Browse code
Merge commit 'f4d5a2cc35fcdf06ec031fabe8b0710e995fe924'
* commit 'f4d5a2cc35fcdf06ec031fabe8b0710e995fe924':
aarch64: NEON float to s16 audio conversion
Merged-by: Michael Niedermayer <michaelni@gmx.at>
Michael Niedermayer authored on 2014/04/23 06:32:55Showing 5 changed files
- libavresample/aarch64/Makefile
- libavresample/aarch64/audio_convert_init.c
- libavresample/aarch64/audio_convert_neon.S
- libavresample/audio_convert.c
- libavresample/audio_convert.h
| 2 | 6 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,49 @@ |
| 0 |
+/* |
|
| 1 |
+ * This file is part of FFmpeg. |
|
| 2 |
+ * |
|
| 3 |
+ * FFmpeg is free software; you can redistribute it and/or |
|
| 4 |
+ * modify it under the terms of the GNU Lesser General Public |
|
| 5 |
+ * License as published by the Free Software Foundation; either |
|
| 6 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
| 7 |
+ * |
|
| 8 |
+ * FFmpeg is distributed in the hope that it will be useful, |
|
| 9 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
| 10 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
| 11 |
+ * Lesser General Public License for more details. |
|
| 12 |
+ * |
|
| 13 |
+ * You should have received a copy of the GNU Lesser General Public |
|
| 14 |
+ * License along with FFmpeg; if not, write to the Free Software |
|
| 15 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
| 16 |
+ */ |
|
| 17 |
+ |
|
| 18 |
+#include <stdint.h> |
|
| 19 |
+ |
|
| 20 |
+#include "config.h" |
|
| 21 |
+#include "libavutil/attributes.h" |
|
| 22 |
+#include "libavutil/cpu.h" |
|
| 23 |
+#include "libavutil/aarch64/cpu.h" |
|
| 24 |
+#include "libavutil/samplefmt.h" |
|
| 25 |
+#include "libavresample/audio_convert.h" |
|
| 26 |
+ |
|
| 27 |
+void ff_conv_flt_to_s16_neon(int16_t *dst, const float *src, int len); |
|
| 28 |
+void ff_conv_fltp_to_s16_neon(int16_t *dst, float *const *src, |
|
| 29 |
+ int len, int channels); |
|
| 30 |
+void ff_conv_fltp_to_s16_2ch_neon(int16_t *dst, float *const *src, |
|
| 31 |
+ int len, int channels); |
|
| 32 |
+ |
|
| 33 |
+av_cold void ff_audio_convert_init_aarch64(AudioConvert *ac) |
|
| 34 |
+{
|
|
| 35 |
+ int cpu_flags = av_get_cpu_flags(); |
|
| 36 |
+ |
|
| 37 |
+ if (have_neon(cpu_flags)) {
|
|
| 38 |
+ ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT, |
|
| 39 |
+ 0, 16, 8, "NEON", |
|
| 40 |
+ ff_conv_flt_to_s16_neon); |
|
| 41 |
+ ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP, |
|
| 42 |
+ 2, 16, 8, "NEON", |
|
| 43 |
+ ff_conv_fltp_to_s16_2ch_neon); |
|
| 44 |
+ ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP, |
|
| 45 |
+ 0, 16, 8, "NEON", |
|
| 46 |
+ ff_conv_fltp_to_s16_neon); |
|
| 47 |
+ } |
|
| 48 |
+} |
| 0 | 49 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,363 @@ |
| 0 |
+/* |
|
| 1 |
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> |
|
| 2 |
+ * Copyright (c) 2014 Janne Grunau <janne-libav@jannau.net> |
|
| 3 |
+ * |
|
| 4 |
+ * This file is part of FFmpeg. |
|
| 5 |
+ * |
|
| 6 |
+ * FFmpeg is free software; you can redistribute it and/or |
|
| 7 |
+ * modify it under the terms of the GNU Lesser General Public |
|
| 8 |
+ * License as published by the Free Software Foundation; either |
|
| 9 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
| 10 |
+ * |
|
| 11 |
+ * FFmpeg is distributed in the hope that it will be useful, |
|
| 12 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
| 13 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
| 14 |
+ * Lesser General Public License for more details. |
|
| 15 |
+ * |
|
| 16 |
+ * You should have received a copy of the GNU Lesser General Public |
|
| 17 |
+ * License along with FFmpeg; if not, write to the Free Software |
|
| 18 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
| 19 |
+ */ |
|
| 20 |
+ |
|
| 21 |
+#include "config.h" |
|
| 22 |
+#include "libavutil/aarch64/asm.S" |
|
| 23 |
+ |
|
| 24 |
+function ff_conv_flt_to_s16_neon, export=1 |
|
| 25 |
+ subs x2, x2, #8 |
|
| 26 |
+ ld1 {v0.4s}, [x1], #16
|
|
| 27 |
+ fcvtzs v4.4s, v0.4s, #31 |
|
| 28 |
+ ld1 {v1.4s}, [x1], #16
|
|
| 29 |
+ fcvtzs v5.4s, v1.4s, #31 |
|
| 30 |
+ b.eq 3f |
|
| 31 |
+ ands x12, x2, #~15 |
|
| 32 |
+ b.eq 2f |
|
| 33 |
+1: subs x12, x12, #16 |
|
| 34 |
+ sqrshrn v4.4h, v4.4s, #16 |
|
| 35 |
+ ld1 {v2.4s}, [x1], #16
|
|
| 36 |
+ fcvtzs v6.4s, v2.4s, #31 |
|
| 37 |
+ sqrshrn2 v4.8h, v5.4s, #16 |
|
| 38 |
+ ld1 {v3.4s}, [x1], #16
|
|
| 39 |
+ fcvtzs v7.4s, v3.4s, #31 |
|
| 40 |
+ sqrshrn v6.4h, v6.4s, #16 |
|
| 41 |
+ st1 {v4.8h}, [x0], #16
|
|
| 42 |
+ sqrshrn2 v6.8h, v7.4s, #16 |
|
| 43 |
+ ld1 {v0.4s}, [x1], #16
|
|
| 44 |
+ fcvtzs v4.4s, v0.4s, #31 |
|
| 45 |
+ ld1 {v1.4s}, [x1], #16
|
|
| 46 |
+ fcvtzs v5.4s, v1.4s, #31 |
|
| 47 |
+ st1 {v6.8h}, [x0], #16
|
|
| 48 |
+ b.ne 1b |
|
| 49 |
+ ands x2, x2, #15 |
|
| 50 |
+ b.eq 3f |
|
| 51 |
+2: ld1 {v2.4s}, [x1], #16
|
|
| 52 |
+ sqrshrn v4.4h, v4.4s, #16 |
|
| 53 |
+ fcvtzs v6.4s, v2.4s, #31 |
|
| 54 |
+ ld1 {v3.4s}, [x1], #16
|
|
| 55 |
+ sqrshrn2 v4.8h, v5.4s, #16 |
|
| 56 |
+ fcvtzs v7.4s, v3.4s, #31 |
|
| 57 |
+ sqrshrn v6.4h, v6.4s, #16 |
|
| 58 |
+ st1 {v4.8h}, [x0], #16
|
|
| 59 |
+ sqrshrn2 v6.8h, v7.4s, #16 |
|
| 60 |
+ st1 {v6.8h}, [x0]
|
|
| 61 |
+ ret |
|
| 62 |
+3: sqrshrn v4.4h, v4.4s, #16 |
|
| 63 |
+ sqrshrn2 v4.8h, v5.4s, #16 |
|
| 64 |
+ st1 {v4.8h}, [x0]
|
|
| 65 |
+ ret |
|
| 66 |
+endfunc |
|
| 67 |
+ |
|
| 68 |
+function ff_conv_fltp_to_s16_2ch_neon, export=1 |
|
| 69 |
+ ldp x4, x5, [x1] |
|
| 70 |
+ subs x2, x2, #8 |
|
| 71 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 72 |
+ fcvtzs v4.4s, v0.4s, #31 |
|
| 73 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 74 |
+ fcvtzs v5.4s, v1.4s, #31 |
|
| 75 |
+ ld1 {v2.4s}, [x5], #16
|
|
| 76 |
+ fcvtzs v6.4s, v2.4s, #31 |
|
| 77 |
+ ld1 {v3.4s}, [x5], #16
|
|
| 78 |
+ fcvtzs v7.4s, v3.4s, #31 |
|
| 79 |
+ b.eq 3f |
|
| 80 |
+ ands x12, x2, #~15 |
|
| 81 |
+ b.eq 2f |
|
| 82 |
+1: subs x12, x12, #16 |
|
| 83 |
+ ld1 {v16.4s}, [x4], #16
|
|
| 84 |
+ fcvtzs v20.4s, v16.4s, #31 |
|
| 85 |
+ sri v6.4s, v4.4s, #16 |
|
| 86 |
+ ld1 {v17.4s}, [x4], #16
|
|
| 87 |
+ fcvtzs v21.4s, v17.4s, #31 |
|
| 88 |
+ ld1 {v18.4s}, [x5], #16
|
|
| 89 |
+ fcvtzs v22.4s, v18.4s, #31 |
|
| 90 |
+ ld1 {v19.4s}, [x5], #16
|
|
| 91 |
+ sri v7.4s, v5.4s, #16 |
|
| 92 |
+ st1 {v6.4s}, [x0], #16
|
|
| 93 |
+ fcvtzs v23.4s, v19.4s, #31 |
|
| 94 |
+ st1 {v7.4s}, [x0], #16
|
|
| 95 |
+ sri v22.4s, v20.4s, #16 |
|
| 96 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 97 |
+ sri v23.4s, v21.4s, #16 |
|
| 98 |
+ st1 {v22.4s}, [x0], #16
|
|
| 99 |
+ fcvtzs v4.4s, v0.4s, #31 |
|
| 100 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 101 |
+ fcvtzs v5.4s, v1.4s, #31 |
|
| 102 |
+ ld1 {v2.4s}, [x5], #16
|
|
| 103 |
+ fcvtzs v6.4s, v2.4s, #31 |
|
| 104 |
+ ld1 {v3.4s}, [x5], #16
|
|
| 105 |
+ fcvtzs v7.4s, v3.4s, #31 |
|
| 106 |
+ st1 {v23.4s}, [x0], #16
|
|
| 107 |
+ b.ne 1b |
|
| 108 |
+ ands x2, x2, #15 |
|
| 109 |
+ b.eq 3f |
|
| 110 |
+2: sri v6.4s, v4.4s, #16 |
|
| 111 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 112 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 113 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 114 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 115 |
+ ld1 {v2.4s}, [x5], #16
|
|
| 116 |
+ fcvtzs v2.4s, v2.4s, #31 |
|
| 117 |
+ sri v7.4s, v5.4s, #16 |
|
| 118 |
+ ld1 {v3.4s}, [x5], #16
|
|
| 119 |
+ fcvtzs v3.4s, v3.4s, #31 |
|
| 120 |
+ sri v2.4s, v0.4s, #16 |
|
| 121 |
+ st1 {v6.4s,v7.4s}, [x0], #32
|
|
| 122 |
+ sri v3.4s, v1.4s, #16 |
|
| 123 |
+ st1 {v2.4s,v3.4s}, [x0], #32
|
|
| 124 |
+ ret |
|
| 125 |
+3: sri v6.4s, v4.4s, #16 |
|
| 126 |
+ sri v7.4s, v5.4s, #16 |
|
| 127 |
+ st1 {v6.4s,v7.4s}, [x0]
|
|
| 128 |
+ ret |
|
| 129 |
+endfunc |
|
| 130 |
+ |
|
| 131 |
+function ff_conv_fltp_to_s16_neon, export=1 |
|
| 132 |
+ cmp w3, #2 |
|
| 133 |
+ b.eq X(ff_conv_fltp_to_s16_2ch_neon) |
|
| 134 |
+ b.gt 1f |
|
| 135 |
+ ldr x1, [x1] |
|
| 136 |
+ b X(ff_conv_flt_to_s16_neon) |
|
| 137 |
+1: |
|
| 138 |
+ cmp w3, #4 |
|
| 139 |
+ lsl x12, x3, #1 |
|
| 140 |
+ b.lt 4f |
|
| 141 |
+ |
|
| 142 |
+5: // 4 channels |
|
| 143 |
+ ldp x4, x5, [x1], #16 |
|
| 144 |
+ ldp x6, x7, [x1], #16 |
|
| 145 |
+ mov w9, w2 |
|
| 146 |
+ mov x8, x0 |
|
| 147 |
+ ld1 {v4.4s}, [x4], #16
|
|
| 148 |
+ fcvtzs v4.4s, v4.4s, #31 |
|
| 149 |
+ ld1 {v5.4s}, [x5], #16
|
|
| 150 |
+ fcvtzs v5.4s, v5.4s, #31 |
|
| 151 |
+ ld1 {v6.4s}, [x6], #16
|
|
| 152 |
+ fcvtzs v6.4s, v6.4s, #31 |
|
| 153 |
+ ld1 {v7.4s}, [x7], #16
|
|
| 154 |
+ fcvtzs v7.4s, v7.4s, #31 |
|
| 155 |
+6: |
|
| 156 |
+ subs w9, w9, #8 |
|
| 157 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 158 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 159 |
+ sri v5.4s, v4.4s, #16 |
|
| 160 |
+ ld1 {v1.4s}, [x5], #16
|
|
| 161 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 162 |
+ sri v7.4s, v6.4s, #16 |
|
| 163 |
+ ld1 {v2.4s}, [x6], #16
|
|
| 164 |
+ fcvtzs v2.4s, v2.4s, #31 |
|
| 165 |
+ zip1 v16.4s, v5.4s, v7.4s |
|
| 166 |
+ ld1 {v3.4s}, [x7], #16
|
|
| 167 |
+ fcvtzs v3.4s, v3.4s, #31 |
|
| 168 |
+ zip2 v17.4s, v5.4s, v7.4s |
|
| 169 |
+ st1 {v16.d}[0], [x8], x12
|
|
| 170 |
+ sri v1.4s, v0.4s, #16 |
|
| 171 |
+ st1 {v16.d}[1], [x8], x12
|
|
| 172 |
+ sri v3.4s, v2.4s, #16 |
|
| 173 |
+ st1 {v17.d}[0], [x8], x12
|
|
| 174 |
+ zip1 v18.4s, v1.4s, v3.4s |
|
| 175 |
+ st1 {v17.d}[1], [x8], x12
|
|
| 176 |
+ zip2 v19.4s, v1.4s, v3.4s |
|
| 177 |
+ b.eq 7f |
|
| 178 |
+ ld1 {v4.4s}, [x4], #16
|
|
| 179 |
+ fcvtzs v4.4s, v4.4s, #31 |
|
| 180 |
+ st1 {v18.d}[0], [x8], x12
|
|
| 181 |
+ ld1 {v5.4s}, [x5], #16
|
|
| 182 |
+ fcvtzs v5.4s, v5.4s, #31 |
|
| 183 |
+ st1 {v18.d}[1], [x8], x12
|
|
| 184 |
+ ld1 {v6.4s}, [x6], #16
|
|
| 185 |
+ fcvtzs v6.4s, v6.4s, #31 |
|
| 186 |
+ st1 {v19.d}[0], [x8], x12
|
|
| 187 |
+ ld1 {v7.4s}, [x7], #16
|
|
| 188 |
+ fcvtzs v7.4s, v7.4s, #31 |
|
| 189 |
+ st1 {v19.d}[1], [x8], x12
|
|
| 190 |
+ b 6b |
|
| 191 |
+7: |
|
| 192 |
+ st1 {v18.d}[0], [x8], x12
|
|
| 193 |
+ st1 {v18.d}[1], [x8], x12
|
|
| 194 |
+ st1 {v19.d}[0], [x8], x12
|
|
| 195 |
+ st1 {v19.d}[1], [x8], x12
|
|
| 196 |
+ subs w3, w3, #4 |
|
| 197 |
+ b.eq end |
|
| 198 |
+ cmp w3, #4 |
|
| 199 |
+ add x0, x0, #8 |
|
| 200 |
+ b.ge 5b |
|
| 201 |
+ |
|
| 202 |
+4: // 2 channels |
|
| 203 |
+ cmp w3, #2 |
|
| 204 |
+ b.lt 4f |
|
| 205 |
+ ldp x4, x5, [x1], #16 |
|
| 206 |
+ mov w9, w2 |
|
| 207 |
+ mov x8, x0 |
|
| 208 |
+ tst w9, #8 |
|
| 209 |
+ ld1 {v4.4s}, [x4], #16
|
|
| 210 |
+ fcvtzs v4.4s, v4.4s, #31 |
|
| 211 |
+ ld1 {v5.4s}, [x5], #16
|
|
| 212 |
+ fcvtzs v5.4s, v5.4s, #31 |
|
| 213 |
+ ld1 {v6.4s}, [x4], #16
|
|
| 214 |
+ fcvtzs v6.4s, v6.4s, #31 |
|
| 215 |
+ ld1 {v7.4s}, [x5], #16
|
|
| 216 |
+ fcvtzs v7.4s, v7.4s, #31 |
|
| 217 |
+ b.eq 6f |
|
| 218 |
+ subs w9, w9, #8 |
|
| 219 |
+ b.eq 7f |
|
| 220 |
+ sri v5.4s, v4.4s, #16 |
|
| 221 |
+ ld1 {v4.4s}, [x4], #16
|
|
| 222 |
+ fcvtzs v4.4s, v4.4s, #31 |
|
| 223 |
+ st1 {v5.s}[0], [x8], x12
|
|
| 224 |
+ sri v7.4s, v6.4s, #16 |
|
| 225 |
+ st1 {v5.s}[1], [x8], x12
|
|
| 226 |
+ ld1 {v6.4s}, [x4], #16
|
|
| 227 |
+ fcvtzs v6.4s, v6.4s, #31 |
|
| 228 |
+ st1 {v5.s}[2], [x8], x12
|
|
| 229 |
+ st1 {v5.s}[3], [x8], x12
|
|
| 230 |
+ st1 {v7.s}[0], [x8], x12
|
|
| 231 |
+ st1 {v7.s}[1], [x8], x12
|
|
| 232 |
+ ld1 {v5.4s}, [x5], #16
|
|
| 233 |
+ fcvtzs v5.4s, v5.4s, #31 |
|
| 234 |
+ st1 {v7.s}[2], [x8], x12
|
|
| 235 |
+ st1 {v7.s}[3], [x8], x12
|
|
| 236 |
+ ld1 {v7.4s}, [x5], #16
|
|
| 237 |
+ fcvtzs v7.4s, v7.4s, #31 |
|
| 238 |
+6: |
|
| 239 |
+ subs w9, w9, #16 |
|
| 240 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 241 |
+ sri v5.4s, v4.4s, #16 |
|
| 242 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 243 |
+ ld1 {v1.4s}, [x5], #16
|
|
| 244 |
+ sri v7.4s, v6.4s, #16 |
|
| 245 |
+ st1 {v5.s}[0], [x8], x12
|
|
| 246 |
+ st1 {v5.s}[1], [x8], x12
|
|
| 247 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 248 |
+ st1 {v5.s}[2], [x8], x12
|
|
| 249 |
+ st1 {v5.s}[3], [x8], x12
|
|
| 250 |
+ ld1 {v2.4s}, [x4], #16
|
|
| 251 |
+ st1 {v7.s}[0], [x8], x12
|
|
| 252 |
+ fcvtzs v2.4s, v2.4s, #31 |
|
| 253 |
+ st1 {v7.s}[1], [x8], x12
|
|
| 254 |
+ ld1 {v3.4s}, [x5], #16
|
|
| 255 |
+ st1 {v7.s}[2], [x8], x12
|
|
| 256 |
+ fcvtzs v3.4s, v3.4s, #31 |
|
| 257 |
+ st1 {v7.s}[3], [x8], x12
|
|
| 258 |
+ sri v1.4s, v0.4s, #16 |
|
| 259 |
+ sri v3.4s, v2.4s, #16 |
|
| 260 |
+ b.eq 6f |
|
| 261 |
+ ld1 {v4.4s}, [x4], #16
|
|
| 262 |
+ st1 {v1.s}[0], [x8], x12
|
|
| 263 |
+ fcvtzs v4.4s, v4.4s, #31 |
|
| 264 |
+ st1 {v1.s}[1], [x8], x12
|
|
| 265 |
+ ld1 {v5.4s}, [x5], #16
|
|
| 266 |
+ st1 {v1.s}[2], [x8], x12
|
|
| 267 |
+ fcvtzs v5.4s, v5.4s, #31 |
|
| 268 |
+ st1 {v1.s}[3], [x8], x12
|
|
| 269 |
+ ld1 {v6.4s}, [x4], #16
|
|
| 270 |
+ st1 {v3.s}[0], [x8], x12
|
|
| 271 |
+ fcvtzs v6.4s, v6.4s, #31 |
|
| 272 |
+ st1 {v3.s}[1], [x8], x12
|
|
| 273 |
+ ld1 {v7.4s}, [x5], #16
|
|
| 274 |
+ st1 {v3.s}[2], [x8], x12
|
|
| 275 |
+ fcvtzs v7.4s, v7.4s, #31 |
|
| 276 |
+ st1 {v3.s}[3], [x8], x12
|
|
| 277 |
+ b.gt 6b |
|
| 278 |
+6: |
|
| 279 |
+ st1 {v1.s}[0], [x8], x12
|
|
| 280 |
+ st1 {v1.s}[1], [x8], x12
|
|
| 281 |
+ st1 {v1.s}[2], [x8], x12
|
|
| 282 |
+ st1 {v1.s}[3], [x8], x12
|
|
| 283 |
+ st1 {v3.s}[0], [x8], x12
|
|
| 284 |
+ st1 {v3.s}[1], [x8], x12
|
|
| 285 |
+ st1 {v3.s}[2], [x8], x12
|
|
| 286 |
+ st1 {v3.s}[3], [x8], x12
|
|
| 287 |
+ b 8f |
|
| 288 |
+7: |
|
| 289 |
+ sri v5.4s, v4.4s, #16 |
|
| 290 |
+ sri v7.4s, v6.4s, #16 |
|
| 291 |
+ st1 {v5.s}[0], [x8], x12
|
|
| 292 |
+ st1 {v5.s}[1], [x8], x12
|
|
| 293 |
+ st1 {v5.s}[2], [x8], x12
|
|
| 294 |
+ st1 {v5.s}[3], [x8], x12
|
|
| 295 |
+ st1 {v7.s}[0], [x8], x12
|
|
| 296 |
+ st1 {v7.s}[1], [x8], x12
|
|
| 297 |
+ st1 {v7.s}[2], [x8], x12
|
|
| 298 |
+ st1 {v7.s}[3], [x8], x12
|
|
| 299 |
+8: |
|
| 300 |
+ subs w3, w3, #2 |
|
| 301 |
+ add x0, x0, #4 |
|
| 302 |
+ b.eq end |
|
| 303 |
+ |
|
| 304 |
+4: // 1 channel |
|
| 305 |
+ ldr x4, [x1] |
|
| 306 |
+ tst w2, #8 |
|
| 307 |
+ mov w9, w2 |
|
| 308 |
+ mov x5, x0 |
|
| 309 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 310 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 311 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 312 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 313 |
+ b.ne 8f |
|
| 314 |
+6: |
|
| 315 |
+ subs w9, w9, #16 |
|
| 316 |
+ ld1 {v2.4s}, [x4], #16
|
|
| 317 |
+ fcvtzs v2.4s, v2.4s, #31 |
|
| 318 |
+ ld1 {v3.4s}, [x4], #16
|
|
| 319 |
+ fcvtzs v3.4s, v3.4s, #31 |
|
| 320 |
+ st1 {v0.h}[1], [x5], x12
|
|
| 321 |
+ st1 {v0.h}[3], [x5], x12
|
|
| 322 |
+ st1 {v0.h}[5], [x5], x12
|
|
| 323 |
+ st1 {v0.h}[7], [x5], x12
|
|
| 324 |
+ st1 {v1.h}[1], [x5], x12
|
|
| 325 |
+ st1 {v1.h}[3], [x5], x12
|
|
| 326 |
+ st1 {v1.h}[5], [x5], x12
|
|
| 327 |
+ st1 {v1.h}[7], [x5], x12
|
|
| 328 |
+ b.eq 7f |
|
| 329 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 330 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 331 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 332 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 333 |
+7: |
|
| 334 |
+ st1 {v2.h}[1], [x5], x12
|
|
| 335 |
+ st1 {v2.h}[3], [x5], x12
|
|
| 336 |
+ st1 {v2.h}[5], [x5], x12
|
|
| 337 |
+ st1 {v2.h}[7], [x5], x12
|
|
| 338 |
+ st1 {v3.h}[1], [x5], x12
|
|
| 339 |
+ st1 {v3.h}[3], [x5], x12
|
|
| 340 |
+ st1 {v3.h}[5], [x5], x12
|
|
| 341 |
+ st1 {v3.h}[7], [x5], x12
|
|
| 342 |
+ b.gt 6b |
|
| 343 |
+ ret |
|
| 344 |
+8: |
|
| 345 |
+ subs w9, w9, #8 |
|
| 346 |
+ st1 {v0.h}[1], [x5], x12
|
|
| 347 |
+ st1 {v0.h}[3], [x5], x12
|
|
| 348 |
+ st1 {v0.h}[5], [x5], x12
|
|
| 349 |
+ st1 {v0.h}[7], [x5], x12
|
|
| 350 |
+ st1 {v1.h}[1], [x5], x12
|
|
| 351 |
+ st1 {v1.h}[3], [x5], x12
|
|
| 352 |
+ st1 {v1.h}[5], [x5], x12
|
|
| 353 |
+ st1 {v1.h}[7], [x5], x12
|
|
| 354 |
+ b.eq end |
|
| 355 |
+ ld1 {v0.4s}, [x4], #16
|
|
| 356 |
+ fcvtzs v0.4s, v0.4s, #31 |
|
| 357 |
+ ld1 {v1.4s}, [x4], #16
|
|
| 358 |
+ fcvtzs v1.4s, v1.4s, #31 |
|
| 359 |
+ b 6b |
|
| 360 |
+end: |
|
| 361 |
+ ret |
|
| 362 |
+endfunc |
| ... | ... |
@@ -96,6 +96,7 @@ int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in); |
| 96 | 96 |
|
| 97 | 97 |
/* arch-specific initialization functions */ |
| 98 | 98 |
|
| 99 |
+void ff_audio_convert_init_aarch64(AudioConvert *ac); |
|
| 99 | 100 |
void ff_audio_convert_init_arm(AudioConvert *ac); |
| 100 | 101 |
void ff_audio_convert_init_x86(AudioConvert *ac); |
| 101 | 102 |
|