1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
|
#ifndef AWS_COMMON_MATH_GCC_X64_ASM_INL
#define AWS_COMMON_MATH_GCC_X64_ASM_INL
/**
* Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
* SPDX-License-Identifier: Apache-2.0.
*/
/*
* This header is already included, but include it again to make editor
* highlighting happier.
*/
#include <aws/common/common.h>
#include <aws/common/math.h>
/* clang-format off */
AWS_EXTERN_C_BEGIN
/**
* Multiplies a * b. If the result overflows, returns 2^64 - 1.
*/
AWS_STATIC_IMPL uint64_t aws_mul_u64_saturating(uint64_t a, uint64_t b) {
/* We can use inline assembly to do this efficiently on x86-64 and x86.
we specify rdx as an output, rather than a clobber, because we want to
allow it to be allocated as an input register */
uint64_t rdx;
__asm__("mulq %q[arg2]\n" /* rax * b, result is in RDX:RAX, OF=CF=(RDX != 0) */
"cmovc %q[saturate], %%rax\n"
: /* in/out: %rax = a, out: rdx (ignored) */ "+&a"(a), "=&d"(rdx)
: /* in: register only */ [arg2] "r"(b),
/* in: saturation value (reg/memory) */ [saturate] "rm"(~0LL)
: /* clobbers: cc */ "cc");
(void)rdx; /* suppress unused warnings */
return a;
}
/**
* If a * b overflows, returns AWS_OP_ERR; otherwise multiplies
* a * b, returns the result in *r, and returns AWS_OP_SUCCESS.
*/
AWS_STATIC_IMPL int aws_mul_u64_checked(uint64_t a, uint64_t b, uint64_t *r) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
char flag;
uint64_t result = a;
__asm__("mulq %q[arg2]\n" /* rax * b, result is in RDX:RAX, OF=CF=(RDX != 0) */
"seto %[flag]\n" /* flag = overflow_bit */
: /* in/out: %rax (first arg & result), %d (flag) */ "+&a"(result), [flag] "=&d"(flag)
: /* in: reg for 2nd operand */
[arg2] "r"(b)
: /* clobbers: cc (d is used for flag so no need to clobber)*/ "cc");
*r = result;
if (flag) {
return aws_raise_error(AWS_ERROR_OVERFLOW_DETECTED);
}
return AWS_OP_SUCCESS;
}
/**
* Multiplies a * b. If the result overflows, returns 2^32 - 1.
*/
AWS_STATIC_IMPL uint32_t aws_mul_u32_saturating(uint32_t a, uint32_t b) {
/* We can use inline assembly to do this efficiently on x86-64 and x86.
we specify edx as an output, rather than a clobber, because we want to
allow it to be allocated as an input register */
uint32_t edx;
__asm__("mull %k[arg2]\n" /* eax * b, result is in EDX:EAX, OF=CF=(EDX != 0) */
/* cmov isn't guaranteed to be available on x86-32 */
"jnc .1f%=\n"
"mov $0xFFFFFFFF, %%eax\n"
".1f%=:"
: /* in/out: %eax = result/a, out: edx (ignored) */ "+&a"(a), "=&d"(edx)
: /* in: operand 2 in reg */ [arg2] "r"(b)
: /* clobbers: cc */ "cc");
(void)edx; /* suppress unused warnings */
return a;
}
/**
* If a * b overflows, returns AWS_OP_ERR; otherwise multiplies
* a * b, returns the result in *r, and returns AWS_OP_SUCCESS.
*/
AWS_STATIC_IMPL int aws_mul_u32_checked(uint32_t a, uint32_t b, uint32_t *r) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
uint32_t result = a;
char flag;
/**
* Note: We use SETNO which only takes a byte register. To make this easy,
* we'll write it to dl (which we throw away anyway) and mask off the high bits.
*/
__asm__("mull %k[arg2]\n" /* eax * b, result is in EDX:EAX, OF=CF=(EDX != 0) */
"seto %[flag]\n" /* flag = overflow_bit */
: /* in/out: %eax (first arg & result), %d (flag) */ "+&a"(result), [flag] "=&d"(flag)
: /* in: reg for 2nd operand */
[arg2] "r"(b)
: /* clobbers: cc (d is used for flag so no need to clobber)*/ "cc");
*r = result;
if (flag) {
return aws_raise_error(AWS_ERROR_OVERFLOW_DETECTED);
}
return AWS_OP_SUCCESS;
}
/**
* If a + b overflows, returns AWS_OP_ERR; otherwise adds
* a + b, returns the result in *r, and returns AWS_OP_SUCCESS.
*/
AWS_STATIC_IMPL int aws_add_u64_checked(uint64_t a, uint64_t b, uint64_t *r) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
char flag;
__asm__("addq %[argb], %[arga]\n" /* [arga] = [arga] + [argb] */
"setc %[flag]\n" /* [flag] = 1 if overflow, 0 otherwise */
: /* in/out: */ [arga] "+r"(a), [flag] "=&r"(flag)
: /* in: */ [argb] "r"(b)
: /* clobbers: */ "cc");
*r = a;
if (flag) {
return aws_raise_error(AWS_ERROR_OVERFLOW_DETECTED);
}
return AWS_OP_SUCCESS;
}
/**
* Adds a + b. If the result overflows, returns 2^64 - 1.
*/
AWS_STATIC_IMPL uint64_t aws_add_u64_saturating(uint64_t a, uint64_t b) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
__asm__("addq %[arg1], %[arg2]\n" /* [arga] = [arga] + [argb] */
"cmovc %q[saturate], %[arg2]\n"
: /* in/out: %rax = a, out: rdx (ignored) */ [arg2] "+r"(b)
: /* in: register only */ [arg1] "r"(a),
/* in: saturation value (reg/memory) */ [saturate] "rm"(~0LL)
: /* clobbers: cc */ "cc");
return b;
}
/**
* If a + b overflows, returns AWS_OP_ERR; otherwise adds
* a + b, returns the result in *r, and returns AWS_OP_SUCCESS.
*/
AWS_STATIC_IMPL int aws_add_u32_checked(uint32_t a, uint32_t b, uint32_t *r) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
char flag;
__asm__("addl %[argb], %[arga]\n" /* [arga] = [arga] + [argb] */
"setc %[flag]\n" /* [flag] = 1 if overflow, 0 otherwise */
: /* in/out: */ [arga] "+r"(a), [flag] "=&r"(flag)
: /* in: */ [argb] "r"(b)
: /* clobbers: */ "cc");
*r = a;
if (flag) {
return aws_raise_error(AWS_ERROR_OVERFLOW_DETECTED);
}
return AWS_OP_SUCCESS;
}
/**
* Adds a + b. If the result overflows, returns 2^32 - 1.
*/
AWS_STATIC_IMPL uint32_t aws_add_u32_saturating(uint32_t a, uint32_t b) {
/* We can use inline assembly to do this efficiently on x86-64 and x86. */
__asm__("addl %[arg1], %[arg2]\n" /* [arga] = [arga] + [argb] */
/* cmov isn't guaranteed to be available on x86-32 */
"jnc .1f%=\n"
"mov $0xFFFFFFFF, %%eax\n"
".1f%=:"
: /* in/out: %rax = a, out: rdx (ignored) */ [arg2] "+a"(b)
: /* in: register only */ [arg1] "r"(a)
: /* clobbers: cc */ "cc");
return b;
}
AWS_EXTERN_C_END
/* clang-format on */
#endif /* AWS_COMMON_MATH_GCC_X64_ASM_INL */
|
