3445
by ths
Add sharable clz/clo inline functions and use them for the mips target. |
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 |
||
3513
by j_mayer
For consistency, move muls64 / mulu64 prototypes to host-utils.h |
26 |
#if defined(__x86_64__)
|
27 |
#define __HAVE_FAST_MULU64__
|
|
28 |
static always_inline void mulu64 (uint64_t *plow, uint64_t *phigh, |
|
29 |
uint64_t a, uint64_t b) |
|
30 |
{
|
|
31 |
__asm__ ("mul %0\n\t" |
|
32 |
: "=d" (*phigh), "=a" (*plow) |
|
33 |
: "a" (a), "0" (b)); |
|
34 |
}
|
|
35 |
#define __HAVE_FAST_MULS64__
|
|
36 |
static always_inline void muls64 (uint64_t *plow, uint64_t *phigh, |
|
37 |
int64_t a, int64_t b) |
|
38 |
{
|
|
39 |
__asm__ ("imul %0\n\t" |
|
40 |
: "=d" (*phigh), "=a" (*plow) |
|
41 |
: "a" (a), "0" (b)); |
|
42 |
}
|
|
43 |
#else
|
|
3527
by j_mayer
Fix muls64 prototype to match the actual implementation. |
44 |
void muls64(uint64_t *phigh, uint64_t *plow, int64_t a, int64_t b); |
3513
by j_mayer
For consistency, move muls64 / mulu64 prototypes to host-utils.h |
45 |
void mulu64(uint64_t *phigh, uint64_t *plow, uint64_t a, uint64_t b); |
46 |
#endif
|
|
47 |
||
3445
by ths
Add sharable clz/clo inline functions and use them for the mips target. |
48 |
/* Note that some of those functions may end up calling libgcc functions,
|
49 |
depending on the host machine. It is up to the target emulation to
|
|
50 |
cope with that. */
|
|
51 |
||
52 |
/* Binary search for leading zeros. */
|
|
53 |
||
54 |
static always_inline int clz32(uint32_t val) |
|
55 |
{
|
|
56 |
int cnt = 0; |
|
57 |
||
58 |
if (!(val & 0xFFFF0000U)) { |
|
59 |
cnt += 16; |
|
60 |
val <<= 16; |
|
61 |
}
|
|
62 |
if (!(val & 0xFF000000U)) { |
|
63 |
cnt += 8; |
|
64 |
val <<= 8; |
|
65 |
}
|
|
66 |
if (!(val & 0xF0000000U)) { |
|
67 |
cnt += 4; |
|
68 |
val <<= 4; |
|
69 |
}
|
|
70 |
if (!(val & 0xC0000000U)) { |
|
71 |
cnt += 2; |
|
72 |
val <<= 2; |
|
73 |
}
|
|
74 |
if (!(val & 0x80000000U)) { |
|
75 |
cnt++; |
|
76 |
val <<= 1; |
|
77 |
}
|
|
78 |
if (!(val & 0x80000000U)) { |
|
79 |
cnt++; |
|
80 |
}
|
|
81 |
return cnt; |
|
82 |
}
|
|
83 |
||
84 |
static always_inline int clo32(uint32_t val) |
|
85 |
{
|
|
86 |
return clz32(~val); |
|
87 |
}
|
|
88 |
||
89 |
static always_inline int clz64(uint64_t val) |
|
90 |
{
|
|
91 |
int cnt = 0; |
|
92 |
||
3513
by j_mayer
For consistency, move muls64 / mulu64 prototypes to host-utils.h |
93 |
if (!(val >> 32)) { |
3445
by ths
Add sharable clz/clo inline functions and use them for the mips target. |
94 |
cnt += 32; |
3513
by j_mayer
For consistency, move muls64 / mulu64 prototypes to host-utils.h |
95 |
} else { |
96 |
val >>= 32; |
|
97 |
}
|
|
98 |
||
99 |
return cnt + clz32(val); |
|
3445
by ths
Add sharable clz/clo inline functions and use them for the mips target. |
100 |
}
|
101 |
||
102 |
static always_inline int clo64(uint64_t val) |
|
103 |
{
|
|
104 |
return clz64(~val); |
|
105 |
}
|
|
3455
by j_mayer
Add shared ctz32, cto32, ctz64, cto64, |
106 |
|
107 |
static always_inline int ctz32 (uint32_t val) |
|
108 |
{
|
|
109 |
int cnt; |
|
110 |
||
111 |
cnt = 0; |
|
112 |
if (!(val & 0x0000FFFFUL)) { |
|
113 |
cnt += 16; |
|
114 |
val >>= 16; |
|
115 |
}
|
|
116 |
if (!(val & 0x000000FFUL)) { |
|
117 |
cnt += 8; |
|
118 |
val >>= 8; |
|
119 |
}
|
|
120 |
if (!(val & 0x0000000FUL)) { |
|
121 |
cnt += 4; |
|
122 |
val >>= 4; |
|
123 |
}
|
|
124 |
if (!(val & 0x00000003UL)) { |
|
125 |
cnt += 2; |
|
126 |
val >>= 2; |
|
127 |
}
|
|
128 |
if (!(val & 0x00000001UL)) { |
|
129 |
cnt++; |
|
130 |
val >>= 1; |
|
131 |
}
|
|
132 |
if (!(val & 0x00000001UL)) { |
|
133 |
cnt++; |
|
134 |
}
|
|
135 |
||
136 |
return cnt; |
|
137 |
}
|
|
138 |
||
139 |
static always_inline int cto32 (uint32_t val) |
|
140 |
{
|
|
141 |
return ctz32(~val); |
|
142 |
}
|
|
143 |
||
144 |
static always_inline int ctz64 (uint64_t val) |
|
145 |
{
|
|
146 |
int cnt; |
|
147 |
||
148 |
cnt = 0; |
|
149 |
if (!((uint32_t)val)) { |
|
150 |
cnt += 32; |
|
151 |
val >>= 32; |
|
152 |
}
|
|
153 |
||
154 |
return cnt + ctz32(val); |
|
155 |
}
|
|
156 |
||
157 |
static always_inline int cto64 (uint64_t val) |
|
158 |
{
|
|
159 |
return ctz64(~val); |
|
160 |
}
|
|
161 |
||
162 |
static always_inline int ctpop8 (uint8_t val) |
|
163 |
{
|
|
164 |
val = (val & 0x55) + ((val >> 1) & 0x55); |
|
165 |
val = (val & 0x33) + ((val >> 2) & 0x33); |
|
166 |
val = (val & 0x0f) + ((val >> 4) & 0x0f); |
|
167 |
||
168 |
return val; |
|
169 |
}
|
|
170 |
||
171 |
static always_inline int ctpop16 (uint16_t val) |
|
172 |
{
|
|
173 |
val = (val & 0x5555) + ((val >> 1) & 0x5555); |
|
174 |
val = (val & 0x3333) + ((val >> 2) & 0x3333); |
|
175 |
val = (val & 0x0f0f) + ((val >> 4) & 0x0f0f); |
|
176 |
val = (val & 0x00ff) + ((val >> 8) & 0x00ff); |
|
177 |
||
178 |
return val; |
|
179 |
}
|
|
180 |
||
181 |
static always_inline int ctpop32 (uint32_t val) |
|
182 |
{
|
|
183 |
val = (val & 0x55555555) + ((val >> 1) & 0x55555555); |
|
184 |
val = (val & 0x33333333) + ((val >> 2) & 0x33333333); |
|
185 |
val = (val & 0x0f0f0f0f) + ((val >> 4) & 0x0f0f0f0f); |
|
186 |
val = (val & 0x00ff00ff) + ((val >> 8) & 0x00ff00ff); |
|
187 |
val = (val & 0x0000ffff) + ((val >> 16) & 0x0000ffff); |
|
188 |
||
189 |
return val; |
|
190 |
}
|
|
191 |
||
192 |
static always_inline int ctpop64 (uint64_t val) |
|
193 |
{
|
|
194 |
val = (val & 0x5555555555555555ULL) + ((val >> 1) & 0x5555555555555555ULL); |
|
195 |
val = (val & 0x3333333333333333ULL) + ((val >> 2) & 0x3333333333333333ULL); |
|
196 |
val = (val & 0x0f0f0f0f0f0f0f0fULL) + ((val >> 4) & 0x0f0f0f0f0f0f0f0fULL); |
|
197 |
val = (val & 0x00ff00ff00ff00ffULL) + ((val >> 8) & 0x00ff00ff00ff00ffULL); |
|
198 |
val = (val & 0x0000ffff0000ffffULL) + ((val >> 16) & 0x0000ffff0000ffffULL); |
|
199 |
val = (val & 0x00000000ffffffffULL) + ((val >> 32) & 0x00000000ffffffffULL); |
|
200 |
||
201 |
return val; |
|
3827
by ths
Code formatting. |
202 |
}
|