2 * VP9 SIMD optimizations
4 * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>
6 * This file is part of FFmpeg.
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.
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.
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
23 #ifndef AVCODEC_X86_VP9DSP_INIT_H
24 #define AVCODEC_X86_VP9DSP_INIT_H
26 #include "libavutil/attributes.h"
27 #include "libavutil/mem_internal.h"
29 #include "libavcodec/vp9dsp.h"
31 // hack to force-expand BPC
32 #define cat(a, bpp, b) a##bpp##b
34 #define decl_fpel_func(avg, sz, bpp, opt) \
35 void ff_vp9_##avg##sz##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
36 const uint8_t *src, ptrdiff_t src_stride, \
37 int h, int mx, int my)
39 #define decl_mc_func(avg, sz, dir, opt, type, f_sz, bpp) \
40 void ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
41 const uint8_t *src, ptrdiff_t src_stride, \
42 int h, const type (*filter)[f_sz])
44 #define decl_mc_funcs(sz, opt, type, fsz, bpp) \
45 decl_mc_func(put, sz, h, opt, type, fsz, bpp); \
46 decl_mc_func(avg, sz, h, opt, type, fsz, bpp); \
47 decl_mc_func(put, sz, v, opt, type, fsz, bpp); \
48 decl_mc_func(avg, sz, v, opt, type, fsz, bpp)
50 #define decl_ipred_fn(type, sz, bpp, opt) \
51 void ff_vp9_ipred_##type##_##sz##x##sz##_##bpp##_##opt(uint8_t *dst, \
56 #define decl_ipred_fns(type, bpp, opt4, opt8_16_32) \
57 decl_ipred_fn(type, 4, bpp, opt4); \
58 decl_ipred_fn(type, 8, bpp, opt8_16_32); \
59 decl_ipred_fn(type, 16, bpp, opt8_16_32); \
60 decl_ipred_fn(type, 32, bpp, opt8_16_32)
62 #define decl_itxfm_func(typea, typeb, size, bpp, opt) \
63 void cat(ff_vp9_##typea##_##typeb##_##size##x##size##_add_, bpp, _##opt)(uint8_t *dst, \
68 #define decl_itxfm_funcs(size, bpp, opt) \
69 decl_itxfm_func(idct, idct, size, bpp, opt); \
70 decl_itxfm_func(iadst, idct, size, bpp, opt); \
71 decl_itxfm_func(idct, iadst, size, bpp, opt); \
72 decl_itxfm_func(iadst, iadst, size, bpp, opt)
74 #define mc_rep_func(avg, sz, hsz, hszb, dir, opt, type, f_sz, bpp) \
75 static av_always_inline void \
76 ff_vp9_##avg##_8tap_1d_##dir##_##sz##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
77 const uint8_t *src, ptrdiff_t src_stride, \
78 int h, const type (*filter)[f_sz]) \
80 ff_vp9_##avg##_8tap_1d_##dir##_##hsz##_##bpp##_##opt(dst, dst_stride, src, \
81 src_stride, h, filter); \
82 ff_vp9_##avg##_8tap_1d_##dir##_##hsz##_##bpp##_##opt(dst + hszb, dst_stride, src + hszb, \
83 src_stride, h, filter); \
86 #define mc_rep_funcs(sz, hsz, hszb, opt, type, fsz, bpp) \
87 mc_rep_func(put, sz, hsz, hszb, h, opt, type, fsz, bpp) \
88 mc_rep_func(avg, sz, hsz, hszb, h, opt, type, fsz, bpp) \
89 mc_rep_func(put, sz, hsz, hszb, v, opt, type, fsz, bpp) \
90 mc_rep_func(avg, sz, hsz, hszb, v, opt, type, fsz, bpp)
92 #define filter_8tap_1d_fn(op, sz, f, f_opt, fname, dir, dvar, bpp, opt) \
93 static void op##_8tap_##fname##_##sz##dir##_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
94 const uint8_t *src, ptrdiff_t src_stride, \
95 int h, int mx, int my) \
97 ff_vp9_##op##_8tap_1d_##dir##_##sz##_##bpp##_##opt(dst, dst_stride, src, src_stride, \
98 h, ff_filters_##f_opt[f][dvar - 1]); \
101 #define filters_8tap_1d_fn(op, sz, dir, dvar, bpp, opt, f_opt) \
102 filter_8tap_1d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, dir, dvar, bpp, opt) \
103 filter_8tap_1d_fn(op, sz, FILTER_8TAP_SHARP, f_opt, sharp, dir, dvar, bpp, opt) \
104 filter_8tap_1d_fn(op, sz, FILTER_8TAP_SMOOTH, f_opt, smooth, dir, dvar, bpp, opt)
106 #define filters_8tap_1d_fn2(op, sz, bpp, opt, f_opt) \
107 filters_8tap_1d_fn(op, sz, h, mx, bpp, opt, f_opt) \
108 filters_8tap_1d_fn(op, sz, v, my, bpp, opt, f_opt)
110 #define filters_8tap_1d_fn3(op, bpp, opt4, opt8, f_opt) \
111 filters_8tap_1d_fn2(op, 64, bpp, opt8, f_opt) \
112 filters_8tap_1d_fn2(op, 32, bpp, opt8, f_opt) \
113 filters_8tap_1d_fn2(op, 16, bpp, opt8, f_opt) \
114 filters_8tap_1d_fn2(op, 8, bpp, opt8, f_opt) \
115 filters_8tap_1d_fn2(op, 4, bpp, opt4, f_opt)
117 #define filter_8tap_2d_fn(op, sz, f, f_opt, fname, align, bpp, bytes, opt) \
118 static void op##_8tap_##fname##_##sz##hv_##bpp##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \
119 const uint8_t *src, ptrdiff_t src_stride, \
120 int h, int mx, int my) \
122 LOCAL_ALIGNED_##align(uint8_t, temp, [71 * 64 * bytes]); \
123 ff_vp9_put_8tap_1d_h_##sz##_##bpp##_##opt(temp, 64 * bytes, src - 3 * src_stride, \
125 ff_filters_##f_opt[f][mx - 1]); \
126 ff_vp9_##op##_8tap_1d_v_##sz##_##bpp##_##opt(dst, dst_stride, temp + 3 * bytes * 64, \
128 ff_filters_##f_opt[f][my - 1]); \
131 #define filters_8tap_2d_fn(op, sz, align, bpp, bytes, opt, f_opt) \
132 filter_8tap_2d_fn(op, sz, FILTER_8TAP_REGULAR, f_opt, regular, align, bpp, bytes, opt) \
133 filter_8tap_2d_fn(op, sz, FILTER_8TAP_SHARP, f_opt, sharp, align, bpp, bytes, opt) \
134 filter_8tap_2d_fn(op, sz, FILTER_8TAP_SMOOTH, f_opt, smooth, align, bpp, bytes, opt)
136 #define filters_8tap_2d_fn2(op, align, bpp, bytes, opt4, opt8, f_opt) \
137 filters_8tap_2d_fn(op, 64, align, bpp, bytes, opt8, f_opt) \
138 filters_8tap_2d_fn(op, 32, align, bpp, bytes, opt8, f_opt) \
139 filters_8tap_2d_fn(op, 16, align, bpp, bytes, opt8, f_opt) \
140 filters_8tap_2d_fn(op, 8, align, bpp, bytes, opt8, f_opt) \
141 filters_8tap_2d_fn(op, 4, align, bpp, bytes, opt4, f_opt)
143 #define init_fpel_func(idx1, idx2, sz, type, bpp, opt) \
144 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
145 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
146 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
147 dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##bpp##_##opt
149 #define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, bpp, opt) \
150 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = \
151 type##_8tap_smooth_##sz##dir##_##bpp##_##opt; \
152 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = \
153 type##_8tap_regular_##sz##dir##_##bpp##_##opt; \
154 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][idxh][idxv] = \
155 type##_8tap_sharp_##sz##dir##_##bpp##_##opt
157 #define init_subpel2(idx1, idx2, sz, type, bpp, opt) \
158 init_subpel1(idx1, idx2, 1, 1, sz, hv, type, bpp, opt); \
159 init_subpel1(idx1, idx2, 0, 1, sz, v, type, bpp, opt); \
160 init_subpel1(idx1, idx2, 1, 0, sz, h, type, bpp, opt)
162 #define init_subpel3_32_64(idx, type, bpp, opt) \
163 init_subpel2(0, idx, 64, type, bpp, opt); \
164 init_subpel2(1, idx, 32, type, bpp, opt)
166 #define init_subpel3_8to64(idx, type, bpp, opt) \
167 init_subpel3_32_64(idx, type, bpp, opt); \
168 init_subpel2(2, idx, 16, type, bpp, opt); \
169 init_subpel2(3, idx, 8, type, bpp, opt)
171 #define init_subpel3(idx, type, bpp, opt) \
172 init_subpel3_8to64(idx, type, bpp, opt); \
173 init_subpel2(4, idx, 4, type, bpp, opt)
175 #define init_ipred_func(type, enum, sz, bpp, opt) \
176 dsp->intra_pred[TX_##sz##X##sz][enum##_PRED] = \
177 cat(ff_vp9_ipred_##type##_##sz##x##sz##_, bpp, _##opt)
179 #define init_8_16_32_ipred_funcs(type, enum, bpp, opt) \
180 init_ipred_func(type, enum, 8, bpp, opt); \
181 init_ipred_func(type, enum, 16, bpp, opt); \
182 init_ipred_func(type, enum, 32, bpp, opt)
184 #define init_ipred_funcs(type, enum, bpp, opt) \
185 init_ipred_func(type, enum, 4, bpp, opt); \
186 init_8_16_32_ipred_funcs(type, enum, bpp, opt)
188 void ff_vp9dsp_init_10bpp_x86(VP9DSPContext *dsp, int bitexact);
189 void ff_vp9dsp_init_12bpp_x86(VP9DSPContext *dsp, int bitexact);
190 void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp);
192 #endif /* AVCODEC_X86_VP9DSP_INIT_H */