1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libyuv/source/scale_mips.cc Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,653 @@
1.4 +/*
1.5 + * Copyright 2012 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/basic_types.h"
1.15 +#include "libyuv/row.h"
1.16 +
1.17 +#ifdef __cplusplus
1.18 +namespace libyuv {
1.19 +extern "C" {
1.20 +#endif
1.21 +
1.22 +// This module is for GCC MIPS DSPR2
1.23 +#if !defined(LIBYUV_DISABLE_MIPS) && \
1.24 + defined(__mips_dsp) && (__mips_dsp_rev >= 2)
1.25 +
1.26 +void ScaleRowDown2_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.27 + uint8* dst, int dst_width) {
1.28 + __asm__ __volatile__(
1.29 + ".set push \n"
1.30 + ".set noreorder \n"
1.31 +
1.32 + "srl $t9, %[dst_width], 4 \n" // iterations -> by 16
1.33 + "beqz $t9, 2f \n"
1.34 + " nop \n"
1.35 +
1.36 + ".p2align 2 \n"
1.37 + "1: \n"
1.38 + "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
1.39 + "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
1.40 + "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
1.41 + "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
1.42 + "lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
1.43 + "lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
1.44 + "lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
1.45 + "lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
1.46 + // TODO(fbarchard): Use odd pixels instead of even.
1.47 + "precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0|
1.48 + "precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8|
1.49 + "precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16|
1.50 + "precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24|
1.51 + "addiu %[src_ptr], %[src_ptr], 32 \n"
1.52 + "addiu $t9, $t9, -1 \n"
1.53 + "sw $t8, 0(%[dst]) \n"
1.54 + "sw $t0, 4(%[dst]) \n"
1.55 + "sw $t1, 8(%[dst]) \n"
1.56 + "sw $t2, 12(%[dst]) \n"
1.57 + "bgtz $t9, 1b \n"
1.58 + " addiu %[dst], %[dst], 16 \n"
1.59 +
1.60 + "2: \n"
1.61 + "andi $t9, %[dst_width], 0xf \n" // residue
1.62 + "beqz $t9, 3f \n"
1.63 + " nop \n"
1.64 +
1.65 + "21: \n"
1.66 + "lbu $t0, 0(%[src_ptr]) \n"
1.67 + "addiu %[src_ptr], %[src_ptr], 2 \n"
1.68 + "addiu $t9, $t9, -1 \n"
1.69 + "sb $t0, 0(%[dst]) \n"
1.70 + "bgtz $t9, 21b \n"
1.71 + " addiu %[dst], %[dst], 1 \n"
1.72 +
1.73 + "3: \n"
1.74 + ".set pop \n"
1.75 + : [src_ptr] "+r" (src_ptr),
1.76 + [dst] "+r" (dst)
1.77 + : [dst_width] "r" (dst_width)
1.78 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.79 + "t6", "t7", "t8", "t9"
1.80 + );
1.81 +}
1.82 +
1.83 +void ScaleRowDown2Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.84 + uint8* dst, int dst_width) {
1.85 + const uint8* t = src_ptr + src_stride;
1.86 +
1.87 + __asm__ __volatile__ (
1.88 + ".set push \n"
1.89 + ".set noreorder \n"
1.90 +
1.91 + "srl $t9, %[dst_width], 3 \n" // iterations -> step 8
1.92 + "bltz $t9, 2f \n"
1.93 + " nop \n"
1.94 +
1.95 + ".p2align 2 \n"
1.96 + "1: \n"
1.97 + "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
1.98 + "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
1.99 + "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
1.100 + "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
1.101 + "lw $t4, 0(%[t]) \n" // |19|18|17|16|
1.102 + "lw $t5, 4(%[t]) \n" // |23|22|21|20|
1.103 + "lw $t6, 8(%[t]) \n" // |27|26|25|24|
1.104 + "lw $t7, 12(%[t]) \n" // |31|30|29|28|
1.105 + "addiu $t9, $t9, -1 \n"
1.106 + "srl $t8, $t0, 16 \n" // |X|X|3|2|
1.107 + "ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
1.108 + "ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
1.109 + "raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
1.110 + "raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
1.111 + "shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
1.112 + "shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
1.113 + "srl $t8, $t1, 16 \n" // |X|X|7|6|
1.114 + "ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
1.115 + "ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
1.116 + "raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
1.117 + "raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
1.118 + "shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
1.119 + "shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
1.120 + "srl $t8, $t2, 16 \n" // |X|X|11|10|
1.121 + "ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
1.122 + "ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
1.123 + "raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
1.124 + "raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
1.125 + "shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
1.126 + "shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
1.127 + "srl $t8, $t3, 16 \n" // |X|X|15|14|
1.128 + "ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
1.129 + "ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
1.130 + "raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
1.131 + "raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
1.132 + "shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
1.133 + "shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
1.134 + "addiu %[src_ptr], %[src_ptr], 16 \n"
1.135 + "addiu %[t], %[t], 16 \n"
1.136 + "sb $t0, 0(%[dst]) \n"
1.137 + "sb $t4, 1(%[dst]) \n"
1.138 + "sb $t1, 2(%[dst]) \n"
1.139 + "sb $t5, 3(%[dst]) \n"
1.140 + "sb $t2, 4(%[dst]) \n"
1.141 + "sb $t6, 5(%[dst]) \n"
1.142 + "sb $t3, 6(%[dst]) \n"
1.143 + "sb $t7, 7(%[dst]) \n"
1.144 + "bgtz $t9, 1b \n"
1.145 + " addiu %[dst], %[dst], 8 \n"
1.146 +
1.147 + "2: \n"
1.148 + "andi $t9, %[dst_width], 0x7 \n" // x = residue
1.149 + "beqz $t9, 3f \n"
1.150 + " nop \n"
1.151 +
1.152 + "21: \n"
1.153 + "lwr $t1, 0(%[src_ptr]) \n"
1.154 + "lwl $t1, 3(%[src_ptr]) \n"
1.155 + "lwr $t2, 0(%[t]) \n"
1.156 + "lwl $t2, 3(%[t]) \n"
1.157 + "srl $t8, $t1, 16 \n"
1.158 + "ins $t1, $t2, 16, 16 \n"
1.159 + "ins $t2, $t8, 0, 16 \n"
1.160 + "raddu.w.qb $t1, $t1 \n"
1.161 + "raddu.w.qb $t2, $t2 \n"
1.162 + "shra_r.w $t1, $t1, 2 \n"
1.163 + "shra_r.w $t2, $t2, 2 \n"
1.164 + "sb $t1, 0(%[dst]) \n"
1.165 + "sb $t2, 1(%[dst]) \n"
1.166 + "addiu %[src_ptr], %[src_ptr], 4 \n"
1.167 + "addiu $t9, $t9, -2 \n"
1.168 + "addiu %[t], %[t], 4 \n"
1.169 + "bgtz $t9, 21b \n"
1.170 + " addiu %[dst], %[dst], 2 \n"
1.171 +
1.172 + "3: \n"
1.173 + ".set pop \n"
1.174 +
1.175 + : [src_ptr] "+r" (src_ptr),
1.176 + [dst] "+r" (dst), [t] "+r" (t)
1.177 + : [dst_width] "r" (dst_width)
1.178 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.179 + "t6", "t7", "t8", "t9"
1.180 + );
1.181 +}
1.182 +
1.183 +void ScaleRowDown4_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.184 + uint8* dst, int dst_width) {
1.185 + __asm__ __volatile__ (
1.186 + ".set push \n"
1.187 + ".set noreorder \n"
1.188 +
1.189 + "srl $t9, %[dst_width], 3 \n"
1.190 + "beqz $t9, 2f \n"
1.191 + " nop \n"
1.192 +
1.193 + ".p2align 2 \n"
1.194 + "1: \n"
1.195 + "lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
1.196 + "lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
1.197 + "lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
1.198 + "lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
1.199 + "lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
1.200 + "lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
1.201 + "lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
1.202 + "lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
1.203 + "precr.qb.ph $t1, $t2, $t1 \n" // |6|4|2|0|
1.204 + "precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8|
1.205 + "precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16|
1.206 + "precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24|
1.207 + "precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0|
1.208 + "precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16|
1.209 + "addiu %[src_ptr], %[src_ptr], 32 \n"
1.210 + "addiu $t9, $t9, -1 \n"
1.211 + "sw $t1, 0(%[dst]) \n"
1.212 + "sw $t5, 4(%[dst]) \n"
1.213 + "bgtz $t9, 1b \n"
1.214 + " addiu %[dst], %[dst], 8 \n"
1.215 +
1.216 + "2: \n"
1.217 + "andi $t9, %[dst_width], 7 \n" // residue
1.218 + "beqz $t9, 3f \n"
1.219 + " nop \n"
1.220 +
1.221 + "21: \n"
1.222 + "lbu $t1, 0(%[src_ptr]) \n"
1.223 + "addiu %[src_ptr], %[src_ptr], 4 \n"
1.224 + "addiu $t9, $t9, -1 \n"
1.225 + "sb $t1, 0(%[dst]) \n"
1.226 + "bgtz $t9, 21b \n"
1.227 + " addiu %[dst], %[dst], 1 \n"
1.228 +
1.229 + "3: \n"
1.230 + ".set pop \n"
1.231 + : [src_ptr] "+r" (src_ptr),
1.232 + [dst] "+r" (dst)
1.233 + : [dst_width] "r" (dst_width)
1.234 + : "t1", "t2", "t3", "t4", "t5",
1.235 + "t6", "t7", "t8", "t9"
1.236 + );
1.237 +}
1.238 +
1.239 +void ScaleRowDown4Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.240 + uint8* dst, int dst_width) {
1.241 + intptr_t stride = src_stride;
1.242 + const uint8* s1 = src_ptr + stride;
1.243 + const uint8* s2 = s1 + stride;
1.244 + const uint8* s3 = s2 + stride;
1.245 +
1.246 + __asm__ __volatile__ (
1.247 + ".set push \n"
1.248 + ".set noreorder \n"
1.249 +
1.250 + "srl $t9, %[dst_width], 1 \n"
1.251 + "andi $t8, %[dst_width], 1 \n"
1.252 +
1.253 + ".p2align 2 \n"
1.254 + "1: \n"
1.255 + "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
1.256 + "lw $t1, 0(%[s1]) \n" // |7|6|5|4|
1.257 + "lw $t2, 0(%[s2]) \n" // |11|10|9|8|
1.258 + "lw $t3, 0(%[s3]) \n" // |15|14|13|12|
1.259 + "lw $t4, 4(%[src_ptr]) \n" // |19|18|17|16|
1.260 + "lw $t5, 4(%[s1]) \n" // |23|22|21|20|
1.261 + "lw $t6, 4(%[s2]) \n" // |27|26|25|24|
1.262 + "lw $t7, 4(%[s3]) \n" // |31|30|29|28|
1.263 + "raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
1.264 + "raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
1.265 + "raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
1.266 + "raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
1.267 + "raddu.w.qb $t4, $t4 \n" // |19 + 18 + 17 + 16|
1.268 + "raddu.w.qb $t5, $t5 \n" // |23 + 22 + 21 + 20|
1.269 + "raddu.w.qb $t6, $t6 \n" // |27 + 26 + 25 + 24|
1.270 + "raddu.w.qb $t7, $t7 \n" // |31 + 30 + 29 + 28|
1.271 + "add $t0, $t0, $t1 \n"
1.272 + "add $t1, $t2, $t3 \n"
1.273 + "add $t0, $t0, $t1 \n"
1.274 + "add $t4, $t4, $t5 \n"
1.275 + "add $t6, $t6, $t7 \n"
1.276 + "add $t4, $t4, $t6 \n"
1.277 + "shra_r.w $t0, $t0, 4 \n"
1.278 + "shra_r.w $t4, $t4, 4 \n"
1.279 + "sb $t0, 0(%[dst]) \n"
1.280 + "sb $t4, 1(%[dst]) \n"
1.281 + "addiu %[src_ptr], %[src_ptr], 8 \n"
1.282 + "addiu %[s1], %[s1], 8 \n"
1.283 + "addiu %[s2], %[s2], 8 \n"
1.284 + "addiu %[s3], %[s3], 8 \n"
1.285 + "addiu $t9, $t9, -1 \n"
1.286 + "bgtz $t9, 1b \n"
1.287 + " addiu %[dst], %[dst], 2 \n"
1.288 + "beqz $t8, 2f \n"
1.289 + " nop \n"
1.290 +
1.291 + "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
1.292 + "lw $t1, 0(%[s1]) \n" // |7|6|5|4|
1.293 + "lw $t2, 0(%[s2]) \n" // |11|10|9|8|
1.294 + "lw $t3, 0(%[s3]) \n" // |15|14|13|12|
1.295 + "raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
1.296 + "raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
1.297 + "raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
1.298 + "raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
1.299 + "add $t0, $t0, $t1 \n"
1.300 + "add $t1, $t2, $t3 \n"
1.301 + "add $t0, $t0, $t1 \n"
1.302 + "shra_r.w $t0, $t0, 4 \n"
1.303 + "sb $t0, 0(%[dst]) \n"
1.304 +
1.305 + "2: \n"
1.306 + ".set pop \n"
1.307 +
1.308 + : [src_ptr] "+r" (src_ptr),
1.309 + [dst] "+r" (dst),
1.310 + [s1] "+r" (s1),
1.311 + [s2] "+r" (s2),
1.312 + [s3] "+r" (s3)
1.313 + : [dst_width] "r" (dst_width)
1.314 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.315 + "t6","t7", "t8", "t9"
1.316 + );
1.317 +}
1.318 +
1.319 +void ScaleRowDown34_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.320 + uint8* dst, int dst_width) {
1.321 + __asm__ __volatile__ (
1.322 + ".set push \n"
1.323 + ".set noreorder \n"
1.324 + ".p2align 2 \n"
1.325 + "1: \n"
1.326 + "lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
1.327 + "lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
1.328 + "lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
1.329 + "lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
1.330 + "lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
1.331 + "lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
1.332 + "lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
1.333 + "lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
1.334 + "precrq.qb.ph $t0, $t2, $t4 \n" // |7|5|15|13|
1.335 + "precrq.qb.ph $t9, $t6, $t8 \n" // |23|21|31|30|
1.336 + "addiu %[dst_width], %[dst_width], -24 \n"
1.337 + "ins $t1, $t1, 8, 16 \n" // |3|1|0|X|
1.338 + "ins $t4, $t0, 8, 16 \n" // |X|15|13|12|
1.339 + "ins $t5, $t5, 8, 16 \n" // |19|17|16|X|
1.340 + "ins $t8, $t9, 8, 16 \n" // |X|31|29|28|
1.341 + "addiu %[src_ptr], %[src_ptr], 32 \n"
1.342 + "packrl.ph $t0, $t3, $t0 \n" // |9|8|7|5|
1.343 + "packrl.ph $t9, $t7, $t9 \n" // |25|24|23|21|
1.344 + "prepend $t1, $t2, 8 \n" // |4|3|1|0|
1.345 + "prepend $t3, $t4, 24 \n" // |15|13|12|11|
1.346 + "prepend $t5, $t6, 8 \n" // |20|19|17|16|
1.347 + "prepend $t7, $t8, 24 \n" // |31|29|28|27|
1.348 + "sw $t1, 0(%[dst]) \n"
1.349 + "sw $t0, 4(%[dst]) \n"
1.350 + "sw $t3, 8(%[dst]) \n"
1.351 + "sw $t5, 12(%[dst]) \n"
1.352 + "sw $t9, 16(%[dst]) \n"
1.353 + "sw $t7, 20(%[dst]) \n"
1.354 + "bnez %[dst_width], 1b \n"
1.355 + " addiu %[dst], %[dst], 24 \n"
1.356 + ".set pop \n"
1.357 + : [src_ptr] "+r" (src_ptr),
1.358 + [dst] "+r" (dst),
1.359 + [dst_width] "+r" (dst_width)
1.360 + :
1.361 + : "t0", "t1", "t2", "t3", "t4", "t5",
1.362 + "t6","t7", "t8", "t9"
1.363 + );
1.364 +}
1.365 +
1.366 +void ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.367 + uint8* d, int dst_width) {
1.368 + __asm__ __volatile__ (
1.369 + ".set push \n"
1.370 + ".set noreorder \n"
1.371 + "repl.ph $t3, 3 \n" // 0x00030003
1.372 +
1.373 + ".p2align 2 \n"
1.374 + "1: \n"
1.375 + "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
1.376 + "lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
1.377 + "rotr $t2, $t0, 8 \n" // |S0|S3|S2|S1|
1.378 + "rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
1.379 + "muleu_s.ph.qbl $t4, $t2, $t3 \n" // |S0*3|S3*3|
1.380 + "muleu_s.ph.qbl $t5, $t6, $t3 \n" // |T0*3|T3*3|
1.381 + "andi $t0, $t2, 0xFFFF \n" // |0|0|S2|S1|
1.382 + "andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
1.383 + "raddu.w.qb $t0, $t0 \n"
1.384 + "raddu.w.qb $t1, $t1 \n"
1.385 + "shra_r.w $t0, $t0, 1 \n"
1.386 + "shra_r.w $t1, $t1, 1 \n"
1.387 + "preceu.ph.qbr $t2, $t2 \n" // |0|S2|0|S1|
1.388 + "preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
1.389 + "rotr $t2, $t2, 16 \n" // |0|S1|0|S2|
1.390 + "rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
1.391 + "addu.ph $t2, $t2, $t4 \n"
1.392 + "addu.ph $t6, $t6, $t5 \n"
1.393 + "sll $t5, $t0, 1 \n"
1.394 + "add $t0, $t5, $t0 \n"
1.395 + "shra_r.ph $t2, $t2, 2 \n"
1.396 + "shra_r.ph $t6, $t6, 2 \n"
1.397 + "shll.ph $t4, $t2, 1 \n"
1.398 + "addq.ph $t4, $t4, $t2 \n"
1.399 + "addu $t0, $t0, $t1 \n"
1.400 + "addiu %[src_ptr], %[src_ptr], 4 \n"
1.401 + "shra_r.w $t0, $t0, 2 \n"
1.402 + "addu.ph $t6, $t6, $t4 \n"
1.403 + "shra_r.ph $t6, $t6, 2 \n"
1.404 + "srl $t1, $t6, 16 \n"
1.405 + "addiu %[dst_width], %[dst_width], -3 \n"
1.406 + "sb $t1, 0(%[d]) \n"
1.407 + "sb $t0, 1(%[d]) \n"
1.408 + "sb $t6, 2(%[d]) \n"
1.409 + "bgtz %[dst_width], 1b \n"
1.410 + " addiu %[d], %[d], 3 \n"
1.411 + "3: \n"
1.412 + ".set pop \n"
1.413 + : [src_ptr] "+r" (src_ptr),
1.414 + [src_stride] "+r" (src_stride),
1.415 + [d] "+r" (d),
1.416 + [dst_width] "+r" (dst_width)
1.417 + :
1.418 + : "t0", "t1", "t2", "t3",
1.419 + "t4", "t5", "t6"
1.420 + );
1.421 +}
1.422 +
1.423 +void ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.424 + uint8* d, int dst_width) {
1.425 + __asm__ __volatile__ (
1.426 + ".set push \n"
1.427 + ".set noreorder \n"
1.428 + "repl.ph $t2, 3 \n" // 0x00030003
1.429 +
1.430 + ".p2align 2 \n"
1.431 + "1: \n"
1.432 + "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
1.433 + "lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
1.434 + "rotr $t4, $t0, 8 \n" // |S0|S3|S2|S1|
1.435 + "rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
1.436 + "muleu_s.ph.qbl $t3, $t4, $t2 \n" // |S0*3|S3*3|
1.437 + "muleu_s.ph.qbl $t5, $t6, $t2 \n" // |T0*3|T3*3|
1.438 + "andi $t0, $t4, 0xFFFF \n" // |0|0|S2|S1|
1.439 + "andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
1.440 + "raddu.w.qb $t0, $t0 \n"
1.441 + "raddu.w.qb $t1, $t1 \n"
1.442 + "shra_r.w $t0, $t0, 1 \n"
1.443 + "shra_r.w $t1, $t1, 1 \n"
1.444 + "preceu.ph.qbr $t4, $t4 \n" // |0|S2|0|S1|
1.445 + "preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
1.446 + "rotr $t4, $t4, 16 \n" // |0|S1|0|S2|
1.447 + "rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
1.448 + "addu.ph $t4, $t4, $t3 \n"
1.449 + "addu.ph $t6, $t6, $t5 \n"
1.450 + "shra_r.ph $t6, $t6, 2 \n"
1.451 + "shra_r.ph $t4, $t4, 2 \n"
1.452 + "addu.ph $t6, $t6, $t4 \n"
1.453 + "addiu %[src_ptr], %[src_ptr], 4 \n"
1.454 + "shra_r.ph $t6, $t6, 1 \n"
1.455 + "addu $t0, $t0, $t1 \n"
1.456 + "addiu %[dst_width], %[dst_width], -3 \n"
1.457 + "shra_r.w $t0, $t0, 1 \n"
1.458 + "srl $t1, $t6, 16 \n"
1.459 + "sb $t1, 0(%[d]) \n"
1.460 + "sb $t0, 1(%[d]) \n"
1.461 + "sb $t6, 2(%[d]) \n"
1.462 + "bgtz %[dst_width], 1b \n"
1.463 + " addiu %[d], %[d], 3 \n"
1.464 + "3: \n"
1.465 + ".set pop \n"
1.466 + : [src_ptr] "+r" (src_ptr),
1.467 + [src_stride] "+r" (src_stride),
1.468 + [d] "+r" (d),
1.469 + [dst_width] "+r" (dst_width)
1.470 + :
1.471 + : "t0", "t1", "t2", "t3",
1.472 + "t4", "t5", "t6"
1.473 + );
1.474 +}
1.475 +
1.476 +void ScaleRowDown38_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.477 + uint8* dst, int dst_width) {
1.478 + __asm__ __volatile__ (
1.479 + ".set push \n"
1.480 + ".set noreorder \n"
1.481 +
1.482 + ".p2align 2 \n"
1.483 + "1: \n"
1.484 + "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
1.485 + "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
1.486 + "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
1.487 + "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
1.488 + "lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
1.489 + "lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
1.490 + "lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
1.491 + "lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
1.492 + "wsbh $t0, $t0 \n" // |2|3|0|1|
1.493 + "wsbh $t6, $t6 \n" // |26|27|24|25|
1.494 + "srl $t0, $t0, 8 \n" // |X|2|3|0|
1.495 + "srl $t3, $t3, 16 \n" // |X|X|15|14|
1.496 + "srl $t5, $t5, 16 \n" // |X|X|23|22|
1.497 + "srl $t7, $t7, 16 \n" // |X|X|31|30|
1.498 + "ins $t1, $t2, 24, 8 \n" // |8|6|5|4|
1.499 + "ins $t6, $t5, 0, 8 \n" // |26|27|24|22|
1.500 + "ins $t1, $t0, 0, 16 \n" // |8|6|3|0|
1.501 + "ins $t6, $t7, 24, 8 \n" // |30|27|24|22|
1.502 + "prepend $t2, $t3, 24 \n" // |X|15|14|11|
1.503 + "ins $t4, $t4, 16, 8 \n" // |19|16|17|X|
1.504 + "ins $t4, $t2, 0, 16 \n" // |19|16|14|11|
1.505 + "addiu %[src_ptr], %[src_ptr], 32 \n"
1.506 + "addiu %[dst_width], %[dst_width], -12 \n"
1.507 + "addiu $t8,%[dst_width], -12 \n"
1.508 + "sw $t1, 0(%[dst]) \n"
1.509 + "sw $t4, 4(%[dst]) \n"
1.510 + "sw $t6, 8(%[dst]) \n"
1.511 + "bgez $t8, 1b \n"
1.512 + " addiu %[dst], %[dst], 12 \n"
1.513 + ".set pop \n"
1.514 + : [src_ptr] "+r" (src_ptr),
1.515 + [dst] "+r" (dst),
1.516 + [dst_width] "+r" (dst_width)
1.517 + :
1.518 + : "t0", "t1", "t2", "t3", "t4",
1.519 + "t5", "t6", "t7", "t8"
1.520 + );
1.521 +}
1.522 +
1.523 +void ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
1.524 + uint8* dst_ptr, int dst_width) {
1.525 + intptr_t stride = src_stride;
1.526 + const uint8* t = src_ptr + stride;
1.527 + const int c = 0x2AAA;
1.528 +
1.529 + __asm__ __volatile__ (
1.530 + ".set push \n"
1.531 + ".set noreorder \n"
1.532 +
1.533 + ".p2align 2 \n"
1.534 + "1: \n"
1.535 + "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
1.536 + "lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
1.537 + "lw $t2, 0(%[t]) \n" // |T3|T2|T1|T0|
1.538 + "lw $t3, 4(%[t]) \n" // |T7|T6|T5|T4|
1.539 + "rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
1.540 + "packrl.ph $t4, $t1, $t3 \n" // |S7|S6|T7|T6|
1.541 + "packrl.ph $t5, $t3, $t1 \n" // |T5|T4|S5|S4|
1.542 + "raddu.w.qb $t4, $t4 \n" // S7+S6+T7+T6
1.543 + "raddu.w.qb $t5, $t5 \n" // T5+T4+S5+S4
1.544 + "precrq.qb.ph $t6, $t0, $t2 \n" // |S3|S1|T3|T1|
1.545 + "precrq.qb.ph $t6, $t6, $t6 \n" // |S3|T3|S3|T3|
1.546 + "srl $t4, $t4, 2 \n" // t4 / 4
1.547 + "srl $t6, $t6, 16 \n" // |0|0|S3|T3|
1.548 + "raddu.w.qb $t6, $t6 \n" // 0+0+S3+T3
1.549 + "addu $t6, $t5, $t6 \n"
1.550 + "mul $t6, $t6, %[c] \n" // t6 * 0x2AAA
1.551 + "sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
1.552 + "sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
1.553 + "raddu.w.qb $t0, $t0 \n" // S2+S1+S0+0
1.554 + "raddu.w.qb $t2, $t2 \n" // T2+T1+T0+0
1.555 + "addu $t0, $t0, $t2 \n"
1.556 + "mul $t0, $t0, %[c] \n" // t0 * 0x2AAA
1.557 + "addiu %[src_ptr], %[src_ptr], 8 \n"
1.558 + "addiu %[t], %[t], 8 \n"
1.559 + "addiu %[dst_width], %[dst_width], -3 \n"
1.560 + "addiu %[dst_ptr], %[dst_ptr], 3 \n"
1.561 + "srl $t6, $t6, 16 \n"
1.562 + "srl $t0, $t0, 16 \n"
1.563 + "sb $t4, -1(%[dst_ptr]) \n"
1.564 + "sb $t6, -2(%[dst_ptr]) \n"
1.565 + "bgtz %[dst_width], 1b \n"
1.566 + " sb $t0, -3(%[dst_ptr]) \n"
1.567 + ".set pop \n"
1.568 + : [src_ptr] "+r" (src_ptr),
1.569 + [dst_ptr] "+r" (dst_ptr),
1.570 + [t] "+r" (t),
1.571 + [dst_width] "+r" (dst_width)
1.572 + : [c] "r" (c)
1.573 + : "t0", "t1", "t2", "t3", "t4", "t5", "t6"
1.574 + );
1.575 +}
1.576 +
1.577 +void ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8* src_ptr,
1.578 + ptrdiff_t src_stride,
1.579 + uint8* dst_ptr, int dst_width) {
1.580 + intptr_t stride = src_stride;
1.581 + const uint8* s1 = src_ptr + stride;
1.582 + stride += stride;
1.583 + const uint8* s2 = src_ptr + stride;
1.584 + const int c1 = 0x1C71;
1.585 + const int c2 = 0x2AAA;
1.586 +
1.587 + __asm__ __volatile__ (
1.588 + ".set push \n"
1.589 + ".set noreorder \n"
1.590 +
1.591 + ".p2align 2 \n"
1.592 + "1: \n"
1.593 + "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
1.594 + "lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
1.595 + "lw $t2, 0(%[s1]) \n" // |T3|T2|T1|T0|
1.596 + "lw $t3, 4(%[s1]) \n" // |T7|T6|T5|T4|
1.597 + "lw $t4, 0(%[s2]) \n" // |R3|R2|R1|R0|
1.598 + "lw $t5, 4(%[s2]) \n" // |R7|R6|R5|R4|
1.599 + "rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
1.600 + "packrl.ph $t6, $t1, $t3 \n" // |S7|S6|T7|T6|
1.601 + "raddu.w.qb $t6, $t6 \n" // S7+S6+T7+T6
1.602 + "packrl.ph $t7, $t3, $t1 \n" // |T5|T4|S5|S4|
1.603 + "raddu.w.qb $t7, $t7 \n" // T5+T4+S5+S4
1.604 + "sll $t8, $t5, 16 \n" // |R5|R4|0|0|
1.605 + "raddu.w.qb $t8, $t8 \n" // R5+R4
1.606 + "addu $t7, $t7, $t8 \n"
1.607 + "srl $t8, $t5, 16 \n" // |0|0|R7|R6|
1.608 + "raddu.w.qb $t8, $t8 \n" // R7 + R6
1.609 + "addu $t6, $t6, $t8 \n"
1.610 + "mul $t6, $t6, %[c2] \n" // t6 * 0x2AAA
1.611 + "precrq.qb.ph $t8, $t0, $t2 \n" // |S3|S1|T3|T1|
1.612 + "precrq.qb.ph $t8, $t8, $t4 \n" // |S3|T3|R3|R1|
1.613 + "srl $t8, $t8, 8 \n" // |0|S3|T3|R3|
1.614 + "raddu.w.qb $t8, $t8 \n" // S3 + T3 + R3
1.615 + "addu $t7, $t7, $t8 \n"
1.616 + "mul $t7, $t7, %[c1] \n" // t7 * 0x1C71
1.617 + "sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
1.618 + "sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
1.619 + "sll $t4, $t4, 8 \n" // |R2|R1|R0|0|
1.620 + "raddu.w.qb $t0, $t0 \n"
1.621 + "raddu.w.qb $t2, $t2 \n"
1.622 + "raddu.w.qb $t4, $t4 \n"
1.623 + "addu $t0, $t0, $t2 \n"
1.624 + "addu $t0, $t0, $t4 \n"
1.625 + "mul $t0, $t0, %[c1] \n" // t0 * 0x1C71
1.626 + "addiu %[src_ptr], %[src_ptr], 8 \n"
1.627 + "addiu %[s1], %[s1], 8 \n"
1.628 + "addiu %[s2], %[s2], 8 \n"
1.629 + "addiu %[dst_width], %[dst_width], -3 \n"
1.630 + "addiu %[dst_ptr], %[dst_ptr], 3 \n"
1.631 + "srl $t6, $t6, 16 \n"
1.632 + "srl $t7, $t7, 16 \n"
1.633 + "srl $t0, $t0, 16 \n"
1.634 + "sb $t6, -1(%[dst_ptr]) \n"
1.635 + "sb $t7, -2(%[dst_ptr]) \n"
1.636 + "bgtz %[dst_width], 1b \n"
1.637 + " sb $t0, -3(%[dst_ptr]) \n"
1.638 + ".set pop \n"
1.639 + : [src_ptr] "+r" (src_ptr),
1.640 + [dst_ptr] "+r" (dst_ptr),
1.641 + [s1] "+r" (s1),
1.642 + [s2] "+r" (s2),
1.643 + [dst_width] "+r" (dst_width)
1.644 + : [c1] "r" (c1), [c2] "r" (c2)
1.645 + : "t0", "t1", "t2", "t3", "t4",
1.646 + "t5", "t6", "t7", "t8"
1.647 + );
1.648 +}
1.649 +
1.650 +#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
1.651 +
1.652 +#ifdef __cplusplus
1.653 +} // extern "C"
1.654 +} // namespace libyuv
1.655 +#endif
1.656 +
