root / host-utils.h @ cb9c377f
History | View | Annotate | Download (5.5 kB)
1 | 05f778c8 | ths | /*
|
---|---|---|---|
2 | 05f778c8 | ths | * Utility compute operations used by translated code.
|
3 | 05f778c8 | ths | *
|
4 | 05f778c8 | ths | * Copyright (c) 2007 Thiemo Seufer
|
5 | 05f778c8 | ths | * Copyright (c) 2007 Jocelyn Mayer
|
6 | 05f778c8 | ths | *
|
7 | 05f778c8 | ths | * Permission is hereby granted, free of charge, to any person obtaining a copy
|
8 | 05f778c8 | ths | * of this software and associated documentation files (the "Software"), to deal
|
9 | 05f778c8 | ths | * in the Software without restriction, including without limitation the rights
|
10 | 05f778c8 | ths | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
11 | 05f778c8 | ths | * copies of the Software, and to permit persons to whom the Software is
|
12 | 05f778c8 | ths | * furnished to do so, subject to the following conditions:
|
13 | 05f778c8 | ths | *
|
14 | 05f778c8 | ths | * The above copyright notice and this permission notice shall be included in
|
15 | 05f778c8 | ths | * all copies or substantial portions of the Software.
|
16 | 05f778c8 | ths | *
|
17 | 05f778c8 | ths | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
18 | 05f778c8 | ths | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
19 | 05f778c8 | ths | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
20 | 05f778c8 | ths | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
21 | 05f778c8 | ths | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
22 | 05f778c8 | ths | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
23 | 05f778c8 | ths | * THE SOFTWARE.
|
24 | 05f778c8 | ths | */
|
25 | cb9c377f | Paolo Bonzini | #ifndef HOST_UTILS_H
|
26 | cb9c377f | Paolo Bonzini | #define HOST_UTILS_H 1 |
27 | 05f778c8 | ths | |
28 | f8b72754 | Stefan Weil | #include "compiler.h" /* QEMU_GNUC_PREREQ */ |
29 | cebdff77 | ths | |
30 | 7a51ad82 | j_mayer | #if defined(__x86_64__)
|
31 | 7a51ad82 | j_mayer | #define __HAVE_FAST_MULU64__
|
32 | facd2857 | Blue Swirl | static inline void mulu64(uint64_t *plow, uint64_t *phigh, |
33 | facd2857 | Blue Swirl | uint64_t a, uint64_t b) |
34 | 7a51ad82 | j_mayer | { |
35 | 7a51ad82 | j_mayer | __asm__ ("mul %0\n\t"
|
36 | 7a51ad82 | j_mayer | : "=d" (*phigh), "=a" (*plow) |
37 | 7a51ad82 | j_mayer | : "a" (a), "0" (b)); |
38 | 7a51ad82 | j_mayer | } |
39 | 7a51ad82 | j_mayer | #define __HAVE_FAST_MULS64__
|
40 | facd2857 | Blue Swirl | static inline void muls64(uint64_t *plow, uint64_t *phigh, |
41 | facd2857 | Blue Swirl | int64_t a, int64_t b) |
42 | 7a51ad82 | j_mayer | { |
43 | 7a51ad82 | j_mayer | __asm__ ("imul %0\n\t"
|
44 | 7a51ad82 | j_mayer | : "=d" (*phigh), "=a" (*plow) |
45 | 7a51ad82 | j_mayer | : "a" (a), "0" (b)); |
46 | 7a51ad82 | j_mayer | } |
47 | 7a51ad82 | j_mayer | #else
|
48 | 05e1d830 | j_mayer | void muls64(uint64_t *phigh, uint64_t *plow, int64_t a, int64_t b);
|
49 | 7a51ad82 | j_mayer | void mulu64(uint64_t *phigh, uint64_t *plow, uint64_t a, uint64_t b);
|
50 | 7a51ad82 | j_mayer | #endif
|
51 | 7a51ad82 | j_mayer | |
52 | 05f778c8 | ths | /* Binary search for leading zeros. */
|
53 | 05f778c8 | ths | |
54 | facd2857 | Blue Swirl | static inline int clz32(uint32_t val) |
55 | 05f778c8 | ths | { |
56 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
57 | 7d019980 | aurel32 | if (val)
|
58 | 7d019980 | aurel32 | return __builtin_clz(val);
|
59 | 7d019980 | aurel32 | else
|
60 | 7d019980 | aurel32 | return 32; |
61 | 7d019980 | aurel32 | #else
|
62 | 05f778c8 | ths | int cnt = 0; |
63 | 05f778c8 | ths | |
64 | 05f778c8 | ths | if (!(val & 0xFFFF0000U)) { |
65 | 05f778c8 | ths | cnt += 16;
|
66 | 05f778c8 | ths | val <<= 16;
|
67 | 05f778c8 | ths | } |
68 | 05f778c8 | ths | if (!(val & 0xFF000000U)) { |
69 | 05f778c8 | ths | cnt += 8;
|
70 | 05f778c8 | ths | val <<= 8;
|
71 | 05f778c8 | ths | } |
72 | 05f778c8 | ths | if (!(val & 0xF0000000U)) { |
73 | 05f778c8 | ths | cnt += 4;
|
74 | 05f778c8 | ths | val <<= 4;
|
75 | 05f778c8 | ths | } |
76 | 05f778c8 | ths | if (!(val & 0xC0000000U)) { |
77 | 05f778c8 | ths | cnt += 2;
|
78 | 05f778c8 | ths | val <<= 2;
|
79 | 05f778c8 | ths | } |
80 | 05f778c8 | ths | if (!(val & 0x80000000U)) { |
81 | 05f778c8 | ths | cnt++; |
82 | 05f778c8 | ths | val <<= 1;
|
83 | 05f778c8 | ths | } |
84 | 05f778c8 | ths | if (!(val & 0x80000000U)) { |
85 | 05f778c8 | ths | cnt++; |
86 | 05f778c8 | ths | } |
87 | 05f778c8 | ths | return cnt;
|
88 | 7d019980 | aurel32 | #endif
|
89 | 05f778c8 | ths | } |
90 | 05f778c8 | ths | |
91 | facd2857 | Blue Swirl | static inline int clo32(uint32_t val) |
92 | 05f778c8 | ths | { |
93 | 05f778c8 | ths | return clz32(~val);
|
94 | 05f778c8 | ths | } |
95 | 05f778c8 | ths | |
96 | facd2857 | Blue Swirl | static inline int clz64(uint64_t val) |
97 | 05f778c8 | ths | { |
98 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
99 | 7d019980 | aurel32 | if (val)
|
100 | 7d019980 | aurel32 | return __builtin_clzll(val);
|
101 | 7d019980 | aurel32 | else
|
102 | 7d019980 | aurel32 | return 64; |
103 | 7d019980 | aurel32 | #else
|
104 | 05f778c8 | ths | int cnt = 0; |
105 | 05f778c8 | ths | |
106 | 7a51ad82 | j_mayer | if (!(val >> 32)) { |
107 | 05f778c8 | ths | cnt += 32;
|
108 | 7a51ad82 | j_mayer | } else {
|
109 | 7a51ad82 | j_mayer | val >>= 32;
|
110 | 05f778c8 | ths | } |
111 | 7a51ad82 | j_mayer | |
112 | 7a51ad82 | j_mayer | return cnt + clz32(val);
|
113 | 7d019980 | aurel32 | #endif
|
114 | 05f778c8 | ths | } |
115 | 05f778c8 | ths | |
116 | facd2857 | Blue Swirl | static inline int clo64(uint64_t val) |
117 | 05f778c8 | ths | { |
118 | 05f778c8 | ths | return clz64(~val);
|
119 | 05f778c8 | ths | } |
120 | b9ef45ff | j_mayer | |
121 | facd2857 | Blue Swirl | static inline int ctz32(uint32_t val) |
122 | b9ef45ff | j_mayer | { |
123 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
124 | 7d019980 | aurel32 | if (val)
|
125 | 7d019980 | aurel32 | return __builtin_ctz(val);
|
126 | 7d019980 | aurel32 | else
|
127 | 7d019980 | aurel32 | return 32; |
128 | 7d019980 | aurel32 | #else
|
129 | b9ef45ff | j_mayer | int cnt;
|
130 | b9ef45ff | j_mayer | |
131 | b9ef45ff | j_mayer | cnt = 0;
|
132 | b9ef45ff | j_mayer | if (!(val & 0x0000FFFFUL)) { |
133 | c8906845 | balrog | cnt += 16;
|
134 | b9ef45ff | j_mayer | val >>= 16;
|
135 | c8906845 | balrog | } |
136 | b9ef45ff | j_mayer | if (!(val & 0x000000FFUL)) { |
137 | c8906845 | balrog | cnt += 8;
|
138 | b9ef45ff | j_mayer | val >>= 8;
|
139 | c8906845 | balrog | } |
140 | b9ef45ff | j_mayer | if (!(val & 0x0000000FUL)) { |
141 | c8906845 | balrog | cnt += 4;
|
142 | b9ef45ff | j_mayer | val >>= 4;
|
143 | c8906845 | balrog | } |
144 | b9ef45ff | j_mayer | if (!(val & 0x00000003UL)) { |
145 | c8906845 | balrog | cnt += 2;
|
146 | b9ef45ff | j_mayer | val >>= 2;
|
147 | c8906845 | balrog | } |
148 | b9ef45ff | j_mayer | if (!(val & 0x00000001UL)) { |
149 | c8906845 | balrog | cnt++; |
150 | b9ef45ff | j_mayer | val >>= 1;
|
151 | c8906845 | balrog | } |
152 | b9ef45ff | j_mayer | if (!(val & 0x00000001UL)) { |
153 | c8906845 | balrog | cnt++; |
154 | c8906845 | balrog | } |
155 | b9ef45ff | j_mayer | |
156 | c8906845 | balrog | return cnt;
|
157 | 7d019980 | aurel32 | #endif
|
158 | c8906845 | balrog | } |
159 | c8906845 | balrog | |
160 | facd2857 | Blue Swirl | static inline int cto32(uint32_t val) |
161 | c8906845 | balrog | { |
162 | b9ef45ff | j_mayer | return ctz32(~val);
|
163 | b9ef45ff | j_mayer | } |
164 | b9ef45ff | j_mayer | |
165 | facd2857 | Blue Swirl | static inline int ctz64(uint64_t val) |
166 | b9ef45ff | j_mayer | { |
167 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
168 | 7d019980 | aurel32 | if (val)
|
169 | 06445248 | Richard Henderson | return __builtin_ctzll(val);
|
170 | 7d019980 | aurel32 | else
|
171 | 7d019980 | aurel32 | return 64; |
172 | 7d019980 | aurel32 | #else
|
173 | b9ef45ff | j_mayer | int cnt;
|
174 | b9ef45ff | j_mayer | |
175 | b9ef45ff | j_mayer | cnt = 0;
|
176 | b9ef45ff | j_mayer | if (!((uint32_t)val)) {
|
177 | b9ef45ff | j_mayer | cnt += 32;
|
178 | b9ef45ff | j_mayer | val >>= 32;
|
179 | b9ef45ff | j_mayer | } |
180 | b9ef45ff | j_mayer | |
181 | b9ef45ff | j_mayer | return cnt + ctz32(val);
|
182 | 7d019980 | aurel32 | #endif
|
183 | b9ef45ff | j_mayer | } |
184 | b9ef45ff | j_mayer | |
185 | facd2857 | Blue Swirl | static inline int cto64(uint64_t val) |
186 | b9ef45ff | j_mayer | { |
187 | b9ef45ff | j_mayer | return ctz64(~val);
|
188 | b9ef45ff | j_mayer | } |
189 | b9ef45ff | j_mayer | |
190 | facd2857 | Blue Swirl | static inline int ctpop8(uint8_t val) |
191 | b9ef45ff | j_mayer | { |
192 | b9ef45ff | j_mayer | val = (val & 0x55) + ((val >> 1) & 0x55); |
193 | b9ef45ff | j_mayer | val = (val & 0x33) + ((val >> 2) & 0x33); |
194 | b9ef45ff | j_mayer | val = (val & 0x0f) + ((val >> 4) & 0x0f); |
195 | b9ef45ff | j_mayer | |
196 | b9ef45ff | j_mayer | return val;
|
197 | b9ef45ff | j_mayer | } |
198 | b9ef45ff | j_mayer | |
199 | facd2857 | Blue Swirl | static inline int ctpop16(uint16_t val) |
200 | b9ef45ff | j_mayer | { |
201 | b9ef45ff | j_mayer | val = (val & 0x5555) + ((val >> 1) & 0x5555); |
202 | b9ef45ff | j_mayer | val = (val & 0x3333) + ((val >> 2) & 0x3333); |
203 | b9ef45ff | j_mayer | val = (val & 0x0f0f) + ((val >> 4) & 0x0f0f); |
204 | b9ef45ff | j_mayer | val = (val & 0x00ff) + ((val >> 8) & 0x00ff); |
205 | b9ef45ff | j_mayer | |
206 | b9ef45ff | j_mayer | return val;
|
207 | b9ef45ff | j_mayer | } |
208 | b9ef45ff | j_mayer | |
209 | facd2857 | Blue Swirl | static inline int ctpop32(uint32_t val) |
210 | b9ef45ff | j_mayer | { |
211 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
212 | 7d019980 | aurel32 | return __builtin_popcount(val);
|
213 | 7d019980 | aurel32 | #else
|
214 | b9ef45ff | j_mayer | val = (val & 0x55555555) + ((val >> 1) & 0x55555555); |
215 | b9ef45ff | j_mayer | val = (val & 0x33333333) + ((val >> 2) & 0x33333333); |
216 | b9ef45ff | j_mayer | val = (val & 0x0f0f0f0f) + ((val >> 4) & 0x0f0f0f0f); |
217 | b9ef45ff | j_mayer | val = (val & 0x00ff00ff) + ((val >> 8) & 0x00ff00ff); |
218 | b9ef45ff | j_mayer | val = (val & 0x0000ffff) + ((val >> 16) & 0x0000ffff); |
219 | b9ef45ff | j_mayer | |
220 | b9ef45ff | j_mayer | return val;
|
221 | 7d019980 | aurel32 | #endif
|
222 | b9ef45ff | j_mayer | } |
223 | b9ef45ff | j_mayer | |
224 | facd2857 | Blue Swirl | static inline int ctpop64(uint64_t val) |
225 | b9ef45ff | j_mayer | { |
226 | bad5b1ec | aurel32 | #if QEMU_GNUC_PREREQ(3, 4) |
227 | 7d019980 | aurel32 | return __builtin_popcountll(val);
|
228 | 7d019980 | aurel32 | #else
|
229 | b9ef45ff | j_mayer | val = (val & 0x5555555555555555ULL) + ((val >> 1) & 0x5555555555555555ULL); |
230 | b9ef45ff | j_mayer | val = (val & 0x3333333333333333ULL) + ((val >> 2) & 0x3333333333333333ULL); |
231 | b9ef45ff | j_mayer | val = (val & 0x0f0f0f0f0f0f0f0fULL) + ((val >> 4) & 0x0f0f0f0f0f0f0f0fULL); |
232 | b9ef45ff | j_mayer | val = (val & 0x00ff00ff00ff00ffULL) + ((val >> 8) & 0x00ff00ff00ff00ffULL); |
233 | b9ef45ff | j_mayer | val = (val & 0x0000ffff0000ffffULL) + ((val >> 16) & 0x0000ffff0000ffffULL); |
234 | b9ef45ff | j_mayer | val = (val & 0x00000000ffffffffULL) + ((val >> 32) & 0x00000000ffffffffULL); |
235 | b9ef45ff | j_mayer | |
236 | b9ef45ff | j_mayer | return val;
|
237 | 7d019980 | aurel32 | #endif
|
238 | 3800af9e | ths | } |
239 | cb9c377f | Paolo Bonzini | |
240 | cb9c377f | Paolo Bonzini | #endif |