Repository navigation
Expand file tree
/
Copy pathmandelbrot.cc
More file actions
250 lines (206 loc) · 9.39 KB
/
Copy pathmandelbrot.cc
File metadata and controls
250 lines (206 loc) · 9.39 KB
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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
#include <cmath>
#include <cstdio>
#include <cstdlib>
#include <string>
#include "timer.h"
#include <VecCore/VecCore>
using namespace vecCore;
extern void write_png(const char *filename, float *data, size_t nx, size_t ny);
// Normalized (smooth) iteration count. For an escaped point the fractional
// part is recovered from how far |z| overshot the bailout radius, which turns
// the integer escape-time bands into a continuous value suitable for smooth
// gradient coloring. Interior points (|z| still bounded) get a sentinel < 0.
static inline float smooth_value(double zr, double zi, long n)
{
double mag2 = zr * zr + zi * zi;
if (mag2 < 4.0)
return -1.0f;
return float(n + 1.0 - std::log2(0.5 * std::log(mag2)));
}
template<typename T>
void mandelbrot(T xmin, T xmax, size_t nx,
T ymin, T ymax, size_t ny,
size_t max_iter, float *image)
{
T dx = (xmax - xmin) / T(nx);
T dy = (ymax - ymin) / T(ny);
for (size_t i = 0; i < nx; ++i) {
for (size_t j = 0; j < ny; ++j) {
size_t k = 0;
T x = xmin + T(i) * dx, cr = x, zr = x;
T y = ymin + T(j) * dy, ci = y, zi = y;
do {
x = zr*zr - zi*zi + cr;
y = T(2.0) * zr*zi + ci;
zr = x;
zi = y;
} while (++k < max_iter && (zr*zr + zi*zi < T(4.0)));
image[ny*i + j] = smooth_value(zr, zi, k);
}
}
}
#ifdef __AVX2__
void mandelbrot_avx2(float xmin, float xmax, size_t nx,
float ymin, float ymax, size_t ny,
size_t max_iter, float *image)
{
__m256 dx = _mm256_set1_ps((xmax - xmin) / float(nx));
__m256 dy = _mm256_set1_ps((ymax - ymin) / float(ny));
__m256 dyv = _mm256_mul_ps(dy, _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f));
for (size_t i = 0; i < nx; ++i) {
for (size_t j = 0; j < ny; j += 8) {
size_t k = 0;
#ifdef __FMA__
__m256 x = _mm256_fmadd_ps(_mm256_set1_ps(i), dx, _mm256_set1_ps(xmin));
__m256 y = _mm256_fmadd_ps(_mm256_set1_ps(j), dy, _mm256_add_ps(_mm256_set1_ps(ymin), dyv));
#else
__m256 x = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(i), dx), _mm256_set1_ps(xmin));
__m256 y = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(j), dy), _mm256_add_ps(_mm256_set1_ps(ymin), dyv));
#endif
__m256 cr = x, zr = x, ci = y, zi = y;
__m256i kv = _mm256_setzero_si256();
__m256 mask = _mm256_cmp_ps(_mm256_setzero_ps(), _mm256_setzero_ps(), _CMP_EQ_OQ);
do {
__m256 zr2 = _mm256_mul_ps(zr, zr);
__m256 zi2 = _mm256_mul_ps(zi, zi);
__m256 zri = _mm256_mul_ps(zr, zi);
x = _mm256_add_ps(_mm256_sub_ps(zr2,zi2), cr);
#ifdef __FMA__
y = _mm256_fmadd_ps(_mm256_set1_ps(2.0f), zri, ci);
#else
y = _mm256_add_ps(_mm256_mul_ps(_mm256_set1_ps(2.0f), zri), ci);
#endif
zr = _mm256_blendv_ps(zr, x, mask);
zi = _mm256_blendv_ps(zi, y, mask);
kv = _mm256_blendv_epi8(kv, _mm256_set1_epi32(++k), _mm256_castps_si256(mask));
mask = _mm256_cmp_ps(_mm256_add_ps(_mm256_mul_ps(zr, zr), _mm256_mul_ps(zi, zi)),
_mm256_set1_ps(4.0f), _CMP_NGE_UQ);
} while (k < max_iter && !_mm256_testz_ps(mask, mask));
for (k = 0; k < 8; ++k)
image[ny*i + j + k] = smooth_value(((float*)&zr)[k], ((float*)&zi)[k], ((int*)&kv)[k]);
}
}
}
#endif
template<typename T>
void mandelbrot_v(Scalar<T> xmin, Scalar<T> xmax, size_t nx,
Scalar<T> ymin, Scalar<T> ymax, size_t ny,
Scalar<Index<T>> max_iter, float *image)
{
T iota;
for (size_t i = 0; i < VectorSize<T>(); ++i)
Set<T>(iota, i, i);
T dx = T(xmax - xmin) / T((Scalar<T>)nx);
T dy = T(ymax - ymin) / T((Scalar<T>)ny), dyv = iota * dy;
for (size_t i = 0; i < nx; ++i) {
for (size_t j = 0; j < ny; j += VectorSize<T>()) {
Scalar<Index<T>> k(0);
T x = xmin + T((Scalar<T>)i) * dx, cr = x, zr = x;
T y = ymin + T((Scalar<T>)j) * dy + dyv, ci = y, zi = y;
Index<T> kv(0);
Mask<T> m(true);
do {
x = zr*zr - zi*zi + cr;
y = T((Scalar<T>)2.0) * zr*zi + ci;
MaskedAssign<T>(zr, m, x);
MaskedAssign<T>(zi, m, y);
MaskedAssign<Index<T>>(kv, m, ++k);
m = zr*zr + zi*zi < T((Scalar<T>)4.0);
} while (k < max_iter && !MaskEmpty(m));
for (size_t k = 0; k < VectorSize<T>(); ++k)
image[ny*i + j + k] = smooth_value(Get(zr, k), Get(zi, k), Get(kv, k));
}
}
}
template<typename T>
void bench_mandelbrot(T xmin, T xmax, size_t nx, T ymin, T ymax, size_t ny,
int max_iter, float *image, const char *backend)
{
std::string filename = "mandelbrot_" + std::string(backend) + ".png";
Timer<milliseconds> timer;
mandelbrot<T>(xmin, xmax, nx, ymin, ymax, ny, max_iter, image);
printf("%25s: %7.2lf ms\n", backend, timer.Elapsed());
write_png(filename.c_str(), image, nx, ny);
}
#ifdef __AVX2__
void bench_mandelbrot_avx2(float xmin, float xmax, size_t nx,
float ymin, float ymax, size_t ny,
int max_iter, float *image,
const char *backend)
{
std::string filename = "mandelbrot_" + std::string(backend) + ".png";
Timer<milliseconds> timer;
mandelbrot_avx2(xmin, xmax, nx, ymin, ymax, ny, max_iter, image);
printf("%25s: %7.2lf ms\n", backend, timer.Elapsed());
write_png(filename.c_str(), image, nx, ny);
}
#endif
template<typename T>
void bench_mandelbrot_v(Scalar<T> xmin, Scalar<T> xmax, size_t nx,
Scalar<T> ymin, Scalar<T> ymax, size_t ny,
int max_iter, float *image, const char *backend)
{
std::string filename = "mandelbrot_" + std::string(backend) + ".png";
Timer<milliseconds> timer;
mandelbrot_v<T>(xmin, xmax, nx, ymin, ymax, ny, max_iter, image);
printf("%25s: %7.2lf ms\n", backend, timer.Elapsed());
write_png(filename.c_str(), image, nx, ny);
}
int main()
{
double xmin = -2.1, xmax = 1.1;
double ymin = -1.35, ymax = 1.35;
size_t nx = 1024, ny = 864, max_iter = 500;
float *image = new float[nx*ny];
/* single precision */
bench_mandelbrot<float>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "float");
bench_mandelbrot_v<float>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "float_v");
#ifdef __AVX2__
bench_mandelbrot_avx2(xmin, xmax, nx, ymin, ymax, ny, max_iter, image, "float_avx2");
#endif
#ifdef VECCORE_ENABLE_VC
bench_mandelbrot_v<backend::VcVector::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "float_vc");
#endif
#ifdef VECCORE_ENABLE_UMESIMD
bench_mandelbrot_v<backend::UMESimd::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "float_umesimd");
#endif
#ifdef VECCORE_ENABLE_STD_SIMD
bench_mandelbrot_v<backend::SIMDScalar::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_scalar_float");
bench_mandelbrot_v<backend::SIMDNative::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_native_float");
bench_mandelbrot_v<backend::SIMDVector<4>::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_vector4_float");
bench_mandelbrot_v<backend::SIMDVector<8>::Float_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_vector8_float");
#endif
/* double precision */
bench_mandelbrot<double>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "double");
bench_mandelbrot_v<backend::Scalar::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "double_v");
#ifdef VECCORE_ENABLE_VC
bench_mandelbrot_v<backend::VcVector::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "double_vc");
#endif
#ifdef VECCORE_ENABLE_UMESIMD
bench_mandelbrot_v<backend::UMESimd::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "double_umesimd");
#endif
#ifdef VECCORE_ENABLE_STD_SIMD
bench_mandelbrot_v<backend::SIMDScalar::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_scalar_double");
bench_mandelbrot_v<backend::SIMDNative::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_native_double");
bench_mandelbrot_v<backend::SIMDVector<4>::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_vector4_double");
bench_mandelbrot_v<backend::SIMDVector<8>::Double_v>(xmin, xmax, nx, ymin, ymax, ny,
max_iter, image, "std_simd_vector8_double");
#endif
delete[] image;
return 0;
}