/src/mupdf/source/fitz/ftoa.c
Line | Count | Source (jump to first uncovered line) |
1 | | #include "mupdf/fitz.h" |
2 | | |
3 | | #include <assert.h> |
4 | | |
5 | | /* |
6 | | Convert IEEE single precision numbers into decimal ASCII strings, while |
7 | | satisfying the following two properties: |
8 | | 1) Calling strtof or '(float) strtod' on the result must produce the |
9 | | original float, independent of the rounding mode used by strtof/strtod. |
10 | | 2) Minimize the number of produced decimal digits. E.g. the float 0.7f |
11 | | should convert to "0.7", not "0.69999999". |
12 | | |
13 | | To solve this we use a dedicated single precision version of |
14 | | Florian Loitsch's Grisu2 algorithm. See |
15 | | http://florian.loitsch.com/publications/dtoa-pldi2010.pdf?attredirects=0 |
16 | | |
17 | | The code below is derived from Loitsch's C code, which |
18 | | implements the same algorithm for IEEE double precision. See |
19 | | http://florian.loitsch.com/publications/bench.tar.gz?attredirects=0 |
20 | | */ |
21 | | |
22 | | /* |
23 | | Copyright (c) 2009 Florian Loitsch |
24 | | |
25 | | Permission is hereby granted, free of charge, to any person |
26 | | obtaining a copy of this software and associated documentation |
27 | | files (the "Software"), to deal in the Software without |
28 | | restriction, including without limitation the rights to use, |
29 | | copy, modify, merge, publish, distribute, sublicense, and/or sell |
30 | | copies of the Software, and to permit persons to whom the |
31 | | Software is furnished to do so, subject to the following |
32 | | conditions: |
33 | | |
34 | | The above copyright notice and this permission notice shall be |
35 | | included in all copies or substantial portions of the Software. |
36 | | |
37 | | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
38 | | EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES |
39 | | OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
40 | | NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT |
41 | | HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, |
42 | | WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
43 | | FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
44 | | OTHER DEALINGS IN THE SOFTWARE. |
45 | | */ |
46 | | |
47 | | static uint32_t |
48 | | float_to_uint32(float d) |
49 | 118k | { |
50 | 118k | union |
51 | 118k | { |
52 | 118k | float d; |
53 | 118k | uint32_t n; |
54 | 118k | } tmp; |
55 | 118k | tmp.d = d; |
56 | 118k | return tmp.n; |
57 | 118k | } |
58 | | |
59 | | typedef struct |
60 | | { |
61 | | uint64_t f; |
62 | | int e; |
63 | | } diy_fp_t; |
64 | | |
65 | 826k | #define DIY_SIGNIFICAND_SIZE 64 |
66 | | #define DIY_LEADING_BIT ((uint64_t) 1 << (DIY_SIGNIFICAND_SIZE - 1)) |
67 | | |
68 | | static diy_fp_t |
69 | | minus(diy_fp_t x, diy_fp_t y) |
70 | 118k | { |
71 | 118k | diy_fp_t result = {x.f - y.f, x.e}; |
72 | 118k | assert(x.e == y.e && x.f >= y.f); |
73 | 118k | return result; |
74 | 118k | } |
75 | | |
76 | | static diy_fp_t |
77 | | multiply(diy_fp_t x, diy_fp_t y) |
78 | 236k | { |
79 | 236k | uint64_t a, b, c, d, ac, bc, ad, bd, tmp; |
80 | 236k | int half = DIY_SIGNIFICAND_SIZE / 2; |
81 | 236k | diy_fp_t r; uint64_t mask = ((uint64_t) 1 << half) - 1; |
82 | 236k | a = x.f >> half; b = x.f & mask; |
83 | 236k | c = y.f >> half; d = y.f & mask; |
84 | 236k | ac = a * c; bc = b * c; ad = a * d; bd = b * d; |
85 | 236k | tmp = (bd >> half) + (ad & mask) + (bc & mask); |
86 | 236k | tmp += ((uint64_t)1U) << (half - 1); /* Round. */ |
87 | 236k | r.f = ac + (ad >> half) + (bc >> half) + (tmp >> half); |
88 | 236k | r.e = x.e + y.e + half * 2; |
89 | 236k | return r; |
90 | 236k | } |
91 | | |
92 | 472k | #define SP_SIGNIFICAND_SIZE 23 |
93 | 118k | #define SP_EXPONENT_BIAS (127 + SP_SIGNIFICAND_SIZE) |
94 | 0 | #define SP_MIN_EXPONENT (-SP_EXPONENT_BIAS) |
95 | 118k | #define SP_EXPONENT_MASK 0x7f800000 |
96 | 118k | #define SP_SIGNIFICAND_MASK 0x7fffff |
97 | 354k | #define SP_HIDDEN_BIT 0x800000 /* 2^23 */ |
98 | | |
99 | | /* Does not normalize the result. */ |
100 | | static diy_fp_t |
101 | | float2diy_fp(float d) |
102 | 118k | { |
103 | 118k | uint32_t d32 = float_to_uint32(d); |
104 | 118k | int biased_e = (d32 & SP_EXPONENT_MASK) >> SP_SIGNIFICAND_SIZE; |
105 | 118k | uint32_t significand = d32 & SP_SIGNIFICAND_MASK; |
106 | 118k | diy_fp_t res; |
107 | | |
108 | 118k | if (biased_e != 0) |
109 | 118k | { |
110 | 118k | res.f = significand + SP_HIDDEN_BIT; |
111 | 118k | res.e = biased_e - SP_EXPONENT_BIAS; |
112 | 118k | } |
113 | 0 | else |
114 | 0 | { |
115 | 0 | res.f = significand; |
116 | 0 | res.e = SP_MIN_EXPONENT + 1; |
117 | 0 | } |
118 | 118k | return res; |
119 | 118k | } |
120 | | |
121 | | static diy_fp_t |
122 | | normalize_boundary(diy_fp_t in) |
123 | 118k | { |
124 | 118k | diy_fp_t res = in; |
125 | | /* The original number could have been a denormal. */ |
126 | 118k | while (! (res.f & (SP_HIDDEN_BIT << 1))) |
127 | 0 | { |
128 | 0 | res.f <<= 1; |
129 | 0 | res.e--; |
130 | 0 | } |
131 | | /* Do the final shifts in one go. */ |
132 | 118k | res.f <<= (DIY_SIGNIFICAND_SIZE - SP_SIGNIFICAND_SIZE - 2); |
133 | 118k | res.e = res.e - (DIY_SIGNIFICAND_SIZE - SP_SIGNIFICAND_SIZE - 2); |
134 | 118k | return res; |
135 | 118k | } |
136 | | |
137 | | static void |
138 | | normalized_boundaries(float f, diy_fp_t* lower_ptr, diy_fp_t* upper_ptr) |
139 | 118k | { |
140 | 118k | diy_fp_t v = float2diy_fp(f); |
141 | 118k | diy_fp_t upper, lower; |
142 | 118k | int significand_is_zero = v.f == SP_HIDDEN_BIT; |
143 | | |
144 | 118k | upper.f = (v.f << 1) + 1; upper.e = v.e - 1; |
145 | 118k | upper = normalize_boundary(upper); |
146 | 118k | if (significand_is_zero) |
147 | 15.1k | { |
148 | 15.1k | lower.f = (v.f << 2) - 1; |
149 | 15.1k | lower.e = v.e - 2; |
150 | 15.1k | } |
151 | 102k | else |
152 | 102k | { |
153 | 102k | lower.f = (v.f << 1) - 1; |
154 | 102k | lower.e = v.e - 1; |
155 | 102k | } |
156 | 118k | lower.f <<= lower.e - upper.e; |
157 | 118k | lower.e = upper.e; |
158 | | |
159 | | /* Adjust to double boundaries, so that we can also read the numbers with '(float) strtod'. */ |
160 | 118k | upper.f -= 1 << 10; |
161 | 118k | lower.f += 1 << 10; |
162 | | |
163 | 118k | *upper_ptr = upper; |
164 | 118k | *lower_ptr = lower; |
165 | 118k | } |
166 | | |
167 | | static int |
168 | | k_comp(int n) |
169 | 118k | { |
170 | | /* Avoid ceil and floating point multiplication for better |
171 | | * performance and portability. Instead use the approximation |
172 | | * log10(2) ~ 1233/(2^12). Tests show that this gives the correct |
173 | | * result for all values of n in the range -500..500. */ |
174 | 118k | int tmp = n + DIY_SIGNIFICAND_SIZE - 1; |
175 | 118k | int k = (tmp * 1233) / (1 << 12); |
176 | 118k | return tmp > 0 ? k + 1 : k; |
177 | 118k | } |
178 | | |
179 | | /* Cached powers of ten from 10**-37..10**46. Produced using GNU MPFR's mpfr_pow_si. */ |
180 | | |
181 | | /* Significands. */ |
182 | | static uint64_t powers_ten[84] = { |
183 | | 0x881cea14545c7575ull, 0xaa242499697392d3ull, 0xd4ad2dbfc3d07788ull, |
184 | | 0x84ec3c97da624ab5ull, 0xa6274bbdd0fadd62ull, 0xcfb11ead453994baull, |
185 | | 0x81ceb32c4b43fcf5ull, 0xa2425ff75e14fc32ull, 0xcad2f7f5359a3b3eull, |
186 | | 0xfd87b5f28300ca0eull, 0x9e74d1b791e07e48ull, 0xc612062576589ddbull, |
187 | | 0xf79687aed3eec551ull, 0x9abe14cd44753b53ull, 0xc16d9a0095928a27ull, |
188 | | 0xf1c90080baf72cb1ull, 0x971da05074da7befull, 0xbce5086492111aebull, |
189 | | 0xec1e4a7db69561a5ull, 0x9392ee8e921d5d07ull, 0xb877aa3236a4b449ull, |
190 | | 0xe69594bec44de15bull, 0x901d7cf73ab0acd9ull, 0xb424dc35095cd80full, |
191 | | 0xe12e13424bb40e13ull, 0x8cbccc096f5088ccull, 0xafebff0bcb24aaffull, |
192 | | 0xdbe6fecebdedd5bfull, 0x89705f4136b4a597ull, 0xabcc77118461cefdull, |
193 | | 0xd6bf94d5e57a42bcull, 0x8637bd05af6c69b6ull, 0xa7c5ac471b478423ull, |
194 | | 0xd1b71758e219652cull, 0x83126e978d4fdf3bull, 0xa3d70a3d70a3d70aull, |
195 | | 0xcccccccccccccccdull, 0x8000000000000000ull, 0xa000000000000000ull, |
196 | | 0xc800000000000000ull, 0xfa00000000000000ull, 0x9c40000000000000ull, |
197 | | 0xc350000000000000ull, 0xf424000000000000ull, 0x9896800000000000ull, |
198 | | 0xbebc200000000000ull, 0xee6b280000000000ull, 0x9502f90000000000ull, |
199 | | 0xba43b74000000000ull, 0xe8d4a51000000000ull, 0x9184e72a00000000ull, |
200 | | 0xb5e620f480000000ull, 0xe35fa931a0000000ull, 0x8e1bc9bf04000000ull, |
201 | | 0xb1a2bc2ec5000000ull, 0xde0b6b3a76400000ull, 0x8ac7230489e80000ull, |
202 | | 0xad78ebc5ac620000ull, 0xd8d726b7177a8000ull, 0x878678326eac9000ull, |
203 | | 0xa968163f0a57b400ull, 0xd3c21bcecceda100ull, 0x84595161401484a0ull, |
204 | | 0xa56fa5b99019a5c8ull, 0xcecb8f27f4200f3aull, 0x813f3978f8940984ull, |
205 | | 0xa18f07d736b90be5ull, 0xc9f2c9cd04674edfull, 0xfc6f7c4045812296ull, |
206 | | 0x9dc5ada82b70b59eull, 0xc5371912364ce305ull, 0xf684df56c3e01bc7ull, |
207 | | 0x9a130b963a6c115cull, 0xc097ce7bc90715b3ull, 0xf0bdc21abb48db20ull, |
208 | | 0x96769950b50d88f4ull, 0xbc143fa4e250eb31ull, 0xeb194f8e1ae525fdull, |
209 | | 0x92efd1b8d0cf37beull, 0xb7abc627050305aeull, 0xe596b7b0c643c719ull, |
210 | | 0x8f7e32ce7bea5c70ull, 0xb35dbf821ae4f38cull, 0xe0352f62a19e306full, |
211 | | }; |
212 | | |
213 | | /* Exponents. */ |
214 | | static int powers_ten_e[84] = { |
215 | | -186, -183, -180, -176, -173, -170, -166, -163, -160, -157, -153, |
216 | | -150, -147, -143, -140, -137, -133, -130, -127, -123, -120, -117, |
217 | | -113, -110, -107, -103, -100, -97, -93, -90, -87, -83, -80, |
218 | | -77, -73, -70, -67, -63, -60, -57, -54, -50, -47, -44, |
219 | | -40, -37, -34, -30, -27, -24, -20, -17, -14, -10, -7, |
220 | | -4, 0, 3, 6, 10, 13, 16, 20, 23, 26, 30, |
221 | | 33, 36, 39, 43, 46, 49, 53, 56, 59, 63, 66, |
222 | | 69, 73, 76, 79, 83, 86, 89 |
223 | | }; |
224 | | |
225 | | static diy_fp_t |
226 | | cached_power(int i) |
227 | 118k | { |
228 | 118k | diy_fp_t result; |
229 | | |
230 | 118k | assert (i >= -37 && i <= 46); |
231 | 118k | result.f = powers_ten[i + 37]; |
232 | 118k | result.e = powers_ten_e[i + 37]; |
233 | 118k | return result; |
234 | 118k | } |
235 | | |
236 | | /* Returns buffer length. */ |
237 | | static int |
238 | | digit_gen_mix_grisu2(diy_fp_t D_upper, diy_fp_t delta, char* buffer, int* K) |
239 | 118k | { |
240 | 118k | int kappa; |
241 | 118k | diy_fp_t one = {(uint64_t) 1 << -D_upper.e, D_upper.e}; |
242 | 118k | unsigned char p1 = D_upper.f >> -one.e; |
243 | 118k | uint64_t p2 = D_upper.f & (one.f - 1); |
244 | 118k | unsigned char div = 10; |
245 | 118k | uint64_t mask = one.f - 1; |
246 | 118k | int len = 0; |
247 | 313k | for (kappa = 2; kappa > 0; --kappa) |
248 | 217k | { |
249 | 217k | unsigned char digit = p1 / div; |
250 | 217k | if (digit || len) |
251 | 194k | buffer[len++] = '0' + digit; |
252 | 217k | p1 %= div; div /= 10; |
253 | 217k | if ((((uint64_t) p1) << -one.e) + p2 <= delta.f) |
254 | 22.2k | { |
255 | 22.2k | *K += kappa - 1; |
256 | 22.2k | return len; |
257 | 22.2k | } |
258 | 217k | } |
259 | 95.8k | do |
260 | 496k | { |
261 | 496k | p2 *= 10; |
262 | 496k | buffer[len++] = '0' + (p2 >> -one.e); |
263 | 496k | p2 &= mask; |
264 | 496k | kappa--; |
265 | 496k | delta.f *= 10; |
266 | 496k | } |
267 | 496k | while (p2 > delta.f); |
268 | 95.8k | *K += kappa; |
269 | 95.8k | return len; |
270 | 118k | } |
271 | | |
272 | | /* |
273 | | Compute decimal integer m, exp such that: |
274 | | f = m * 10^exp |
275 | | m is as short as possible without losing exactness |
276 | | Assumes special cases (0, NaN, +Inf, -Inf) have been handled. |
277 | | */ |
278 | | int |
279 | | fz_grisu(float v, char* buffer, int* K) |
280 | 118k | { |
281 | 118k | diy_fp_t w_lower, w_upper, D_upper, D_lower, c_mk, delta; |
282 | 118k | int length, mk, alpha = -DIY_SIGNIFICAND_SIZE + 4; |
283 | | |
284 | 118k | normalized_boundaries(v, &w_lower, &w_upper); |
285 | 118k | mk = k_comp(alpha - w_upper.e - DIY_SIGNIFICAND_SIZE); |
286 | 118k | c_mk = cached_power(mk); |
287 | | |
288 | 118k | D_upper = multiply(w_upper, c_mk); |
289 | 118k | D_lower = multiply(w_lower, c_mk); |
290 | | |
291 | 118k | D_upper.f--; |
292 | 118k | D_lower.f++; |
293 | | |
294 | 118k | delta = minus(D_upper, D_lower); |
295 | | |
296 | 118k | *K = -mk; |
297 | 118k | length = digit_gen_mix_grisu2(D_upper, delta, buffer, K); |
298 | | |
299 | 118k | buffer[length] = 0; |
300 | 118k | return length; |
301 | 118k | } |