1 | /* Floating-point remainder function. |
2 | Copyright (C) 2023-2024 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | |
5 | The GNU C Library is free software; you can redistribute it and/or |
6 | modify it under the terms of the GNU Lesser General Public |
7 | License as published by the Free Software Foundation; either |
8 | version 2.1 of the License, or (at your option) any later version. |
9 | |
10 | The GNU C Library is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 | Lesser General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU Lesser General Public |
16 | License along with the GNU C Library; if not, see |
17 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <libm-alias-finite.h> |
20 | #include <libm-alias-float.h> |
21 | #include <math-svid-compat.h> |
22 | #include <math.h> |
23 | #include "math_config.h" |
24 | |
25 | /* With x = mx * 2^ex and y = my * 2^ey (mx, my, ex, ey being integers), the |
26 | simplest implementation is: |
27 | |
28 | mx * 2^ex == 2 * mx * 2^(ex - 1) |
29 | |
30 | or |
31 | |
32 | while (ex > ey) |
33 | { |
34 | mx *= 2; |
35 | --ex; |
36 | mx %= my; |
37 | } |
38 | |
39 | With the mathematical equivalence of: |
40 | |
41 | r == x % y == (x % (N * y)) % y |
42 | |
43 | And with mx/my being mantissa of a single floating point number (which uses |
44 | less bits than the storage type), on each step the argument reduction can |
45 | be improved by 8 (which is the size of uint32_t minus MANTISSA_WIDTH plus |
46 | the implicit one bit): |
47 | |
48 | mx * 2^ex == 2^8 * mx * 2^(ex - 8) |
49 | |
50 | or |
51 | |
52 | while (ex > ey) |
53 | { |
54 | mx << 8; |
55 | ex -= 8; |
56 | mx %= my; |
57 | } |
58 | |
59 | Special cases: |
60 | - If x or y is a NaN, a NaN is returned. |
61 | - If x is an infinity, or y is zero, a NaN is returned and EDOM is set. |
62 | - If x is +0/-0, and y is not zero, +0/-0 is returned. */ |
63 | |
64 | float |
65 | __fmodf (float x, float y) |
66 | { |
67 | uint32_t hx = asuint (f: x); |
68 | uint32_t hy = asuint (f: y); |
69 | |
70 | uint32_t sx = hx & SIGN_MASK; |
71 | /* Get |x| and |y|. */ |
72 | hx ^= sx; |
73 | hy &= ~SIGN_MASK; |
74 | |
75 | if (__glibc_likely (hx < hy)) |
76 | { |
77 | /* If y is a NaN, return a NaN. */ |
78 | if (__glibc_unlikely (hy > EXPONENT_MASK)) |
79 | return x * y; |
80 | return x; |
81 | } |
82 | |
83 | int ex = hx >> MANTISSA_WIDTH; |
84 | int ey = hy >> MANTISSA_WIDTH; |
85 | int exp_diff = ex - ey; |
86 | |
87 | /* Common case where exponents are close: |x/y| < 2^9, x not inf/NaN |
88 | and |x%y| not denormal. */ |
89 | if (__glibc_likely (ey < (EXPONENT_MASK >> MANTISSA_WIDTH) - EXPONENT_WIDTH |
90 | && ey > MANTISSA_WIDTH |
91 | && exp_diff <= EXPONENT_WIDTH)) |
92 | { |
93 | uint32_t mx = (hx << EXPONENT_WIDTH) | SIGN_MASK; |
94 | uint32_t my = (hy << EXPONENT_WIDTH) | SIGN_MASK; |
95 | |
96 | mx %= (my >> exp_diff); |
97 | |
98 | if (__glibc_unlikely (mx == 0)) |
99 | return asfloat (i: sx); |
100 | int shift = __builtin_clz (mx); |
101 | ex -= shift + 1; |
102 | mx <<= shift; |
103 | mx = sx | (mx >> EXPONENT_WIDTH); |
104 | return asfloat (i: mx + ((uint32_t)ex << MANTISSA_WIDTH)); |
105 | } |
106 | |
107 | if (__glibc_unlikely (hy == 0 || hx >= EXPONENT_MASK)) |
108 | { |
109 | /* If x is a NaN, return a NaN. */ |
110 | if (hx > EXPONENT_MASK) |
111 | return x * y; |
112 | |
113 | /* If x is an infinity or y is zero, return a NaN and set EDOM. */ |
114 | return __math_edomf (x: (x * y) / (x * y)); |
115 | } |
116 | |
117 | /* Special case, both x and y are denormal. */ |
118 | if (__glibc_unlikely (ex == 0)) |
119 | return asfloat (i: sx | hx % hy); |
120 | |
121 | /* Extract normalized mantissas - hx is not denormal and hy != 0. */ |
122 | uint32_t mx = get_mantissa (x: hx) | (MANTISSA_MASK + 1); |
123 | uint32_t my = get_mantissa (x: hy) | (MANTISSA_MASK + 1); |
124 | int lead_zeros_my = EXPONENT_WIDTH; |
125 | |
126 | ey--; |
127 | /* Special case for denormal y. */ |
128 | if (__glibc_unlikely (ey < 0)) |
129 | { |
130 | my = hy; |
131 | ey = 0; |
132 | exp_diff--; |
133 | lead_zeros_my = __builtin_clz (my); |
134 | } |
135 | |
136 | int tail_zeros_my = __builtin_ctz (my); |
137 | int sides_zeroes = lead_zeros_my + tail_zeros_my; |
138 | |
139 | int right_shift = exp_diff < tail_zeros_my ? exp_diff : tail_zeros_my; |
140 | my >>= right_shift; |
141 | exp_diff -= right_shift; |
142 | ey += right_shift; |
143 | |
144 | int left_shift = exp_diff < EXPONENT_WIDTH ? exp_diff : EXPONENT_WIDTH; |
145 | mx <<= left_shift; |
146 | exp_diff -= left_shift; |
147 | |
148 | mx %= my; |
149 | |
150 | if (__glibc_unlikely (mx == 0)) |
151 | return asfloat (i: sx); |
152 | |
153 | if (exp_diff == 0) |
154 | return make_float (x: mx, ep: ey, s: sx); |
155 | |
156 | /* Multiplication with the inverse is faster than repeated modulo. */ |
157 | uint32_t inv_hy = UINT32_MAX / my; |
158 | while (exp_diff > sides_zeroes) { |
159 | exp_diff -= sides_zeroes; |
160 | uint32_t hd = (mx * inv_hy) >> (BIT_WIDTH - sides_zeroes); |
161 | mx <<= sides_zeroes; |
162 | mx -= hd * my; |
163 | while (__glibc_unlikely (mx > my)) |
164 | mx -= my; |
165 | } |
166 | uint32_t hd = (mx * inv_hy) >> (BIT_WIDTH - exp_diff); |
167 | mx <<= exp_diff; |
168 | mx -= hd * my; |
169 | while (__glibc_unlikely (mx > my)) |
170 | mx -= my; |
171 | |
172 | return make_float (x: mx, ep: ey, s: sx); |
173 | } |
174 | strong_alias (__fmodf, __ieee754_fmodf) |
175 | #if LIBM_SVID_COMPAT |
176 | versioned_symbol (libm, __fmodf, fmodf, GLIBC_2_38); |
177 | libm_alias_float_other (__fmod, fmod) |
178 | #else |
179 | libm_alias_float (__fmod, fmod) |
180 | #endif |
181 | libm_alias_finite (__ieee754_fmodf, __fmodf) |
182 | |