| 1 |
/* |
| 2 |
* Copyright (C) 2020 Linux Studio Plugins Project <https://lsp-plug.in/> |
| 3 |
* (C) 2020 Vladimir Sadovnikov <[email protected]> |
| 4 |
* |
| 5 |
* This file is part of lsp-dsp-lib |
| 6 |
* Created on: 31 мар. 2020 г. |
| 7 |
* |
| 8 |
* lsp-dsp-lib is free software: you can redistribute it and/or modify |
| 9 |
* it under the terms of the GNU Lesser General Public License as published by |
| 10 |
* the Free Software Foundation, either version 3 of the License, or |
| 11 |
* any later version. |
| 12 |
* |
| 13 |
* lsp-dsp-lib is distributed in the hope that it will be useful, |
| 14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 |
* GNU Lesser General Public License for more details. |
| 17 |
* |
| 18 |
* You should have received a copy of the GNU Lesser General Public License |
| 19 |
* along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>. |
| 20 |
*/ |
| 21 |
|
| 22 |
#include <lsp-plug.in/common/alloc.h> |
| 23 |
#include <lsp-plug.in/common/types.h> |
| 24 |
#include <lsp-plug.in/dsp/dsp.h> |
| 25 |
#include <lsp-plug.in/test-fw/helpers.h> |
| 26 |
#include <lsp-plug.in/test-fw/ptest.h> |
| 27 |
|
| 28 |
#define MIN_RANK 8 |
| 29 |
#define MAX_RANK 16 |
| 30 |
|
| 31 |
namespace lsp |
| 32 |
{ |
| 33 |
namespace generic |
| 34 |
{ |
| 35 |
void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 36 |
void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 37 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 38 |
void add2(float *dst, const float *src, size_t count); |
| 39 |
|
| 40 |
void fastconv_parse(float *dst, const float *src, size_t rank); |
| 41 |
void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 42 |
} |
| 43 |
|
| 44 |
IF_ARCH_X86( |
| 45 |
namespace sse |
| 46 |
{ |
| 47 |
void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 48 |
void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 49 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 50 |
void add2(float *dst, const float *src, size_t count); |
| 51 |
|
| 52 |
void fastconv_parse(float *dst, const float *src, size_t rank); |
| 53 |
void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 54 |
} |
| 55 |
|
| 56 |
namespace avx |
| 57 |
{ |
| 58 |
void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 59 |
void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 60 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 61 |
void add2(float *dst, const float *src, size_t count); |
| 62 |
|
| 63 |
void fastconv_parse(float *dst, const float *src, size_t rank); |
| 64 |
void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 65 |
|
| 66 |
void direct_fft_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 67 |
void reverse_fft_fma3(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 68 |
void complex_mul3_fma3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 69 |
|
| 70 |
void fastconv_parse_fma3(float *dst, const float *src, size_t rank); |
| 71 |
void fastconv_parse_apply_fma3(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 72 |
} |
| 73 |
) |
| 74 |
|
| 75 |
IF_ARCH_ARM( |
| 76 |
namespace neon_d32 |
| 77 |
{ |
| 78 |
void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 79 |
void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 80 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 81 |
void add2(float *dst, const float *src, size_t count); |
| 82 |
|
| 83 |
void fastconv_parse(float *dst, const float *src, size_t rank); |
| 84 |
void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 85 |
} |
| 86 |
) |
| 87 |
|
| 88 |
IF_ARCH_AARCH64( |
| 89 |
namespace asimd |
| 90 |
{ |
| 91 |
void direct_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 92 |
void reverse_fft(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 93 |
void complex_mul3(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 94 |
void add2(float *dst, const float *src, size_t count); |
| 95 |
|
| 96 |
void fastconv_parse(float *dst, const float *src, size_t rank); |
| 97 |
void fastconv_parse_apply(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 98 |
} |
| 99 |
) |
| 100 |
|
| 101 |
typedef void (* direct_fft_t)(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 102 |
typedef void (* reverse_fft_t)(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank); |
| 103 |
typedef void (* complex_mul3_t)(float *dst_re, float *dst_im, const float *src1_re, const float *src1_im, const float *src2_re, const float *src2_im, size_t count); |
| 104 |
typedef void (* add2_t)(float *dst, const float *src, size_t count); |
| 105 |
|
| 106 |
typedef void (* fastconv_parse_t)(float *dst, const float *src, size_t rank); |
| 107 |
typedef void (* fastconv_parse_apply_t)(float *dst, float *tmp, const float *c, const float *src, size_t rank); |
| 108 |
} |
| 109 |
|
| 110 |
//----------------------------------------------------------------------------- |
| 111 |
// Performance test for complex multiplication |
| 112 |
PTEST_BEGIN("dsp.fft", fastconv, 10, 1000) |
| 113 |
|
| 114 |
void call( |
| 115 |
const char *label, |
| 116 |
float *out, float *tmp, float *tmp2, float *conv, const float *in, const float *cv, size_t rank, |
| 117 |
direct_fft_t direct, complex_mul3_t cmul, reverse_fft_t reverse, add2_t add2 |
| 118 |
) |
| 119 |
{ |
| 120 |
if (!(PTEST_SUPPORTED(direct) && (PTEST_SUPPORTED(cmul)) && (PTEST_SUPPORTED(reverse)) && (PTEST_SUPPORTED(add2)))) |
| 121 |
return; |
| 122 |
|
| 123 |
char buf[80]; |
| 124 |
snprintf(buf, sizeof(buf), "%s x %d", label, int(1 << rank)); |
| 125 |
printf("Testing %s samples (rank = %d)...\n", buf, int(rank)); |
| 126 |
size_t bin_size = 1 << rank; |
| 127 |
|
| 128 |
// Prepare data |
| 129 |
dsp::fill_zero(out, bin_size); |
| 130 |
direct(conv, &conv[bin_size], cv, &cv[bin_size], rank); |
| 131 |
|
| 132 |
// Run test |
| 133 |
PTEST_LOOP(buf, |
| 134 |
direct(tmp, &tmp[bin_size], in, &in[bin_size], rank); |
| 135 |
cmul(tmp, &tmp[bin_size], tmp, &tmp[bin_size], conv, &conv[bin_size], bin_size); |
| 136 |
reverse(tmp2, &tmp2[bin_size], tmp, &tmp[bin_size], rank); |
| 137 |
add2(out, tmp2, bin_size); |
| 138 |
) |
| 139 |
} |
| 140 |
|
| 141 |
void call( |
| 142 |
const char *label, |
| 143 |
float *out, float *tmp, float *conv, const float *in, const float *cv, size_t rank, |
| 144 |
fastconv_parse_t parse, fastconv_parse_apply_t apply |
| 145 |
) |
| 146 |
{ |
| 147 |
if (!(PTEST_SUPPORTED(parse) && (PTEST_SUPPORTED(apply)))) |
| 148 |
return; |
| 149 |
|
| 150 |
char buf[80]; |
| 151 |
snprintf(buf, sizeof(buf), "%s x %d", label, int(1 << rank)); |
| 152 |
printf("Testing %s samples (rank = %d)...\n", buf, int(rank)); |
| 153 |
|
| 154 |
// Prepare data |
| 155 |
dsp::fill_zero(out, 1 << rank); |
| 156 |
parse(conv, cv, rank); |
| 157 |
|
| 158 |
PTEST_LOOP(buf, |
| 159 |
apply(out, tmp, conv, in, rank); |
| 160 |
); |
| 161 |
} |
| 162 |
|
| 163 |
PTEST_MAIN |
| 164 |
{ |
| 165 |
size_t fft_size = 1 << MAX_RANK; |
| 166 |
size_t alloc = fft_size * 2 * 2 // in + cv size |
| 167 |
+ fft_size * 2 // conv size |
| 168 |
+ fft_size * 2 // tmp size |
| 169 |
+ fft_size * 2 // tmp2 size |
| 170 |
+ fft_size; // out size |
| 171 |
|
| 172 |
uint8_t *data = NULL; |
| 173 |
float *in = alloc_aligned<float>(data, alloc, 64); |
| 174 |
dsp::fill_zero(in, alloc); |
| 175 |
|
| 176 |
float *cv = &in[fft_size * 2]; |
| 177 |
float *conv = &cv[fft_size * 2]; |
| 178 |
float *tmp = &conv[fft_size * 2]; |
| 179 |
float *tmp2 = &tmp[fft_size * 2]; |
| 180 |
float *out = &tmp2[fft_size * 2]; |
| 181 |
|
| 182 |
for (size_t i=0; i < (fft_size / 2); ++i) |
| 183 |
{ |
| 184 |
in[i] = randf(0.0f, 1.0f); |
| 185 |
cv[i] = randf(0.0f, 1.0f); |
| 186 |
} |
| 187 |
|
| 188 |
for (size_t rank=MIN_RANK; rank <= MAX_RANK; ++rank) |
| 189 |
{ |
| 190 |
call("generic::fft", out, tmp, tmp2, conv, in, cv, rank, |
| 191 |
generic::direct_fft, generic::complex_mul3, generic::reverse_fft, generic::add2); |
| 192 |
call("generic::fastconv_fft", out, tmp, conv, in, cv, rank, |
| 193 |
generic::fastconv_parse, generic::fastconv_parse_apply); |
| 194 |
|
| 195 |
IF_ARCH_X86( |
| 196 |
call("sse::fft", out, tmp, tmp2, conv, in, cv, rank, |
| 197 |
sse::direct_fft, sse::complex_mul3, sse::reverse_fft, sse::add2); |
| 198 |
call("avx::fft", out, tmp, tmp2, conv, in, cv, rank, |
| 199 |
avx::direct_fft, avx::complex_mul3, avx::reverse_fft, avx::add2); |
| 200 |
call("avx::fft_fma3", out, tmp, tmp2, conv, in, cv, rank, |
| 201 |
avx::direct_fft_fma3, avx::complex_mul3_fma3, avx::reverse_fft_fma3, avx::add2); |
| 202 |
|
| 203 |
call("sse::fastconv_fft", out, tmp, conv, in, cv, rank, |
| 204 |
sse::fastconv_parse, sse::fastconv_parse_apply); |
| 205 |
call("avx::fastconv_fft", out, tmp, conv, in, cv, rank, |
| 206 |
avx::fastconv_parse, avx::fastconv_parse_apply); |
| 207 |
call("avx::fastconv_fft_fma3", out, tmp, conv, in, cv, rank, |
| 208 |
avx::fastconv_parse_fma3, avx::fastconv_parse_apply_fma3); |
| 209 |
) |
| 210 |
|
| 211 |
IF_ARCH_ARM( |
| 212 |
call("neon_d32::fft", out, tmp, tmp2, conv, in, cv, rank, |
| 213 |
neon_d32::direct_fft, neon_d32::complex_mul3, neon_d32::reverse_fft, neon_d32::add2); |
| 214 |
call("neon_d32::fastconv_fft", out, tmp, conv, in, cv, rank, |
| 215 |
neon_d32::fastconv_parse, neon_d32::fastconv_parse_apply); |
| 216 |
) |
| 217 |
|
| 218 |
IF_ARCH_AARCH64( |
| 219 |
call("asimd::fft", out, tmp, tmp2, conv, in, cv, rank, |
| 220 |
asimd::direct_fft, asimd::complex_mul3, asimd::reverse_fft, asimd::add2); |
| 221 |
call("asimd::fastconv_fft", out, tmp, conv, in, cv, rank, |
| 222 |
asimd::fastconv_parse, asimd::fastconv_parse_apply); |
| 223 |
) |
| 224 |
|
| 225 |
PTEST_SEPARATOR; |
| 226 |
} |
| 227 |
|
| 228 |
free_aligned(data); |
| 229 |
} |
| 230 |
|
| 231 |
PTEST_END |
| 232 |
|
| 233 |
|