1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libyuv/source/rotate_mips.cc Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,486 @@
1.4 +/*
1.5 + * Copyright 2011 The LibYuv Project Authors. All rights reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include "libyuv/row.h"
1.15 +
1.16 +#include "libyuv/basic_types.h"
1.17 +
1.18 +#ifdef __cplusplus
1.19 +namespace libyuv {
1.20 +extern "C" {
1.21 +#endif
1.22 +
1.23 +#if !defined(LIBYUV_DISABLE_MIPS) && \
1.24 + defined(__mips_dsp) && (__mips_dsp_rev >= 2)
1.25 +
1.26 +void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride,
1.27 + uint8* dst, int dst_stride,
1.28 + int width) {
1.29 + __asm__ __volatile__ (
1.30 + ".set push \n"
1.31 + ".set noreorder \n"
1.32 + "sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
1.33 + "sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
1.34 + "sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
1.35 + "addu $t3, $t2, %[src_stride] \n"
1.36 + "addu $t5, $t4, %[src_stride] \n"
1.37 + "addu $t6, $t2, $t4 \n"
1.38 + "andi $t0, %[dst], 0x3 \n"
1.39 + "andi $t1, %[dst_stride], 0x3 \n"
1.40 + "or $t0, $t0, $t1 \n"
1.41 + "bnez $t0, 11f \n"
1.42 + " subu $t7, $t9, %[src_stride] \n"
1.43 +//dst + dst_stride word aligned
1.44 + "1: \n"
1.45 + "lbu $t0, 0(%[src]) \n"
1.46 + "lbux $t1, %[src_stride](%[src]) \n"
1.47 + "lbux $t8, $t2(%[src]) \n"
1.48 + "lbux $t9, $t3(%[src]) \n"
1.49 + "sll $t1, $t1, 16 \n"
1.50 + "sll $t9, $t9, 16 \n"
1.51 + "or $t0, $t0, $t1 \n"
1.52 + "or $t8, $t8, $t9 \n"
1.53 + "precr.qb.ph $s0, $t8, $t0 \n"
1.54 + "lbux $t0, $t4(%[src]) \n"
1.55 + "lbux $t1, $t5(%[src]) \n"
1.56 + "lbux $t8, $t6(%[src]) \n"
1.57 + "lbux $t9, $t7(%[src]) \n"
1.58 + "sll $t1, $t1, 16 \n"
1.59 + "sll $t9, $t9, 16 \n"
1.60 + "or $t0, $t0, $t1 \n"
1.61 + "or $t8, $t8, $t9 \n"
1.62 + "precr.qb.ph $s1, $t8, $t0 \n"
1.63 + "sw $s0, 0(%[dst]) \n"
1.64 + "addiu %[width], -1 \n"
1.65 + "addiu %[src], 1 \n"
1.66 + "sw $s1, 4(%[dst]) \n"
1.67 + "bnez %[width], 1b \n"
1.68 + " addu %[dst], %[dst], %[dst_stride] \n"
1.69 + "b 2f \n"
1.70 +//dst + dst_stride unaligned
1.71 + "11: \n"
1.72 + "lbu $t0, 0(%[src]) \n"
1.73 + "lbux $t1, %[src_stride](%[src]) \n"
1.74 + "lbux $t8, $t2(%[src]) \n"
1.75 + "lbux $t9, $t3(%[src]) \n"
1.76 + "sll $t1, $t1, 16 \n"
1.77 + "sll $t9, $t9, 16 \n"
1.78 + "or $t0, $t0, $t1 \n"
1.79 + "or $t8, $t8, $t9 \n"
1.80 + "precr.qb.ph $s0, $t8, $t0 \n"
1.81 + "lbux $t0, $t4(%[src]) \n"
1.82 + "lbux $t1, $t5(%[src]) \n"
1.83 + "lbux $t8, $t6(%[src]) \n"
1.84 + "lbux $t9, $t7(%[src]) \n"
1.85 + "sll $t1, $t1, 16 \n"
1.86 + "sll $t9, $t9, 16 \n"
1.87 + "or $t0, $t0, $t1 \n"
1.88 + "or $t8, $t8, $t9 \n"
1.89 + "precr.qb.ph $s1, $t8, $t0 \n"
1.90 + "swr $s0, 0(%[dst]) \n"
1.91 + "swl $s0, 3(%[dst]) \n"
1.92 + "addiu %[width], -1 \n"
1.93 + "addiu %[src], 1 \n"
1.94 + "swr $s1, 4(%[dst]) \n"
1.95 + "swl $s1, 7(%[dst]) \n"
1.96 + "bnez %[width], 11b \n"
1.97 + "addu %[dst], %[dst], %[dst_stride] \n"
1.98 + "2: \n"
1.99 + ".set pop \n"
1.100 + :[src] "+r" (src),
1.101 + [dst] "+r" (dst),
1.102 + [width] "+r" (width)
1.103 + :[src_stride] "r" (src_stride),
1.104 + [dst_stride] "r" (dst_stride)
1.105 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.106 + "t6", "t7", "t8", "t9",
1.107 + "s0", "s1"
1.108 + );
1.109 +}
1.110 +
1.111 +void TransposeWx8_FAST_MIPS_DSPR2(const uint8* src, int src_stride,
1.112 + uint8* dst, int dst_stride,
1.113 + int width) {
1.114 + __asm__ __volatile__ (
1.115 + ".set noat \n"
1.116 + ".set push \n"
1.117 + ".set noreorder \n"
1.118 + "beqz %[width], 2f \n"
1.119 + " sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
1.120 + "sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
1.121 + "sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
1.122 + "addu $t3, $t2, %[src_stride] \n"
1.123 + "addu $t5, $t4, %[src_stride] \n"
1.124 + "addu $t6, $t2, $t4 \n"
1.125 +
1.126 + "srl $AT, %[width], 0x2 \n"
1.127 + "andi $t0, %[dst], 0x3 \n"
1.128 + "andi $t1, %[dst_stride], 0x3 \n"
1.129 + "or $t0, $t0, $t1 \n"
1.130 + "bnez $t0, 11f \n"
1.131 + " subu $t7, $t9, %[src_stride] \n"
1.132 +//dst + dst_stride word aligned
1.133 + "1: \n"
1.134 + "lw $t0, 0(%[src]) \n"
1.135 + "lwx $t1, %[src_stride](%[src]) \n"
1.136 + "lwx $t8, $t2(%[src]) \n"
1.137 + "lwx $t9, $t3(%[src]) \n"
1.138 +
1.139 +// t0 = | 30 | 20 | 10 | 00 |
1.140 +// t1 = | 31 | 21 | 11 | 01 |
1.141 +// t8 = | 32 | 22 | 12 | 02 |
1.142 +// t9 = | 33 | 23 | 13 | 03 |
1.143 +
1.144 + "precr.qb.ph $s0, $t1, $t0 \n"
1.145 + "precr.qb.ph $s1, $t9, $t8 \n"
1.146 + "precrq.qb.ph $s2, $t1, $t0 \n"
1.147 + "precrq.qb.ph $s3, $t9, $t8 \n"
1.148 +
1.149 + // s0 = | 21 | 01 | 20 | 00 |
1.150 + // s1 = | 23 | 03 | 22 | 02 |
1.151 + // s2 = | 31 | 11 | 30 | 10 |
1.152 + // s3 = | 33 | 13 | 32 | 12 |
1.153 +
1.154 + "precr.qb.ph $s4, $s1, $s0 \n"
1.155 + "precrq.qb.ph $s5, $s1, $s0 \n"
1.156 + "precr.qb.ph $s6, $s3, $s2 \n"
1.157 + "precrq.qb.ph $s7, $s3, $s2 \n"
1.158 +
1.159 + // s4 = | 03 | 02 | 01 | 00 |
1.160 + // s5 = | 23 | 22 | 21 | 20 |
1.161 + // s6 = | 13 | 12 | 11 | 10 |
1.162 + // s7 = | 33 | 32 | 31 | 30 |
1.163 +
1.164 + "lwx $t0, $t4(%[src]) \n"
1.165 + "lwx $t1, $t5(%[src]) \n"
1.166 + "lwx $t8, $t6(%[src]) \n"
1.167 + "lwx $t9, $t7(%[src]) \n"
1.168 +
1.169 +// t0 = | 34 | 24 | 14 | 04 |
1.170 +// t1 = | 35 | 25 | 15 | 05 |
1.171 +// t8 = | 36 | 26 | 16 | 06 |
1.172 +// t9 = | 37 | 27 | 17 | 07 |
1.173 +
1.174 + "precr.qb.ph $s0, $t1, $t0 \n"
1.175 + "precr.qb.ph $s1, $t9, $t8 \n"
1.176 + "precrq.qb.ph $s2, $t1, $t0 \n"
1.177 + "precrq.qb.ph $s3, $t9, $t8 \n"
1.178 +
1.179 + // s0 = | 25 | 05 | 24 | 04 |
1.180 + // s1 = | 27 | 07 | 26 | 06 |
1.181 + // s2 = | 35 | 15 | 34 | 14 |
1.182 + // s3 = | 37 | 17 | 36 | 16 |
1.183 +
1.184 + "precr.qb.ph $t0, $s1, $s0 \n"
1.185 + "precrq.qb.ph $t1, $s1, $s0 \n"
1.186 + "precr.qb.ph $t8, $s3, $s2 \n"
1.187 + "precrq.qb.ph $t9, $s3, $s2 \n"
1.188 +
1.189 + // t0 = | 07 | 06 | 05 | 04 |
1.190 + // t1 = | 27 | 26 | 25 | 24 |
1.191 + // t8 = | 17 | 16 | 15 | 14 |
1.192 + // t9 = | 37 | 36 | 35 | 34 |
1.193 +
1.194 + "addu $s0, %[dst], %[dst_stride] \n"
1.195 + "addu $s1, $s0, %[dst_stride] \n"
1.196 + "addu $s2, $s1, %[dst_stride] \n"
1.197 +
1.198 + "sw $s4, 0(%[dst]) \n"
1.199 + "sw $t0, 4(%[dst]) \n"
1.200 + "sw $s6, 0($s0) \n"
1.201 + "sw $t8, 4($s0) \n"
1.202 + "sw $s5, 0($s1) \n"
1.203 + "sw $t1, 4($s1) \n"
1.204 + "sw $s7, 0($s2) \n"
1.205 + "sw $t9, 4($s2) \n"
1.206 +
1.207 + "addiu $AT, -1 \n"
1.208 + "addiu %[src], 4 \n"
1.209 +
1.210 + "bnez $AT, 1b \n"
1.211 + " addu %[dst], $s2, %[dst_stride] \n"
1.212 + "b 2f \n"
1.213 +//dst + dst_stride unaligned
1.214 + "11: \n"
1.215 + "lw $t0, 0(%[src]) \n"
1.216 + "lwx $t1, %[src_stride](%[src]) \n"
1.217 + "lwx $t8, $t2(%[src]) \n"
1.218 + "lwx $t9, $t3(%[src]) \n"
1.219 +
1.220 +// t0 = | 30 | 20 | 10 | 00 |
1.221 +// t1 = | 31 | 21 | 11 | 01 |
1.222 +// t8 = | 32 | 22 | 12 | 02 |
1.223 +// t9 = | 33 | 23 | 13 | 03 |
1.224 +
1.225 + "precr.qb.ph $s0, $t1, $t0 \n"
1.226 + "precr.qb.ph $s1, $t9, $t8 \n"
1.227 + "precrq.qb.ph $s2, $t1, $t0 \n"
1.228 + "precrq.qb.ph $s3, $t9, $t8 \n"
1.229 +
1.230 + // s0 = | 21 | 01 | 20 | 00 |
1.231 + // s1 = | 23 | 03 | 22 | 02 |
1.232 + // s2 = | 31 | 11 | 30 | 10 |
1.233 + // s3 = | 33 | 13 | 32 | 12 |
1.234 +
1.235 + "precr.qb.ph $s4, $s1, $s0 \n"
1.236 + "precrq.qb.ph $s5, $s1, $s0 \n"
1.237 + "precr.qb.ph $s6, $s3, $s2 \n"
1.238 + "precrq.qb.ph $s7, $s3, $s2 \n"
1.239 +
1.240 + // s4 = | 03 | 02 | 01 | 00 |
1.241 + // s5 = | 23 | 22 | 21 | 20 |
1.242 + // s6 = | 13 | 12 | 11 | 10 |
1.243 + // s7 = | 33 | 32 | 31 | 30 |
1.244 +
1.245 + "lwx $t0, $t4(%[src]) \n"
1.246 + "lwx $t1, $t5(%[src]) \n"
1.247 + "lwx $t8, $t6(%[src]) \n"
1.248 + "lwx $t9, $t7(%[src]) \n"
1.249 +
1.250 +// t0 = | 34 | 24 | 14 | 04 |
1.251 +// t1 = | 35 | 25 | 15 | 05 |
1.252 +// t8 = | 36 | 26 | 16 | 06 |
1.253 +// t9 = | 37 | 27 | 17 | 07 |
1.254 +
1.255 + "precr.qb.ph $s0, $t1, $t0 \n"
1.256 + "precr.qb.ph $s1, $t9, $t8 \n"
1.257 + "precrq.qb.ph $s2, $t1, $t0 \n"
1.258 + "precrq.qb.ph $s3, $t9, $t8 \n"
1.259 +
1.260 + // s0 = | 25 | 05 | 24 | 04 |
1.261 + // s1 = | 27 | 07 | 26 | 06 |
1.262 + // s2 = | 35 | 15 | 34 | 14 |
1.263 + // s3 = | 37 | 17 | 36 | 16 |
1.264 +
1.265 + "precr.qb.ph $t0, $s1, $s0 \n"
1.266 + "precrq.qb.ph $t1, $s1, $s0 \n"
1.267 + "precr.qb.ph $t8, $s3, $s2 \n"
1.268 + "precrq.qb.ph $t9, $s3, $s2 \n"
1.269 +
1.270 + // t0 = | 07 | 06 | 05 | 04 |
1.271 + // t1 = | 27 | 26 | 25 | 24 |
1.272 + // t8 = | 17 | 16 | 15 | 14 |
1.273 + // t9 = | 37 | 36 | 35 | 34 |
1.274 +
1.275 + "addu $s0, %[dst], %[dst_stride] \n"
1.276 + "addu $s1, $s0, %[dst_stride] \n"
1.277 + "addu $s2, $s1, %[dst_stride] \n"
1.278 +
1.279 + "swr $s4, 0(%[dst]) \n"
1.280 + "swl $s4, 3(%[dst]) \n"
1.281 + "swr $t0, 4(%[dst]) \n"
1.282 + "swl $t0, 7(%[dst]) \n"
1.283 + "swr $s6, 0($s0) \n"
1.284 + "swl $s6, 3($s0) \n"
1.285 + "swr $t8, 4($s0) \n"
1.286 + "swl $t8, 7($s0) \n"
1.287 + "swr $s5, 0($s1) \n"
1.288 + "swl $s5, 3($s1) \n"
1.289 + "swr $t1, 4($s1) \n"
1.290 + "swl $t1, 7($s1) \n"
1.291 + "swr $s7, 0($s2) \n"
1.292 + "swl $s7, 3($s2) \n"
1.293 + "swr $t9, 4($s2) \n"
1.294 + "swl $t9, 7($s2) \n"
1.295 +
1.296 + "addiu $AT, -1 \n"
1.297 + "addiu %[src], 4 \n"
1.298 +
1.299 + "bnez $AT, 11b \n"
1.300 + " addu %[dst], $s2, %[dst_stride] \n"
1.301 + "2: \n"
1.302 + ".set pop \n"
1.303 + ".set at \n"
1.304 + :[src] "+r" (src),
1.305 + [dst] "+r" (dst),
1.306 + [width] "+r" (width)
1.307 + :[src_stride] "r" (src_stride),
1.308 + [dst_stride] "r" (dst_stride)
1.309 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.310 + "t6", "t7", "t8", "t9",
1.311 + "s0", "s1", "s2", "s3", "s4",
1.312 + "s5", "s6", "s7"
1.313 + );
1.314 +}
1.315 +
1.316 +void TransposeUVWx8_MIPS_DSPR2(const uint8* src, int src_stride,
1.317 + uint8* dst_a, int dst_stride_a,
1.318 + uint8* dst_b, int dst_stride_b,
1.319 + int width) {
1.320 + __asm__ __volatile__ (
1.321 + ".set push \n"
1.322 + ".set noreorder \n"
1.323 + "beqz %[width], 2f \n"
1.324 + " sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
1.325 + "sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
1.326 + "sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
1.327 + "addu $t3, $t2, %[src_stride] \n"
1.328 + "addu $t5, $t4, %[src_stride] \n"
1.329 + "addu $t6, $t2, $t4 \n"
1.330 + "subu $t7, $t9, %[src_stride] \n"
1.331 + "srl $t1, %[width], 1 \n"
1.332 +
1.333 +// check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
1.334 + "andi $t0, %[dst_a], 0x3 \n"
1.335 + "andi $t8, %[dst_b], 0x3 \n"
1.336 + "or $t0, $t0, $t8 \n"
1.337 + "andi $t8, %[dst_stride_a], 0x3 \n"
1.338 + "andi $s5, %[dst_stride_b], 0x3 \n"
1.339 + "or $t8, $t8, $s5 \n"
1.340 + "or $t0, $t0, $t8 \n"
1.341 + "bnez $t0, 11f \n"
1.342 + " nop \n"
1.343 +// dst + dst_stride word aligned (both, a & b dst addresses)
1.344 + "1: \n"
1.345 + "lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
1.346 + "lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
1.347 + "addu $s5, %[dst_a], %[dst_stride_a] \n"
1.348 + "lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
1.349 + "lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
1.350 + "addu $s6, %[dst_b], %[dst_stride_b] \n"
1.351 +
1.352 + "precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
1.353 + "precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
1.354 + "precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
1.355 + "precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
1.356 +
1.357 + "sll $t0, $t0, 16 \n"
1.358 + "packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
1.359 + "sll $t9, $t9, 16 \n"
1.360 + "packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
1.361 +
1.362 + "sw $s3, 0($s5) \n"
1.363 + "sw $s4, 0($s6) \n"
1.364 +
1.365 + "precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
1.366 + "precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
1.367 +
1.368 + "lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
1.369 + "lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
1.370 + "lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
1.371 + "lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
1.372 + "sw $s3, 0(%[dst_a]) \n"
1.373 + "sw $s4, 0(%[dst_b]) \n"
1.374 +
1.375 + "precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
1.376 + "precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
1.377 + "precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
1.378 + "precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
1.379 +
1.380 + "sll $t0, $t0, 16 \n"
1.381 + "packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
1.382 + "sll $t9, $t9, 16 \n"
1.383 + "packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
1.384 + "sw $s3, 4($s5) \n"
1.385 + "sw $s4, 4($s6) \n"
1.386 +
1.387 + "precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
1.388 + "precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
1.389 +
1.390 + "addiu %[src], 4 \n"
1.391 + "addiu $t1, -1 \n"
1.392 + "sll $t0, %[dst_stride_a], 1 \n"
1.393 + "sll $t8, %[dst_stride_b], 1 \n"
1.394 + "sw $s3, 4(%[dst_a]) \n"
1.395 + "sw $s4, 4(%[dst_b]) \n"
1.396 + "addu %[dst_a], %[dst_a], $t0 \n"
1.397 + "bnez $t1, 1b \n"
1.398 + " addu %[dst_b], %[dst_b], $t8 \n"
1.399 + "b 2f \n"
1.400 + " nop \n"
1.401 +
1.402 +// dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
1.403 + "11: \n"
1.404 + "lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
1.405 + "lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
1.406 + "addu $s5, %[dst_a], %[dst_stride_a] \n"
1.407 + "lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
1.408 + "lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
1.409 + "addu $s6, %[dst_b], %[dst_stride_b] \n"
1.410 +
1.411 + "precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
1.412 + "precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
1.413 + "precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
1.414 + "precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
1.415 +
1.416 + "sll $t0, $t0, 16 \n"
1.417 + "packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
1.418 + "sll $t9, $t9, 16 \n"
1.419 + "packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
1.420 +
1.421 + "swr $s3, 0($s5) \n"
1.422 + "swl $s3, 3($s5) \n"
1.423 + "swr $s4, 0($s6) \n"
1.424 + "swl $s4, 3($s6) \n"
1.425 +
1.426 + "precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
1.427 + "precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
1.428 +
1.429 + "lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
1.430 + "lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
1.431 + "lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
1.432 + "lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
1.433 + "swr $s3, 0(%[dst_a]) \n"
1.434 + "swl $s3, 3(%[dst_a]) \n"
1.435 + "swr $s4, 0(%[dst_b]) \n"
1.436 + "swl $s4, 3(%[dst_b]) \n"
1.437 +
1.438 + "precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
1.439 + "precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
1.440 + "precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
1.441 + "precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
1.442 +
1.443 + "sll $t0, $t0, 16 \n"
1.444 + "packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
1.445 + "sll $t9, $t9, 16 \n"
1.446 + "packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
1.447 +
1.448 + "swr $s3, 4($s5) \n"
1.449 + "swl $s3, 7($s5) \n"
1.450 + "swr $s4, 4($s6) \n"
1.451 + "swl $s4, 7($s6) \n"
1.452 +
1.453 + "precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
1.454 + "precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
1.455 +
1.456 + "addiu %[src], 4 \n"
1.457 + "addiu $t1, -1 \n"
1.458 + "sll $t0, %[dst_stride_a], 1 \n"
1.459 + "sll $t8, %[dst_stride_b], 1 \n"
1.460 + "swr $s3, 4(%[dst_a]) \n"
1.461 + "swl $s3, 7(%[dst_a]) \n"
1.462 + "swr $s4, 4(%[dst_b]) \n"
1.463 + "swl $s4, 7(%[dst_b]) \n"
1.464 + "addu %[dst_a], %[dst_a], $t0 \n"
1.465 + "bnez $t1, 11b \n"
1.466 + " addu %[dst_b], %[dst_b], $t8 \n"
1.467 +
1.468 + "2: \n"
1.469 + ".set pop \n"
1.470 + : [src] "+r" (src),
1.471 + [dst_a] "+r" (dst_a),
1.472 + [dst_b] "+r" (dst_b),
1.473 + [width] "+r" (width),
1.474 + [src_stride] "+r" (src_stride)
1.475 + : [dst_stride_a] "r" (dst_stride_a),
1.476 + [dst_stride_b] "r" (dst_stride_b)
1.477 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.478 + "t6", "t7", "t8", "t9",
1.479 + "s0", "s1", "s2", "s3",
1.480 + "s4", "s5", "s6"
1.481 + );
1.482 +}
1.483 +
1.484 +#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
1.485 +
1.486 +#ifdef __cplusplus
1.487 +} // extern "C"
1.488 +} // namespace libyuv
1.489 +#endif
