Statistics
| Branch: | Revision:

ffmpeg / libavcodec / ppc / vp8dsp_altivec.c @ ccf22d3e

History | View | Annotate | Download (10 KB)

1
/**
2
 * VP8 compatible video decoder
3
 *
4
 * Copyright (C) 2010 David Conrad
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 "libavcodec/vp8dsp.h"
24
#include "dsputil_altivec.h"
25
#include "types_altivec.h"
26
#include "util_altivec.h"
27

    
28
#define REPT4(...) { __VA_ARGS__, __VA_ARGS__, __VA_ARGS__, __VA_ARGS__ }
29

    
30
// h subpel filter uses msum to multiply+add 4 pixel taps at once
31
static const vec_s8 h_subpel_filters_inner[7] =
32
{
33
    REPT4( -6, 123,  12,  -1),
34
    REPT4(-11, 108,  36,  -8),
35
    REPT4( -9,  93,  50,  -6),
36
    REPT4(-16,  77,  77, -16),
37
    REPT4( -6,  50,  93,  -9),
38
    REPT4( -8,  36, 108, -11),
39
    REPT4( -1,  12, 123,  -6),
40
};
41

    
42
// for 6tap filters, these are the outer two taps
43
// The zeros mask off pixels 4-7 when filtering 0-3
44
// and vice-versa
45
static const vec_s8 h_subpel_filters_outer[3] =
46
{
47
    REPT4(0, 0, 2, 1),
48
    REPT4(0, 0, 3, 3),
49
    REPT4(0, 0, 1, 2),
50
};
51

    
52
#define LOAD_H_SUBPEL_FILTER(i) \
53
    vec_s8 filter_inner  = h_subpel_filters_inner[i]; \
54
    vec_s8 filter_outerh = h_subpel_filters_outer[(i)>>1]; \
55
    vec_s8 filter_outerl = vec_sld(filter_outerh, filter_outerh, 2)
56

    
57
#define FILTER_H(dstv, off) \
58
    a = vec_ld((off)-2,    src); \
59
    b = vec_ld((off)-2+15, src); \
60
\
61
    pixh  = vec_perm(a, b, permh##off); \
62
    pixl  = vec_perm(a, b, perml##off); \
63
    filth = vec_msum(filter_inner, pixh, c64); \
64
    filtl = vec_msum(filter_inner, pixl, c64); \
65
\
66
    if (is6tap) { \
67
        outer = vec_perm(a, b, perm_6tap##off); \
68
        filth = vec_msum(filter_outerh, outer, filth); \
69
        filtl = vec_msum(filter_outerl, outer, filtl); \
70
    } \
71
    if (w == 4) \
72
        filtl = filth; /* discard pixels 4-7 */ \
73
    dstv = vec_packs(filth, filtl); \
74
    dstv = vec_sra(dstv, c7)
75

    
76
static av_always_inline
77
void put_vp8_epel_h_altivec_core(uint8_t *dst, int dst_stride,
78
                                 uint8_t *src, int src_stride,
79
                                 int h, int mx, int w, int is6tap)
80
{
81
    LOAD_H_SUBPEL_FILTER(mx-1);
82
    vec_u8 align_vec0, align_vec8, permh0, permh8, filt;
83
    vec_u8 perm_6tap0, perm_6tap8, perml0, perml8;
84
    vec_u8 a, b, pixh, pixl, outer;
85
    vec_s16 f16h, f16l;
86
    vec_s32 filth, filtl;
87

    
88
    vec_u8 perm_inner = { 1,2,3,4, 2,3,4,5, 3,4,5,6, 4,5,6,7 };
89
    vec_u8 perm_outer = { 4,9, 0,5, 5,10, 1,6, 6,11, 2,7, 7,12, 3,8 };
90
    vec_s32 c64 = vec_sl(vec_splat_s32(1), vec_splat_u32(6));
91
    vec_u16 c7  = vec_splat_u16(7);
92

    
93
    align_vec0 = vec_lvsl( -2, src);
94
    align_vec8 = vec_lvsl(8-2, src);
95

    
96
    permh0     = vec_perm(align_vec0, align_vec0, perm_inner);
97
    permh8     = vec_perm(align_vec8, align_vec8, perm_inner);
98
    perm_inner = vec_add(perm_inner, vec_splat_u8(4));
99
    perml0     = vec_perm(align_vec0, align_vec0, perm_inner);
100
    perml8     = vec_perm(align_vec8, align_vec8, perm_inner);
101
    perm_6tap0 = vec_perm(align_vec0, align_vec0, perm_outer);
102
    perm_6tap8 = vec_perm(align_vec8, align_vec8, perm_outer);
103

    
104
    while (h --> 0) {
105
        FILTER_H(f16h, 0);
106

    
107
        if (w == 16) {
108
            FILTER_H(f16l, 8);
109
            filt = vec_packsu(f16h, f16l);
110
            vec_st(filt, 0, dst);
111
        } else {
112
            filt = vec_packsu(f16h, f16h);
113
            vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
114
            if (w == 8)
115
                vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
116
        }
117
        src += src_stride;
118
        dst += dst_stride;
119
    }
120
}
121

    
122
// v subpel filter does a simple vertical multiply + add
123
static const vec_u8 v_subpel_filters[7] =
124
{
125
    { 0,   6, 123,  12,   1,   0 },
126
    { 2,  11, 108,  36,   8,   1 },
127
    { 0,   9,  93,  50,   6,   0 },
128
    { 3,  16,  77,  77,  16,   3 },
129
    { 0,   6,  50,  93,   9,   0 },
130
    { 1,   8,  36, 108,  11,   2 },
131
    { 0,   1,  12, 123,   6,   0 },
132
};
133

    
134
#define LOAD_V_SUBPEL_FILTER(i) \
135
    vec_u8 subpel_filter = v_subpel_filters[i]; \
136
    vec_u8 f0 = vec_splat(subpel_filter, 0); \
137
    vec_u8 f1 = vec_splat(subpel_filter, 1); \
138
    vec_u8 f2 = vec_splat(subpel_filter, 2); \
139
    vec_u8 f3 = vec_splat(subpel_filter, 3); \
140
    vec_u8 f4 = vec_splat(subpel_filter, 4); \
141
    vec_u8 f5 = vec_splat(subpel_filter, 5)
142

    
143
#define FILTER_V(dstv, vec_mul) \
144
    s1f = (vec_s16)vec_mul(s1, f1); \
145
    s2f = (vec_s16)vec_mul(s2, f2); \
146
    s3f = (vec_s16)vec_mul(s3, f3); \
147
    s4f = (vec_s16)vec_mul(s4, f4); \
148
    s2f = vec_subs(s2f, s1f); \
149
    s3f = vec_subs(s3f, s4f); \
150
    if (is6tap) { \
151
        s0f = (vec_s16)vec_mul(s0, f0); \
152
        s5f = (vec_s16)vec_mul(s5, f5); \
153
        s2f = vec_adds(s2f, s0f); \
154
        s3f = vec_adds(s3f, s5f); \
155
    } \
156
    dstv = vec_adds(s2f, s3f); \
157
    dstv = vec_adds(dstv, c64); \
158
    dstv = vec_sra(dstv, c7)
159

    
160
static av_always_inline
161
void put_vp8_epel_v_altivec_core(uint8_t *dst, int dst_stride,
162
                                 uint8_t *src, int src_stride,
163
                                 int h, int my, int w, int is6tap)
164
{
165
    LOAD_V_SUBPEL_FILTER(my-1);
166
    vec_u8 s0, s1, s2, s3, s4, s5, filt, align_vech, perm_vec, align_vecl;
167
    vec_s16 s0f, s1f, s2f, s3f, s4f, s5f, f16h, f16l;
168
    vec_s16 c64 = vec_sl(vec_splat_s16(1), vec_splat_u16(6));
169
    vec_u16 c7  = vec_splat_u16(7);
170

    
171
    // we want pixels 0-7 to be in the even positions and 8-15 in the odd,
172
    // so combine this permute with the alignment permute vector
173
    align_vech = vec_lvsl(0, src);
174
    align_vecl = vec_sld(align_vech, align_vech, 8);
175
    if (w ==16)
176
        perm_vec = vec_mergeh(align_vech, align_vecl);
177
    else
178
        perm_vec = vec_mergeh(align_vech, align_vech);
179

    
180
    if (is6tap)
181
        s0 = load_with_perm_vec(-2*src_stride, src, perm_vec);
182
    s1 = load_with_perm_vec(-1*src_stride, src, perm_vec);
183
    s2 = load_with_perm_vec( 0*src_stride, src, perm_vec);
184
    s3 = load_with_perm_vec( 1*src_stride, src, perm_vec);
185
    if (is6tap)
186
        s4 = load_with_perm_vec( 2*src_stride, src, perm_vec);
187

    
188
    src += (2+is6tap)*src_stride;
189

    
190
    while (h --> 0) {
191
        if (is6tap)
192
            s5 = load_with_perm_vec(0, src, perm_vec);
193
        else
194
            s4 = load_with_perm_vec(0, src, perm_vec);
195

    
196
        FILTER_V(f16h, vec_mule);
197

    
198
        if (w == 16) {
199
            FILTER_V(f16l, vec_mulo);
200
            filt = vec_packsu(f16h, f16l);
201
            vec_st(filt, 0, dst);
202
        } else {
203
            filt = vec_packsu(f16h, f16h);
204
            if (w == 4)
205
                filt = (vec_u8)vec_splat((vec_u32)filt, 0);
206
            else
207
                vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
208
            vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
209
        }
210

    
211
        if (is6tap)
212
            s0 = s1;
213
        s1 = s2;
214
        s2 = s3;
215
        s3 = s4;
216
        if (is6tap)
217
            s4 = s5;
218

    
219
        dst += dst_stride;
220
        src += src_stride;
221
    }
222
}
223

    
224
#define EPEL_FUNCS(WIDTH, TAPS) \
225
static av_noinline \
226
void put_vp8_epel ## WIDTH ## _h ## TAPS ## _altivec(uint8_t *dst, int dst_stride, uint8_t *src, int src_stride, int h, int mx, int my) \
227
{ \
228
    put_vp8_epel_h_altivec_core(dst, dst_stride, src, src_stride, h, mx, WIDTH, TAPS == 6); \
229
} \
230
\
231
static av_noinline \
232
void put_vp8_epel ## WIDTH ## _v ## TAPS ## _altivec(uint8_t *dst, int dst_stride, uint8_t *src, int src_stride, int h, int mx, int my) \
233
{ \
234
    put_vp8_epel_v_altivec_core(dst, dst_stride, src, src_stride, h, my, WIDTH, TAPS == 6); \
235
}
236

    
237
#define EPEL_HV(WIDTH, HTAPS, VTAPS) \
238
static void put_vp8_epel ## WIDTH ## _h ## HTAPS ## v ## VTAPS ## _altivec(uint8_t *dst, int stride, uint8_t *src, int s, int h, int mx, int my) \
239
{ \
240
    DECLARE_ALIGNED(16, uint8_t, tmp)[(2*WIDTH+5)*16]; \
241
    put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16,     src-2*stride, stride, h+5, mx, my); \
242
    put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, stride, tmp+2*16,     16,     h,   mx, my); \
243
}
244

    
245
EPEL_FUNCS(16,6)
246
EPEL_FUNCS(8, 6)
247
EPEL_FUNCS(8, 4)
248
EPEL_FUNCS(4, 6)
249
EPEL_FUNCS(4, 4)
250

    
251
EPEL_HV(16, 6,6)
252
EPEL_HV(8,  6,6)
253
EPEL_HV(8,  4,6)
254
EPEL_HV(8,  6,4)
255
EPEL_HV(8,  4,4)
256
EPEL_HV(4,  6,6)
257
EPEL_HV(4,  4,6)
258
EPEL_HV(4,  6,4)
259
EPEL_HV(4,  4,4)
260

    
261
static void put_vp8_pixels16_altivec(uint8_t *dst, int stride, uint8_t *src, int s, int h, int mx, int my)
262
{
263
    put_pixels16_altivec(dst, src, stride, h);
264
}
265

    
266
av_cold void ff_vp8dsp_init_altivec(VP8DSPContext *c)
267
{
268
    if (!(mm_support() & AV_CPU_FLAG_ALTIVEC))
269
        return;
270

    
271
    c->put_vp8_epel_pixels_tab[0][0][0] = put_vp8_pixels16_altivec;
272
    c->put_vp8_epel_pixels_tab[0][0][2] = put_vp8_epel16_h6_altivec;
273
    c->put_vp8_epel_pixels_tab[0][2][0] = put_vp8_epel16_v6_altivec;
274
    c->put_vp8_epel_pixels_tab[0][2][2] = put_vp8_epel16_h6v6_altivec;
275

    
276
    c->put_vp8_epel_pixels_tab[1][0][2] = put_vp8_epel8_h6_altivec;
277
    c->put_vp8_epel_pixels_tab[1][2][0] = put_vp8_epel8_v6_altivec;
278
    c->put_vp8_epel_pixels_tab[1][0][1] = put_vp8_epel8_h4_altivec;
279
    c->put_vp8_epel_pixels_tab[1][1][0] = put_vp8_epel8_v4_altivec;
280

    
281
    c->put_vp8_epel_pixels_tab[1][2][2] = put_vp8_epel8_h6v6_altivec;
282
    c->put_vp8_epel_pixels_tab[1][1][1] = put_vp8_epel8_h4v4_altivec;
283
    c->put_vp8_epel_pixels_tab[1][1][2] = put_vp8_epel8_h6v4_altivec;
284
    c->put_vp8_epel_pixels_tab[1][2][1] = put_vp8_epel8_h4v6_altivec;
285

    
286
    c->put_vp8_epel_pixels_tab[2][0][2] = put_vp8_epel4_h6_altivec;
287
    c->put_vp8_epel_pixels_tab[2][2][0] = put_vp8_epel4_v6_altivec;
288
    c->put_vp8_epel_pixels_tab[2][0][1] = put_vp8_epel4_h4_altivec;
289
    c->put_vp8_epel_pixels_tab[2][1][0] = put_vp8_epel4_v4_altivec;
290

    
291
    c->put_vp8_epel_pixels_tab[2][2][2] = put_vp8_epel4_h6v6_altivec;
292
    c->put_vp8_epel_pixels_tab[2][1][1] = put_vp8_epel4_h4v4_altivec;
293
    c->put_vp8_epel_pixels_tab[2][1][2] = put_vp8_epel4_h6v4_altivec;
294
    c->put_vp8_epel_pixels_tab[2][2][1] = put_vp8_epel4_h4v6_altivec;
295
}