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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
|
/*
* Copyright (C) 2020 Linux Studio Plugins Project <https://lsp-plug.in/>
* (C) 2020 Vladimir Sadovnikov <sadko4u@gmail.com>
*
* This file is part of lsp-dsp-lib
* Created on: 31 мар. 2020 г.
*
* lsp-dsp-lib is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* any later version.
*
* lsp-dsp-lib is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
*/
#ifndef PRIVATE_DSP_ARCH_X86_SSE_FFT_BUTTERFLY_H_
#define PRIVATE_DSP_ARCH_X86_SSE_FFT_BUTTERFLY_H_
#ifndef PRIVATE_DSP_ARCH_X86_SSE_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_X86_SSE_IMPL */
/* This is the SSE implementation of the function:
static void butterfly_direct_fft(float *dst_re, float *dst_im, size_t pairs, size_t blocks)
{
float w[4], w_re[4], w_im[4], c_re[4], c_im[4];
size_t stride = pairs << 1;
for (size_t p=0; p < pairs; p += 4)
{
// Calculate sines and cosines
float l = M_PI / pairs;
w[0] = p * l;
w[1] = (p + 1) * l;
w[2] = (p + 2) * l;
w[3] = (p + 3) * l;
w_re[0] = cosf(w[0]);
w_re[1] = cosf(w[1]);
w_re[2] = cosf(w[2]);
w_re[3] = cosf(w[3]);
w_im[0] = sinf(w[0]);
w_im[1] = sinf(w[1]);
w_im[2] = sinf(w[2]);
w_im[3] = sinf(w[3]);
// Init pointers
float *a_re = &dst_re[p];
float *a_im = &dst_im[p];
float *b_re = &a_re[pairs];
float *b_im = &a_im[pairs];
for (size_t b=0; b < blocks; ++b)
{
// Perform 4x butterflies
// Calculate complex c[i] = w[i] * b[i]
c_re[0] = w_re[0] * b_re[0] + w_im[0] * b_im[0];
c_re[1] = w_re[1] * b_re[1] + w_im[1] * b_im[1];
c_re[2] = w_re[2] * b_re[2] + w_im[2] * b_im[2];
c_re[3] = w_re[3] * b_re[3] + w_im[3] * b_im[3];
c_im[0] = w_re[0] * b_im[0] - w_im[0] * b_re[0];
c_im[1] = w_re[1] * b_im[1] - w_im[1] * b_re[1];
c_im[2] = w_re[2] * b_im[2] - w_im[2] * b_re[2];
c_im[3] = w_re[3] * b_im[3] - w_im[3] * b_re[3];
// Calculate the output values:
// a[i]' = a[i] + c[i]
// b[i]' = a[i] - c[i]
b_re[0] = a_re[0] - c_re[0];
b_re[1] = a_re[1] - c_re[1];
b_re[2] = a_re[2] - c_re[2];
b_re[3] = a_re[3] - c_re[3];
b_im[0] = a_im[0] - c_im[0];
b_im[1] = a_im[1] - c_im[1];
b_im[2] = a_im[2] - c_im[2];
b_im[3] = a_im[3] - c_im[3];
a_re[0] = a_re[0] + c_re[0];
a_re[1] = a_re[1] + c_re[1];
a_re[2] = a_re[2] + c_re[2];
a_re[3] = a_re[3] + c_re[3];
a_im[0] = a_im[0] + c_im[0];
a_im[1] = a_im[1] + c_im[1];
a_im[2] = a_im[2] + c_im[2];
a_im[3] = a_im[3] + c_im[3];
// Move pointers
a_re += stride;
a_im += stride;
b_re += stride;
b_im += stride;
}
}
}
*/
namespace lsp
{
namespace sse
{
#ifdef ARCH_I386
#define FFT_BUTTERFLY_BODY(add_b, add_a) \
/* Init pointers */ \
float *a_re = &dst_re[b*pairs*2]; \
float *a_im = &dst_im[b*pairs*2]; \
float *b_re = &a_re[pairs]; \
float *b_im = &a_im[pairs]; \
size_t p = pairs; \
size_t off = 0; \
float *tmp_re, *tmp_im; \
\
ARCH_X86_ASM \
( \
/* Prepare angle */ \
__ASM_EMIT("mov %[XFFT_A_RE], %[tmp_re]") \
__ASM_EMIT("mov %[XFFT_A_IM], %[tmp_im]") \
__ASM_EMIT("movaps (%[tmp_re], %[rank]), %%xmm6") /* xmm6 = angle_re[0..3] */ \
__ASM_EMIT("movaps (%[tmp_im], %[rank]), %%xmm7") /* xmm7 = angle_im[0..3] */ \
/* Start loop */ \
__ASM_EMIT(".align 16") \
__ASM_EMIT("1:") \
\
/* Load complex values */ \
/* predicate: xmm6 = w_re[0..3] */ \
/* predicate: xmm7 = w_im[0..3] */ \
__ASM_EMIT("mov %[a_re], %[tmp_re]") \
__ASM_EMIT("mov %[a_im], %[tmp_im]") \
__ASM_EMIT("movups (%[tmp_re], %[off]), %%xmm0") /* xmm0 = a_re[0..3] */ \
__ASM_EMIT("movups (%[tmp_im], %[off]), %%xmm1") /* xmm1 = a_im[0..3] */ \
__ASM_EMIT("mov %[b_re], %[tmp_re]") \
__ASM_EMIT("mov %[b_im], %[tmp_im]") \
__ASM_EMIT("movups (%[tmp_re], %[off]), %%xmm2") /* xmm2 = b_re[0..3] */ \
__ASM_EMIT("movups (%[tmp_im], %[off]), %%xmm3") /* xmm3 = b_im[0..3] */ \
\
/* Calculate complex multiplication */ \
__ASM_EMIT("movaps %%xmm2, %%xmm4") /* xmm4 = b_re[0..3] */ \
__ASM_EMIT("movaps %%xmm3, %%xmm5") /* xmm5 = b_im[0..3] */ \
__ASM_EMIT("mulps %%xmm6, %%xmm2") /* xmm2 = w_re[0..3] * b_re[0..3] */ \
__ASM_EMIT("mulps %%xmm6, %%xmm3") /* xmm3 = w_re[0..3] * b_im[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm4") /* xmm4 = w_im[0..3] * b_re[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm5") /* xmm5 = w_im[0..3] * b_im[0..3] */ \
__ASM_EMIT(add_b " %%xmm5, %%xmm2") /* xmm2 = c_re[0..3] = w_re[0..3] * b_re[0..3] +- w_im[0..3] * b_im[0..3] */ \
__ASM_EMIT(add_a " %%xmm4, %%xmm3") /* xmm3 = c_im[0..3] = w_re[0..3] * b_im[0..3] -+ w_im[0..3] * b_re[0..3] */ \
\
/* Perform butterfly */ \
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = a_re[0..3] */ \
__ASM_EMIT("movaps %%xmm1, %%xmm5") /* xmm5 = a_im[0..3] */ \
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = a_re[0..3] - c_re[0..3] */ \
__ASM_EMIT("subps %%xmm3, %%xmm1") /* xmm1 = a_im[0..3] - c_im[0..3] */ \
__ASM_EMIT("addps %%xmm4, %%xmm2") /* xmm2 = a_re[0..3] + c_re[0..3] */ \
__ASM_EMIT("addps %%xmm5, %%xmm3") /* xmm3 = a_im[0..3] + c_im[0..3] */ \
\
/* Store values */ \
__ASM_EMIT("movups %%xmm0, (%[tmp_re], %[off])") \
__ASM_EMIT("movups %%xmm1, (%[tmp_im], %[off])") \
__ASM_EMIT("mov %[a_re], %[tmp_re]") \
__ASM_EMIT("mov %[a_im], %[tmp_im]") \
__ASM_EMIT("movups %%xmm2, (%[tmp_re], %[off])") \
__ASM_EMIT("movups %%xmm3, (%[tmp_im], %[off])") \
__ASM_EMIT("add $0x10, %[off]") \
\
/* Repeat loop */ \
__ASM_EMIT("subl $4, %[p]") \
__ASM_EMIT("jz 2f") \
\
/* Rotate angle */ \
__ASM_EMIT("mov %[XFFT_W_RE], %[tmp_re]") \
__ASM_EMIT("mov %[XFFT_W_IM], %[tmp_im]") \
__ASM_EMIT("movaps (%[tmp_re], %[rank]), %%xmm0") /* xmm0 = w_re[0..3] */ \
__ASM_EMIT("movaps (%[tmp_im], %[rank]), %%xmm1") /* xmm1 = w_im[0..3] */ \
__ASM_EMIT("movaps %%xmm1, %%xmm3") /* xmm3 = w_im[0..3] */ \
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = w_re[0..3] */ \
\
__ASM_EMIT("mulps %%xmm6, %%xmm3") /* xmm3 = a_re[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm1") /* xmm1 = a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm0, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] */ \
__ASM_EMIT("mulps %%xmm2, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] */ \
__ASM_EMIT("subps %%xmm1, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] + a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("addps %%xmm3, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] - a_re[0..3] * w_im[0..3] */ \
\
/* Repeat loop */ \
__ASM_EMIT("jmp 1b") \
__ASM_EMIT("2:") \
\
: [tmp_re] "=&r" (tmp_re), [tmp_im] "=&r" (tmp_im), \
[off] "+r" (off), [p] X86_PGREG (p) \
: [rank] "r" (rank), \
[a_re] "m" (a_re), [a_im] "m" (a_im), \
[b_re] "m" (b_re), [b_im] "m" (b_im), \
[XFFT_A_RE] "g" (&XFFT_A_RE[0]), [XFFT_A_IM] "g" (&XFFT_A_IM[0]), \
[XFFT_W_RE] "g" (&XFFT_W_RE[0]), [XFFT_W_IM] "g" (&XFFT_W_IM[0]) \
: "cc", "memory", \
"%xmm0", "%xmm1", "%xmm2", "%xmm3", \
"%xmm4", "%xmm5", "%xmm6", "%xmm7" \
);
#else /* ARCH_X86_64 */
#define FFT_BUTTERFLY_BODY(add_b, add_a) \
/* Init pointers */ \
float *a_re = &dst_re[b*pairs*2]; \
float *a_im = &dst_im[b*pairs*2]; \
float *b_re = &a_re[pairs]; \
float *b_im = &a_im[pairs]; \
size_t p = pairs; \
size_t off = 0; \
\
ARCH_X86_ASM \
( \
/* Prepare angle */ \
__ASM_EMIT("movaps (%[XFFT_A_RE], %[rank]), %%xmm6") /* xmm6 = angle_re[0..3] */ \
__ASM_EMIT("movaps (%[XFFT_A_IM], %[rank]), %%xmm7") /* xmm7 = angle_im[0..3] */ \
/* Start loop */ \
__ASM_EMIT(".align 16") \
__ASM_EMIT("1:") \
\
/* Load complex values */ \
/* predicate: xmm6 = w_re[0..3] */ \
/* predicate: xmm7 = w_im[0..3] */ \
__ASM_EMIT("movups (%[a_re], %[off]), %%xmm0") /* xmm0 = a_re[0..3] */ \
__ASM_EMIT("movups (%[a_im], %[off]), %%xmm1") /* xmm1 = a_im[0..3] */ \
__ASM_EMIT("movups (%[b_re], %[off]), %%xmm2") /* xmm2 = b_re[0..3] */ \
__ASM_EMIT("movups (%[b_im], %[off]), %%xmm3") /* xmm3 = b_im[0..3] */ \
\
/* Calculate complex multiplication */ \
__ASM_EMIT("movaps %%xmm2, %%xmm4") /* xmm4 = b_re[0..3] */ \
__ASM_EMIT("movaps %%xmm3, %%xmm5") /* xmm5 = b_im[0..3] */ \
__ASM_EMIT("mulps %%xmm6, %%xmm2") /* xmm2 = w_re[0..3] * b_re[0..3] */ \
__ASM_EMIT("mulps %%xmm6, %%xmm3") /* xmm3 = w_re[0..3] * b_im[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm4") /* xmm4 = w_im[0..3] * b_re[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm5") /* xmm5 = w_im[0..3] * b_im[0..3] */ \
__ASM_EMIT(add_b " %%xmm5, %%xmm2") /* xmm2 = c_re[0..3] = w_re[0..3] * b_re[0..3] +- w_im[0..3] * b_im[0..3] */ \
__ASM_EMIT(add_a " %%xmm4, %%xmm3") /* xmm3 = c_im[0..3] = w_re[0..3] * b_im[0..3] -+ w_im[0..3] * b_re[0..3] */ \
\
/* Perform butterfly */ \
__ASM_EMIT("movaps %%xmm0, %%xmm4") /* xmm4 = a_re[0..3] */ \
__ASM_EMIT("movaps %%xmm1, %%xmm5") /* xmm5 = a_im[0..3] */ \
__ASM_EMIT("subps %%xmm2, %%xmm0") /* xmm0 = a_re[0..3] - c_re[0..3] */ \
__ASM_EMIT("subps %%xmm3, %%xmm1") /* xmm1 = a_im[0..3] - c_im[0..3] */ \
__ASM_EMIT("addps %%xmm4, %%xmm2") /* xmm2 = a_re[0..3] + c_re[0..3] */ \
__ASM_EMIT("addps %%xmm5, %%xmm3") /* xmm3 = a_im[0..3] + c_im[0..3] */ \
\
/* Store values */ \
__ASM_EMIT("movups %%xmm2, (%[a_re], %[off])") \
__ASM_EMIT("movups %%xmm3, (%[a_im], %[off])") \
__ASM_EMIT("movups %%xmm0, (%[b_re], %[off])") \
__ASM_EMIT("movups %%xmm1, (%[b_im], %[off])") \
/* Update pointers */ \
__ASM_EMIT("add $0x10, %[off]") \
/* Repeat loop */ \
__ASM_EMIT("sub $4, %[p]") \
__ASM_EMIT("jz 2f") \
\
/* Rotate angle */ \
__ASM_EMIT("movaps (%[XFFT_W_IM], %[rank]), %%xmm1") /* xmm1 = w_im[0..3] */ \
__ASM_EMIT("movaps (%[XFFT_W_RE], %[rank]), %%xmm0") /* xmm0 = w_re[0..3] */ \
__ASM_EMIT("movaps %%xmm1, %%xmm3") /* xmm3 = w_im[0..3] */ \
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = w_re[0..3] */ \
__ASM_EMIT("mulps %%xmm6, %%xmm3") /* xmm3 = a_re[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm1") /* xmm1 = a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm0, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] */ \
__ASM_EMIT("mulps %%xmm2, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] */ \
__ASM_EMIT("subps %%xmm1, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] + a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("addps %%xmm3, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] - a_re[0..3] * w_im[0..3] */ \
\
/* Repeat loop */ \
__ASM_EMIT("jmp 1b") \
__ASM_EMIT("2:") \
\
: [off] "+r" (off), [p] "+r" (p) \
: [a_re] "r"(a_re), [a_im] "r"(a_im), \
[b_re] "r"(b_re), [b_im] "r"(b_im), \
[rank] "r" (rank), \
[XFFT_A_RE] "r" (XFFT_A_RE), [XFFT_A_IM] "r" (XFFT_A_IM), \
[XFFT_W_RE] "r" (XFFT_W_RE), [XFFT_W_IM] "r"(XFFT_W_IM) \
: "cc", "memory", \
"%xmm0", "%xmm1", "%xmm2", "%xmm3", \
"%xmm4", "%xmm5", "%xmm6", "%xmm7" \
);
#endif /* ARCH_X86_64 */
#define FFT_ANGLE_INIT \
ARCH_X86_ASM \
( \
__ASM_EMIT("sub $2, %[rank]") /* rank -= 2 */ \
__ASM_EMIT("shl $4, %[rank]") /* rank *= 16 */ \
__ASM_EMIT("movaps (%[XFFT_A_RE], %[rank]), %%xmm6") /* xmm6 = angle_re[0..3] */ \
__ASM_EMIT("movaps (%[XFFT_A_IM], %[rank]), %%xmm7") /* xmm7 = angle_im[0..3] */ \
: [rank] "+r" (rank) \
: \
[XFFT_A_RE] "r"(XFFT_A_RE), \
[XFFT_A_IM] "r"(XFFT_A_IM) \
: "cc", \
"%xmm6", "%xmm7" \
);
#define FFT_ANGLE_ROTATE \
ARCH_X86_ASM \
( \
/* xmm6 = a_re[0..3] */ \
/* xmm7 = a_im[0..3] */ \
__ASM_EMIT("movaps (%[XFFT_W_IM], %[rank]), %%xmm1") /* xmm1 = w_im[0..3] */ \
__ASM_EMIT("movaps (%[XFFT_W_RE], %[rank]), %%xmm0") /* xmm0 = w_re[0..3] */ \
__ASM_EMIT("movaps %%xmm1, %%xmm3") /* xmm3 = w_im[0..3] */ \
__ASM_EMIT("movaps %%xmm0, %%xmm2") /* xmm2 = w_re[0..3] */ \
\
__ASM_EMIT("mulps %%xmm6, %%xmm3") /* xmm3 = a_re[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm7, %%xmm1") /* xmm1 = a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("mulps %%xmm0, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] */ \
__ASM_EMIT("mulps %%xmm2, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] */ \
__ASM_EMIT("subps %%xmm1, %%xmm6") /* xmm6 = a_re[0..3] * w_re[0..3] + a_im[0..3] * w_im[0..3] */ \
__ASM_EMIT("addps %%xmm3, %%xmm7") /* xmm7 = a_im[0..3] * w_re[0..3] - a_re[0..3] * w_im[0..3] */ \
: \
: [rank] "r" (rank), \
[XFFT_W_RE] "r"(XFFT_W_RE), \
[XFFT_W_IM] "r"(XFFT_W_IM) \
: "cc", \
"%xmm0", "%xmm1", "%xmm2", "%xmm3", \
"%xmm6", "%xmm7" \
);
static inline void butterfly_direct(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
rank = (rank - 2) << 4;
for (size_t b=0; b<blocks; ++b)
{
FFT_BUTTERFLY_BODY("addps", "subps");
}
}
static inline void butterfly_reverse(float *dst_re, float *dst_im, size_t rank, size_t blocks)
{
size_t pairs = 1 << rank;
rank = (rank - 2) << 4;
for (size_t b=0; b<blocks; ++b)
{
FFT_BUTTERFLY_BODY("subps", "addps");
}
}
#undef FFT_ANGLE_INIT
#undef FFT_ANGLE_ROTATE
#undef FFT_BUTTERFLY_BODY
} /* namespace sse */
} /* namespace lsp */
#endif /* PRIVATE_DSP_ARCH_X86_SSE_FFT_BUTTERFLY_H_ */
|
