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
|
#include "benchmark_helpers.hpp"
#include <cmath>
#ifdef __SSE__
#include <xmmintrin.h>
#endif
using namespace std;
nova::aligned_array<float, 64> out, in, in2;
typedef float afloat __attribute__ ((__aligned__(16)));
void __noinline__ bench_1(float * out, float * in1, float in2, float slope, unsigned int n)
{
for (unsigned int i = 0; i != n; ++i)
{
out[i] = in1[i] + in2;
in2 += slope;
}
}
void __noinline__ bench_2(float * out, float * in1, float in2, float slope, unsigned int n)
{
for (unsigned int i = 0; i != n; i += 4)
{
out[i] = in1[i] + in2; in2 += slope;
out[i+1] = in1[i+1] + in2; in2 += slope;
out[i+2] = in1[i+2] + in2; in2 += slope;
out[i+3] = in1[i+3] + in2; in2 += slope;
}
}
#ifdef __SSE__
void __noinline__ bench_3(float * out, float * in1, float in2, float slope, unsigned int n)
{
__m128 arg2 = _mm_set_ps(in2, in2+slope, in2+slope+slope, in2+slope+slope+slope);
const __m128 vslope = _mm_set_ps1(slope+slope+slope+slope);
std::size_t loops = n / 4;
do {
__m128 arg1 = _mm_load_ps(in1);
__m128 result = _mm_add_ps(arg1, arg2);
arg2 = _mm_add_ps(arg2, vslope);
_mm_store_ps(out, result);
in1+=4;
out+=4;
} while (--loops);
}
void __noinline__ bench_3a(float * out, float * in1, float in2, float slope, unsigned int n)
{
__m128 arg2 = _mm_set_ps(in2, in2+slope, in2+ 2*slope, in2+3*slope);
const __m128 vslope = _mm_set_ps1(4 * slope);
std::size_t loops = n / 4;
do {
__m128 arg1 = _mm_load_ps(in1);
__m128 result = _mm_add_ps(arg1, arg2);
arg2 = _mm_add_ps(arg2, vslope);
_mm_store_ps(out, result);
in1+=4;
out+=4;
} while (--loops);
}
#endif
void __noinline__ bench_4(float * out, float * in1, float in2, float slope, unsigned int n)
{
for (unsigned int i = 0; i != n; ++i)
{
out[i] = in1[i] * in2;
in2 += slope;
}
}
void __noinline__ bench_5(float * out, float * in1, float in2, float slope, unsigned int n)
{
for (unsigned int i = 0; i != n; i += 4)
{
out[i] = in1[i] * in2; in2 += slope;
out[i+1] = in1[i+1] * in2; in2 += slope;
out[i+2] = in1[i+2] * in2; in2 += slope;
out[i+3] = in1[i+3] * in2; in2 += slope;
}
}
#ifdef __SSE__
void __noinline__ bench_6(float * out, float * in1, float in2, float slope, unsigned int n)
{
__m128 arg2 = _mm_set_ps(in2, in2+slope, in2+slope+slope, in2+slope+slope+slope);
const __m128 vslope = _mm_set_ps1(slope+slope+slope+slope);
std::size_t loops = n / 4;
do {
__m128 arg1 = _mm_load_ps(in1);
__m128 result = _mm_mul_ps(arg1, arg2);
arg2 = _mm_add_ps(arg2, vslope);
_mm_store_ps(out, result);
in1+=4;
out+=4;
} while (--loops);
}
void __noinline__ bench_6a(float * out, float * in1, float in2, float slope, unsigned int n)
{
__m128 arg2 = _mm_set_ps(in2, in2+slope, in2+2*slope, in2+3*slope);
const __m128 vslope = _mm_set_ps1(4*slope);
std::size_t loops = n / 4;
do {
__m128 arg1 = _mm_load_ps(in1);
__m128 result = _mm_mul_ps(arg1, arg2);
arg2 = _mm_add_ps(arg2, vslope);
_mm_store_ps(out, result);
in1+=4;
out+=4;
} while (--loops);
}
#endif
int main(void)
{
out.assign(0.f);
in.assign(0.2f);
in2.assign(0.3f);
const unsigned int iterations = 50000000;
run_bench(boost::bind(bench_1, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
run_bench(boost::bind(bench_2, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
#ifdef __SSE__
run_bench(boost::bind(bench_3, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
run_bench(boost::bind(bench_3a, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
#endif
run_bench(boost::bind(bench_4, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
run_bench(boost::bind(bench_5, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
#ifdef __SSE__
run_bench(boost::bind(bench_6, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
run_bench(boost::bind(bench_6a, out.begin(), in.begin(), 0.1f, 0.001f, 64), iterations);
#endif
}
|
