2 * VP8 compatible video decoder
4 * Copyright (C) 2010 David Conrad
6 * This file is part of Libav.
8 * Libav 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.
13 * Libav 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.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/cpu.h"
24 #include "libavcodec/vp8dsp.h"
25 #include "dsputil_altivec.h"
26 #include "types_altivec.h"
27 #include "util_altivec.h"
29 #define REPT4(...) { __VA_ARGS__, __VA_ARGS__, __VA_ARGS__, __VA_ARGS__ }
31 // h subpel filter uses msum to multiply+add 4 pixel taps at once
32 static const vec_s8 h_subpel_filters_inner[7] =
34 REPT4( -6, 123, 12, -1),
35 REPT4(-11, 108, 36, -8),
36 REPT4( -9, 93, 50, -6),
37 REPT4(-16, 77, 77, -16),
38 REPT4( -6, 50, 93, -9),
39 REPT4( -8, 36, 108, -11),
40 REPT4( -1, 12, 123, -6),
43 // for 6tap filters, these are the outer two taps
44 // The zeros mask off pixels 4-7 when filtering 0-3
46 static const vec_s8 h_subpel_filters_outer[3] =
53 #define LOAD_H_SUBPEL_FILTER(i) \
54 vec_s8 filter_inner = h_subpel_filters_inner[i]; \
55 vec_s8 filter_outerh = h_subpel_filters_outer[(i)>>1]; \
56 vec_s8 filter_outerl = vec_sld(filter_outerh, filter_outerh, 2)
58 #define FILTER_H(dstv, off) \
59 a = vec_ld((off)-is6tap-1, src); \
60 b = vec_ld((off)-is6tap-1+15, src); \
62 pixh = vec_perm(a, b, permh##off); \
63 pixl = vec_perm(a, b, perml##off); \
64 filth = vec_msum(filter_inner, pixh, c64); \
65 filtl = vec_msum(filter_inner, pixl, c64); \
68 outer = vec_perm(a, b, perm_6tap##off); \
69 filth = vec_msum(filter_outerh, outer, filth); \
70 filtl = vec_msum(filter_outerl, outer, filtl); \
73 filtl = filth; /* discard pixels 4-7 */ \
74 dstv = vec_packs(filth, filtl); \
75 dstv = vec_sra(dstv, c7)
77 static av_always_inline
78 void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride,
79 uint8_t *src, ptrdiff_t src_stride,
80 int h, int mx, int w, int is6tap)
82 LOAD_H_SUBPEL_FILTER(mx-1);
83 vec_u8 align_vec0, align_vec8, permh0, permh8, filt;
84 vec_u8 perm_6tap0, perm_6tap8, perml0, perml8;
85 vec_u8 a, b, pixh, pixl, outer;
89 vec_u8 perm_inner6 = { 1,2,3,4, 2,3,4,5, 3,4,5,6, 4,5,6,7 };
90 vec_u8 perm_inner4 = { 0,1,2,3, 1,2,3,4, 2,3,4,5, 3,4,5,6 };
91 vec_u8 perm_inner = is6tap ? perm_inner6 : perm_inner4;
92 vec_u8 perm_outer = { 4,9, 0,5, 5,10, 1,6, 6,11, 2,7, 7,12, 3,8 };
93 vec_s32 c64 = vec_sl(vec_splat_s32(1), vec_splat_u32(6));
94 vec_u16 c7 = vec_splat_u16(7);
96 align_vec0 = vec_lvsl( -is6tap-1, src);
97 align_vec8 = vec_lvsl(8-is6tap-1, src);
99 permh0 = vec_perm(align_vec0, align_vec0, perm_inner);
100 permh8 = vec_perm(align_vec8, align_vec8, perm_inner);
101 perm_inner = vec_add(perm_inner, vec_splat_u8(4));
102 perml0 = vec_perm(align_vec0, align_vec0, perm_inner);
103 perml8 = vec_perm(align_vec8, align_vec8, perm_inner);
104 perm_6tap0 = vec_perm(align_vec0, align_vec0, perm_outer);
105 perm_6tap8 = vec_perm(align_vec8, align_vec8, perm_outer);
112 filt = vec_packsu(f16h, f16l);
113 vec_st(filt, 0, dst);
115 filt = vec_packsu(f16h, f16h);
116 vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
118 vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
125 // v subpel filter does a simple vertical multiply + add
126 static const vec_u8 v_subpel_filters[7] =
128 { 0, 6, 123, 12, 1, 0 },
129 { 2, 11, 108, 36, 8, 1 },
130 { 0, 9, 93, 50, 6, 0 },
131 { 3, 16, 77, 77, 16, 3 },
132 { 0, 6, 50, 93, 9, 0 },
133 { 1, 8, 36, 108, 11, 2 },
134 { 0, 1, 12, 123, 6, 0 },
137 #define LOAD_V_SUBPEL_FILTER(i) \
138 vec_u8 subpel_filter = v_subpel_filters[i]; \
139 vec_u8 f0 = vec_splat(subpel_filter, 0); \
140 vec_u8 f1 = vec_splat(subpel_filter, 1); \
141 vec_u8 f2 = vec_splat(subpel_filter, 2); \
142 vec_u8 f3 = vec_splat(subpel_filter, 3); \
143 vec_u8 f4 = vec_splat(subpel_filter, 4); \
144 vec_u8 f5 = vec_splat(subpel_filter, 5)
146 #define FILTER_V(dstv, vec_mul) \
147 s1f = (vec_s16)vec_mul(s1, f1); \
148 s2f = (vec_s16)vec_mul(s2, f2); \
149 s3f = (vec_s16)vec_mul(s3, f3); \
150 s4f = (vec_s16)vec_mul(s4, f4); \
151 s2f = vec_subs(s2f, s1f); \
152 s3f = vec_subs(s3f, s4f); \
154 s0f = (vec_s16)vec_mul(s0, f0); \
155 s5f = (vec_s16)vec_mul(s5, f5); \
156 s2f = vec_adds(s2f, s0f); \
157 s3f = vec_adds(s3f, s5f); \
159 dstv = vec_adds(s2f, s3f); \
160 dstv = vec_adds(dstv, c64); \
161 dstv = vec_sra(dstv, c7)
163 static av_always_inline
164 void put_vp8_epel_v_altivec_core(uint8_t *dst, ptrdiff_t dst_stride,
165 uint8_t *src, ptrdiff_t src_stride,
166 int h, int my, int w, int is6tap)
168 LOAD_V_SUBPEL_FILTER(my-1);
169 vec_u8 s0, s1, s2, s3, s4, s5, filt, align_vech, perm_vec, align_vecl;
170 vec_s16 s0f, s1f, s2f, s3f, s4f, s5f, f16h, f16l;
171 vec_s16 c64 = vec_sl(vec_splat_s16(1), vec_splat_u16(6));
172 vec_u16 c7 = vec_splat_u16(7);
174 // we want pixels 0-7 to be in the even positions and 8-15 in the odd,
175 // so combine this permute with the alignment permute vector
176 align_vech = vec_lvsl(0, src);
177 align_vecl = vec_sld(align_vech, align_vech, 8);
179 perm_vec = vec_mergeh(align_vech, align_vecl);
181 perm_vec = vec_mergeh(align_vech, align_vech);
184 s0 = load_with_perm_vec(-2*src_stride, src, perm_vec);
185 s1 = load_with_perm_vec(-1*src_stride, src, perm_vec);
186 s2 = load_with_perm_vec( 0*src_stride, src, perm_vec);
187 s3 = load_with_perm_vec( 1*src_stride, src, perm_vec);
189 s4 = load_with_perm_vec( 2*src_stride, src, perm_vec);
191 src += (2+is6tap)*src_stride;
195 s5 = load_with_perm_vec(0, src, perm_vec);
197 s4 = load_with_perm_vec(0, src, perm_vec);
199 FILTER_V(f16h, vec_mule);
202 FILTER_V(f16l, vec_mulo);
203 filt = vec_packsu(f16h, f16l);
204 vec_st(filt, 0, dst);
206 filt = vec_packsu(f16h, f16h);
208 filt = (vec_u8)vec_splat((vec_u32)filt, 0);
210 vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
211 vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
227 #define EPEL_FUNCS(WIDTH, TAPS) \
229 void put_vp8_epel ## WIDTH ## _h ## TAPS ## _altivec(uint8_t *dst, ptrdiff_t dst_stride, uint8_t *src, ptrdiff_t src_stride, int h, int mx, int my) \
231 put_vp8_epel_h_altivec_core(dst, dst_stride, src, src_stride, h, mx, WIDTH, TAPS == 6); \
235 void put_vp8_epel ## WIDTH ## _v ## TAPS ## _altivec(uint8_t *dst, ptrdiff_t dst_stride, uint8_t *src, ptrdiff_t src_stride, int h, int mx, int my) \
237 put_vp8_epel_v_altivec_core(dst, dst_stride, src, src_stride, h, my, WIDTH, TAPS == 6); \
240 #define EPEL_HV(WIDTH, HTAPS, VTAPS) \
241 static void put_vp8_epel ## WIDTH ## _h ## HTAPS ## v ## VTAPS ## _altivec(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s, int h, int mx, int my) \
243 DECLARE_ALIGNED(16, uint8_t, tmp)[(2*WIDTH+5)*16]; \
245 put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-2*stride, stride, h+5, mx, my); \
246 put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, stride, tmp+2*16, 16, h, mx, my); \
248 put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-stride, stride, h+4, mx, my); \
249 put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, stride, tmp+16, 16, h, mx, my); \
269 static void put_vp8_pixels16_altivec(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s, int h, int mx, int my)
271 ff_put_pixels16_altivec(dst, src, stride, h);
274 av_cold void ff_vp8dsp_init_altivec(VP8DSPContext *c)
276 if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
279 c->put_vp8_epel_pixels_tab[0][0][0] = put_vp8_pixels16_altivec;
280 c->put_vp8_epel_pixels_tab[0][0][2] = put_vp8_epel16_h6_altivec;
281 c->put_vp8_epel_pixels_tab[0][2][0] = put_vp8_epel16_v6_altivec;
282 c->put_vp8_epel_pixels_tab[0][2][2] = put_vp8_epel16_h6v6_altivec;
284 c->put_vp8_epel_pixels_tab[1][0][2] = put_vp8_epel8_h6_altivec;
285 c->put_vp8_epel_pixels_tab[1][2][0] = put_vp8_epel8_v6_altivec;
286 c->put_vp8_epel_pixels_tab[1][0][1] = put_vp8_epel8_h4_altivec;
287 c->put_vp8_epel_pixels_tab[1][1][0] = put_vp8_epel8_v4_altivec;
289 c->put_vp8_epel_pixels_tab[1][2][2] = put_vp8_epel8_h6v6_altivec;
290 c->put_vp8_epel_pixels_tab[1][1][1] = put_vp8_epel8_h4v4_altivec;
291 c->put_vp8_epel_pixels_tab[1][1][2] = put_vp8_epel8_h6v4_altivec;
292 c->put_vp8_epel_pixels_tab[1][2][1] = put_vp8_epel8_h4v6_altivec;
294 c->put_vp8_epel_pixels_tab[2][0][2] = put_vp8_epel4_h6_altivec;
295 c->put_vp8_epel_pixels_tab[2][2][0] = put_vp8_epel4_v6_altivec;
296 c->put_vp8_epel_pixels_tab[2][0][1] = put_vp8_epel4_h4_altivec;
297 c->put_vp8_epel_pixels_tab[2][1][0] = put_vp8_epel4_v4_altivec;
299 c->put_vp8_epel_pixels_tab[2][2][2] = put_vp8_epel4_h6v6_altivec;
300 c->put_vp8_epel_pixels_tab[2][1][1] = put_vp8_epel4_h4v4_altivec;
301 c->put_vp8_epel_pixels_tab[2][1][2] = put_vp8_epel4_h6v4_altivec;
302 c->put_vp8_epel_pixels_tab[2][2][1] = put_vp8_epel4_h4v6_altivec;