1 /* str2num.c: bcmath library file. */
2 /*
3 Copyright (C) 1991, 1992, 1993, 1994, 1997 Free Software Foundation, Inc.
4 Copyright (C) 2000 Philip A. Nelson
5
6 This library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2 of the License, or (at your option) any later version.
10
11 This library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details. (LICENSE)
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with this library; if not, write to:
18
19 The Free Software Foundation, Inc.
20 59 Temple Place, Suite 330
21 Boston, MA 02111-1307 USA.
22
23 You may contact the author by:
24 e-mail: philnelson@acm.org
25 us-mail: Philip A. Nelson
26 Computer Science Department, 9062
27 Western Washington University
28 Bellingham, WA 98226-9062
29
30 *************************************************************************/
31
32 #include "bcmath.h"
33 #include "convert.h"
34 #include "private.h"
35 #include <stdbool.h>
36 #include <stddef.h>
37 #ifdef __SSE2__
38 # include <emmintrin.h>
39 #endif
40
41 /* Convert strings to bc numbers. Base 10 only.*/
bc_count_digits(const char * str,const char * end)42 static const char *bc_count_digits(const char *str, const char *end)
43 {
44 /* Process in bulk */
45 #ifdef __SSE2__
46 const __m128i offset = _mm_set1_epi8((signed char) (SCHAR_MIN - '0'));
47 /* we use the less than comparator, so add 1 */
48 const __m128i threshold = _mm_set1_epi8(SCHAR_MIN + ('9' + 1 - '0'));
49
50 while (str + sizeof(__m128i) <= end) {
51 __m128i bytes = _mm_loadu_si128((const __m128i *) str);
52 /* Wrapping-add the offset to the bytes, such that all bytes below '0' are positive and others are negative.
53 * More specifically, '0' will be -128 and '9' will be -119. */
54 bytes = _mm_add_epi8(bytes, offset);
55 /* Now mark all bytes that are <= '9', i.e. <= -119, i.e. < -118, i.e. the threshold. */
56 bytes = _mm_cmplt_epi8(bytes, threshold);
57
58 int mask = _mm_movemask_epi8(bytes);
59 if (mask != 0xffff) {
60 /* At least one of the bytes is not within range. Move to the first offending byte. */
61 #ifdef PHP_HAVE_BUILTIN_CTZL
62 return str + __builtin_ctz(~mask);
63 #else
64 break;
65 #endif
66 }
67
68 str += sizeof(__m128i);
69 }
70 #endif
71
72 while (*str >= '0' && *str <= '9') {
73 str++;
74 }
75
76 return str;
77 }
78
bc_skip_zero_reverse(const char * scanner,const char * stop)79 static inline const char *bc_skip_zero_reverse(const char *scanner, const char *stop)
80 {
81 /* Check in bulk */
82 #ifdef __SSE2__
83 const __m128i c_zero_repeat = _mm_set1_epi8('0');
84 while (scanner - sizeof(__m128i) >= stop) {
85 scanner -= sizeof(__m128i);
86 __m128i bytes = _mm_loadu_si128((const __m128i *) scanner);
87 /* Checks if all numeric strings are equal to '0'. */
88 bytes = _mm_cmpeq_epi8(bytes, c_zero_repeat);
89
90 int mask = _mm_movemask_epi8(bytes);
91 /* The probability of having 16 trailing 0s in a row is very low, so we use EXPECTED. */
92 if (EXPECTED(mask != 0xffff)) {
93 /* Move the pointer back and check each character in loop. */
94 scanner += sizeof(__m128i);
95 break;
96 }
97 }
98 #endif
99
100 /* Exclude trailing zeros. */
101 while (scanner - 1 >= stop && scanner[-1] == '0') {
102 scanner--;
103 }
104
105 return scanner;
106 }
107
108 /* Assumes `num` points to NULL, i.e. does yet not hold a number. */
bc_str2num(bc_num * num,const char * str,const char * end,size_t scale,size_t * full_scale,bool auto_scale)109 bool bc_str2num(bc_num *num, const char *str, const char *end, size_t scale, size_t *full_scale, bool auto_scale)
110 {
111 size_t str_scale = 0;
112 const char *ptr = str;
113 const char *fractional_ptr = NULL;
114 const char *fractional_end = NULL;
115 bool zero_int = false;
116
117 ZEND_ASSERT(*num == NULL);
118
119 /* Check for valid number and count digits. */
120 if ((*ptr == '+') || (*ptr == '-')) {
121 /* Skip Sign */
122 ptr++;
123 }
124 /* Skip leading zeros. */
125 while (*ptr == '0') {
126 ptr++;
127 }
128 const char *integer_ptr = ptr;
129 /* digits before the decimal point */
130 ptr = bc_count_digits(ptr, end);
131 size_t digits = ptr - integer_ptr;
132 /* decimal point */
133 const char *decimal_point = (*ptr == '.') ? ptr : NULL;
134
135 /* If a non-digit and non-decimal-point indicator is in the string, i.e. an invalid character */
136 if (UNEXPECTED(!decimal_point && *ptr != '\0')) {
137 goto fail;
138 }
139
140 /* search and validate fractional end if exists */
141 if (decimal_point) {
142 /* search */
143 fractional_ptr = fractional_end = decimal_point + 1;
144 /* For strings that end with a decimal point, such as "012." */
145 if (UNEXPECTED(*fractional_ptr == '\0')) {
146 if (full_scale) {
147 *full_scale = 0;
148 }
149 goto after_fractional;
150 }
151
152 /* validate */
153 fractional_end = bc_count_digits(fractional_ptr, end);
154 if (UNEXPECTED(*fractional_end != '\0')) {
155 /* invalid num */
156 goto fail;
157 }
158
159 if (full_scale) {
160 *full_scale = fractional_end - fractional_ptr;
161 }
162
163 /* Exclude trailing zeros. */
164 fractional_end = bc_skip_zero_reverse(fractional_end, fractional_ptr);
165
166 /* Calculate the length of the fraction excluding trailing zero. */
167 str_scale = fractional_end - fractional_ptr;
168
169 /*
170 * If set the scale manually and it is smaller than the automatically calculated scale,
171 * adjust it to match the manual setting.
172 */
173 if (str_scale > scale && !auto_scale) {
174 fractional_end -= str_scale - scale;
175 str_scale = scale;
176 }
177 } else {
178 if (full_scale) {
179 *full_scale = 0;
180 }
181 }
182
183 after_fractional:
184
185 if (digits + str_scale == 0) {
186 goto zero;
187 }
188
189 /* Adjust numbers and allocate storage and initialize fields. */
190 if (digits == 0) {
191 zero_int = true;
192 digits = 1;
193 }
194 *num = bc_new_num_nonzeroed(digits, str_scale);
195 (*num)->n_sign = *str == '-' ? MINUS : PLUS;
196 char *nptr = (*num)->n_value;
197
198 if (zero_int) {
199 *nptr++ = 0;
200 /*
201 * If zero_int is true and the str_scale is 0, there is an early return,
202 * so here str_scale is always greater than 0.
203 */
204 nptr = bc_copy_and_toggle_bcd(nptr, fractional_ptr, fractional_end);
205 } else {
206 const char *integer_end = integer_ptr + digits;
207 nptr = bc_copy_and_toggle_bcd(nptr, integer_ptr, integer_end);
208 if (str_scale > 0) {
209 nptr = bc_copy_and_toggle_bcd(nptr, fractional_ptr, fractional_end);
210 }
211 }
212
213 return true;
214
215 zero:
216 *num = bc_copy_num(BCG(_zero_));
217 return true;
218
219 fail:
220 *num = bc_copy_num(BCG(_zero_));
221 return false;
222 }
223
