TABLE_SIZE;
+  const uint32_t offset = i - base2;
+  const uint64_t* const mul = DOUBLE_POW5_SPLIT2[base];
+  if (offset == 0) {
+    result[0] = mul[0];
+    result[1] = mul[1];
+    return;
+  }
+  const uint64_t m = DOUBLE_POW5_TABLE[offset];
+  const uint128_t b0 = ((uint128_t) m) * mul[0];
+  const uint128_t b2 = ((uint128_t) m) * mul[1];
+  const uint32_t delta = pow5bits(i) - pow5bits(base2);
+  const uint128_t shiftedSum = (b0 >> delta) + (b2 << (64 - delta)) + ((POW5_OFFSETS[i / 16] >> ((i % 16) << 1)) & 3);
+  result[0] = (uint64_t) shiftedSum;
+  result[1] = (uint64_t) (shiftedSum >> 64);
+}
+
+// Computes 5^-i in the form required by Ryu, and stores it in the given pointer.
+static inline void double_computeInvPow5(const uint32_t i, uint64_t* const result) {
+  const uint32_t base = (i + POW5_TABLE_SIZE - 1) / POW5_TABLE_SIZE;
+  const uint32_t base2 = base * POW5_TABLE_SIZE;
+  const uint32_t offset = base2 - i;
+  const uint64_t* const mul = DOUBLE_POW5_INV_SPLIT2[base]; // 1/5^base2
+  if (offset == 0) {
+    result[0] = mul[0];
+    result[1] = mul[1];
+    return;
+  }
+  const uint64_t m = DOUBLE_POW5_TABLE[offset]; // 5^offset
+  const uint128_t b0 = ((uint128_t) m) * (mul[0] - 1);
+  const uint128_t b2 = ((uint128_t) m) * mul[1]; // 1/5^base2 * 5^offset = 1/5^(base2-offset) = 1/5^i
+  const uint32_t delta = pow5bits(base2) - pow5bits(i);
+  assert(i / 16 < sizeof(POW5_INV_OFFSETS) / sizeof(POW5_INV_OFFSETS[0]));
+  const uint128_t shiftedSum =
+    ((b0 >> delta) + (b2 << (64 - delta))) + 1 + ((POW5_INV_OFFSETS[i / 16] >> ((i % 16) << 1)) & 3);
+  result[0] = (uint64_t) shiftedSum;
+  result[1] = (uint64_t) (shiftedSum >> 64);
+}
+
+#else // defined(HAS_UINT128)
+
+// Computes 5^i in the form required by Ryu, and stores it in the given pointer.
+static inline void double_computePow5(const uint32_t i, uint64_t* const result) {
+  const uint32_t base = i / POW5_TABLE_SIZE;
+  const uint32_t base2 = base * POW5_TABLE_SIZE;
+  const uint32_t offset = i - base2;
+  const uint64_t* const mul = DOUBLE_POW5_SPLIT2[base];
+  if (offset == 0) {
+    result[0] = mul[0];
+    result[1] = mul[1];
+    return;
+  }
+  const uint64_t m = DOUBLE_POW5_TABLE[offset];
+  uint64_t high1;
+  const uint64_t low1 = umul128(m, mul[1], &high1);
+  uint64_t high0;
+  const uint64_t low0 = umul128(m, mul[0], &high0);
+  const uint64_t sum = high0 + low1;
+  if (sum < high0) {
+    ++high1; // overflow into high1
+  }
+  // high1 | sum | low0
+  const uint32_t delta = pow5bits(i) - pow5bits(base2);
+  result[0] = shiftright128(low0, sum, delta) + ((POW5_OFFSETS[i / 16] >> ((i % 16) << 1)) & 3);
+  result[1] = shiftright128(sum, high1, delta);
+}
+
+// Computes 5^-i in the form required by Ryu, and stores it in the given pointer.
+static inline void double_computeInvPow5(const uint32_t i, uint64_t* const result) {
+  const uint32_t base = (i + POW5_TABLE_SIZE - 1) / POW5_TABLE_SIZE;
+  const uint32_t base2 = base * POW5_TABLE_SIZE;
+  const uint32_t offset = base2 - i;
+  const uint64_t* const mul = DOUBLE_POW5_INV_SPLIT2[base]; // 1/5^base2
+  if (offset == 0) {
+    result[0] = mul[0];
+    result[1] = mul[1];
+    return;
+  }
+  const uint64_t m = DOUBLE_POW5_TABLE[offset];
+  uint64_t high1;
+  const uint64_t low1 = umul128(m, mul[1], &high1);
+  uint64_t high0;
+  const uint64_t low0 = umul128(m, mul[0] - 1, &high0);
+  const uint64_t sum = high0 + low1;
+  if (sum < high0) {
+    ++high1; // overflow into high1
+  }
+  // high1 | sum | low0
+  const uint32_t delta = pow5bits(base2) - pow5bits(i);
+  assert(i / 16 < sizeof(POW5_INV_OFFSETS) / sizeof(POW5_INV_OFFSETS[0]));
+  result[0] = shiftright128(low0, sum, delta) + 1 + ((POW5_INV_OFFSETS[i / 16] >> ((i % 16) << 1)) & 3);
+  result[1] = shiftright128(sum, high1, delta);
+}
+
+#endif // defined(HAS_UINT128)
+
+#endif // RYU_D2S_SMALL_TABLE_H

package/vendor/ryu/digit_table.h

@@ -0,0 +1,35 @@
+// Copyright 2018 Ulf Adams
+//
+// The contents of this file may be used under the terms of the Apache License,
+// Version 2.0.
+//
+//    (See accompanying file LICENSE-Apache or copy at
+//     http://www.apache.org/licenses/LICENSE-2.0)
+//
+// Alternatively, the contents of this file may be used under the terms of
+// the Boost Software License, Version 1.0.
+//    (See accompanying file LICENSE-Boost or copy at
+//     https://www.boost.org/LICENSE_1_0.txt)
+//
+// Unless required by applicable law or agreed to in writing, this software
+// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.
+#ifndef RYU_DIGIT_TABLE_H
+#define RYU_DIGIT_TABLE_H
+
+// A table of all two-digit numbers. This is used to speed up decimal digit
+// generation by copying pairs of digits into the final output.
+static const char DIGIT_TABLE[200] = {
+  '0','0','0','1','0','2','0','3','0','4','0','5','0','6','0','7','0','8','0','9',
+  '1','0','1','1','1','2','1','3','1','4','1','5','1','6','1','7','1','8','1','9',
+  '2','0','2','1','2','2','2','3','2','4','2','5','2','6','2','7','2','8','2','9',
+  '3','0','3','1','3','2','3','3','3','4','3','5','3','6','3','7','3','8','3','9',
+  '4','0','4','1','4','2','4','3','4','4','4','5','4','6','4','7','4','8','4','9',
+  '5','0','5','1','5','2','5','3','5','4','5','5','5','6','5','7','5','8','5','9',
+  '6','0','6','1','6','2','6','3','6','4','6','5','6','6','6','7','6','8','6','9',
+  '7','0','7','1','7','2','7','3','7','4','7','5','7','6','7','7','7','8','7','9',
+  '8','0','8','1','8','2','8','3','8','4','8','5','8','6','8','7','8','8','8','9',
+  '9','0','9','1','9','2','9','3','9','4','9','5','9','6','9','7','9','8','9','9'
+};
+
+#endif // RYU_DIGIT_TABLE_H

package/vendor/ryu/ryu.h

@@ -0,0 +1,46 @@
+// Copyright 2018 Ulf Adams
+//
+// The contents of this file may be used under the terms of the Apache License,
+// Version 2.0.
+//
+//    (See accompanying file LICENSE-Apache or copy at
+//     http://www.apache.org/licenses/LICENSE-2.0)
+//
+// Alternatively, the contents of this file may be used under the terms of
+// the Boost Software License, Version 1.0.
+//    (See accompanying file LICENSE-Boost or copy at
+//     https://www.boost.org/LICENSE_1_0.txt)
+//
+// Unless required by applicable law or agreed to in writing, this software
+// is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.
+#ifndef RYU_H
+#define RYU_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <inttypes.h>
+
+int d2s_buffered_n(double f, char* result);
+void d2s_buffered(double f, char* result);
+char* d2s(double f);
+
+int f2s_buffered_n(float f, char* result);
+void f2s_buffered(float f, char* result);
+char* f2s(float f);
+
+int d2fixed_buffered_n(double d, uint32_t precision, char* result);
+void d2fixed_buffered(double d, uint32_t precision, char* result);
+char* d2fixed(double d, uint32_t precision);
+
+int d2exp_buffered_n(double d, uint32_t precision, char* result);
+void d2exp_buffered(double d, uint32_t precision, char* result);
+char* d2exp(double d, uint32_t precision);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // RYU_H

package/vendor/zlib/LICENSE

@@ -0,0 +1,22 @@
+Copyright notice:
+
+ (C) 1995-2022 Jean-loup Gailly and Mark Adler
+
+  This software is provided 'as-is', without any express or implied
+  warranty.  In no event will the authors be held liable for any damages
+  arising from the use of this software.
+
+  Permission is granted to anyone to use this software for any purpose,
+  including commercial applications, and to alter it and redistribute it
+  freely, subject to the following restrictions:
+
+  1. The origin of this software must not be misrepresented; you must not
+     claim that you wrote the original software. If you use this software
+     in a product, an acknowledgment in the product documentation would be
+     appreciated but is not required.
+  2. Altered source versions must be plainly marked as such, and must not be
+     misrepresented as being the original software.
+  3. This notice may not be removed or altered from any source distribution.
+
+  Jean-loup Gailly        Mark Adler
+  jloup@gzip.org          madler@alumni.caltech.edu