1 /*
2  * VC-1 and WMV3 - DSP functions Loongson MMI-optimized
3  *
4  * Copyright (c) 2016 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
5  *
6  * This file is part of FFmpeg.
7  *
8  * FFmpeg is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * FFmpeg is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with FFmpeg; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include "libavutil/avassert.h"
24 #include "libavcodec/vc1dsp.h"
25 #include "constants.h"
26 #include "vc1dsp_mips.h"
27 #include "hpeldsp_mips.h"
28 #include "libavutil/mips/mmiutils.h"
29 
30 #define VC1_INV_TRANCS_8_TYPE1(o1, o2, r1, r2, r3, r4, c0)                  \
31         "li         %[tmp0],    "#r1"                                 \n\t" \
32         "mtc1       %[tmp0],    %[ftmp13]                             \n\t" \
33         "punpcklwd  %[ftmp13],  %[ftmp13],  %[ftmp13]                 \n\t" \
34         "li         %[tmp0],    "#r2"                                 \n\t" \
35         "mtc1       %[tmp0],    %[ftmp14]                             \n\t" \
36         "punpcklwd  %[ftmp14],  %[ftmp14],  %[ftmp14]                 \n\t" \
37         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp13]                 \n\t" \
38         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp14]                 \n\t" \
39         "paddw      %[ftmp1],   %[ftmp1],   %[ftmp2]                  \n\t" \
40         "pmaddhw    %[ftmp2],   %[ftmp6],   %[ftmp13]                 \n\t" \
41         "pmaddhw    %[ftmp3],   %[ftmp8],   %[ftmp14]                 \n\t" \
42         "paddw      %[ftmp2],   %[ftmp2],   %[ftmp3]                  \n\t" \
43                                                                             \
44         "li         %[tmp0],    "#r3"                                 \n\t" \
45         "mtc1       %[tmp0],    %[ftmp13]                             \n\t" \
46         "punpcklwd  %[ftmp13],  %[ftmp13],  %[ftmp13]                 \n\t" \
47         "li         %[tmp0],    "#r4"                                 \n\t" \
48         "mtc1       %[tmp0],    %[ftmp14]                             \n\t" \
49         "punpcklwd  %[ftmp14],  %[ftmp14],  %[ftmp14]                 \n\t" \
50         "pmaddhw    %[ftmp3],   %[ftmp9],   %[ftmp13]                 \n\t" \
51         "pmaddhw    %[ftmp4],   %[ftmp11],  %[ftmp14]                 \n\t" \
52         "paddw      %[ftmp3],   %[ftmp3],   %[ftmp4]                  \n\t" \
53         "pmaddhw    %[ftmp4],   %[ftmp10],  %[ftmp13]                 \n\t" \
54         "pmaddhw    %[ftmp13],  %[ftmp12],  %[ftmp14]                 \n\t" \
55         "paddw      %[ftmp4],   %[ftmp4],   %[ftmp13]                 \n\t" \
56                                                                             \
57         "paddw      %[ftmp1],   %[ftmp1],   "#c0"                     \n\t" \
58         "paddw      %[ftmp2],   %[ftmp2],   "#c0"                     \n\t" \
59         "paddw      %[ftmp13],  %[ftmp1],   %[ftmp3]                  \n\t" \
60         "psubw      %[ftmp14],  %[ftmp1],   %[ftmp3]                  \n\t" \
61         "paddw      %[ftmp1],   %[ftmp2],   %[ftmp4]                  \n\t" \
62         "psubw      %[ftmp3],   %[ftmp2],   %[ftmp4]                  \n\t" \
63         "psraw      %[ftmp13],  %[ftmp13],  %[ftmp0]                  \n\t" \
64         "psraw      %[ftmp1],   %[ftmp1],   %[ftmp0]                  \n\t" \
65         "psraw      %[ftmp14],  %[ftmp14],  %[ftmp0]                  \n\t" \
66         "psraw      %[ftmp3],   %[ftmp3],   %[ftmp0]                  \n\t" \
67         "punpcklhw  %[ftmp2],   %[ftmp13],  %[ftmp1]                  \n\t" \
68         "punpckhhw  %[ftmp4],   %[ftmp13],  %[ftmp1]                  \n\t" \
69         "punpcklhw  "#o1",      %[ftmp2],   %[ftmp4]                  \n\t" \
70         "punpcklhw  %[ftmp2],   %[ftmp14],  %[ftmp3]                  \n\t" \
71         "punpckhhw  %[ftmp4],   %[ftmp14],  %[ftmp3]                  \n\t" \
72         "punpcklhw  "#o2",      %[ftmp2],   %[ftmp4]                  \n\t"
73 
74 #define VC1_INV_TRANCS_8_TYPE2(o1, o2, r1, r2, r3, r4, c0, c1)              \
75         "li         %[tmp0],    "#r1"                                 \n\t" \
76         "mtc1       %[tmp0],    %[ftmp13]                             \n\t" \
77         "punpcklwd  %[ftmp13],  %[ftmp13],  %[ftmp13]                 \n\t" \
78         "li         %[tmp0],    "#r2"                                 \n\t" \
79         "mtc1       %[tmp0],    %[ftmp14]                             \n\t" \
80         "punpcklwd  %[ftmp14],  %[ftmp14],  %[ftmp14]                 \n\t" \
81         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp13]                 \n\t" \
82         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp14]                 \n\t" \
83         "paddw      %[ftmp1],   %[ftmp1],   %[ftmp2]                  \n\t" \
84         "pmaddhw    %[ftmp2],   %[ftmp6],   %[ftmp13]                 \n\t" \
85         "pmaddhw    %[ftmp3],   %[ftmp8],   %[ftmp14]                 \n\t" \
86         "paddw      %[ftmp2],   %[ftmp2],   %[ftmp3]                  \n\t" \
87                                                                             \
88         "li         %[tmp0],    "#r3"                                 \n\t" \
89         "mtc1       %[tmp0],    %[ftmp13]                             \n\t" \
90         "punpcklwd  %[ftmp13],  %[ftmp13],  %[ftmp13]                 \n\t" \
91         "li         %[tmp0],    "#r4"                                 \n\t" \
92         "mtc1       %[tmp0],    %[ftmp14]                             \n\t" \
93         "punpcklwd  %[ftmp14],  %[ftmp14],  %[ftmp14]                 \n\t" \
94         "pmaddhw    %[ftmp3],   %[ftmp9],   %[ftmp13]                 \n\t" \
95         "pmaddhw    %[ftmp4],   %[ftmp11],  %[ftmp14]                 \n\t" \
96         "paddw      %[ftmp3],   %[ftmp3],   %[ftmp4]                  \n\t" \
97         "pmaddhw    %[ftmp4],   %[ftmp10],  %[ftmp13]                 \n\t" \
98         "pmaddhw    %[ftmp13],  %[ftmp12],  %[ftmp14]                 \n\t" \
99         "paddw      %[ftmp4],   %[ftmp4],   %[ftmp13]                 \n\t" \
100                                                                             \
101         "paddw      %[ftmp13],  %[ftmp1],   %[ftmp3]                  \n\t" \
102         "psubw      %[ftmp14],  %[ftmp1],   %[ftmp3]                  \n\t" \
103         "paddw      %[ftmp14],  %[ftmp14],  "#c1"                     \n\t" \
104         "paddw      %[ftmp1],   %[ftmp2],   %[ftmp4]                  \n\t" \
105         "psubw      %[ftmp3],   %[ftmp2],   %[ftmp4]                  \n\t" \
106         "paddw      %[ftmp3],   %[ftmp3],   "#c1"                     \n\t" \
107         "paddw      %[ftmp13],  %[ftmp13],  "#c0"                     \n\t" \
108         "paddw      %[ftmp14],  %[ftmp14],  "#c0"                     \n\t" \
109         "paddw      %[ftmp1],   %[ftmp1],   "#c0"                     \n\t" \
110         "paddw      %[ftmp3],   %[ftmp3],   "#c0"                     \n\t" \
111         "psraw      %[ftmp13],  %[ftmp13],  %[ftmp0]                  \n\t" \
112         "psraw      %[ftmp1],   %[ftmp1],   %[ftmp0]                  \n\t" \
113         "psraw      %[ftmp14],  %[ftmp14],  %[ftmp0]                  \n\t" \
114         "psraw      %[ftmp3],   %[ftmp3],   %[ftmp0]                  \n\t" \
115         "punpcklhw  %[ftmp2],   %[ftmp13],  %[ftmp1]                  \n\t" \
116         "punpckhhw  %[ftmp4],   %[ftmp13],  %[ftmp1]                  \n\t" \
117         "punpcklhw  "#o1",      %[ftmp2],   %[ftmp4]                  \n\t" \
118         "punpcklhw  %[ftmp2],   %[ftmp14],  %[ftmp3]                  \n\t" \
119         "punpckhhw  %[ftmp4],   %[ftmp14],  %[ftmp3]                  \n\t" \
120         "punpcklhw  "#o2",      %[ftmp2],   %[ftmp4]                  \n\t"
121 
122 /* Do inverse transform on 8x8 block */
ff_vc1_inv_trans_8x8_dc_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)123 void ff_vc1_inv_trans_8x8_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
124 {
125     int dc = block[0];
126     double ftmp[9];
127     mips_reg addr[1];
128     int count;
129 
130     dc = (3 * dc +  1) >> 1;
131     dc = (3 * dc + 16) >> 5;
132 
133     __asm__ volatile(
134         "xor        %[ftmp0],   %[ftmp0],       %[ftmp0]                \n\t"
135         "pshufh     %[dc],      %[dc],          %[ftmp0]                \n\t"
136         "li         %[count],   0x02                                    \n\t"
137 
138         "1:                                                             \n\t"
139         MMI_LDC1(%[ftmp1], %[dest], 0x00)
140         PTR_ADDU   "%[addr0],   %[dest],        %[linesize]             \n\t"
141         MMI_LDC1(%[ftmp2], %[addr0], 0x00)
142         PTR_ADDU   "%[addr0],   %[addr0],       %[linesize]             \n\t"
143         MMI_LDC1(%[ftmp3], %[addr0], 0x00)
144         PTR_ADDU   "%[addr0],   %[addr0],       %[linesize]             \n\t"
145         MMI_LDC1(%[ftmp4], %[addr0], 0x00)
146 
147         "punpckhbh  %[ftmp5],   %[ftmp1],       %[ftmp0]                \n\t"
148         "punpcklbh  %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
149         "punpckhbh  %[ftmp6],   %[ftmp2],       %[ftmp0]                \n\t"
150         "punpcklbh  %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
151         "punpckhbh  %[ftmp7],   %[ftmp3],       %[ftmp0]                \n\t"
152         "punpcklbh  %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
153         "punpckhbh  %[ftmp8],   %[ftmp4],       %[ftmp0]                \n\t"
154         "punpcklbh  %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
155 
156         "paddsh     %[ftmp1],   %[ftmp1],       %[dc]                   \n\t"
157         "paddsh     %[ftmp2],   %[ftmp2],       %[dc]                   \n\t"
158         "paddsh     %[ftmp3],   %[ftmp3],       %[dc]                   \n\t"
159         "paddsh     %[ftmp4],   %[ftmp4],       %[dc]                   \n\t"
160         "paddsh     %[ftmp5],   %[ftmp5],       %[dc]                   \n\t"
161         "paddsh     %[ftmp6],   %[ftmp6],       %[dc]                   \n\t"
162         "paddsh     %[ftmp7],   %[ftmp7],       %[dc]                   \n\t"
163         "paddsh     %[ftmp8],   %[ftmp8],       %[dc]                   \n\t"
164 
165         "packushb   %[ftmp1],   %[ftmp1],       %[ftmp5]                \n\t"
166         "packushb   %[ftmp2],   %[ftmp2],       %[ftmp6]                \n\t"
167         "packushb   %[ftmp3],   %[ftmp3],       %[ftmp7]                \n\t"
168         "packushb   %[ftmp4],   %[ftmp4],       %[ftmp8]                \n\t"
169 
170         MMI_SDC1(%[ftmp1], %[dest], 0x00)
171         PTR_ADDU   "%[addr0],   %[dest],        %[linesize]             \n\t"
172         MMI_SDC1(%[ftmp2], %[addr0], 0x00)
173         PTR_ADDU   "%[addr0],   %[addr0],       %[linesize]             \n\t"
174         MMI_SDC1(%[ftmp3], %[addr0], 0x00)
175         PTR_ADDU   "%[addr0],   %[addr0],       %[linesize]             \n\t"
176         MMI_SDC1(%[ftmp4], %[addr0], 0x00)
177 
178         "addiu      %[count],   %[count],       -0x01                   \n\t"
179         PTR_ADDU   "%[dest],    %[addr0],       %[linesize]             \n\t"
180         "bnez       %[count],   1b                                      \n\t"
181         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
182           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
183           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
184           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
185           [ftmp8]"=&f"(ftmp[8]),
186           [addr0]"=&r"(addr[0]),
187           [count]"=&r"(count),          [dest]"+&r"(dest)
188         : [linesize]"r"((mips_reg)linesize),
189           [dc]"f"(dc)
190         : "memory"
191     );
192 }
193 
194 #if _MIPS_SIM != _ABIO32
ff_vc1_inv_trans_8x8_mmi(int16_t block[64])195 void ff_vc1_inv_trans_8x8_mmi(int16_t block[64])
196 {
197     DECLARE_ALIGNED(16, int16_t, temp[64]);
198     DECLARE_ALIGNED(8, const uint64_t, ff_pw_1_local) = {0x0000000100000001ULL};
199     DECLARE_ALIGNED(8, const uint64_t, ff_pw_4_local) = {0x0000000400000004ULL};
200     DECLARE_ALIGNED(8, const uint64_t, ff_pw_64_local)= {0x0000004000000040ULL};
201     double ftmp[23];
202     uint64_t tmp[1];
203 
204     __asm__ volatile (
205         /* 1st loop: start */
206         "li         %[tmp0],    0x03                                    \n\t"
207         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
208 
209        // 1st part
210         MMI_LDC1(%[ftmp1], %[block], 0x00)
211         MMI_LDC1(%[ftmp11], %[block], 0x10)
212         MMI_LDC1(%[ftmp2], %[block], 0x20)
213         MMI_LDC1(%[ftmp12], %[block], 0x30)
214         MMI_LDC1(%[ftmp3], %[block], 0x40)
215         MMI_LDC1(%[ftmp13], %[block], 0x50)
216         MMI_LDC1(%[ftmp4], %[block], 0x60)
217         MMI_LDC1(%[ftmp14], %[block], 0x70)
218         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
219         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
220         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
221         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
222 
223         "punpcklhw  %[ftmp9],  %[ftmp11],  %[ftmp12]                    \n\t"
224         "punpckhhw  %[ftmp10], %[ftmp11],  %[ftmp12]                    \n\t"
225         "punpcklhw  %[ftmp11], %[ftmp13],  %[ftmp14]                    \n\t"
226         "punpckhhw  %[ftmp12], %[ftmp13],  %[ftmp14]                    \n\t"
227 
228         /* ftmp15:dst03,dst02,dst01,dst00 ftmp22:dst73,dst72,dst71,dst70 */
229         VC1_INV_TRANCS_8_TYPE1(%[ftmp15], %[ftmp22], 0x0010000c, 0x0006000c,
230                                0x000f0010, 0x00040009, %[ff_pw_4])
231 
232         /* ftmp16:dst13,dst12,dst11,dst10 ftmp21:dst63,dst62,dst61,dst60 */
233         VC1_INV_TRANCS_8_TYPE1(%[ftmp16], %[ftmp21], 0x0006000c, 0xfff0fff4,
234                                0xfffc000f, 0xfff7fff0, %[ff_pw_4])
235 
236         /* ftmp17:dst23,dst22,dst21,dst20 ftmp20:dst53,dst52,dst51,dst50 */
237         VC1_INV_TRANCS_8_TYPE1(%[ftmp17], %[ftmp20], 0xfffa000c, 0x0010fff4,
238                                0xfff00009, 0x000f0004, %[ff_pw_4])
239 
240         /* ftmp18:dst33,dst32,dst31,dst30 ftmp19:dst43,dst42,dst41,dst40 */
241         VC1_INV_TRANCS_8_TYPE1(%[ftmp18], %[ftmp19], 0xfff0000c, 0xfffa000c,
242                                0xfff70004, 0xfff0000f, %[ff_pw_4])
243 
244         TRANSPOSE_4H(%[ftmp15], %[ftmp16], %[ftmp17], %[ftmp18],
245                      %[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4])
246 
247         TRANSPOSE_4H(%[ftmp19], %[ftmp20], %[ftmp21], %[ftmp22],
248                      %[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4])
249 
250         MMI_SDC1(%[ftmp15], %[temp], 0x00)
251         MMI_SDC1(%[ftmp19], %[temp], 0x08)
252         MMI_SDC1(%[ftmp16], %[temp], 0x10)
253         MMI_SDC1(%[ftmp20], %[temp], 0x18)
254         MMI_SDC1(%[ftmp17], %[temp], 0x20)
255         MMI_SDC1(%[ftmp21], %[temp], 0x28)
256         MMI_SDC1(%[ftmp18], %[temp], 0x30)
257         MMI_SDC1(%[ftmp22], %[temp], 0x38)
258 
259        // 2nd part
260         MMI_LDC1(%[ftmp1], %[block], 0x08)
261         MMI_LDC1(%[ftmp11], %[block], 0x18)
262         MMI_LDC1(%[ftmp2], %[block], 0x28)
263         MMI_LDC1(%[ftmp12], %[block], 0x38)
264         MMI_LDC1(%[ftmp3], %[block], 0x48)
265         MMI_LDC1(%[ftmp13], %[block], 0x58)
266         MMI_LDC1(%[ftmp4], %[block], 0x68)
267         MMI_LDC1(%[ftmp14], %[block], 0x78)
268         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
269         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
270         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
271         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
272 
273         "punpcklhw  %[ftmp9],   %[ftmp11],  %[ftmp12]                   \n\t"
274         "punpckhhw  %[ftmp10],  %[ftmp11],  %[ftmp12]                   \n\t"
275         "punpcklhw  %[ftmp11],  %[ftmp13],  %[ftmp14]                   \n\t"
276         "punpckhhw  %[ftmp12],  %[ftmp13],  %[ftmp14]                   \n\t"
277 
278         /* ftmp15:dst03,dst02,dst01,dst00 ftmp22:dst73,dst72,dst71,dst70 */
279         VC1_INV_TRANCS_8_TYPE1(%[ftmp15], %[ftmp22], 0x0010000c, 0x0006000c,
280                                0x000f0010, 0x00040009, %[ff_pw_4])
281 
282         /* ftmp16:dst13,dst12,dst11,dst10 ftmp21:dst63,dst62,dst61,dst60 */
283         VC1_INV_TRANCS_8_TYPE1(%[ftmp16], %[ftmp21], 0x0006000c, 0xfff0fff4,
284                                0xfffc000f, 0xfff7fff0, %[ff_pw_4])
285 
286         /* ftmp17:dst23,dst22,dst21,dst20 ftmp20:dst53,dst52,dst51,dst50 */
287         VC1_INV_TRANCS_8_TYPE1(%[ftmp17], %[ftmp20], 0xfffa000c, 0x0010fff4,
288                                0xfff00009, 0x000f0004, %[ff_pw_4])
289 
290         /* ftmp18:dst33,dst32,dst31,dst30 ftmp19:dst43,dst42,dst41,dst40 */
291         VC1_INV_TRANCS_8_TYPE1(%[ftmp18], %[ftmp19], 0xfff0000c, 0xfffa000c,
292                                0xfff70004, 0xfff0000f, %[ff_pw_4])
293 
294         TRANSPOSE_4H(%[ftmp15], %[ftmp16], %[ftmp17], %[ftmp18],
295                      %[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4])
296 
297         TRANSPOSE_4H(%[ftmp19], %[ftmp20], %[ftmp21], %[ftmp22],
298                      %[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4])
299 
300         MMI_SDC1(%[ftmp19], %[temp], 0x48)
301         MMI_SDC1(%[ftmp20], %[temp], 0x58)
302         MMI_SDC1(%[ftmp21], %[temp], 0x68)
303         MMI_SDC1(%[ftmp22], %[temp], 0x78)
304         /* 1st loop: end */
305 
306         /* 2nd loop: start */
307         "li         %[tmp0],    0x07                                    \n\t"
308         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
309 
310         // 1st part
311         MMI_LDC1(%[ftmp1], %[temp], 0x00)
312         MMI_LDC1(%[ftmp11], %[temp], 0x10)
313         MMI_LDC1(%[ftmp2], %[temp], 0x20)
314         MMI_LDC1(%[ftmp12], %[temp], 0x30)
315         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
316         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
317         "punpcklhw  %[ftmp7],   %[ftmp15],  %[ftmp17]                   \n\t"
318         "punpckhhw  %[ftmp8],   %[ftmp15],  %[ftmp17]                   \n\t"
319 
320         "punpcklhw  %[ftmp9],   %[ftmp11],  %[ftmp12]                   \n\t"
321         "punpckhhw  %[ftmp10],  %[ftmp11],  %[ftmp12]                   \n\t"
322         "punpcklhw  %[ftmp11],  %[ftmp16],  %[ftmp18]                   \n\t"
323         "punpckhhw  %[ftmp12],  %[ftmp16],  %[ftmp18]                   \n\t"
324 
325         /* ftmp15:dst03,dst02,dst01,dst00 ftmp22:dst73,dst72,dst71,dst70 */
326         VC1_INV_TRANCS_8_TYPE2(%[ftmp15], %[ftmp22], 0x0010000c, 0x0006000c,
327                                0x000f0010, 0x00040009, %[ff_pw_64], %[ff_pw_1])
328 
329         /* ftmp16:dst13,dst12,dst11,dst10 ftmp21:dst63,dst62,dst61,dst60 */
330         VC1_INV_TRANCS_8_TYPE2(%[ftmp16], %[ftmp21], 0x0006000c, 0xfff0fff4,
331                                0xfffc000f, 0xfff7fff0, %[ff_pw_64], %[ff_pw_1])
332 
333         /* ftmp17:dst23,dst22,dst21,dst20 ftmp20:dst53,dst52,dst51,dst50 */
334         VC1_INV_TRANCS_8_TYPE2(%[ftmp17], %[ftmp20], 0xfffa000c, 0x0010fff4,
335                                0xfff00009, 0x000f0004, %[ff_pw_64], %[ff_pw_1])
336 
337         /* ftmp18:dst33,dst32,dst31,dst30 ftmp19:dst43,dst42,dst41,dst40 */
338         VC1_INV_TRANCS_8_TYPE2(%[ftmp18], %[ftmp19], 0xfff0000c, 0xfffa000c,
339                                0xfff70004, 0xfff0000f, %[ff_pw_64], %[ff_pw_1])
340 
341         MMI_SDC1(%[ftmp15], %[block], 0x00)
342         MMI_SDC1(%[ftmp16], %[block], 0x10)
343         MMI_SDC1(%[ftmp17], %[block], 0x20)
344         MMI_SDC1(%[ftmp18], %[block], 0x30)
345         MMI_SDC1(%[ftmp19], %[block], 0x40)
346         MMI_SDC1(%[ftmp20], %[block], 0x50)
347         MMI_SDC1(%[ftmp21], %[block], 0x60)
348         MMI_SDC1(%[ftmp22], %[block], 0x70)
349 
350        // 2nd part
351         MMI_LDC1(%[ftmp1], %[temp], 0x08)
352         MMI_LDC1(%[ftmp11], %[temp], 0x18)
353         MMI_LDC1(%[ftmp2], %[temp], 0x28)
354         MMI_LDC1(%[ftmp12], %[temp], 0x38)
355         MMI_LDC1(%[ftmp3], %[temp], 0x48)
356         MMI_LDC1(%[ftmp13], %[temp], 0x58)
357         MMI_LDC1(%[ftmp4], %[temp], 0x68)
358         MMI_LDC1(%[ftmp14], %[temp], 0x78)
359         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
360         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
361         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
362         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
363 
364         "punpcklhw  %[ftmp9],   %[ftmp11],  %[ftmp12]                   \n\t"
365         "punpckhhw  %[ftmp10],  %[ftmp11],  %[ftmp12]                   \n\t"
366         "punpcklhw  %[ftmp11],  %[ftmp13],  %[ftmp14]                   \n\t"
367         "punpckhhw  %[ftmp12],  %[ftmp13],  %[ftmp14]                   \n\t"
368 
369         /* ftmp15:dst03,dst02,dst01,dst00 ftmp22:dst73,dst72,dst71,dst70 */
370         VC1_INV_TRANCS_8_TYPE2(%[ftmp15], %[ftmp22], 0x0010000c, 0x0006000c,
371                                0x000f0010, 0x00040009, %[ff_pw_64], %[ff_pw_1])
372 
373         /* ftmp16:dst13,dst12,dst11,dst10 ftmp21:dst63,dst62,dst61,dst60 */
374         VC1_INV_TRANCS_8_TYPE2(%[ftmp16], %[ftmp21], 0x0006000c, 0xfff0fff4,
375                                0xfffc000f, 0xfff7fff0, %[ff_pw_64], %[ff_pw_1])
376 
377         /* ftmp17:dst23,dst22,dst21,dst20 ftmp20:dst53,dst52,dst51,dst50 */
378         VC1_INV_TRANCS_8_TYPE2(%[ftmp17], %[ftmp20], 0xfffa000c, 0x0010fff4,
379                                0xfff00009, 0x000f0004, %[ff_pw_64], %[ff_pw_1])
380 
381         /* ftmp18:dst33,dst32,dst31,dst30 ftmp19:dst43,dst42,dst41,dst40 */
382         VC1_INV_TRANCS_8_TYPE2(%[ftmp18], %[ftmp19], 0xfff0000c, 0xfffa000c,
383                                0xfff70004, 0xfff0000f, %[ff_pw_64], %[ff_pw_1])
384 
385         MMI_SDC1(%[ftmp15], %[block], 0x08)
386         MMI_SDC1(%[ftmp16], %[block], 0x18)
387         MMI_SDC1(%[ftmp17], %[block], 0x28)
388         MMI_SDC1(%[ftmp18], %[block], 0x38)
389         MMI_SDC1(%[ftmp19], %[block], 0x48)
390         MMI_SDC1(%[ftmp20], %[block], 0x58)
391         MMI_SDC1(%[ftmp21], %[block], 0x68)
392         MMI_SDC1(%[ftmp22], %[block], 0x78)
393         /* 2nd loop: end */
394         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
395           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
396           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
397           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
398           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
399           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
400           [ftmp12]"=&f"(ftmp[12]),      [ftmp13]"=&f"(ftmp[13]),
401           [ftmp14]"=&f"(ftmp[14]),      [ftmp15]"=&f"(ftmp[15]),
402           [ftmp16]"=&f"(ftmp[16]),      [ftmp17]"=&f"(ftmp[17]),
403           [ftmp18]"=&f"(ftmp[18]),      [ftmp19]"=&f"(ftmp[19]),
404           [ftmp20]"=&f"(ftmp[20]),      [ftmp21]"=&f"(ftmp[21]),
405           [ftmp22]"=&f"(ftmp[22]),
406           [tmp0]"=&r"(tmp[0])
407         : [ff_pw_1]"f"(ff_pw_1_local),  [ff_pw_64]"f"(ff_pw_64_local),
408           [ff_pw_4]"f"(ff_pw_4_local), [block]"r"(block),
409           [temp]"r"(temp)
410         : "memory"
411     );
412 }
413 #endif
414 
415 /* Do inverse transform on 8x4 part of block */
ff_vc1_inv_trans_8x4_dc_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)416 void ff_vc1_inv_trans_8x4_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
417 {
418     int dc = block[0];
419     double ftmp[9];
420 
421     dc = ( 3 * dc +  1) >> 1;
422     dc = (17 * dc + 64) >> 7;
423 
424     __asm__ volatile(
425         "xor        %[ftmp0],   %[ftmp0],       %[ftmp0]                \n\t"
426         "pshufh     %[dc],      %[dc],          %[ftmp0]                \n\t"
427 
428         MMI_LDC1(%[ftmp1], %[dest0], 0x00)
429         MMI_LDC1(%[ftmp2], %[dest1], 0x00)
430         MMI_LDC1(%[ftmp3], %[dest2], 0x00)
431         MMI_LDC1(%[ftmp4], %[dest3], 0x00)
432 
433         "punpckhbh  %[ftmp5],   %[ftmp1],       %[ftmp0]                \n\t"
434         "punpcklbh  %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
435         "punpckhbh  %[ftmp6],   %[ftmp2],       %[ftmp0]                \n\t"
436         "punpcklbh  %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
437         "punpckhbh  %[ftmp7],   %[ftmp3],       %[ftmp0]                \n\t"
438         "punpcklbh  %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
439         "punpckhbh  %[ftmp8],   %[ftmp4],       %[ftmp0]                \n\t"
440         "punpcklbh  %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
441 
442         "paddsh     %[ftmp1],   %[ftmp1],       %[dc]                   \n\t"
443         "paddsh     %[ftmp2],   %[ftmp2],       %[dc]                   \n\t"
444         "paddsh     %[ftmp3],   %[ftmp3],       %[dc]                   \n\t"
445         "paddsh     %[ftmp4],   %[ftmp4],       %[dc]                   \n\t"
446         "paddsh     %[ftmp5],   %[ftmp5],       %[dc]                   \n\t"
447         "paddsh     %[ftmp6],   %[ftmp6],       %[dc]                   \n\t"
448         "paddsh     %[ftmp7],   %[ftmp7],       %[dc]                   \n\t"
449         "paddsh     %[ftmp8],   %[ftmp8],       %[dc]                   \n\t"
450 
451         "packushb   %[ftmp1],   %[ftmp1],       %[ftmp5]                \n\t"
452         "packushb   %[ftmp2],   %[ftmp2],       %[ftmp6]                \n\t"
453         "packushb   %[ftmp3],   %[ftmp3],       %[ftmp7]                \n\t"
454         "packushb   %[ftmp4],   %[ftmp4],       %[ftmp8]                \n\t"
455 
456         MMI_SDC1(%[ftmp1], %[dest0], 0x00)
457         MMI_SDC1(%[ftmp2], %[dest1], 0x00)
458         MMI_SDC1(%[ftmp3], %[dest2], 0x00)
459         MMI_SDC1(%[ftmp4], %[dest3], 0x00)
460         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
461           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
462           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
463           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
464           [ftmp8]"=&f"(ftmp[8])
465         : [dest0]"r"(dest+0*linesize),  [dest1]"r"(dest+1*linesize),
466           [dest2]"r"(dest+2*linesize),  [dest3]"r"(dest+3*linesize),
467           [dc]"f"(dc)
468         : "memory"
469     );
470 }
471 
472 #if _MIPS_SIM != _ABIO32
ff_vc1_inv_trans_8x4_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)473 void ff_vc1_inv_trans_8x4_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
474 {
475     int16_t *src = block;
476     int16_t *dst = block;
477     double ftmp[16];
478     uint32_t tmp[1];
479     int16_t count = 4;
480     DECLARE_ALIGNED(16, const uint64_t, ff_pw_4_local) = {0x0000000400000004ULL};
481     DECLARE_ALIGNED(16, const uint64_t, ff_pw_64_local)= {0x0000004000000040ULL};
482     int16_t coeff[64] = {12, 16,  16,  15,  12,   9,   6,   4,
483                          12, 15,   6,  -4, -12, -16, -16,  -9,
484                          12,  9,  -6, -16, -12,   4,  16,  15,
485                          12,  4, -16,  -9,  12,  15,  -6, -16,
486                          12, -4, -16,   9,  12, -15,  -6,  16,
487                          12, -9,  -6,  16, -12,  -4,  16, -15,
488                          12, -15,  6,   4, -12,  16, -16,   9,
489                          12, -16, 16, -15,  12,  -9,   6,  -4};
490 
491     // 1st loop
492     __asm__ volatile (
493         "li         %[tmp0],    0x03                                    \n\t"
494         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
495 
496         "1:                                                             \n\t"
497         MMI_LDC1(%[ftmp1], %[src], 0x00)
498         MMI_LDC1(%[ftmp2], %[src], 0x08)
499 
500         /* ftmp11: dst1,dst0 */
501         MMI_LDC1(%[ftmp3], %[coeff], 0x00)
502         MMI_LDC1(%[ftmp4], %[coeff], 0x08)
503         MMI_LDC1(%[ftmp5], %[coeff], 0x10)
504         MMI_LDC1(%[ftmp6], %[coeff], 0x18)
505         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp3]                    \n\t"
506         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp4]                    \n\t"
507         "paddw      %[ftmp9],   %[ftmp7],   %[ftmp8]                    \n\t"
508         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp5]                    \n\t"
509         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp6]                    \n\t"
510         "paddw      %[ftmp10],  %[ftmp7],   %[ftmp8]                    \n\t"
511         "punpcklwd  %[ftmp7],   %[ftmp9],   %[ftmp10]                   \n\t"
512         "punpckhwd  %[ftmp8],   %[ftmp9],   %[ftmp10]                   \n\t"
513         "paddw      %[ftmp11],  %[ftmp7],   %[ftmp8]                    \n\t"
514         "paddw      %[ftmp11],  %[ftmp11],  %[ff_pw_4]                  \n\t"
515 
516         /* ftmp12: dst3,dst2 */
517         MMI_LDC1(%[ftmp3], %[coeff], 0x20)
518         MMI_LDC1(%[ftmp4], %[coeff], 0x28)
519         MMI_LDC1(%[ftmp5], %[coeff], 0x30)
520         MMI_LDC1(%[ftmp6], %[coeff], 0x38)
521         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp3]                    \n\t"
522         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp4]                    \n\t"
523         "paddw      %[ftmp9],   %[ftmp7],   %[ftmp8]                    \n\t"
524         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp5]                    \n\t"
525         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp6]                    \n\t"
526         "paddw      %[ftmp10],  %[ftmp7],   %[ftmp8]                    \n\t"
527         "punpcklwd  %[ftmp7],   %[ftmp9],   %[ftmp10]                   \n\t"
528         "punpckhwd  %[ftmp8],   %[ftmp9],   %[ftmp10]                   \n\t"
529         "paddw      %[ftmp12],  %[ftmp7],   %[ftmp8]                    \n\t"
530         "paddw      %[ftmp12],  %[ftmp12],  %[ff_pw_4]                  \n\t"
531 
532         /* ftmp13: dst5,dst4 */
533         MMI_LDC1(%[ftmp3], %[coeff], 0x40)
534         MMI_LDC1(%[ftmp4], %[coeff], 0x48)
535         MMI_LDC1(%[ftmp5], %[coeff], 0x50)
536         MMI_LDC1(%[ftmp6], %[coeff], 0x58)
537         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp3]                    \n\t"
538         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp4]                    \n\t"
539         "paddw      %[ftmp9],   %[ftmp7],   %[ftmp8]                    \n\t"
540         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp5]                    \n\t"
541         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp6]                    \n\t"
542         "paddw      %[ftmp10],  %[ftmp7],   %[ftmp8]                    \n\t"
543         "punpcklwd  %[ftmp7],   %[ftmp9],   %[ftmp10]                   \n\t"
544         "punpckhwd  %[ftmp8],   %[ftmp9],   %[ftmp10]                   \n\t"
545         "paddw      %[ftmp13],  %[ftmp7],   %[ftmp8]                    \n\t"
546         "paddw      %[ftmp13],  %[ftmp13],  %[ff_pw_4]                  \n\t"
547 
548         /* ftmp14: dst7,dst6 */
549         MMI_LDC1(%[ftmp3], %[coeff], 0x60)
550         MMI_LDC1(%[ftmp4], %[coeff], 0x68)
551         MMI_LDC1(%[ftmp5], %[coeff], 0x70)
552         MMI_LDC1(%[ftmp6], %[coeff], 0x78)
553         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp3]                    \n\t"
554         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp4]                    \n\t"
555         "paddw      %[ftmp9],   %[ftmp7],   %[ftmp8]                    \n\t"
556         "pmaddhw    %[ftmp7],   %[ftmp1],   %[ftmp5]                    \n\t"
557         "pmaddhw    %[ftmp8],   %[ftmp2],   %[ftmp6]                    \n\t"
558         "paddw      %[ftmp10],  %[ftmp7],   %[ftmp8]                    \n\t"
559         "punpcklwd  %[ftmp7],   %[ftmp9],   %[ftmp10]                   \n\t"
560         "punpckhwd  %[ftmp8],   %[ftmp9],   %[ftmp10]                   \n\t"
561         "paddw      %[ftmp14],  %[ftmp7],   %[ftmp8]                    \n\t"
562         "paddw      %[ftmp14],  %[ftmp14],  %[ff_pw_4]                  \n\t"
563 
564         /* ftmp9: dst3,dst2,dst1,dst0    ftmp10: dst7,dst6,dst5,dst4 */
565         "psraw      %[ftmp11],  %[ftmp11],  %[ftmp0]                    \n\t"
566         "psraw      %[ftmp12],  %[ftmp12],  %[ftmp0]                    \n\t"
567         "psraw      %[ftmp13],  %[ftmp13],  %[ftmp0]                    \n\t"
568         "psraw      %[ftmp14],  %[ftmp14],  %[ftmp0]                    \n\t"
569         "punpcklhw  %[ftmp7],   %[ftmp11],  %[ftmp12]                   \n\t"
570         "punpckhhw  %[ftmp8],   %[ftmp11],  %[ftmp12]                   \n\t"
571         "punpcklhw  %[ftmp9],   %[ftmp7],   %[ftmp8]                    \n\t"
572         "punpcklhw  %[ftmp7],   %[ftmp13],  %[ftmp14]                   \n\t"
573         "punpckhhw  %[ftmp8],   %[ftmp13],  %[ftmp14]                   \n\t"
574         "punpcklhw  %[ftmp10],  %[ftmp7],   %[ftmp8]                    \n\t"
575         MMI_SDC1(%[ftmp9], %[dst], 0x00)
576         MMI_SDC1(%[ftmp10], %[dst], 0x08)
577 
578         PTR_ADDIU  "%[src],     %[src],     0x10                        \n\t"
579         PTR_ADDIU  "%[dst],     %[dst],     0x10                        \n\t"
580         "addiu      %[count],   %[count],   -0x01                       \n\t"
581         "bnez       %[count],   1b                                      \n\t"
582         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
583           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
584           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
585           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
586           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
587           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
588           [ftmp12]"=&f"(ftmp[12]),      [ftmp13]"=&f"(ftmp[13]),
589           [ftmp14]"=&f"(ftmp[14]),      [tmp0]"=&r"(tmp[0]),
590           [src]"+&r"(src), [dst]"+&r"(dst), [count]"+&r"(count)
591         : [ff_pw_4]"f"(ff_pw_4_local),  [coeff]"r"(coeff)
592         : "memory"
593     );
594 
595     src = block;
596 
597     // 2nd loop
598     __asm__ volatile (
599         "li         %[tmp0],    0x44                                    \n\t"
600         "mtc1       %[tmp0],    %[ftmp15]                               \n\t"
601 
602         // 1st part
603         "li         %[tmp0],    0x07                                    \n\t"
604         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
605         MMI_LDC1(%[ftmp1], %[src], 0x00)
606         MMI_LDC1(%[ftmp2], %[src], 0x10)
607         MMI_LDC1(%[ftmp3], %[src], 0x20)
608         MMI_LDC1(%[ftmp4], %[src], 0x30)
609         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
610         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
611         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
612         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
613 
614         /* ftmp11: dst03,dst02,dst01,dst00 */
615         "li         %[tmp0],    0x00160011                              \n\t"
616         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
617         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
618         "li         %[tmp0],    0x000a0011                              \n\t"
619         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
620         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
621         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
622         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
623         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
624         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
625         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
626         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
627         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
628         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
629         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
630         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
631         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
632         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
633         "punpcklhw  %[ftmp11],  %[ftmp1],   %[ftmp2]                    \n\t"
634 
635         /* ftmp12: dst13,dst12,dst11,dst10 */
636         "li         %[tmp0],    0x000a0011                              \n\t"
637         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
638         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
639         "li         %[tmp0],    0xffeaffef                              \n\t"
640         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
641         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
642         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
643         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
644         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
645         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
646         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
647         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
648         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
649         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
650         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
651         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
652         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
653         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
654         "punpcklhw  %[ftmp12],  %[ftmp1],   %[ftmp2]                    \n\t"
655 
656         /* ftmp13: dst23,dst22,dst21,dst20 */
657         "li         %[tmp0],    0xfff60011                              \n\t"
658         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
659         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
660         "li         %[tmp0],    0x0016ffef                              \n\t"
661         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
662         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
663         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
664         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
665         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
666         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
667         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
668         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
669         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
670         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
671         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
672         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
673         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
674         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
675         "punpcklhw  %[ftmp13],  %[ftmp1],   %[ftmp2]                    \n\t"
676 
677         /* ftmp14: dst33,dst32,dst31,dst30 */
678         "li         %[tmp0],    0xffea0011                              \n\t"
679         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
680         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
681         "li         %[tmp0],    0xfff60011                              \n\t"
682         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
683         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
684         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
685         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
686         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
687         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
688         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
689         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
690         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
691         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
692         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
693         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
694         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
695         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
696         "punpcklhw  %[ftmp14],  %[ftmp1],   %[ftmp2]                    \n\t"
697 
698         MMI_LWC1(%[ftmp1], %[dest], 0x00)
699         PTR_ADDU    "%[tmp0],   %[dest],    %[linesize]                 \n\t"
700         MMI_LWC1(%[ftmp2], %[tmp0], 0x00)
701         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
702         MMI_LWC1(%[ftmp3], %[tmp0], 0x00)
703         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
704         MMI_LWC1(%[ftmp4], %[tmp0], 0x00)
705         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
706         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
707         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
708         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
709         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
710         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp11]                   \n\t"
711         "paddh      %[ftmp2],   %[ftmp2],   %[ftmp12]                   \n\t"
712         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp13]                   \n\t"
713         "paddh      %[ftmp4],   %[ftmp4],   %[ftmp14]                   \n\t"
714         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
715         "packushb   %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
716         "packushb   %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
717         "packushb   %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
718         MMI_SWC1(%[ftmp1], %[dest], 0x00)
719         PTR_ADDU   "%[tmp0],    %[dest],    %[linesize]                 \n\t"
720         MMI_SWC1(%[ftmp2], %[tmp0], 0x00)
721         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
722         MMI_SWC1(%[ftmp3], %[tmp0], 0x00)
723         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
724         MMI_SWC1(%[ftmp4], %[tmp0], 0x00)
725 
726         // 2nd part
727         "li         %[tmp0],    0x07                                    \n\t"
728         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
729         MMI_LDC1(%[ftmp1], %[src], 0x08)
730         MMI_LDC1(%[ftmp2], %[src], 0x18)
731         MMI_LDC1(%[ftmp3], %[src], 0x28)
732         MMI_LDC1(%[ftmp4], %[src], 0x38)
733         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
734         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
735         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
736         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
737 
738         /* ftmp11: dst03,dst02,dst01,dst00 */
739         "li         %[tmp0],    0x00160011                              \n\t"
740         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
741         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
742         "li         %[tmp0],    0x000a0011                              \n\t"
743         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
744         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
745         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
746         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
747         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
748         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
749         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
750         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
751         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
752         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
753         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
754         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
755         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
756         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
757         "punpcklhw  %[ftmp11],  %[ftmp1],   %[ftmp2]                    \n\t"
758 
759         /* ftmp12: dst13,dst12,dst11,dst10 */
760         "li         %[tmp0],    0x000a0011                              \n\t"
761         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
762         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
763         "li         %[tmp0],    0xffeaffef                              \n\t"
764         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
765         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
766         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
767         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
768         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
769         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
770         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
771         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
772         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
773         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
774         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
775         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
776         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
777         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
778         "punpcklhw  %[ftmp12],  %[ftmp1],   %[ftmp2]                    \n\t"
779 
780         /* ftmp13: dst23,dst22,dst21,dst20 */
781         "li         %[tmp0],    0xfff60011                              \n\t"
782         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
783         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
784         "li         %[tmp0],    0x0016ffef                              \n\t"
785         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
786         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
787         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
788         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
789         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
790         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
791         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
792         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
793         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
794         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
795         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
796         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
797         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
798         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
799         "punpcklhw  %[ftmp13],  %[ftmp1],   %[ftmp2]                    \n\t"
800 
801         /* ftmp14: dst33,dst32,dst31,dst30 */
802         "li         %[tmp0],    0xffea0011                              \n\t"
803         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
804         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
805         "li         %[tmp0],    0xfff60011                              \n\t"
806         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
807         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
808         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
809         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
810         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
811         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
812         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
813         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
814         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
815         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
816         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
817         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
818         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
819         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
820         "punpcklhw  %[ftmp14],  %[ftmp1],   %[ftmp2]                    \n\t"
821 
822         MMI_LWC1(%[ftmp1], %[dest], 0x04)
823         PTR_ADDU    "%[tmp0],   %[dest],    %[linesize]                 \n\t"
824         MMI_LWC1(%[ftmp2], %[tmp0], 0x04)
825         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
826         MMI_LWC1(%[ftmp3], %[tmp0], 0x04)
827         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
828         MMI_LWC1(%[ftmp4], %[tmp0], 0x04)
829         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
830         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
831         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
832         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
833         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
834         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp11]                   \n\t"
835         "paddh      %[ftmp2],   %[ftmp2],   %[ftmp12]                   \n\t"
836         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp13]                   \n\t"
837         "paddh      %[ftmp4],   %[ftmp4],   %[ftmp14]                   \n\t"
838         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
839         "packushb   %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
840         "packushb   %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
841         "packushb   %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
842         MMI_SWC1(%[ftmp1], %[dest], 0x04)
843         PTR_ADDU   "%[tmp0],    %[dest],    %[linesize]                 \n\t"
844         MMI_SWC1(%[ftmp2], %[tmp0], 0x04)
845         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
846         MMI_SWC1(%[ftmp3], %[tmp0], 0x04)
847         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
848         MMI_SWC1(%[ftmp4], %[tmp0], 0x04)
849 
850         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
851           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
852           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
853           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
854           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
855           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
856           [ftmp12]"=&f"(ftmp[12]),      [ftmp13]"=&f"(ftmp[13]),
857           [ftmp14]"=&f"(ftmp[14]),      [ftmp15]"=&f"(ftmp[15]),
858           [tmp0]"=&r"(tmp[0])
859         : [ff_pw_64]"f"(ff_pw_64_local),
860           [src]"r"(src), [dest]"r"(dest), [linesize]"r"(linesize)
861         :"memory"
862     );
863 }
864 #endif
865 
866 /* Do inverse transform on 4x8 parts of block */
ff_vc1_inv_trans_4x8_dc_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)867 void ff_vc1_inv_trans_4x8_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
868 {
869     int dc = block[0];
870     double ftmp[9];
871     DECLARE_VAR_LOW32;
872 
873     dc = (17 * dc +  4) >> 3;
874     dc = (12 * dc + 64) >> 7;
875 
876     __asm__ volatile(
877         "xor        %[ftmp0],   %[ftmp0],       %[ftmp0]                \n\t"
878         "pshufh     %[dc],      %[dc],          %[ftmp0]                \n\t"
879 
880         MMI_LWC1(%[ftmp1], %[dest0], 0x00)
881         MMI_LWC1(%[ftmp2], %[dest1], 0x00)
882         MMI_LWC1(%[ftmp3], %[dest2], 0x00)
883         MMI_LWC1(%[ftmp4], %[dest3], 0x00)
884         MMI_LWC1(%[ftmp5], %[dest4], 0x00)
885         MMI_LWC1(%[ftmp6], %[dest5], 0x00)
886         MMI_LWC1(%[ftmp7], %[dest6], 0x00)
887         MMI_LWC1(%[ftmp8], %[dest7], 0x00)
888 
889         "punpcklbh  %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
890         "punpcklbh  %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
891         "punpcklbh  %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
892         "punpcklbh  %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
893         "punpcklbh  %[ftmp5],   %[ftmp5],       %[ftmp0]                \n\t"
894         "punpcklbh  %[ftmp6],   %[ftmp6],       %[ftmp0]                \n\t"
895         "punpcklbh  %[ftmp7],   %[ftmp7],       %[ftmp0]                \n\t"
896         "punpcklbh  %[ftmp8],   %[ftmp8],       %[ftmp0]                \n\t"
897 
898         "paddsh     %[ftmp1],   %[ftmp1],       %[dc]                   \n\t"
899         "paddsh     %[ftmp2],   %[ftmp2],       %[dc]                   \n\t"
900         "paddsh     %[ftmp3],   %[ftmp3],       %[dc]                   \n\t"
901         "paddsh     %[ftmp4],   %[ftmp4],       %[dc]                   \n\t"
902         "paddsh     %[ftmp5],   %[ftmp5],       %[dc]                   \n\t"
903         "paddsh     %[ftmp6],   %[ftmp6],       %[dc]                   \n\t"
904         "paddsh     %[ftmp7],   %[ftmp7],       %[dc]                   \n\t"
905         "paddsh     %[ftmp8],   %[ftmp8],       %[dc]                   \n\t"
906 
907         "packushb   %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
908         "packushb   %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
909         "packushb   %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
910         "packushb   %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
911         "packushb   %[ftmp5],   %[ftmp5],       %[ftmp0]                \n\t"
912         "packushb   %[ftmp6],   %[ftmp6],       %[ftmp0]                \n\t"
913         "packushb   %[ftmp7],   %[ftmp7],       %[ftmp0]                \n\t"
914         "packushb   %[ftmp8],   %[ftmp8],       %[ftmp0]                \n\t"
915 
916         MMI_SWC1(%[ftmp1], %[dest0], 0x00)
917         MMI_SWC1(%[ftmp2], %[dest1], 0x00)
918         MMI_SWC1(%[ftmp3], %[dest2], 0x00)
919         MMI_SWC1(%[ftmp4], %[dest3], 0x00)
920         MMI_SWC1(%[ftmp5], %[dest4], 0x00)
921         MMI_SWC1(%[ftmp6], %[dest5], 0x00)
922         MMI_SWC1(%[ftmp7], %[dest6], 0x00)
923         MMI_SWC1(%[ftmp8], %[dest7], 0x00)
924         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
925           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
926           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
927           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
928           RESTRICT_ASM_LOW32
929           [ftmp8]"=&f"(ftmp[8])
930         : [dest0]"r"(dest+0*linesize),  [dest1]"r"(dest+1*linesize),
931           [dest2]"r"(dest+2*linesize),  [dest3]"r"(dest+3*linesize),
932           [dest4]"r"(dest+4*linesize),  [dest5]"r"(dest+5*linesize),
933           [dest6]"r"(dest+6*linesize),  [dest7]"r"(dest+7*linesize),
934           [dc]"f"(dc)
935         : "memory"
936     );
937 }
938 
939 #if _MIPS_SIM != _ABIO32
ff_vc1_inv_trans_4x8_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)940 void ff_vc1_inv_trans_4x8_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
941 {
942     int16_t *src = block;
943     int16_t *dst = block;
944     double ftmp[23];
945     uint32_t count = 8, tmp[1];
946     int16_t coeff[16] = {17, 22, 17, 10,
947                          17, 10,-17,-22,
948                          17,-10,-17, 22,
949                          17,-22, 17,-10};
950     DECLARE_ALIGNED(8, const uint64_t, ff_pw_1_local) = {0x0000000100000001ULL};
951     DECLARE_ALIGNED(8, const uint64_t, ff_pw_4_local) = {0x0000000400000004ULL};
952     DECLARE_ALIGNED(8, const uint64_t, ff_pw_64_local)= {0x0000004000000040ULL};
953 
954     // 1st loop
955     __asm__ volatile (
956 
957         "li         %[tmp0],    0x03                                    \n\t"
958         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
959 
960         MMI_LDC1(%[ftmp2], %[coeff], 0x00)
961         MMI_LDC1(%[ftmp3], %[coeff], 0x08)
962         MMI_LDC1(%[ftmp4], %[coeff], 0x10)
963         MMI_LDC1(%[ftmp5], %[coeff], 0x18)
964         "1:                                                             \n\t"
965         /* ftmp8: dst3,dst2,dst1,dst0 */
966         MMI_LDC1(%[ftmp1], %[src], 0x00)
967         "pmaddhw    %[ftmp6],   %[ftmp2],   %[ftmp1]                    \n\t"
968         "pmaddhw    %[ftmp7],   %[ftmp3],   %[ftmp1]                    \n\t"
969         "pmaddhw    %[ftmp8],   %[ftmp4],   %[ftmp1]                    \n\t"
970         "pmaddhw    %[ftmp9],   %[ftmp5],   %[ftmp1]                    \n\t"
971         "punpcklwd  %[ftmp10],  %[ftmp6],   %[ftmp7]                    \n\t"
972         "punpckhwd  %[ftmp11],  %[ftmp6],   %[ftmp7]                    \n\t"
973         "punpcklwd  %[ftmp6],   %[ftmp8],   %[ftmp9]                    \n\t"
974         "punpckhwd  %[ftmp7],   %[ftmp8],   %[ftmp9]                    \n\t"
975         "paddw      %[ftmp8],   %[ftmp10],  %[ftmp11]                   \n\t"
976         "paddw      %[ftmp9],   %[ftmp6],   %[ftmp7]                    \n\t"
977         "paddw      %[ftmp8],   %[ftmp8],   %[ff_pw_4]                  \n\t"
978         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_4]                  \n\t"
979         "psraw      %[ftmp8],   %[ftmp8],   %[ftmp0]                    \n\t"
980         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
981         "punpcklhw  %[ftmp6],   %[ftmp8],   %[ftmp9]                    \n\t"
982         "punpckhhw  %[ftmp7],   %[ftmp8],   %[ftmp9]                    \n\t"
983         "punpcklhw  %[ftmp8],   %[ftmp6],   %[ftmp7]                    \n\t"
984         MMI_SDC1(%[ftmp8], %[dst], 0x00)
985 
986         PTR_ADDIU  "%[src],     %[src],     0x10                        \n\t"
987         PTR_ADDIU  "%[dst],     %[dst],     0x10                        \n\t"
988         "addiu      %[count],   %[count],   -0x01                       \n\t"
989         "bnez       %[count],   1b                                      \n\t"
990         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
991           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
992           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
993           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
994           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
995           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
996           [tmp0]"=&r"(tmp[0]),          [count]"+&r"(count),
997           [src]"+&r"(src),              [dst]"+&r"(dst)
998         : [ff_pw_4]"f"(ff_pw_4_local),  [coeff]"r"(coeff)
999         : "memory"
1000     );
1001 
1002     src = block;
1003 
1004     // 2nd loop
1005     __asm__ volatile (
1006         "li         %[tmp0],    0x07                                    \n\t"
1007         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
1008 
1009         MMI_LDC1(%[ftmp1], %[src], 0x00)
1010         MMI_LDC1(%[ftmp2], %[src], 0x20)
1011         MMI_LDC1(%[ftmp3], %[src], 0x40)
1012         MMI_LDC1(%[ftmp4], %[src], 0x60)
1013         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
1014         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
1015         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
1016         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
1017 
1018         MMI_LDC1(%[ftmp1], %[src], 0x10)
1019         MMI_LDC1(%[ftmp2], %[src], 0x30)
1020         MMI_LDC1(%[ftmp3], %[src], 0x50)
1021         MMI_LDC1(%[ftmp4], %[src], 0x70)
1022         "punpcklhw  %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
1023         "punpckhhw  %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
1024         "punpcklhw  %[ftmp11],  %[ftmp3],   %[ftmp4]                    \n\t"
1025         "punpckhhw  %[ftmp12],  %[ftmp3],   %[ftmp4]                    \n\t"
1026 
1027         /* ftmp15:dst03,dst02,dst01,dst00 ftmp22:dst73,dst72,dst71,dst70 */
1028         VC1_INV_TRANCS_8_TYPE2(%[ftmp15], %[ftmp22], 0x0010000c, 0x0006000c,
1029                                0x000f0010, 0x00040009, %[ff_pw_64], %[ff_pw_1])
1030 
1031         /* ftmp16:dst13,dst12,dst11,dst10 ftmp21:dst63,dst62,dst61,dst60 */
1032         VC1_INV_TRANCS_8_TYPE2(%[ftmp16], %[ftmp21], 0x0006000c, 0xfff0fff4,
1033                                0xfffc000f, 0xfff7fff0, %[ff_pw_64], %[ff_pw_1])
1034 
1035         /* ftmp17:dst23,dst22,dst21,dst20 ftmp20:dst53,dst52,dst51,dst50 */
1036         VC1_INV_TRANCS_8_TYPE2(%[ftmp17], %[ftmp20], 0xfffa000c, 0x0010fff4,
1037                                0xfff00009, 0x000f0004, %[ff_pw_64], %[ff_pw_1])
1038 
1039         /* ftmp18:dst33,dst32,dst31,dst30 ftmp19:dst43,dst42,dst41,dst40 */
1040         VC1_INV_TRANCS_8_TYPE2(%[ftmp18], %[ftmp19], 0xfff0000c, 0xfffa000c,
1041                                0xfff70004, 0xfff0000f, %[ff_pw_64], %[ff_pw_1])
1042 
1043         MMI_LWC1(%[ftmp1], %[dest], 0x00)
1044         PTR_ADDU  "%[tmp0],   %[dest],    %[linesize]                 \n\t"
1045         MMI_LWC1(%[ftmp2], %[tmp0], 0x00)
1046         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1047         MMI_LWC1(%[ftmp3], %[tmp0], 0x00)
1048         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1049         MMI_LWC1(%[ftmp4], %[tmp0], 0x00)
1050         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1051         MMI_LWC1(%[ftmp5], %[tmp0], 0x00)
1052         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1053         MMI_LWC1(%[ftmp6], %[tmp0], 0x00)
1054         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1055         MMI_LWC1(%[ftmp7], %[tmp0], 0x00)
1056         PTR_ADDU  "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1057         MMI_LWC1(%[ftmp8], %[tmp0], 0x00)
1058         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
1059         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
1060         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
1061         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
1062         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
1063         "punpcklbh  %[ftmp5],   %[ftmp5],   %[ftmp0]                    \n\t"
1064         "punpcklbh  %[ftmp6],   %[ftmp6],   %[ftmp0]                    \n\t"
1065         "punpcklbh  %[ftmp7],   %[ftmp7],   %[ftmp0]                    \n\t"
1066         "punpcklbh  %[ftmp8],   %[ftmp8],   %[ftmp0]                    \n\t"
1067 
1068         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp15]                   \n\t"
1069         "paddh      %[ftmp2],   %[ftmp2],   %[ftmp16]                   \n\t"
1070         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp17]                   \n\t"
1071         "paddh      %[ftmp4],   %[ftmp4],   %[ftmp18]                   \n\t"
1072         "paddh      %[ftmp5],   %[ftmp5],   %[ftmp19]                   \n\t"
1073         "paddh      %[ftmp6],   %[ftmp6],   %[ftmp20]                   \n\t"
1074         "paddh      %[ftmp7],   %[ftmp7],   %[ftmp21]                   \n\t"
1075         "paddh      %[ftmp8],   %[ftmp8],   %[ftmp22]                   \n\t"
1076 
1077         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
1078         "packushb   %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
1079         "packushb   %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
1080         "packushb   %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
1081         "packushb   %[ftmp5],   %[ftmp5],   %[ftmp0]                    \n\t"
1082         "packushb   %[ftmp6],   %[ftmp6],   %[ftmp0]                    \n\t"
1083         "packushb   %[ftmp7],   %[ftmp7],   %[ftmp0]                    \n\t"
1084         "packushb   %[ftmp8],   %[ftmp8],   %[ftmp0]                    \n\t"
1085 
1086         MMI_SWC1(%[ftmp1], %[dest], 0x00)
1087         PTR_ADDU   "%[tmp0],    %[dest],    %[linesize]                 \n\t"
1088         MMI_SWC1(%[ftmp2], %[tmp0], 0x00)
1089         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1090         MMI_SWC1(%[ftmp3], %[tmp0], 0x00)
1091         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1092         MMI_SWC1(%[ftmp4], %[tmp0], 0x00)
1093         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1094         MMI_SWC1(%[ftmp5], %[tmp0], 0x00)
1095         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1096         MMI_SWC1(%[ftmp6], %[tmp0], 0x00)
1097         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1098         MMI_SWC1(%[ftmp7], %[tmp0], 0x00)
1099         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1100         MMI_SWC1(%[ftmp8], %[tmp0], 0x00)
1101 
1102         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
1103           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
1104           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
1105           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
1106           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
1107           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
1108           [ftmp12]"=&f"(ftmp[12]),      [ftmp13]"=&f"(ftmp[13]),
1109           [ftmp14]"=&f"(ftmp[14]),      [ftmp15]"=&f"(ftmp[15]),
1110           [ftmp16]"=&f"(ftmp[16]),      [ftmp17]"=&f"(ftmp[17]),
1111           [ftmp18]"=&f"(ftmp[18]),      [ftmp19]"=&f"(ftmp[19]),
1112           [ftmp20]"=&f"(ftmp[20]),      [ftmp21]"=&f"(ftmp[21]),
1113           [ftmp22]"=&f"(ftmp[22]),
1114           [tmp0]"=&r"(tmp[0])
1115         : [ff_pw_1]"f"(ff_pw_1_local),  [ff_pw_64]"f"(ff_pw_64_local),
1116           [src]"r"(src), [dest]"r"(dest), [linesize]"r"(linesize)
1117         : "memory"
1118     );
1119 }
1120 #endif
1121 
1122 /* Do inverse transform on 4x4 part of block */
ff_vc1_inv_trans_4x4_dc_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)1123 void ff_vc1_inv_trans_4x4_dc_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
1124 {
1125     int dc = block[0];
1126     double ftmp[5];
1127     DECLARE_VAR_LOW32;
1128 
1129     dc = (17 * dc +  4) >> 3;
1130     dc = (17 * dc + 64) >> 7;
1131 
1132     __asm__ volatile(
1133         "xor        %[ftmp0],   %[ftmp0],       %[ftmp0]                \n\t"
1134         "pshufh     %[dc],      %[dc],          %[ftmp0]                \n\t"
1135 
1136         MMI_LWC1(%[ftmp1], %[dest0], 0x00)
1137         MMI_LWC1(%[ftmp2], %[dest1], 0x00)
1138         MMI_LWC1(%[ftmp3], %[dest2], 0x00)
1139         MMI_LWC1(%[ftmp4], %[dest3], 0x00)
1140 
1141         "punpcklbh  %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
1142         "punpcklbh  %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
1143         "punpcklbh  %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
1144         "punpcklbh  %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
1145 
1146         "paddsh     %[ftmp1],   %[ftmp1],       %[dc]                   \n\t"
1147         "paddsh     %[ftmp2],   %[ftmp2],       %[dc]                   \n\t"
1148         "paddsh     %[ftmp3],   %[ftmp3],       %[dc]                   \n\t"
1149         "paddsh     %[ftmp4],   %[ftmp4],       %[dc]                   \n\t"
1150 
1151         "packushb   %[ftmp1],   %[ftmp1],       %[ftmp0]                \n\t"
1152         "packushb   %[ftmp2],   %[ftmp2],       %[ftmp0]                \n\t"
1153         "packushb   %[ftmp3],   %[ftmp3],       %[ftmp0]                \n\t"
1154         "packushb   %[ftmp4],   %[ftmp4],       %[ftmp0]                \n\t"
1155 
1156         MMI_SWC1(%[ftmp1], %[dest0], 0x00)
1157         MMI_SWC1(%[ftmp2], %[dest1], 0x00)
1158         MMI_SWC1(%[ftmp3], %[dest2], 0x00)
1159         MMI_SWC1(%[ftmp4], %[dest3], 0x00)
1160         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
1161           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
1162           RESTRICT_ASM_LOW32
1163           [ftmp4]"=&f"(ftmp[4])
1164         : [dest0]"r"(dest+0*linesize),  [dest1]"r"(dest+1*linesize),
1165           [dest2]"r"(dest+2*linesize),  [dest3]"r"(dest+3*linesize),
1166           [dc]"f"(dc)
1167         : "memory"
1168     );
1169 }
1170 
ff_vc1_inv_trans_4x4_mmi(uint8_t * dest,ptrdiff_t linesize,int16_t * block)1171 void ff_vc1_inv_trans_4x4_mmi(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
1172 {
1173     int16_t *src = block;
1174     int16_t *dst = block;
1175     double ftmp[16];
1176     uint32_t count = 4, tmp[1];
1177     int16_t coeff[16] = {17, 22, 17, 10,
1178                          17, 10,-17,-22,
1179                          17,-10,-17, 22,
1180                          17,-22, 17,-10};
1181     DECLARE_ALIGNED(8, const uint64_t, ff_pw_4_local) = {0x0000000400000004ULL};
1182     DECLARE_ALIGNED(8, const uint64_t, ff_pw_64_local)= {0x0000004000000040ULL};
1183     // 1st loop
1184     __asm__ volatile (
1185 
1186         "li         %[tmp0],    0x03                                    \n\t"
1187         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
1188         MMI_LDC1(%[ftmp2], %[coeff], 0x00)
1189         MMI_LDC1(%[ftmp3], %[coeff], 0x08)
1190         MMI_LDC1(%[ftmp4], %[coeff], 0x10)
1191         MMI_LDC1(%[ftmp5], %[coeff], 0x18)
1192         "1:                                                             \n\t"
1193         /* ftmp8: dst3,dst2,dst1,dst0 */
1194         MMI_LDC1(%[ftmp1], %[src], 0x00)
1195         "pmaddhw    %[ftmp6],   %[ftmp2],   %[ftmp1]                    \n\t"
1196         "pmaddhw    %[ftmp7],   %[ftmp3],   %[ftmp1]                    \n\t"
1197         "pmaddhw    %[ftmp8],   %[ftmp4],   %[ftmp1]                    \n\t"
1198         "pmaddhw    %[ftmp9],   %[ftmp5],   %[ftmp1]                    \n\t"
1199         "punpcklwd  %[ftmp10],  %[ftmp6],   %[ftmp7]                    \n\t"
1200         "punpckhwd  %[ftmp11],  %[ftmp6],   %[ftmp7]                    \n\t"
1201         "punpcklwd  %[ftmp6],   %[ftmp8],   %[ftmp9]                    \n\t"
1202         "punpckhwd  %[ftmp7],   %[ftmp8],   %[ftmp9]                    \n\t"
1203         "paddw      %[ftmp8],   %[ftmp10],  %[ftmp11]                   \n\t"
1204         "paddw      %[ftmp9],   %[ftmp6],   %[ftmp7]                    \n\t"
1205         "paddw      %[ftmp8],   %[ftmp8],   %[ff_pw_4]                  \n\t"
1206         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_4]                  \n\t"
1207         "psraw      %[ftmp8],   %[ftmp8],   %[ftmp0]                    \n\t"
1208         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
1209         "punpcklhw  %[ftmp6],   %[ftmp8],   %[ftmp9]                    \n\t"
1210         "punpckhhw  %[ftmp7],   %[ftmp8],   %[ftmp9]                    \n\t"
1211         "punpcklhw  %[ftmp8],   %[ftmp6],   %[ftmp7]                    \n\t"
1212         MMI_SDC1(%[ftmp8], %[dst], 0x00)
1213 
1214         PTR_ADDIU  "%[src],     %[src],     0x10                        \n\t"
1215         PTR_ADDIU  "%[dst],     %[dst],     0x10                        \n\t"
1216         "addiu      %[count],   %[count],   -0x01                       \n\t"
1217         "bnez       %[count],   1b                                      \n\t"
1218         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
1219           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
1220           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
1221           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
1222           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
1223           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
1224           [tmp0]"=&r"(tmp[0]),          [count]"+&r"(count),
1225           [src]"+&r"(src),              [dst]"+&r"(dst)
1226         : [ff_pw_4]"f"(ff_pw_4_local),  [coeff]"r"(coeff)
1227         : "memory"
1228     );
1229 
1230     src = block;
1231 
1232     // 2nd loop
1233     __asm__ volatile (
1234         "li         %[tmp0],    0x07                                    \n\t"
1235         "mtc1       %[tmp0],    %[ftmp0]                                \n\t"
1236         "li         %[tmp0],    0x44                                    \n\t"
1237         "mtc1       %[tmp0],    %[ftmp15]                               \n\t"
1238 
1239         MMI_LDC1(%[ftmp1], %[src], 0x00)
1240         MMI_LDC1(%[ftmp2], %[src], 0x10)
1241         MMI_LDC1(%[ftmp3], %[src], 0x20)
1242         MMI_LDC1(%[ftmp4], %[src], 0x30)
1243         "punpcklhw  %[ftmp5],   %[ftmp1],   %[ftmp2]                    \n\t"
1244         "punpckhhw  %[ftmp6],   %[ftmp1],   %[ftmp2]                    \n\t"
1245         "punpcklhw  %[ftmp7],   %[ftmp3],   %[ftmp4]                    \n\t"
1246         "punpckhhw  %[ftmp8],   %[ftmp3],   %[ftmp4]                    \n\t"
1247 
1248         /* ftmp11: dst03,dst02,dst01,dst00 */
1249         "li         %[tmp0],    0x00160011                              \n\t"
1250         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
1251         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
1252         "li         %[tmp0],    0x000a0011                              \n\t"
1253         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
1254         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
1255         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
1256         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
1257         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
1258         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
1259         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
1260         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
1261         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
1262         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
1263         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
1264         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
1265         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
1266         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
1267         "punpcklhw  %[ftmp11],  %[ftmp1],   %[ftmp2]                    \n\t"
1268 
1269         /* ftmp12: dst13,dst12,dst11,dst10 */
1270         "li         %[tmp0],    0x000a0011                              \n\t"
1271         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
1272         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
1273         "li         %[tmp0],    0xffeaffef                              \n\t"
1274         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
1275         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
1276         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
1277         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
1278         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
1279         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
1280         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
1281         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
1282         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
1283         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
1284         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
1285         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
1286         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
1287         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
1288         "punpcklhw  %[ftmp12],  %[ftmp1],   %[ftmp2]                    \n\t"
1289 
1290         /* ftmp13: dst23,dst22,dst21,dst20 */
1291         "li         %[tmp0],    0xfff60011                              \n\t"
1292         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
1293         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
1294         "li         %[tmp0],    0x0016ffef                              \n\t"
1295         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
1296         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
1297         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
1298         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
1299         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
1300         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
1301         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
1302         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
1303         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
1304         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
1305         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
1306         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
1307         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
1308         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
1309         "punpcklhw  %[ftmp13],  %[ftmp1],   %[ftmp2]                    \n\t"
1310 
1311         /* ftmp14: dst33,dst32,dst31,dst30 */
1312         "li         %[tmp0],    0xffea0011                              \n\t"
1313         "mtc1       %[tmp0],    %[ftmp3]                                \n\t"
1314         "pshufh     %[ftmp3],   %[ftmp3],   %[ftmp15]                   \n\t"
1315         "li         %[tmp0],    0xfff60011                              \n\t"
1316         "mtc1       %[tmp0],    %[ftmp4]                                \n\t"
1317         "pshufh     %[ftmp4],   %[ftmp4],   %[ftmp15]                   \n\t"
1318         "pmaddhw    %[ftmp1],   %[ftmp5],   %[ftmp3]                    \n\t"
1319         "pmaddhw    %[ftmp2],   %[ftmp7],   %[ftmp4]                    \n\t"
1320         "paddw      %[ftmp9],   %[ftmp1],   %[ftmp2]                    \n\t"
1321         "pmaddhw    %[ftmp1],   %[ftmp6],   %[ftmp3]                    \n\t"
1322         "pmaddhw    %[ftmp2],   %[ftmp8],   %[ftmp4]                    \n\t"
1323         "paddw      %[ftmp10],  %[ftmp1],   %[ftmp2]                    \n\t"
1324         "paddw      %[ftmp9],   %[ftmp9],   %[ff_pw_64]                 \n\t"
1325         "paddw      %[ftmp10],  %[ftmp10],  %[ff_pw_64]                 \n\t"
1326         "psraw      %[ftmp9],   %[ftmp9],   %[ftmp0]                    \n\t"
1327         "psraw      %[ftmp10],  %[ftmp10],  %[ftmp0]                    \n\t"
1328         "punpcklhw  %[ftmp1],   %[ftmp9],   %[ftmp10]                   \n\t"
1329         "punpckhhw  %[ftmp2],   %[ftmp9],   %[ftmp10]                   \n\t"
1330         "punpcklhw  %[ftmp14],  %[ftmp1],   %[ftmp2]                    \n\t"
1331 
1332         MMI_LWC1(%[ftmp1], %[dest], 0x00)
1333         PTR_ADDU    "%[tmp0],   %[dest],    %[linesize]                 \n\t"
1334         MMI_LWC1(%[ftmp2], %[tmp0], 0x00)
1335         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1336         MMI_LWC1(%[ftmp3], %[tmp0], 0x00)
1337         PTR_ADDU    "%[tmp0],   %[tmp0],    %[linesize]                 \n\t"
1338         MMI_LWC1(%[ftmp4], %[tmp0], 0x00)
1339         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
1340         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
1341         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
1342         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
1343         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
1344         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp11]                   \n\t"
1345         "paddh      %[ftmp2],   %[ftmp2],   %[ftmp12]                   \n\t"
1346         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp13]                   \n\t"
1347         "paddh      %[ftmp4],   %[ftmp4],   %[ftmp14]                   \n\t"
1348         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]                    \n\t"
1349         "packushb   %[ftmp2],   %[ftmp2],   %[ftmp0]                    \n\t"
1350         "packushb   %[ftmp3],   %[ftmp3],   %[ftmp0]                    \n\t"
1351         "packushb   %[ftmp4],   %[ftmp4],   %[ftmp0]                    \n\t"
1352 
1353         MMI_SWC1(%[ftmp1], %[dest], 0x00)
1354         PTR_ADDU   "%[tmp0],    %[dest],    %[linesize]                 \n\t"
1355         MMI_SWC1(%[ftmp2], %[tmp0], 0x00)
1356         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1357         MMI_SWC1(%[ftmp3], %[tmp0], 0x00)
1358         PTR_ADDU   "%[tmp0],    %[tmp0],    %[linesize]                 \n\t"
1359         MMI_SWC1(%[ftmp4], %[tmp0], 0x00)
1360 
1361         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
1362           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
1363           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
1364           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
1365           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
1366           [ftmp10]"=&f"(ftmp[10]),      [ftmp11]"=&f"(ftmp[11]),
1367           [ftmp12]"=&f"(ftmp[12]),      [ftmp13]"=&f"(ftmp[13]),
1368           [ftmp14]"=&f"(ftmp[14]),      [ftmp15]"=&f"(ftmp[15]),
1369           [tmp0]"=&r"(tmp[0])
1370         : [ff_pw_64]"f"(ff_pw_64_local),
1371           [src]"r"(src), [dest]"r"(dest), [linesize]"r"(linesize)
1372         :"memory"
1373     );
1374 }
1375 
1376 /* Apply overlap transform to horizontal edge */
ff_vc1_h_overlap_mmi(uint8_t * src,int stride)1377 void ff_vc1_h_overlap_mmi(uint8_t *src, int stride)
1378 {
1379     int i;
1380     int a, b, c, d;
1381     int d1, d2;
1382     int rnd = 1;
1383     for (i = 0; i < 8; i++) {
1384         a  = src[-2];
1385         b  = src[-1];
1386         c  = src[0];
1387         d  = src[1];
1388         d1 = (a - d + 3 + rnd) >> 3;
1389         d2 = (a - d + b - c + 4 - rnd) >> 3;
1390 
1391         src[-2] = a - d1;
1392         src[-1] = av_clip_uint8(b - d2);
1393         src[0]  = av_clip_uint8(c + d2);
1394         src[1]  = d + d1;
1395         src    += stride;
1396         rnd     = !rnd;
1397     }
1398 }
1399 
ff_vc1_h_s_overlap_mmi(int16_t * left,int16_t * right,int left_stride,int right_stride,int flags)1400 void ff_vc1_h_s_overlap_mmi(int16_t *left, int16_t *right, int left_stride, int right_stride, int flags)
1401 {
1402     int i;
1403     int a, b, c, d;
1404     int d1, d2;
1405     int rnd1 = flags & 2 ? 3 : 4;
1406     int rnd2 = 7 - rnd1;
1407     for (i = 0; i < 8; i++) {
1408         a  = left[6];
1409         b  = left[7];
1410         c  = right[0];
1411         d  = right[1];
1412         d1 = a - d;
1413         d2 = a - d + b - c;
1414 
1415         left[6]  = ((a << 3) - d1 + rnd1) >> 3;
1416         left[7]  = ((b << 3) - d2 + rnd2) >> 3;
1417         right[0] = ((c << 3) + d2 + rnd1) >> 3;
1418         right[1] = ((d << 3) + d1 + rnd2) >> 3;
1419 
1420         right += right_stride;
1421         left  += left_stride;
1422         if (flags & 1) {
1423             rnd2   = 7 - rnd2;
1424             rnd1   = 7 - rnd1;
1425         }
1426     }
1427 }
1428 
1429 /* Apply overlap transform to vertical edge */
ff_vc1_v_overlap_mmi(uint8_t * src,int stride)1430 void ff_vc1_v_overlap_mmi(uint8_t *src, int stride)
1431 {
1432     int i;
1433     int a, b, c, d;
1434     int d1, d2;
1435     int rnd = 1;
1436     for (i = 0; i < 8; i++) {
1437         a  = src[-2 * stride];
1438         b  = src[-stride];
1439         c  = src[0];
1440         d  = src[stride];
1441         d1 = (a - d + 3 + rnd) >> 3;
1442         d2 = (a - d + b - c + 4 - rnd) >> 3;
1443 
1444         src[-2 * stride] = a - d1;
1445         src[-stride]     = av_clip_uint8(b - d2);
1446         src[0]           = av_clip_uint8(c + d2);
1447         src[stride]      = d + d1;
1448         src++;
1449         rnd = !rnd;
1450     }
1451 }
1452 
ff_vc1_v_s_overlap_mmi(int16_t * top,int16_t * bottom)1453 void ff_vc1_v_s_overlap_mmi(int16_t *top, int16_t *bottom)
1454 {
1455     int i;
1456     int a, b, c, d;
1457     int d1, d2;
1458     int rnd1 = 4, rnd2 = 3;
1459     for (i = 0; i < 8; i++) {
1460         a  = top[48];
1461         b  = top[56];
1462         c  = bottom[0];
1463         d  = bottom[8];
1464         d1 = a - d;
1465         d2 = a - d + b - c;
1466 
1467         top[48]   = ((a << 3) - d1 + rnd1) >> 3;
1468         top[56]   = ((b << 3) - d2 + rnd2) >> 3;
1469         bottom[0] = ((c << 3) + d2 + rnd1) >> 3;
1470         bottom[8] = ((d << 3) + d1 + rnd2) >> 3;
1471 
1472         bottom++;
1473         top++;
1474         rnd2 = 7 - rnd2;
1475         rnd1 = 7 - rnd1;
1476     }
1477 }
1478 
1479 /**
1480  * VC-1 in-loop deblocking filter for one line
1481  * @param src source block type
1482  * @param stride block stride
1483  * @param pq block quantizer
1484  * @return whether other 3 pairs should be filtered or not
1485  * @see 8.6
1486  */
vc1_filter_line(uint8_t * src,int stride,int pq)1487 static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
1488 {
1489     int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -
1490               5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;
1491     int a0_sign = a0 >> 31;        /* Store sign */
1492 
1493     a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
1494     if (a0 < pq) {
1495         int a1 = FFABS((2 * (src[-4 * stride] - src[-1 * stride]) -
1496                         5 * (src[-3 * stride] - src[-2 * stride]) + 4) >> 3);
1497         int a2 = FFABS((2 * (src[ 0 * stride] - src[ 3 * stride]) -
1498                         5 * (src[ 1 * stride] - src[ 2 * stride]) + 4) >> 3);
1499         if (a1 < a0 || a2 < a0) {
1500             int clip      = src[-1 * stride] - src[0 * stride];
1501             int clip_sign = clip >> 31;
1502 
1503             clip = ((clip ^ clip_sign) - clip_sign) >> 1;
1504             if (clip) {
1505                 int a3     = FFMIN(a1, a2);
1506                 int d      = 5 * (a3 - a0);
1507                 int d_sign = (d >> 31);
1508 
1509                 d       = ((d ^ d_sign) - d_sign) >> 3;
1510                 d_sign ^= a0_sign;
1511 
1512                 if (d_sign ^ clip_sign)
1513                     d = 0;
1514                 else {
1515                     d = FFMIN(d, clip);
1516                     d = (d ^ d_sign) - d_sign; /* Restore sign */
1517                     src[-1 * stride] = av_clip_uint8(src[-1 * stride] - d);
1518                     src[ 0 * stride] = av_clip_uint8(src[ 0 * stride] + d);
1519                 }
1520                 return 1;
1521             }
1522         }
1523     }
1524     return 0;
1525 }
1526 
1527 /**
1528  * VC-1 in-loop deblocking filter
1529  * @param src source block type
1530  * @param step distance between horizontally adjacent elements
1531  * @param stride distance between vertically adjacent elements
1532  * @param len edge length to filter (4 or 8 pixels)
1533  * @param pq block quantizer
1534  * @see 8.6
1535  */
vc1_loop_filter(uint8_t * src,int step,int stride,int len,int pq)1536 static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
1537                                    int len, int pq)
1538 {
1539     int i;
1540     int filt3;
1541 
1542     for (i = 0; i < len; i += 4) {
1543         filt3 = vc1_filter_line(src + 2 * step, stride, pq);
1544         if (filt3) {
1545             vc1_filter_line(src + 0 * step, stride, pq);
1546             vc1_filter_line(src + 1 * step, stride, pq);
1547             vc1_filter_line(src + 3 * step, stride, pq);
1548         }
1549         src += step * 4;
1550     }
1551 }
1552 
ff_vc1_v_loop_filter4_mmi(uint8_t * src,int stride,int pq)1553 void ff_vc1_v_loop_filter4_mmi(uint8_t *src, int stride, int pq)
1554 {
1555     vc1_loop_filter(src, 1, stride, 4, pq);
1556 }
1557 
ff_vc1_h_loop_filter4_mmi(uint8_t * src,int stride,int pq)1558 void ff_vc1_h_loop_filter4_mmi(uint8_t *src, int stride, int pq)
1559 {
1560     vc1_loop_filter(src, stride, 1, 4, pq);
1561 }
1562 
ff_vc1_v_loop_filter8_mmi(uint8_t * src,int stride,int pq)1563 void ff_vc1_v_loop_filter8_mmi(uint8_t *src, int stride, int pq)
1564 {
1565     vc1_loop_filter(src, 1, stride, 8, pq);
1566 }
1567 
ff_vc1_h_loop_filter8_mmi(uint8_t * src,int stride,int pq)1568 void ff_vc1_h_loop_filter8_mmi(uint8_t *src, int stride, int pq)
1569 {
1570     vc1_loop_filter(src, stride, 1, 8, pq);
1571 }
1572 
ff_vc1_v_loop_filter16_mmi(uint8_t * src,int stride,int pq)1573 void ff_vc1_v_loop_filter16_mmi(uint8_t *src, int stride, int pq)
1574 {
1575     vc1_loop_filter(src, 1, stride, 16, pq);
1576 }
1577 
ff_vc1_h_loop_filter16_mmi(uint8_t * src,int stride,int pq)1578 void ff_vc1_h_loop_filter16_mmi(uint8_t *src, int stride, int pq)
1579 {
1580     vc1_loop_filter(src, stride, 1, 16, pq);
1581 }
1582 
ff_put_vc1_mspel_mc00_mmi(uint8_t * dst,const uint8_t * src,ptrdiff_t stride,int rnd)1583 void ff_put_vc1_mspel_mc00_mmi(uint8_t *dst, const uint8_t *src,
1584                                ptrdiff_t stride, int rnd)
1585 {
1586     ff_put_pixels8_8_mmi(dst, src, stride, 8);
1587 }
ff_put_vc1_mspel_mc00_16_mmi(uint8_t * dst,const uint8_t * src,ptrdiff_t stride,int rnd)1588 void ff_put_vc1_mspel_mc00_16_mmi(uint8_t *dst, const uint8_t *src,
1589                                   ptrdiff_t stride, int rnd)
1590 {
1591     ff_put_pixels16_8_mmi(dst, src, stride, 16);
1592 }
ff_avg_vc1_mspel_mc00_mmi(uint8_t * dst,const uint8_t * src,ptrdiff_t stride,int rnd)1593 void ff_avg_vc1_mspel_mc00_mmi(uint8_t *dst, const uint8_t *src,
1594                                ptrdiff_t stride, int rnd)
1595 {
1596     ff_avg_pixels8_8_mmi(dst, src, stride, 8);
1597 }
ff_avg_vc1_mspel_mc00_16_mmi(uint8_t * dst,const uint8_t * src,ptrdiff_t stride,int rnd)1598 void ff_avg_vc1_mspel_mc00_16_mmi(uint8_t *dst, const uint8_t *src,
1599                                   ptrdiff_t stride, int rnd)
1600 {
1601     ff_avg_pixels16_8_mmi(dst, src, stride, 16);
1602 }
1603 
1604 #define OP_PUT(S, D)
1605 #define OP_AVG(S, D)                                                        \
1606     "ldc1       $f16,   "#S"                        \n\t"                   \
1607     "pavgb      "#D",   "#D",   $f16                \n\t"
1608 
1609 /** Add rounder from $f14 to $f6 and pack result at destination */
1610 #define NORMALIZE_MMI(SHIFT)                                                \
1611     "paddh      $f6,    $f6,    $f14                \n\t" /* +bias-r */     \
1612     "paddh      $f8,    $f8,    $f14                \n\t" /* +bias-r */     \
1613     "psrah      $f6,    $f6,    "SHIFT"             \n\t"                   \
1614     "psrah      $f8,    $f8,    "SHIFT"             \n\t"
1615 
1616 #define TRANSFER_DO_PACK(OP)                                                \
1617     "packushb   $f6,    $f6,    $f8                 \n\t"                   \
1618     OP((%[dst]), $f6)                                                       \
1619     "sdc1       $f6,    0x00(%[dst])                \n\t"
1620 
1621 #define TRANSFER_DONT_PACK(OP)                                              \
1622      OP(0(%[dst]), $f6)                                                     \
1623      OP(8(%[dst]), $f8)                                                     \
1624      "sdc1      $f6,    0x00(%[dst])                \n\t"                   \
1625      "sdc1      $f8,    0x08(%[dst])                \n\t"
1626 
1627 /** @see MSPEL_FILTER13_CORE for use as UNPACK macro */
1628 #define DO_UNPACK(reg)                                                      \
1629     "punpcklbh  "reg",  "reg",  $f0                 \n\t"
1630 #define DONT_UNPACK(reg)
1631 
1632 /** Compute the rounder 32-r or 8-r and unpacks it to $f14 */
1633 #define LOAD_ROUNDER_MMI(ROUND)                                             \
1634     "lwc1       $f14,   "ROUND"                     \n\t"                   \
1635     "punpcklhw  $f14,   $f14,   $f14                \n\t"                   \
1636     "punpcklwd  $f14,   $f14,   $f14                \n\t"
1637 
1638 
1639 #define SHIFT2_LINE(OFF, R0, R1, R2, R3)                                    \
1640     "paddh      "#R1",      "#R1",  "#R2"           \n\t"                   \
1641     PTR_ADDU    "$9,        %[src], %[stride1]      \n\t"                   \
1642     MMI_ULWC1(R0, $9, 0x00)                                                 \
1643     "pmullh     "#R1",      "#R1",  $f6             \n\t"                   \
1644     "punpcklbh  "#R0",      "#R0",  $f0             \n\t"                   \
1645     PTR_ADDU    "$9,        %[src], %[stride]       \n\t"                   \
1646     MMI_ULWC1(R3, $9, 0x00)                                                 \
1647     "psubh      "#R1",      "#R1",  "#R0"           \n\t"                   \
1648     "punpcklbh  "#R3",      "#R3",  $f0             \n\t"                   \
1649     "paddh      "#R1",      "#R1",  $f14            \n\t"                   \
1650     "psubh      "#R1",      "#R1",  "#R3"           \n\t"                   \
1651     "psrah      "#R1",      "#R1",  %[shift]        \n\t"                   \
1652     MMI_SDC1(R1, %[dst], OFF)                                               \
1653     PTR_ADDU    "%[src],    %[src], %[stride]       \n\t"
1654 
1655 /** Sacrificing $f12 makes it possible to pipeline loads from src */
vc1_put_ver_16b_shift2_mmi(int16_t * dst,const uint8_t * src,mips_reg stride,int rnd,int64_t shift)1656 static void vc1_put_ver_16b_shift2_mmi(int16_t *dst,
1657                                        const uint8_t *src, mips_reg stride,
1658                                        int rnd, int64_t shift)
1659 {
1660     DECLARE_VAR_LOW32;
1661     DECLARE_VAR_ADDRT;
1662 
1663     __asm__ volatile(
1664         "xor        $f0,    $f0,    $f0             \n\t"
1665         "li         $8,     0x03                    \n\t"
1666         LOAD_ROUNDER_MMI("%[rnd]")
1667         "ldc1       $f12,   %[ff_pw_9]              \n\t"
1668         "1:                                         \n\t"
1669         MMI_ULWC1($f4, %[src], 0x00)
1670         PTR_ADDU   "%[src], %[src], %[stride]       \n\t"
1671         MMI_ULWC1($f6, %[src], 0x00)
1672         "punpcklbh  $f4,    $f4,    $f0             \n\t"
1673         "punpcklbh  $f6,    $f6,    $f0             \n\t"
1674         SHIFT2_LINE(  0, $f2, $f4, $f6, $f8)
1675         SHIFT2_LINE( 24, $f4, $f6, $f8, $f2)
1676         SHIFT2_LINE( 48, $f6, $f8, $f2, $f4)
1677         SHIFT2_LINE( 72, $f8, $f2, $f4, $f6)
1678         SHIFT2_LINE( 96, $f2, $f4, $f6, $f8)
1679         SHIFT2_LINE(120, $f4, $f6, $f8, $f2)
1680         SHIFT2_LINE(144, $f6, $f8, $f2, $f4)
1681         SHIFT2_LINE(168, $f8, $f2, $f4, $f6)
1682         PTR_SUBU   "%[src], %[src], %[stride2]      \n\t"
1683         PTR_ADDIU  "%[dst], %[dst], 0x08            \n\t"
1684         "addiu      $8,     $8,    -0x01            \n\t"
1685         "bnez       $8,     1b                      \n\t"
1686         : RESTRICT_ASM_LOW32            RESTRICT_ASM_ADDRT
1687           [src]"+r"(src),               [dst]"+r"(dst)
1688         : [stride]"r"(stride),          [stride1]"r"(-2*stride),
1689           [shift]"f"(shift),            [rnd]"m"(rnd),
1690           [stride2]"r"(9*stride-4),     [ff_pw_9]"m"(ff_pw_9)
1691         : "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
1692           "$f14", "$f16", "memory"
1693     );
1694 }
1695 
1696 /**
1697  * Data is already unpacked, so some operations can directly be made from
1698  * memory.
1699  */
1700 #define VC1_HOR_16B_SHIFT2(OP, OPNAME)                                      \
1701 static void OPNAME ## vc1_hor_16b_shift2_mmi(uint8_t *dst, mips_reg stride, \
1702                                              const int16_t *src, int rnd)   \
1703 {                                                                           \
1704     int h = 8;                                                              \
1705     DECLARE_VAR_ALL64;                                                      \
1706     DECLARE_VAR_ADDRT;                                                      \
1707                                                                             \
1708     src -= 1;                                                               \
1709     rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */                            \
1710                                                                             \
1711     __asm__ volatile(                                                       \
1712         LOAD_ROUNDER_MMI("%[rnd]")                                          \
1713         "ldc1       $f12,   %[ff_pw_128]            \n\t"                   \
1714         "ldc1       $f10,   %[ff_pw_9]              \n\t"                   \
1715         "1:                                         \n\t"                   \
1716         MMI_ULDC1($f2, %[src], 0x00)                                        \
1717         MMI_ULDC1($f4, %[src], 0x08)                                        \
1718         MMI_ULDC1($f6, %[src], 0x02)                                        \
1719         MMI_ULDC1($f8, %[src], 0x0a)                                        \
1720         MMI_ULDC1($f0, %[src], 0x06)                                        \
1721         "paddh      $f2,    $f2,    $f0             \n\t"                   \
1722         MMI_ULDC1($f0, %[src], 0x0e)                                        \
1723         "paddh      $f4,    $f4,    $f0             \n\t"                   \
1724         MMI_ULDC1($f0, %[src], 0x04)                                        \
1725         "paddh      $f6,    $f6,    $f0             \n\t"                   \
1726         MMI_ULDC1($f0, %[src], 0x0b)                                        \
1727         "paddh      $f8,    $f8,    $f0             \n\t"                   \
1728         "pmullh     $f6,    $f6,    $f10            \n\t"                   \
1729         "pmullh     $f8,    $f8,    $f10            \n\t"                   \
1730         "psubh      $f6,    $f6,    $f2             \n\t"                   \
1731         "psubh      $f8,    $f8,    $f4             \n\t"                   \
1732         "li         $8,     0x07                    \n\t"                   \
1733         "mtc1       $8,     $f16                    \n\t"                   \
1734         NORMALIZE_MMI("$f16")                                               \
1735         /* Remove bias */                                                   \
1736         "paddh      $f6,    $f6,    $f12            \n\t"                   \
1737         "paddh      $f8,    $f8,    $f12            \n\t"                   \
1738         TRANSFER_DO_PACK(OP)                                                \
1739         "addiu      %[h],   %[h],  -0x01            \n\t"                   \
1740         PTR_ADDIU  "%[src], %[src], 0x18            \n\t"                   \
1741         PTR_ADDU   "%[dst], %[dst], %[stride]       \n\t"                   \
1742         "bnez       %[h],   1b                      \n\t"                   \
1743         : RESTRICT_ASM_ALL64            RESTRICT_ASM_ADDRT                  \
1744           [h]"+r"(h),                                                       \
1745           [src]"+r"(src),               [dst]"+r"(dst)                      \
1746         : [stride]"r"(stride),          [rnd]"m"(rnd),                      \
1747           [ff_pw_9]"m"(ff_pw_9),        [ff_pw_128]"m"(ff_pw_128)           \
1748         : "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12", "$f14",  \
1749           "$f16", "memory"                                                  \
1750     );                                                                      \
1751 }
1752 
1753 VC1_HOR_16B_SHIFT2(OP_PUT, put_)
1754 VC1_HOR_16B_SHIFT2(OP_AVG, avg_)
1755 
1756 /**
1757  * Purely vertical or horizontal 1/2 shift interpolation.
1758  * Sacrify $f12 for *9 factor.
1759  */
1760 #define VC1_SHIFT2(OP, OPNAME)\
1761 static void OPNAME ## vc1_shift2_mmi(uint8_t *dst, const uint8_t *src,      \
1762                                      mips_reg stride, int rnd,              \
1763                                      mips_reg offset)                       \
1764 {                                                                           \
1765     DECLARE_VAR_LOW32;                                                      \
1766     DECLARE_VAR_ADDRT;                                                      \
1767                                                                             \
1768     rnd = 8 - rnd;                                                          \
1769                                                                             \
1770     __asm__ volatile(                                                       \
1771         "xor        $f0,    $f0,    $f0             \n\t"                   \
1772         "li         $10,    0x08                    \n\t"                   \
1773         LOAD_ROUNDER_MMI("%[rnd]")                                          \
1774         "ldc1       $f12,   %[ff_pw_9]              \n\t"                   \
1775         "1:                                         \n\t"                   \
1776         MMI_ULWC1($f6, %[src], 0x00)                                        \
1777         MMI_ULWC1($f8, %[src], 0x04)                                        \
1778         PTR_ADDU   "$9,     %[src], %[offset]       \n\t"                   \
1779         MMI_ULWC1($f2, $9, 0x00)                                            \
1780         MMI_ULWC1($f4, $9, 0x04)                                            \
1781         PTR_ADDU   "%[src], %[src], %[offset]       \n\t"                   \
1782         "punpcklbh  $f6,    $f6,    $f0             \n\t"                   \
1783         "punpcklbh  $f8,    $f8,    $f0             \n\t"                   \
1784         "punpcklbh  $f2,    $f2,    $f0             \n\t"                   \
1785         "punpcklbh  $f4,    $f4,    $f0             \n\t"                   \
1786         "paddh      $f6,    $f6,    $f2             \n\t"                   \
1787         "paddh      $f8,    $f8,    $f4             \n\t"                   \
1788         PTR_ADDU   "$9,     %[src], %[offset_x2n]   \n\t"                   \
1789         MMI_ULWC1($f2, $9, 0x00)                                            \
1790         MMI_ULWC1($f4, $9, 0x04)                                            \
1791         "pmullh     $f6,    $f6,    $f12            \n\t" /* 0,9,9,0*/      \
1792         "pmullh     $f8,    $f8,    $f12            \n\t" /* 0,9,9,0*/      \
1793         "punpcklbh  $f2,    $f2,    $f0             \n\t"                   \
1794         "punpcklbh  $f4,    $f4,    $f0             \n\t"                   \
1795         "psubh      $f6,    $f6,    $f2             \n\t" /*-1,9,9,0*/      \
1796         "psubh      $f8,    $f8,    $f4             \n\t" /*-1,9,9,0*/      \
1797         PTR_ADDU   "$9,     %[src], %[offset]       \n\t"                   \
1798         MMI_ULWC1($f2, $9, 0x00)                                            \
1799         MMI_ULWC1($f4, $9, 0x04)                                            \
1800         "punpcklbh  $f2,    $f2,    $f0             \n\t"                   \
1801         "punpcklbh  $f4,    $f4,    $f0             \n\t"                   \
1802         "psubh      $f6,    $f6,    $f2             \n\t" /*-1,9,9,-1*/     \
1803         "psubh      $f8,    $f8,    $f4             \n\t" /*-1,9,9,-1*/     \
1804         "li         $8,     0x04                    \n\t"                   \
1805         "mtc1       $8,     $f16                    \n\t"                   \
1806         NORMALIZE_MMI("$f16")                                               \
1807         "packushb   $f6,    $f6,    $f8             \n\t"                   \
1808         OP((%[dst]), $f6)                                                   \
1809         "sdc1       $f6,    0x00(%[dst])            \n\t"                   \
1810         "addiu      $10,    $10,   -0x01            \n\t"                   \
1811         PTR_ADDU   "%[src], %[src], %[stride1]      \n\t"                   \
1812         PTR_ADDU   "%[dst], %[dst], %[stride]       \n\t"                   \
1813         "bnez       $10,    1b                      \n\t"                   \
1814         : RESTRICT_ASM_LOW32            RESTRICT_ASM_ADDRT                  \
1815           [src]"+r"(src),               [dst]"+r"(dst)                      \
1816         : [offset]"r"(offset),          [offset_x2n]"r"(-2*offset),         \
1817           [stride]"r"(stride),          [rnd]"m"(rnd),                      \
1818           [stride1]"r"(stride-offset),                                      \
1819           [ff_pw_9]"m"(ff_pw_9)                                             \
1820         : "$8", "$9", "$10", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10",     \
1821           "$f12", "$f14", "$f16", "memory"                                  \
1822     );                                                                      \
1823 }
1824 
1825 VC1_SHIFT2(OP_PUT, put_)
1826 VC1_SHIFT2(OP_AVG, avg_)
1827 
1828 /**
1829  * Core of the 1/4 and 3/4 shift bicubic interpolation.
1830  *
1831  * @param UNPACK  Macro unpacking arguments from 8 to 16bits (can be empty).
1832  * @param LOAD    "MMI_ULWC1" or "MMI_ULDC1", if data read is already unpacked.
1833  * @param M       "1" for MMI_ULWC1, "2" for MMI_ULDC1.
1834  * @param A1      Stride address of 1st tap (beware of unpacked/packed).
1835  * @param A2      Stride address of 2nd tap
1836  * @param A3      Stride address of 3rd tap
1837  * @param A4      Stride address of 4th tap
1838  */
1839 #define MSPEL_FILTER13_CORE(UNPACK, LOAD, M, A1, A2, A3, A4)                \
1840     PTR_ADDU   "$9,     %[src], "#A1"           \n\t"                       \
1841     LOAD($f2, $9, M*0)                                                      \
1842     LOAD($f4, $9, M*4)                                                      \
1843     UNPACK("$f2")                                                           \
1844     UNPACK("$f4")                                                           \
1845     "pmullh     $f2,    $f2,    %[ff_pw_3]      \n\t"                       \
1846     "pmullh     $f4,    $f4,    %[ff_pw_3]      \n\t"                       \
1847     PTR_ADDU   "$9,     %[src], "#A2"           \n\t"                       \
1848     LOAD($f6, $9, M*0)                                                      \
1849     LOAD($f8, $9, M*4)                                                      \
1850     UNPACK("$f6")                                                           \
1851     UNPACK("$f8")                                                           \
1852     "pmullh     $f6,    $f6,    $f12            \n\t" /* *18 */             \
1853     "pmullh     $f8,    $f8,    $f12            \n\t" /* *18 */             \
1854     "psubh      $f6,    $f6,    $f2             \n\t" /* *18, -3 */         \
1855     "psubh      $f8,    $f8,    $f4             \n\t" /* *18, -3 */         \
1856     PTR_ADDU   "$9,     %[src], "#A4"           \n\t"                       \
1857     LOAD($f2, $9, M*0)                                                      \
1858     LOAD($f4, $9, M*4)                                                      \
1859     UNPACK("$f2")                                                           \
1860     UNPACK("$f4")                                                           \
1861     "li         $8,     0x02                    \n\t"                       \
1862     "mtc1       $8,     $f16                    \n\t"                       \
1863     "psllh      $f2,    $f2,    $f16            \n\t" /* 4* */              \
1864     "psllh      $f4,    $f4,    $f16            \n\t" /* 4* */              \
1865     "psubh      $f6,    $f6,    $f2             \n\t" /* -4,18,-3 */        \
1866     "psubh      $f8,    $f8,    $f4             \n\t" /* -4,18,-3 */        \
1867     PTR_ADDU   "$9,     %[src], "#A3"           \n\t"                       \
1868     LOAD($f2, $9, M*0)                                                      \
1869     LOAD($f4, $9, M*4)                                                      \
1870     UNPACK("$f2")                                                           \
1871     UNPACK("$f4")                                                           \
1872     "pmullh     $f2,    $f2,    $f10            \n\t" /* *53 */             \
1873     "pmullh     $f4,    $f4,    $f10            \n\t" /* *53 */             \
1874     "paddh      $f6,    $f6,    $f2             \n\t" /* 4,53,18,-3 */      \
1875     "paddh      $f8,    $f8,    $f4             \n\t" /* 4,53,18,-3 */
1876 
1877 /**
1878  * Macro to build the vertical 16bits version of vc1_put_shift[13].
1879  * Here, offset=src_stride. Parameters passed A1 to A4 must use
1880  * %3 (src_stride), %4 (2*src_stride) and %5 (3*src_stride).
1881  *
1882  * @param  NAME   Either 1 or 3
1883  * @see MSPEL_FILTER13_CORE for information on A1->A4
1884  */
1885 #define MSPEL_FILTER13_VER_16B(NAME, A1, A2, A3, A4)                        \
1886 static void                                                                 \
1887 vc1_put_ver_16b_ ## NAME ## _mmi(int16_t *dst, const uint8_t *src,          \
1888                                  mips_reg src_stride,                       \
1889                                  int rnd, int64_t shift)                    \
1890 {                                                                           \
1891     int h = 8;                                                              \
1892     DECLARE_VAR_LOW32;                                                      \
1893     DECLARE_VAR_ADDRT;                                                      \
1894                                                                             \
1895     src -= src_stride;                                                      \
1896                                                                             \
1897     __asm__ volatile(                                                       \
1898         "xor        $f0,    $f0,    $f0             \n\t"                   \
1899         LOAD_ROUNDER_MMI("%[rnd]")                                          \
1900         "ldc1       $f10,   %[ff_pw_53]             \n\t"                   \
1901         "ldc1       $f12,   %[ff_pw_18]             \n\t"                   \
1902         ".p2align 3                                 \n\t"                   \
1903         "1:                                         \n\t"                   \
1904         MSPEL_FILTER13_CORE(DO_UNPACK, MMI_ULWC1, 1, A1, A2, A3, A4)        \
1905         NORMALIZE_MMI("%[shift]")                                           \
1906         TRANSFER_DONT_PACK(OP_PUT)                                          \
1907         /* Last 3 (in fact 4) bytes on the line */                          \
1908         PTR_ADDU   "$9,     %[src], "#A1"           \n\t"                   \
1909         MMI_ULWC1($f2, $9, 0x08)                                            \
1910         DO_UNPACK("$f2")                                                    \
1911         "mov.d      $f6,    $f2                     \n\t"                   \
1912         "paddh      $f2,    $f2,    $f2             \n\t"                   \
1913         "paddh      $f2,    $f2,    $f6             \n\t" /* 3* */          \
1914         PTR_ADDU   "$9,     %[src], "#A2"           \n\t"                   \
1915         MMI_ULWC1($f6, $9, 0x08)                                            \
1916         DO_UNPACK("$f6")                                                    \
1917         "pmullh     $f6,    $f6,    $f12            \n\t" /* *18 */         \
1918         "psubh      $f6,    $f6,    $f2             \n\t" /* *18,-3 */      \
1919         PTR_ADDU   "$9,     %[src], "#A3"           \n\t"                   \
1920         MMI_ULWC1($f2, $9, 0x08)                                            \
1921         DO_UNPACK("$f2")                                                    \
1922         "pmullh     $f2,    $f2,    $f10            \n\t" /* *53 */         \
1923         "paddh      $f6,    $f6,    $f2             \n\t" /* *53,18,-3 */   \
1924         PTR_ADDU   "$9,     %[src], "#A4"           \n\t"                   \
1925         MMI_ULWC1($f2, $9, 0x08)                                            \
1926         DO_UNPACK("$f2")                                                    \
1927         "li         $8,     0x02                    \n\t"                   \
1928         "mtc1       $8,     $f16                    \n\t"                   \
1929         "psllh      $f2,    $f2,    $f16            \n\t" /* 4* */          \
1930         "psubh      $f6,    $f6,    $f2             \n\t"                   \
1931         "paddh      $f6,    $f6,    $f14            \n\t"                   \
1932         "li         $8,     0x06                    \n\t"                   \
1933         "mtc1       $8,     $f16                    \n\t"                   \
1934         "psrah      $f6,    $f6,    $f16            \n\t"                   \
1935         "sdc1       $f6,    0x10(%[dst])            \n\t"                   \
1936         "addiu      %[h],   %[h],  -0x01            \n\t"                   \
1937         PTR_ADDU   "%[src], %[src], %[stride_x1]    \n\t"                   \
1938         PTR_ADDIU  "%[dst], %[dst], 0x18            \n\t"                   \
1939         "bnez       %[h],   1b                      \n\t"                   \
1940         : RESTRICT_ASM_LOW32            RESTRICT_ASM_ADDRT                  \
1941           [h]"+r"(h),                                                       \
1942           [src]"+r"(src),               [dst]"+r"(dst)                      \
1943         : [stride_x1]"r"(src_stride),   [stride_x2]"r"(2*src_stride),       \
1944           [stride_x3]"r"(3*src_stride),                                     \
1945           [rnd]"m"(rnd),                [shift]"f"(shift),                  \
1946           [ff_pw_53]"m"(ff_pw_53),      [ff_pw_18]"m"(ff_pw_18),            \
1947           [ff_pw_3]"f"(ff_pw_3)                                             \
1948         : "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",    \
1949           "$f14", "$f16", "memory"                                          \
1950     );                                                                      \
1951 }
1952 
1953 /**
1954  * Macro to build the horizontal 16bits version of vc1_put_shift[13].
1955  * Here, offset=16bits, so parameters passed A1 to A4 should be simple.
1956  *
1957  * @param  NAME   Either 1 or 3
1958  * @see MSPEL_FILTER13_CORE for information on A1->A4
1959  */
1960 #define MSPEL_FILTER13_HOR_16B(NAME, A1, A2, A3, A4, OP, OPNAME)            \
1961 static void                                                                 \
1962 OPNAME ## vc1_hor_16b_ ## NAME ## _mmi(uint8_t *dst, mips_reg stride,       \
1963                                        const int16_t *src, int rnd)         \
1964 {                                                                           \
1965     int h = 8;                                                              \
1966     DECLARE_VAR_ALL64;                                                      \
1967     DECLARE_VAR_ADDRT;                                                      \
1968                                                                             \
1969     src -= 1;                                                               \
1970     rnd -= (-4+58+13-3)*256; /* Add -256 bias */                            \
1971                                                                             \
1972     __asm__ volatile(                                                       \
1973         "xor        $f0,    $f0,    $f0             \n\t"                   \
1974         LOAD_ROUNDER_MMI("%[rnd]")                                          \
1975         "ldc1       $f10,   %[ff_pw_53]             \n\t"                   \
1976         "ldc1       $f12,   %[ff_pw_18]             \n\t"                   \
1977         ".p2align 3                                 \n\t"                   \
1978         "1:                                         \n\t"                   \
1979         MSPEL_FILTER13_CORE(DONT_UNPACK, MMI_ULDC1, 2, A1, A2, A3, A4)      \
1980         "li         $8,     0x07                    \n\t"                   \
1981         "mtc1       $8,     $f16                    \n\t"                   \
1982         NORMALIZE_MMI("$f16")                                               \
1983         /* Remove bias */                                                   \
1984         "paddh      $f6,    $f6,    %[ff_pw_128]    \n\t"                   \
1985         "paddh      $f8,    $f8,    %[ff_pw_128]    \n\t"                   \
1986         TRANSFER_DO_PACK(OP)                                                \
1987         "addiu      %[h],   %[h],  -0x01            \n\t"                   \
1988         PTR_ADDU   "%[src], %[src], 0x18            \n\t"                   \
1989         PTR_ADDU   "%[dst], %[dst], %[stride]       \n\t"                   \
1990         "bnez       %[h],   1b                      \n\t"                   \
1991         : RESTRICT_ASM_ALL64            RESTRICT_ASM_ADDRT                  \
1992           [h]"+r"(h),                                                       \
1993           [src]"+r"(src),               [dst]"+r"(dst)                      \
1994         : [stride]"r"(stride),          [rnd]"m"(rnd),                      \
1995           [ff_pw_53]"m"(ff_pw_53),      [ff_pw_18]"m"(ff_pw_18),            \
1996           [ff_pw_3]"f"(ff_pw_3),        [ff_pw_128]"f"(ff_pw_128)           \
1997         : "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",    \
1998           "$f14", "$f16", "memory"                                          \
1999     );                                                                      \
2000 }
2001 
2002 /**
2003  * Macro to build the 8bits, any direction, version of vc1_put_shift[13].
2004  * Here, offset=src_stride. Parameters passed A1 to A4 must use
2005  * %3 (offset), %4 (2*offset) and %5 (3*offset).
2006  *
2007  * @param  NAME   Either 1 or 3
2008  * @see MSPEL_FILTER13_CORE for information on A1->A4
2009  */
2010 #define MSPEL_FILTER13_8B(NAME, A1, A2, A3, A4, OP, OPNAME)                 \
2011 static void                                                                 \
2012 OPNAME ## vc1_## NAME ## _mmi(uint8_t *dst, const uint8_t *src,             \
2013                               mips_reg stride, int rnd, mips_reg offset)    \
2014 {                                                                           \
2015     int h = 8;                                                              \
2016     DECLARE_VAR_LOW32;                                                      \
2017     DECLARE_VAR_ADDRT;                                                      \
2018                                                                             \
2019     src -= offset;                                                          \
2020     rnd = 32-rnd;                                                           \
2021                                                                             \
2022     __asm__ volatile (                                                      \
2023         "xor        $f0,    $f0,    $f0             \n\t"                   \
2024         LOAD_ROUNDER_MMI("%[rnd]")                                          \
2025         "ldc1       $f10,   %[ff_pw_53]             \n\t"                   \
2026         "ldc1       $f12,   %[ff_pw_18]             \n\t"                   \
2027         ".p2align 3                                 \n\t"                   \
2028         "1:                                         \n\t"                   \
2029         MSPEL_FILTER13_CORE(DO_UNPACK, MMI_ULWC1, 1, A1, A2, A3, A4)        \
2030         "li         $8,     0x06                    \n\t"                   \
2031         "mtc1       $8,     $f16                    \n\t"                   \
2032         NORMALIZE_MMI("$f16")                                               \
2033         TRANSFER_DO_PACK(OP)                                                \
2034         "addiu      %[h],   %[h],      -0x01        \n\t"                   \
2035         PTR_ADDU   "%[src], %[src],     %[stride]   \n\t"                   \
2036         PTR_ADDU   "%[dst], %[dst],     %[stride]   \n\t"                   \
2037         "bnez       %[h],   1b                      \n\t"                   \
2038         : RESTRICT_ASM_LOW32            RESTRICT_ASM_ADDRT                  \
2039           [h]"+r"(h),                                                       \
2040           [src]"+r"(src),               [dst]"+r"(dst)                      \
2041         : [offset_x1]"r"(offset),       [offset_x2]"r"(2*offset),           \
2042           [offset_x3]"r"(3*offset),     [stride]"r"(stride),                \
2043           [rnd]"m"(rnd),                                                    \
2044           [ff_pw_53]"m"(ff_pw_53),      [ff_pw_18]"m"(ff_pw_18),            \
2045           [ff_pw_3]"f"(ff_pw_3)                                             \
2046         : "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",    \
2047           "$f14", "$f16", "memory"                                          \
2048     );                                                                      \
2049 }
2050 
2051 
2052 /** 1/4 shift bicubic interpolation */
2053 MSPEL_FILTER13_8B(shift1, %[offset_x3], %[offset_x2], %[offset_x1], $0, OP_PUT, put_)
2054 MSPEL_FILTER13_8B(shift1, %[offset_x3], %[offset_x2], %[offset_x1], $0, OP_AVG, avg_)
2055 MSPEL_FILTER13_VER_16B(shift1, %[stride_x3], %[stride_x2], %[stride_x1], $0)
2056 MSPEL_FILTER13_HOR_16B(shift1, 6, 4, 2, 0, OP_PUT, put_)
2057 MSPEL_FILTER13_HOR_16B(shift1, 6, 4, 2, 0, OP_AVG, avg_)
2058 
2059 /** 3/4 shift bicubic interpolation */
2060 MSPEL_FILTER13_8B(shift3, $0, %[offset_x1], %[offset_x2], %[offset_x3], OP_PUT, put_)
2061 MSPEL_FILTER13_8B(shift3, $0, %[offset_x1], %[offset_x2], %[offset_x3], OP_AVG, avg_)
2062 MSPEL_FILTER13_VER_16B(shift3, $0, %[stride_x1], %[stride_x2], %[stride_x3])
2063 MSPEL_FILTER13_HOR_16B(shift3, 0, 2, 4, 6, OP_PUT, put_)
2064 MSPEL_FILTER13_HOR_16B(shift3, 0, 2, 4, 6, OP_AVG, avg_)
2065 
2066 typedef void (*vc1_mspel_mc_filter_ver_16bits)
2067              (int16_t *dst, const uint8_t *src, mips_reg src_stride, int rnd,
2068               int64_t shift);
2069 typedef void (*vc1_mspel_mc_filter_hor_16bits)
2070              (uint8_t *dst, mips_reg dst_stride, const int16_t *src, int rnd);
2071 typedef void (*vc1_mspel_mc_filter_8bits)
2072              (uint8_t *dst, const uint8_t *src, mips_reg stride, int rnd,
2073               mips_reg offset);
2074 
2075 /**
2076  * Interpolate fractional pel values by applying proper vertical then
2077  * horizontal filter.
2078  *
2079  * @param  dst     Destination buffer for interpolated pels.
2080  * @param  src     Source buffer.
2081  * @param  stride  Stride for both src and dst buffers.
2082  * @param  hmode   Horizontal filter (expressed in quarter pixels shift).
2083  * @param  hmode   Vertical filter.
2084  * @param  rnd     Rounding bias.
2085  */
2086 #define VC1_MSPEL_MC(OP)                                                    \
2087 static void OP ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride,\
2088                                int hmode, int vmode, int rnd)               \
2089 {                                                                           \
2090     static const vc1_mspel_mc_filter_ver_16bits vc1_put_shift_ver_16bits[] =\
2091          { NULL, vc1_put_ver_16b_shift1_mmi,                                \
2092                  vc1_put_ver_16b_shift2_mmi,                                \
2093                  vc1_put_ver_16b_shift3_mmi };                              \
2094     static const vc1_mspel_mc_filter_hor_16bits vc1_put_shift_hor_16bits[] =\
2095          { NULL, OP ## vc1_hor_16b_shift1_mmi,                              \
2096                  OP ## vc1_hor_16b_shift2_mmi,                              \
2097                  OP ## vc1_hor_16b_shift3_mmi };                            \
2098     static const vc1_mspel_mc_filter_8bits vc1_put_shift_8bits[] =          \
2099          { NULL, OP ## vc1_shift1_mmi,                                      \
2100                  OP ## vc1_shift2_mmi,                                      \
2101                  OP ## vc1_shift3_mmi };                                    \
2102                                                                             \
2103     if (vmode) { /* Vertical filter to apply */                             \
2104         if (hmode) { /* Horizontal filter to apply, output to tmp */        \
2105             static const int shift_value[] = { 0, 5, 1, 5 };                \
2106             int    shift = (shift_value[hmode]+shift_value[vmode])>>1;      \
2107             int    r;                                                       \
2108             LOCAL_ALIGNED(16, int16_t, tmp, [12*8]);                        \
2109                                                                             \
2110             r = (1<<(shift-1)) + rnd-1;                                     \
2111             vc1_put_shift_ver_16bits[vmode](tmp, src-1, stride, r, shift);  \
2112                                                                             \
2113             vc1_put_shift_hor_16bits[hmode](dst, stride, tmp+1, 64-rnd);    \
2114             return;                                                         \
2115         }                                                                   \
2116         else { /* No horizontal filter, output 8 lines to dst */            \
2117             vc1_put_shift_8bits[vmode](dst, src, stride, 1-rnd, stride);    \
2118             return;                                                         \
2119         }                                                                   \
2120     }                                                                       \
2121                                                                             \
2122     /* Horizontal mode with no vertical mode */                             \
2123     vc1_put_shift_8bits[hmode](dst, src, stride, rnd, 1);                   \
2124 }                                                                           \
2125 static void OP ## vc1_mspel_mc_16(uint8_t *dst, const uint8_t *src,         \
2126                                   int stride, int hmode, int vmode, int rnd)\
2127 {                                                                           \
2128     OP ## vc1_mspel_mc(dst + 0, src + 0, stride, hmode, vmode, rnd);        \
2129     OP ## vc1_mspel_mc(dst + 8, src + 8, stride, hmode, vmode, rnd);        \
2130     dst += 8*stride; src += 8*stride;                                       \
2131     OP ## vc1_mspel_mc(dst + 0, src + 0, stride, hmode, vmode, rnd);        \
2132     OP ## vc1_mspel_mc(dst + 8, src + 8, stride, hmode, vmode, rnd);        \
2133 }
2134 
2135 VC1_MSPEL_MC(put_)
VC1_MSPEL_MC(avg_)2136 VC1_MSPEL_MC(avg_)
2137 
2138 /** Macro to ease bicubic filter interpolation functions declarations */
2139 #define DECLARE_FUNCTION(a, b)                                              \
2140 void ff_put_vc1_mspel_mc ## a ## b ## _mmi(uint8_t *dst,                    \
2141                                            const uint8_t *src,              \
2142                                            ptrdiff_t stride,                \
2143                                            int rnd)                         \
2144 {                                                                           \
2145      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \
2146 }                                                                           \
2147 void ff_avg_vc1_mspel_mc ## a ## b ## _mmi(uint8_t *dst,                    \
2148                                            const uint8_t *src,              \
2149                                            ptrdiff_t stride,                \
2150                                            int rnd)                         \
2151 {                                                                           \
2152      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \
2153 }                                                                           \
2154 void ff_put_vc1_mspel_mc ## a ## b ## _16_mmi(uint8_t *dst,                 \
2155                                               const uint8_t *src,           \
2156                                               ptrdiff_t stride,             \
2157                                               int rnd)                      \
2158 {                                                                           \
2159      put_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                      \
2160 }                                                                           \
2161 void ff_avg_vc1_mspel_mc ## a ## b ## _16_mmi(uint8_t *dst,                 \
2162                                               const uint8_t *src,           \
2163                                               ptrdiff_t stride,             \
2164                                               int rnd)                      \
2165 {                                                                           \
2166      avg_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                      \
2167 }
2168 
2169 DECLARE_FUNCTION(0, 1)
2170 DECLARE_FUNCTION(0, 2)
2171 DECLARE_FUNCTION(0, 3)
2172 
2173 DECLARE_FUNCTION(1, 0)
2174 DECLARE_FUNCTION(1, 1)
2175 DECLARE_FUNCTION(1, 2)
2176 DECLARE_FUNCTION(1, 3)
2177 
2178 DECLARE_FUNCTION(2, 0)
2179 DECLARE_FUNCTION(2, 1)
2180 DECLARE_FUNCTION(2, 2)
2181 DECLARE_FUNCTION(2, 3)
2182 
2183 DECLARE_FUNCTION(3, 0)
2184 DECLARE_FUNCTION(3, 1)
2185 DECLARE_FUNCTION(3, 2)
2186 DECLARE_FUNCTION(3, 3)
2187 
2188 #define CHROMA_MC_8_MMI                                                     \
2189         "punpckhbh  %[ftmp5],   %[ftmp1],   %[ftmp0]                \n\t"   \
2190         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                \n\t"   \
2191         "punpckhbh  %[ftmp6],   %[ftmp2],   %[ftmp0]                \n\t"   \
2192         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                \n\t"   \
2193         "punpckhbh  %[ftmp7],   %[ftmp3],   %[ftmp0]                \n\t"   \
2194         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                \n\t"   \
2195         "punpckhbh  %[ftmp8],   %[ftmp4],   %[ftmp0]                \n\t"   \
2196         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                \n\t"   \
2197                                                                             \
2198         "pmullh     %[ftmp1],   %[ftmp1],   %[A]                    \n\t"   \
2199         "pmullh     %[ftmp5],   %[ftmp5],   %[A]                    \n\t"   \
2200         "pmullh     %[ftmp2],   %[ftmp2],   %[B]                    \n\t"   \
2201         "pmullh     %[ftmp6],   %[ftmp6],   %[B]                    \n\t"   \
2202         "pmullh     %[ftmp3],   %[ftmp3],   %[C]                    \n\t"   \
2203         "pmullh     %[ftmp7],   %[ftmp7],   %[C]                    \n\t"   \
2204         "pmullh     %[ftmp4],   %[ftmp4],   %[D]                    \n\t"   \
2205         "pmullh     %[ftmp8],   %[ftmp8],   %[D]                    \n\t"   \
2206                                                                             \
2207         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp2]                \n\t"   \
2208         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp4]                \n\t"   \
2209         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp3]                \n\t"   \
2210         "paddh      %[ftmp1],   %[ftmp1],   %[ff_pw_28]             \n\t"   \
2211                                                                             \
2212         "paddh      %[ftmp5],   %[ftmp5],   %[ftmp6]                \n\t"   \
2213         "paddh      %[ftmp7],   %[ftmp7],   %[ftmp8]                \n\t"   \
2214         "paddh      %[ftmp5],   %[ftmp5],   %[ftmp7]                \n\t"   \
2215         "paddh      %[ftmp5],   %[ftmp5],   %[ff_pw_28]             \n\t"   \
2216                                                                             \
2217         "psrlh      %[ftmp1],   %[ftmp1],   %[ftmp9]                \n\t"   \
2218         "psrlh      %[ftmp5],   %[ftmp5],   %[ftmp9]                \n\t"   \
2219         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp5]                \n\t"
2220 
2221 
2222 #define CHROMA_MC_4_MMI                                                     \
2223         "punpcklbh  %[ftmp1],   %[ftmp1],   %[ftmp0]                \n\t"   \
2224         "punpcklbh  %[ftmp2],   %[ftmp2],   %[ftmp0]                \n\t"   \
2225         "punpcklbh  %[ftmp3],   %[ftmp3],   %[ftmp0]                \n\t"   \
2226         "punpcklbh  %[ftmp4],   %[ftmp4],   %[ftmp0]                \n\t"   \
2227                                                                             \
2228         "pmullh     %[ftmp1],   %[ftmp1],   %[A]                    \n\t"   \
2229         "pmullh     %[ftmp2],   %[ftmp2],   %[B]                    \n\t"   \
2230         "pmullh     %[ftmp3],   %[ftmp3],   %[C]                    \n\t"   \
2231         "pmullh     %[ftmp4],   %[ftmp4],   %[D]                    \n\t"   \
2232                                                                             \
2233         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp2]                \n\t"   \
2234         "paddh      %[ftmp3],   %[ftmp3],   %[ftmp4]                \n\t"   \
2235         "paddh      %[ftmp1],   %[ftmp1],   %[ftmp3]                \n\t"   \
2236         "paddh      %[ftmp1],   %[ftmp1],   %[ff_pw_28]             \n\t"   \
2237                                                                             \
2238         "psrlh      %[ftmp1],   %[ftmp1],   %[ftmp5]                \n\t"   \
2239         "packushb   %[ftmp1],   %[ftmp1],   %[ftmp0]                \n\t"
2240 
2241 
2242 void ff_put_no_rnd_vc1_chroma_mc8_mmi(uint8_t *dst /* align 8 */,
2243                                       uint8_t *src /* align 1 */,
2244                                       int stride, int h, int x, int y)
2245 {
2246     const int A = (8 - x) * (8 - y);
2247     const int B =     (x) * (8 - y);
2248     const int C = (8 - x) *     (y);
2249     const int D =     (x) *     (y);
2250     double ftmp[10];
2251     uint32_t tmp[1];
2252     DECLARE_VAR_ALL64;
2253     DECLARE_VAR_ADDRT;
2254 
2255     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
2256 
2257     __asm__ volatile(
2258         "li         %[tmp0],    0x06                                    \n\t"
2259         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
2260         "mtc1       %[tmp0],    %[ftmp9]                                \n\t"
2261         "pshufh     %[A],       %[A],       %[ftmp0]                    \n\t"
2262         "pshufh     %[B],       %[B],       %[ftmp0]                    \n\t"
2263         "pshufh     %[C],       %[C],       %[ftmp0]                    \n\t"
2264         "pshufh     %[D],       %[D],       %[ftmp0]                    \n\t"
2265 
2266         "1:                                                             \n\t"
2267         MMI_ULDC1(%[ftmp1], %[src], 0x00)
2268         MMI_ULDC1(%[ftmp2], %[src], 0x01)
2269         PTR_ADDU   "%[src],     %[src],     %[stride]                   \n\t"
2270         MMI_ULDC1(%[ftmp3], %[src], 0x00)
2271         MMI_ULDC1(%[ftmp4], %[src], 0x01)
2272 
2273         CHROMA_MC_8_MMI
2274 
2275         MMI_SDC1(%[ftmp1], %[dst], 0x00)
2276         "addiu      %[h],       %[h],      -0x01                        \n\t"
2277         PTR_ADDU   "%[dst],     %[dst],     %[stride]                   \n\t"
2278         "bnez       %[h],       1b                                      \n\t"
2279         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
2280           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
2281           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
2282           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
2283           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
2284           RESTRICT_ASM_ALL64
2285           RESTRICT_ASM_ADDRT
2286           [tmp0]"=&r"(tmp[0]),
2287           [src]"+&r"(src),              [dst]"+&r"(dst),
2288           [h]"+&r"(h)
2289         : [stride]"r"((mips_reg)stride),
2290           [A]"f"(A),                    [B]"f"(B),
2291           [C]"f"(C),                    [D]"f"(D),
2292           [ff_pw_28]"f"(ff_pw_28)
2293         : "memory"
2294     );
2295 }
2296 
ff_put_no_rnd_vc1_chroma_mc4_mmi(uint8_t * dst,uint8_t * src,int stride,int h,int x,int y)2297 void ff_put_no_rnd_vc1_chroma_mc4_mmi(uint8_t *dst /* align 8 */,
2298                                       uint8_t *src /* align 1 */,
2299                                       int stride, int h, int x, int y)
2300 {
2301     const int A = (8 - x) * (8 - y);
2302     const int B =     (x) * (8 - y);
2303     const int C = (8 - x) *     (y);
2304     const int D =     (x) *     (y);
2305     double ftmp[6];
2306     uint32_t tmp[1];
2307     DECLARE_VAR_LOW32;
2308     DECLARE_VAR_ADDRT;
2309 
2310     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
2311 
2312     __asm__ volatile(
2313         "li         %[tmp0],    0x06                                    \n\t"
2314         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
2315         "mtc1       %[tmp0],    %[ftmp5]                                \n\t"
2316         "pshufh     %[A],       %[A],       %[ftmp0]                    \n\t"
2317         "pshufh     %[B],       %[B],       %[ftmp0]                    \n\t"
2318         "pshufh     %[C],       %[C],       %[ftmp0]                    \n\t"
2319         "pshufh     %[D],       %[D],       %[ftmp0]                    \n\t"
2320 
2321         "1:                                                             \n\t"
2322         MMI_ULWC1(%[ftmp1], %[src], 0x00)
2323         MMI_ULWC1(%[ftmp2], %[src], 0x01)
2324         PTR_ADDU   "%[src],     %[src],     %[stride]                   \n\t"
2325         MMI_ULWC1(%[ftmp3], %[src], 0x00)
2326         MMI_ULWC1(%[ftmp4], %[src], 0x01)
2327 
2328         CHROMA_MC_4_MMI
2329 
2330         MMI_SWC1(%[ftmp1], %[dst], 0x00)
2331         "addiu      %[h],       %[h],      -0x01                        \n\t"
2332         PTR_ADDU   "%[dst],     %[dst],     %[stride]                   \n\t"
2333         "bnez       %[h],       1b                                      \n\t"
2334         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
2335           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
2336           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
2337           [tmp0]"=&r"(tmp[0]),
2338           RESTRICT_ASM_LOW32
2339           RESTRICT_ASM_ADDRT
2340           [src]"+&r"(src),              [dst]"+&r"(dst),
2341           [h]"+&r"(h)
2342         : [stride]"r"((mips_reg)stride),
2343           [A]"f"(A),                    [B]"f"(B),
2344           [C]"f"(C),                    [D]"f"(D),
2345           [ff_pw_28]"f"(ff_pw_28)
2346         : "memory"
2347     );
2348 }
2349 
ff_avg_no_rnd_vc1_chroma_mc8_mmi(uint8_t * dst,uint8_t * src,int stride,int h,int x,int y)2350 void ff_avg_no_rnd_vc1_chroma_mc8_mmi(uint8_t *dst /* align 8 */,
2351                                       uint8_t *src /* align 1 */,
2352                                       int stride, int h, int x, int y)
2353 {
2354     const int A = (8 - x) * (8 - y);
2355     const int B =     (x) * (8 - y);
2356     const int C = (8 - x) *     (y);
2357     const int D =     (x) *     (y);
2358     double ftmp[10];
2359     uint32_t tmp[1];
2360     DECLARE_VAR_ALL64;
2361     DECLARE_VAR_ADDRT;
2362 
2363     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
2364 
2365     __asm__ volatile(
2366         "li         %[tmp0],    0x06                                    \n\t"
2367         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
2368         "mtc1       %[tmp0],    %[ftmp9]                                \n\t"
2369         "pshufh     %[A],       %[A],       %[ftmp0]                    \n\t"
2370         "pshufh     %[B],       %[B],       %[ftmp0]                    \n\t"
2371         "pshufh     %[C],       %[C],       %[ftmp0]                    \n\t"
2372         "pshufh     %[D],       %[D],       %[ftmp0]                    \n\t"
2373 
2374         "1:                                                             \n\t"
2375         MMI_ULDC1(%[ftmp1], %[src], 0x00)
2376         MMI_ULDC1(%[ftmp2], %[src], 0x01)
2377         PTR_ADDU   "%[src],     %[src],     %[stride]                   \n\t"
2378         MMI_ULDC1(%[ftmp3], %[src], 0x00)
2379         MMI_ULDC1(%[ftmp4], %[src], 0x01)
2380 
2381         CHROMA_MC_8_MMI
2382 
2383         MMI_LDC1(%[ftmp2], %[dst], 0x00)
2384         "pavgb      %[ftmp1],   %[ftmp1],   %[ftmp2]                    \n\t"
2385 
2386         MMI_SDC1(%[ftmp1], %[dst], 0x00)
2387         "addiu      %[h],       %[h],      -0x01                        \n\t"
2388         PTR_ADDU   "%[dst],     %[dst],     %[stride]                   \n\t"
2389         "bnez       %[h],       1b                                      \n\t"
2390         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
2391           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
2392           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
2393           [ftmp6]"=&f"(ftmp[6]),        [ftmp7]"=&f"(ftmp[7]),
2394           [ftmp8]"=&f"(ftmp[8]),        [ftmp9]"=&f"(ftmp[9]),
2395           [tmp0]"=&r"(tmp[0]),
2396           RESTRICT_ASM_ALL64
2397           RESTRICT_ASM_ADDRT
2398           [src]"+&r"(src),              [dst]"+&r"(dst),
2399           [h]"+&r"(h)
2400         : [stride]"r"((mips_reg)stride),
2401           [A]"f"(A),                    [B]"f"(B),
2402           [C]"f"(C),                    [D]"f"(D),
2403           [ff_pw_28]"f"(ff_pw_28)
2404         : "memory"
2405     );
2406 }
2407 
ff_avg_no_rnd_vc1_chroma_mc4_mmi(uint8_t * dst,uint8_t * src,int stride,int h,int x,int y)2408 void ff_avg_no_rnd_vc1_chroma_mc4_mmi(uint8_t *dst /* align 8 */,
2409                                       uint8_t *src /* align 1 */,
2410                                       int stride, int h, int x, int y)
2411 {
2412     const int A = (8 - x) * (8 - y);
2413     const int B = (    x) * (8 - y);
2414     const int C = (8 - x) * (    y);
2415     const int D = (    x) * (    y);
2416     double ftmp[6];
2417     uint32_t tmp[1];
2418     DECLARE_VAR_LOW32;
2419     DECLARE_VAR_ADDRT;
2420 
2421     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
2422 
2423     __asm__ volatile(
2424         "li         %[tmp0],    0x06                                    \n\t"
2425         "xor        %[ftmp0],   %[ftmp0],   %[ftmp0]                    \n\t"
2426         "mtc1       %[tmp0],    %[ftmp5]                                \n\t"
2427         "pshufh     %[A],       %[A],       %[ftmp0]                    \n\t"
2428         "pshufh     %[B],       %[B],       %[ftmp0]                    \n\t"
2429         "pshufh     %[C],       %[C],       %[ftmp0]                    \n\t"
2430         "pshufh     %[D],       %[D],       %[ftmp0]                    \n\t"
2431 
2432         "1:                                                             \n\t"
2433         MMI_ULWC1(%[ftmp1], %[src], 0x00)
2434         MMI_ULWC1(%[ftmp2], %[src], 0x01)
2435         PTR_ADDU   "%[src],     %[src],     %[stride]                   \n\t"
2436         MMI_ULWC1(%[ftmp3], %[src], 0x00)
2437         MMI_ULWC1(%[ftmp4], %[src], 0x01)
2438 
2439         CHROMA_MC_4_MMI
2440 
2441         MMI_LWC1(%[ftmp2], %[dst], 0x00)
2442         "pavgb      %[ftmp1],   %[ftmp1],   %[ftmp2]                    \n\t"
2443 
2444         MMI_SWC1(%[ftmp1], %[dst], 0x00)
2445         "addiu      %[h],       %[h],      -0x01                        \n\t"
2446         PTR_ADDU   "%[dst],     %[dst],     %[stride]                   \n\t"
2447         "bnez       %[h],       1b                                      \n\t"
2448         : [ftmp0]"=&f"(ftmp[0]),        [ftmp1]"=&f"(ftmp[1]),
2449           [ftmp2]"=&f"(ftmp[2]),        [ftmp3]"=&f"(ftmp[3]),
2450           [ftmp4]"=&f"(ftmp[4]),        [ftmp5]"=&f"(ftmp[5]),
2451           [tmp0]"=&r"(tmp[0]),
2452           RESTRICT_ASM_LOW32
2453           RESTRICT_ASM_ADDRT
2454           [src]"+&r"(src),              [dst]"+&r"(dst),
2455           [h]"+&r"(h)
2456         : [stride]"r"((mips_reg)stride),
2457           [A]"f"(A),                    [B]"f"(B),
2458           [C]"f"(C),                    [D]"f"(D),
2459           [ff_pw_28]"f"(ff_pw_28)
2460         : "memory"
2461     );
2462 }
2463