scale_mips.cc revision f047e7ca6983218eed7703c7afd51fed7bd3b5c9
1/*
2 *  Copyright 2012 The LibYuv Project Authors. All rights reserved.
3 *
4 *  Use of this source code is governed by a BSD-style license
5 *  that can be found in the LICENSE file in the root of the source
6 *  tree. An additional intellectual property rights grant can be found
7 *  in the file PATENTS. All contributing project authors may
8 *  be found in the AUTHORS file in the root of the source tree.
9 */
10
11#include "libyuv/basic_types.h"
12#include "libyuv/row.h"
13
14#ifdef __cplusplus
15namespace libyuv {
16extern "C" {
17#endif
18
19// This module is for GCC MIPS DSPR2
20#if !defined(LIBYUV_DISABLE_MIPS) && \
21    defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \
22    (_MIPS_SIM == _MIPS_SIM_ABI32)
23
24void ScaleRowDown2_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
25                         uint8* dst, int dst_width) {
26  __asm__ __volatile__(
27    ".set push                                     \n"
28    ".set noreorder                                \n"
29
30    "srl            $t9, %[dst_width], 4           \n"  // iterations -> by 16
31    "beqz           $t9, 2f                        \n"
32    " nop                                          \n"
33
34  "1:                                              \n"
35    "lw             $t0, 0(%[src_ptr])             \n"  // |3|2|1|0|
36    "lw             $t1, 4(%[src_ptr])             \n"  // |7|6|5|4|
37    "lw             $t2, 8(%[src_ptr])             \n"  // |11|10|9|8|
38    "lw             $t3, 12(%[src_ptr])            \n"  // |15|14|13|12|
39    "lw             $t4, 16(%[src_ptr])            \n"  // |19|18|17|16|
40    "lw             $t5, 20(%[src_ptr])            \n"  // |23|22|21|20|
41    "lw             $t6, 24(%[src_ptr])            \n"  // |27|26|25|24|
42    "lw             $t7, 28(%[src_ptr])            \n"  // |31|30|29|28|
43    // TODO(fbarchard): Use odd pixels instead of even.
44    "precr.qb.ph    $t8, $t1, $t0                  \n"  // |6|4|2|0|
45    "precr.qb.ph    $t0, $t3, $t2                  \n"  // |14|12|10|8|
46    "precr.qb.ph    $t1, $t5, $t4                  \n"  // |22|20|18|16|
47    "precr.qb.ph    $t2, $t7, $t6                  \n"  // |30|28|26|24|
48    "addiu          %[src_ptr], %[src_ptr], 32     \n"
49    "addiu          $t9, $t9, -1                   \n"
50    "sw             $t8, 0(%[dst])                 \n"
51    "sw             $t0, 4(%[dst])                 \n"
52    "sw             $t1, 8(%[dst])                 \n"
53    "sw             $t2, 12(%[dst])                \n"
54    "bgtz           $t9, 1b                        \n"
55    " addiu         %[dst], %[dst], 16             \n"
56
57  "2:                                              \n"
58    "andi           $t9, %[dst_width], 0xf         \n"  // residue
59    "beqz           $t9, 3f                        \n"
60    " nop                                          \n"
61
62  "21:                                             \n"
63    "lbu            $t0, 0(%[src_ptr])             \n"
64    "addiu          %[src_ptr], %[src_ptr], 2      \n"
65    "addiu          $t9, $t9, -1                   \n"
66    "sb             $t0, 0(%[dst])                 \n"
67    "bgtz           $t9, 21b                       \n"
68    " addiu         %[dst], %[dst], 1              \n"
69
70  "3:                                              \n"
71    ".set pop                                      \n"
72  : [src_ptr] "+r" (src_ptr),
73    [dst] "+r" (dst)
74  : [dst_width] "r" (dst_width)
75  : "t0", "t1", "t2", "t3", "t4", "t5",
76    "t6", "t7", "t8", "t9"
77  );
78}
79
80void ScaleRowDown2Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
81                            uint8* dst, int dst_width) {
82  const uint8* t = src_ptr + src_stride;
83
84  __asm__ __volatile__ (
85    ".set push                                    \n"
86    ".set noreorder                               \n"
87
88    "srl            $t9, %[dst_width], 3          \n"  // iterations -> step 8
89    "bltz           $t9, 2f                       \n"
90    " nop                                         \n"
91
92  "1:                                             \n"
93    "lw             $t0, 0(%[src_ptr])            \n"  // |3|2|1|0|
94    "lw             $t1, 4(%[src_ptr])            \n"  // |7|6|5|4|
95    "lw             $t2, 8(%[src_ptr])            \n"  // |11|10|9|8|
96    "lw             $t3, 12(%[src_ptr])           \n"  // |15|14|13|12|
97    "lw             $t4, 0(%[t])                  \n"  // |19|18|17|16|
98    "lw             $t5, 4(%[t])                  \n"  // |23|22|21|20|
99    "lw             $t6, 8(%[t])                  \n"  // |27|26|25|24|
100    "lw             $t7, 12(%[t])                 \n"  // |31|30|29|28|
101    "addiu          $t9, $t9, -1                  \n"
102    "srl            $t8, $t0, 16                  \n"  // |X|X|3|2|
103    "ins            $t0, $t4, 16, 16              \n"  // |17|16|1|0|
104    "ins            $t4, $t8, 0, 16               \n"  // |19|18|3|2|
105    "raddu.w.qb     $t0, $t0                      \n"  // |17+16+1+0|
106    "raddu.w.qb     $t4, $t4                      \n"  // |19+18+3+2|
107    "shra_r.w       $t0, $t0, 2                   \n"  // |t0+2|>>2
108    "shra_r.w       $t4, $t4, 2                   \n"  // |t4+2|>>2
109    "srl            $t8, $t1, 16                  \n"  // |X|X|7|6|
110    "ins            $t1, $t5, 16, 16              \n"  // |21|20|5|4|
111    "ins            $t5, $t8, 0, 16               \n"  // |22|23|7|6|
112    "raddu.w.qb     $t1, $t1                      \n"  // |21+20+5+4|
113    "raddu.w.qb     $t5, $t5                      \n"  // |23+22+7+6|
114    "shra_r.w       $t1, $t1, 2                   \n"  // |t1+2|>>2
115    "shra_r.w       $t5, $t5, 2                   \n"  // |t5+2|>>2
116    "srl            $t8, $t2, 16                  \n"  // |X|X|11|10|
117    "ins            $t2, $t6, 16, 16              \n"  // |25|24|9|8|
118    "ins            $t6, $t8, 0, 16               \n"  // |27|26|11|10|
119    "raddu.w.qb     $t2, $t2                      \n"  // |25+24+9+8|
120    "raddu.w.qb     $t6, $t6                      \n"  // |27+26+11+10|
121    "shra_r.w       $t2, $t2, 2                   \n"  // |t2+2|>>2
122    "shra_r.w       $t6, $t6, 2                   \n"  // |t5+2|>>2
123    "srl            $t8, $t3, 16                  \n"  // |X|X|15|14|
124    "ins            $t3, $t7, 16, 16              \n"  // |29|28|13|12|
125    "ins            $t7, $t8, 0, 16               \n"  // |31|30|15|14|
126    "raddu.w.qb     $t3, $t3                      \n"  // |29+28+13+12|
127    "raddu.w.qb     $t7, $t7                      \n"  // |31+30+15+14|
128    "shra_r.w       $t3, $t3, 2                   \n"  // |t3+2|>>2
129    "shra_r.w       $t7, $t7, 2                   \n"  // |t7+2|>>2
130    "addiu          %[src_ptr], %[src_ptr], 16    \n"
131    "addiu          %[t], %[t], 16                \n"
132    "sb             $t0, 0(%[dst])                \n"
133    "sb             $t4, 1(%[dst])                \n"
134    "sb             $t1, 2(%[dst])                \n"
135    "sb             $t5, 3(%[dst])                \n"
136    "sb             $t2, 4(%[dst])                \n"
137    "sb             $t6, 5(%[dst])                \n"
138    "sb             $t3, 6(%[dst])                \n"
139    "sb             $t7, 7(%[dst])                \n"
140    "bgtz           $t9, 1b                       \n"
141    " addiu         %[dst], %[dst], 8             \n"
142
143  "2:                                             \n"
144    "andi           $t9, %[dst_width], 0x7        \n"  // x = residue
145    "beqz           $t9, 3f                       \n"
146    " nop                                         \n"
147
148    "21:                                          \n"
149    "lwr            $t1, 0(%[src_ptr])            \n"
150    "lwl            $t1, 3(%[src_ptr])            \n"
151    "lwr            $t2, 0(%[t])                  \n"
152    "lwl            $t2, 3(%[t])                  \n"
153    "srl            $t8, $t1, 16                  \n"
154    "ins            $t1, $t2, 16, 16              \n"
155    "ins            $t2, $t8, 0, 16               \n"
156    "raddu.w.qb     $t1, $t1                      \n"
157    "raddu.w.qb     $t2, $t2                      \n"
158    "shra_r.w       $t1, $t1, 2                   \n"
159    "shra_r.w       $t2, $t2, 2                   \n"
160    "sb             $t1, 0(%[dst])                \n"
161    "sb             $t2, 1(%[dst])                \n"
162    "addiu          %[src_ptr], %[src_ptr], 4     \n"
163    "addiu          $t9, $t9, -2                  \n"
164    "addiu          %[t], %[t], 4                 \n"
165    "bgtz           $t9, 21b                      \n"
166    " addiu         %[dst], %[dst], 2             \n"
167
168  "3:                                             \n"
169    ".set pop                                     \n"
170
171  : [src_ptr] "+r" (src_ptr),
172    [dst] "+r" (dst), [t] "+r" (t)
173  : [dst_width] "r" (dst_width)
174  : "t0", "t1", "t2", "t3", "t4", "t5",
175    "t6", "t7", "t8", "t9"
176  );
177}
178
179void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
180                         uint8* dst, int dst_width) {
181  __asm__ __volatile__ (
182      ".set push                                    \n"
183      ".set noreorder                               \n"
184
185      "srl            $t9, %[dst_width], 3          \n"
186      "beqz           $t9, 2f                       \n"
187      " nop                                         \n"
188
189     "1:                                            \n"
190      "lw             $t1, 0(%[src_ptr])            \n"  // |3|2|1|0|
191      "lw             $t2, 4(%[src_ptr])            \n"  // |7|6|5|4|
192      "lw             $t3, 8(%[src_ptr])            \n"  // |11|10|9|8|
193      "lw             $t4, 12(%[src_ptr])           \n"  // |15|14|13|12|
194      "lw             $t5, 16(%[src_ptr])           \n"  // |19|18|17|16|
195      "lw             $t6, 20(%[src_ptr])           \n"  // |23|22|21|20|
196      "lw             $t7, 24(%[src_ptr])           \n"  // |27|26|25|24|
197      "lw             $t8, 28(%[src_ptr])           \n"  // |31|30|29|28|
198      "precr.qb.ph    $t1, $t2, $t1                 \n"  // |6|4|2|0|
199      "precr.qb.ph    $t2, $t4, $t3                 \n"  // |14|12|10|8|
200      "precr.qb.ph    $t5, $t6, $t5                 \n"  // |22|20|18|16|
201      "precr.qb.ph    $t6, $t8, $t7                 \n"  // |30|28|26|24|
202      "precr.qb.ph    $t1, $t2, $t1                 \n"  // |12|8|4|0|
203      "precr.qb.ph    $t5, $t6, $t5                 \n"  // |28|24|20|16|
204      "addiu          %[src_ptr], %[src_ptr], 32    \n"
205      "addiu          $t9, $t9, -1                  \n"
206      "sw             $t1, 0(%[dst])                \n"
207      "sw             $t5, 4(%[dst])                \n"
208      "bgtz           $t9, 1b                       \n"
209      " addiu         %[dst], %[dst], 8             \n"
210
211    "2:                                             \n"
212      "andi           $t9, %[dst_width], 7          \n"  // residue
213      "beqz           $t9, 3f                       \n"
214      " nop                                         \n"
215
216    "21:                                            \n"
217      "lbu            $t1, 0(%[src_ptr])            \n"
218      "addiu          %[src_ptr], %[src_ptr], 4     \n"
219      "addiu          $t9, $t9, -1                  \n"
220      "sb             $t1, 0(%[dst])                \n"
221      "bgtz           $t9, 21b                      \n"
222      " addiu         %[dst], %[dst], 1             \n"
223
224    "3:                                             \n"
225      ".set pop                                     \n"
226      : [src_ptr] "+r" (src_ptr),
227        [dst] "+r" (dst)
228      : [dst_width] "r" (dst_width)
229      : "t1", "t2", "t3", "t4", "t5",
230        "t6", "t7", "t8", "t9"
231  );
232}
233
234void ScaleRowDown4Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
235                            uint8* dst, int dst_width) {
236  intptr_t stride = src_stride;
237  const uint8* s1 = src_ptr + stride;
238  const uint8* s2 = s1 + stride;
239  const uint8* s3 = s2 + stride;
240
241  __asm__ __volatile__ (
242      ".set push                                  \n"
243      ".set noreorder                             \n"
244
245      "srl           $t9, %[dst_width], 1         \n"
246      "andi          $t8, %[dst_width], 1         \n"
247
248     "1:                                          \n"
249      "lw            $t0, 0(%[src_ptr])           \n"  // |3|2|1|0|
250      "lw            $t1, 0(%[s1])                \n"  // |7|6|5|4|
251      "lw            $t2, 0(%[s2])                \n"  // |11|10|9|8|
252      "lw            $t3, 0(%[s3])                \n"  // |15|14|13|12|
253      "lw            $t4, 4(%[src_ptr])           \n"  // |19|18|17|16|
254      "lw            $t5, 4(%[s1])                \n"  // |23|22|21|20|
255      "lw            $t6, 4(%[s2])                \n"  // |27|26|25|24|
256      "lw            $t7, 4(%[s3])                \n"  // |31|30|29|28|
257      "raddu.w.qb    $t0, $t0                     \n"  // |3 + 2 + 1 + 0|
258      "raddu.w.qb    $t1, $t1                     \n"  // |7 + 6 + 5 + 4|
259      "raddu.w.qb    $t2, $t2                     \n"  // |11 + 10 + 9 + 8|
260      "raddu.w.qb    $t3, $t3                     \n"  // |15 + 14 + 13 + 12|
261      "raddu.w.qb    $t4, $t4                     \n"  // |19 + 18 + 17 + 16|
262      "raddu.w.qb    $t5, $t5                     \n"  // |23 + 22 + 21 + 20|
263      "raddu.w.qb    $t6, $t6                     \n"  // |27 + 26 + 25 + 24|
264      "raddu.w.qb    $t7, $t7                     \n"  // |31 + 30 + 29 + 28|
265      "add           $t0, $t0, $t1                \n"
266      "add           $t1, $t2, $t3                \n"
267      "add           $t0, $t0, $t1                \n"
268      "add           $t4, $t4, $t5                \n"
269      "add           $t6, $t6, $t7                \n"
270      "add           $t4, $t4, $t6                \n"
271      "shra_r.w      $t0, $t0, 4                  \n"
272      "shra_r.w      $t4, $t4, 4                  \n"
273      "sb            $t0, 0(%[dst])               \n"
274      "sb            $t4, 1(%[dst])               \n"
275      "addiu         %[src_ptr], %[src_ptr], 8    \n"
276      "addiu         %[s1], %[s1], 8              \n"
277      "addiu         %[s2], %[s2], 8              \n"
278      "addiu         %[s3], %[s3], 8              \n"
279      "addiu         $t9, $t9, -1                 \n"
280      "bgtz          $t9, 1b                      \n"
281      " addiu        %[dst], %[dst], 2            \n"
282      "beqz          $t8, 2f                      \n"
283      " nop                                       \n"
284
285      "lw            $t0, 0(%[src_ptr])           \n"  // |3|2|1|0|
286      "lw            $t1, 0(%[s1])                \n"  // |7|6|5|4|
287      "lw            $t2, 0(%[s2])                \n"  // |11|10|9|8|
288      "lw            $t3, 0(%[s3])                \n"  // |15|14|13|12|
289      "raddu.w.qb    $t0, $t0                     \n"  // |3 + 2 + 1 + 0|
290      "raddu.w.qb    $t1, $t1                     \n"  // |7 + 6 + 5 + 4|
291      "raddu.w.qb    $t2, $t2                     \n"  // |11 + 10 + 9 + 8|
292      "raddu.w.qb    $t3, $t3                     \n"  // |15 + 14 + 13 + 12|
293      "add           $t0, $t0, $t1                \n"
294      "add           $t1, $t2, $t3                \n"
295      "add           $t0, $t0, $t1                \n"
296      "shra_r.w      $t0, $t0, 4                  \n"
297      "sb            $t0, 0(%[dst])               \n"
298
299      "2:                                         \n"
300      ".set pop                                   \n"
301
302      : [src_ptr] "+r" (src_ptr),
303        [dst] "+r" (dst),
304        [s1] "+r" (s1),
305        [s2] "+r" (s2),
306        [s3] "+r" (s3)
307      : [dst_width] "r" (dst_width)
308      : "t0", "t1", "t2", "t3", "t4", "t5",
309        "t6","t7", "t8", "t9"
310  );
311}
312
313void ScaleRowDown34_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
314                          uint8* dst, int dst_width) {
315  __asm__ __volatile__ (
316      ".set push                                          \n"
317      ".set noreorder                                     \n"
318    "1:                                                   \n"
319      "lw              $t1, 0(%[src_ptr])                 \n"  // |3|2|1|0|
320      "lw              $t2, 4(%[src_ptr])                 \n"  // |7|6|5|4|
321      "lw              $t3, 8(%[src_ptr])                 \n"  // |11|10|9|8|
322      "lw              $t4, 12(%[src_ptr])                \n"  // |15|14|13|12|
323      "lw              $t5, 16(%[src_ptr])                \n"  // |19|18|17|16|
324      "lw              $t6, 20(%[src_ptr])                \n"  // |23|22|21|20|
325      "lw              $t7, 24(%[src_ptr])                \n"  // |27|26|25|24|
326      "lw              $t8, 28(%[src_ptr])                \n"  // |31|30|29|28|
327      "precrq.qb.ph    $t0, $t2, $t4                      \n"  // |7|5|15|13|
328      "precrq.qb.ph    $t9, $t6, $t8                      \n"  // |23|21|31|30|
329      "addiu           %[dst_width], %[dst_width], -24    \n"
330      "ins             $t1, $t1, 8, 16                    \n"  // |3|1|0|X|
331      "ins             $t4, $t0, 8, 16                    \n"  // |X|15|13|12|
332      "ins             $t5, $t5, 8, 16                    \n"  // |19|17|16|X|
333      "ins             $t8, $t9, 8, 16                    \n"  // |X|31|29|28|
334      "addiu           %[src_ptr], %[src_ptr], 32         \n"
335      "packrl.ph       $t0, $t3, $t0                      \n"  // |9|8|7|5|
336      "packrl.ph       $t9, $t7, $t9                      \n"  // |25|24|23|21|
337      "prepend         $t1, $t2, 8                        \n"  // |4|3|1|0|
338      "prepend         $t3, $t4, 24                       \n"  // |15|13|12|11|
339      "prepend         $t5, $t6, 8                        \n"  // |20|19|17|16|
340      "prepend         $t7, $t8, 24                       \n"  // |31|29|28|27|
341      "sw              $t1, 0(%[dst])                     \n"
342      "sw              $t0, 4(%[dst])                     \n"
343      "sw              $t3, 8(%[dst])                     \n"
344      "sw              $t5, 12(%[dst])                    \n"
345      "sw              $t9, 16(%[dst])                    \n"
346      "sw              $t7, 20(%[dst])                    \n"
347      "bnez            %[dst_width], 1b                   \n"
348      " addiu          %[dst], %[dst], 24                 \n"
349      ".set pop                                           \n"
350      : [src_ptr] "+r" (src_ptr),
351        [dst] "+r" (dst),
352        [dst_width] "+r" (dst_width)
353      :
354      : "t0", "t1", "t2", "t3", "t4", "t5",
355        "t6","t7", "t8", "t9"
356  );
357}
358
359void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
360                                uint8* d, int dst_width) {
361  __asm__ __volatile__ (
362      ".set push                                         \n"
363      ".set noreorder                                    \n"
364      "repl.ph           $t3, 3                          \n"  // 0x00030003
365
366    "1:                                                  \n"
367      "lw                $t0, 0(%[src_ptr])              \n"  // |S3|S2|S1|S0|
368      "lwx               $t1, %[src_stride](%[src_ptr])  \n"  // |T3|T2|T1|T0|
369      "rotr              $t2, $t0, 8                     \n"  // |S0|S3|S2|S1|
370      "rotr              $t6, $t1, 8                     \n"  // |T0|T3|T2|T1|
371      "muleu_s.ph.qbl    $t4, $t2, $t3                   \n"  // |S0*3|S3*3|
372      "muleu_s.ph.qbl    $t5, $t6, $t3                   \n"  // |T0*3|T3*3|
373      "andi              $t0, $t2, 0xFFFF                \n"  // |0|0|S2|S1|
374      "andi              $t1, $t6, 0xFFFF                \n"  // |0|0|T2|T1|
375      "raddu.w.qb        $t0, $t0                        \n"
376      "raddu.w.qb        $t1, $t1                        \n"
377      "shra_r.w          $t0, $t0, 1                     \n"
378      "shra_r.w          $t1, $t1, 1                     \n"
379      "preceu.ph.qbr     $t2, $t2                        \n"  // |0|S2|0|S1|
380      "preceu.ph.qbr     $t6, $t6                        \n"  // |0|T2|0|T1|
381      "rotr              $t2, $t2, 16                    \n"  // |0|S1|0|S2|
382      "rotr              $t6, $t6, 16                    \n"  // |0|T1|0|T2|
383      "addu.ph           $t2, $t2, $t4                   \n"
384      "addu.ph           $t6, $t6, $t5                   \n"
385      "sll               $t5, $t0, 1                     \n"
386      "add               $t0, $t5, $t0                   \n"
387      "shra_r.ph         $t2, $t2, 2                     \n"
388      "shra_r.ph         $t6, $t6, 2                     \n"
389      "shll.ph           $t4, $t2, 1                     \n"
390      "addq.ph           $t4, $t4, $t2                   \n"
391      "addu              $t0, $t0, $t1                   \n"
392      "addiu             %[src_ptr], %[src_ptr], 4       \n"
393      "shra_r.w          $t0, $t0, 2                     \n"
394      "addu.ph           $t6, $t6, $t4                   \n"
395      "shra_r.ph         $t6, $t6, 2                     \n"
396      "srl               $t1, $t6, 16                    \n"
397      "addiu             %[dst_width], %[dst_width], -3  \n"
398      "sb                $t1, 0(%[d])                    \n"
399      "sb                $t0, 1(%[d])                    \n"
400      "sb                $t6, 2(%[d])                    \n"
401      "bgtz              %[dst_width], 1b                \n"
402      " addiu            %[d], %[d], 3                   \n"
403    "3:                                                  \n"
404      ".set pop                                          \n"
405      : [src_ptr] "+r" (src_ptr),
406        [src_stride] "+r" (src_stride),
407        [d] "+r" (d),
408        [dst_width] "+r" (dst_width)
409      :
410      : "t0", "t1", "t2", "t3",
411        "t4", "t5", "t6"
412  );
413}
414
415void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
416                                uint8* d, int dst_width) {
417  __asm__ __volatile__ (
418      ".set push                                           \n"
419      ".set noreorder                                      \n"
420      "repl.ph           $t2, 3                            \n"  // 0x00030003
421
422    "1:                                                    \n"
423      "lw                $t0, 0(%[src_ptr])                \n"  // |S3|S2|S1|S0|
424      "lwx               $t1, %[src_stride](%[src_ptr])    \n"  // |T3|T2|T1|T0|
425      "rotr              $t4, $t0, 8                       \n"  // |S0|S3|S2|S1|
426      "rotr              $t6, $t1, 8                       \n"  // |T0|T3|T2|T1|
427      "muleu_s.ph.qbl    $t3, $t4, $t2                     \n"  // |S0*3|S3*3|
428      "muleu_s.ph.qbl    $t5, $t6, $t2                     \n"  // |T0*3|T3*3|
429      "andi              $t0, $t4, 0xFFFF                  \n"  // |0|0|S2|S1|
430      "andi              $t1, $t6, 0xFFFF                  \n"  // |0|0|T2|T1|
431      "raddu.w.qb        $t0, $t0                          \n"
432      "raddu.w.qb        $t1, $t1                          \n"
433      "shra_r.w          $t0, $t0, 1                       \n"
434      "shra_r.w          $t1, $t1, 1                       \n"
435      "preceu.ph.qbr     $t4, $t4                          \n"  // |0|S2|0|S1|
436      "preceu.ph.qbr     $t6, $t6                          \n"  // |0|T2|0|T1|
437      "rotr              $t4, $t4, 16                      \n"  // |0|S1|0|S2|
438      "rotr              $t6, $t6, 16                      \n"  // |0|T1|0|T2|
439      "addu.ph           $t4, $t4, $t3                     \n"
440      "addu.ph           $t6, $t6, $t5                     \n"
441      "shra_r.ph         $t6, $t6, 2                       \n"
442      "shra_r.ph         $t4, $t4, 2                       \n"
443      "addu.ph           $t6, $t6, $t4                     \n"
444      "addiu             %[src_ptr], %[src_ptr], 4         \n"
445      "shra_r.ph         $t6, $t6, 1                       \n"
446      "addu              $t0, $t0, $t1                     \n"
447      "addiu             %[dst_width], %[dst_width], -3    \n"
448      "shra_r.w          $t0, $t0, 1                       \n"
449      "srl               $t1, $t6, 16                      \n"
450      "sb                $t1, 0(%[d])                      \n"
451      "sb                $t0, 1(%[d])                      \n"
452      "sb                $t6, 2(%[d])                      \n"
453      "bgtz              %[dst_width], 1b                  \n"
454      " addiu            %[d], %[d], 3                     \n"
455    "3:                                                    \n"
456      ".set pop                                            \n"
457      : [src_ptr] "+r" (src_ptr),
458        [src_stride] "+r" (src_stride),
459        [d] "+r" (d),
460        [dst_width] "+r" (dst_width)
461      :
462      : "t0", "t1", "t2", "t3",
463        "t4", "t5", "t6"
464  );
465}
466
467void ScaleRowDown38_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
468                          uint8* dst, int dst_width) {
469  __asm__ __volatile__ (
470      ".set push                                     \n"
471      ".set noreorder                                \n"
472
473    "1:                                              \n"
474      "lw         $t0, 0(%[src_ptr])                 \n"  // |3|2|1|0|
475      "lw         $t1, 4(%[src_ptr])                 \n"  // |7|6|5|4|
476      "lw         $t2, 8(%[src_ptr])                 \n"  // |11|10|9|8|
477      "lw         $t3, 12(%[src_ptr])                \n"  // |15|14|13|12|
478      "lw         $t4, 16(%[src_ptr])                \n"  // |19|18|17|16|
479      "lw         $t5, 20(%[src_ptr])                \n"  // |23|22|21|20|
480      "lw         $t6, 24(%[src_ptr])                \n"  // |27|26|25|24|
481      "lw         $t7, 28(%[src_ptr])                \n"  // |31|30|29|28|
482      "wsbh       $t0, $t0                           \n"  // |2|3|0|1|
483      "wsbh       $t6, $t6                           \n"  // |26|27|24|25|
484      "srl        $t0, $t0, 8                        \n"  // |X|2|3|0|
485      "srl        $t3, $t3, 16                       \n"  // |X|X|15|14|
486      "srl        $t5, $t5, 16                       \n"  // |X|X|23|22|
487      "srl        $t7, $t7, 16                       \n"  // |X|X|31|30|
488      "ins        $t1, $t2, 24, 8                    \n"  // |8|6|5|4|
489      "ins        $t6, $t5, 0, 8                     \n"  // |26|27|24|22|
490      "ins        $t1, $t0, 0, 16                    \n"  // |8|6|3|0|
491      "ins        $t6, $t7, 24, 8                    \n"  // |30|27|24|22|
492      "prepend    $t2, $t3, 24                       \n"  // |X|15|14|11|
493      "ins        $t4, $t4, 16, 8                    \n"  // |19|16|17|X|
494      "ins        $t4, $t2, 0, 16                    \n"  // |19|16|14|11|
495      "addiu      %[src_ptr], %[src_ptr], 32         \n"
496      "addiu      %[dst_width], %[dst_width], -12    \n"
497      "addiu      $t8,%[dst_width], -12              \n"
498      "sw         $t1, 0(%[dst])                     \n"
499      "sw         $t4, 4(%[dst])                     \n"
500      "sw         $t6, 8(%[dst])                     \n"
501      "bgez       $t8, 1b                            \n"
502      " addiu     %[dst], %[dst], 12                 \n"
503      ".set pop                                      \n"
504      : [src_ptr] "+r" (src_ptr),
505        [dst] "+r" (dst),
506        [dst_width] "+r" (dst_width)
507      :
508      : "t0", "t1", "t2", "t3", "t4",
509        "t5", "t6", "t7", "t8"
510  );
511}
512
513void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
514                                uint8* dst_ptr, int dst_width) {
515  intptr_t stride = src_stride;
516  const uint8* t = src_ptr + stride;
517  const int c = 0x2AAA;
518
519  __asm__ __volatile__ (
520      ".set push                                         \n"
521      ".set noreorder                                    \n"
522
523    "1:                                                  \n"
524      "lw              $t0, 0(%[src_ptr])                \n"  // |S3|S2|S1|S0|
525      "lw              $t1, 4(%[src_ptr])                \n"  // |S7|S6|S5|S4|
526      "lw              $t2, 0(%[t])                      \n"  // |T3|T2|T1|T0|
527      "lw              $t3, 4(%[t])                      \n"  // |T7|T6|T5|T4|
528      "rotr            $t1, $t1, 16                      \n"  // |S5|S4|S7|S6|
529      "packrl.ph       $t4, $t1, $t3                     \n"  // |S7|S6|T7|T6|
530      "packrl.ph       $t5, $t3, $t1                     \n"  // |T5|T4|S5|S4|
531      "raddu.w.qb      $t4, $t4                          \n"  // S7+S6+T7+T6
532      "raddu.w.qb      $t5, $t5                          \n"  // T5+T4+S5+S4
533      "precrq.qb.ph    $t6, $t0, $t2                     \n"  // |S3|S1|T3|T1|
534      "precrq.qb.ph    $t6, $t6, $t6                     \n"  // |S3|T3|S3|T3|
535      "srl             $t4, $t4, 2                       \n"  // t4 / 4
536      "srl             $t6, $t6, 16                      \n"  // |0|0|S3|T3|
537      "raddu.w.qb      $t6, $t6                          \n"  // 0+0+S3+T3
538      "addu            $t6, $t5, $t6                     \n"
539      "mul             $t6, $t6, %[c]                    \n"  // t6 * 0x2AAA
540      "sll             $t0, $t0, 8                       \n"  // |S2|S1|S0|0|
541      "sll             $t2, $t2, 8                       \n"  // |T2|T1|T0|0|
542      "raddu.w.qb      $t0, $t0                          \n"  // S2+S1+S0+0
543      "raddu.w.qb      $t2, $t2                          \n"  // T2+T1+T0+0
544      "addu            $t0, $t0, $t2                     \n"
545      "mul             $t0, $t0, %[c]                    \n"  // t0 * 0x2AAA
546      "addiu           %[src_ptr], %[src_ptr], 8         \n"
547      "addiu           %[t], %[t], 8                     \n"
548      "addiu           %[dst_width], %[dst_width], -3    \n"
549      "addiu           %[dst_ptr], %[dst_ptr], 3         \n"
550      "srl             $t6, $t6, 16                      \n"
551      "srl             $t0, $t0, 16                      \n"
552      "sb              $t4, -1(%[dst_ptr])               \n"
553      "sb              $t6, -2(%[dst_ptr])               \n"
554      "bgtz            %[dst_width], 1b                  \n"
555      " sb             $t0, -3(%[dst_ptr])               \n"
556      ".set pop                                          \n"
557      : [src_ptr] "+r" (src_ptr),
558        [dst_ptr] "+r" (dst_ptr),
559        [t] "+r" (t),
560        [dst_width] "+r" (dst_width)
561      : [c] "r" (c)
562      : "t0", "t1", "t2", "t3", "t4", "t5", "t6"
563  );
564}
565
566void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
567                                ptrdiff_t src_stride,
568                                uint8* dst_ptr, int dst_width) {
569  intptr_t stride = src_stride;
570  const uint8* s1 = src_ptr + stride;
571  stride += stride;
572  const uint8* s2 = src_ptr + stride;
573  const int c1 = 0x1C71;
574  const int c2 = 0x2AAA;
575
576  __asm__ __volatile__ (
577      ".set push                                         \n"
578      ".set noreorder                                    \n"
579
580    "1:                                                  \n"
581      "lw              $t0, 0(%[src_ptr])                \n"  // |S3|S2|S1|S0|
582      "lw              $t1, 4(%[src_ptr])                \n"  // |S7|S6|S5|S4|
583      "lw              $t2, 0(%[s1])                     \n"  // |T3|T2|T1|T0|
584      "lw              $t3, 4(%[s1])                     \n"  // |T7|T6|T5|T4|
585      "lw              $t4, 0(%[s2])                     \n"  // |R3|R2|R1|R0|
586      "lw              $t5, 4(%[s2])                     \n"  // |R7|R6|R5|R4|
587      "rotr            $t1, $t1, 16                      \n"  // |S5|S4|S7|S6|
588      "packrl.ph       $t6, $t1, $t3                     \n"  // |S7|S6|T7|T6|
589      "raddu.w.qb      $t6, $t6                          \n"  // S7+S6+T7+T6
590      "packrl.ph       $t7, $t3, $t1                     \n"  // |T5|T4|S5|S4|
591      "raddu.w.qb      $t7, $t7                          \n"  // T5+T4+S5+S4
592      "sll             $t8, $t5, 16                      \n"  // |R5|R4|0|0|
593      "raddu.w.qb      $t8, $t8                          \n"  // R5+R4
594      "addu            $t7, $t7, $t8                     \n"
595      "srl             $t8, $t5, 16                      \n"  // |0|0|R7|R6|
596      "raddu.w.qb      $t8, $t8                          \n"  // R7 + R6
597      "addu            $t6, $t6, $t8                     \n"
598      "mul             $t6, $t6, %[c2]                   \n"  // t6 * 0x2AAA
599      "precrq.qb.ph    $t8, $t0, $t2                     \n"  // |S3|S1|T3|T1|
600      "precrq.qb.ph    $t8, $t8, $t4                     \n"  // |S3|T3|R3|R1|
601      "srl             $t8, $t8, 8                       \n"  // |0|S3|T3|R3|
602      "raddu.w.qb      $t8, $t8                          \n"  // S3 + T3 + R3
603      "addu            $t7, $t7, $t8                     \n"
604      "mul             $t7, $t7, %[c1]                   \n"  // t7 * 0x1C71
605      "sll             $t0, $t0, 8                       \n"  // |S2|S1|S0|0|
606      "sll             $t2, $t2, 8                       \n"  // |T2|T1|T0|0|
607      "sll             $t4, $t4, 8                       \n"  // |R2|R1|R0|0|
608      "raddu.w.qb      $t0, $t0                          \n"
609      "raddu.w.qb      $t2, $t2                          \n"
610      "raddu.w.qb      $t4, $t4                          \n"
611      "addu            $t0, $t0, $t2                     \n"
612      "addu            $t0, $t0, $t4                     \n"
613      "mul             $t0, $t0, %[c1]                   \n"  // t0 * 0x1C71
614      "addiu           %[src_ptr], %[src_ptr], 8         \n"
615      "addiu           %[s1], %[s1], 8                   \n"
616      "addiu           %[s2], %[s2], 8                   \n"
617      "addiu           %[dst_width], %[dst_width], -3    \n"
618      "addiu           %[dst_ptr], %[dst_ptr], 3         \n"
619      "srl             $t6, $t6, 16                      \n"
620      "srl             $t7, $t7, 16                      \n"
621      "srl             $t0, $t0, 16                      \n"
622      "sb              $t6, -1(%[dst_ptr])               \n"
623      "sb              $t7, -2(%[dst_ptr])               \n"
624      "bgtz            %[dst_width], 1b                  \n"
625      " sb             $t0, -3(%[dst_ptr])               \n"
626      ".set pop                                          \n"
627      : [src_ptr] "+r" (src_ptr),
628        [dst_ptr] "+r" (dst_ptr),
629        [s1] "+r" (s1),
630        [s2] "+r" (s2),
631        [dst_width] "+r" (dst_width)
632      : [c1] "r" (c1), [c2] "r" (c2)
633      : "t0", "t1", "t2", "t3", "t4",
634        "t5", "t6", "t7", "t8"
635  );
636}
637
638#endif  // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
639
640#ifdef __cplusplus
641}  // extern "C"
642}  // namespace libyuv
643#endif
644
645