root / host-utils.h @ f97572e5
History | View | Annotate | Download (5.1 kB)
1 |
/*
|
---|---|
2 |
* Utility compute operations used by translated code.
|
3 |
*
|
4 |
* Copyright (c) 2007 Thiemo Seufer
|
5 |
* Copyright (c) 2007 Jocelyn Mayer
|
6 |
*
|
7 |
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
8 |
* of this software and associated documentation files (the "Software"), to deal
|
9 |
* in the Software without restriction, including without limitation the rights
|
10 |
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
11 |
* copies of the Software, and to permit persons to whom the Software is
|
12 |
* furnished to do so, subject to the following conditions:
|
13 |
*
|
14 |
* The above copyright notice and this permission notice shall be included in
|
15 |
* all copies or substantial portions of the Software.
|
16 |
*
|
17 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
18 |
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
19 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
20 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
21 |
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
22 |
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
23 |
* THE SOFTWARE.
|
24 |
*/
|
25 |
|
26 |
#include "osdep.h" |
27 |
|
28 |
#if defined(__x86_64__)
|
29 |
#define __HAVE_FAST_MULU64__
|
30 |
static always_inline void mulu64 (uint64_t *plow, uint64_t *phigh, |
31 |
uint64_t a, uint64_t b) |
32 |
{ |
33 |
__asm__ ("mul %0\n\t"
|
34 |
: "=d" (*phigh), "=a" (*plow) |
35 |
: "a" (a), "0" (b)); |
36 |
} |
37 |
#define __HAVE_FAST_MULS64__
|
38 |
static always_inline void muls64 (uint64_t *plow, uint64_t *phigh, |
39 |
int64_t a, int64_t b) |
40 |
{ |
41 |
__asm__ ("imul %0\n\t"
|
42 |
: "=d" (*phigh), "=a" (*plow) |
43 |
: "a" (a), "0" (b)); |
44 |
} |
45 |
#else
|
46 |
void muls64(uint64_t *phigh, uint64_t *plow, int64_t a, int64_t b);
|
47 |
void mulu64(uint64_t *phigh, uint64_t *plow, uint64_t a, uint64_t b);
|
48 |
#endif
|
49 |
|
50 |
/* Note that some of those functions may end up calling libgcc functions,
|
51 |
depending on the host machine. It is up to the target emulation to
|
52 |
cope with that. */
|
53 |
|
54 |
/* Binary search for leading zeros. */
|
55 |
|
56 |
static always_inline int clz32(uint32_t val) |
57 |
{ |
58 |
int cnt = 0; |
59 |
|
60 |
if (!(val & 0xFFFF0000U)) { |
61 |
cnt += 16;
|
62 |
val <<= 16;
|
63 |
} |
64 |
if (!(val & 0xFF000000U)) { |
65 |
cnt += 8;
|
66 |
val <<= 8;
|
67 |
} |
68 |
if (!(val & 0xF0000000U)) { |
69 |
cnt += 4;
|
70 |
val <<= 4;
|
71 |
} |
72 |
if (!(val & 0xC0000000U)) { |
73 |
cnt += 2;
|
74 |
val <<= 2;
|
75 |
} |
76 |
if (!(val & 0x80000000U)) { |
77 |
cnt++; |
78 |
val <<= 1;
|
79 |
} |
80 |
if (!(val & 0x80000000U)) { |
81 |
cnt++; |
82 |
} |
83 |
return cnt;
|
84 |
} |
85 |
|
86 |
static always_inline int clo32(uint32_t val) |
87 |
{ |
88 |
return clz32(~val);
|
89 |
} |
90 |
|
91 |
static always_inline int clz64(uint64_t val) |
92 |
{ |
93 |
int cnt = 0; |
94 |
|
95 |
if (!(val >> 32)) { |
96 |
cnt += 32;
|
97 |
} else {
|
98 |
val >>= 32;
|
99 |
} |
100 |
|
101 |
return cnt + clz32(val);
|
102 |
} |
103 |
|
104 |
static always_inline int clo64(uint64_t val) |
105 |
{ |
106 |
return clz64(~val);
|
107 |
} |
108 |
|
109 |
static always_inline int ctz32 (uint32_t val) |
110 |
{ |
111 |
int cnt;
|
112 |
|
113 |
cnt = 0;
|
114 |
if (!(val & 0x0000FFFFUL)) { |
115 |
cnt += 16;
|
116 |
val >>= 16;
|
117 |
} |
118 |
if (!(val & 0x000000FFUL)) { |
119 |
cnt += 8;
|
120 |
val >>= 8;
|
121 |
} |
122 |
if (!(val & 0x0000000FUL)) { |
123 |
cnt += 4;
|
124 |
val >>= 4;
|
125 |
} |
126 |
if (!(val & 0x00000003UL)) { |
127 |
cnt += 2;
|
128 |
val >>= 2;
|
129 |
} |
130 |
if (!(val & 0x00000001UL)) { |
131 |
cnt++; |
132 |
val >>= 1;
|
133 |
} |
134 |
if (!(val & 0x00000001UL)) { |
135 |
cnt++; |
136 |
} |
137 |
|
138 |
return cnt;
|
139 |
} |
140 |
|
141 |
static always_inline int cto32 (uint32_t val) |
142 |
{ |
143 |
return ctz32(~val);
|
144 |
} |
145 |
|
146 |
static always_inline int ctz64 (uint64_t val) |
147 |
{ |
148 |
int cnt;
|
149 |
|
150 |
cnt = 0;
|
151 |
if (!((uint32_t)val)) {
|
152 |
cnt += 32;
|
153 |
val >>= 32;
|
154 |
} |
155 |
|
156 |
return cnt + ctz32(val);
|
157 |
} |
158 |
|
159 |
static always_inline int cto64 (uint64_t val) |
160 |
{ |
161 |
return ctz64(~val);
|
162 |
} |
163 |
|
164 |
static always_inline int ctpop8 (uint8_t val) |
165 |
{ |
166 |
val = (val & 0x55) + ((val >> 1) & 0x55); |
167 |
val = (val & 0x33) + ((val >> 2) & 0x33); |
168 |
val = (val & 0x0f) + ((val >> 4) & 0x0f); |
169 |
|
170 |
return val;
|
171 |
} |
172 |
|
173 |
static always_inline int ctpop16 (uint16_t val) |
174 |
{ |
175 |
val = (val & 0x5555) + ((val >> 1) & 0x5555); |
176 |
val = (val & 0x3333) + ((val >> 2) & 0x3333); |
177 |
val = (val & 0x0f0f) + ((val >> 4) & 0x0f0f); |
178 |
val = (val & 0x00ff) + ((val >> 8) & 0x00ff); |
179 |
|
180 |
return val;
|
181 |
} |
182 |
|
183 |
static always_inline int ctpop32 (uint32_t val) |
184 |
{ |
185 |
val = (val & 0x55555555) + ((val >> 1) & 0x55555555); |
186 |
val = (val & 0x33333333) + ((val >> 2) & 0x33333333); |
187 |
val = (val & 0x0f0f0f0f) + ((val >> 4) & 0x0f0f0f0f); |
188 |
val = (val & 0x00ff00ff) + ((val >> 8) & 0x00ff00ff); |
189 |
val = (val & 0x0000ffff) + ((val >> 16) & 0x0000ffff); |
190 |
|
191 |
return val;
|
192 |
} |
193 |
|
194 |
static always_inline int ctpop64 (uint64_t val) |
195 |
{ |
196 |
val = (val & 0x5555555555555555ULL) + ((val >> 1) & 0x5555555555555555ULL); |
197 |
val = (val & 0x3333333333333333ULL) + ((val >> 2) & 0x3333333333333333ULL); |
198 |
val = (val & 0x0f0f0f0f0f0f0f0fULL) + ((val >> 4) & 0x0f0f0f0f0f0f0f0fULL); |
199 |
val = (val & 0x00ff00ff00ff00ffULL) + ((val >> 8) & 0x00ff00ff00ff00ffULL); |
200 |
val = (val & 0x0000ffff0000ffffULL) + ((val >> 16) & 0x0000ffff0000ffffULL); |
201 |
val = (val & 0x00000000ffffffffULL) + ((val >> 32) & 0x00000000ffffffffULL); |
202 |
|
203 |
return val;
|
204 |
} |