vp9: add 16x16 idct avx2 (8-bit).
[ffmpeg.git] / libavcodec / qpel_template.c
1 /*
2  * quarterpel DSP function templates
3  * Copyright (c) 2000, 2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
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 /**
24  * @file
25  * quarterpel DSP function templates
26  */
27
28 #define PIXOP2(OPNAME, OP)                                              \
29 static inline void OPNAME ## _no_rnd_pixels8_l2_8(uint8_t *dst,         \
30                                                   const uint8_t *src1,  \
31                                                   const uint8_t *src2,  \
32                                                   int dst_stride,       \
33                                                   int src_stride1,      \
34                                                   int src_stride2,      \
35                                                   int h)                \
36 {                                                                       \
37     int i;                                                              \
38                                                                         \
39     for (i = 0; i < h; i++) {                                           \
40         uint32_t a, b;                                                  \
41         a = AV_RN32(&src1[i * src_stride1]);                            \
42         b = AV_RN32(&src2[i * src_stride2]);                            \
43         OP(*((uint32_t *) &dst[i * dst_stride]),                        \
44            no_rnd_avg32(a, b));                                         \
45         a = AV_RN32(&src1[i * src_stride1 + 4]);                        \
46         b = AV_RN32(&src2[i * src_stride2 + 4]);                        \
47         OP(*((uint32_t *) &dst[i * dst_stride + 4]),                    \
48            no_rnd_avg32(a, b));                                         \
49     }                                                                   \
50 }                                                                       \
51                                                                         \
52 static inline void OPNAME ## _no_rnd_pixels16_l2_8(uint8_t *dst,        \
53                                                    const uint8_t *src1, \
54                                                    const uint8_t *src2, \
55                                                    int dst_stride,      \
56                                                    int src_stride1,     \
57                                                    int src_stride2,     \
58                                                    int h)               \
59 {                                                                       \
60     OPNAME ## _no_rnd_pixels8_l2_8(dst, src1, src2, dst_stride,         \
61                                    src_stride1, src_stride2, h);        \
62     OPNAME ## _no_rnd_pixels8_l2_8(dst  + 8,                            \
63                                    src1 + 8,                            \
64                                    src2 + 8,                            \
65                                    dst_stride, src_stride1,             \
66                                    src_stride2, h);                     \
67 }                                                                       \
68                                                                         \
69 static inline void OPNAME ## _pixels8_l4_8(uint8_t *dst,                \
70                                            const uint8_t *src1,         \
71                                            const uint8_t *src2,         \
72                                            const uint8_t *src3,         \
73                                            const uint8_t *src4,         \
74                                            int dst_stride,              \
75                                            int src_stride1,             \
76                                            int src_stride2,             \
77                                            int src_stride3,             \
78                                            int src_stride4,             \
79                                            int h)                       \
80 {                                                                       \
81     /* FIXME HIGH BIT DEPTH */                                          \
82     int i;                                                              \
83                                                                         \
84     for (i = 0; i < h; i++) {                                           \
85         uint32_t a, b, c, d, l0, l1, h0, h1;                            \
86         a  = AV_RN32(&src1[i * src_stride1]);                           \
87         b  = AV_RN32(&src2[i * src_stride2]);                           \
88         c  = AV_RN32(&src3[i * src_stride3]);                           \
89         d  = AV_RN32(&src4[i * src_stride4]);                           \
90         l0 = (a & 0x03030303UL) +                                       \
91              (b & 0x03030303UL) +                                       \
92                   0x02020202UL;                                         \
93         h0 = ((a & 0xFCFCFCFCUL) >> 2) +                                \
94              ((b & 0xFCFCFCFCUL) >> 2);                                 \
95         l1 = (c & 0x03030303UL) +                                       \
96              (d & 0x03030303UL);                                        \
97         h1 = ((c & 0xFCFCFCFCUL) >> 2) +                                \
98              ((d & 0xFCFCFCFCUL) >> 2);                                 \
99         OP(*((uint32_t *) &dst[i * dst_stride]),                        \
100            h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));                \
101         a  = AV_RN32(&src1[i * src_stride1 + 4]);                       \
102         b  = AV_RN32(&src2[i * src_stride2 + 4]);                       \
103         c  = AV_RN32(&src3[i * src_stride3 + 4]);                       \
104         d  = AV_RN32(&src4[i * src_stride4 + 4]);                       \
105         l0 = (a & 0x03030303UL) +                                       \
106              (b & 0x03030303UL) +                                       \
107                   0x02020202UL;                                         \
108         h0 = ((a & 0xFCFCFCFCUL) >> 2) +                                \
109              ((b & 0xFCFCFCFCUL) >> 2);                                 \
110         l1 = (c & 0x03030303UL) +                                       \
111              (d & 0x03030303UL);                                        \
112         h1 = ((c & 0xFCFCFCFCUL) >> 2) +                                \
113              ((d & 0xFCFCFCFCUL) >> 2);                                 \
114         OP(*((uint32_t *) &dst[i * dst_stride + 4]),                    \
115            h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));                \
116     }                                                                   \
117 }                                                                       \
118                                                                         \
119 static inline void OPNAME ## _no_rnd_pixels8_l4_8(uint8_t *dst,         \
120                                                   const uint8_t *src1,  \
121                                                   const uint8_t *src2,  \
122                                                   const uint8_t *src3,  \
123                                                   const uint8_t *src4,  \
124                                                   int dst_stride,       \
125                                                   int src_stride1,      \
126                                                   int src_stride2,      \
127                                                   int src_stride3,      \
128                                                   int src_stride4,      \
129                                                   int h)                \
130 {                                                                       \
131     /* FIXME HIGH BIT DEPTH */                                          \
132     int i;                                                              \
133                                                                         \
134     for (i = 0; i < h; i++) {                                           \
135         uint32_t a, b, c, d, l0, l1, h0, h1;                            \
136         a  = AV_RN32(&src1[i * src_stride1]);                           \
137         b  = AV_RN32(&src2[i * src_stride2]);                           \
138         c  = AV_RN32(&src3[i * src_stride3]);                           \
139         d  = AV_RN32(&src4[i * src_stride4]);                           \
140         l0 = (a & 0x03030303UL) +                                       \
141              (b & 0x03030303UL) +                                       \
142                   0x01010101UL;                                         \
143         h0 = ((a & 0xFCFCFCFCUL) >> 2) +                                \
144              ((b & 0xFCFCFCFCUL) >> 2);                                 \
145         l1 = (c & 0x03030303UL) +                                       \
146              (d & 0x03030303UL);                                        \
147         h1 = ((c & 0xFCFCFCFCUL) >> 2) +                                \
148              ((d & 0xFCFCFCFCUL) >> 2);                                 \
149         OP(*((uint32_t *) &dst[i * dst_stride]),                        \
150            h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));                \
151         a  = AV_RN32(&src1[i * src_stride1 + 4]);                       \
152         b  = AV_RN32(&src2[i * src_stride2 + 4]);                       \
153         c  = AV_RN32(&src3[i * src_stride3 + 4]);                       \
154         d  = AV_RN32(&src4[i * src_stride4 + 4]);                       \
155         l0 = (a & 0x03030303UL) +                                       \
156              (b & 0x03030303UL) +                                       \
157                   0x01010101UL;                                         \
158         h0 = ((a & 0xFCFCFCFCUL) >> 2) +                                \
159              ((b & 0xFCFCFCFCUL) >> 2);                                 \
160         l1 = (c & 0x03030303UL) +                                       \
161              (d & 0x03030303UL);                                        \
162         h1 = ((c & 0xFCFCFCFCUL) >> 2) +                                \
163              ((d & 0xFCFCFCFCUL) >> 2);                                 \
164         OP(*((uint32_t *) &dst[i * dst_stride + 4]),                    \
165            h0 + h1 + (((l0 + l1) >> 2) & 0x0F0F0F0FUL));                \
166     }                                                                   \
167 }                                                                       \
168                                                                         \
169 static inline void OPNAME ## _pixels16_l4_8(uint8_t *dst,               \
170                                             const uint8_t *src1,        \
171                                             const uint8_t *src2,        \
172                                             const uint8_t *src3,        \
173                                             const uint8_t *src4,        \
174                                             int dst_stride,             \
175                                             int src_stride1,            \
176                                             int src_stride2,            \
177                                             int src_stride3,            \
178                                             int src_stride4,            \
179                                             int h)                      \
180 {                                                                       \
181     OPNAME ## _pixels8_l4_8(dst, src1, src2, src3, src4, dst_stride,    \
182                             src_stride1, src_stride2, src_stride3,      \
183                             src_stride4, h);                            \
184     OPNAME ## _pixels8_l4_8(dst  + 8,                                   \
185                             src1 + 8, src2 + 8,                         \
186                             src3 + 8, src4 + 8,                         \
187                             dst_stride, src_stride1, src_stride2,       \
188                             src_stride3, src_stride4, h);               \
189 }                                                                       \
190                                                                         \
191 static inline void OPNAME ## _no_rnd_pixels16_l4_8(uint8_t *dst,        \
192                                                    const uint8_t *src1, \
193                                                    const uint8_t *src2, \
194                                                    const uint8_t *src3, \
195                                                    const uint8_t *src4, \
196                                                    int dst_stride,      \
197                                                    int src_stride1,     \
198                                                    int src_stride2,     \
199                                                    int src_stride3,     \
200                                                    int src_stride4,     \
201                                                    int h)               \
202 {                                                                       \
203     OPNAME ## _no_rnd_pixels8_l4_8(dst, src1, src2, src3, src4,         \
204                                    dst_stride, src_stride1,             \
205                                    src_stride2, src_stride3,            \
206                                    src_stride4, h);                     \
207     OPNAME ## _no_rnd_pixels8_l4_8(dst  + 8,                            \
208                                    src1 + 8, src2 + 8,                  \
209                                    src3 + 8, src4 + 8,                  \
210                                    dst_stride, src_stride1,             \
211                                    src_stride2, src_stride3,            \
212                                    src_stride4, h);                     \
213 }                                                                       \
214
215 #define op_avg(a, b) a = rnd_avg32(a, b)
216 #define op_put(a, b) a = b
217 #define put_no_rnd_pixels8_8_c put_pixels8_8_c
218 PIXOP2(avg, op_avg)
219 PIXOP2(put, op_put)
220 #undef op_avg
221 #undef op_put