Debian Codegraph @

modules/lsp-dsp-lib/include/private/dsp/arch/aarch64/asimd/fft/scramble.h

raw
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 #ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_FFT_SCRAMBLE_H_
23 #define PRIVATE_DSP_ARCH_AARCH64_ASIMD_FFT_SCRAMBLE_H_
24
25 #ifndef PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL
26 #error "This header should not be included directly"
27 #endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_IMPL */
28
29 namespace lsp
30 {
31 namespace asimd
32 {
33 static inline void scramble_self_direct(float *dst_re, float *dst_im, size_t rank)
34 {
35 IF_ARCH_AARCH64(
36 size_t i, j;
37 float *d_re, *d_im;
38 float *src_re = dst_re, *src_im = dst_im;
39 size_t count = 1 << rank, rrank = 64 - rank;
40 );
41
42 // Self algorithm
43 ARCH_AARCH64_ASM(
44 // Do bit-reverse shuffle
45 __ASM_EMIT("mov %[i], #1") // i = 1
46 __ASM_EMIT("1:")
47 __ASM_EMIT("rbit %[j], %[i]") // j = reverse_bits(i)
48 __ASM_EMIT("add %[src_re], %[src_re], #4")
49 __ASM_EMIT("add %[src_im], %[src_im], #4")
50 __ASM_EMIT("lsr %[j], %[j], %[rrank]") // j = reverse_bits(i) >> rrank
51 __ASM_EMIT("cmp %[i], %[j]") // i <=> j
52 __ASM_EMIT("b.hs 2f") // if (i >= j) continue
53 __ASM_EMIT("add %[d_re], %[dst_re], %[j], LSL #2") // d_re = &dst_re[j]
54 __ASM_EMIT("add %[d_im], %[dst_im], %[j], LSL #2") // d_im = &dst_im[j]
55 __ASM_EMIT("ld1 {v0.s}[0], [%[src_re]]") // v0 = *src_re
56 __ASM_EMIT("ld1 {v1.s}[0], [%[src_im]]") // v1 = *src_im
57 __ASM_EMIT("ld1 {v2.s}[0], [%[d_re]]") // v2 = *td_re
58 __ASM_EMIT("ld1 {v3.s}[0], [%[d_im]]") // v3 = *td_im
59 __ASM_EMIT("st1 {v2.s}[0], [%[src_re]]") // *(src_re++) = s2
60 __ASM_EMIT("st1 {v3.s}[0], [%[src_im]]") // *(src_im++) = s3
61 __ASM_EMIT("st1 {v0.s}[0], [%[d_re]]") // *d_re = s0
62 __ASM_EMIT("st1 {v1.s}[0], [%[d_im]]") // *d_im = s1
63 __ASM_EMIT("2:")
64 __ASM_EMIT("add %[i], %[i], #1") // i++
65 __ASM_EMIT("cmp %[i], %[count]") // i <=> count
66 __ASM_EMIT("b.lo 1b")
67
68 // Perform x8 butterflies
69 __ASM_EMIT("3:")
70 __ASM_EMIT("ld2 {v0.4s, v1.4s}, [%[dst_re]]") // v0 = r0 r2 r4 r6, v1 = r1 r3 r5 r7
71 __ASM_EMIT("ld2 {v2.4s, v3.4s}, [%[dst_im]]") // v2 = i0 i2 i4 i6, v3 = i1 i3 i5 i7
72 __ASM_EMIT("fadd v6.4s, v0.4s, v1.4s") // v6 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6'
73 __ASM_EMIT("fadd v7.4s, v2.4s, v3.4s") // v7 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6'
74 __ASM_EMIT("fsub v0.4s, v0.4s, v1.4s") // v0 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7'
75 __ASM_EMIT("fsub v1.4s, v2.4s, v3.4s") // v1 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7'
76 // v6 = r0' r2' r4' r6'
77 // v0 = r1' r3' r5' r7'
78 // v7 = i0' i2' i4' i6'
79 // v1 = i1' i3' i5' i7'
80 __ASM_EMIT("uzp1 v4.4s, v6.4s, v0.4s") // v4 = r0' r4' r1' r5'
81 __ASM_EMIT("uzp2 v0.4s, v6.4s, v0.4s") // v0 = r2' r6' r3' r7'
82 __ASM_EMIT("uzp1 v5.4s, v7.4s, v1.4s") // v5 = i0' i4' i1' i5'
83 __ASM_EMIT("uzp2 v1.4s, v7.4s, v1.4s") // v1 = i2' i6' i3' i7'
84 __ASM_EMIT("mov v2.16b, v0.16b") // v2 = r2' r6' r3' r7'
85 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = r2' r6' i3' i7'
86 __ASM_EMIT("ins v1.d[1], v2.d[1]") // v1 = i2' i6' r3' r7'
87 __ASM_EMIT("fadd v2.4s, v4.4s, v0.4s") // v2 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r1" r5"
88 __ASM_EMIT("fsub v3.4s, v4.4s, v0.4s") // v3 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r3" r7"
89 __ASM_EMIT("fadd v0.4s, v5.4s, v1.4s") // v0 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i3" i7"
90 __ASM_EMIT("fsub v1.4s, v5.4s, v1.4s") // v1 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i1" i5"
91 // v0 = i0" i4" i3" i7"
92 // v1 = i2" i6" i1" i5"
93 // v2 = r0" r4" r1" r5"
94 // v3 = r2" r6" r3" r7"
95 __ASM_EMIT("mov v4.16b, v0.16b") // v4 = i0" i4" i3" i7"
96 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = i0" i4" i1" i5"
97 __ASM_EMIT("ins v1.d[1], v4.d[1]") // v1 = i2" i6" i3" i7"
98 __ASM_EMIT("uzp1 v4.4s, v2.4s, v3.4s") // v4 = r0" r1" r2" r3"
99 __ASM_EMIT("uzp2 v5.4s, v2.4s, v3.4s") // v5 = r4" r5" r6" r7"
100 __ASM_EMIT("uzp1 v6.4s, v0.4s, v1.4s") // v6 = i0" i1" i2" i3"
101 __ASM_EMIT("uzp2 v7.4s, v0.4s, v1.4s") // v7 = i4" i5" i6" i7"
102 // Store
103 __ASM_EMIT("stp q4, q5, [%[dst_re]]")
104 __ASM_EMIT("stp q6, q7, [%[dst_im]]")
105 __ASM_EMIT("subs %[count], %[count], #8") // i <=> count
106 __ASM_EMIT("add %[dst_re], %[dst_re], #0x20")
107 __ASM_EMIT("add %[dst_im], %[dst_im], #0x20")
108 __ASM_EMIT("b.ne 3b")
109
110 : [src_re] "+r" (src_re), [src_im] "+r" (src_im),
111 [dst_re] "+r" (dst_re), [dst_im] "+r" (dst_im),
112 [d_re] "=&r" (d_re), [d_im] "=&r" (d_im),
113 [i] "=&r" (i), [j] "=&r" (j),
114 [count] "+r" (count)
115 : [rrank] "r" (rrank)
116 : "cc", "memory",
117 "v0", "v1", "v2", "v3",
118 "v4", "v5", "v6", "v7"
119 );
120 }
121
122 static inline void scramble_copy_direct(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
123 {
124 IF_ARCH_AARCH64(
125 size_t i, j;
126 float *s_re, *s_im;
127 size_t regs = 1 << (rank - 3), rrank = 64 - (rank - 3);
128 );
129
130 ARCH_AARCH64_ASM(
131 __ASM_EMIT("eor %[i], %[i], %[i]") // i = 0
132 __ASM_EMIT("1:")
133 __ASM_EMIT("rbit %[j], %[i]") // j = reverse_bits(i)
134 __ASM_EMIT("lsr %[j], %[j], %[rrank]") // j = reverse_bits(i) >> rrank
135
136 __ASM_EMIT("add %[s_re], %[src_re], %[j], LSL #2") // s_re = &src_re[i]
137 __ASM_EMIT("add %[s_im], %[src_im], %[j], LSL #2") // s_re = &src_im[i]
138 __ASM_EMIT("ld1 {v0.s}[0], [%[s_re]]") // v0 = r0 ? ? ?
139 __ASM_EMIT("ld1 {v2.s}[0], [%[s_im]]") // v2 = i0 ? ? ?
140 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs]
141 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs]
142 __ASM_EMIT("ld1 {v0.s}[2], [%[s_re]]") // v0 = r0 ? r4 ?
143 __ASM_EMIT("ld1 {v2.s}[2], [%[s_im]]") // v2 = i0 ? i4 ?
144 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*2]
145 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*2]
146 __ASM_EMIT("ld1 {v0.s}[1], [%[s_re]]") // v0 = r0 r2 r4 ?
147 __ASM_EMIT("ld1 {v2.s}[1], [%[s_im]]") // v2 = i0 i2 i4 ?
148 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*3]
149 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*3]
150 __ASM_EMIT("ld1 {v0.s}[3], [%[s_re]]") // v0 = r0 r2 r4 r6
151 __ASM_EMIT("ld1 {v2.s}[3], [%[s_im]]") // v2 = i0 i2 i4 i6
152
153 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*4]
154 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*4]
155 __ASM_EMIT("ld1 {v1.s}[0], [%[s_re]]") // v1 = r1 ? ? ?
156 __ASM_EMIT("ld1 {v3.s}[0], [%[s_im]]") // v3 = i1 ? ? ?
157 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*5]
158 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*5]
159 __ASM_EMIT("ld1 {v1.s}[2], [%[s_re]]") // v1 = r1 ? r5 ?
160 __ASM_EMIT("ld1 {v3.s}[2], [%[s_im]]") // v3 = i1 ? i5 ?
161 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*6]
162 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*6]
163 __ASM_EMIT("ld1 {v1.s}[1], [%[s_re]]") // v1 = r1 r3 r5 ?
164 __ASM_EMIT("ld1 {v3.s}[1], [%[s_im]]") // v3 = i1 i3 i5 ?
165 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*7]
166 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*7]
167 __ASM_EMIT("ld1 {v1.s}[3], [%[s_re]]") // v1 = r1 r3 r5 r7
168 __ASM_EMIT("ld1 {v3.s}[3], [%[s_im]]") // v3 = i1 i3 i5 i7
169 __ASM_EMIT("add %[i], %[i], #1")
170 // v0 = r0 r2 r4 r6
171 // v1 = r1 r3 r5 r7
172 // v2 = i0 i2 i4 i6
173 // v3 = i1 i3 i5 i7
174 __ASM_EMIT("fadd v6.4s, v0.4s, v1.4s") // v6 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6'
175 __ASM_EMIT("fadd v7.4s, v2.4s, v3.4s") // v7 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6'
176 __ASM_EMIT("fsub v0.4s, v0.4s, v1.4s") // v0 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7'
177 __ASM_EMIT("fsub v1.4s, v2.4s, v3.4s") // v1 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7'
178 // v6 = r0' r2' r4' r6'
179 // v0 = r1' r3' r5' r7'
180 // v7 = i0' i2' i4' i6'
181 // v1 = i1' i3' i5' i7'
182 __ASM_EMIT("uzp1 v4.4s, v6.4s, v0.4s") // v4 = r0' r4' r1' r5'
183 __ASM_EMIT("uzp2 v0.4s, v6.4s, v0.4s") // v0 = r2' r6' r3' r7'
184 __ASM_EMIT("uzp1 v5.4s, v7.4s, v1.4s") // v5 = i0' i4' i1' i5'
185 __ASM_EMIT("uzp2 v1.4s, v7.4s, v1.4s") // v1 = i2' i6' i3' i7'
186 __ASM_EMIT("mov v2.16b, v0.16b") // v2 = r2' r6' r3' r7'
187 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = r2' r6' i3' i7'
188 __ASM_EMIT("ins v1.d[1], v2.d[1]") // v1 = i2' i6' r3' r7'
189 __ASM_EMIT("fadd v2.4s, v4.4s, v0.4s") // v2 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r1" r5"
190 __ASM_EMIT("fsub v3.4s, v4.4s, v0.4s") // v3 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r3" r7"
191 __ASM_EMIT("fadd v0.4s, v5.4s, v1.4s") // v0 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i3" i7"
192 __ASM_EMIT("fsub v1.4s, v5.4s, v1.4s") // v1 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i1" i5"
193 // v0 = i0" i4" i3" i7"
194 // v1 = i2" i6" i1" i5"
195 // v2 = r0" r4" r1" r5"
196 // v3 = r2" r6" r3" r7"
197 __ASM_EMIT("mov v4.16b, v0.16b") // v4 = i0" i4" i3" i7"
198 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = i0" i4" i1" i5"
199 __ASM_EMIT("ins v1.d[1], v4.d[1]") // v1 = i2" i6" i3" i7"
200 __ASM_EMIT("uzp1 v4.4s, v2.4s, v3.4s") // v4 = r0" r1" r2" r3"
201 __ASM_EMIT("uzp2 v5.4s, v2.4s, v3.4s") // v5 = r4" r5" r6" r7"
202 __ASM_EMIT("uzp1 v6.4s, v0.4s, v1.4s") // v6 = i0" i1" i2" i3"
203 __ASM_EMIT("uzp2 v7.4s, v0.4s, v1.4s") // v7 = i4" i5" i6" i7"
204 // Store
205 __ASM_EMIT("stp q4, q5, [%[dst_re]]")
206 __ASM_EMIT("stp q6, q7, [%[dst_im]]")
207 __ASM_EMIT("cmp %[i], %[regs]")
208 __ASM_EMIT("add %[dst_re], %[dst_re], #0x20")
209 __ASM_EMIT("add %[dst_im], %[dst_im], #0x20")
210 __ASM_EMIT("b.lo 1b")
211
212 : [dst_re] "+r" (dst_re), [dst_im] "+r" (dst_im),
213 [s_re] "=&r" (s_re), [s_im] "=&r" (s_im),
214 [i] "=&r" (i), [j] "=&r" (j)
215 : [src_re] "r" (src_re), [src_im] "r" (src_im),
216 [rrank] "r" (rrank), [regs] "r" (regs)
217 : "cc", "memory",
218 "v0", "v1", "v2", "v3",
219 "v4", "v5", "v6", "v7"
220 );
221 }
222
223 static inline void scramble_self_reverse(float *dst_re, float *dst_im, size_t rank)
224 {
225 IF_ARCH_AARCH64(
226 size_t i, j;
227 float *d_re, *d_im;
228 size_t count = 1 << rank, rrank = 64 - rank;
229 float *src_re = dst_re, *src_im = dst_im;
230 );
231
232 // Self algorithm
233 ARCH_AARCH64_ASM(
234 // Do bit-reverse shuffle
235 __ASM_EMIT("mov %[i], #1") // i = 1
236 __ASM_EMIT("1:")
237 __ASM_EMIT("rbit %[j], %[i]") // j = reverse_bits(i)
238 __ASM_EMIT("add %[src_re], %[src_re], #4")
239 __ASM_EMIT("add %[src_im], %[src_im], #4")
240 __ASM_EMIT("lsr %[j], %[j], %[rrank]") // j = reverse_bits(i) >> rrank
241 __ASM_EMIT("cmp %[i], %[j]") // i <=> j
242 __ASM_EMIT("b.hs 2f") // if (i >= j) continue
243 __ASM_EMIT("add %[d_re], %[dst_re], %[j], LSL #2") // d_re = &dst_re[j]
244 __ASM_EMIT("add %[d_im], %[dst_im], %[j], LSL #2") // d_im = &dst_im[j]
245 __ASM_EMIT("ld1 {v0.s}[0], [%[src_re]]") // v0 = *src_re
246 __ASM_EMIT("ld1 {v1.s}[0], [%[src_im]]") // v1 = *src_im
247 __ASM_EMIT("ld1 {v2.s}[0], [%[d_re]]") // v2 = *td_re
248 __ASM_EMIT("ld1 {v3.s}[0], [%[d_im]]") // v3 = *td_im
249 __ASM_EMIT("st1 {v2.s}[0], [%[src_re]]") // *(src_re++) = s2
250 __ASM_EMIT("st1 {v3.s}[0], [%[src_im]]") // *(src_im++) = s3
251 __ASM_EMIT("st1 {v0.s}[0], [%[d_re]]") // *d_re = s0
252 __ASM_EMIT("st1 {v1.s}[0], [%[d_im]]") // *d_im = s1
253 __ASM_EMIT("2:")
254 __ASM_EMIT("add %[i], %[i], #1") // i++
255 __ASM_EMIT("cmp %[i], %[count]") // i <=> count
256 __ASM_EMIT("b.lo 1b")
257 // Perform x8 butterflies
258 __ASM_EMIT("3:")
259 __ASM_EMIT("ld2 {v0.4s, v1.4s}, [%[dst_re]]") // v0 = r0 r2 r4 r6, v1 = r1 r3 r5 r7
260 __ASM_EMIT("ld2 {v2.4s, v3.4s}, [%[dst_im]]") // v2 = i0 i2 i4 i6, v3 = i1 i3 i5 i7
261 __ASM_EMIT("fadd v6.4s, v0.4s, v1.4s") // v6 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6'
262 __ASM_EMIT("fadd v7.4s, v2.4s, v3.4s") // v7 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6'
263 __ASM_EMIT("fsub v0.4s, v0.4s, v1.4s") // v0 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7'
264 __ASM_EMIT("fsub v1.4s, v2.4s, v3.4s") // v1 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7'
265 // v6 = r0' r2' r4' r6'
266 // v0 = r1' r3' r5' r7'
267 // v7 = i0' i2' i4' i6'
268 // v1 = i1' i3' i5' i7'
269 __ASM_EMIT("uzp1 v4.4s, v6.4s, v0.4s") // v4 = r0' r4' r1' r5'
270 __ASM_EMIT("uzp2 v0.4s, v6.4s, v0.4s") // v0 = r2' r6' r3' r7'
271 __ASM_EMIT("uzp1 v5.4s, v7.4s, v1.4s") // v5 = i0' i4' i1' i5'
272 __ASM_EMIT("uzp2 v1.4s, v7.4s, v1.4s") // v1 = i2' i6' i3' i7'
273 __ASM_EMIT("mov v2.16b, v0.16b") // v2 = r2' r6' r3' r7'
274 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = r2' r6' i3' i7'
275 __ASM_EMIT("ins v1.d[1], v2.d[1]") // v1 = i2' i6' r3' r7'
276 __ASM_EMIT("fadd v2.4s, v4.4s, v0.4s") // v2 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7"
277 __ASM_EMIT("fsub v3.4s, v4.4s, v0.4s") // v3 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5"
278 __ASM_EMIT("fadd v0.4s, v5.4s, v1.4s") // v0 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5"
279 __ASM_EMIT("fsub v1.4s, v5.4s, v1.4s") // v1 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7"
280 // v0 = i0" i4" i1" i5"
281 // v1 = i2" i6" i3" i7"
282 // v2 = r0" r4" r3" r7"
283 // v3 = r2" r6" r1" r5"
284 __ASM_EMIT("mov v4.16b, v2.16b") // v4 = r0" r4" r3" r7"
285 __ASM_EMIT("ins v2.d[1], v3.d[1]") // v2 = r0" r4" r1" r5"
286 __ASM_EMIT("ins v3.d[1], v4.d[1]") // v3 = r2" r6" r3" r7"
287 __ASM_EMIT("uzp1 v4.4s, v2.4s, v3.4s") // v4 = r0" r1" r2" r3"
288 __ASM_EMIT("uzp2 v5.4s, v2.4s, v3.4s") // v5 = r4" r5" r6" r7"
289 __ASM_EMIT("uzp1 v6.4s, v0.4s, v1.4s") // v6 = i0" i1" i2" i3"
290 __ASM_EMIT("uzp2 v7.4s, v0.4s, v1.4s") // v7 = i4" i5" i6" i7"
291 // Store
292 __ASM_EMIT("stp q4, q5, [%[dst_re]]")
293 __ASM_EMIT("stp q6, q7, [%[dst_im]]")
294 __ASM_EMIT("subs %[count], %[count], #8") // i <=> count
295 __ASM_EMIT("add %[dst_re], %[dst_re], #0x20")
296 __ASM_EMIT("add %[dst_im], %[dst_im], #0x20")
297 __ASM_EMIT("b.ne 3b")
298
299 : [src_re] "+r" (src_re), [src_im] "+r" (src_im),
300 [dst_re] "+r" (dst_re), [dst_im] "+r" (dst_im),
301 [d_re] "=&r" (d_re), [d_im] "=&r" (d_im),
302 [i] "=&r" (i), [j] "=&r" (j),
303 [count] "+r" (count)
304 : [rrank] "r" (rrank)
305 : "cc", "memory",
306 "v0", "v1", "v2", "v3",
307 "v4", "v5", "v6", "v7"
308 );
309 }
310
311 static inline void scramble_copy_reverse(float *dst_re, float *dst_im, const float *src_re, const float *src_im, size_t rank)
312 {
313 IF_ARCH_AARCH64(
314 size_t i, j;
315 float *s_re, *s_im;
316 size_t regs = 1 << (rank - 3), rrank = 64 - (rank - 3);
317 );
318
319 ARCH_AARCH64_ASM(
320 __ASM_EMIT("eor %[i], %[i], %[i]") // i = 0
321 __ASM_EMIT("1:")
322 __ASM_EMIT("rbit %[j], %[i]") // j = reverse_bits(i)
323 __ASM_EMIT("lsr %[j], %[j], %[rrank]") // j = reverse_bits(i) >> rrank
324
325 __ASM_EMIT("add %[s_re], %[src_re], %[j], LSL #2") // s_re = &src_re[i]
326 __ASM_EMIT("add %[s_im], %[src_im], %[j], LSL #2") // s_re = &src_im[i]
327 __ASM_EMIT("ld1 {v0.s}[0], [%[s_re]]") // v0 = r0 ? ? ?
328 __ASM_EMIT("ld1 {v2.s}[0], [%[s_im]]") // v2 = i0 ? ? ?
329 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs]
330 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs]
331 __ASM_EMIT("ld1 {v0.s}[2], [%[s_re]]") // v0 = r0 ? r4 ?
332 __ASM_EMIT("ld1 {v2.s}[2], [%[s_im]]") // v2 = i0 ? i4 ?
333 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*2]
334 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*2]
335 __ASM_EMIT("ld1 {v0.s}[1], [%[s_re]]") // v0 = r0 r2 r4 ?
336 __ASM_EMIT("ld1 {v2.s}[1], [%[s_im]]") // v2 = i0 i2 i4 ?
337 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*3]
338 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*3]
339 __ASM_EMIT("ld1 {v0.s}[3], [%[s_re]]") // v0 = r0 r2 r4 r6
340 __ASM_EMIT("ld1 {v2.s}[3], [%[s_im]]") // v2 = i0 i2 i4 i6
341
342 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*4]
343 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*4]
344 __ASM_EMIT("ld1 {v1.s}[0], [%[s_re]]") // v1 = r1 ? ? ?
345 __ASM_EMIT("ld1 {v3.s}[0], [%[s_im]]") // v3 = i1 ? ? ?
346 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*5]
347 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*5]
348 __ASM_EMIT("ld1 {v1.s}[2], [%[s_re]]") // v1 = r1 ? r5 ?
349 __ASM_EMIT("ld1 {v3.s}[2], [%[s_im]]") // v3 = i1 ? i5 ?
350 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*6]
351 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*6]
352 __ASM_EMIT("ld1 {v1.s}[1], [%[s_re]]") // v1 = r1 r3 r5 ?
353 __ASM_EMIT("ld1 {v3.s}[1], [%[s_im]]") // v3 = i1 i3 i5 ?
354 __ASM_EMIT("add %[s_re], %[s_re], %[regs], LSL #2") // s_re = &src_re[i + regs*7]
355 __ASM_EMIT("add %[s_im], %[s_im], %[regs], LSL #2") // s_re = &src_im[i + regs*7]
356 __ASM_EMIT("ld1 {v1.s}[3], [%[s_re]]") // v1 = r1 r3 r5 r7
357 __ASM_EMIT("ld1 {v3.s}[3], [%[s_im]]") // v3 = i1 i3 i5 i7
358 __ASM_EMIT("add %[i], %[i], #1")
359 // v0 = r0 r2 r4 r6
360 // v1 = r1 r3 r5 r7
361 // v2 = i0 i2 i4 i6
362 // v3 = i1 i3 i5 i7
363 __ASM_EMIT("fadd v6.4s, v0.4s, v1.4s") // v6 = r0+r1 r2+r3 r4+r5 r6+r7 = r0' r2' r4' r6'
364 __ASM_EMIT("fadd v7.4s, v2.4s, v3.4s") // v7 = i0+i1 i2+i3 i4+i5 i6+i7 = i0' i2' i4' i6'
365 __ASM_EMIT("fsub v0.4s, v0.4s, v1.4s") // v0 = r0-r1 r2-r3 r4-r5 r6-r7 = r1' r3' r5' r7'
366 __ASM_EMIT("fsub v1.4s, v2.4s, v3.4s") // v1 = i0-i1 i2-i3 i4-i5 i6-i7 = i1' i3' i5' i7'
367 // v6 = r0' r2' r4' r6'
368 // v0 = r1' r3' r5' r7'
369 // v7 = i0' i2' i4' i6'
370 // v1 = i1' i3' i5' i7'
371 __ASM_EMIT("uzp1 v4.4s, v6.4s, v0.4s") // v4 = r0' r4' r1' r5'
372 __ASM_EMIT("uzp2 v0.4s, v6.4s, v0.4s") // v0 = r2' r6' r3' r7'
373 __ASM_EMIT("uzp1 v5.4s, v7.4s, v1.4s") // v5 = i0' i4' i1' i5'
374 __ASM_EMIT("uzp2 v1.4s, v7.4s, v1.4s") // v1 = i2' i6' i3' i7'
375 __ASM_EMIT("mov v2.16b, v0.16b") // v2 = r2' r6' r3' r7'
376 __ASM_EMIT("ins v0.d[1], v1.d[1]") // v0 = r2' r6' i3' i7'
377 __ASM_EMIT("ins v1.d[1], v2.d[1]") // v1 = i2' i6' r3' r7'
378 __ASM_EMIT("fadd v2.4s, v4.4s, v0.4s") // v2 = r0'+r2' r4'+r6' r1'+i3' r5'+i7' = r0" r4" r3" r7"
379 __ASM_EMIT("fsub v3.4s, v4.4s, v0.4s") // v3 = r0'-r2' r4'-r6' r1'-i3' r5'-i7' = r2" r6" r1" r5"
380 __ASM_EMIT("fadd v0.4s, v5.4s, v1.4s") // v0 = i0'+i2' i4'+i6' i1'+r3' i5'+r7' = i0" i4" i1" i5"
381 __ASM_EMIT("fsub v1.4s, v5.4s, v1.4s") // v1 = i0'-i2' i4'-i6' i1'-r3' i5'-r7' = i2" i6" i3" i7"
382 // v0 = i0" i4" i1" i5"
383 // v1 = i2" i6" i3" i7"
384 // v2 = r0" r4" r3" r7"
385 // v3 = r2" r6" r1" r5"
386 __ASM_EMIT("mov v4.16b, v2.16b") // v4 = r0" r4" r3" r7"
387 __ASM_EMIT("ins v2.d[1], v3.d[1]") // v2 = r0" r4" r1" r5"
388 __ASM_EMIT("ins v3.d[1], v4.d[1]") // v3 = r2" r6" r3" r7"
389 __ASM_EMIT("uzp1 v4.4s, v2.4s, v3.4s") // v4 = r0" r1" r2" r3"
390 __ASM_EMIT("uzp2 v5.4s, v2.4s, v3.4s") // v5 = r4" r5" r6" r7"
391 __ASM_EMIT("uzp1 v6.4s, v0.4s, v1.4s") // v6 = i0" i1" i2" i3"
392 __ASM_EMIT("uzp2 v7.4s, v0.4s, v1.4s") // v7 = i4" i5" i6" i7"
393 // Store
394 __ASM_EMIT("stp q4, q5, [%[dst_re]]")
395 __ASM_EMIT("stp q6, q7, [%[dst_im]]")
396 __ASM_EMIT("cmp %[i], %[regs]")
397 __ASM_EMIT("add %[dst_re], %[dst_re], #0x20")
398 __ASM_EMIT("add %[dst_im], %[dst_im], #0x20")
399 __ASM_EMIT("blo 1b")
400
401 : [dst_re] "+r" (dst_re), [dst_im] "+r" (dst_im),
402 [s_re] "=&r" (s_re), [s_im] "=&r" (s_im),
403 [i] "=&r" (i), [j] "=&r" (j)
404 : [src_re] "r" (src_re), [src_im] "r" (src_im),
405 [rrank] "r" (rrank), [regs] "r" (regs)
406 : "cc", "memory",
407 "v0", "v1", "v2", "v3",
408 "v4", "v5", "v6", "v7"
409 );
410 }
411 }
412 }
413
414 #endif /* PRIVATE_DSP_ARCH_AARCH64_ASIMD_FFT_SCRAMBLE_H_ */