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
|
#include "benchmark_helpers.hpp"
#include "../simd_unary_arithmetic.hpp"
#define sc_abs(a) ((a)>=0?(a) : -(a))
nova::aligned_array<float, 1024> in;
nova::aligned_array<float, 1024> out;
void __noinline__ bench_1(unsigned int numSamples)
{
for (unsigned int i = 0; i != numSamples; ++i)
out[i] = sc_abs(in[i]);
}
void __noinline__ bench_2(unsigned int numSamples)
{
for (unsigned int i = 0; i != numSamples; ++i)
out[i] = std::fabs(in[i]);
}
void __noinline__ bench_3(unsigned int numSamples)
{
typedef float afloat __attribute__ ((__aligned__(16)));
float * __restrict__ o = out.data();
float * __restrict__ i = in.data();
for (unsigned int n = 0; n != numSamples; n += 8)
{
o[n] = std::fabs(i[n]);
o[n+1] = std::fabs(i[n+1]);
o[n+2] = std::fabs(i[n+2]);
o[n+3] = std::fabs(i[n+3]);
o[n+4] = std::fabs(i[n+4]);
o[n+5] = std::fabs(i[n+5]);
o[n+6] = std::fabs(i[n+6]);
o[n+7] = std::fabs(i[n+7]);
}
}
using namespace nova;
void __noinline__ bench_4(unsigned int numSamples)
{
abs_vec_simd<64>(out.begin(), in.begin());
}
void __noinline__ bench_5(unsigned int numSamples)
{
abs_vec_simd(out.begin(), in.begin(), numSamples);
}
int main(void)
{
/* touch buffers */
out.assign(0);
fill_container(in);
const unsigned int iterations = 50000000;
run_bench(boost::bind(bench_1, 64), iterations);
run_bench(boost::bind(bench_2, 64), iterations);
run_bench(boost::bind(bench_3, 64), iterations);
run_bench(boost::bind(bench_4, 64), iterations);
run_bench(boost::bind(bench_5, 64), iterations);
}
|
