Austin Schuh | bb1338c | 2024-06-15 19:31:16 -0700 | [diff] [blame] | 1 | /* double mpq_get_d (mpq_t src) -- mpq to double, rounding towards zero. |
| 2 | |
| 3 | Copyright 1995, 1996, 2001-2005, 2018, 2019 Free Software Foundation, Inc. |
| 4 | |
| 5 | This file is part of the GNU MP Library. |
| 6 | |
| 7 | The GNU MP Library is free software; you can redistribute it and/or modify |
| 8 | it under the terms of either: |
| 9 | |
| 10 | * the GNU Lesser General Public License as published by the Free |
| 11 | Software Foundation; either version 3 of the License, or (at your |
| 12 | option) any later version. |
| 13 | |
| 14 | or |
| 15 | |
| 16 | * the GNU General Public License as published by the Free Software |
| 17 | Foundation; either version 2 of the License, or (at your option) any |
| 18 | later version. |
| 19 | |
| 20 | or both in parallel, as here. |
| 21 | |
| 22 | The GNU MP Library is distributed in the hope that it will be useful, but |
| 23 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 24 | or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 25 | for more details. |
| 26 | |
| 27 | You should have received copies of the GNU General Public License and the |
| 28 | GNU Lesser General Public License along with the GNU MP Library. If not, |
| 29 | see https://www.gnu.org/licenses/. */ |
| 30 | |
| 31 | #include <stdio.h> /* for NULL */ |
| 32 | #include "gmp-impl.h" |
| 33 | #include "longlong.h" |
| 34 | |
| 35 | |
| 36 | /* All that's needed is to get the high 53 bits of the quotient num/den, |
| 37 | rounded towards zero. More than 53 bits is fine, any excess is ignored |
| 38 | by mpn_get_d. |
| 39 | |
| 40 | N_QLIMBS is how many quotient limbs we need to satisfy the mantissa of a |
| 41 | double, assuming the highest of those limbs is non-zero. The target |
| 42 | qsize for mpn_tdiv_qr is then 1 more than this, since that function may |
| 43 | give a zero in the high limb (and non-zero in the second highest). |
| 44 | |
| 45 | The use of 8*sizeof(double) in N_QLIMBS is an overestimate of the |
| 46 | mantissa bits, but it gets the same result as the true value (53 or 48 or |
| 47 | whatever) when rounded up to a multiple of GMP_NUMB_BITS, for non-nails. |
| 48 | |
| 49 | Enhancements: |
| 50 | |
| 51 | Use the true mantissa size in the N_QLIMBS formula, to save a divide step |
| 52 | in nails. |
| 53 | |
| 54 | Examine the high limbs of num and den to see if the highest 1 bit of the |
| 55 | quotient will fall high enough that just N_QLIMBS-1 limbs is enough to |
| 56 | get the necessary bits, thereby saving a division step. |
| 57 | |
| 58 | Bit shift either num or den to arrange for the above condition on the |
| 59 | high 1 bit of the quotient, to save a division step always. A shift to |
| 60 | save a division step is definitely worthwhile with mpn_tdiv_qr, though we |
| 61 | may want to reassess this on big num/den when a quotient-only division |
| 62 | exists. |
| 63 | |
| 64 | Maybe we could estimate the final exponent using nsize-dsize (and |
| 65 | possibly the high limbs of num and den), so as to detect overflow and |
| 66 | return infinity or zero quickly. Overflow is never very helpful to an |
| 67 | application, and can therefore probably be regarded as abnormal, but we |
| 68 | may still like to optimize it if the conditions are easy. (This would |
| 69 | only be for float formats we know, unknown formats are not important and |
| 70 | can be left to mpn_get_d.) |
| 71 | |
| 72 | Future: |
| 73 | |
| 74 | If/when mpn_tdiv_qr supports its qxn parameter we can use that instead of |
| 75 | padding n with zeros in temporary space. |
| 76 | |
| 77 | Alternatives: |
| 78 | |
| 79 | An alternative algorithm, that may be faster: |
| 80 | 0. Let n be somewhat larger than the number of significant bits in a double. |
| 81 | 1. Extract the most significant n bits of the denominator, and an equal |
| 82 | number of bits from the numerator. |
| 83 | 2. Interpret the extracted numbers as integers, call them a and b |
| 84 | respectively, and develop n bits of the fractions ((a + 1) / b) and |
| 85 | (a / (b + 1)) using mpn_divrem. |
| 86 | 3. If the computed values are identical UP TO THE POSITION WE CARE ABOUT, |
| 87 | we are done. If they are different, repeat the algorithm from step 1, |
| 88 | but first let n = n * 2. |
| 89 | 4. If we end up using all bits from the numerator and denominator, fall |
| 90 | back to a plain division. |
| 91 | 5. Just to make life harder, The computation of a + 1 and b + 1 above |
| 92 | might give carry-out... Needs special handling. It might work to |
| 93 | subtract 1 in both cases instead. |
| 94 | |
| 95 | Not certain if this approach would be faster than a quotient-only |
| 96 | division. Presumably such optimizations are the sort of thing we would |
| 97 | like to have helping everywhere that uses a quotient-only division. */ |
| 98 | |
| 99 | double |
| 100 | mpq_get_d (mpq_srcptr src) |
| 101 | { |
| 102 | double res; |
| 103 | mp_srcptr np, dp; |
| 104 | mp_ptr temp; |
| 105 | mp_size_t nsize = SIZ(NUM(src)); |
| 106 | mp_size_t dsize = SIZ(DEN(src)); |
| 107 | mp_size_t qsize, prospective_qsize, zeros; |
| 108 | mp_size_t sign_quotient = nsize; |
| 109 | long exp; |
| 110 | #define N_QLIMBS (1 + (sizeof (double) + GMP_LIMB_BYTES-1) / GMP_LIMB_BYTES) |
| 111 | mp_limb_t qarr[N_QLIMBS + 1]; |
| 112 | mp_ptr qp = qarr; |
| 113 | TMP_DECL; |
| 114 | |
| 115 | ASSERT (dsize > 0); /* canonical src */ |
| 116 | |
| 117 | /* mpn_get_d below requires a non-zero operand */ |
| 118 | if (UNLIKELY (nsize == 0)) |
| 119 | return 0.0; |
| 120 | |
| 121 | TMP_MARK; |
| 122 | nsize = ABS (nsize); |
| 123 | dsize = ABS (dsize); |
| 124 | np = PTR(NUM(src)); |
| 125 | dp = PTR(DEN(src)); |
| 126 | |
| 127 | prospective_qsize = nsize - dsize; /* from using given n,d */ |
| 128 | qsize = N_QLIMBS; /* desired qsize */ |
| 129 | |
| 130 | zeros = qsize - prospective_qsize; /* padding n to get qsize */ |
| 131 | exp = (long) -zeros * GMP_NUMB_BITS; /* relative to low of qp */ |
| 132 | |
| 133 | /* zero extend n into temporary space, if necessary */ |
| 134 | if (zeros > 0) |
| 135 | { |
| 136 | mp_size_t tsize; |
| 137 | tsize = nsize + zeros; /* size for copy of n */ |
| 138 | |
| 139 | temp = TMP_ALLOC_LIMBS (tsize + 1); |
| 140 | MPN_FILL (temp, zeros, 0); |
| 141 | MPN_COPY (temp + zeros, np, nsize); |
| 142 | np = temp; |
| 143 | nsize = tsize; |
| 144 | } |
| 145 | else /* negative zeros means shorten n */ |
| 146 | { |
| 147 | np -= zeros; |
| 148 | nsize += zeros; |
| 149 | |
| 150 | temp = TMP_ALLOC_LIMBS (nsize + 1); |
| 151 | } |
| 152 | |
| 153 | ASSERT (qsize == nsize - dsize); |
| 154 | mpn_div_q (qp, np, nsize, dp, dsize, temp); |
| 155 | |
| 156 | /* strip possible zero high limb */ |
| 157 | qsize += (qp[qsize] != 0); |
| 158 | |
| 159 | res = mpn_get_d (qp, qsize, sign_quotient, exp); |
| 160 | TMP_FREE; |
| 161 | return res; |
| 162 | } |