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
|
// Copyright 2017 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef BASE_NUMERICS_SAFE_MATH_ARM_IMPL_H_
#define BASE_NUMERICS_SAFE_MATH_ARM_IMPL_H_
#include <cassert>
#include <type_traits>
#include "anglebase/numerics/safe_conversions.h"
namespace angle
{
namespace base
{
namespace internal
{
template <typename T, typename U>
struct CheckedMulFastAsmOp
{
static const bool is_supported = FastIntegerArithmeticPromotion<T, U>::is_contained;
// The following is much more efficient than the Clang and GCC builtins for
// performing overflow-checked multiplication when a twice wider type is
// available. The below compiles down to 2-3 instructions, depending on the
// width of the types in use.
// As an example, an int32_t multiply compiles to:
// smull r0, r1, r0, r1
// cmp r1, r1, asr #31
// And an int16_t multiply compiles to:
// smulbb r1, r1, r0
// asr r2, r1, #16
// cmp r2, r1, asr #15
template <typename V>
__attribute__((always_inline)) static bool Do(T x, U y, V *result)
{
using Promotion = typename FastIntegerArithmeticPromotion<T, U>::type;
Promotion presult;
presult = static_cast<Promotion>(x) * static_cast<Promotion>(y);
if (!IsValueInRangeForNumericType<V>(presult))
return false;
*result = static_cast<V>(presult);
return true;
}
};
template <typename T, typename U>
struct ClampedAddFastAsmOp
{
static const bool is_supported =
BigEnoughPromotion<T, U>::is_contained &&
IsTypeInRangeForNumericType<int32_t, typename BigEnoughPromotion<T, U>::type>::value;
template <typename V>
__attribute__((always_inline)) static V Do(T x, U y)
{
// This will get promoted to an int, so let the compiler do whatever is
// clever and rely on the saturated cast to bounds check.
if (IsIntegerArithmeticSafe<int, T, U>::value)
return saturated_cast<V>(x + y);
int32_t result;
int32_t x_i32 = checked_cast<int32_t>(x);
int32_t y_i32 = checked_cast<int32_t>(y);
asm("qadd %[result], %[first], %[second]"
: [result] "=r"(result)
: [first] "r"(x_i32), [second] "r"(y_i32));
return saturated_cast<V>(result);
}
};
template <typename T, typename U>
struct ClampedSubFastAsmOp
{
static const bool is_supported =
BigEnoughPromotion<T, U>::is_contained &&
IsTypeInRangeForNumericType<int32_t, typename BigEnoughPromotion<T, U>::type>::value;
template <typename V>
__attribute__((always_inline)) static V Do(T x, U y)
{
// This will get promoted to an int, so let the compiler do whatever is
// clever and rely on the saturated cast to bounds check.
if (IsIntegerArithmeticSafe<int, T, U>::value)
return saturated_cast<V>(x - y);
int32_t result;
int32_t x_i32 = checked_cast<int32_t>(x);
int32_t y_i32 = checked_cast<int32_t>(y);
asm("qsub %[result], %[first], %[second]"
: [result] "=r"(result)
: [first] "r"(x_i32), [second] "r"(y_i32));
return saturated_cast<V>(result);
}
};
template <typename T, typename U>
struct ClampedMulFastAsmOp
{
static const bool is_supported = CheckedMulFastAsmOp<T, U>::is_supported;
template <typename V>
__attribute__((always_inline)) static V Do(T x, U y)
{
// Use the CheckedMulFastAsmOp for full-width 32-bit values, because
// it's fewer instructions than promoting and then saturating.
if (!IsIntegerArithmeticSafe<int32_t, T, U>::value &&
!IsIntegerArithmeticSafe<uint32_t, T, U>::value)
{
V result;
return CheckedMulFastAsmOp<T, U>::Do(x, y, &result)
? result
: CommonMaxOrMin<V>(IsValueNegative(x) ^ IsValueNegative(y));
}
assert((FastIntegerArithmeticPromotion<T, U>::is_contained));
using Promotion = typename FastIntegerArithmeticPromotion<T, U>::type;
return saturated_cast<V>(static_cast<Promotion>(x) * static_cast<Promotion>(y));
}
};
} // namespace internal
} // namespace base
} // namespace angle
#endif // BASE_NUMERICS_SAFE_MATH_ARM_IMPL_H_
|
