-
Notifications
You must be signed in to change notification settings - Fork 1
/
batched_fft.cxx
194 lines (167 loc) · 6.27 KB
/
batched_fft.cxx
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
#include <CL/sycl.hpp>
#include <complex>
#include <ctime>
#include <iostream>
#include <oneapi/mkl.hpp>
constexpr double PI = 3.141592653589793;
template <typename T>
inline void _expect_near_array(const char* file, int line, const char* xname,
T* x, const char* yname, T* y, int n,
double max_err = -1.0) {
bool equal = true;
int i;
double err;
if (max_err == -1.0) {
max_err = 1e-10;
}
for (i = 0; i < n; i++) {
err = std::abs(x[i] - y[i]);
if (err > max_err) {
equal = false;
break;
}
}
if (!equal) {
// const int max_view = 10;
std::cerr << "Arrays not close (max " << max_err << ") at " << file
<< ":" << line << std::endl
<< " err " << err << " at [" << i << "]" << std::endl
<< " " << xname << ":" << std::endl
<< x[i] << std::endl
<< " " << yname << ":" << std::endl
<< y[i] << std::endl;
}
}
#define GT_EXPECT_NEAR_ARRAY(x, y, n) \
_expect_near_array(__FILE__, __LINE__, #x, x, #y, y, n)
#define GT_EXPECT_NEAR_ARRAY_ERR(x, y, n, max_err) \
_expect_near_array(__FILE__, __LINE__, #x, x, #y, y, n, max_err)
template <typename E, typename Desc>
void test_fft_r2c_1d_many_batches(sycl::queue& q) {
constexpr int Nx = 48;
constexpr int Nxc = Nx / 2 + 1;
constexpr int batch_size = 1024 * 256;
using T = std::complex<E>;
E* h_A = sycl::malloc_host<E>(Nx * batch_size, q);
q.fill(h_A, 0, Nx * batch_size);
E* d_A = sycl::malloc_device<E>(Nx * batch_size, q);
E* h_A2 = sycl::malloc_host<E>(Nx * batch_size, q);
q.fill(h_A2, 0, Nx * batch_size);
E* d_A2 = sycl::malloc_device<E>(Nx * batch_size, q);
T* h_B = sycl::malloc_host<T>(Nxc * batch_size, q);
T* h_B0_expected = sycl::malloc_host<T>(Nxc, q);
T* d_B = sycl::malloc_device<T>(Nxc * batch_size, q);
double x, y;
struct timespec start, end;
double elapsed;
std::cout << "init h_A" << std::endl;
// Set up periodic domain with frequency 4 and 8 for batch 0 and 1
// m = [sin(2*pi*x) for x in -2:4/Nx:2-4/Nx]
for (int i = 0; i < Nx; i++) {
x = -2.0 + 4.0 * i / static_cast<E>(Nx);
y = sin(2 * PI * x);
for (int b = 0; b < batch_size; b++) {
h_A[i + b * Nx] = y;
}
}
std::cout << "copy h_A -> d_A" << std::endl;
q.copy(h_A, d_A, Nx * batch_size).wait();
std::int64_t rstrides[2] = {0, 1};
std::int64_t cstrides[2] = {0, 1};
std::cout << "init plan" << std::endl;
Desc plan(Nx);
plan.set_value(oneapi::mkl::dft::config_param::NUMBER_OF_TRANSFORMS,
batch_size);
plan.set_value(oneapi::mkl::dft::config_param::INPUT_STRIDES, rstrides);
plan.set_value(oneapi::mkl::dft::config_param::OUTPUT_STRIDES, cstrides);
plan.set_value(oneapi::mkl::dft::config_param::FWD_DISTANCE, Nx);
plan.set_value(oneapi::mkl::dft::config_param::BWD_DISTANCE, Nxc);
plan.set_value(oneapi::mkl::dft::config_param::PLACEMENT,
DFTI_NOT_INPLACE);
plan.set_value(oneapi::mkl::dft::config_param::CONJUGATE_EVEN_STORAGE,
DFTI_COMPLEX_COMPLEX);
plan.commit(q);
std::cout << "compute forward fft" << std::endl;
clock_gettime(CLOCK_MONOTONIC, &start);
auto e =
oneapi::mkl::dft::compute_forward(plan, d_A, reinterpret_cast<E*>(d_B));
e.wait();
clock_gettime(CLOCK_MONOTONIC, &end);
elapsed =
(end.tv_sec - start.tv_sec) + (end.tv_nsec - start.tv_nsec) * 1.0e-9;
std::cout << "fft time(s): " << elapsed << std::endl;
std::cout << "copy fft result to host" << std::endl;
q.copy(d_B, h_B, Nxc * batch_size).wait();
// NB: allow greater error than for other tests
double max_err = 0.0005;
std::cout << "init expected result" << std::endl;
// Expect denormalized -0.5i at positions 4 for all batches
for (int i = 0; i < Nxc; i++) {
if (i == 4) {
h_B0_expected[i] = T(0, -0.5 * Nx);
} else {
h_B0_expected[i] = T(0, 0);
}
}
std::cout << "check result" << std::endl;
for (int b = 0; b < batch_size; b++) {
GT_EXPECT_NEAR_ARRAY_ERR(h_B0_expected, h_B + b * Nxc, Nxc, max_err);
}
/*
std::cout << "round trip test" << std::endl;
// test roundtripping data, with normalization
plan.set_value(oneapi::mkl::dft::config_param::INPUT_STRIDES, cstrides);
plan.set_value(oneapi::mkl::dft::config_param::OUTPUT_STRIDES, rstrides);
plan.commit(q);
e = oneapi::mkl::dft::compute_backward(plan, reinterpret_cast<E*>(d_B),
d_A2);
e.wait();
q.copy(d_A2, h_A2, Nx * batch_size);
for (int i = 0; i < Nx * batch_size; i++) {
h_A2[i] /= E(Nx);
}
GT_EXPECT_NEAR_ARRAY(h_A, h_A2, Nx * batch_size);
*/
sycl::free(h_A, q);
sycl::free(h_A2, q);
sycl::free(d_A, q);
sycl::free(h_B, q);
sycl::free(h_B0_expected, q);
sycl::free(d_B, q);
}
inline auto get_exception_handler() {
static auto exception_handler = [](cl::sycl::exception_list exceptions) {
for (std::exception_ptr const& e : exceptions) {
try {
std::rethrow_exception(e);
} catch (cl::sycl::exception const& e) {
std::cerr << "Caught asynchronous SYCL exception:" << std::endl
<< e.what() << std::endl;
abort();
}
}
};
return exception_handler;
}
int main(int argc, char** argv) {
auto q = cl::sycl::queue{get_exception_handler()};
auto dev = q.get_device();
std::string type;
if (dev.is_cpu()) {
type = "CPU ";
} else if (dev.is_gpu()) {
type = "GPU ";
} else if (dev.is_host()) {
type = "HOST ";
} else {
type = "OTHER";
}
std::cout << "[" << type << "] "
<< dev.get_info<cl::sycl::info::device::name>() << " {"
<< dev.get_info<cl::sycl::info::device::vendor>() << "}"
<< std::endl;
using Desc =
oneapi::mkl::dft::descriptor<oneapi::mkl::dft::precision::DOUBLE,
oneapi::mkl::dft::domain::REAL>;
test_fft_r2c_1d_many_batches<double, Desc>(q);
}