1/*
2 * Copyright (C) 2009 David Conrad
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include "dsputil.h"
22#include "diracdsp.h"
23#include "libavcodec/x86/diracdsp_mmx.h"
24
25#define FILTER(src, stride)                                     \
26    ((21*((src)[ 0*stride] + (src)[1*stride])                   \
27      -7*((src)[-1*stride] + (src)[2*stride])                   \
28      +3*((src)[-2*stride] + (src)[3*stride])                   \
29      -1*((src)[-3*stride] + (src)[4*stride]) + 16) >> 5)
30
31static void dirac_hpel_filter(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, const uint8_t *src,
32                              int stride, int width, int height)
33{
34    int x, y;
35
36    for (y = 0; y < height; y++) {
37        for (x = -3; x < width+5; x++)
38            dstv[x] = av_clip_uint8(FILTER(src+x, stride));
39
40        for (x = 0; x < width; x++)
41            dstc[x] = av_clip_uint8(FILTER(dstv+x, 1));
42
43        for (x = 0; x < width; x++)
44            dsth[x] = av_clip_uint8(FILTER(src+x, 1));
45
46        src  += stride;
47        dsth += stride;
48        dstv += stride;
49        dstc += stride;
50    }
51}
52
53#define PIXOP_BILINEAR(PFX, OP, WIDTH)                                  \
54    static void ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c(uint8_t *dst, const uint8_t *src[5], int stride, int h) \
55    {                                                                   \
56        int x;                                                          \
57        const uint8_t *s0 = src[0];                                     \
58        const uint8_t *s1 = src[1];                                     \
59        const uint8_t *s2 = src[2];                                     \
60        const uint8_t *s3 = src[3];                                     \
61        const uint8_t *w  = src[4];                                     \
62                                                                        \
63        while (h--) {                                                   \
64            for (x = 0; x < WIDTH; x++) {                               \
65                OP(dst[x], (s0[x]*w[0] + s1[x]*w[1] + s2[x]*w[2] + s3[x]*w[3] + 8) >> 4); \
66            }                                                           \
67                                                                        \
68            dst += stride;                                              \
69            s0 += stride;                                               \
70            s1 += stride;                                               \
71            s2 += stride;                                               \
72            s3 += stride;                                               \
73        }                                                               \
74    }
75
76#define OP_PUT(dst, val) (dst) = (val)
77#define OP_AVG(dst, val) (dst) = (((dst) + (val) + 1)>>1)
78
79PIXOP_BILINEAR(put, OP_PUT, 8)
80PIXOP_BILINEAR(put, OP_PUT, 16)
81PIXOP_BILINEAR(put, OP_PUT, 32)
82PIXOP_BILINEAR(avg, OP_AVG, 8)
83PIXOP_BILINEAR(avg, OP_AVG, 16)
84PIXOP_BILINEAR(avg, OP_AVG, 32)
85
86#define op_scale1(x)  block[x] = av_clip_uint8( (block[x]*weight + (1<<(log2_denom-1))) >> log2_denom)
87#define op_scale2(x)  dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + (1<<(log2_denom-1))) >> log2_denom)
88
89#define DIRAC_WEIGHT(W)                                                 \
90    static void weight_dirac_pixels ## W ## _c(uint8_t *block, int stride, int log2_denom, \
91                                               int weight, int h) {     \
92        int x;                                                          \
93        while (h--) {                                                   \
94            for (x = 0; x < W; x++) {                                   \
95                op_scale1(x);                                           \
96                op_scale1(x+1);                                         \
97            }                                                           \
98            block += stride;                                            \
99        }                                                               \
100    }                                                                   \
101    static void biweight_dirac_pixels ## W ## _c(uint8_t *dst, const uint8_t *src, int stride, int log2_denom, \
102                                                 int weightd, int weights, int h) { \
103        int x;                                                          \
104        while (h--) {                                                   \
105            for (x = 0; x < W; x++) {                                   \
106                op_scale2(x);                                           \
107                op_scale2(x+1);                                         \
108            }                                                           \
109            dst += stride;                                              \
110            src += stride;                                              \
111        }                                                               \
112    }
113
114DIRAC_WEIGHT(8)
115DIRAC_WEIGHT(16)
116DIRAC_WEIGHT(32)
117
118#define ADD_OBMC(xblen)                                                 \
119    static void add_obmc ## xblen ## _c(uint16_t *dst, const uint8_t *src, int stride, \
120                                        const uint8_t *obmc_weight, int yblen) \
121    {                                                                   \
122        int x;                                                          \
123        while (yblen--) {                                               \
124            for (x = 0; x < xblen; x += 2) {                            \
125                dst[x  ] += src[x  ] * obmc_weight[x  ];                \
126                dst[x+1] += src[x+1] * obmc_weight[x+1];                \
127            }                                                           \
128            dst += stride;                                              \
129            src += stride;                                              \
130            obmc_weight += 32;                                          \
131        }                                                               \
132    }
133
134ADD_OBMC(8)
135ADD_OBMC(16)
136ADD_OBMC(32)
137
138static void put_signed_rect_clamped_c(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height)
139{
140    int x, y;
141    for (y = 0; y < height; y++) {
142        for (x = 0; x < width; x+=4) {
143            dst[x  ] = av_clip_uint8(src[x  ] + 128);
144            dst[x+1] = av_clip_uint8(src[x+1] + 128);
145            dst[x+2] = av_clip_uint8(src[x+2] + 128);
146            dst[x+3] = av_clip_uint8(src[x+3] + 128);
147        }
148        dst += dst_stride;
149        src += src_stride;
150    }
151}
152
153static void add_rect_clamped_c(uint8_t *dst, const uint16_t *src, int stride,
154                               const int16_t *idwt, int idwt_stride,
155                               int width, int height)
156{
157    int x, y;
158
159    for (y = 0; y < height; y++) {
160        for (x = 0; x < width; x+=2) {
161            dst[x  ] = av_clip_uint8(((src[x  ]+32)>>6) + idwt[x  ]);
162            dst[x+1] = av_clip_uint8(((src[x+1]+32)>>6) + idwt[x+1]);
163        }
164        dst += stride;
165        src += stride;
166        idwt += idwt_stride;
167    }
168}
169
170#define PIXFUNC(PFX, WIDTH)                                             \
171    c->PFX ## _dirac_pixels_tab[WIDTH>>4][0] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _c; \
172    c->PFX ## _dirac_pixels_tab[WIDTH>>4][1] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l2_c; \
173    c->PFX ## _dirac_pixels_tab[WIDTH>>4][2] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l4_c; \
174    c->PFX ## _dirac_pixels_tab[WIDTH>>4][3] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c
175
176void ff_diracdsp_init(DiracDSPContext *c)
177{
178    c->dirac_hpel_filter = dirac_hpel_filter;
179    c->add_rect_clamped = add_rect_clamped_c;
180    c->put_signed_rect_clamped = put_signed_rect_clamped_c;
181
182    c->add_dirac_obmc[0] = add_obmc8_c;
183    c->add_dirac_obmc[1] = add_obmc16_c;
184    c->add_dirac_obmc[2] = add_obmc32_c;
185
186    c->weight_dirac_pixels_tab[0] = weight_dirac_pixels8_c;
187    c->weight_dirac_pixels_tab[1] = weight_dirac_pixels16_c;
188    c->weight_dirac_pixels_tab[2] = weight_dirac_pixels32_c;
189    c->biweight_dirac_pixels_tab[0] = biweight_dirac_pixels8_c;
190    c->biweight_dirac_pixels_tab[1] = biweight_dirac_pixels16_c;
191    c->biweight_dirac_pixels_tab[2] = biweight_dirac_pixels32_c;
192
193    PIXFUNC(put, 8);
194    PIXFUNC(put, 16);
195    PIXFUNC(put, 32);
196    PIXFUNC(avg, 8);
197    PIXFUNC(avg, 16);
198    PIXFUNC(avg, 32);
199
200    if (HAVE_MMX && HAVE_YASM) ff_diracdsp_init_mmx(c);
201}
202