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
|
/*
* 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_AARCH64_ASIMD_FASTCONV_PARSE_H_
#define PRIVATE_DSP_ARCH_AARCH64_ASIMD_FASTCONV_PARSE_H_
#ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL
#error "This header should not be included directly"
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL */
namespace lsp
{
namespace asimd
{
void fastconv_parse(float *dst, const float *src, size_t rank)
{
size_t items = 1 << rank; // number of complex numbers per half of block (16 for rank=4)
size_t n = items >> 1; // number of butterflies per block (8 for rank=4)
IF_ARCH_AARCH64(
const float *fdw = &XFFT_DW[(rank - 3) << 3];
const float *fw = &XFFT_A[(rank - 3) << 4];
size_t k, p;
float *a, *b;
);
if (n > 4)
{
// First loop
ARCH_AARCH64_ASM
(
__ASM_EMIT("mov %[a], %[dst]") // a = dst
__ASM_EMIT("add %[b], %[a], %[n], LSL #3") // b = &a[n*2]
__ASM_EMIT("mov %[k], %[n]")
__ASM_EMIT("ldp q16, q17, [%[XFFT_A], #0x00]") // v8 = wr0, v9 = wr1
__ASM_EMIT("ldp q18, q19, [%[XFFT_A], #0x20]") // v10 = wi0, v11 = wi1
__ASM_EMIT("ldp q24, q25, [%[XFFT_DW]]") // v24 = dr, w25 = di
// 8x butterflies
__ASM_EMIT("1:")
__ASM_EMIT("ldp q0, q2, [%[src]]") // v0 = ar0, v2 = ar1
__ASM_EMIT("eor v1.16b, v1.16b, v1.16b") // v1 = 0
__ASM_EMIT("eor v3.16b, v3.16b, v3.16b") // v3 = 0
__ASM_EMIT("fmul v5.4s, v0.4s, v18.4s") // v5 = ar0 * wi0
__ASM_EMIT("fmul v7.4s, v2.4s, v19.4s") // v7 = ar1 * wi1
__ASM_EMIT("fmul v4.4s, v0.4s, v16.4s") // v4 = ar0 * wr0 = br0
__ASM_EMIT("fmul v6.4s, v2.4s, v17.4s") // v6 = ar1 * wr1 = br1
__ASM_EMIT("fneg v5.4s, v5.4s") // v5 = - ar0 * wi0 = bi0
__ASM_EMIT("fneg v7.4s, v7.4s") // v7 = - ar1 * wi1 = bi1
__ASM_EMIT("stp q0, q1, [%[a], #0x00]")
__ASM_EMIT("stp q2, q3, [%[a], #0x20]")
__ASM_EMIT("add %[src], %[src], #0x20")
__ASM_EMIT("stp q4, q5, [%[b], #0x00]")
__ASM_EMIT("stp q6, q7, [%[b], #0x20]")
__ASM_EMIT("subs %[k], %[k], #8")
__ASM_EMIT("add %[a], %[a], #0x40")
__ASM_EMIT("add %[b], %[b], #0x40")
__ASM_EMIT("b.ls 2f")
// Rotate angle
__ASM_EMIT("fmul v20.4s, v16.4s, v25.4s") // v20.4s = wr0 * di
__ASM_EMIT("fmul v21.4s, v17.4s, v25.4s") // v21.4s = wr1 * di
__ASM_EMIT("fmul v22.4s, v18.4s, v25.4s") // v22.4s = wi0 * di
__ASM_EMIT("fmul v23.4s, v19.4s, v25.4s") // v23.4s = wi1 * di
__ASM_EMIT("fmul v16.4s, v16.4s, v24.4s") // v16.4s = wr0 * dr
__ASM_EMIT("fmul v17.4s, v17.4s, v24.4s") // v17.4s = wr1 * dr
__ASM_EMIT("fmul v18.4s, v18.4s, v24.4s") // v18.4s = wi0 * dr
__ASM_EMIT("fmul v19.4s, v19.4s, v24.4s") // v19.4s = wi1 * dr
__ASM_EMIT("fsub v16.4s, v16.4s, v22.4s") // v16.4s = wr0*dr - wi0*di
__ASM_EMIT("fsub v17.4s, v17.4s, v23.4s") // v17.4s = wr1*dr - wi1*di
__ASM_EMIT("fadd v18.4s, v18.4s, v20.4s") // v18.4s = wi0*dr + wr0*di
__ASM_EMIT("fadd v19.4s, v19.4s, v21.4s") // v19.4s = wi1*dr + wr1*di
__ASM_EMIT("b 1b")
// Post-process
__ASM_EMIT("2:")
__ASM_EMIT("sub %[XFFT_A], %[XFFT_A], #0x40") // fw -= 16
__ASM_EMIT("sub %[XFFT_DW], %[XFFT_DW], #0x20") // fdw -= 8
__ASM_EMIT("lsr %[n], %[n], #1") // n >>= 1
: [src] "+r" (src),
[a] "=&r" (a), [b] "=&r" (b),
[XFFT_A] "+r" (fw), [XFFT_DW] "+r" (fdw),
[k] "=&r" (k), [n] "+r" (n)
: [dst] "r" (dst)
: "cc", "memory",
"v0", "v1", "v2", "v3",
"v4", "v5", "v6", "v7",
"v16", "v17", "v18", "v19",
"v20", "v21", "v22", "v23",
"v24", "v25"
);
// All other loops while n > 4
ARCH_AARCH64_ASM
(
__ASM_EMIT("cmp %[n], #8")
__ASM_EMIT("b.lo 8f")
__ASM_EMIT("1:")
__ASM_EMIT("mov %[a], %[dst]") // a = dst
__ASM_EMIT("mov %[p], %[items]") // p = items
__ASM_EMIT("3:")
__ASM_EMIT("add %[b], %[a], %[n], LSL #3") // b = &a[n*2]
__ASM_EMIT("ldp q16, q17, [%[XFFT_A], #0x00]") // v16 = wr0, v17 = wr1
__ASM_EMIT("ldp q18, q19, [%[XFFT_A], #0x20]") // v18 = wi0, v19 = wi1
__ASM_EMIT("ldp q24, q25, [%[XFFT_DW]]") // v24 = dr, v25 = di
__ASM_EMIT("mov %[k], %[n]")
// 8x butterflies
__ASM_EMIT("5:")
__ASM_EMIT("ldp q0, q1, [%[a], #0x00]") // v0 = ar0, v1 = ai0
__ASM_EMIT("ldp q2, q3, [%[a], #0x20]") // v2 = ar1, v3 = ai1
__ASM_EMIT("ldp q4, q5, [%[b], #0x00]") // v4 = br0, v5 = bi0
__ASM_EMIT("ldp q6, q7, [%[b], #0x20]") // v6 = br1, v7 = bi1
__ASM_EMIT("fsub v20.4s, v0.4s, v4.4s") // v20 = ar0 - br0 = cr0
__ASM_EMIT("fsub v21.4s, v1.4s, v5.4s") // v21 = ai0 - bi0 = ci0
__ASM_EMIT("fsub v22.4s, v2.4s, v6.4s") // v22 = ar1 - br1 = cr1
__ASM_EMIT("fsub v23.4s, v3.4s, v7.4s") // v23 = ai1 - bi1 = ci1
__ASM_EMIT("fadd v0.4s, v0.4s, v4.4s") // v0 = ar0 + br0 = ar0'
__ASM_EMIT("fadd v1.4s, v1.4s, v5.4s") // v1 = ai0 + bi0 = ai0'
__ASM_EMIT("fadd v2.4s, v2.4s, v6.4s") // v2 = ar1 + br1 = ar1'
__ASM_EMIT("fadd v3.4s, v3.4s, v7.4s") // v3 = ai1 + bi1 = ai1'
__ASM_EMIT("stp q0, q1, [%[a], #0x00]")
__ASM_EMIT("stp q2, q3, [%[a], #0x20]")
__ASM_EMIT("fmul v0.4s, v16.4s, v20.4s") // v0 = wr0 * cr0
__ASM_EMIT("fmul v2.4s, v17.4s, v22.4s") // v2 = wr1 * cr1
__ASM_EMIT("fmul v1.4s, v16.4s, v21.4s") // v1 = wr0 * ci0
__ASM_EMIT("fmul v3.4s, v17.4s, v23.4s") // v3 = wr1 * ci1
__ASM_EMIT("fmla v0.4s, v18.4s, v21.4s") // v0 = wr0*cr0 + wi0*ci0 = br0'
__ASM_EMIT("fmla v2.4s, v19.4s, v23.4s") // v2 = wr1*cr1 + wi1*ci1 = bi0'
__ASM_EMIT("fmls v1.4s, v18.4s, v20.4s") // v1 = wr0*ci0 - wi0*cr0 = br1'
__ASM_EMIT("fmls v3.4s, v19.4s, v22.4s") // v3 = wr1*ci1 - wi1*cr1 = bi1'
__ASM_EMIT("stp q0, q1, [%[b], #0x00]")
__ASM_EMIT("stp q2, q3, [%[b], #0x20]")
__ASM_EMIT("subs %[k], %[k], #8")
__ASM_EMIT("add %[a], %[a], #0x40")
__ASM_EMIT("add %[b], %[b], #0x40")
__ASM_EMIT("b.ls 6f")
// Rotate angle & repeat loop
__ASM_EMIT("fmul v20.4s, v16.4s, v25.4s") // v20 = wr0 * di
__ASM_EMIT("fmul v21.4s, v17.4s, v25.4s") // v21 = wr1 * di
__ASM_EMIT("fmul v22.4s, v18.4s, v25.4s") // v22 = wi0 * di
__ASM_EMIT("fmul v23.4s, v19.4s, v25.4s") // v23 = wi1 * di
__ASM_EMIT("fmul v16.4s, v16.4s, v24.4s") // v16 = wr0 * dr
__ASM_EMIT("fmul v17.4s, v17.4s, v24.4s") // v17 = wr1 * dr
__ASM_EMIT("fmul v18.4s, v18.4s, v24.4s") // v18 = wi0 * dr
__ASM_EMIT("fmul v19.4s, v19.4s, v24.4s") // v19 = wi1 * dr
__ASM_EMIT("fsub v16.4s, v16.4s, v22.4s") // v16 = wr0*dr - wi0*di
__ASM_EMIT("fsub v17.4s, v17.4s, v23.4s") // v17 = wr1*dr - wi1*di
__ASM_EMIT("fadd v18.4s, v18.4s, v20.4s") // v18 = wi0*dr + wr0*di
__ASM_EMIT("fadd v19.4s, v19.4s, v21.4s") // v19 = wi1*dr + wr1*di
__ASM_EMIT("b 5b")
__ASM_EMIT("6:")
__ASM_EMIT("mov %[a], %[b]") // a = b
__ASM_EMIT("subs %[p], %[p], %[n], LSL #1") // p -= n*2
__ASM_EMIT("bne 3b")
__ASM_EMIT("lsr %[n], %[n], #1") // n >>= 1
__ASM_EMIT("sub %[XFFT_A], %[XFFT_A], #0x40") // fw -= 16
__ASM_EMIT("sub %[XFFT_DW], %[XFFT_DW], #0x20") // fdw -= 8
__ASM_EMIT("cmp %[n], #8")
__ASM_EMIT("b.ge 1b")
// Loop for n=4
__ASM_EMIT("8:")
__ASM_EMIT("mov %[a], %[dst]") // a = dst
__ASM_EMIT("mov %[p], %[items]") // p = items
__ASM_EMIT("ldp q16, q17, [%[XFFT_A], #0x00]") // v16 = wr0, v17 = wr1
__ASM_EMIT("ldp q18, q19, [%[XFFT_A], #0x20]") // v18 = wi0, v19 = wi1
// 8x butterflies
__ASM_EMIT("9:")
__ASM_EMIT("ldp q0, q1, [%[a], #0x00]") // v0 = ar0, v1 = ai0
__ASM_EMIT("ldp q2, q3, [%[a], #0x20]") // v2 = ar1, v3 = ai1
__ASM_EMIT("ldp q4, q5, [%[a], #0x40]") // v4 = ar2, v5 = ai2
__ASM_EMIT("ldp q6, q7, [%[a], #0x60]") // v6 = ar3, v7 = ai3
__ASM_EMIT("fsub v20.4s, v0.4s, v2.4s") // v20 = ar0 - br0 = cr0
__ASM_EMIT("fsub v21.4s, v1.4s, v3.4s") // v21 = ai0 - bi0 = ci0
__ASM_EMIT("fsub v22.4s, v4.4s, v6.4s") // v22 = ar1 - br1 = cr1
__ASM_EMIT("fsub v23.4s, v5.4s, v7.4s") // v23 = ai1 - bi1 = ci1
__ASM_EMIT("fadd v0.4s, v0.4s, v2.4s") // v0 = ar0 + br0 = ar0'
__ASM_EMIT("fadd v1.4s, v1.4s, v3.4s") // v1 = ai0 + bi0 = ai0'
__ASM_EMIT("fadd v4.4s, v4.4s, v6.4s") // v4 = ar1 + br1 = ar1'
__ASM_EMIT("fadd v5.4s, v5.4s, v7.4s") // v5 = ai1 + bi1 = ai1'
__ASM_EMIT("fmul v2.4s, v16.4s, v20.4s") // v2 = wr0 * cr0
__ASM_EMIT("fmul v6.4s, v17.4s, v22.4s") // v6 = wr1 * cr1
__ASM_EMIT("fmul v3.4s, v16.4s, v21.4s") // v3 = wr0 * ci0
__ASM_EMIT("fmul v7.4s, v17.4s, v23.4s") // v7 = wr1 * ci1
__ASM_EMIT("fmla v2.4s, v18.4s, v21.4s") // v2 = wr0*cr0 + wi0*ci0 = br0'
__ASM_EMIT("fmla v6.4s, v19.4s, v23.4s") // v6 = wr1*cr1 + wi1*ci1 = br1'
__ASM_EMIT("fmls v3.4s, v18.4s, v20.4s") // v3 = wr0*ci0 - wi0*cr0 = bi0'
__ASM_EMIT("fmls v7.4s, v19.4s, v22.4s") // v7 = wr1*ci1 - wi1*cr1 = bi1'
__ASM_EMIT("stp q0, q1, [%[a], #0x00]")
__ASM_EMIT("stp q2, q3, [%[a], #0x20]")
__ASM_EMIT("stp q4, q5, [%[a], #0x40]")
__ASM_EMIT("stp q6, q7, [%[a], #0x60]")
__ASM_EMIT("subs %[p], %[p], #16") // p -= 16
__ASM_EMIT("add %[a], %[a], #0x80")
__ASM_EMIT("bne 9b")
: [a] "=&r" (a), [b] "=&r" (b),
[k] "=&r" (k), [p] "=&r" (p),
[XFFT_A] "+r" (fw), [XFFT_DW] "+r" (fdw),
[n] "+r" (n)
: [dst] "r" (dst), [items] "r" (items)
: "cc", "memory",
"v0", "v1", "v2", "v3",
"v4", "v5", "v6", "v7",
"v16", "v17", "v18", "v19",
"v20", "v21", "v22", "v23",
"v24", "v25"
);
}
else
{
ARCH_AARCH64_ASM
(
__ASM_EMIT("ldr q0, [%[src]]")
__ASM_EMIT("eor v1.16b, v1.16b, v1.16b")
__ASM_EMIT("stp q0, q1, [%[dst]]")
:
: [src] "r" (src), [dst] "r" (dst)
: "memory",
"v0", "v1"
);
}
// Perform small 4x4 post-processing
ARCH_AARCH64_ASM(
// Loop for n=4
__ASM_EMIT("subs %[items], %[items], #8")
__ASM_EMIT("ldp q16, q17, [%[MASK]]")
__ASM_EMIT("b.lo 2f")
// 8x butterflies
__ASM_EMIT("1:")
__ASM_EMIT("ld4 {v0.4s-v3.4s}, [%[dst]]") // v0 = r0 i0 r4 i4, v1 = r1 i1 r5 i5, v2 = r2 i2 r6 i6, v3 = r3 i3 r7 i7
// v0 = r0 i0 r4 i4
// v1 = r1 i1 r5 i5
// v2 = r2 i2 r6 i6
// v3 = r3 i3 r7 i7
__ASM_EMIT("fsub v4.4s, v0.4s, v2.4s") // v4 = r0-r2 i0-i2 r4-r6 i4-i6 = r1' i1' r5' i5'
__ASM_EMIT("fsub v5.4s, v1.4s, v3.4s") // v5 = r1-r3 i1-i3 r5-r7 i5-i7 = r3' i3' r7' i7'
__ASM_EMIT("fadd v6.4s, v0.4s, v2.4s") // v6 = r0+r2 i0+i2 r4+r6 i4+i6 = r0' i0' r4' i4'
__ASM_EMIT("fadd v7.4s, v1.4s, v3.4s") // v7 = r1+r3 i1+i3 r5+r7 i5+i7 = r2' i2' r6' i6'
__ASM_EMIT("rev64 v5.4s, v5.4s") // v5 = i3' r3' i7' r7'
// v4 = r1' i1' r5' i5'
// v5 = i3' r3' i7' r7'
// v6 = r0' i0' r4' i4'
// v7 = r2' i2' r6' i6'
__ASM_EMIT("fadd v0.4s, v6.4s, v7.4s") // v0 = r0'+r2' i0'+i2' r4'+r6' i4'+i6' = r0" i0" r4" i4"
__ASM_EMIT("fsub v1.4s, v6.4s, v7.4s") // v1 = r0'-r2' i0'-i2' r4'-r6' i4'-i6' = r1" i1" r5" i5"
__ASM_EMIT("fadd v2.4s, v4.4s, v5.4s") // v2 = r1'+i3' i1'+r3' r5'+i7' i5'+r7' = r2" i3" r6" i7"
__ASM_EMIT("fsub v3.4s, v4.4s, v5.4s") // v3 = r1'-i3' i1'-r3' r5'-i7' i5'-r7' = r3" i2" r7" i6"
__ASM_EMIT("mov v6.16b, v2.16b") // v6 = r2" i3" r6" i7"
__ASM_EMIT("mov v7.16b, v3.16b") // v7 = r3" i2" r7" i6"
__ASM_EMIT("bit v2.16b, v7.16b, v16.16b") // v2 = r2" i2" r6" i6"
__ASM_EMIT("bit v3.16b, v6.16b, v17.16b") // v3 = r3" i3" r7" i7"
__ASM_EMIT("st4 {v0.4s-v3.4s}, [%[dst]]")
__ASM_EMIT("subs %[items], %[items], #8") // items -= 8
__ASM_EMIT("add %[dst], %[dst], #0x40")
__ASM_EMIT("b.hs 1b")
__ASM_EMIT("2:")
// 4x butterflies
__ASM_EMIT("adds %[items], %[items], #4")
__ASM_EMIT("b.lt 4f")
__ASM_EMIT("ldp q2, q3, [%[dst]]") // v2 = r0 r1 r2 r3, v3 = i0 i1 i2 i3
__ASM_EMIT("zip1 v0.4s, v2.4s, v3.4s") // v0 = r0 i0 r1 i1
__ASM_EMIT("zip2 v1.4s, v2.4s, v3.4s") // v1 = r2 i2 r3 i3
__ASM_EMIT("fsub v4.4s, v0.4s, v1.4s") // v4 = r0-r2 i0-i2 r1-r3 i1-i3 = r1' i1' r3' i3'
__ASM_EMIT("fadd v5.4s, v0.4s, v1.4s") // v5 = r0+r2 i0+i2 r1+r3 i1+i3 = r0' i0' r2' i2'
__ASM_EMIT("rev64 v6.4s, v4.4s") // v6 = i1' r1' i3' r3'
__ASM_EMIT("ins v6.d[0], v5.d[1]") // v6 = r2' i2' i3' r3'
__ASM_EMIT("ins v5.d[1], v4.d[0]") // v5 = r0' i0' r1' i1'
__ASM_EMIT("fadd v0.4s, v5.4s, v6.4s") // v0 = r0'+r2' i0'+r2' r1'+i3' i1'+r3' = r0" i0" r2" i3"
__ASM_EMIT("fsub v1.4s, v5.4s, v6.4s") // v1 = r0'-r2' i0'-r2' r1'-i3' i1'-r3' = r1" i1" r3" i2"
__ASM_EMIT("zip1 v4.4s, v0.4s, v1.4s") // v4 = r0" r1" i0" i1"
__ASM_EMIT("zip2 v5.4s, v0.4s, v1.4s") // v5 = r2" r3" i3" i2"
__ASM_EMIT("rev64 v6.4s, v5.4s") // v6 = r3" r2" i2" i3"
__ASM_EMIT("ins v6.d[0], v4.d[1]") // v6 = i0" i1" i2" i3"
__ASM_EMIT("ins v4.d[1], v5.d[0]") // v4 = r0" r1" r2" r3"
__ASM_EMIT("stp q4, q6, [%[dst]]")
__ASM_EMIT("4:")
: [dst] "+r" (dst), [items] "+r" (items)
: [MASK] "r" (&fastconv_swp_mask[0])
: "cc", "memory",
"v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
"v16", "v17"
);
}
}
}
#endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_FASTCONV_PARSE_H_ */
|
