| 1 |
/* |
| 2 |
* Copyright © 2018, VideoLAN and dav1d authors |
| 3 |
* Copyright © 2018, Janne Grunau |
| 4 |
* Copyright © 2020, Martin Storsjo |
| 5 |
* All rights reserved. |
| 6 |
* |
| 7 |
* Redistribution and use in source and binary forms, with or without |
| 8 |
* modification, are permitted provided that the following conditions are met: |
| 9 |
* |
| 10 |
* 1. Redistributions of source code must retain the above copyright notice, this |
| 11 |
* list of conditions and the following disclaimer. |
| 12 |
* |
| 13 |
* 2. Redistributions in binary form must reproduce the above copyright notice, |
| 14 |
* this list of conditions and the following disclaimer in the documentation |
| 15 |
* and/or other materials provided with the distribution. |
| 16 |
* |
| 17 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND |
| 18 |
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
| 19 |
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 20 |
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR |
| 21 |
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
| 22 |
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 23 |
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
| 24 |
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 25 |
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 26 |
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 27 |
*/ |
| 28 |
|
| 29 |
#include "src/arm/asm.S" |
| 30 |
#include "util.S" |
| 31 |
|
| 32 |
#define PREP_BIAS 8192 |
| 33 |
|
| 34 |
.macro avg d0, d00, d01, d1, d10, d11 |
| 35 |
vld1.16 {q0, q1}, [r2, :128]! |
| 36 |
vld1.16 {q2, q3}, [r3, :128]! |
| 37 |
vqadd.s16 q0, q0, q2 |
| 38 |
vqadd.s16 q1, q1, q3 |
| 39 |
vmax.s16 q0, q0, q12 // -2*PREP_BIAS - 1 << intermediate_bits |
| 40 |
vmax.s16 q1, q1, q12 // -2*PREP_BIAS - 1 << intermediate_bits |
| 41 |
vqsub.s16 q0, q0, q12 // -2*PREP_BIAS - 1 << intermediate_bits |
| 42 |
vqsub.s16 q1, q1, q12 // -2*PREP_BIAS - 1 << intermediate_bits |
| 43 |
vshl.s16 \d0, q0, q13 // -(intermediate_bits+1) |
| 44 |
vshl.s16 \d1, q1, q13 // -(intermediate_bits+1) |
| 45 |
.endm |
| 46 |
|
| 47 |
.macro w_avg d0, d00, d01, d1, d10, d11 |
| 48 |
vld1.16 {q0, q1}, [r2, :128]! |
| 49 |
vld1.16 {q2, q3}, [r3, :128]! |
| 50 |
// This difference requires a 17 bit range, and all bits are |
| 51 |
// significant for the following multiplication. |
| 52 |
vsubl.s16 \d0, d4, d0 |
| 53 |
vsubl.s16 q0, d5, d1 |
| 54 |
vsubl.s16 \d1, d6, d2 |
| 55 |
vsubl.s16 q1, d7, d3 |
| 56 |
vmul.s32 \d0, \d0, q4 |
| 57 |
vmul.s32 q0, q0, q4 |
| 58 |
vmul.s32 \d1, \d1, q4 |
| 59 |
vmul.s32 q1, q1, q4 |
| 60 |
vshr.s32 \d0, \d0, #4 |
| 61 |
vshr.s32 q0, q0, #4 |
| 62 |
vshr.s32 \d1, \d1, #4 |
| 63 |
vshr.s32 q1, q1, #4 |
| 64 |
vaddw.s16 \d0, \d0, d4 |
| 65 |
vaddw.s16 q0, q0, d5 |
| 66 |
vaddw.s16 \d1, \d1, d6 |
| 67 |
vaddw.s16 q1, q1, d7 |
| 68 |
vmovn.i32 \d00, \d0 |
| 69 |
vmovn.i32 \d01, q0 |
| 70 |
vmovn.i32 \d10, \d1 |
| 71 |
vmovn.i32 \d11, q1 |
| 72 |
vrshl.s16 \d0, \d0, q13 // -intermediate_bits |
| 73 |
vrshl.s16 \d1, \d1, q13 // -intermediate_bits |
| 74 |
vadd.s16 \d0, \d0, q12 // PREP_BIAS >> intermediate_bits |
| 75 |
vadd.s16 \d1, \d1, q12 // PREP_BIAS >> intermediate_bits |
| 76 |
vmin.s16 \d0, \d0, q15 // bitdepth_max |
| 77 |
vmin.s16 \d1, \d1, q15 // bitdepth_max |
| 78 |
vmax.s16 \d0, \d0, q14 // 0 |
| 79 |
vmax.s16 \d1, \d1, q14 // 0 |
| 80 |
.endm |
| 81 |
|
| 82 |
.macro mask d0, d00, d01, d1, d10, d11 |
| 83 |
vld1.8 {q7}, [r6, :128]! |
| 84 |
vld1.16 {q0, q1}, [r2, :128]! |
| 85 |
vneg.s8 q7, q7 |
| 86 |
vld1.16 {q2, q3}, [r3, :128]! |
| 87 |
vmovl.s8 q6, d14 |
| 88 |
vmovl.s8 q7, d15 |
| 89 |
vmovl.s16 q4, d12 |
| 90 |
vmovl.s16 q5, d13 |
| 91 |
vmovl.s16 q6, d14 |
| 92 |
vmovl.s16 q7, d15 |
| 93 |
vsubl.s16 \d0, d4, d0 |
| 94 |
vsubl.s16 q0, d5, d1 |
| 95 |
vsubl.s16 \d1, d6, d2 |
| 96 |
vsubl.s16 q1, d7, d3 |
| 97 |
vmul.s32 \d0, \d0, q4 |
| 98 |
vmul.s32 q0, q0, q5 |
| 99 |
vmul.s32 \d1, \d1, q6 |
| 100 |
vmul.s32 q1, q1, q7 |
| 101 |
vshr.s32 \d0, \d0, #6 |
| 102 |
vshr.s32 q0, q0, #6 |
| 103 |
vshr.s32 \d1, \d1, #6 |
| 104 |
vshr.s32 q1, q1, #6 |
| 105 |
vaddw.s16 \d0, \d0, d4 |
| 106 |
vaddw.s16 q0, q0, d5 |
| 107 |
vaddw.s16 \d1, \d1, d6 |
| 108 |
vaddw.s16 q1, q1, d7 |
| 109 |
vmovn.i32 \d00, \d0 |
| 110 |
vmovn.i32 \d01, q0 |
| 111 |
vmovn.i32 \d10, \d1 |
| 112 |
vmovn.i32 \d11, q1 |
| 113 |
vrshl.s16 \d0, \d0, q13 // -intermediate_bits |
| 114 |
vrshl.s16 \d1, \d1, q13 // -intermediate_bits |
| 115 |
vadd.s16 \d0, \d0, q12 // PREP_BIAS >> intermediate_bits |
| 116 |
vadd.s16 \d1, \d1, q12 // PREP_BIAS >> intermediate_bits |
| 117 |
vmin.s16 \d0, \d0, q15 // bitdepth_max |
| 118 |
vmin.s16 \d1, \d1, q15 // bitdepth_max |
| 119 |
vmax.s16 \d0, \d0, q14 // 0 |
| 120 |
vmax.s16 \d1, \d1, q14 // 0 |
| 121 |
.endm |
| 122 |
|
| 123 |
.macro bidir_fn type, bdmax |
| 124 |
function \type\()_16bpc_neon, export=1 |
| 125 |
push {r4-r7,lr} |
| 126 |
ldrd r4, r5, [sp, #20] |
| 127 |
ldr r6, [sp, #28] |
| 128 |
clz r4, r4 |
| 129 |
.ifnc \type, avg |
| 130 |
ldr r7, [sp, #32] |
| 131 |
vmov.i16 q14, #0 |
| 132 |
vdup.16 q15, r7 // bitdepth_max |
| 133 |
.endif |
| 134 |
.ifc \type, w_avg |
| 135 |
vpush {q4} |
| 136 |
.endif |
| 137 |
.ifc \type, mask |
| 138 |
vpush {q4-q7} |
| 139 |
.endif |
| 140 |
clz r7, \bdmax |
| 141 |
sub r7, r7, #18 // intermediate_bits = clz(bitdepth_max) - 18 |
| 142 |
.ifc \type, avg |
| 143 |
mov lr, #1 |
| 144 |
movw r12, #2*PREP_BIAS |
| 145 |
lsl lr, lr, r7 // 1 << intermediate_bits |
| 146 |
neg r12, r12 // -2*PREP_BIAS |
| 147 |
add r7, r7, #1 |
| 148 |
sub r12, r12, lr // -2*PREP_BIAS - 1 << intermediate_bits |
| 149 |
neg r7, r7 // -(intermediate_bits+1) |
| 150 |
vdup.16 q12, r12 // -2*PREP_BIAS - 1 << intermediate_bits |
| 151 |
vdup.16 q13, r7 // -(intermediate_bits+1) |
| 152 |
.else |
| 153 |
mov r12, #PREP_BIAS |
| 154 |
lsr r12, r12, r7 // PREP_BIAS >> intermediate_bits |
| 155 |
neg r7, r7 // -intermediate_bits |
| 156 |
vdup.16 q12, r12 // PREP_BIAS >> intermediate_bits |
| 157 |
vdup.16 q13, r7 // -intermediate_bits |
| 158 |
.endif |
| 159 |
.ifc \type, w_avg |
| 160 |
vdup.32 q4, r6 |
| 161 |
vneg.s32 q4, q4 |
| 162 |
.endif |
| 163 |
adr r7, L(\type\()_tbl) |
| 164 |
sub r4, r4, #24 |
| 165 |
\type q8, d16, d17, q9, d18, d19 |
| 166 |
ldr r4, [r7, r4, lsl #2] |
| 167 |
add r7, r7, r4 |
| 168 |
bx r7 |
| 169 |
|
| 170 |
.align 2 |
| 171 |
L(\type\()_tbl): |
| 172 |
.word 1280f - L(\type\()_tbl) + CONFIG_THUMB |
| 173 |
.word 640f - L(\type\()_tbl) + CONFIG_THUMB |
| 174 |
.word 320f - L(\type\()_tbl) + CONFIG_THUMB |
| 175 |
.word 160f - L(\type\()_tbl) + CONFIG_THUMB |
| 176 |
.word 80f - L(\type\()_tbl) + CONFIG_THUMB |
| 177 |
.word 40f - L(\type\()_tbl) + CONFIG_THUMB |
| 178 |
|
| 179 |
40: |
| 180 |
add r7, r0, r1 |
| 181 |
lsl r1, r1, #1 |
| 182 |
4: |
| 183 |
subs r5, r5, #4 |
| 184 |
vst1.16 {d16}, [r0, :64], r1 |
| 185 |
vst1.16 {d17}, [r7, :64], r1 |
| 186 |
vst1.16 {d18}, [r0, :64], r1 |
| 187 |
vst1.16 {d19}, [r7, :64], r1 |
| 188 |
ble 0f |
| 189 |
\type q8, d16, d17, q9, d18, d19 |
| 190 |
b 4b |
| 191 |
80: |
| 192 |
add r7, r0, r1 |
| 193 |
lsl r1, r1, #1 |
| 194 |
8: |
| 195 |
vst1.16 {q8}, [r0, :128], r1 |
| 196 |
subs r5, r5, #2 |
| 197 |
vst1.16 {q9}, [r7, :128], r1 |
| 198 |
ble 0f |
| 199 |
\type q8, d16, d17, q9, d18, d19 |
| 200 |
b 8b |
| 201 |
160: |
| 202 |
16: |
| 203 |
\type q10, d20, d21, q11, d22, d23 |
| 204 |
vst1.16 {q8, q9}, [r0, :128], r1 |
| 205 |
subs r5, r5, #2 |
| 206 |
vst1.16 {q10, q11}, [r0, :128], r1 |
| 207 |
ble 0f |
| 208 |
\type q8, d16, d17, q9, d18, d19 |
| 209 |
b 16b |
| 210 |
320: |
| 211 |
add r7, r0, #32 |
| 212 |
32: |
| 213 |
\type q10, d20, d21, q11, d22, d23 |
| 214 |
vst1.16 {q8, q9}, [r0, :128], r1 |
| 215 |
subs r5, r5, #1 |
| 216 |
vst1.16 {q10, q11}, [r7, :128], r1 |
| 217 |
ble 0f |
| 218 |
\type q8, d16, d17, q9, d18, d19 |
| 219 |
b 32b |
| 220 |
640: |
| 221 |
add r7, r0, #32 |
| 222 |
mov r12, #64 |
| 223 |
sub r1, r1, #64 |
| 224 |
64: |
| 225 |
\type q10, d20, d21, q11, d22, d23 |
| 226 |
vst1.16 {q8, q9}, [r0, :128], r12 |
| 227 |
\type q8, d16, d17, q9, d18, d19 |
| 228 |
vst1.16 {q10, q11}, [r7, :128], r12 |
| 229 |
\type q10, d20, d21, q11, d22, d23 |
| 230 |
vst1.16 {q8, q9}, [r0, :128], r1 |
| 231 |
subs r5, r5, #1 |
| 232 |
vst1.16 {q10, q11}, [r7, :128], r1 |
| 233 |
ble 0f |
| 234 |
\type q8, d16, d17, q9, d18, d19 |
| 235 |
b 64b |
| 236 |
1280: |
| 237 |
add r7, r0, #32 |
| 238 |
mov r12, #64 |
| 239 |
sub r1, r1, #192 |
| 240 |
128: |
| 241 |
\type q10, d20, d21, q11, d22, d23 |
| 242 |
vst1.16 {q8, q9}, [r0, :128], r12 |
| 243 |
\type q8, d16, d17, q9, d18, d19 |
| 244 |
vst1.16 {q10, q11}, [r7, :128], r12 |
| 245 |
\type q10, d20, d21, q11, d22, d23 |
| 246 |
vst1.16 {q8, q9}, [r0, :128], r12 |
| 247 |
\type q8, d16, d17, q9, d18, d19 |
| 248 |
vst1.16 {q10, q11}, [r7, :128], r12 |
| 249 |
\type q10, d20, d21, q11, d22, d23 |
| 250 |
vst1.16 {q8, q9}, [r0, :128], r12 |
| 251 |
\type q8, d16, d17, q9, d18, d19 |
| 252 |
vst1.16 {q10, q11}, [r7, :128], r12 |
| 253 |
\type q10, d20, d21, q11, d22, d23 |
| 254 |
vst1.16 {q8, q9}, [r0, :128], r1 |
| 255 |
subs r5, r5, #1 |
| 256 |
vst1.16 {q10, q11}, [r7, :128], r1 |
| 257 |
ble 0f |
| 258 |
\type q8, d16, d17, q9, d18, d19 |
| 259 |
b 128b |
| 260 |
0: |
| 261 |
.ifc \type, mask |
| 262 |
vpop {q4-q7} |
| 263 |
.endif |
| 264 |
.ifc \type, w_avg |
| 265 |
vpop {q4} |
| 266 |
.endif |
| 267 |
pop {r4-r7,pc} |
| 268 |
endfunc |
| 269 |
.endm |
| 270 |
|
| 271 |
bidir_fn avg, r6 |
| 272 |
bidir_fn w_avg, r7 |
| 273 |
bidir_fn mask, r7 |
| 274 |
|
| 275 |
|
| 276 |
.macro w_mask_fn type |
| 277 |
function w_mask_\type\()_16bpc_neon, export=1 |
| 278 |
push {r4-r10,lr} |
| 279 |
vpush {q4-q7} |
| 280 |
ldrd r4, r5, [sp, #96] |
| 281 |
ldrd r6, r7, [sp, #104] |
| 282 |
ldr r8, [sp, #112] |
| 283 |
clz r9, r4 |
| 284 |
adr lr, L(w_mask_\type\()_tbl) |
| 285 |
vdup.16 q15, r8 // bitdepth_max |
| 286 |
sub r9, r9, #24 |
| 287 |
clz r8, r8 // clz(bitdepth_max) |
| 288 |
ldr r9, [lr, r9, lsl #2] |
| 289 |
add r9, lr, r9 |
| 290 |
sub r8, r8, #12 // sh = intermediate_bits + 6 = clz(bitdepth_max) - 12 |
| 291 |
mov r10, #PREP_BIAS*64 |
| 292 |
neg r8, r8 // -sh |
| 293 |
movw r12, #27615 // (64 + 1 - 38)<<mask_sh - 1 - mask_rnd |
| 294 |
vdup.32 q14, r8 // -sh |
| 295 |
vdup.16 q0, r12 |
| 296 |
.if \type == 444 |
| 297 |
vmov.i8 q1, #64 |
| 298 |
.elseif \type == 422 |
| 299 |
vdup.8 d4, r7 |
| 300 |
vmov.i8 d2, #129 |
| 301 |
vsub.i16 d2, d2, d4 |
| 302 |
.elseif \type == 420 |
| 303 |
vdup.16 q2, r7 |
| 304 |
vmov.i16 q1, #0x100 |
| 305 |
vsub.i16 q1, q1, q2 |
| 306 |
.endif |
| 307 |
add r12, r0, r1 |
| 308 |
lsl r1, r1, #1 |
| 309 |
bx r9 |
| 310 |
|
| 311 |
.align 2 |
| 312 |
L(w_mask_\type\()_tbl): |
| 313 |
.word 1280f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 314 |
.word 640f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 315 |
.word 320f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 316 |
.word 160f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 317 |
.word 8f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 318 |
.word 4f - L(w_mask_\type\()_tbl) + CONFIG_THUMB |
| 319 |
|
| 320 |
4: |
| 321 |
vld1.16 {q2, q3}, [r2, :128]! // tmp1 (four rows at once) |
| 322 |
vld1.16 {q4, q5}, [r3, :128]! // tmp2 (four rows at once) |
| 323 |
subs r5, r5, #4 |
| 324 |
vdup.32 q13, r10 // PREP_BIAS*64 |
| 325 |
vabd.s16 q6, q2, q4 // abs(tmp1 - tmp2) |
| 326 |
vabd.s16 q7, q3, q5 |
| 327 |
vsubl.s16 q8, d8, d4 // tmp2 - tmp1 (requires 17 bit) |
| 328 |
vsubl.s16 q9, d9, d5 |
| 329 |
vsubl.s16 q10, d10, d6 |
| 330 |
vsubl.s16 q11, d11, d7 |
| 331 |
vqsub.u16 q6, q0, q6 // 27615 - abs() |
| 332 |
vqsub.u16 q7, q0, q7 |
| 333 |
vshll.s16 q5, d7, #6 // tmp1 << 6 |
| 334 |
vshll.s16 q4, d6, #6 |
| 335 |
vshll.s16 q3, d5, #6 |
| 336 |
vshll.s16 q2, d4, #6 |
| 337 |
vshr.u16 q6, q6, #10 // 64-m = (27615 - abs()) >> mask_sh |
| 338 |
vshr.u16 q7, q7, #10 |
| 339 |
vadd.i32 q2, q2, q13 // += PREP_BIAS*64 |
| 340 |
vadd.i32 q3, q3, q13 |
| 341 |
vadd.i32 q4, q4, q13 |
| 342 |
vadd.i32 q5, q5, q13 |
| 343 |
vmovl.u16 q12, d12 |
| 344 |
vmovl.u16 q13, d13 |
| 345 |
vmla.i32 q2, q8, q12 // (tmp2-tmp1)*(64-m) |
| 346 |
vmovl.u16 q12, d14 |
| 347 |
vmla.i32 q3, q9, q13 |
| 348 |
vmovl.u16 q13, d15 |
| 349 |
vmla.i32 q4, q10, q12 |
| 350 |
vmla.i32 q5, q11, q13 |
| 351 |
vrshl.s32 q2, q2, q14 // (tmp1<<6 + (tmp2-tmp1)*(64-m) + (1 << (sh-1)) + PREP_BIAS*64) >> sh |
| 352 |
vrshl.s32 q3, q3, q14 |
| 353 |
vrshl.s32 q4, q4, q14 |
| 354 |
vrshl.s32 q5, q5, q14 |
| 355 |
vqmovun.s32 d4, q2 // iclip_pixel |
| 356 |
vqmovun.s32 d5, q3 |
| 357 |
vqmovun.s32 d6, q4 |
| 358 |
vqmovun.s32 d7, q5 |
| 359 |
vmin.u16 q2, q2, q15 // iclip_pixel |
| 360 |
vmin.u16 q3, q3, q15 // iclip_pixel |
| 361 |
.if \type == 444 |
| 362 |
vmovn.i16 d12, q6 // 64 - m |
| 363 |
vmovn.i16 d13, q7 |
| 364 |
vsub.i16 q6, q1, q6 // m |
| 365 |
vst1.8 {q6}, [r6, :128]! |
| 366 |
.elseif \type == 422 |
| 367 |
vpadd.i16 d12, d12, d13 // (64 - m) + (64 - n) (column wise addition) |
| 368 |
vpadd.i16 d13, d14, d15 |
| 369 |
vmovn.i16 d12, q6 |
| 370 |
vhsub.u8 d12, d2, d12 // ((129 - sign) - ((64 - m) + (64 - n)) >> 1 |
| 371 |
vst1.8 {d12}, [r6, :64]! |
| 372 |
.elseif \type == 420 |
| 373 |
vadd.i16 d12, d12, d13 // (64 - my1) + (64 - my2) (row wise addition) |
| 374 |
vadd.i16 d13, d14, d15 |
| 375 |
vpadd.i16 d12, d12, d13 // (128 - m) + (128 - n) (column wise addition) |
| 376 |
vsub.i16 d12, d2, d12 // (256 - sign) - ((128 - m) + (128 - n)) |
| 377 |
vrshrn.i16 d12, q6, #2 // ((256 - sign) - ((128 - m) + (128 - n)) + 2) >> 2 |
| 378 |
vst1.32 {d12[0]}, [r6, :32]! |
| 379 |
.endif |
| 380 |
vst1.16 {d4}, [r0, :64], r1 |
| 381 |
vst1.16 {d5}, [r12, :64], r1 |
| 382 |
vst1.16 {d6}, [r0, :64], r1 |
| 383 |
vst1.16 {d7}, [r12, :64], r1 |
| 384 |
bgt 4b |
| 385 |
vpop {q4-q7} |
| 386 |
pop {r4-r10,pc} |
| 387 |
8: |
| 388 |
vld1.16 {q2, q3}, [r2, :128]! // tmp1 |
| 389 |
vld1.16 {q4, q5}, [r3, :128]! // tmp2 |
| 390 |
subs r5, r5, #2 |
| 391 |
vdup.32 q13, r10 // PREP_BIAS*64 |
| 392 |
vabd.s16 q6, q2, q4 // abs(tmp1 - tmp2) |
| 393 |
vabd.s16 q7, q3, q5 |
| 394 |
vsubl.s16 q8, d8, d4 // tmp2 - tmp1 (requires 17 bit) |
| 395 |
vsubl.s16 q9, d9, d5 |
| 396 |
vsubl.s16 q10, d10, d6 |
| 397 |
vsubl.s16 q11, d11, d7 |
| 398 |
vqsub.u16 q6, q0, q6 // 27615 - abs() |
| 399 |
vqsub.u16 q7, q0, q7 |
| 400 |
vshll.s16 q5, d7, #6 // tmp1 << 6 |
| 401 |
vshll.s16 q4, d6, #6 |
| 402 |
vshll.s16 q3, d5, #6 |
| 403 |
vshll.s16 q2, d4, #6 |
| 404 |
vshr.u16 q6, q6, #10 // 64-m = (27615 - abs()) >> mask_sh |
| 405 |
vshr.u16 q7, q7, #10 |
| 406 |
vadd.i32 q2, q2, q13 // += PREP_BIAS*64 |
| 407 |
vadd.i32 q3, q3, q13 |
| 408 |
vadd.i32 q4, q4, q13 |
| 409 |
vadd.i32 q5, q5, q13 |
| 410 |
vmovl.u16 q12, d12 |
| 411 |
vmovl.u16 q13, d13 |
| 412 |
vmla.i32 q2, q8, q12 // (tmp2-tmp1)*(64-m) |
| 413 |
vmovl.u16 q12, d14 |
| 414 |
vmla.i32 q3, q9, q13 |
| 415 |
vmovl.u16 q13, d15 |
| 416 |
vmla.i32 q4, q10, q12 |
| 417 |
vmla.i32 q5, q11, q13 |
| 418 |
vrshl.s32 q2, q2, q14 // (tmp1<<6 + (tmp2-tmp1)*(64-m) + (1 << (sh-1)) + PREP_BIAS*64) >> sh |
| 419 |
vrshl.s32 q3, q3, q14 |
| 420 |
vrshl.s32 q4, q4, q14 |
| 421 |
vrshl.s32 q5, q5, q14 |
| 422 |
vqmovun.s32 d4, q2 // iclip_pixel |
| 423 |
vqmovun.s32 d5, q3 |
| 424 |
vqmovun.s32 d6, q4 |
| 425 |
vqmovun.s32 d7, q5 |
| 426 |
vmin.u16 q2, q2, q15 // iclip_pixel |
| 427 |
vmin.u16 q3, q3, q15 // iclip_pixel |
| 428 |
.if \type == 444 |
| 429 |
vmovn.i16 d12, q6 // 64 - m |
| 430 |
vmovn.i16 d13, q7 |
| 431 |
vsub.i16 q6, q1, q6 // m |
| 432 |
vst1.8 {q6}, [r6, :128]! |
| 433 |
.elseif \type == 422 |
| 434 |
vpadd.i16 d12, d12, d13 // (64 - m) + (64 - n) (column wise addition) |
| 435 |
vpadd.i16 d13, d14, d15 |
| 436 |
vmovn.i16 d12, q6 |
| 437 |
vhsub.u8 d12, d2, d12 // ((129 - sign) - ((64 - m) + (64 - n)) >> 1 |
| 438 |
vst1.8 {d12}, [r6, :64]! |
| 439 |
.elseif \type == 420 |
| 440 |
vadd.i16 q6, q6, q7 // (64 - my1) + (64 - my2) (row wise addition) |
| 441 |
vpadd.i16 d12, d12, d13 // (128 - m) + (128 - n) (column wise addition) |
| 442 |
vsub.i16 d12, d2, d12 // (256 - sign) - ((128 - m) + (128 - n)) |
| 443 |
vrshrn.i16 d12, q6, #2 // ((256 - sign) - ((128 - m) + (128 - n)) + 2) >> 2 |
| 444 |
vst1.32 {d12[0]}, [r6, :32]! |
| 445 |
.endif |
| 446 |
vst1.16 {q2}, [r0, :128], r1 |
| 447 |
vst1.16 {q3}, [r12, :128], r1 |
| 448 |
bgt 8b |
| 449 |
vpop {q4-q7} |
| 450 |
pop {r4-r10,pc} |
| 451 |
1280: |
| 452 |
640: |
| 453 |
320: |
| 454 |
160: |
| 455 |
sub r1, r1, r4, lsl #1 |
| 456 |
.if \type == 444 |
| 457 |
add lr, r6, r4 |
| 458 |
.elseif \type == 422 |
| 459 |
add lr, r6, r4, lsr #1 |
| 460 |
.endif |
| 461 |
add r7, r2, r4, lsl #1 |
| 462 |
add r9, r3, r4, lsl #1 |
| 463 |
161: |
| 464 |
mov r8, r4 |
| 465 |
16: |
| 466 |
vld1.16 {q2}, [r2, :128]! // tmp1 |
| 467 |
vld1.16 {q4}, [r3, :128]! // tmp2 |
| 468 |
vld1.16 {q3}, [r7, :128]! |
| 469 |
vld1.16 {q5}, [r9, :128]! |
| 470 |
subs r8, r8, #8 |
| 471 |
vdup.32 q13, r10 // PREP_BIAS*64 |
| 472 |
vabd.s16 q6, q2, q4 // abs(tmp1 - tmp2) |
| 473 |
vabd.s16 q7, q3, q5 |
| 474 |
vsubl.s16 q8, d8, d4 // tmp2 - tmp1 (requires 17 bit) |
| 475 |
vsubl.s16 q9, d9, d5 |
| 476 |
vsubl.s16 q10, d10, d6 |
| 477 |
vsubl.s16 q11, d11, d7 |
| 478 |
vqsub.u16 q6, q0, q6 // 27615 - abs() |
| 479 |
vqsub.u16 q7, q0, q7 |
| 480 |
vshll.s16 q5, d7, #6 // tmp1 << 6 |
| 481 |
vshll.s16 q4, d6, #6 |
| 482 |
vshll.s16 q3, d5, #6 |
| 483 |
vshll.s16 q2, d4, #6 |
| 484 |
vshr.u16 q6, q6, #10 // 64-m = (27615 - abs()) >> mask_sh |
| 485 |
vshr.u16 q7, q7, #10 |
| 486 |
vadd.i32 q2, q2, q13 // += PREP_BIAS*64 |
| 487 |
vadd.i32 q3, q3, q13 |
| 488 |
vadd.i32 q4, q4, q13 |
| 489 |
vadd.i32 q5, q5, q13 |
| 490 |
vmovl.u16 q12, d12 |
| 491 |
vmovl.u16 q13, d13 |
| 492 |
vmla.i32 q2, q8, q12 // (tmp2-tmp1)*(64-m) |
| 493 |
vmovl.u16 q12, d14 |
| 494 |
vmla.i32 q3, q9, q13 |
| 495 |
vmovl.u16 q13, d15 |
| 496 |
vmla.i32 q4, q10, q12 |
| 497 |
vmla.i32 q5, q11, q13 |
| 498 |
vrshl.s32 q2, q2, q14 // (tmp1<<6 + (tmp2-tmp1)*(64-m) + (1 << (sh-1)) + PREP_BIAS*64) >> sh |
| 499 |
vrshl.s32 q3, q3, q14 |
| 500 |
vrshl.s32 q4, q4, q14 |
| 501 |
vrshl.s32 q5, q5, q14 |
| 502 |
vqmovun.s32 d4, q2 // iclip_pixel |
| 503 |
vqmovun.s32 d5, q3 |
| 504 |
vqmovun.s32 d6, q4 |
| 505 |
vqmovun.s32 d7, q5 |
| 506 |
vmin.u16 q2, q2, q15 // iclip_pixel |
| 507 |
vmin.u16 q3, q3, q15 // iclip_pixel |
| 508 |
.if \type == 444 |
| 509 |
vmovn.i16 d12, q6 // 64 - m |
| 510 |
vmovn.i16 d13, q7 |
| 511 |
vsub.i16 q6, q1, q6 // m |
| 512 |
vst1.8 {d12}, [r6, :64]! |
| 513 |
vst1.8 {d13}, [lr, :64]! |
| 514 |
.elseif \type == 422 |
| 515 |
vpadd.i16 d12, d12, d13 // (64 - m) + (64 - n) (column wise addition) |
| 516 |
vpadd.i16 d13, d14, d15 |
| 517 |
vmovn.i16 d12, q6 |
| 518 |
vhsub.u8 d12, d2, d12 // ((129 - sign) - ((64 - m) + (64 - n)) >> 1 |
| 519 |
vst1.32 {d12[0]}, [r6, :32]! |
| 520 |
vst1.32 {d12[1]}, [lr, :32]! |
| 521 |
.elseif \type == 420 |
| 522 |
vadd.i16 q6, q6, q7 // (64 - my1) + (64 - my2) (row wise addition) |
| 523 |
vpadd.i16 d12, d12, d13 // (128 - m) + (128 - n) (column wise addition) |
| 524 |
vsub.i16 d12, d2, d12 // (256 - sign) - ((128 - m) + (128 - n)) |
| 525 |
vrshrn.i16 d12, q6, #2 // ((256 - sign) - ((128 - m) + (128 - n)) + 2) >> 2 |
| 526 |
vst1.32 {d12[0]}, [r6, :32]! |
| 527 |
.endif |
| 528 |
vst1.16 {q2}, [r0, :128]! |
| 529 |
vst1.16 {q3}, [r12, :128]! |
| 530 |
bgt 16b |
| 531 |
subs r5, r5, #2 |
| 532 |
add r2, r2, r4, lsl #1 |
| 533 |
add r3, r3, r4, lsl #1 |
| 534 |
add r7, r7, r4, lsl #1 |
| 535 |
add r9, r9, r4, lsl #1 |
| 536 |
.if \type == 444 |
| 537 |
add r6, r6, r4 |
| 538 |
add lr, lr, r4 |
| 539 |
.elseif \type == 422 |
| 540 |
add r6, r6, r4, lsr #1 |
| 541 |
add lr, lr, r4, lsr #1 |
| 542 |
.endif |
| 543 |
add r0, r0, r1 |
| 544 |
add r12, r12, r1 |
| 545 |
bgt 161b |
| 546 |
vpop {q4-q7} |
| 547 |
pop {r4-r10,pc} |
| 548 |
endfunc |
| 549 |
.endm |
| 550 |
|
| 551 |
w_mask_fn 444 |
| 552 |
w_mask_fn 422 |
| 553 |
w_mask_fn 420 |
| 554 |
|
| 555 |
function blend_16bpc_neon, export=1 |
| 556 |
push {r4-r5,lr} |
| 557 |
ldrd r4, r5, [sp, #12] |
| 558 |
clz lr, r3 |
| 559 |
adr r3, L(blend_tbl) |
| 560 |
sub lr, lr, #26 |
| 561 |
ldr lr, [r3, lr, lsl #2] |
| 562 |
add r3, r3, lr |
| 563 |
bx r3 |
| 564 |
|
| 565 |
.align 2 |
| 566 |
L(blend_tbl): |
| 567 |
.word 320f - L(blend_tbl) + CONFIG_THUMB |
| 568 |
.word 160f - L(blend_tbl) + CONFIG_THUMB |
| 569 |
.word 80f - L(blend_tbl) + CONFIG_THUMB |
| 570 |
.word 40f - L(blend_tbl) + CONFIG_THUMB |
| 571 |
|
| 572 |
40: |
| 573 |
add r12, r0, r1 |
| 574 |
lsl r1, r1, #1 |
| 575 |
4: |
| 576 |
vld1.8 {d4}, [r5, :64]! |
| 577 |
vld1.16 {q1}, [r2, :128]! |
| 578 |
vld1.16 {d0}, [r0, :64] |
| 579 |
vneg.s8 d4, d4 // -m |
| 580 |
subs r4, r4, #2 |
| 581 |
vld1.16 {d1}, [r12, :64] |
| 582 |
vmovl.s8 q2, d4 |
| 583 |
vshl.i16 q2, q2, #9 // -m << 9 |
| 584 |
vsub.i16 q1, q0, q1 // a - b |
| 585 |
vqrdmulh.s16 q1, q1, q2 // ((a-b)*-m + 32) >> 6 |
| 586 |
vadd.i16 q0, q0, q1 |
| 587 |
vst1.16 {d0}, [r0, :64], r1 |
| 588 |
vst1.16 {d1}, [r12, :64], r1 |
| 589 |
bgt 4b |
| 590 |
pop {r4-r5,pc} |
| 591 |
80: |
| 592 |
add r12, r0, r1 |
| 593 |
lsl r1, r1, #1 |
| 594 |
8: |
| 595 |
vld1.8 {q8}, [r5, :128]! |
| 596 |
vld1.16 {q2, q3}, [r2, :128]! |
| 597 |
vneg.s8 q9, q8 // -m |
| 598 |
vld1.16 {q0}, [r0, :128] |
| 599 |
vld1.16 {q1}, [r12, :128] |
| 600 |
vmovl.s8 q8, d18 |
| 601 |
vmovl.s8 q9, d19 |
| 602 |
vshl.i16 q8, q8, #9 // -m << 9 |
| 603 |
vshl.i16 q9, q9, #9 |
| 604 |
vsub.i16 q2, q0, q2 // a - b |
| 605 |
vsub.i16 q3, q1, q3 |
| 606 |
subs r4, r4, #2 |
| 607 |
vqrdmulh.s16 q2, q2, q8 // ((a-b)*-m + 32) >> 6 |
| 608 |
vqrdmulh.s16 q3, q3, q9 |
| 609 |
vadd.i16 q0, q0, q2 |
| 610 |
vadd.i16 q1, q1, q3 |
| 611 |
vst1.16 {q0}, [r0, :128], r1 |
| 612 |
vst1.16 {q1}, [r12, :128], r1 |
| 613 |
bgt 8b |
| 614 |
pop {r4-r5,pc} |
| 615 |
160: |
| 616 |
add r12, r0, r1 |
| 617 |
lsl r1, r1, #1 |
| 618 |
16: |
| 619 |
vld1.8 {q12, q13}, [r5, :128]! |
| 620 |
vld1.16 {q8, q9}, [r2, :128]! |
| 621 |
subs r4, r4, #2 |
| 622 |
vneg.s8 q14, q12 // -m |
| 623 |
vld1.16 {q0, q1}, [r0, :128] |
| 624 |
vneg.s8 q15, q13 |
| 625 |
vld1.16 {q10, q11}, [r2, :128]! |
| 626 |
vmovl.s8 q12, d28 |
| 627 |
vmovl.s8 q13, d29 |
| 628 |
vmovl.s8 q14, d30 |
| 629 |
vmovl.s8 q15, d31 |
| 630 |
vld1.16 {q2, q3}, [r12, :128] |
| 631 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 632 |
vshl.i16 q13, q13, #9 |
| 633 |
vshl.i16 q14, q14, #9 |
| 634 |
vshl.i16 q15, q15, #9 |
| 635 |
vsub.i16 q8, q0, q8 // a - b |
| 636 |
vsub.i16 q9, q1, q9 |
| 637 |
vsub.i16 q10, q2, q10 |
| 638 |
vsub.i16 q11, q3, q11 |
| 639 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 640 |
vqrdmulh.s16 q9, q9, q13 |
| 641 |
vqrdmulh.s16 q10, q10, q14 |
| 642 |
vqrdmulh.s16 q11, q11, q15 |
| 643 |
vadd.i16 q0, q0, q8 |
| 644 |
vadd.i16 q1, q1, q9 |
| 645 |
vadd.i16 q2, q2, q10 |
| 646 |
vst1.16 {q0, q1}, [r0, :128], r1 |
| 647 |
vadd.i16 q3, q3, q11 |
| 648 |
vst1.16 {q2, q3}, [r12, :128], r1 |
| 649 |
bgt 16b |
| 650 |
pop {r4-r5,pc} |
| 651 |
320: |
| 652 |
add r12, r0, #32 |
| 653 |
32: |
| 654 |
vld1.8 {q12, q13}, [r5, :128]! |
| 655 |
vld1.16 {q8, q9}, [r2, :128]! |
| 656 |
subs r4, r4, #1 |
| 657 |
vneg.s8 q14, q12 // -m |
| 658 |
vld1.16 {q0, q1}, [r0, :128] |
| 659 |
vneg.s8 q15, q13 |
| 660 |
vld1.16 {q10, q11}, [r2, :128]! |
| 661 |
vmovl.s8 q12, d28 |
| 662 |
vmovl.s8 q13, d29 |
| 663 |
vmovl.s8 q14, d30 |
| 664 |
vmovl.s8 q15, d31 |
| 665 |
vld1.16 {q2, q3}, [r12, :128] |
| 666 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 667 |
vshl.i16 q13, q13, #9 |
| 668 |
vshl.i16 q14, q14, #9 |
| 669 |
vshl.i16 q15, q15, #9 |
| 670 |
vsub.i16 q8, q0, q8 // a - b |
| 671 |
vsub.i16 q9, q1, q9 |
| 672 |
vsub.i16 q10, q2, q10 |
| 673 |
vsub.i16 q11, q3, q11 |
| 674 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 675 |
vqrdmulh.s16 q9, q9, q13 |
| 676 |
vqrdmulh.s16 q10, q10, q14 |
| 677 |
vqrdmulh.s16 q11, q11, q15 |
| 678 |
vadd.i16 q0, q0, q8 |
| 679 |
vadd.i16 q1, q1, q9 |
| 680 |
vadd.i16 q2, q2, q10 |
| 681 |
vst1.16 {q0, q1}, [r0, :128], r1 |
| 682 |
vadd.i16 q3, q3, q11 |
| 683 |
vst1.16 {q2, q3}, [r12, :128], r1 |
| 684 |
bgt 32b |
| 685 |
pop {r4-r5,pc} |
| 686 |
endfunc |
| 687 |
|
| 688 |
function blend_h_16bpc_neon, export=1 |
| 689 |
push {r4-r5,lr} |
| 690 |
ldr r4, [sp, #12] |
| 691 |
movrel r5, X(obmc_masks) |
| 692 |
add r5, r5, r4 |
| 693 |
sub r4, r4, r4, lsr #2 |
| 694 |
clz lr, r3 |
| 695 |
adr r12, L(blend_h_tbl) |
| 696 |
sub lr, lr, #24 |
| 697 |
ldr lr, [r12, lr, lsl #2] |
| 698 |
add r12, r12, lr |
| 699 |
bx r12 |
| 700 |
|
| 701 |
.align 2 |
| 702 |
L(blend_h_tbl): |
| 703 |
.word 1280f - L(blend_h_tbl) + CONFIG_THUMB |
| 704 |
.word 640f - L(blend_h_tbl) + CONFIG_THUMB |
| 705 |
.word 320f - L(blend_h_tbl) + CONFIG_THUMB |
| 706 |
.word 160f - L(blend_h_tbl) + CONFIG_THUMB |
| 707 |
.word 80f - L(blend_h_tbl) + CONFIG_THUMB |
| 708 |
.word 40f - L(blend_h_tbl) + CONFIG_THUMB |
| 709 |
.word 20f - L(blend_h_tbl) + CONFIG_THUMB |
| 710 |
|
| 711 |
20: |
| 712 |
add r12, r0, r1 |
| 713 |
lsl r1, r1, #1 |
| 714 |
2: |
| 715 |
vld2.8 {d4[], d5[]}, [r5, :16]! |
| 716 |
vld1.16 {d2}, [r2, :64]! |
| 717 |
vext.8 d4, d4, d5, #6 |
| 718 |
subs r4, r4, #2 |
| 719 |
vneg.s8 d4, d4 // -m |
| 720 |
vld1.32 {d0[]}, [r0, :32] |
| 721 |
vld1.32 {d0[1]}, [r12, :32] |
| 722 |
vmovl.s8 q2, d4 |
| 723 |
vshl.i16 d4, d4, #9 // -m << 9 |
| 724 |
vsub.i16 d2, d0, d2 // a - b |
| 725 |
vqrdmulh.s16 d2, d2, d4 // ((a-b)*-m + 32) >> 6 |
| 726 |
vadd.i16 d0, d0, d2 |
| 727 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 728 |
vst1.32 {d0[1]}, [r12, :32], r1 |
| 729 |
bgt 2b |
| 730 |
pop {r4-r5,pc} |
| 731 |
40: |
| 732 |
add r12, r0, r1 |
| 733 |
lsl r1, r1, #1 |
| 734 |
4: |
| 735 |
vld2.8 {d4[], d5[]}, [r5, :16]! |
| 736 |
vld1.16 {q1}, [r2, :128]! |
| 737 |
vext.8 d4, d4, d5, #4 |
| 738 |
subs r4, r4, #2 |
| 739 |
vneg.s8 d4, d4 // -m |
| 740 |
vld1.16 {d0}, [r0, :64] |
| 741 |
vld1.16 {d1}, [r12, :64] |
| 742 |
vmovl.s8 q2, d4 |
| 743 |
vshl.i16 q2, q2, #9 // -m << 9 |
| 744 |
vsub.i16 q1, q0, q1 // a - b |
| 745 |
vqrdmulh.s16 q1, q1, q2 // ((a-b)*-m + 32) >> 6 |
| 746 |
vadd.i16 q0, q0, q1 |
| 747 |
vst1.16 {d0}, [r0, :64], r1 |
| 748 |
vst1.16 {d1}, [r12, :64], r1 |
| 749 |
bgt 4b |
| 750 |
pop {r4-r5,pc} |
| 751 |
80: |
| 752 |
add r12, r0, r1 |
| 753 |
lsl r1, r1, #1 |
| 754 |
8: |
| 755 |
vld2.8 {d16[], d17[]}, [r5, :16]! |
| 756 |
vld1.16 {q2, q3}, [r2, :128]! |
| 757 |
vneg.s8 q9, q8 // -m |
| 758 |
vld1.16 {q0}, [r0, :128] |
| 759 |
subs r4, r4, #2 |
| 760 |
vmovl.s8 q8, d18 |
| 761 |
vmovl.s8 q9, d19 |
| 762 |
vld1.16 {q1}, [r12, :128] |
| 763 |
vshl.i16 q8, q8, #9 // -m << 9 |
| 764 |
vshl.i16 q9, q9, #9 |
| 765 |
vsub.i16 q2, q0, q2 // a - b |
| 766 |
vsub.i16 q3, q1, q3 |
| 767 |
vqrdmulh.s16 q2, q2, q8 // ((a-b)*-m + 32) >> 6 |
| 768 |
vqrdmulh.s16 q3, q3, q9 |
| 769 |
vadd.i16 q0, q0, q2 |
| 770 |
vadd.i16 q1, q1, q3 |
| 771 |
vst1.16 {q0}, [r0, :128], r1 |
| 772 |
vst1.16 {q1}, [r12, :128], r1 |
| 773 |
bgt 8b |
| 774 |
pop {r4-r5,pc} |
| 775 |
160: |
| 776 |
add r12, r0, r1 |
| 777 |
lsl r1, r1, #1 |
| 778 |
16: |
| 779 |
vld2.8 {d24[], d25[]}, [r5, :16]! |
| 780 |
vld1.16 {q8, q9}, [r2, :128]! |
| 781 |
subs r4, r4, #2 |
| 782 |
vneg.s8 q13, q12 // -m |
| 783 |
vld1.16 {q0, q1}, [r0, :128] |
| 784 |
vmovl.s8 q12, d26 |
| 785 |
vld1.16 {q10, q11}, [r2, :128]! |
| 786 |
vmovl.s8 q13, d27 |
| 787 |
vld1.16 {q2, q3}, [r12, :128] |
| 788 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 789 |
vshl.i16 q13, q13, #9 |
| 790 |
vsub.i16 q8, q0, q8 // a - b |
| 791 |
vsub.i16 q9, q1, q9 |
| 792 |
vsub.i16 q10, q2, q10 |
| 793 |
vsub.i16 q11, q3, q11 |
| 794 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 795 |
vqrdmulh.s16 q9, q9, q12 |
| 796 |
vqrdmulh.s16 q10, q10, q13 |
| 797 |
vqrdmulh.s16 q11, q11, q13 |
| 798 |
vadd.i16 q0, q0, q8 |
| 799 |
vadd.i16 q1, q1, q9 |
| 800 |
vadd.i16 q2, q2, q10 |
| 801 |
vadd.i16 q3, q3, q11 |
| 802 |
vst1.16 {q0, q1}, [r0, :128], r1 |
| 803 |
vst1.16 {q2, q3}, [r12, :128], r1 |
| 804 |
bgt 16b |
| 805 |
pop {r4-r5,pc} |
| 806 |
1280: |
| 807 |
640: |
| 808 |
320: |
| 809 |
sub r1, r1, r3, lsl #1 |
| 810 |
321: |
| 811 |
vld1.8 {d24[]}, [r5]! |
| 812 |
mov r12, r3 |
| 813 |
vneg.s8 d24, d24 // -m |
| 814 |
vmovl.s8 q12, d24 |
| 815 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 816 |
32: |
| 817 |
vld1.16 {q8, q9}, [r2, :128]! |
| 818 |
vld1.16 {q0, q1}, [r0, :128]! |
| 819 |
subs r12, r12, #32 |
| 820 |
vld1.16 {q10, q11}, [r2, :128]! |
| 821 |
vld1.16 {q2, q3}, [r0, :128] |
| 822 |
vsub.i16 q8, q0, q8 // a - b |
| 823 |
vsub.i16 q9, q1, q9 |
| 824 |
vsub.i16 q10, q2, q10 |
| 825 |
vsub.i16 q11, q3, q11 |
| 826 |
sub r0, r0, #32 |
| 827 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 828 |
vqrdmulh.s16 q9, q9, q12 |
| 829 |
vqrdmulh.s16 q10, q10, q12 |
| 830 |
vqrdmulh.s16 q11, q11, q12 |
| 831 |
vadd.i16 q0, q0, q8 |
| 832 |
vadd.i16 q1, q1, q9 |
| 833 |
vadd.i16 q2, q2, q10 |
| 834 |
vst1.16 {q0, q1}, [r0, :128]! |
| 835 |
vadd.i16 q3, q3, q11 |
| 836 |
vst1.16 {q2, q3}, [r0, :128]! |
| 837 |
bgt 32b |
| 838 |
subs r4, r4, #1 |
| 839 |
add r0, r0, r1 |
| 840 |
bgt 321b |
| 841 |
pop {r4-r5,pc} |
| 842 |
endfunc |
| 843 |
|
| 844 |
function blend_v_16bpc_neon, export=1 |
| 845 |
push {r4,lr} |
| 846 |
ldr r4, [sp, #8] |
| 847 |
movrel lr, X(obmc_masks) |
| 848 |
add lr, lr, r3 |
| 849 |
clz r12, r3 |
| 850 |
adr r3, L(blend_v_tbl) |
| 851 |
sub r12, r12, #26 |
| 852 |
ldr r12, [r3, r12, lsl #2] |
| 853 |
add r3, r3, r12 |
| 854 |
bx r3 |
| 855 |
|
| 856 |
.align 2 |
| 857 |
L(blend_v_tbl): |
| 858 |
.word 320f - L(blend_v_tbl) + CONFIG_THUMB |
| 859 |
.word 160f - L(blend_v_tbl) + CONFIG_THUMB |
| 860 |
.word 80f - L(blend_v_tbl) + CONFIG_THUMB |
| 861 |
.word 40f - L(blend_v_tbl) + CONFIG_THUMB |
| 862 |
.word 20f - L(blend_v_tbl) + CONFIG_THUMB |
| 863 |
|
| 864 |
20: |
| 865 |
add r12, r0, r1 |
| 866 |
lsl r1, r1, #1 |
| 867 |
vld1.8 {d4[]}, [lr] |
| 868 |
vneg.s8 d4, d4 // -m |
| 869 |
vmovl.s8 q2, d4 |
| 870 |
vshl.i16 d4, d4, #9 // -m << 9 |
| 871 |
2: |
| 872 |
vld1.32 {d2[]}, [r2, :32]! |
| 873 |
vld1.16 {d0[]}, [r0, :16] |
| 874 |
subs r4, r4, #2 |
| 875 |
vld1.16 {d2[1]}, [r2, :16] |
| 876 |
vld1.16 {d0[1]}, [r12, :16] |
| 877 |
add r2, r2, #4 |
| 878 |
vsub.i16 d2, d0, d2 // a - b |
| 879 |
vqrdmulh.s16 d2, d2, d4 // ((a-b)*-m + 32) >> 6 |
| 880 |
vadd.i16 d0, d0, d2 |
| 881 |
vst1.16 {d0[0]}, [r0, :16], r1 |
| 882 |
vst1.16 {d0[1]}, [r12, :16], r1 |
| 883 |
bgt 2b |
| 884 |
pop {r4,pc} |
| 885 |
40: |
| 886 |
vld1.32 {d4[]}, [lr, :32] |
| 887 |
add r12, r0, r1 |
| 888 |
vneg.s8 d4, d4 // -m |
| 889 |
lsl r1, r1, #1 |
| 890 |
vmovl.s8 q2, d4 |
| 891 |
sub r1, r1, #4 |
| 892 |
vshl.i16 q2, q2, #9 // -m << 9 |
| 893 |
4: |
| 894 |
vld1.16 {q1}, [r2, :128]! |
| 895 |
vld1.16 {d0}, [r0, :64] |
| 896 |
vld1.16 {d1}, [r12, :64] |
| 897 |
subs r4, r4, #2 |
| 898 |
vsub.i16 q1, q0, q1 // a - b |
| 899 |
vqrdmulh.s16 q1, q1, q2 // ((a-b)*-m + 32) >> 6 |
| 900 |
vadd.i16 q0, q0, q1 |
| 901 |
vst1.32 {d0[0]}, [r0, :32]! |
| 902 |
vst1.32 {d1[0]}, [r12, :32]! |
| 903 |
vst1.16 {d0[2]}, [r0, :16], r1 |
| 904 |
vst1.16 {d1[2]}, [r12, :16], r1 |
| 905 |
bgt 4b |
| 906 |
pop {r4,pc} |
| 907 |
80: |
| 908 |
vld1.8 {d16}, [lr, :64] |
| 909 |
add r12, r0, r1 |
| 910 |
vneg.s8 d16, d16 // -m |
| 911 |
lsl r1, r1, #1 |
| 912 |
vmovl.s8 q8, d16 |
| 913 |
sub r1, r1, #8 |
| 914 |
vshl.i16 q8, q8, #9 // -m << 9 |
| 915 |
8: |
| 916 |
vld1.16 {q2, q3}, [r2, :128]! |
| 917 |
vld1.16 {q0}, [r0, :128] |
| 918 |
vld1.16 {q1}, [r12, :128] |
| 919 |
subs r4, r4, #2 |
| 920 |
vsub.i16 q2, q0, q2 // a - b |
| 921 |
vsub.i16 q3, q1, q3 |
| 922 |
vqrdmulh.s16 q2, q2, q8 // ((a-b)*-m + 32) >> 6 |
| 923 |
vqrdmulh.s16 q3, q3, q8 |
| 924 |
vadd.i16 q0, q0, q2 |
| 925 |
vadd.i16 q1, q1, q3 |
| 926 |
vst1.16 {d0}, [r0, :64]! |
| 927 |
vst1.16 {d2}, [r12, :64]! |
| 928 |
vst1.32 {d1[0]}, [r0, :32], r1 |
| 929 |
vst1.32 {d3[0]}, [r12, :32], r1 |
| 930 |
bgt 8b |
| 931 |
pop {r4,pc} |
| 932 |
160: |
| 933 |
vld1.8 {q12}, [lr, :128] |
| 934 |
add r12, r0, r1 |
| 935 |
vneg.s8 q13, q12 // -m |
| 936 |
lsl r1, r1, #1 |
| 937 |
vmovl.s8 q12, d26 |
| 938 |
vmovl.s8 q13, d27 |
| 939 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 940 |
vshl.i16 d26, d26, #9 |
| 941 |
16: |
| 942 |
vld1.16 {q8, q9}, [r2, :128]! |
| 943 |
vld1.16 {d0, d1, d2}, [r0, :64] |
| 944 |
subs r4, r4, #2 |
| 945 |
vld1.16 {q10, q11}, [r2, :128]! |
| 946 |
vsub.i16 q8, q0, q8 // a - b |
| 947 |
vld1.16 {d4, d5, d6}, [r12, :64] |
| 948 |
vsub.i16 d18, d2, d18 |
| 949 |
vsub.i16 q10, q2, q10 |
| 950 |
vsub.i16 d22, d6, d22 |
| 951 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 952 |
vqrdmulh.s16 d18, d18, d26 |
| 953 |
vqrdmulh.s16 q10, q10, q12 |
| 954 |
vqrdmulh.s16 d22, d22, d26 |
| 955 |
vadd.i16 q0, q0, q8 |
| 956 |
vadd.i16 d2, d2, d18 |
| 957 |
vadd.i16 q2, q2, q10 |
| 958 |
vst1.16 {d0, d1, d2}, [r0, :64], r1 |
| 959 |
vadd.i16 d6, d6, d22 |
| 960 |
vst1.16 {d4, d5, d6}, [r12, :64], r1 |
| 961 |
bgt 16b |
| 962 |
pop {r4,pc} |
| 963 |
320: |
| 964 |
vld1.8 {d24, d25, d26}, [lr, :64] |
| 965 |
vneg.s8 q14, q12 // -m |
| 966 |
vneg.s8 d30, d26 |
| 967 |
vmovl.s8 q12, d28 |
| 968 |
vmovl.s8 q13, d29 |
| 969 |
vmovl.s8 q14, d30 |
| 970 |
sub r1, r1, #32 |
| 971 |
vshl.i16 q12, q12, #9 // -m << 9 |
| 972 |
vshl.i16 q13, q13, #9 |
| 973 |
vshl.i16 q14, q14, #9 |
| 974 |
32: |
| 975 |
vld1.16 {q8, q9}, [r2, :128]! |
| 976 |
vld1.16 {q0, q1}, [r0, :128]! |
| 977 |
subs r4, r4, #1 |
| 978 |
vld1.16 {q10}, [r2, :128] |
| 979 |
vsub.i16 q8, q0, q8 // a - b |
| 980 |
vld1.16 {q2}, [r0, :128] |
| 981 |
sub r0, r0, #32 |
| 982 |
vsub.i16 q9, q1, q9 |
| 983 |
vsub.i16 q10, q2, q10 |
| 984 |
vqrdmulh.s16 q8, q8, q12 // ((a-b)*-m + 32) >> 6 |
| 985 |
vqrdmulh.s16 q9, q9, q13 |
| 986 |
vqrdmulh.s16 q10, q10, q14 |
| 987 |
vadd.i16 q0, q0, q8 |
| 988 |
vadd.i16 q1, q1, q9 |
| 989 |
vadd.i16 q2, q2, q10 |
| 990 |
vst1.16 {q0, q1}, [r0, :128]! |
| 991 |
add r2, r2, #32 |
| 992 |
vst1.16 {q2}, [r0, :128], r1 |
| 993 |
bgt 32b |
| 994 |
pop {r4,pc} |
| 995 |
endfunc |
| 996 |
|
| 997 |
// This has got the same signature as the put_8tap functions, |
| 998 |
// and assumes that r9 is set to (clz(w)-24). |
| 999 |
function put_neon |
| 1000 |
adr r10, L(put_tbl) |
| 1001 |
ldr r9, [r10, r9, lsl #2] |
| 1002 |
add r10, r10, r9 |
| 1003 |
bx r10 |
| 1004 |
|
| 1005 |
.align 2 |
| 1006 |
L(put_tbl): |
| 1007 |
.word 1280f - L(put_tbl) + CONFIG_THUMB |
| 1008 |
.word 640f - L(put_tbl) + CONFIG_THUMB |
| 1009 |
.word 320f - L(put_tbl) + CONFIG_THUMB |
| 1010 |
.word 16f - L(put_tbl) + CONFIG_THUMB |
| 1011 |
.word 80f - L(put_tbl) + CONFIG_THUMB |
| 1012 |
.word 4f - L(put_tbl) + CONFIG_THUMB |
| 1013 |
.word 2f - L(put_tbl) + CONFIG_THUMB |
| 1014 |
|
| 1015 |
2: |
| 1016 |
vld1.32 {d0[]}, [r2], r3 |
| 1017 |
vld1.32 {d1[]}, [r2], r3 |
| 1018 |
subs r5, r5, #2 |
| 1019 |
vst1.32 {d0[0]}, [r0, :32], r1 |
| 1020 |
vst1.32 {d1[1]}, [r0, :32], r1 |
| 1021 |
bgt 2b |
| 1022 |
pop {r4-r11,pc} |
| 1023 |
4: |
| 1024 |
vld1.16 {d0}, [r2], r3 |
| 1025 |
vld1.16 {d1}, [r2], r3 |
| 1026 |
subs r5, r5, #2 |
| 1027 |
vst1.16 {d0}, [r0, :64], r1 |
| 1028 |
vst1.16 {d1}, [r0, :64], r1 |
| 1029 |
bgt 4b |
| 1030 |
pop {r4-r11,pc} |
| 1031 |
80: |
| 1032 |
add r8, r0, r1 |
| 1033 |
lsl r1, r1, #1 |
| 1034 |
add r9, r2, r3 |
| 1035 |
lsl r3, r3, #1 |
| 1036 |
8: |
| 1037 |
vld1.16 {q0}, [r2], r3 |
| 1038 |
vld1.16 {q1}, [r9], r3 |
| 1039 |
subs r5, r5, #2 |
| 1040 |
vst1.16 {q0}, [r0, :128], r1 |
| 1041 |
vst1.16 {q1}, [r8, :128], r1 |
| 1042 |
bgt 8b |
| 1043 |
pop {r4-r11,pc} |
| 1044 |
16: |
| 1045 |
vld1.16 {q0, q1}, [r2], r3 |
| 1046 |
subs r5, r5, #1 |
| 1047 |
vst1.16 {q0, q1}, [r0, :128], r1 |
| 1048 |
bgt 16b |
| 1049 |
pop {r4-r11,pc} |
| 1050 |
320: |
| 1051 |
sub r1, r1, #32 |
| 1052 |
sub r3, r3, #32 |
| 1053 |
32: |
| 1054 |
vld1.16 {q0, q1}, [r2]! |
| 1055 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1056 |
vld1.16 {q2, q3}, [r2], r3 |
| 1057 |
subs r5, r5, #1 |
| 1058 |
vst1.16 {q2, q3}, [r0, :128], r1 |
| 1059 |
bgt 32b |
| 1060 |
pop {r4-r11,pc} |
| 1061 |
640: |
| 1062 |
sub r1, r1, #96 |
| 1063 |
sub r3, r3, #96 |
| 1064 |
64: |
| 1065 |
vld1.16 {q8, q9}, [r2]! |
| 1066 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1067 |
vld1.16 {q10, q11}, [r2]! |
| 1068 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1069 |
vld1.16 {q12, q13}, [r2]! |
| 1070 |
vst1.16 {q12, q13}, [r0, :128]! |
| 1071 |
vld1.16 {q14, q15}, [r2], r3 |
| 1072 |
subs r5, r5, #1 |
| 1073 |
vst1.16 {q14, q15}, [r0, :128], r1 |
| 1074 |
bgt 64b |
| 1075 |
pop {r4-r11,pc} |
| 1076 |
1280: |
| 1077 |
sub r1, r1, #224 |
| 1078 |
sub r3, r3, #224 |
| 1079 |
128: |
| 1080 |
vld1.16 {q8, q9}, [r2]! |
| 1081 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1082 |
vld1.16 {q10, q11}, [r2]! |
| 1083 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1084 |
vld1.16 {q12, q13}, [r2]! |
| 1085 |
vst1.16 {q12, q13}, [r0, :128]! |
| 1086 |
vld1.16 {q14, q15}, [r2]! |
| 1087 |
vst1.16 {q14, q15}, [r0, :128]! |
| 1088 |
vld1.16 {q8, q9}, [r2]! |
| 1089 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1090 |
vld1.16 {q10, q11}, [r2]! |
| 1091 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1092 |
vld1.16 {q12, q13}, [r2]! |
| 1093 |
vst1.16 {q12, q13}, [r0, :128]! |
| 1094 |
vld1.16 {q14, q15}, [r2], r3 |
| 1095 |
subs r5, r5, #1 |
| 1096 |
vst1.16 {q14, q15}, [r0, :128], r1 |
| 1097 |
bgt 128b |
| 1098 |
pop {r4-r11,pc} |
| 1099 |
endfunc |
| 1100 |
|
| 1101 |
// This has got the same signature as the prep_8tap functions, |
| 1102 |
// and assumes that r9 is set to (clz(w)-24), r7 to intermediate_bits and |
| 1103 |
// r8 to w*2. |
| 1104 |
function prep_neon |
| 1105 |
adr r10, L(prep_tbl) |
| 1106 |
ldr r9, [r10, r9, lsl #2] |
| 1107 |
vdup.16 q15, r7 // intermediate_bits |
| 1108 |
vmov.i16 q14, #PREP_BIAS |
| 1109 |
add r10, r10, r9 |
| 1110 |
bx r10 |
| 1111 |
|
| 1112 |
.align 2 |
| 1113 |
L(prep_tbl): |
| 1114 |
.word 1280f - L(prep_tbl) + CONFIG_THUMB |
| 1115 |
.word 640f - L(prep_tbl) + CONFIG_THUMB |
| 1116 |
.word 320f - L(prep_tbl) + CONFIG_THUMB |
| 1117 |
.word 16f - L(prep_tbl) + CONFIG_THUMB |
| 1118 |
.word 80f - L(prep_tbl) + CONFIG_THUMB |
| 1119 |
.word 40f - L(prep_tbl) + CONFIG_THUMB |
| 1120 |
|
| 1121 |
40: |
| 1122 |
add r9, r1, r2 |
| 1123 |
lsl r2, r2, #1 |
| 1124 |
4: |
| 1125 |
vld1.16 {d0}, [r1], r2 |
| 1126 |
vld1.16 {d1}, [r9], r2 |
| 1127 |
subs r4, r4, #2 |
| 1128 |
vshl.s16 q0, q0, q15 |
| 1129 |
vsub.i16 q0, q0, q14 |
| 1130 |
vst1.16 {q0}, [r0, :128]! |
| 1131 |
bgt 4b |
| 1132 |
pop {r4-r11,pc} |
| 1133 |
80: |
| 1134 |
add r9, r1, r2 |
| 1135 |
lsl r2, r2, #1 |
| 1136 |
8: |
| 1137 |
vld1.16 {q0}, [r1], r2 |
| 1138 |
vld1.16 {q1}, [r9], r2 |
| 1139 |
subs r4, r4, #2 |
| 1140 |
vshl.s16 q0, q0, q15 |
| 1141 |
vshl.s16 q1, q1, q15 |
| 1142 |
vsub.i16 q0, q0, q14 |
| 1143 |
vsub.i16 q1, q1, q14 |
| 1144 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1145 |
bgt 8b |
| 1146 |
pop {r4-r11,pc} |
| 1147 |
16: |
| 1148 |
vld1.16 {q0, q1}, [r1], r2 |
| 1149 |
vshl.s16 q0, q0, q15 |
| 1150 |
vld1.16 {q2, q3}, [r1], r2 |
| 1151 |
subs r4, r4, #2 |
| 1152 |
vshl.s16 q1, q1, q15 |
| 1153 |
vshl.s16 q2, q2, q15 |
| 1154 |
vshl.s16 q3, q3, q15 |
| 1155 |
vsub.i16 q0, q0, q14 |
| 1156 |
vsub.i16 q1, q1, q14 |
| 1157 |
vsub.i16 q2, q2, q14 |
| 1158 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1159 |
vsub.i16 q3, q3, q14 |
| 1160 |
vst1.16 {q2, q3}, [r0, :128]! |
| 1161 |
bgt 16b |
| 1162 |
pop {r4-r11,pc} |
| 1163 |
320: |
| 1164 |
sub r2, r2, #32 |
| 1165 |
32: |
| 1166 |
vld1.16 {q0, q1}, [r1]! |
| 1167 |
subs r4, r4, #1 |
| 1168 |
vshl.s16 q0, q0, q15 |
| 1169 |
vld1.16 {q2, q3}, [r1], r2 |
| 1170 |
vshl.s16 q1, q1, q15 |
| 1171 |
vshl.s16 q2, q2, q15 |
| 1172 |
vshl.s16 q3, q3, q15 |
| 1173 |
vsub.i16 q0, q0, q14 |
| 1174 |
vsub.i16 q1, q1, q14 |
| 1175 |
vsub.i16 q2, q2, q14 |
| 1176 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1177 |
vsub.i16 q3, q3, q14 |
| 1178 |
vst1.16 {q2, q3}, [r0, :128]! |
| 1179 |
bgt 32b |
| 1180 |
pop {r4-r11,pc} |
| 1181 |
640: |
| 1182 |
sub r2, r2, #96 |
| 1183 |
64: |
| 1184 |
vld1.16 {q0, q1}, [r1]! |
| 1185 |
subs r4, r4, #1 |
| 1186 |
vshl.s16 q0, q0, q15 |
| 1187 |
vld1.16 {q2, q3}, [r1]! |
| 1188 |
vshl.s16 q1, q1, q15 |
| 1189 |
vld1.16 {q8, q9}, [r1]! |
| 1190 |
vshl.s16 q2, q2, q15 |
| 1191 |
vld1.16 {q10, q11}, [r1], r2 |
| 1192 |
vshl.s16 q3, q3, q15 |
| 1193 |
vshl.s16 q8, q8, q15 |
| 1194 |
vshl.s16 q9, q9, q15 |
| 1195 |
vshl.s16 q10, q10, q15 |
| 1196 |
vshl.s16 q11, q11, q15 |
| 1197 |
vsub.i16 q0, q0, q14 |
| 1198 |
vsub.i16 q1, q1, q14 |
| 1199 |
vsub.i16 q2, q2, q14 |
| 1200 |
vsub.i16 q3, q3, q14 |
| 1201 |
vsub.i16 q8, q8, q14 |
| 1202 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1203 |
vsub.i16 q9, q9, q14 |
| 1204 |
vst1.16 {q2, q3}, [r0, :128]! |
| 1205 |
vsub.i16 q10, q10, q14 |
| 1206 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1207 |
vsub.i16 q11, q11, q14 |
| 1208 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1209 |
bgt 64b |
| 1210 |
pop {r4-r11,pc} |
| 1211 |
1280: |
| 1212 |
sub r2, r2, #224 |
| 1213 |
128: |
| 1214 |
vld1.16 {q0, q1}, [r1]! |
| 1215 |
subs r4, r4, #1 |
| 1216 |
vshl.s16 q0, q0, q15 |
| 1217 |
vld1.16 {q2, q3}, [r1]! |
| 1218 |
vshl.s16 q1, q1, q15 |
| 1219 |
vld1.16 {q8, q9}, [r1]! |
| 1220 |
vshl.s16 q2, q2, q15 |
| 1221 |
vld1.16 {q10, q11}, [r1]! |
| 1222 |
vshl.s16 q3, q3, q15 |
| 1223 |
vshl.s16 q8, q8, q15 |
| 1224 |
vshl.s16 q9, q9, q15 |
| 1225 |
vshl.s16 q10, q10, q15 |
| 1226 |
vshl.s16 q11, q11, q15 |
| 1227 |
vsub.i16 q0, q0, q14 |
| 1228 |
vsub.i16 q1, q1, q14 |
| 1229 |
vsub.i16 q2, q2, q14 |
| 1230 |
vsub.i16 q3, q3, q14 |
| 1231 |
vsub.i16 q8, q8, q14 |
| 1232 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1233 |
vld1.16 {q0, q1}, [r1]! |
| 1234 |
vsub.i16 q9, q9, q14 |
| 1235 |
vsub.i16 q10, q10, q14 |
| 1236 |
vst1.16 {q2, q3}, [r0, :128]! |
| 1237 |
vld1.16 {q2, q3}, [r1]! |
| 1238 |
vsub.i16 q11, q11, q14 |
| 1239 |
vshl.s16 q0, q0, q15 |
| 1240 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1241 |
vld1.16 {q8, q9}, [r1]! |
| 1242 |
vshl.s16 q1, q1, q15 |
| 1243 |
vshl.s16 q2, q2, q15 |
| 1244 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1245 |
vld1.16 {q10, q11}, [r1], r2 |
| 1246 |
vshl.s16 q3, q3, q15 |
| 1247 |
vshl.s16 q8, q8, q15 |
| 1248 |
vshl.s16 q9, q9, q15 |
| 1249 |
vshl.s16 q10, q10, q15 |
| 1250 |
vshl.s16 q11, q11, q15 |
| 1251 |
vsub.i16 q0, q0, q14 |
| 1252 |
vsub.i16 q1, q1, q14 |
| 1253 |
vsub.i16 q2, q2, q14 |
| 1254 |
vsub.i16 q3, q3, q14 |
| 1255 |
vsub.i16 q8, q8, q14 |
| 1256 |
vst1.16 {q0, q1}, [r0, :128]! |
| 1257 |
vsub.i16 q9, q9, q14 |
| 1258 |
vst1.16 {q2, q3}, [r0, :128]! |
| 1259 |
vsub.i16 q10, q10, q14 |
| 1260 |
vst1.16 {q8, q9}, [r0, :128]! |
| 1261 |
vsub.i16 q11, q11, q14 |
| 1262 |
vst1.16 {q10, q11}, [r0, :128]! |
| 1263 |
bgt 128b |
| 1264 |
pop {r4-r11,pc} |
| 1265 |
endfunc |
| 1266 |
|
| 1267 |
.macro load_slice s0, s1, strd, wd, d0, d1, d2, d3, d4, d5, d6 |
| 1268 |
vld1.\wd {\d0[]}, [\s0], \strd |
| 1269 |
vld1.\wd {\d1[]}, [\s1], \strd |
| 1270 |
.ifnb \d2 |
| 1271 |
vld1.\wd {\d2[]}, [\s0], \strd |
| 1272 |
vld1.\wd {\d3[]}, [\s1], \strd |
| 1273 |
.endif |
| 1274 |
.ifnb \d4 |
| 1275 |
vld1.\wd {\d4[]}, [\s0], \strd |
| 1276 |
.endif |
| 1277 |
.ifnb \d5 |
| 1278 |
vld1.\wd {\d5[]}, [\s1], \strd |
| 1279 |
.endif |
| 1280 |
.ifnb \d6 |
| 1281 |
vld1.\wd {\d6[]}, [\s0], \strd |
| 1282 |
.endif |
| 1283 |
.endm |
| 1284 |
.macro load_reg s0, s1, strd, d0, d1, d2, d3, d4, d5, d6 |
| 1285 |
vld1.16 {\d0}, [\s0], \strd |
| 1286 |
vld1.16 {\d1}, [\s1], \strd |
| 1287 |
.ifnb \d2 |
| 1288 |
vld1.16 {\d2}, [\s0], \strd |
| 1289 |
vld1.16 {\d3}, [\s1], \strd |
| 1290 |
.endif |
| 1291 |
.ifnb \d4 |
| 1292 |
vld1.16 {\d4}, [\s0], \strd |
| 1293 |
.endif |
| 1294 |
.ifnb \d5 |
| 1295 |
vld1.16 {\d5}, [\s1], \strd |
| 1296 |
.endif |
| 1297 |
.ifnb \d6 |
| 1298 |
vld1.16 {\d6}, [\s0], \strd |
| 1299 |
.endif |
| 1300 |
.endm |
| 1301 |
.macro load_regpair s0, s1, strd, d0, d1, d2, d3, d4, d5 |
| 1302 |
vld1.16 {\d0, \d1}, [\s0], \strd |
| 1303 |
.ifnb \d2 |
| 1304 |
vld1.16 {\d2, \d3}, [\s1], \strd |
| 1305 |
.endif |
| 1306 |
.ifnb \d4 |
| 1307 |
vld1.16 {\d4, \d5}, [\s0], \strd |
| 1308 |
.endif |
| 1309 |
.endm |
| 1310 |
.macro load_32 s0, s1, strd, d0, d1, d2, d3, d4, d5, d6 |
| 1311 |
load_slice \s0, \s1, \strd, 32, \d0, \d1, \d2, \d3, \d4, \d5, \d6 |
| 1312 |
.endm |
| 1313 |
.macro load_16s16 s0, s1, strd, d0, d1, d2, d3, d4, d5 |
| 1314 |
load_regpair \s0, \s1, \strd, \d0, \d1, \d2, \d3, \d4, \d5 |
| 1315 |
.endm |
| 1316 |
.macro interleave_1_32 r0, r1, r2, r3, r4 |
| 1317 |
vext.8 \r0, \r0, \r1, #4 |
| 1318 |
vext.8 \r1, \r1, \r2, #4 |
| 1319 |
.ifnb \r3 |
| 1320 |
vext.8 \r2, \r2, \r3, #4 |
| 1321 |
vext.8 \r3, \r3, \r4, #4 |
| 1322 |
.endif |
| 1323 |
.endm |
| 1324 |
.macro vmin_u16 c, r0, r1, r2, r3 |
| 1325 |
vmin.u16 \r0, \r0, \c |
| 1326 |
.ifnb \r1 |
| 1327 |
vmin.u16 \r1, \r1, \c |
| 1328 |
.endif |
| 1329 |
.ifnb \r2 |
| 1330 |
vmin.u16 \r2, \r2, \c |
| 1331 |
vmin.u16 \r3, \r3, \c |
| 1332 |
.endif |
| 1333 |
.endm |
| 1334 |
.macro vsub_i16 c, r0, r1, r2, r3 |
| 1335 |
vsub.i16 \r0, \r0, \c |
| 1336 |
.ifnb \r1 |
| 1337 |
vsub.i16 \r1, \r1, \c |
| 1338 |
.endif |
| 1339 |
.ifnb \r2 |
| 1340 |
vsub.i16 \r2, \r2, \c |
| 1341 |
vsub.i16 \r3, \r3, \c |
| 1342 |
.endif |
| 1343 |
.endm |
| 1344 |
.macro vmull_vmlal_4 d, s0, s1, s2, s3 |
| 1345 |
vmull.s16 \d, \s0, d0[0] |
| 1346 |
vmlal.s16 \d, \s1, d0[1] |
| 1347 |
vmlal.s16 \d, \s2, d0[2] |
| 1348 |
vmlal.s16 \d, \s3, d0[3] |
| 1349 |
.endm |
| 1350 |
.macro vmull_vmlal_8 d, s0, s1, s2, s3, s4, s5, s6, s7 |
| 1351 |
vmull.s16 \d, \s0, d0[0] |
| 1352 |
vmlal.s16 \d, \s1, d0[1] |
| 1353 |
vmlal.s16 \d, \s2, d0[2] |
| 1354 |
vmlal.s16 \d, \s3, d0[3] |
| 1355 |
vmlal.s16 \d, \s4, d1[0] |
| 1356 |
vmlal.s16 \d, \s5, d1[1] |
| 1357 |
vmlal.s16 \d, \s6, d1[2] |
| 1358 |
vmlal.s16 \d, \s7, d1[3] |
| 1359 |
.endm |
| 1360 |
.macro vqrshrun_s32 shift, q0, d0, q1, d1, q2, d2, q3, d3 |
| 1361 |
vqrshrun.s32 \d0, \q0, #\shift |
| 1362 |
.ifnb \q1 |
| 1363 |
vqrshrun.s32 \d1, \q1, #\shift |
| 1364 |
.endif |
| 1365 |
.ifnb \q2 |
| 1366 |
vqrshrun.s32 \d2, \q2, #\shift |
| 1367 |
vqrshrun.s32 \d3, \q3, #\shift |
| 1368 |
.endif |
| 1369 |
.endm |
| 1370 |
.macro vmovn_i32 q0, d0, q1, d1, q2, d2, q3, d3 |
| 1371 |
vmovn.i32 \d0, \q0 |
| 1372 |
.ifnb \q1 |
| 1373 |
vmovn.i32 \d1, \q1 |
| 1374 |
.endif |
| 1375 |
.ifnb \q2 |
| 1376 |
vmovn.i32 \d2, \q2 |
| 1377 |
vmovn.i32 \d3, \q3 |
| 1378 |
.endif |
| 1379 |
.endm |
| 1380 |
.macro vrshl_s32 shift, r0, r1, r2, r3 |
| 1381 |
vrshl.s32 \r0, \r0, \shift |
| 1382 |
vrshl.s32 \r1, \r1, \shift |
| 1383 |
.ifnb \r2 |
| 1384 |
vrshl.s32 \r2, \r2, \shift |
| 1385 |
vrshl.s32 \r3, \r3, \shift |
| 1386 |
.endif |
| 1387 |
.endm |
| 1388 |
.macro vst1_32 strd, r0, r1 |
| 1389 |
vst1.32 {\r0[0]}, [r0, :32], \strd |
| 1390 |
vst1.32 {\r0[1]}, [r9, :32], \strd |
| 1391 |
.ifnb \r1 |
| 1392 |
vst1.32 {\r1[0]}, [r0, :32], \strd |
| 1393 |
vst1.32 {\r1[1]}, [r9, :32], \strd |
| 1394 |
.endif |
| 1395 |
.endm |
| 1396 |
.macro vst1_reg strd, align, r0, r1, r2, r3, r4, r5, r6, r7 |
| 1397 |
vst1.16 {\r0}, [r0, \align], \strd |
| 1398 |
vst1.16 {\r1}, [r9, \align], \strd |
| 1399 |
.ifnb \r2 |
| 1400 |
vst1.16 {\r2}, [r0, \align], \strd |
| 1401 |
vst1.16 {\r3}, [r9, \align], \strd |
| 1402 |
.endif |
| 1403 |
.ifnb \r4 |
| 1404 |
vst1.16 {\r4}, [r0, \align], \strd |
| 1405 |
vst1.16 {\r5}, [r9, \align], \strd |
| 1406 |
vst1.16 {\r6}, [r0, \align], \strd |
| 1407 |
vst1.16 {\r7}, [r9, \align], \strd |
| 1408 |
.endif |
| 1409 |
.endm |
| 1410 |
.macro finalize type, q0, q1, d0, d1, q2, q3, d2, d3 |
| 1411 |
.ifc \type, put |
| 1412 |
vqrshrun_s32 6, \q0, \d0, \q1, \d1, \q2, \d2, \q3, \d3 |
| 1413 |
vmin_u16 q15, \q0, \q1 |
| 1414 |
.else |
| 1415 |
vrshl_s32 q14, \q0, \q1, \q2, \q3 // -(6-intermediate_bits) |
| 1416 |
vmovn_i32 \q0, \d0, \q1, \d1, \q2, \d2, \q3, \d3 |
| 1417 |
vsub_i16 q15, \q0, \q1 // PREP_BIAS |
| 1418 |
.endif |
| 1419 |
.endm |
| 1420 |
.macro shift_store_4 type, strd, q0, q1, d0, d1, q2, q3, d2, d3 |
| 1421 |
finalize \type, \q0, \q1, \d0, \d1, \q2, \q3, \d2, \d3 |
| 1422 |
vst1_reg \strd, :64, \d0, \d1, \d2, \d3 |
| 1423 |
.endm |
| 1424 |
.macro shift_store_8 type, strd, q0, q1, d0, d1, q2, q3, d2, d3 |
| 1425 |
finalize \type, \q0, \q1, \d0, \d1, \q2, \q3, \d2, \d3 |
| 1426 |
vst1_reg \strd, :128, \q0, \q1 |
| 1427 |
.endm |
| 1428 |
.macro shift_store_16 type, strd, q0, q1, d0, d1, q2, q3, d2, d3 |
| 1429 |
finalize \type, \q0, \q1, \d0, \d1, \q2, \q3, \d2, \d3 |
| 1430 |
vst1.16 {\q0, \q1}, [r0, :128], \strd |
| 1431 |
.endm |
| 1432 |
|
| 1433 |
.macro make_8tap_fn op, type, type_h, type_v |
| 1434 |
function \op\()_8tap_\type\()_16bpc_neon, export=1 |
| 1435 |
push {r4-r11,lr} |
| 1436 |
movw r9, \type_h |
| 1437 |
movw r10, \type_v |
| 1438 |
b \op\()_8tap_neon |
| 1439 |
endfunc |
| 1440 |
.endm |
| 1441 |
|
| 1442 |
// No spaces in these expressions, due to gas-preprocessor. |
| 1443 |
#define REGULAR ((0*15<<7)|3*15) |
| 1444 |
#define SMOOTH ((1*15<<7)|4*15) |
| 1445 |
#define SHARP ((2*15<<7)|3*15) |
| 1446 |
|
| 1447 |
.macro filter_fn type, dst, d_strd, src, s_strd, w, h, mx, my, bdmax, ds2, sr2 |
| 1448 |
make_8tap_fn \type, regular, REGULAR, REGULAR |
| 1449 |
make_8tap_fn \type, regular_smooth, REGULAR, SMOOTH |
| 1450 |
make_8tap_fn \type, regular_sharp, REGULAR, SHARP |
| 1451 |
make_8tap_fn \type, smooth, SMOOTH, SMOOTH |
| 1452 |
make_8tap_fn \type, smooth_regular, SMOOTH, REGULAR |
| 1453 |
make_8tap_fn \type, smooth_sharp, SMOOTH, SHARP |
| 1454 |
make_8tap_fn \type, sharp, SHARP, SHARP |
| 1455 |
make_8tap_fn \type, sharp_regular, SHARP, REGULAR |
| 1456 |
make_8tap_fn \type, sharp_smooth, SHARP, SMOOTH |
| 1457 |
|
| 1458 |
function \type\()_8tap_neon |
| 1459 |
ldrd r4, r5, [sp, #36] |
| 1460 |
ldrd r6, r7, [sp, #44] |
| 1461 |
.ifc \bdmax, r8 |
| 1462 |
ldr r8, [sp, #52] |
| 1463 |
.endif |
| 1464 |
movw r11, #0x4081 // (1 << 14) | (1 << 7) | (1 << 0) |
| 1465 |
mul \mx, \mx, r11 |
| 1466 |
mul \my, \my, r11 |
| 1467 |
add \mx, \mx, r9 // mx, 8tap_h, 4tap_h |
| 1468 |
add \my, \my, r10 // my, 8tap_v, 4tap_v |
| 1469 |
|
| 1470 |
.ifc \type, prep |
| 1471 |
lsl \d_strd, \w, #1 |
| 1472 |
.endif |
| 1473 |
|
| 1474 |
vdup.16 q15, \bdmax // bitdepth_max |
| 1475 |
clz \bdmax, \bdmax |
| 1476 |
clz r9, \w |
| 1477 |
sub \bdmax, \bdmax, #18 // intermediate_bits = clz(bitdepth_max) - 18 |
| 1478 |
tst \mx, #(0x7f << 14) |
| 1479 |
sub r9, r9, #24 |
| 1480 |
add lr, \bdmax, #6 // 6 + intermediate_bits |
| 1481 |
rsb r12, \bdmax, #6 // 6 - intermediate_bits |
| 1482 |
movrel r11, X(mc_subpel_filters), -8 |
| 1483 |
bne L(\type\()_8tap_h) |
| 1484 |
tst \my, #(0x7f << 14) |
| 1485 |
bne L(\type\()_8tap_v) |
| 1486 |
b \type\()_neon |
| 1487 |
|
| 1488 |
L(\type\()_8tap_h): |
| 1489 |
cmp \w, #4 |
| 1490 |
ubfx r10, \mx, #7, #7 |
| 1491 |
and \mx, \mx, #0x7f |
| 1492 |
it gt |
| 1493 |
movgt \mx, r10 |
| 1494 |
tst \my, #(0x7f << 14) |
| 1495 |
add \mx, r11, \mx, lsl #3 |
| 1496 |
bne L(\type\()_8tap_hv) |
| 1497 |
|
| 1498 |
adr r10, L(\type\()_8tap_h_tbl) |
| 1499 |
vdup.32 q14, r12 // 6 - intermediate_bits |
| 1500 |
ldr r9, [r10, r9, lsl #2] |
| 1501 |
vneg.s32 q14, q14 // -(6-intermediate_bits) |
| 1502 |
.ifc \type, put |
| 1503 |
vdup.16 q13, \bdmax // intermediate_bits |
| 1504 |
.else |
| 1505 |
vmov.i16 q13, #PREP_BIAS |
| 1506 |
.endif |
| 1507 |
add r10, r10, r9 |
| 1508 |
.ifc \type, put |
| 1509 |
vneg.s16 q13, q13 // -intermediate_bits |
| 1510 |
.endif |
| 1511 |
bx r10 |
| 1512 |
|
| 1513 |
.align 2 |
| 1514 |
L(\type\()_8tap_h_tbl): |
| 1515 |
.word 1280f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1516 |
.word 640f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1517 |
.word 320f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1518 |
.word 160f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1519 |
.word 80f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1520 |
.word 40f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1521 |
.word 20f - L(\type\()_8tap_h_tbl) + CONFIG_THUMB |
| 1522 |
|
| 1523 |
20: // 2xN h |
| 1524 |
.ifc \type, put |
| 1525 |
add \mx, \mx, #2 |
| 1526 |
vld1.32 {d0[]}, [\mx] |
| 1527 |
sub \src, \src, #2 |
| 1528 |
add \ds2, \dst, \d_strd |
| 1529 |
add \sr2, \src, \s_strd |
| 1530 |
lsl \d_strd, \d_strd, #1 |
| 1531 |
lsl \s_strd, \s_strd, #1 |
| 1532 |
vmovl.s8 q0, d0 |
| 1533 |
2: |
| 1534 |
vld1.16 {q2}, [\src], \s_strd |
| 1535 |
vld1.16 {q3}, [\sr2], \s_strd |
| 1536 |
vext.8 d5, d4, d5, #2 |
| 1537 |
vext.8 d7, d6, d7, #2 |
| 1538 |
subs \h, \h, #2 |
| 1539 |
vtrn.32 d4, d6 |
| 1540 |
vtrn.32 d5, d7 |
| 1541 |
vmull.s16 q1, d4, d0[0] |
| 1542 |
vmlal.s16 q1, d5, d0[1] |
| 1543 |
vmlal.s16 q1, d6, d0[2] |
| 1544 |
vmlal.s16 q1, d7, d0[3] |
| 1545 |
vrshl.s32 q1, q1, q14 // -(6-intermediate_bits) |
| 1546 |
vqmovun.s32 d2, q1 |
| 1547 |
vrshl.s16 d2, d2, d26 // -intermediate_bits |
| 1548 |
vmin.u16 d2, d2, d30 |
| 1549 |
vst1.32 {d2[0]}, [\dst, :32], \d_strd |
| 1550 |
vst1.32 {d2[1]}, [\ds2, :32], \d_strd |
| 1551 |
bgt 2b |
| 1552 |
pop {r4-r11,pc} |
| 1553 |
.endif |
| 1554 |
|
| 1555 |
40: // 4xN h |
| 1556 |
add \mx, \mx, #2 |
| 1557 |
vld1.32 {d0[]}, [\mx] |
| 1558 |
sub \src, \src, #2 |
| 1559 |
add \ds2, \dst, \d_strd |
| 1560 |
add \sr2, \src, \s_strd |
| 1561 |
lsl \d_strd, \d_strd, #1 |
| 1562 |
lsl \s_strd, \s_strd, #1 |
| 1563 |
vmovl.s8 q0, d0 |
| 1564 |
4: |
| 1565 |
vld1.16 {q8}, [\src], \s_strd |
| 1566 |
vld1.16 {q11}, [\sr2], \s_strd |
| 1567 |
vext.8 d18, d16, d17, #2 |
| 1568 |
vext.8 d19, d16, d17, #4 |
| 1569 |
vext.8 d20, d16, d17, #6 |
| 1570 |
vext.8 d24, d22, d23, #2 |
| 1571 |
vext.8 d25, d22, d23, #4 |
| 1572 |
vext.8 d21, d22, d23, #6 |
| 1573 |
subs \h, \h, #2 |
| 1574 |
vmull.s16 q2, d16, d0[0] |
| 1575 |
vmlal.s16 q2, d18, d0[1] |
| 1576 |
vmlal.s16 q2, d19, d0[2] |
| 1577 |
vmlal.s16 q2, d20, d0[3] |
| 1578 |
vmull.s16 q3, d22, d0[0] |
| 1579 |
vmlal.s16 q3, d24, d0[1] |
| 1580 |
vmlal.s16 q3, d25, d0[2] |
| 1581 |
vmlal.s16 q3, d21, d0[3] |
| 1582 |
vrshl.s32 q2, q2, q14 // -(6-intermediate_bits) |
| 1583 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 1584 |
.ifc \type, put |
| 1585 |
vqmovun.s32 d4, q2 |
| 1586 |
vqmovun.s32 d5, q3 |
| 1587 |
vrshl.s16 q2, q2, q13 // -intermediate_bits |
| 1588 |
vmin.u16 q2, q2, q15 |
| 1589 |
.else |
| 1590 |
vmovn.s32 d4, q2 |
| 1591 |
vmovn.s32 d5, q3 |
| 1592 |
vsub.i16 q2, q2, q13 // PREP_BIAS |
| 1593 |
.endif |
| 1594 |
vst1.16 {d4}, [\dst, :64], \d_strd |
| 1595 |
vst1.16 {d5}, [\ds2, :64], \d_strd |
| 1596 |
bgt 4b |
| 1597 |
pop {r4-r11,pc} |
| 1598 |
|
| 1599 |
80: |
| 1600 |
160: |
| 1601 |
320: |
| 1602 |
640: |
| 1603 |
1280: // 8xN, 16xN, 32xN, ... h |
| 1604 |
vpush {q4-q5} |
| 1605 |
vld1.8 {d0}, [\mx, :64] |
| 1606 |
sub \src, \src, #6 |
| 1607 |
add \ds2, \dst, \d_strd |
| 1608 |
add \sr2, \src, \s_strd |
| 1609 |
lsl \s_strd, \s_strd, #1 |
| 1610 |
vmovl.s8 q0, d0 |
| 1611 |
|
| 1612 |
sub \s_strd, \s_strd, \w, lsl #1 |
| 1613 |
sub \s_strd, \s_strd, #16 |
| 1614 |
.ifc \type, put |
| 1615 |
lsl \d_strd, \d_strd, #1 |
| 1616 |
sub \d_strd, \d_strd, \w, lsl #1 |
| 1617 |
.endif |
| 1618 |
81: |
| 1619 |
vld1.16 {q8, q9}, [\src]! |
| 1620 |
vld1.16 {q10, q11}, [\sr2]! |
| 1621 |
mov \mx, \w |
| 1622 |
|
| 1623 |
8: |
| 1624 |
vmull.s16 q1, d16, d0[0] |
| 1625 |
vmull.s16 q2, d17, d0[0] |
| 1626 |
vmull.s16 q3, d20, d0[0] |
| 1627 |
vmull.s16 q4, d21, d0[0] |
| 1628 |
.irpc i, 1234567 |
| 1629 |
vext.8 q12, q8, q9, #(2*\i) |
| 1630 |
vext.8 q5, q10, q11, #(2*\i) |
| 1631 |
.if \i < 4 |
| 1632 |
vmlal.s16 q1, d24, d0[\i] |
| 1633 |
vmlal.s16 q2, d25, d0[\i] |
| 1634 |
vmlal.s16 q3, d10, d0[\i] |
| 1635 |
vmlal.s16 q4, d11, d0[\i] |
| 1636 |
.else |
| 1637 |
vmlal.s16 q1, d24, d1[\i-4] |
| 1638 |
vmlal.s16 q2, d25, d1[\i-4] |
| 1639 |
vmlal.s16 q3, d10, d1[\i-4] |
| 1640 |
vmlal.s16 q4, d11, d1[\i-4] |
| 1641 |
.endif |
| 1642 |
.endr |
| 1643 |
subs \mx, \mx, #8 |
| 1644 |
vrshl.s32 q1, q1, q14 // -(6-intermediate_bits) |
| 1645 |
vrshl.s32 q2, q2, q14 // -(6-intermediate_bits) |
| 1646 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 1647 |
vrshl.s32 q4, q4, q14 // -(6-intermediate_bits) |
| 1648 |
.ifc \type, put |
| 1649 |
vqmovun.s32 d2, q1 |
| 1650 |
vqmovun.s32 d3, q2 |
| 1651 |
vqmovun.s32 d4, q3 |
| 1652 |
vqmovun.s32 d5, q4 |
| 1653 |
vrshl.s16 q1, q1, q13 // -intermediate_bits |
| 1654 |
vrshl.s16 q2, q2, q13 // -intermediate_bits |
| 1655 |
vmin.u16 q1, q1, q15 |
| 1656 |
vmin.u16 q2, q2, q15 |
| 1657 |
.else |
| 1658 |
vmovn.s32 d2, q1 |
| 1659 |
vmovn.s32 d3, q2 |
| 1660 |
vmovn.s32 d4, q3 |
| 1661 |
vmovn.s32 d5, q4 |
| 1662 |
vsub.i16 q1, q1, q13 // PREP_BIAS |
| 1663 |
vsub.i16 q2, q2, q13 // PREP_BIAS |
| 1664 |
.endif |
| 1665 |
vst1.16 {q1}, [\dst, :128]! |
| 1666 |
vst1.16 {q2}, [\ds2, :128]! |
| 1667 |
ble 9f |
| 1668 |
|
| 1669 |
vmov q8, q9 |
| 1670 |
vmov q10, q11 |
| 1671 |
vld1.16 {q9}, [\src]! |
| 1672 |
vld1.16 {q11}, [\sr2]! |
| 1673 |
b 8b |
| 1674 |
|
| 1675 |
9: |
| 1676 |
add \dst, \dst, \d_strd |
| 1677 |
add \ds2, \ds2, \d_strd |
| 1678 |
add \src, \src, \s_strd |
| 1679 |
add \sr2, \sr2, \s_strd |
| 1680 |
|
| 1681 |
subs \h, \h, #2 |
| 1682 |
bgt 81b |
| 1683 |
vpop {q4-q5} |
| 1684 |
pop {r4-r11,pc} |
| 1685 |
|
| 1686 |
|
| 1687 |
L(\type\()_8tap_v): |
| 1688 |
cmp \h, #4 |
| 1689 |
ubfx r10, \my, #7, #7 |
| 1690 |
and \my, \my, #0x7f |
| 1691 |
it gt |
| 1692 |
movgt \my, r10 |
| 1693 |
add \my, r11, \my, lsl #3 |
| 1694 |
|
| 1695 |
.ifc \type, prep |
| 1696 |
vdup.32 q14, r12 // 6 - intermediate_bits |
| 1697 |
vmov.i16 q15, #PREP_BIAS |
| 1698 |
.endif |
| 1699 |
adr r10, L(\type\()_8tap_v_tbl) |
| 1700 |
ldr r9, [r10, r9, lsl #2] |
| 1701 |
.ifc \type, prep |
| 1702 |
vneg.s32 q14, q14 // -(6-intermediate_bits) |
| 1703 |
.endif |
| 1704 |
add r10, r10, r9 |
| 1705 |
bx r10 |
| 1706 |
|
| 1707 |
.align 2 |
| 1708 |
L(\type\()_8tap_v_tbl): |
| 1709 |
.word 1280f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1710 |
.word 640f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1711 |
.word 320f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1712 |
.word 160f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1713 |
.word 80f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1714 |
.word 40f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1715 |
.word 20f - L(\type\()_8tap_v_tbl) + CONFIG_THUMB |
| 1716 |
|
| 1717 |
20: // 2xN v |
| 1718 |
.ifc \type, put |
| 1719 |
bgt 28f |
| 1720 |
|
| 1721 |
cmp \h, #2 |
| 1722 |
add \my, \my, #2 |
| 1723 |
vld1.32 {d0[]}, [\my] |
| 1724 |
sub \src, \src, \s_strd |
| 1725 |
add \ds2, \dst, \d_strd |
| 1726 |
add \sr2, \src, \s_strd |
| 1727 |
lsl \s_strd, \s_strd, #1 |
| 1728 |
lsl \d_strd, \d_strd, #1 |
| 1729 |
vmovl.s8 q0, d0 |
| 1730 |
|
| 1731 |
// 2x2 v |
| 1732 |
load_32 \src, \sr2, \s_strd, d1, d2, d3, d4, d5 |
| 1733 |
interleave_1_32 d1, d2, d3, d4, d5 |
| 1734 |
bgt 24f |
| 1735 |
vmull_vmlal_4 q8, d1, d2, d3, d4 |
| 1736 |
vqrshrun_s32 6, q8, d16 |
| 1737 |
vmin_u16 d30, d16 |
| 1738 |
vst1_32 \d_strd, d16 |
| 1739 |
pop {r4-r11,pc} |
| 1740 |
|
| 1741 |
24: // 2x4 v |
| 1742 |
load_32 \sr2, \src, \s_strd, d6, d7 |
| 1743 |
interleave_1_32 d5, d6, d7 |
| 1744 |
vmull_vmlal_4 q8, d1, d2, d3, d4 |
| 1745 |
vmull_vmlal_4 q9, d3, d4, d5, d6 |
| 1746 |
vqrshrun_s32 6, q8, d16, q9, d17 |
| 1747 |
vmin_u16 q15, q8 |
| 1748 |
vst1_32 \d_strd, d16, d17 |
| 1749 |
pop {r4-r11,pc} |
| 1750 |
|
| 1751 |
28: // 2x6, 2x8, 2x12, 2x16 v |
| 1752 |
vld1.8 {d0}, [\my, :64] |
| 1753 |
sub \sr2, \src, \s_strd, lsl #1 |
| 1754 |
add \ds2, \dst, \d_strd |
| 1755 |
sub \src, \sr2, \s_strd |
| 1756 |
lsl \d_strd, \d_strd, #1 |
| 1757 |
lsl \s_strd, \s_strd, #1 |
| 1758 |
vmovl.s8 q0, d0 |
| 1759 |
|
| 1760 |
load_32 \src, \sr2, \s_strd, d2, d3, d4, d5, d6, d7, d16 |
| 1761 |
interleave_1_32 d2, d3, d4, d5, d6 |
| 1762 |
interleave_1_32 d6, d7, d16 |
| 1763 |
216: |
| 1764 |
subs \h, \h, #4 |
| 1765 |
load_32 \sr2, \src, \s_strd, d17, d18, d19, d20 |
| 1766 |
interleave_1_32 d16, d17, d18, d19, d20 |
| 1767 |
vmull_vmlal_8 q13, d2, d3, d4, d5, d6, d7, d16, d17 |
| 1768 |
vmull_vmlal_8 q1, d4, d5, d6, d7, d16, d17, d18, d19 |
| 1769 |
vqrshrun_s32 6, q13, d26, q1, d27 |
| 1770 |
vmin_u16 q15, q13 |
| 1771 |
vst1_32 \d_strd, d26, d27 |
| 1772 |
ble 0f |
| 1773 |
cmp \h, #2 |
| 1774 |
vmov q1, q3 |
| 1775 |
vmov q2, q8 |
| 1776 |
vmov q3, q9 |
| 1777 |
vmov d16, d20 |
| 1778 |
beq 26f |
| 1779 |
b 216b |
| 1780 |
26: |
| 1781 |
load_32 \sr2, \src, \s_strd, d17, d18 |
| 1782 |
interleave_1_32 d16, d17, d18 |
| 1783 |
vmull_vmlal_8 q13, d2, d3, d4, d5, d6, d7, d16, d17 |
| 1784 |
vqrshrun_s32 6, q13, d26 |
| 1785 |
vmin_u16 d30, d26 |
| 1786 |
vst1_32 \d_strd, d26 |
| 1787 |
0: |
| 1788 |
pop {r4-r11,pc} |
| 1789 |
.endif |
| 1790 |
|
| 1791 |
40: |
| 1792 |
bgt 480f |
| 1793 |
|
| 1794 |
// 4x2, 4x4 v |
| 1795 |
cmp \h, #2 |
| 1796 |
add \my, \my, #2 |
| 1797 |
vld1.32 {d0[]}, [\my] |
| 1798 |
sub \src, \src, \s_strd |
| 1799 |
add \ds2, \dst, \d_strd |
| 1800 |
add \sr2, \src, \s_strd |
| 1801 |
lsl \s_strd, \s_strd, #1 |
| 1802 |
lsl \d_strd, \d_strd, #1 |
| 1803 |
vmovl.s8 q0, d0 |
| 1804 |
|
| 1805 |
load_reg \src, \sr2, \s_strd, d1, d2, d3, d4, d5 |
| 1806 |
vmull_vmlal_4 q8, d1, d2, d3, d4 |
| 1807 |
vmull_vmlal_4 q9, d2, d3, d4, d5 |
| 1808 |
shift_store_4 \type, \d_strd, q8, q9, d16, d17 |
| 1809 |
ble 0f |
| 1810 |
load_reg \sr2, \src, \s_strd, d6, d7 |
| 1811 |
vmull_vmlal_4 q8, d3, d4, d5, d6 |
| 1812 |
vmull_vmlal_4 q9, d4, d5, d6, d7 |
| 1813 |
shift_store_4 \type, \d_strd, q8, q9, d16, d17 |
| 1814 |
0: |
| 1815 |
pop {r4-r11,pc} |
| 1816 |
|
| 1817 |
480: // 4x6, 4x8, 4x12, 4x16 v |
| 1818 |
vld1.8 {d0}, [\my, :64] |
| 1819 |
sub \sr2, \src, \s_strd, lsl #1 |
| 1820 |
add \ds2, \dst, \d_strd |
| 1821 |
sub \src, \sr2, \s_strd |
| 1822 |
lsl \s_strd, \s_strd, #1 |
| 1823 |
lsl \d_strd, \d_strd, #1 |
| 1824 |
vmovl.s8 q0, d0 |
| 1825 |
|
| 1826 |
load_reg \src, \sr2, \s_strd, d16, d17, d18, d19, d20, d21, d22 |
| 1827 |
|
| 1828 |
48: |
| 1829 |
subs \h, \h, #4 |
| 1830 |
load_reg \sr2, \src, \s_strd, d23, d24, d25, d26 |
| 1831 |
vmull_vmlal_8 q1, d16, d17, d18, d19, d20, d21, d22, d23 |
| 1832 |
vmull_vmlal_8 q2, d17, d18, d19, d20, d21, d22, d23, d24 |
| 1833 |
vmull_vmlal_8 q3, d18, d19, d20, d21, d22, d23, d24, d25 |
| 1834 |
vmull_vmlal_8 q8, d19, d20, d21, d22, d23, d24, d25, d26 |
| 1835 |
shift_store_4 \type, \d_strd, q1, q2, d2, d3, q3, q8, d4, d5 |
| 1836 |
ble 0f |
| 1837 |
cmp \h, #2 |
| 1838 |
vmov q8, q10 |
| 1839 |
vmov q9, q11 |
| 1840 |
vmov q10, q12 |
| 1841 |
vmov d22, d26 |
| 1842 |
beq 46f |
| 1843 |
b 48b |
| 1844 |
46: |
| 1845 |
load_reg \sr2, \src, \s_strd, d23, d24 |
| 1846 |
vmull_vmlal_8 q1, d16, d17, d18, d19, d20, d21, d22, d23 |
| 1847 |
vmull_vmlal_8 q2, d17, d18, d19, d20, d21, d22, d23, d24 |
| 1848 |
shift_store_4 \type, \d_strd, q1, q2, d2, d3 |
| 1849 |
0: |
| 1850 |
pop {r4-r11,pc} |
| 1851 |
|
| 1852 |
80: |
| 1853 |
bgt 880f |
| 1854 |
|
| 1855 |
// 8x2, 8x4 v |
| 1856 |
cmp \h, #2 |
| 1857 |
add \my, \my, #2 |
| 1858 |
vld1.32 {d0[]}, [\my] |
| 1859 |
sub \src, \src, \s_strd |
| 1860 |
add \ds2, \dst, \d_strd |
| 1861 |
add \sr2, \src, \s_strd |
| 1862 |
lsl \s_strd, \s_strd, #1 |
| 1863 |
lsl \d_strd, \d_strd, #1 |
| 1864 |
vmovl.s8 q0, d0 |
| 1865 |
|
| 1866 |
load_reg \src, \sr2, \s_strd, q1, q2, q3, q8, q9 |
| 1867 |
vmull_vmlal_4 q10, d2, d4, d6, d16 |
| 1868 |
vmull_vmlal_4 q11, d3, d5, d7, d17 |
| 1869 |
vmull_vmlal_4 q12, d4, d6, d16, d18 |
| 1870 |
vmull_vmlal_4 q13, d5, d7, d17, d19 |
| 1871 |
shift_store_8 \type, \d_strd, q10, q11, d20, d21, q12, q13, d22, d23 |
| 1872 |
ble 0f |
| 1873 |
load_reg \sr2, \src, \s_strd, q10, q11 |
| 1874 |
vmull_vmlal_4 q1, d6, d16, d18, d20 |
| 1875 |
vmull_vmlal_4 q2, d7, d17, d19, d21 |
| 1876 |
vmull_vmlal_4 q12, d16, d18, d20, d22 |
| 1877 |
vmull_vmlal_4 q13, d17, d19, d21, d23 |
| 1878 |
shift_store_8 \type, \d_strd, q1, q2, d2, d3, q12, q13, d4, d5 |
| 1879 |
0: |
| 1880 |
pop {r4-r11,pc} |
| 1881 |
|
| 1882 |
880: // 8x6, 8x8, 8x16, 8x32 v |
| 1883 |
1680: // 16x8, 16x16, ... |
| 1884 |
320: // 32x8, 32x16, ... |
| 1885 |
640: |
| 1886 |
1280: |
| 1887 |
vpush {q4-q7} |
| 1888 |
vld1.8 {d0}, [\my, :64] |
| 1889 |
sub \src, \src, \s_strd |
| 1890 |
sub \src, \src, \s_strd, lsl #1 |
| 1891 |
vmovl.s8 q0, d0 |
| 1892 |
mov \my, \h |
| 1893 |
168: |
| 1894 |
add \ds2, \dst, \d_strd |
| 1895 |
add \sr2, \src, \s_strd |
| 1896 |
lsl \s_strd, \s_strd, #1 |
| 1897 |
lsl \d_strd, \d_strd, #1 |
| 1898 |
|
| 1899 |
load_reg \src, \sr2, \s_strd, q5, q6, q7, q8, q9, q10, q11 |
| 1900 |
|
| 1901 |
88: |
| 1902 |
subs \h, \h, #2 |
| 1903 |
load_reg \sr2, \src, \s_strd, q12, q13 |
| 1904 |
vmull_vmlal_8 q1, d10, d12, d14, d16, d18, d20, d22, d24 |
| 1905 |
vmull_vmlal_8 q2, d11, d13, d15, d17, d19, d21, d23, d25 |
| 1906 |
vmull_vmlal_8 q3, d12, d14, d16, d18, d20, d22, d24, d26 |
| 1907 |
vmull_vmlal_8 q4, d13, d15, d17, d19, d21, d23, d25, d27 |
| 1908 |
shift_store_8 \type, \d_strd, q1, q2, d2, d3, q3, q4, d4, d5 |
| 1909 |
ble 9f |
| 1910 |
subs \h, \h, #2 |
| 1911 |
load_reg \sr2, \src, \s_strd, q1, q2 |
| 1912 |
vmull_vmlal_8 q3, d14, d16, d18, d20, d22, d24, d26, d2 |
| 1913 |
vmull_vmlal_8 q4, d15, d17, d19, d21, d23, d25, d27, d3 |
| 1914 |
vmull_vmlal_8 q5, d16, d18, d20, d22, d24, d26, d2, d4 |
| 1915 |
vmull_vmlal_8 q6, d17, d19, d21, d23, d25, d27, d3, d5 |
| 1916 |
shift_store_8 \type, \d_strd, q3, q4, d6, d7, q5, q6, d8, d9 |
| 1917 |
ble 9f |
| 1918 |
vmov q5, q9 |
| 1919 |
vmov q6, q10 |
| 1920 |
vmov q7, q11 |
| 1921 |
vmov q8, q12 |
| 1922 |
vmov q9, q13 |
| 1923 |
vmov q10, q1 |
| 1924 |
vmov q11, q2 |
| 1925 |
b 88b |
| 1926 |
9: |
| 1927 |
subs \w, \w, #8 |
| 1928 |
ble 0f |
| 1929 |
asr \s_strd, \s_strd, #1 |
| 1930 |
asr \d_strd, \d_strd, #1 |
| 1931 |
mls \src, \s_strd, \my, \src |
| 1932 |
mls \dst, \d_strd, \my, \dst |
| 1933 |
sub \src, \src, \s_strd, lsl #3 |
| 1934 |
mov \h, \my |
| 1935 |
add \src, \src, #16 |
| 1936 |
add \dst, \dst, #16 |
| 1937 |
b 168b |
| 1938 |
0: |
| 1939 |
vpop {q4-q7} |
| 1940 |
pop {r4-r11,pc} |
| 1941 |
|
| 1942 |
160: |
| 1943 |
bgt 1680b |
| 1944 |
|
| 1945 |
// 16x2, 16x4 v |
| 1946 |
vpush {q6-q7} |
| 1947 |
add \my, \my, #2 |
| 1948 |
vld1.32 {d0[]}, [\my] |
| 1949 |
sub \src, \src, \s_strd |
| 1950 |
vmovl.s8 q0, d0 |
| 1951 |
|
| 1952 |
load_16s16 \src, \src, \s_strd, q6, q7, q8, q9, q10, q11 |
| 1953 |
16: |
| 1954 |
load_16s16 \src, \src, \s_strd, q12, q13 |
| 1955 |
subs \h, \h, #1 |
| 1956 |
vmull_vmlal_4 q1, d12, d16, d20, d24 |
| 1957 |
vmull_vmlal_4 q2, d13, d17, d21, d25 |
| 1958 |
vmull_vmlal_4 q3, d14, d18, d22, d26 |
| 1959 |
vmull_vmlal_4 q6, d15, d19, d23, d27 |
| 1960 |
shift_store_16 \type, \d_strd, q1, q2, d2, d3, q3, q6, d4, d5 |
| 1961 |
ble 0f |
| 1962 |
vmov q6, q8 |
| 1963 |
vmov q7, q9 |
| 1964 |
vmov q8, q10 |
| 1965 |
vmov q9, q11 |
| 1966 |
vmov q10, q12 |
| 1967 |
vmov q11, q13 |
| 1968 |
b 16b |
| 1969 |
0: |
| 1970 |
vpop {q6-q7} |
| 1971 |
pop {r4-r11,pc} |
| 1972 |
|
| 1973 |
|
| 1974 |
L(\type\()_8tap_hv): |
| 1975 |
cmp \h, #4 |
| 1976 |
ubfx r10, \my, #7, #7 |
| 1977 |
and \my, \my, #0x7f |
| 1978 |
it gt |
| 1979 |
movgt \my, r10 |
| 1980 |
4: |
| 1981 |
add \my, r11, \my, lsl #3 |
| 1982 |
|
| 1983 |
adr r10, L(\type\()_8tap_hv_tbl) |
| 1984 |
neg r12, r12 // -(6-intermediate_bits) |
| 1985 |
ldr r9, [r10, r9, lsl #2] |
| 1986 |
vdup.32 q14, r12 // -(6-intermediate_bits) |
| 1987 |
.ifc \type, put |
| 1988 |
neg r8, lr // -(6+intermeidate_bits) |
| 1989 |
.else |
| 1990 |
vmov.i16 q13, #PREP_BIAS |
| 1991 |
.endif |
| 1992 |
add r10, r10, r9 |
| 1993 |
.ifc \type, put |
| 1994 |
vdup.32 q13, r8 // -(6+intermediate_bits) |
| 1995 |
.endif |
| 1996 |
bx r10 |
| 1997 |
|
| 1998 |
.align 2 |
| 1999 |
L(\type\()_8tap_hv_tbl): |
| 2000 |
.word 1280f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2001 |
.word 640f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2002 |
.word 320f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2003 |
.word 160f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2004 |
.word 80f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2005 |
.word 40f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2006 |
.word 20f - L(\type\()_8tap_hv_tbl) + CONFIG_THUMB |
| 2007 |
|
| 2008 |
20: |
| 2009 |
.ifc \type, put |
| 2010 |
add \mx, \mx, #2 |
| 2011 |
vld1.32 {d0[]}, [\mx] |
| 2012 |
bgt 280f |
| 2013 |
add \my, \my, #2 |
| 2014 |
vld1.32 {d2[]}, [\my] |
| 2015 |
|
| 2016 |
// 2x2, 2x4 hv |
| 2017 |
sub \sr2, \src, #2 |
| 2018 |
sub \src, \sr2, \s_strd |
| 2019 |
add \ds2, \dst, \d_strd |
| 2020 |
lsl \s_strd, \s_strd, #1 |
| 2021 |
lsl \d_strd, \d_strd, #1 |
| 2022 |
vmovl.s8 q0, d0 |
| 2023 |
vmovl.s8 q1, d2 |
| 2024 |
|
| 2025 |
vld1.16 {q11}, [\src], \s_strd |
| 2026 |
vext.8 d24, d22, d23, #2 |
| 2027 |
vmull.s16 q11, d22, d0 |
| 2028 |
vmull.s16 q12, d24, d0 |
| 2029 |
vpadd.s32 d22, d22, d23 |
| 2030 |
vpadd.s32 d23, d24, d25 |
| 2031 |
vpadd.s32 d22, d22, d23 |
| 2032 |
vrshl.s32 d16, d22, d28 // -(6-intermediate_bits) |
| 2033 |
vmovn.i32 d16, q8 |
| 2034 |
bl L(\type\()_8tap_filter_2) |
| 2035 |
|
| 2036 |
vext.8 d16, d16, d16, #4 |
| 2037 |
vext.8 d16, d16, d24, #4 |
| 2038 |
vmov d17, d24 |
| 2039 |
|
| 2040 |
2: |
| 2041 |
bl L(\type\()_8tap_filter_2) |
| 2042 |
|
| 2043 |
vext.8 d18, d17, d24, #4 |
| 2044 |
vmull.s16 q2, d16, d2[0] |
| 2045 |
vmlal.s16 q2, d17, d2[1] |
| 2046 |
vmlal.s16 q2, d18, d2[2] |
| 2047 |
vmlal.s16 q2, d24, d2[3] |
| 2048 |
|
| 2049 |
vrshl.s32 q2, q2, q13 // -(6+intermediate_bits) |
| 2050 |
vqmovun.s32 d4, q2 |
| 2051 |
vmin.u16 d4, d4, d30 |
| 2052 |
subs \h, \h, #2 |
| 2053 |
vst1.32 {d4[0]}, [\dst, :32], \d_strd |
| 2054 |
vst1.32 {d4[1]}, [\ds2, :32], \d_strd |
| 2055 |
ble 0f |
| 2056 |
vmov d16, d18 |
| 2057 |
vmov d17, d24 |
| 2058 |
b 2b |
| 2059 |
|
| 2060 |
280: // 2x8, 2x16, 2x32 hv |
| 2061 |
vld1.8 {d2}, [\my, :64] |
| 2062 |
sub \src, \src, #2 |
| 2063 |
sub \sr2, \src, \s_strd, lsl #1 |
| 2064 |
sub \src, \sr2, \s_strd |
| 2065 |
add \ds2, \dst, \d_strd |
| 2066 |
lsl \s_strd, \s_strd, #1 |
| 2067 |
lsl \d_strd, \d_strd, #1 |
| 2068 |
vmovl.s8 q0, d0 |
| 2069 |
vmovl.s8 q1, d2 |
| 2070 |
|
| 2071 |
vld1.16 {q11}, [\src], \s_strd |
| 2072 |
vext.8 d24, d22, d23, #2 |
| 2073 |
vmull.s16 q11, d22, d0 |
| 2074 |
vmull.s16 q12, d24, d0 |
| 2075 |
vpadd.s32 d22, d22, d23 |
| 2076 |
vpadd.s32 d23, d24, d25 |
| 2077 |
vpadd.s32 d22, d22, d23 |
| 2078 |
vrshl.s32 d16, d22, d28 // -(6-intermediate_bits) |
| 2079 |
vmovn.i32 d16, q8 |
| 2080 |
|
| 2081 |
bl L(\type\()_8tap_filter_2) |
| 2082 |
|
| 2083 |
vext.8 d16, d16, d16, #4 |
| 2084 |
vext.8 d16, d16, d24, #4 |
| 2085 |
vmov d17, d24 |
| 2086 |
bl L(\type\()_8tap_filter_2) |
| 2087 |
vext.8 d18, d17, d24, #4 |
| 2088 |
vmov d19, d24 |
| 2089 |
bl L(\type\()_8tap_filter_2) |
| 2090 |
vext.8 d20, d19, d24, #4 |
| 2091 |
vmov d21, d24 |
| 2092 |
|
| 2093 |
28: |
| 2094 |
bl L(\type\()_8tap_filter_2) |
| 2095 |
vext.8 d22, d21, d24, #4 |
| 2096 |
vmull.s16 q3, d16, d2[0] |
| 2097 |
vmlal.s16 q3, d17, d2[1] |
| 2098 |
vmlal.s16 q3, d18, d2[2] |
| 2099 |
vmlal.s16 q3, d19, d2[3] |
| 2100 |
vmlal.s16 q3, d20, d3[0] |
| 2101 |
vmlal.s16 q3, d21, d3[1] |
| 2102 |
vmlal.s16 q3, d22, d3[2] |
| 2103 |
vmlal.s16 q3, d24, d3[3] |
| 2104 |
|
| 2105 |
vrshl.s32 q3, q3, q13 // -(6+intermediate_bits) |
| 2106 |
vqmovun.s32 d6, q3 |
| 2107 |
vmin.u16 d6, d6, d30 |
| 2108 |
subs \h, \h, #2 |
| 2109 |
vst1.32 {d6[0]}, [\dst, :32], \d_strd |
| 2110 |
vst1.32 {d6[1]}, [\ds2, :32], \d_strd |
| 2111 |
ble 0f |
| 2112 |
vmov q8, q9 |
| 2113 |
vmov q9, q10 |
| 2114 |
vmov d20, d22 |
| 2115 |
vmov d21, d24 |
| 2116 |
b 28b |
| 2117 |
0: |
| 2118 |
pop {r4-r11,pc} |
| 2119 |
|
| 2120 |
L(\type\()_8tap_filter_2): |
| 2121 |
vld1.16 {q11}, [\sr2], \s_strd |
| 2122 |
vld1.16 {q12}, [\src], \s_strd |
| 2123 |
vext.8 d23, d22, d23, #2 |
| 2124 |
vext.8 d25, d24, d25, #2 |
| 2125 |
vtrn.32 q11, q12 |
| 2126 |
vmull.s16 q3, d22, d0[0] |
| 2127 |
vmlal.s16 q3, d23, d0[1] |
| 2128 |
vmlal.s16 q3, d24, d0[2] |
| 2129 |
vmlal.s16 q3, d25, d0[3] |
| 2130 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 2131 |
vmovn.i32 d24, q3 |
| 2132 |
bx lr |
| 2133 |
.endif |
| 2134 |
|
| 2135 |
40: |
| 2136 |
add \mx, \mx, #2 |
| 2137 |
vld1.32 {d0[]}, [\mx] |
| 2138 |
bgt 480f |
| 2139 |
add \my, \my, #2 |
| 2140 |
vld1.32 {d2[]}, [\my] |
| 2141 |
sub \sr2, \src, #2 |
| 2142 |
sub \src, \sr2, \s_strd |
| 2143 |
add \ds2, \dst, \d_strd |
| 2144 |
lsl \s_strd, \s_strd, #1 |
| 2145 |
lsl \d_strd, \d_strd, #1 |
| 2146 |
vmovl.s8 q0, d0 |
| 2147 |
vmovl.s8 q1, d2 |
| 2148 |
|
| 2149 |
// 4x2, 4x4 hv |
| 2150 |
vld1.16 {q11}, [\src], \s_strd |
| 2151 |
vext.8 d24, d22, d23, #2 |
| 2152 |
vext.8 d25, d22, d23, #4 |
| 2153 |
vext.8 d23, d22, d23, #6 |
| 2154 |
vmull.s16 q10, d22, d0[0] |
| 2155 |
vmlal.s16 q10, d24, d0[1] |
| 2156 |
vmlal.s16 q10, d25, d0[2] |
| 2157 |
vmlal.s16 q10, d23, d0[3] |
| 2158 |
vrshl.s32 q10, q10, q14 // -(6-intermediate_bits) |
| 2159 |
vmovn.i32 d17, q10 |
| 2160 |
|
| 2161 |
bl L(\type\()_8tap_filter_4) |
| 2162 |
vmov q9, q12 |
| 2163 |
|
| 2164 |
4: |
| 2165 |
bl L(\type\()_8tap_filter_4) |
| 2166 |
vmull.s16 q2, d17, d2[0] |
| 2167 |
vmlal.s16 q2, d18, d2[1] |
| 2168 |
vmlal.s16 q2, d19, d2[2] |
| 2169 |
vmlal.s16 q2, d24, d2[3] |
| 2170 |
vmull.s16 q3, d18, d2[0] |
| 2171 |
vmlal.s16 q3, d19, d2[1] |
| 2172 |
vmlal.s16 q3, d24, d2[2] |
| 2173 |
vmlal.s16 q3, d25, d2[3] |
| 2174 |
.ifc \type, put |
| 2175 |
vrshl.s32 q2, q2, q13 // -(6+intermediate_bits) |
| 2176 |
vrshl.s32 q3, q3, q13 // -(6+intermediate_bits) |
| 2177 |
vqmovun.s32 d4, q2 |
| 2178 |
vqmovun.s32 d5, q3 |
| 2179 |
vmin.u16 q2, q2, q15 |
| 2180 |
.else |
| 2181 |
vrshrn.i32 d4, q2, #6 |
| 2182 |
vrshrn.i32 d5, q3, #6 |
| 2183 |
vsub.i16 q2, q2, q13 // PREP_BIAS |
| 2184 |
.endif |
| 2185 |
subs \h, \h, #2 |
| 2186 |
|
| 2187 |
vst1.16 {d4}, [\dst, :64], \d_strd |
| 2188 |
vst1.16 {d5}, [\ds2, :64], \d_strd |
| 2189 |
ble 0f |
| 2190 |
vmov d17, d19 |
| 2191 |
vmov q9, q12 |
| 2192 |
b 4b |
| 2193 |
0: |
| 2194 |
pop {r4-r11,pc} |
| 2195 |
|
| 2196 |
480: // 4x8, 4x16, 4x32 hv |
| 2197 |
vpush {d13-d15} |
| 2198 |
vld1.8 {d2}, [\my, :64] |
| 2199 |
sub \src, \src, #2 |
| 2200 |
sub \sr2, \src, \s_strd, lsl #1 |
| 2201 |
sub \src, \sr2, \s_strd |
| 2202 |
add \ds2, \dst, \d_strd |
| 2203 |
lsl \s_strd, \s_strd, #1 |
| 2204 |
lsl \d_strd, \d_strd, #1 |
| 2205 |
vmovl.s8 q0, d0 |
| 2206 |
vmovl.s8 q1, d2 |
| 2207 |
|
| 2208 |
vld1.16 {q11}, [\src], \s_strd |
| 2209 |
vext.8 d24, d22, d23, #2 |
| 2210 |
vext.8 d25, d22, d23, #4 |
| 2211 |
vext.8 d23, d22, d23, #6 |
| 2212 |
vmull.s16 q10, d22, d0[0] |
| 2213 |
vmlal.s16 q10, d24, d0[1] |
| 2214 |
vmlal.s16 q10, d25, d0[2] |
| 2215 |
vmlal.s16 q10, d23, d0[3] |
| 2216 |
vrshl.s32 q10, q10, q14 // -(6-intermediate_bits) |
| 2217 |
vmovn.i32 d13, q10 |
| 2218 |
|
| 2219 |
bl L(\type\()_8tap_filter_4) |
| 2220 |
vmov q7, q12 |
| 2221 |
bl L(\type\()_8tap_filter_4) |
| 2222 |
vmov q8, q12 |
| 2223 |
bl L(\type\()_8tap_filter_4) |
| 2224 |
vmov q9, q12 |
| 2225 |
|
| 2226 |
48: |
| 2227 |
bl L(\type\()_8tap_filter_4) |
| 2228 |
vmull.s16 q2, d13, d2[0] |
| 2229 |
vmlal.s16 q2, d14, d2[1] |
| 2230 |
vmlal.s16 q2, d15, d2[2] |
| 2231 |
vmlal.s16 q2, d16, d2[3] |
| 2232 |
vmlal.s16 q2, d17, d3[0] |
| 2233 |
vmlal.s16 q2, d18, d3[1] |
| 2234 |
vmlal.s16 q2, d19, d3[2] |
| 2235 |
vmlal.s16 q2, d24, d3[3] |
| 2236 |
vmull.s16 q3, d14, d2[0] |
| 2237 |
vmlal.s16 q3, d15, d2[1] |
| 2238 |
vmlal.s16 q3, d16, d2[2] |
| 2239 |
vmlal.s16 q3, d17, d2[3] |
| 2240 |
vmlal.s16 q3, d18, d3[0] |
| 2241 |
vmlal.s16 q3, d19, d3[1] |
| 2242 |
vmlal.s16 q3, d24, d3[2] |
| 2243 |
vmlal.s16 q3, d25, d3[3] |
| 2244 |
.ifc \type, put |
| 2245 |
vrshl.s32 q2, q2, q13 // -(6+intermediate_bits) |
| 2246 |
vrshl.s32 q3, q3, q13 // -(6+intermediate_bits) |
| 2247 |
vqmovun.s32 d4, q2 |
| 2248 |
vqmovun.s32 d5, q3 |
| 2249 |
vmin.u16 q2, q2, q15 |
| 2250 |
.else |
| 2251 |
vrshrn.i32 d4, q2, #6 |
| 2252 |
vrshrn.i32 d5, q3, #6 |
| 2253 |
vsub.i16 q2, q2, q13 // PREP_BIAS |
| 2254 |
.endif |
| 2255 |
subs \h, \h, #2 |
| 2256 |
vst1.16 {d4}, [\dst, :64], \d_strd |
| 2257 |
vst1.16 {d5}, [\ds2, :64], \d_strd |
| 2258 |
ble 0f |
| 2259 |
vmov d13, d15 |
| 2260 |
vmov q7, q8 |
| 2261 |
vmov q8, q9 |
| 2262 |
vmov q9, q12 |
| 2263 |
b 48b |
| 2264 |
0: |
| 2265 |
vpop {d13-d15} |
| 2266 |
pop {r4-r11,pc} |
| 2267 |
|
| 2268 |
L(\type\()_8tap_filter_4): |
| 2269 |
vld1.16 {q10}, [\sr2], \s_strd |
| 2270 |
vld1.16 {q11}, [\src], \s_strd |
| 2271 |
vext.8 d24, d20, d21, #2 |
| 2272 |
vext.8 d25, d20, d21, #4 |
| 2273 |
vext.8 d21, d20, d21, #6 |
| 2274 |
vmull.s16 q3, d20, d0[0] |
| 2275 |
vmlal.s16 q3, d24, d0[1] |
| 2276 |
vmlal.s16 q3, d25, d0[2] |
| 2277 |
vmlal.s16 q3, d21, d0[3] |
| 2278 |
vext.8 d24, d22, d23, #2 |
| 2279 |
vext.8 d25, d22, d23, #4 |
| 2280 |
vext.8 d23, d22, d23, #6 |
| 2281 |
vmull.s16 q10, d22, d0[0] |
| 2282 |
vmlal.s16 q10, d24, d0[1] |
| 2283 |
vmlal.s16 q10, d25, d0[2] |
| 2284 |
vmlal.s16 q10, d23, d0[3] |
| 2285 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 2286 |
vrshl.s32 q10, q10, q14 // -(6-intermediate_bits) |
| 2287 |
vmovn.i32 d24, q3 |
| 2288 |
vmovn.i32 d25, q10 |
| 2289 |
bx lr |
| 2290 |
|
| 2291 |
80: |
| 2292 |
160: |
| 2293 |
320: |
| 2294 |
bgt 880f |
| 2295 |
add \my, \my, #2 |
| 2296 |
vld1.8 {d0}, [\mx, :64] |
| 2297 |
vld1.32 {d2[]}, [\my] |
| 2298 |
sub \src, \src, #6 |
| 2299 |
sub \src, \src, \s_strd |
| 2300 |
vmovl.s8 q0, d0 |
| 2301 |
vmovl.s8 q1, d2 |
| 2302 |
mov \my, \h |
| 2303 |
|
| 2304 |
164: // 8x2, 8x4, 16x2, 16x4, 32x2, 32x4 hv |
| 2305 |
add \ds2, \dst, \d_strd |
| 2306 |
add \sr2, \src, \s_strd |
| 2307 |
lsl \d_strd, \d_strd, #1 |
| 2308 |
lsl \s_strd, \s_strd, #1 |
| 2309 |
|
| 2310 |
vld1.16 {q11, q12}, [\src], \s_strd |
| 2311 |
vmull.s16 q2, d22, d0[0] |
| 2312 |
vmull.s16 q3, d23, d0[0] |
| 2313 |
vdup.32 q14, r12 // -(6-intermediate_bits) |
| 2314 |
.irpc i, 1234567 |
| 2315 |
vext.8 q10, q11, q12, #(2*\i) |
| 2316 |
.if \i < 4 |
| 2317 |
vmlal.s16 q2, d20, d0[\i] |
| 2318 |
vmlal.s16 q3, d21, d0[\i] |
| 2319 |
.else |
| 2320 |
vmlal.s16 q2, d20, d1[\i - 4] |
| 2321 |
vmlal.s16 q3, d21, d1[\i - 4] |
| 2322 |
.endif |
| 2323 |
.endr |
| 2324 |
vrshl.s32 q2, q2, q14 // -(6-intermediate_bits) |
| 2325 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 2326 |
vmovn.i32 d16, q2 |
| 2327 |
vmovn.i32 d17, q3 |
| 2328 |
|
| 2329 |
bl L(\type\()_8tap_filter_8) |
| 2330 |
vmov q9, q11 |
| 2331 |
vmov q10, q12 |
| 2332 |
|
| 2333 |
8: |
| 2334 |
bl L(\type\()_8tap_filter_8) |
| 2335 |
vmull.s16 q2, d16, d2[0] |
| 2336 |
vmull.s16 q3, d17, d2[0] |
| 2337 |
vmull.s16 q13, d18, d2[0] |
| 2338 |
vmull.s16 q14, d19, d2[0] |
| 2339 |
.ifc \type, put |
| 2340 |
vdup.32 q8, r8 // -(6+intermediate_bits) |
| 2341 |
.endif |
| 2342 |
vmlal.s16 q2, d18, d2[1] |
| 2343 |
vmlal.s16 q3, d19, d2[1] |
| 2344 |
vmlal.s16 q13, d20, d2[1] |
| 2345 |
vmlal.s16 q14, d21, d2[1] |
| 2346 |
vmlal.s16 q2, d20, d2[2] |
| 2347 |
vmlal.s16 q3, d21, d2[2] |
| 2348 |
vmlal.s16 q13, d22, d2[2] |
| 2349 |
vmlal.s16 q14, d23, d2[2] |
| 2350 |
vmlal.s16 q2, d22, d2[3] |
| 2351 |
vmlal.s16 q3, d23, d2[3] |
| 2352 |
vmlal.s16 q13, d24, d2[3] |
| 2353 |
vmlal.s16 q14, d25, d2[3] |
| 2354 |
.ifc \type, put |
| 2355 |
vdup.16 q9, \bdmax // bitdepth_max |
| 2356 |
vrshl.s32 q2, q2, q8 // -(6+intermediate_bits) |
| 2357 |
vrshl.s32 q3, q3, q8 // -(6+intermediate_bits) |
| 2358 |
vrshl.s32 q13, q13, q8 // -(6+intermediate_bits) |
| 2359 |
vrshl.s32 q14, q14, q8 // -(6+intermediate_bits) |
| 2360 |
vqmovun.s32 d4, q2 |
| 2361 |
vqmovun.s32 d5, q3 |
| 2362 |
vqmovun.s32 d6, q13 |
| 2363 |
vqmovun.s32 d7, q14 |
| 2364 |
vmin.u16 q2, q2, q15 |
| 2365 |
vmin.u16 q3, q3, q15 |
| 2366 |
.else |
| 2367 |
vmov.i16 q9, #PREP_BIAS |
| 2368 |
vrshrn.i32 d4, q2, #6 |
| 2369 |
vrshrn.i32 d5, q3, #6 |
| 2370 |
vrshrn.i32 d6, q13, #6 |
| 2371 |
vrshrn.i32 d7, q14, #6 |
| 2372 |
vsub.i16 q2, q2, q9 // PREP_BIAS |
| 2373 |
vsub.i16 q3, q3, q9 // PREP_BIAS |
| 2374 |
.endif |
| 2375 |
subs \h, \h, #2 |
| 2376 |
vst1.16 {q2}, [\dst, :128], \d_strd |
| 2377 |
vst1.16 {q3}, [\ds2, :128], \d_strd |
| 2378 |
ble 9f |
| 2379 |
vmov q8, q10 |
| 2380 |
vmov q9, q11 |
| 2381 |
vmov q10, q12 |
| 2382 |
b 8b |
| 2383 |
9: |
| 2384 |
subs \w, \w, #8 |
| 2385 |
ble 0f |
| 2386 |
asr \s_strd, \s_strd, #1 |
| 2387 |
asr \d_strd, \d_strd, #1 |
| 2388 |
mls \src, \s_strd, \my, \src |
| 2389 |
mls \dst, \d_strd, \my, \dst |
| 2390 |
sub \src, \src, \s_strd, lsl #2 |
| 2391 |
mov \h, \my |
| 2392 |
add \src, \src, #16 |
| 2393 |
add \dst, \dst, #16 |
| 2394 |
b 164b |
| 2395 |
0: |
| 2396 |
pop {r4-r11,pc} |
| 2397 |
|
| 2398 |
880: // 8x8, 8x16, ..., 16x8, ..., 32x8, ... hv |
| 2399 |
640: |
| 2400 |
1280: |
| 2401 |
vpush {q4-q7} |
| 2402 |
vld1.8 {d0}, [\mx, :64] |
| 2403 |
vld1.8 {d2}, [\my, :64] |
| 2404 |
sub \src, \src, #6 |
| 2405 |
sub \src, \src, \s_strd |
| 2406 |
sub \src, \src, \s_strd, lsl #1 |
| 2407 |
vmovl.s8 q0, d0 |
| 2408 |
vmovl.s8 q1, d2 |
| 2409 |
mov \my, \h |
| 2410 |
|
| 2411 |
168: |
| 2412 |
add \ds2, \dst, \d_strd |
| 2413 |
add \sr2, \src, \s_strd |
| 2414 |
lsl \d_strd, \d_strd, #1 |
| 2415 |
lsl \s_strd, \s_strd, #1 |
| 2416 |
|
| 2417 |
vld1.16 {q11, q12}, [\src], \s_strd |
| 2418 |
vmull.s16 q2, d22, d0[0] |
| 2419 |
vmull.s16 q3, d23, d0[0] |
| 2420 |
vdup.32 q14, r12 // -(6-intermediate_bits) |
| 2421 |
.irpc i, 1234567 |
| 2422 |
vext.8 q10, q11, q12, #(2*\i) |
| 2423 |
.if \i < 4 |
| 2424 |
vmlal.s16 q2, d20, d0[\i] |
| 2425 |
vmlal.s16 q3, d21, d0[\i] |
| 2426 |
.else |
| 2427 |
vmlal.s16 q2, d20, d1[\i - 4] |
| 2428 |
vmlal.s16 q3, d21, d1[\i - 4] |
| 2429 |
.endif |
| 2430 |
.endr |
| 2431 |
vrshl.s32 q2, q2, q14 // -(6-intermediate_bits) |
| 2432 |
vrshl.s32 q3, q3, q14 // -(6-intermediate_bits) |
| 2433 |
vmovn.i32 d8, q2 |
| 2434 |
vmovn.i32 d9, q3 |
| 2435 |
|
| 2436 |
bl L(\type\()_8tap_filter_8) |
| 2437 |
vmov q5, q11 |
| 2438 |
vmov q6, q12 |
| 2439 |
bl L(\type\()_8tap_filter_8) |
| 2440 |
vmov q7, q11 |
| 2441 |
vmov q8, q12 |
| 2442 |
bl L(\type\()_8tap_filter_8) |
| 2443 |
vmov q9, q11 |
| 2444 |
vmov q10, q12 |
| 2445 |
|
| 2446 |
88: |
| 2447 |
bl L(\type\()_8tap_filter_8) |
| 2448 |
vmull.s16 q2, d8, d2[0] |
| 2449 |
vmull.s16 q3, d9, d2[0] |
| 2450 |
vmull.s16 q13, d10, d2[0] |
| 2451 |
vmull.s16 q14, d11, d2[0] |
| 2452 |
.ifc \type, put |
| 2453 |
vdup.32 q4, r8 // -(6+intermediate_bits) |
| 2454 |
.endif |
| 2455 |
vmlal.s16 q2, d10, d2[1] |
| 2456 |
vmlal.s16 q3, d11, d2[1] |
| 2457 |
vmlal.s16 q13, d12, d2[1] |
| 2458 |
vmlal.s16 q14, d13, d2[1] |
| 2459 |
vmlal.s16 q2, d12, d2[2] |
| 2460 |
vmlal.s16 q3, d13, d2[2] |
| 2461 |
vmlal.s16 q13, d14, d2[2] |
| 2462 |
vmlal.s16 q14, d15, d2[2] |
| 2463 |
vmlal.s16 q2, d14, d2[3] |
| 2464 |
vmlal.s16 q3, d15, d2[3] |
| 2465 |
vmlal.s16 q13, d16, d2[3] |
| 2466 |
vmlal.s16 q14, d17, d2[3] |
| 2467 |
vmlal.s16 q2, d16, d3[0] |
| 2468 |
vmlal.s16 q3, d17, d3[0] |
| 2469 |
vmlal.s16 q13, d18, d3[0] |
| 2470 |
vmlal.s16 q14, d19, d3[0] |
| 2471 |
vmlal.s16 q2, d18, d3[1] |
| 2472 |
vmlal.s16 q3, d19, d3[1] |
| 2473 |
vmlal.s16 q13, d20, d3[1] |
| 2474 |
vmlal.s16 q14, d21, d3[1] |
| 2475 |
vmlal.s16 q2, d20, d3[2] |
| 2476 |
vmlal.s16 q3, d21, d3[2] |
| 2477 |
vmlal.s16 q13, d22, d3[2] |
| 2478 |
vmlal.s16 q14, d23, d3[2] |
| 2479 |
vmlal.s16 q2, d22, d3[3] |
| 2480 |
vmlal.s16 q3, d23, d3[3] |
| 2481 |
vmlal.s16 q13, d24, d3[3] |
| 2482 |
vmlal.s16 q14, d25, d3[3] |
| 2483 |
.ifc \type, put |
| 2484 |
vrshl.s32 q2, q2, q4 // -(6+intermediate_bits) |
| 2485 |
vrshl.s32 q3, q3, q4 // -(6+intermediate_bits) |
| 2486 |
vrshl.s32 q13, q13, q4 // -(6+intermediate_bits) |
| 2487 |
vrshl.s32 q14, q14, q4 // -(6+intermediate_bits) |
| 2488 |
vqmovun.s32 d4, q2 |
| 2489 |
vqmovun.s32 d5, q3 |
| 2490 |
vqmovun.s32 d6, q13 |
| 2491 |
vqmovun.s32 d7, q14 |
| 2492 |
vmin.u16 q2, q2, q15 |
| 2493 |
vmin.u16 q3, q3, q15 |
| 2494 |
.else |
| 2495 |
vmov.i16 q5, #PREP_BIAS |
| 2496 |
vrshrn.i32 d4, q2, #6 |
| 2497 |
vrshrn.i32 d5, q3, #6 |
| 2498 |
vrshrn.i32 d6, q13, #6 |
| 2499 |
vrshrn.i32 d7, q14, #6 |
| 2500 |
vsub.i16 q2, q2, q5 // PREP_BIAS |
| 2501 |
vsub.i16 q3, q3, q5 // PREP_BIAS |
| 2502 |
.endif |
| 2503 |
subs \h, \h, #2 |
| 2504 |
vst1.16 {q2}, [\dst, :128], \d_strd |
| 2505 |
vst1.16 {q3}, [\ds2, :128], \d_strd |
| 2506 |
ble 9f |
| 2507 |
vmov q4, q6 |
| 2508 |
vmov q5, q7 |
| 2509 |
vmov q6, q8 |
| 2510 |
vmov q7, q9 |
| 2511 |
vmov q8, q10 |
| 2512 |
vmov q9, q11 |
| 2513 |
vmov q10, q12 |
| 2514 |
b 88b |
| 2515 |
9: |
| 2516 |
subs \w, \w, #8 |
| 2517 |
ble 0f |
| 2518 |
asr \s_strd, \s_strd, #1 |
| 2519 |
asr \d_strd, \d_strd, #1 |
| 2520 |
mls \src, \s_strd, \my, \src |
| 2521 |
mls \dst, \d_strd, \my, \dst |
| 2522 |
sub \src, \src, \s_strd, lsl #3 |
| 2523 |
mov \h, \my |
| 2524 |
add \src, \src, #16 |
| 2525 |
add \dst, \dst, #16 |
| 2526 |
b 168b |
| 2527 |
0: |
| 2528 |
vpop {q4-q7} |
| 2529 |
pop {r4-r11,pc} |
| 2530 |
|
| 2531 |
L(\type\()_8tap_filter_8): |
| 2532 |
vld1.16 {q13, q14}, [\sr2], \s_strd |
| 2533 |
vmull.s16 q2, d26, d0[0] |
| 2534 |
vmull.s16 q3, d27, d0[0] |
| 2535 |
.irpc i, 1234567 |
| 2536 |
vext.8 q12, q13, q14, #(2*\i) |
| 2537 |
.if \i < 4 |
| 2538 |
vmlal.s16 q2, d24, d0[\i] |
| 2539 |
vmlal.s16 q3, d25, d0[\i] |
| 2540 |
.else |
| 2541 |
vmlal.s16 q2, d24, d1[\i - 4] |
| 2542 |
vmlal.s16 q3, d25, d1[\i - 4] |
| 2543 |
.endif |
| 2544 |
.endr |
| 2545 |
vdup.32 q12, r12 // -(6-intermediate_bits) |
| 2546 |
vld1.16 {q13, q14}, [\src], \s_strd |
| 2547 |
vrshl.s32 q2, q2, q12 // -(6-intermediate_bits) |
| 2548 |
vrshl.s32 q3, q3, q12 // -(6-intermediate_bits) |
| 2549 |
vmovn.i32 d4, q2 |
| 2550 |
vmovn.i32 d5, q3 |
| 2551 |
|
| 2552 |
vmull.s16 q3, d26, d0[0] |
| 2553 |
vmull.s16 q11, d27, d0[0] |
| 2554 |
.irpc i, 1234567 |
| 2555 |
vext.8 q12, q13, q14, #(2*\i) |
| 2556 |
.if \i < 4 |
| 2557 |
vmlal.s16 q3, d24, d0[\i] |
| 2558 |
vmlal.s16 q11, d25, d0[\i] |
| 2559 |
.else |
| 2560 |
vmlal.s16 q3, d24, d1[\i - 4] |
| 2561 |
vmlal.s16 q11, d25, d1[\i - 4] |
| 2562 |
.endif |
| 2563 |
.endr |
| 2564 |
vdup.32 q13, r12 // -(6-intermediate_bits) |
| 2565 |
vrshl.s32 q3, q3, q13 // -(6-intermediate_bits) |
| 2566 |
vrshl.s32 q11, q11, q13 // -(6-intermediate_bits) |
| 2567 |
|
| 2568 |
vmovn.i32 d24, q3 |
| 2569 |
vmovn.i32 d25, q11 |
| 2570 |
vmov q11, q2 |
| 2571 |
bx lr |
| 2572 |
endfunc |
| 2573 |
|
| 2574 |
function \type\()_bilin_16bpc_neon, export=1 |
| 2575 |
push {r4-r11,lr} |
| 2576 |
ldrd r4, r5, [sp, #36] |
| 2577 |
ldrd r6, r7, [sp, #44] |
| 2578 |
.ifc \bdmax, r8 |
| 2579 |
ldr r8, [sp, #52] |
| 2580 |
.endif |
| 2581 |
vdup.16 q1, \mx |
| 2582 |
vdup.16 q3, \my |
| 2583 |
rsb r9, \mx, #16 |
| 2584 |
rsb r10, \my, #16 |
| 2585 |
vdup.16 q0, r9 |
| 2586 |
vdup.16 q2, r10 |
| 2587 |
.ifc \type, prep |
| 2588 |
lsl \d_strd, \w, #1 |
| 2589 |
.endif |
| 2590 |
clz \bdmax, \bdmax // bitdepth_max |
| 2591 |
clz r9, \w |
| 2592 |
sub \bdmax, \bdmax, #18 // intermediate_bits = clz(bitdepth_max) - 18 |
| 2593 |
cmp \mx, #0 |
| 2594 |
sub r9, r9, #24 |
| 2595 |
rsb r11, \bdmax, #4 // 4 - intermediate_bits |
| 2596 |
add r12, \bdmax, #4 // 4 + intermediate_bits |
| 2597 |
bne L(\type\()_bilin_h) |
| 2598 |
cmp \my, #0 |
| 2599 |
bne L(\type\()_bilin_v) |
| 2600 |
b \type\()_neon |
| 2601 |
|
| 2602 |
L(\type\()_bilin_h): |
| 2603 |
cmp \my, #0 |
| 2604 |
bne L(\type\()_bilin_hv) |
| 2605 |
|
| 2606 |
adr r10, L(\type\()_bilin_h_tbl) |
| 2607 |
vdup.16 q15, r11 // 4 - intermediate_bits |
| 2608 |
ldr r9, [r10, r9, lsl #2] |
| 2609 |
vneg.s16 q15, q15 // -(4-intermediate_bits) |
| 2610 |
.ifc \type, put |
| 2611 |
vdup.16 q14, \bdmax // intermediate_bits |
| 2612 |
.else |
| 2613 |
vmov.i16 q14, #PREP_BIAS |
| 2614 |
.endif |
| 2615 |
add r10, r10, r9 |
| 2616 |
.ifc \type, put |
| 2617 |
vneg.s16 q14, q14 // -intermediate_bits |
| 2618 |
.endif |
| 2619 |
bx r10 |
| 2620 |
|
| 2621 |
.align 2 |
| 2622 |
L(\type\()_bilin_h_tbl): |
| 2623 |
.word 1280f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2624 |
.word 640f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2625 |
.word 320f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2626 |
.word 160f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2627 |
.word 80f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2628 |
.word 40f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2629 |
.word 20f - L(\type\()_bilin_h_tbl) + CONFIG_THUMB |
| 2630 |
|
| 2631 |
20: // 2xN h |
| 2632 |
.ifc \type, put |
| 2633 |
add \ds2, \dst, \d_strd |
| 2634 |
add \sr2, \src, \s_strd |
| 2635 |
lsl \d_strd, \d_strd, #1 |
| 2636 |
lsl \s_strd, \s_strd, #1 |
| 2637 |
2: |
| 2638 |
vld1.16 {d16}, [\src], \s_strd |
| 2639 |
vld1.16 {d18}, [\sr2], \s_strd |
| 2640 |
vext.8 d17, d16, d16, #2 |
| 2641 |
vext.8 d19, d18, d18, #2 |
| 2642 |
vtrn.32 d16, d18 |
| 2643 |
vtrn.32 d17, d19 |
| 2644 |
subs \h, \h, #2 |
| 2645 |
vmul.i16 d16, d16, d0 |
| 2646 |
vmla.i16 d16, d17, d2 |
| 2647 |
vrshl.u16 d16, d16, d30 |
| 2648 |
vrshl.u16 d16, d16, d28 |
| 2649 |
vst1.32 {d16[0]}, [\dst, :32], \d_strd |
| 2650 |
vst1.32 {d16[1]}, [\ds2, :32], \d_strd |
| 2651 |
bgt 2b |
| 2652 |
pop {r4-r11,pc} |
| 2653 |
.endif |
| 2654 |
|
| 2655 |
40: // 4xN h |
| 2656 |
add \ds2, \dst, \d_strd |
| 2657 |
add \sr2, \src, \s_strd |
| 2658 |
lsl \d_strd, \d_strd, #1 |
| 2659 |
lsl \s_strd, \s_strd, #1 |
| 2660 |
4: |
| 2661 |
vld1.16 {q8}, [\src], \s_strd |
| 2662 |
vld1.16 {q10}, [\sr2], \s_strd |
| 2663 |
vext.8 q9, q8, q8, #2 |
| 2664 |
vext.8 q11, q10, q10, #2 |
| 2665 |
vmov d17, d20 |
| 2666 |
vmov d19, d22 |
| 2667 |
subs \h, \h, #2 |
| 2668 |
vmul.i16 q8, q8, q0 |
| 2669 |
vmla.i16 q8, q9, q1 |
| 2670 |
vrshl.u16 q8, q8, q15 |
| 2671 |
.ifc \type, put |
| 2672 |
vrshl.u16 q8, q8, q14 |
| 2673 |
.else |
| 2674 |
vsub.i16 q8, q8, q14 |
| 2675 |
.endif |
| 2676 |
vst1.16 {d16}, [\dst, :64], \d_strd |
| 2677 |
vst1.16 {d17}, [\ds2, :64], \d_strd |
| 2678 |
bgt 4b |
| 2679 |
pop {r4-r11,pc} |
| 2680 |
|
| 2681 |
80: // 8xN h |
| 2682 |
add \ds2, \dst, \d_strd |
| 2683 |
add \sr2, \src, \s_strd |
| 2684 |
lsl \d_strd, \d_strd, #1 |
| 2685 |
lsl \s_strd, \s_strd, #1 |
| 2686 |
8: |
| 2687 |
vld1.16 {d16, d17, d18}, [\src], \s_strd |
| 2688 |
vld1.16 {d20, d21, d22}, [\sr2], \s_strd |
| 2689 |
vext.8 q9, q8, q9, #2 |
| 2690 |
vext.8 q11, q10, q11, #2 |
| 2691 |
subs \h, \h, #2 |
| 2692 |
vmul.i16 q8, q8, q0 |
| 2693 |
vmla.i16 q8, q9, q1 |
| 2694 |
vmul.i16 q10, q10, q0 |
| 2695 |
vmla.i16 q10, q11, q1 |
| 2696 |
vrshl.u16 q8, q8, q15 |
| 2697 |
vrshl.u16 q10, q10, q15 |
| 2698 |
.ifc \type, put |
| 2699 |
vrshl.u16 q8, q8, q14 |
| 2700 |
vrshl.u16 q10, q10, q14 |
| 2701 |
.else |
| 2702 |
vsub.i16 q8, q8, q14 |
| 2703 |
vsub.i16 q10, q10, q14 |
| 2704 |
.endif |
| 2705 |
vst1.16 {q8}, [\dst, :128], \d_strd |
| 2706 |
vst1.16 {q10}, [\ds2, :128], \d_strd |
| 2707 |
bgt 8b |
| 2708 |
pop {r4-r11,pc} |
| 2709 |
160: |
| 2710 |
320: |
| 2711 |
640: |
| 2712 |
1280: // 16xN, 32xN, ... h |
| 2713 |
vpush {q4-q7} |
| 2714 |
add \ds2, \dst, \d_strd |
| 2715 |
add \sr2, \src, \s_strd |
| 2716 |
lsl \s_strd, \s_strd, #1 |
| 2717 |
|
| 2718 |
sub \s_strd, \s_strd, \w, lsl #1 |
| 2719 |
sub \s_strd, \s_strd, #16 |
| 2720 |
.ifc \type, put |
| 2721 |
lsl \d_strd, \d_strd, #1 |
| 2722 |
sub \d_strd, \d_strd, \w, lsl #1 |
| 2723 |
.endif |
| 2724 |
161: |
| 2725 |
vld1.16 {q4}, [\src]! |
| 2726 |
vld1.16 {q9}, [\sr2]! |
| 2727 |
mov \mx, \w |
| 2728 |
|
| 2729 |
16: |
| 2730 |
vld1.16 {q5, q6}, [\src]! |
| 2731 |
vld1.16 {q10, q11}, [\sr2]! |
| 2732 |
vext.8 q7, q4, q5, #2 |
| 2733 |
vext.8 q8, q5, q6, #2 |
| 2734 |
vext.8 q12, q9, q10, #2 |
| 2735 |
vext.8 q13, q10, q11, #2 |
| 2736 |
vmul.i16 q4, q4, q0 |
| 2737 |
vmla.i16 q4, q7, q1 |
| 2738 |
vmul.i16 q5, q5, q0 |
| 2739 |
vmla.i16 q5, q8, q1 |
| 2740 |
vmul.i16 q9, q9, q0 |
| 2741 |
vmla.i16 q9, q12, q1 |
| 2742 |
vmul.i16 q10, q10, q0 |
| 2743 |
vmla.i16 q10, q13, q1 |
| 2744 |
vrshl.u16 q4, q4, q15 |
| 2745 |
vrshl.u16 q5, q5, q15 |
| 2746 |
vrshl.u16 q9, q9, q15 |
| 2747 |
vrshl.u16 q10, q10, q15 |
| 2748 |
subs \mx, \mx, #16 |
| 2749 |
.ifc \type, put |
| 2750 |
vrshl.u16 q4, q4, q14 |
| 2751 |
vrshl.u16 q5, q5, q14 |
| 2752 |
vrshl.u16 q9, q9, q14 |
| 2753 |
vrshl.u16 q10, q10, q14 |
| 2754 |
.else |
| 2755 |
vsub.i16 q4, q4, q14 |
| 2756 |
vsub.i16 q5, q5, q14 |
| 2757 |
vsub.i16 q9, q9, q14 |
| 2758 |
vsub.i16 q10, q10, q14 |
| 2759 |
.endif |
| 2760 |
vst1.16 {q4, q5}, [\dst, :128]! |
| 2761 |
vst1.16 {q9, q10}, [\ds2, :128]! |
| 2762 |
ble 9f |
| 2763 |
|
| 2764 |
vmov q4, q6 |
| 2765 |
vmov q9, q11 |
| 2766 |
b 16b |
| 2767 |
|
| 2768 |
9: |
| 2769 |
add \dst, \dst, \d_strd |
| 2770 |
add \ds2, \ds2, \d_strd |
| 2771 |
add \src, \src, \s_strd |
| 2772 |
add \sr2, \sr2, \s_strd |
| 2773 |
|
| 2774 |
subs \h, \h, #2 |
| 2775 |
bgt 161b |
| 2776 |
vpop {q4-q7} |
| 2777 |
pop {r4-r11,pc} |
| 2778 |
|
| 2779 |
|
| 2780 |
L(\type\()_bilin_v): |
| 2781 |
cmp \h, #4 |
| 2782 |
adr r10, L(\type\()_bilin_v_tbl) |
| 2783 |
.ifc \type, prep |
| 2784 |
vdup.16 q15, r11 // 4 - intermediate_bits |
| 2785 |
.endif |
| 2786 |
ldr r9, [r10, r9, lsl #2] |
| 2787 |
.ifc \type, prep |
| 2788 |
vmov.i16 q14, #PREP_BIAS |
| 2789 |
vneg.s16 q15, q15 // -(4-intermediate_bits) |
| 2790 |
.endif |
| 2791 |
add r10, r10, r9 |
| 2792 |
bx r10 |
| 2793 |
|
| 2794 |
.align 2 |
| 2795 |
L(\type\()_bilin_v_tbl): |
| 2796 |
.word 1280f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2797 |
.word 640f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2798 |
.word 320f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2799 |
.word 160f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2800 |
.word 80f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2801 |
.word 40f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2802 |
.word 20f - L(\type\()_bilin_v_tbl) + CONFIG_THUMB |
| 2803 |
|
| 2804 |
20: // 2xN v |
| 2805 |
.ifc \type, put |
| 2806 |
cmp \h, #2 |
| 2807 |
add \ds2, \dst, \d_strd |
| 2808 |
add \sr2, \src, \s_strd |
| 2809 |
lsl \s_strd, \s_strd, #1 |
| 2810 |
lsl \d_strd, \d_strd, #1 |
| 2811 |
|
| 2812 |
// 2x2 v |
| 2813 |
vld1.32 {d16[]}, [\src], \s_strd |
| 2814 |
bgt 24f |
| 2815 |
22: |
| 2816 |
vld1.32 {d17[]}, [\sr2], \s_strd |
| 2817 |
vld1.32 {d18[]}, [\src], \s_strd |
| 2818 |
vext.8 d16, d16, d17, #4 |
| 2819 |
vext.8 d17, d17, d18, #4 |
| 2820 |
vmul.i16 d16, d16, d4 |
| 2821 |
vmla.i16 d16, d17, d6 |
| 2822 |
vrshr.u16 d16, d16, #4 |
| 2823 |
vst1.32 {d16[0]}, [\dst, :32] |
| 2824 |
vst1.32 {d16[1]}, [\ds2, :32] |
| 2825 |
pop {r4-r11,pc} |
| 2826 |
24: // 2x4, 2x6, 2x8, ... v |
| 2827 |
vld1.32 {d17[]}, [\sr2], \s_strd |
| 2828 |
vld1.32 {d18[]}, [\src], \s_strd |
| 2829 |
vld1.32 {d19[]}, [\sr2], \s_strd |
| 2830 |
vld1.32 {d20[]}, [\src], \s_strd |
| 2831 |
subs \h, \h, #4 |
| 2832 |
vext.8 d16, d16, d17, #4 |
| 2833 |
vext.8 d17, d17, d18, #4 |
| 2834 |
vext.8 d18, d18, d19, #4 |
| 2835 |
vext.8 d19, d19, d20, #4 |
| 2836 |
vswp d17, d18 |
| 2837 |
vmul.i16 q8, q8, q2 |
| 2838 |
vmla.i16 q8, q9, q3 |
| 2839 |
cmp \h, #2 |
| 2840 |
vrshr.u16 q8, q8, #4 |
| 2841 |
vst1.32 {d16[0]}, [\dst, :32], \d_strd |
| 2842 |
vst1.32 {d16[1]}, [\ds2, :32], \d_strd |
| 2843 |
vst1.32 {d17[0]}, [\dst, :32], \d_strd |
| 2844 |
vst1.32 {d17[1]}, [\ds2, :32], \d_strd |
| 2845 |
blt 0f |
| 2846 |
vmov d16, d20 |
| 2847 |
beq 22b |
| 2848 |
b 24b |
| 2849 |
0: |
| 2850 |
pop {r4-r11,pc} |
| 2851 |
.endif |
| 2852 |
|
| 2853 |
40: // 4xN v |
| 2854 |
add \ds2, \dst, \d_strd |
| 2855 |
add \sr2, \src, \s_strd |
| 2856 |
lsl \s_strd, \s_strd, #1 |
| 2857 |
lsl \d_strd, \d_strd, #1 |
| 2858 |
vld1.16 {d16}, [\src], \s_strd |
| 2859 |
4: |
| 2860 |
vld1.16 {d17}, [\sr2], \s_strd |
| 2861 |
vld1.16 {d19}, [\src], \s_strd |
| 2862 |
vmov d18, d17 |
| 2863 |
vmul.i16 q8, q8, q2 |
| 2864 |
vmla.i16 q8, q9, q3 |
| 2865 |
subs \h, \h, #2 |
| 2866 |
.ifc \type, put |
| 2867 |
vrshr.u16 q8, q8, #4 |
| 2868 |
.else |
| 2869 |
vrshl.u16 q8, q8, q15 |
| 2870 |
vsub.i16 q8, q8, q14 |
| 2871 |
.endif |
| 2872 |
vst1.16 {d16}, [\dst, :64], \d_strd |
| 2873 |
vst1.16 {d17}, [\ds2, :64], \d_strd |
| 2874 |
ble 0f |
| 2875 |
vmov d16, d19 |
| 2876 |
b 4b |
| 2877 |
0: |
| 2878 |
pop {r4-r11,pc} |
| 2879 |
|
| 2880 |
80: // 8xN v |
| 2881 |
add \ds2, \dst, \d_strd |
| 2882 |
add \sr2, \src, \s_strd |
| 2883 |
lsl \s_strd, \s_strd, #1 |
| 2884 |
lsl \d_strd, \d_strd, #1 |
| 2885 |
vld1.16 {q8}, [\src], \s_strd |
| 2886 |
8: |
| 2887 |
vld1.16 {q9}, [\sr2], \s_strd |
| 2888 |
vld1.16 {q10}, [\src], \s_strd |
| 2889 |
vmul.i16 q8, q8, q2 |
| 2890 |
vmla.i16 q8, q9, q3 |
| 2891 |
vmul.i16 q9, q9, q2 |
| 2892 |
vmla.i16 q9, q10, q3 |
| 2893 |
subs \h, \h, #2 |
| 2894 |
.ifc \type, put |
| 2895 |
vrshr.u16 q8, q8, #4 |
| 2896 |
vrshr.u16 q9, q9, #4 |
| 2897 |
.else |
| 2898 |
vrshl.u16 q8, q8, q15 |
| 2899 |
vrshl.u16 q9, q9, q15 |
| 2900 |
vsub.i16 q8, q8, q14 |
| 2901 |
vsub.i16 q9, q9, q14 |
| 2902 |
.endif |
| 2903 |
vst1.16 {q8}, [\dst, :128], \d_strd |
| 2904 |
vst1.16 {q9}, [\ds2, :128], \d_strd |
| 2905 |
ble 0f |
| 2906 |
vmov q8, q10 |
| 2907 |
b 8b |
| 2908 |
0: |
| 2909 |
pop {r4-r11,pc} |
| 2910 |
|
| 2911 |
160: // 16xN, 32xN, ... |
| 2912 |
320: |
| 2913 |
640: |
| 2914 |
1280: |
| 2915 |
mov \my, \h |
| 2916 |
1: |
| 2917 |
add \ds2, \dst, \d_strd |
| 2918 |
add \sr2, \src, \s_strd |
| 2919 |
lsl \s_strd, \s_strd, #1 |
| 2920 |
lsl \d_strd, \d_strd, #1 |
| 2921 |
|
| 2922 |
vld1.16 {q8, q9}, [\src], \s_strd |
| 2923 |
2: |
| 2924 |
vld1.16 {q10, q11}, [\sr2], \s_strd |
| 2925 |
vld1.16 {q12, q13}, [\src], \s_strd |
| 2926 |
vmul.i16 q8, q8, q2 |
| 2927 |
vmla.i16 q8, q10, q3 |
| 2928 |
vmul.i16 q9, q9, q2 |
| 2929 |
vmla.i16 q9, q11, q3 |
| 2930 |
vmul.i16 q10, q10, q2 |
| 2931 |
vmla.i16 q10, q12, q3 |
| 2932 |
vmul.i16 q11, q11, q2 |
| 2933 |
vmla.i16 q11, q13, q3 |
| 2934 |
subs \h, \h, #2 |
| 2935 |
.ifc \type, put |
| 2936 |
vrshr.u16 q8, q8, #4 |
| 2937 |
vrshr.u16 q9, q9, #4 |
| 2938 |
vrshr.u16 q10, q10, #4 |
| 2939 |
vrshr.u16 q11, q11, #4 |
| 2940 |
.else |
| 2941 |
vrshl.u16 q8, q8, q15 |
| 2942 |
vrshl.u16 q9, q9, q15 |
| 2943 |
vrshl.u16 q10, q10, q15 |
| 2944 |
vrshl.u16 q11, q11, q15 |
| 2945 |
vsub.i16 q8, q8, q14 |
| 2946 |
vsub.i16 q9, q9, q14 |
| 2947 |
vsub.i16 q10, q10, q14 |
| 2948 |
vsub.i16 q11, q11, q14 |
| 2949 |
.endif |
| 2950 |
vst1.16 {q8, q9}, [\dst, :128], \d_strd |
| 2951 |
vst1.16 {q10, q11}, [\ds2, :128], \d_strd |
| 2952 |
ble 9f |
| 2953 |
vmov q8, q12 |
| 2954 |
vmov q9, q13 |
| 2955 |
b 2b |
| 2956 |
9: |
| 2957 |
subs \w, \w, #16 |
| 2958 |
ble 0f |
| 2959 |
asr \s_strd, \s_strd, #1 |
| 2960 |
asr \d_strd, \d_strd, #1 |
| 2961 |
mls \src, \s_strd, \my, \src |
| 2962 |
mls \dst, \d_strd, \my, \dst |
| 2963 |
sub \src, \src, \s_strd, lsl #1 |
| 2964 |
mov \h, \my |
| 2965 |
add \src, \src, #32 |
| 2966 |
add \dst, \dst, #32 |
| 2967 |
b 1b |
| 2968 |
0: |
| 2969 |
pop {r4-r11,pc} |
| 2970 |
|
| 2971 |
L(\type\()_bilin_hv): |
| 2972 |
adr r10, L(\type\()_bilin_hv_tbl) |
| 2973 |
vdup.16 q15, r11 // 4 - intermediate_bits |
| 2974 |
ldr r9, [r10, r9, lsl #2] |
| 2975 |
vneg.s16 q15, q15 // -(4-intermediate_bits) |
| 2976 |
.ifc \type, put |
| 2977 |
vdup.32 q14, r12 // 4 + intermediate_bits |
| 2978 |
.else |
| 2979 |
vmov.i16 q14, #PREP_BIAS |
| 2980 |
.endif |
| 2981 |
add r10, r10, r9 |
| 2982 |
.ifc \type, put |
| 2983 |
vneg.s32 q14, q14 // -(4+intermediate_bits) |
| 2984 |
.endif |
| 2985 |
bx r10 |
| 2986 |
|
| 2987 |
.align 2 |
| 2988 |
L(\type\()_bilin_hv_tbl): |
| 2989 |
.word 1280f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2990 |
.word 640f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2991 |
.word 320f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2992 |
.word 160f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2993 |
.word 80f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2994 |
.word 40f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2995 |
.word 20f - L(\type\()_bilin_hv_tbl) + CONFIG_THUMB |
| 2996 |
|
| 2997 |
20: // 2xN hv |
| 2998 |
.ifc \type, put |
| 2999 |
add \sr2, \src, \s_strd |
| 3000 |
add \ds2, \dst, \d_strd |
| 3001 |
lsl \s_strd, \s_strd, #1 |
| 3002 |
lsl \d_strd, \d_strd, #1 |
| 3003 |
|
| 3004 |
vld1.16 {d20}, [\src], \s_strd |
| 3005 |
vext.8 d21, d20, d20, #2 |
| 3006 |
vmul.i16 d16, d20, d0 |
| 3007 |
vmla.i16 d16, d21, d2 |
| 3008 |
vrshl.u16 d16, d16, d30 |
| 3009 |
vext.8 d16, d16, d16, #4 |
| 3010 |
|
| 3011 |
2: |
| 3012 |
vld1.16 {d20}, [\sr2], \s_strd |
| 3013 |
vld1.16 {d22}, [\src], \s_strd |
| 3014 |
vext.8 d21, d20, d20, #2 |
| 3015 |
vext.8 d23, d22, d22, #2 |
| 3016 |
vtrn.32 d20, d22 |
| 3017 |
vtrn.32 d21, d23 |
| 3018 |
vmul.i16 d18, d20, d0 |
| 3019 |
vmla.i16 d18, d21, d2 |
| 3020 |
vrshl.u16 d18, d18, d30 |
| 3021 |
|
| 3022 |
vext.8 d16, d16, d18, #4 |
| 3023 |
|
| 3024 |
vmull.u16 q8, d16, d4 |
| 3025 |
vmlal.u16 q8, d18, d6 |
| 3026 |
vrshl.u32 q8, q8, q14 |
| 3027 |
vmovn.i32 d16, q8 |
| 3028 |
subs \h, \h, #2 |
| 3029 |
vst1.32 {d16[0]}, [\dst, :32], \d_strd |
| 3030 |
vst1.32 {d16[1]}, [\ds2, :32], \d_strd |
| 3031 |
ble 0f |
| 3032 |
vmov d16, d18 |
| 3033 |
b 2b |
| 3034 |
0: |
| 3035 |
pop {r4-r11,pc} |
| 3036 |
.endif |
| 3037 |
|
| 3038 |
40: // 4xN hv |
| 3039 |
add \sr2, \src, \s_strd |
| 3040 |
add \ds2, \dst, \d_strd |
| 3041 |
lsl \s_strd, \s_strd, #1 |
| 3042 |
lsl \d_strd, \d_strd, #1 |
| 3043 |
|
| 3044 |
vld1.16 {q10}, [\src], \s_strd |
| 3045 |
vext.8 d21, d20, d21, #2 |
| 3046 |
vmul.i16 d16, d20, d0 |
| 3047 |
vmla.i16 d16, d21, d2 |
| 3048 |
vrshl.u16 d16, d16, d30 |
| 3049 |
|
| 3050 |
4: |
| 3051 |
vld1.16 {q10}, [\sr2], \s_strd |
| 3052 |
vld1.16 {q11}, [\src], \s_strd |
| 3053 |
vext.8 d21, d20, d21, #2 |
| 3054 |
vext.8 d23, d22, d23, #2 |
| 3055 |
vswp d21, d22 |
| 3056 |
vmul.i16 q9, q10, q0 |
| 3057 |
vmla.i16 q9, q11, q1 |
| 3058 |
vrshl.u16 q9, q9, q15 |
| 3059 |
|
| 3060 |
vmull.u16 q10, d16, d4 |
| 3061 |
vmlal.u16 q10, d18, d6 |
| 3062 |
vmull.u16 q11, d18, d4 |
| 3063 |
vmlal.u16 q11, d19, d6 |
| 3064 |
.ifc \type, put |
| 3065 |
vrshl.u32 q10, q10, q14 |
| 3066 |
vrshl.u32 q11, q11, q14 |
| 3067 |
vmovn.i32 d20, q10 |
| 3068 |
vmovn.i32 d21, q11 |
| 3069 |
.else |
| 3070 |
vrshrn.i32 d20, q10, #4 |
| 3071 |
vrshrn.i32 d21, q11, #4 |
| 3072 |
vsub.i16 q10, q10, q14 |
| 3073 |
.endif |
| 3074 |
subs \h, \h, #2 |
| 3075 |
vst1.16 {d20}, [\dst, :64], \d_strd |
| 3076 |
vst1.16 {d21}, [\ds2, :64], \d_strd |
| 3077 |
ble 0f |
| 3078 |
vmov d16, d19 |
| 3079 |
b 4b |
| 3080 |
0: |
| 3081 |
pop {r4-r11,pc} |
| 3082 |
|
| 3083 |
80: // 8xN, 16xN, ... hv |
| 3084 |
160: |
| 3085 |
320: |
| 3086 |
640: |
| 3087 |
1280: |
| 3088 |
mov \my, \h |
| 3089 |
|
| 3090 |
1: |
| 3091 |
add \sr2, \src, \s_strd |
| 3092 |
add \ds2, \dst, \d_strd |
| 3093 |
lsl \s_strd, \s_strd, #1 |
| 3094 |
lsl \d_strd, \d_strd, #1 |
| 3095 |
|
| 3096 |
vld1.16 {d20, d21, d22}, [\src], \s_strd |
| 3097 |
vext.8 q11, q10, q11, #2 |
| 3098 |
vmul.i16 q8, q10, q0 |
| 3099 |
vmla.i16 q8, q11, q1 |
| 3100 |
vrshl.u16 q8, q8, q15 |
| 3101 |
|
| 3102 |
2: |
| 3103 |
vld1.16 {d20, d21, d22}, [\sr2], \s_strd |
| 3104 |
vld1.16 {d24, d25, d26}, [\src], \s_strd |
| 3105 |
vext.8 q11, q10, q11, #2 |
| 3106 |
vext.8 q13, q12, q13, #2 |
| 3107 |
vmul.i16 q9, q10, q0 |
| 3108 |
vmla.i16 q9, q11, q1 |
| 3109 |
vmul.i16 q10, q12, q0 |
| 3110 |
vmla.i16 q10, q13, q1 |
| 3111 |
vrshl.u16 q9, q9, q15 |
| 3112 |
vrshl.u16 q10, q10, q15 |
| 3113 |
|
| 3114 |
vmull.u16 q11, d16, d4 |
| 3115 |
vmlal.u16 q11, d18, d6 |
| 3116 |
vmull.u16 q12, d17, d4 |
| 3117 |
vmlal.u16 q12, d19, d6 |
| 3118 |
vmull.u16 q8, d18, d4 |
| 3119 |
vmlal.u16 q8, d20, d6 |
| 3120 |
vmull.u16 q9, d19, d4 |
| 3121 |
vmlal.u16 q9, d21, d6 |
| 3122 |
.ifc \type, put |
| 3123 |
vrshl.u32 q11, q11, q14 |
| 3124 |
vrshl.u32 q12, q12, q14 |
| 3125 |
vrshl.u32 q8, q8, q14 |
| 3126 |
vrshl.u32 q9, q9, q14 |
| 3127 |
vmovn.i32 d22, q11 |
| 3128 |
vmovn.i32 d23, q12 |
| 3129 |
vmovn.i32 d16, q8 |
| 3130 |
vmovn.i32 d17, q9 |
| 3131 |
.else |
| 3132 |
vrshrn.i32 d22, q11, #4 |
| 3133 |
vrshrn.i32 d23, q12, #4 |
| 3134 |
vrshrn.i32 d16, q8, #4 |
| 3135 |
vrshrn.i32 d17, q9, #4 |
| 3136 |
vsub.i16 q11, q11, q14 |
| 3137 |
vsub.i16 q8, q8, q14 |
| 3138 |
.endif |
| 3139 |
subs \h, \h, #2 |
| 3140 |
vst1.16 {q11}, [\dst, :128], \d_strd |
| 3141 |
vst1.16 {q8}, [\ds2, :128], \d_strd |
| 3142 |
ble 9f |
| 3143 |
vmov q8, q10 |
| 3144 |
b 2b |
| 3145 |
9: |
| 3146 |
subs \w, \w, #8 |
| 3147 |
ble 0f |
| 3148 |
asr \s_strd, \s_strd, #1 |
| 3149 |
asr \d_strd, \d_strd, #1 |
| 3150 |
mls \src, \s_strd, \my, \src |
| 3151 |
mls \dst, \d_strd, \my, \dst |
| 3152 |
sub \src, \src, \s_strd, lsl #1 |
| 3153 |
mov \h, \my |
| 3154 |
add \src, \src, #16 |
| 3155 |
add \dst, \dst, #16 |
| 3156 |
b 1b |
| 3157 |
0: |
| 3158 |
pop {r4-r11,pc} |
| 3159 |
endfunc |
| 3160 |
.endm |
| 3161 |
|
| 3162 |
filter_fn put, r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10 |
| 3163 |
filter_fn prep, r0, r8, r1, r2, r3, r4, r5, r6, r7, r9, r10 |
| 3164 |
|
| 3165 |
.macro load_filter_ptr src |
| 3166 |
asr r12, \src, #10 |
| 3167 |
add r12, r11, r12, lsl #3 |
| 3168 |
.endm |
| 3169 |
|
| 3170 |
.macro load_filter_coef dst, src, inc |
| 3171 |
add \src, \src, \inc |
| 3172 |
vld1.8 {\dst}, [r12, :64] |
| 3173 |
.endm |
| 3174 |
|
| 3175 |
.macro load_filter_row dst, src, inc |
| 3176 |
load_filter_ptr \src |
| 3177 |
load_filter_coef \dst, \src, \inc |
| 3178 |
.endm |
| 3179 |
|
| 3180 |
function warp_filter_horz_neon |
| 3181 |
load_filter_ptr r5 // filter 0 |
| 3182 |
vld1.16 {q6,q7}, [r2], r3 |
| 3183 |
|
| 3184 |
load_filter_coef d0, r5, r7 // filter 0 |
| 3185 |
load_filter_row d2, r5, r7 // filter 1 |
| 3186 |
vmovl.s8 q0, d0 // filter 0 |
| 3187 |
vext.8 q3, q6, q7, #2*1 // filter 1 pixels |
| 3188 |
vmovl.s8 q1, d2 // filter 1 |
| 3189 |
|
| 3190 |
vmull.s16 q4, d12, d0 // filter 0 output (0-3) |
| 3191 |
vmull.s16 q5, d13, d1 // filter 0 output (4-7) |
| 3192 |
|
| 3193 |
load_filter_ptr r5 // filter 2 |
| 3194 |
|
| 3195 |
vmull.s16 q2, d6, d2 // filter 1 output (0-3) |
| 3196 |
vmull.s16 q3, d7, d3 // filter 1 output (4-7) |
| 3197 |
|
| 3198 |
load_filter_coef d0, r5, r7 // filter 2 |
| 3199 |
|
| 3200 |
vpadd.i32 d8, d8, d9 // half pixel 0 (2x32) |
| 3201 |
vpadd.i32 d9, d10, d11 // half pixel 0 (2x32) |
| 3202 |
|
| 3203 |
load_filter_ptr r5 // filter 3 |
| 3204 |
|
| 3205 |
vpadd.i32 d4, d4, d5 // half pixel 1 (2x32) |
| 3206 |
vpadd.i32 d5, d6, d7 // half pixel 1 (2x32) |
| 3207 |
|
| 3208 |
vmovl.s8 q0, d0 // filter 2 |
| 3209 |
vext.8 q3, q6, q7, #2*2 // filter 2 pixels |
| 3210 |
|
| 3211 |
vpadd.i32 d8, d8, d9 // pixel 0 (2x32) |
| 3212 |
vpadd.i32 d9, d4, d5 // pixel 1 (2x32) |
| 3213 |
|
| 3214 |
load_filter_coef d2, r5, r7 // filter 3 |
| 3215 |
|
| 3216 |
vmull.s16 q2, d6, d0 // filter 2 output (0-3) |
| 3217 |
vmull.s16 q3, d7, d1 // filter 2 output (4-7) |
| 3218 |
|
| 3219 |
load_filter_ptr r5 // filter 4 |
| 3220 |
|
| 3221 |
vpadd.i32 d8, d8, d9 // pixel 0,1 |
| 3222 |
|
| 3223 |
vpadd.i32 d9, d4, d5 // half pixel 2 (2x32) |
| 3224 |
vpadd.i32 d10, d6, d7 // half pixel 2 (2x32) |
| 3225 |
|
| 3226 |
vmovl.s8 q1, d2 // filter 3 |
| 3227 |
vext.8 q3, q6, q7, #2*3 // filter 3 pixels |
| 3228 |
|
| 3229 |
load_filter_coef d0, r5, r7 // filter 4 |
| 3230 |
|
| 3231 |
vpadd.i32 d9, d9, d10 // pixel 2 (2x32) |
| 3232 |
|
| 3233 |
vmull.s16 q2, d6, d2 // filter 3 output (0-3) |
| 3234 |
vmull.s16 q3, d7, d3 // filter 3 output (4-7) |
| 3235 |
|
| 3236 |
vmovl.s8 q0, d0 // filter 4 |
| 3237 |
load_filter_ptr r5 // filter 5 |
| 3238 |
|
| 3239 |
vpadd.i32 d10, d4, d5 // half pixel 3 (2x32) |
| 3240 |
vpadd.i32 d11, d6, d7 // half pixel 3 (2x32) |
| 3241 |
|
| 3242 |
vext.8 q3, q6, q7, #2*4 // filter 4 pixels |
| 3243 |
load_filter_coef d2, r5, r7 // filter 5 |
| 3244 |
|
| 3245 |
vpadd.i32 d10, d10, d11 // pixel 3 (2x32) |
| 3246 |
|
| 3247 |
vpadd.i32 d9, d9, d10 // pixel 2,3 |
| 3248 |
|
| 3249 |
vmull.s16 q2, d6, d0 // filter 4 output (0-3) |
| 3250 |
vmull.s16 q3, d7, d1 // filter 4 output (4-7) |
| 3251 |
|
| 3252 |
vmovl.s8 q1, d2 // filter 5 |
| 3253 |
load_filter_ptr r5 // filter 6 |
| 3254 |
|
| 3255 |
vpadd.i32 d10, d4, d5 // half pixel 4 (2x32) |
| 3256 |
vpadd.i32 d11, d6, d7 // half pixel 4 (2x32) |
| 3257 |
|
| 3258 |
vext.8 q3, q6, q7, #2*5 // filter 5 pixels |
| 3259 |
load_filter_coef d0, r5, r7 // filter 6 |
| 3260 |
|
| 3261 |
vpadd.i32 d10, d10, d11 // pixel 4 (2x32) |
| 3262 |
|
| 3263 |
vmull.s16 q2, d6, d2 // filter 5 output (0-3) |
| 3264 |
vmull.s16 q3, d7, d3 // filter 5 output (4-7) |
| 3265 |
|
| 3266 |
vmovl.s8 q0, d0 // filter 6 |
| 3267 |
load_filter_ptr r5 // filter 7 |
| 3268 |
|
| 3269 |
vpadd.i32 d4, d4, d5 // half pixel 5 (2x32) |
| 3270 |
vpadd.i32 d5, d6, d7 // half pixel 5 (2x32) |
| 3271 |
|
| 3272 |
vext.8 q3, q6, q7, #2*6 // filter 6 pixels |
| 3273 |
load_filter_coef d2, r5, r7 // filter 7 |
| 3274 |
|
| 3275 |
vpadd.i32 d11, d4, d5 // pixel 5 (2x32) |
| 3276 |
|
| 3277 |
vmull.s16 q2, d6, d0 // filter 6 output (0-3) |
| 3278 |
vmull.s16 q3, d7, d1 // filter 6 output (4-7) |
| 3279 |
|
| 3280 |
vmovl.s8 q1, d2 // filter 7 |
| 3281 |
|
| 3282 |
vpadd.i32 d10, d10, d11 // pixel 4,5 |
| 3283 |
|
| 3284 |
vpadd.i32 d4, d4, d5 // half pixel 6 (2x32) |
| 3285 |
vpadd.i32 d5, d6, d7 // half pixel 6 (2x32) |
| 3286 |
|
| 3287 |
vext.8 q3, q6, q7, #2*7 // filter 7 pixels |
| 3288 |
|
| 3289 |
vpadd.i32 d11, d4, d5 // pixel 6 (2x32) |
| 3290 |
|
| 3291 |
vmull.s16 q2, d6, d2 // filter 7 output (0-3) |
| 3292 |
vmull.s16 q3, d7, d3 // filter 7 output (4-7) |
| 3293 |
|
| 3294 |
vld1.32 {d14[],d15[]}, [sp] // -(7 - intermediate_bits) |
| 3295 |
|
| 3296 |
vpadd.i32 d4, d4, d5 // half pixel 7 (2x32) |
| 3297 |
vpadd.i32 d5, d6, d7 // half pixel 7 (2x32) |
| 3298 |
|
| 3299 |
sub r5, r5, r7, lsl #3 |
| 3300 |
|
| 3301 |
vpadd.i32 d4, d4, d5 // pixel 7 (2x32) |
| 3302 |
|
| 3303 |
add r5, r5, r8 |
| 3304 |
|
| 3305 |
vpadd.i32 d11, d11, d4 // pixel 6,7 |
| 3306 |
|
| 3307 |
vrshl.s32 q4, q4, q7 // -(7 - intermediate_bits) |
| 3308 |
vrshl.s32 q5, q5, q7 // -(7 - intermediate_bits) |
| 3309 |
|
| 3310 |
bx lr |
| 3311 |
endfunc |
| 3312 |
|
| 3313 |
// void dav1d_warp_affine_8x8_16bpc_neon( |
| 3314 |
// pixel *dst, const ptrdiff_t dst_stride, |
| 3315 |
// const pixel *src, const ptrdiff_t src_stride, |
| 3316 |
// const int16_t *const abcd, int mx, int my, |
| 3317 |
// const int bitdepth_max) |
| 3318 |
.macro warp t |
| 3319 |
function warp_affine_8x8\t\()_16bpc_neon, export=1 |
| 3320 |
push {r4-r11,lr} |
| 3321 |
vpush {q4-q7} |
| 3322 |
ldrd r4, r5, [sp, #100] |
| 3323 |
ldrd r6, r7, [sp, #108] |
| 3324 |
sub sp, sp, #8 |
| 3325 |
|
| 3326 |
clz r7, r7 |
| 3327 |
// intermediate_bits = clz(bitdepth_max) - 18 |
| 3328 |
.ifb \t |
| 3329 |
sub r8, r7, #11 // 7 + intermediate_bits = clz(bitdepth_max) - 18 + 7 |
| 3330 |
.endif |
| 3331 |
sub r7, r7, #25 // -(7 - intermediate_bits) |
| 3332 |
.ifb \t |
| 3333 |
neg r8, r8 // -(7 + intermediate_bits) |
| 3334 |
.endif |
| 3335 |
str r7, [sp] // spill -(7 - intermediate_bits) on stack |
| 3336 |
.ifb \t |
| 3337 |
str r8, [sp, #4] // spill -(7 + intermediate_bits) on stack |
| 3338 |
.endif |
| 3339 |
|
| 3340 |
ldrd r8, r9, [r4] |
| 3341 |
sxth r7, r8 |
| 3342 |
asr r8, r8, #16 |
| 3343 |
asr r4, r9, #16 |
| 3344 |
sxth r9, r9 |
| 3345 |
mov r10, #8 |
| 3346 |
sub r2, r2, r3, lsl #1 |
| 3347 |
sub r2, r2, r3 |
| 3348 |
sub r2, r2, #6 |
| 3349 |
movrel r11, X(mc_warp_filter), 64*8 |
| 3350 |
.ifnb \t |
| 3351 |
lsl r1, r1, #1 |
| 3352 |
.endif |
| 3353 |
add r5, r5, #512 |
| 3354 |
add r6, r6, #512 |
| 3355 |
|
| 3356 |
bl warp_filter_horz_neon |
| 3357 |
vmovn.i32 d16, q4 |
| 3358 |
vmovn.i32 d17, q5 |
| 3359 |
bl warp_filter_horz_neon |
| 3360 |
vmovn.i32 d18, q4 |
| 3361 |
vmovn.i32 d19, q5 |
| 3362 |
bl warp_filter_horz_neon |
| 3363 |
vmovn.i32 d20, q4 |
| 3364 |
vmovn.i32 d21, q5 |
| 3365 |
bl warp_filter_horz_neon |
| 3366 |
vmovn.i32 d22, q4 |
| 3367 |
vmovn.i32 d23, q5 |
| 3368 |
bl warp_filter_horz_neon |
| 3369 |
vmovn.i32 d24, q4 |
| 3370 |
vmovn.i32 d25, q5 |
| 3371 |
bl warp_filter_horz_neon |
| 3372 |
vmovn.i32 d26, q4 |
| 3373 |
vmovn.i32 d27, q5 |
| 3374 |
bl warp_filter_horz_neon |
| 3375 |
vmovn.i32 d28, q4 |
| 3376 |
vmovn.i32 d29, q5 |
| 3377 |
|
| 3378 |
1: |
| 3379 |
bl warp_filter_horz_neon |
| 3380 |
vmovn.i32 d30, q4 |
| 3381 |
vmovn.i32 d31, q5 |
| 3382 |
|
| 3383 |
load_filter_row d8, r6, r9 |
| 3384 |
load_filter_row d9, r6, r9 |
| 3385 |
load_filter_row d10, r6, r9 |
| 3386 |
load_filter_row d11, r6, r9 |
| 3387 |
load_filter_row d12, r6, r9 |
| 3388 |
load_filter_row d13, r6, r9 |
| 3389 |
load_filter_row d14, r6, r9 |
| 3390 |
load_filter_row d15, r6, r9 |
| 3391 |
transpose_8x8b q4, q5, q6, q7, d8, d9, d10, d11, d12, d13, d14, d15 |
| 3392 |
vmovl.s8 q1, d8 |
| 3393 |
vmovl.s8 q2, d9 |
| 3394 |
vmovl.s8 q3, d10 |
| 3395 |
vmovl.s8 q4, d11 |
| 3396 |
vmovl.s8 q5, d12 |
| 3397 |
vmovl.s8 q6, d13 |
| 3398 |
|
| 3399 |
sub r6, r6, r9, lsl #3 |
| 3400 |
|
| 3401 |
// This ordering of vmull/vmlal is highly beneficial for |
| 3402 |
// Cortex A8/A9/A53 here, but harmful for Cortex A7. |
| 3403 |
vmull.s16 q0, d16, d2 |
| 3404 |
vmlal.s16 q0, d18, d4 |
| 3405 |
vmlal.s16 q0, d20, d6 |
| 3406 |
vmlal.s16 q0, d22, d8 |
| 3407 |
vmlal.s16 q0, d24, d10 |
| 3408 |
vmlal.s16 q0, d26, d12 |
| 3409 |
vmull.s16 q1, d17, d3 |
| 3410 |
vmlal.s16 q1, d19, d5 |
| 3411 |
vmlal.s16 q1, d21, d7 |
| 3412 |
vmlal.s16 q1, d23, d9 |
| 3413 |
vmlal.s16 q1, d25, d11 |
| 3414 |
vmlal.s16 q1, d27, d13 |
| 3415 |
|
| 3416 |
vmovl.s8 q2, d14 |
| 3417 |
vmovl.s8 q3, d15 |
| 3418 |
|
| 3419 |
vmlal.s16 q0, d28, d4 |
| 3420 |
vmlal.s16 q0, d30, d6 |
| 3421 |
vmlal.s16 q1, d29, d5 |
| 3422 |
vmlal.s16 q1, d31, d7 |
| 3423 |
|
| 3424 |
.ifb \t |
| 3425 |
ldr lr, [sp, #4] // -(7 + intermediate_bits) |
| 3426 |
ldr r12, [sp, #120] // bitdepth_max |
| 3427 |
vdup.32 q2, lr // -(7 + intermediate_bits) |
| 3428 |
vdup.16 q3, r12 // bitdepth_max |
| 3429 |
.endif |
| 3430 |
|
| 3431 |
vmov q8, q9 |
| 3432 |
vmov q9, q10 |
| 3433 |
.ifb \t |
| 3434 |
vrshl.s32 q0, q0, q2 // -(7 + intermediate_bits) |
| 3435 |
vrshl.s32 q1, q1, q2 // -(7 + intermediate_bits) |
| 3436 |
.else |
| 3437 |
vrshrn.s32 d0, q0, #7 |
| 3438 |
vrshrn.s32 d1, q1, #7 |
| 3439 |
vmov.i16 q3, #PREP_BIAS |
| 3440 |
.endif |
| 3441 |
vmov q10, q11 |
| 3442 |
.ifb \t |
| 3443 |
vqmovun.s32 d0, q0 |
| 3444 |
vqmovun.s32 d1, q1 |
| 3445 |
.else |
| 3446 |
vsub.i16 q0, q0, q3 // PREP_BIAS |
| 3447 |
.endif |
| 3448 |
vmov q11, q12 |
| 3449 |
vmov q12, q13 |
| 3450 |
.ifb \t |
| 3451 |
vmin.u16 q0, q0, q3 // bitdepth_max |
| 3452 |
.endif |
| 3453 |
vmov q13, q14 |
| 3454 |
vmov q14, q15 |
| 3455 |
subs r10, r10, #1 |
| 3456 |
vst1.16 {q0}, [r0, :128], r1 |
| 3457 |
|
| 3458 |
add r6, r6, r4 |
| 3459 |
bgt 1b |
| 3460 |
|
| 3461 |
add sp, sp, #8 |
| 3462 |
vpop {q4-q7} |
| 3463 |
pop {r4-r11,pc} |
| 3464 |
endfunc |
| 3465 |
.endm |
| 3466 |
|
| 3467 |
warp |
| 3468 |
warp t |
| 3469 |
|
| 3470 |
// void dav1d_emu_edge_16bpc_neon( |
| 3471 |
// const intptr_t bw, const intptr_t bh, |
| 3472 |
// const intptr_t iw, const intptr_t ih, |
| 3473 |
// const intptr_t x, const intptr_t y, |
| 3474 |
// pixel *dst, const ptrdiff_t dst_stride, |
| 3475 |
// const pixel *ref, const ptrdiff_t ref_stride) |
| 3476 |
function emu_edge_16bpc_neon, export=1 |
| 3477 |
push {r4-r11,lr} |
| 3478 |
ldrd r4, r5, [sp, #36] |
| 3479 |
ldrd r6, r7, [sp, #44] |
| 3480 |
ldrd r8, r9, [sp, #52] |
| 3481 |
|
| 3482 |
// ref += iclip(y, 0, ih - 1) * PXSTRIDE(ref_stride) |
| 3483 |
// ref += iclip(x, 0, iw - 1) |
| 3484 |
sub r12, r3, #1 // ih - 1 |
| 3485 |
cmp r5, r3 |
| 3486 |
sub lr, r2, #1 // iw - 1 |
| 3487 |
it lt |
| 3488 |
movlt r12, r5 // min(y, ih - 1) |
| 3489 |
cmp r4, r2 |
| 3490 |
bic r12, r12, r12, asr #31 // max(min(y, ih - 1), 0) |
| 3491 |
it lt |
| 3492 |
movlt lr, r4 // min(x, iw - 1) |
| 3493 |
bic lr, lr, lr, asr #31 // max(min(x, iw - 1), 0) |
| 3494 |
mla r8, r12, r9, r8 // ref += iclip() * stride |
| 3495 |
add r8, r8, lr, lsl #1 // ref += iclip() |
| 3496 |
|
| 3497 |
// bottom_ext = iclip(y + bh - ih, 0, bh - 1) |
| 3498 |
// top_ext = iclip(-y, 0, bh - 1) |
| 3499 |
add r10, r5, r1 // y + bh |
| 3500 |
neg r5, r5 // -y |
| 3501 |
sub r10, r10, r3 // y + bh - ih |
| 3502 |
sub r12, r1, #1 // bh - 1 |
| 3503 |
cmp r10, r1 |
| 3504 |
bic r5, r5, r5, asr #31 // max(-y, 0) |
| 3505 |
it ge |
| 3506 |
movge r10, r12 // min(y + bh - ih, bh-1) |
| 3507 |
cmp r5, r1 |
| 3508 |
bic r10, r10, r10, asr #31 // max(min(y + bh - ih, bh-1), 0) |
| 3509 |
it ge |
| 3510 |
movge r5, r12 // min(max(-y, 0), bh-1) |
| 3511 |
|
| 3512 |
// right_ext = iclip(x + bw - iw, 0, bw - 1) |
| 3513 |
// left_ext = iclip(-x, 0, bw - 1) |
| 3514 |
add r11, r4, r0 // x + bw |
| 3515 |
neg r4, r4 // -x |
| 3516 |
sub r11, r11, r2 // x + bw - iw |
| 3517 |
sub lr, r0, #1 // bw - 1 |
| 3518 |
cmp r11, r0 |
| 3519 |
bic r4, r4, r4, asr #31 // max(-x, 0) |
| 3520 |
it ge |
| 3521 |
movge r11, lr // min(x + bw - iw, bw-1) |
| 3522 |
cmp r4, r0 |
| 3523 |
bic r11, r11, r11, asr #31 // max(min(x + bw - iw, bw-1), 0) |
| 3524 |
it ge |
| 3525 |
movge r4, lr // min(max(-x, 0), bw - 1) |
| 3526 |
|
| 3527 |
// center_h = bh - top_ext - bottom_ext |
| 3528 |
// dst += top_ext * PXSTRIDE(dst_stride) |
| 3529 |
// center_w = bw - left_ext - right_ext |
| 3530 |
sub r1, r1, r5 // bh - top_ext |
| 3531 |
mla r6, r5, r7, r6 |
| 3532 |
sub r2, r0, r4 // bw - left_ext |
| 3533 |
sub r1, r1, r10 // center_h = bh - top_ext - bottom_ext |
| 3534 |
sub r2, r2, r11 // center_w = bw - left_ext - right_ext |
| 3535 |
|
| 3536 |
mov r0, r6 // backup of dst |
| 3537 |
|
| 3538 |
.macro v_loop need_left, need_right |
| 3539 |
0: |
| 3540 |
.if \need_left |
| 3541 |
vld1.16 {d0[], d1[]}, [r8] |
| 3542 |
mov r12, r6 // out = dst |
| 3543 |
mov r3, r4 |
| 3544 |
vmov q1, q0 |
| 3545 |
1: |
| 3546 |
subs r3, r3, #16 |
| 3547 |
vst1.16 {q0, q1}, [r12, :128]! |
| 3548 |
bgt 1b |
| 3549 |
.endif |
| 3550 |
mov lr, r8 |
| 3551 |
add r12, r6, r4, lsl #1 // out = dst + left_ext |
| 3552 |
mov r3, r2 |
| 3553 |
1: |
| 3554 |
vld1.16 {q0, q1}, [lr]! |
| 3555 |
subs r3, r3, #32 |
| 3556 |
vld1.16 {q2, q3}, [lr]! |
| 3557 |
.if \need_left |
| 3558 |
vst1.16 {q0, q1}, [r12]! |
| 3559 |
vst1.16 {q2, q3}, [r12]! |
| 3560 |
.else |
| 3561 |
vst1.16 {q0, q1}, [r12, :128]! |
| 3562 |
vst1.16 {q2, q3}, [r12, :128]! |
| 3563 |
.endif |
| 3564 |
bgt 1b |
| 3565 |
.if \need_right |
| 3566 |
add r3, r8, r2, lsl #1 // in + center_w |
| 3567 |
sub r3, r3, #2 // in + center_w - 1 |
| 3568 |
add r12, r6, r4, lsl #1 // dst + left_ext |
| 3569 |
vld1.16 {d0[], d1[]}, [r3] |
| 3570 |
add r12, r12, r2, lsl #1 // out = dst + left_ext + center_w |
| 3571 |
mov r3, r11 |
| 3572 |
vmov q1, q0 |
| 3573 |
1: |
| 3574 |
subs r3, r3, #16 |
| 3575 |
vst1.16 {q0, q1}, [r12]! |
| 3576 |
bgt 1b |
| 3577 |
.endif |
| 3578 |
|
| 3579 |
subs r1, r1, #1 // center_h-- |
| 3580 |
add r6, r6, r7 |
| 3581 |
add r8, r8, r9 |
| 3582 |
bgt 0b |
| 3583 |
.endm |
| 3584 |
|
| 3585 |
cmp r4, #0 |
| 3586 |
beq 2f |
| 3587 |
// need_left |
| 3588 |
cmp r11, #0 |
| 3589 |
beq 3f |
| 3590 |
// need_left + need_right |
| 3591 |
v_loop 1, 1 |
| 3592 |
b 5f |
| 3593 |
|
| 3594 |
2: |
| 3595 |
// !need_left |
| 3596 |
cmp r11, #0 |
| 3597 |
beq 4f |
| 3598 |
// !need_left + need_right |
| 3599 |
v_loop 0, 1 |
| 3600 |
b 5f |
| 3601 |
|
| 3602 |
3: |
| 3603 |
// need_left + !need_right |
| 3604 |
v_loop 1, 0 |
| 3605 |
b 5f |
| 3606 |
|
| 3607 |
4: |
| 3608 |
// !need_left + !need_right |
| 3609 |
v_loop 0, 0 |
| 3610 |
|
| 3611 |
5: |
| 3612 |
cmp r10, #0 |
| 3613 |
// Storing the original dst in r0 overwrote bw, recalculate it here |
| 3614 |
add r2, r2, r4 // center_w + left_ext |
| 3615 |
add r2, r2, r11 // bw = center_w + left_ext + right_ext |
| 3616 |
|
| 3617 |
beq 3f |
| 3618 |
// need_bottom |
| 3619 |
sub r8, r6, r7 // ref = dst - stride |
| 3620 |
mov r4, r2 |
| 3621 |
sub r12, r7, #32 |
| 3622 |
1: |
| 3623 |
vld1.16 {q0, q1}, [r8, :128]! |
| 3624 |
mov r3, r10 |
| 3625 |
vld1.16 {q2, q3}, [r8, :128]! |
| 3626 |
2: |
| 3627 |
vst1.16 {q0, q1}, [r6, :128]! |
| 3628 |
subs r3, r3, #1 |
| 3629 |
vst1.16 {q2, q3}, [r6, :128], r12 |
| 3630 |
bgt 2b |
| 3631 |
mls r6, r7, r10, r6 // dst -= bottom_ext * stride |
| 3632 |
subs r4, r4, #32 // bw -= 32 |
| 3633 |
add r6, r6, #64 // dst += 32 |
| 3634 |
bgt 1b |
| 3635 |
|
| 3636 |
3: |
| 3637 |
cmp r5, #0 |
| 3638 |
beq 3f |
| 3639 |
// need_top |
| 3640 |
mls r6, r7, r5, r0 // dst = stored_dst - top_ext * stride |
| 3641 |
sub r12, r7, #32 |
| 3642 |
1: |
| 3643 |
vld1.16 {q0, q1}, [r0, :128]! |
| 3644 |
mov r3, r5 |
| 3645 |
vld1.16 {q2, q3}, [r0, :128]! |
| 3646 |
2: |
| 3647 |
vst1.16 {q0, q1}, [r6, :128]! |
| 3648 |
subs r3, r3, #1 |
| 3649 |
vst1.16 {q2, q3}, [r6, :128], r12 |
| 3650 |
bgt 2b |
| 3651 |
mls r6, r7, r5, r6 // dst -= top_ext * stride |
| 3652 |
subs r2, r2, #32 // bw -= 32 |
| 3653 |
add r6, r6, #64 // dst += 32 |
| 3654 |
bgt 1b |
| 3655 |
|
| 3656 |
3: |
| 3657 |
pop {r4-r11,pc} |
| 3658 |
endfunc |