2 * VP9 SIMD optimizations
4 * Copyright (c) 2013 Ronald S. Bultje <rsbultje@gmail.com>
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/attributes.h"
24 #include "libavutil/cpu.h"
25 #include "libavutil/internal.h"
26 #include "libavutil/mem.h"
27 #include "libavutil/x86/asm.h"
28 #include "libavutil/x86/cpu.h"
29 #include "libavcodec/vp9.h"
33 #define fpel_func(avg, sz, opt) \
34 void ff_ ## avg ## sz ## _ ## opt(uint8_t *dst, const uint8_t *src, \
35 ptrdiff_t dst_stride, \
36 ptrdiff_t src_stride, \
37 int h, int mx, int my)
39 fpel_func(put, 4, mmx);
40 fpel_func(put, 8, mmx);
41 fpel_func(put, 16, sse);
42 fpel_func(put, 32, sse);
43 fpel_func(put, 64, sse);
44 fpel_func(avg, 4, sse);
45 fpel_func(avg, 8, sse);
46 fpel_func(avg, 16, sse2);
47 fpel_func(avg, 32, sse2);
48 fpel_func(avg, 64, sse2);
51 #define mc_func(avg, sz, dir, opt) \
53 ff_ ## avg ## _8tap_1d_ ## dir ## _ ## sz ## _ ## opt(uint8_t *dst, \
55 ptrdiff_t dst_stride, \
56 ptrdiff_t src_stride, \
58 const int8_t (*filter)[16])
60 #define mc_funcs(sz) \
61 mc_func(put, sz, h, ssse3); \
62 mc_func(avg, sz, h, ssse3); \
63 mc_func(put, sz, v, ssse3); \
64 mc_func(avg, sz, v, ssse3)
72 #define mc_rep_func(avg, sz, hsz, dir, opt) \
73 static av_always_inline void \
74 ff_ ## avg ## _8tap_1d_ ## dir ## _ ## sz ## _ ## opt(uint8_t *dst, \
76 ptrdiff_t dst_stride, \
77 ptrdiff_t src_stride, \
79 const int8_t (*filter)[16]) \
81 ff_ ## avg ## _8tap_1d_ ## dir ## _ ## hsz ## _ ## opt(dst, src, \
86 ff_ ## avg ## _8tap_1d_ ## dir ## _ ## hsz ## _ ## opt(dst + hsz, \
93 #define mc_rep_funcs(sz, hsz) \
94 mc_rep_func(put, sz, hsz, h, ssse3); \
95 mc_rep_func(avg, sz, hsz, h, ssse3); \
96 mc_rep_func(put, sz, hsz, v, ssse3); \
97 mc_rep_func(avg, sz, hsz, v, ssse3)
100 mc_rep_funcs(32, 16);
101 mc_rep_funcs(64, 32);
106 extern const int8_t ff_filters_ssse3[3][15][4][16];
108 #define filter_8tap_2d_fn(op, sz, f, fname) \
110 op ## _8tap_ ## fname ## _ ## sz ## hv_ssse3(uint8_t *dst, \
111 const uint8_t *src, \
112 ptrdiff_t dst_stride, \
113 ptrdiff_t src_stride, \
114 int h, int mx, int my) \
116 LOCAL_ALIGNED_16(uint8_t, temp, [71 * 64]); \
117 ff_put_8tap_1d_h_ ## sz ## _ssse3(temp, src - 3 * src_stride, \
120 ff_filters_ssse3[f][mx - 1]); \
121 ff_ ## op ## _8tap_1d_v_ ## sz ## _ssse3(dst, temp + 3 * 64, \
124 ff_filters_ssse3[f][my - 1]); \
127 #define filters_8tap_2d_fn(op, sz) \
128 filter_8tap_2d_fn(op, sz, FILTER_8TAP_REGULAR, regular) \
129 filter_8tap_2d_fn(op, sz, FILTER_8TAP_SHARP, sharp) \
130 filter_8tap_2d_fn(op, sz, FILTER_8TAP_SMOOTH, smooth)
132 #define filters_8tap_2d_fn2(op) \
133 filters_8tap_2d_fn(op, 64) \
134 filters_8tap_2d_fn(op, 32) \
135 filters_8tap_2d_fn(op, 16) \
136 filters_8tap_2d_fn(op, 8) \
137 filters_8tap_2d_fn(op, 4)
139 filters_8tap_2d_fn2(put)
140 filters_8tap_2d_fn2(avg)
142 #undef filters_8tap_2d_fn2
143 #undef filters_8tap_2d_fn
144 #undef filter_8tap_2d_fn
146 #define filter_8tap_1d_fn(op, sz, f, fname, dir, dvar) \
148 op ## _8tap_ ## fname ## _ ## sz ## dir ## _ssse3(uint8_t *dst, \
149 const uint8_t *src, \
150 ptrdiff_t dst_stride, \
151 ptrdiff_t src_stride, \
155 ff_ ## op ## _8tap_1d_ ## dir ## _ ## sz ## _ssse3(dst, src, \
158 ff_filters_ssse3[f][dvar - 1]); \
161 #define filters_8tap_1d_fn(op, sz, dir, dvar) \
162 filter_8tap_1d_fn(op, sz, FILTER_8TAP_REGULAR, regular, dir, dvar) \
163 filter_8tap_1d_fn(op, sz, FILTER_8TAP_SHARP, sharp, dir, dvar) \
164 filter_8tap_1d_fn(op, sz, FILTER_8TAP_SMOOTH, smooth, dir, dvar)
166 #define filters_8tap_1d_fn2(op, sz) \
167 filters_8tap_1d_fn(op, sz, h, mx) \
168 filters_8tap_1d_fn(op, sz, v, my)
170 #define filters_8tap_1d_fn3(op) \
171 filters_8tap_1d_fn2(op, 64) \
172 filters_8tap_1d_fn2(op, 32) \
173 filters_8tap_1d_fn2(op, 16) \
174 filters_8tap_1d_fn2(op, 8) \
175 filters_8tap_1d_fn2(op, 4)
177 filters_8tap_1d_fn3(put)
178 filters_8tap_1d_fn3(avg)
180 #undef filters_8tap_1d_fn
181 #undef filters_8tap_1d_fn2
182 #undef filters_8tap_1d_fn3
183 #undef filter_8tap_1d_fn
185 #endif /* HAVE_YASM */
187 av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
190 int cpu_flags = av_get_cpu_flags();
192 #define init_fpel(idx1, idx2, sz, type, opt) \
193 dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
194 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
195 dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
196 dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_ ## type ## sz ## _ ## opt
199 #define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \
200 dsp->mc[idx1][FILTER_8TAP_SMOOTH][idx2][idxh][idxv] = type ## _8tap_smooth_ ## sz ## dir ## _ ## opt; \
201 dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type ## _8tap_regular_ ## sz ## dir ## _ ## opt; \
202 dsp->mc[idx1][FILTER_8TAP_SHARP][idx2][idxh][idxv] = type ## _8tap_sharp_ ## sz ## dir ## _ ## opt
204 #define init_subpel2(idx, idxh, idxv, dir, type, opt) \
205 init_subpel1(0, idx, idxh, idxv, 64, dir, type, opt); \
206 init_subpel1(1, idx, idxh, idxv, 32, dir, type, opt); \
207 init_subpel1(2, idx, idxh, idxv, 16, dir, type, opt); \
208 init_subpel1(3, idx, idxh, idxv, 8, dir, type, opt); \
209 init_subpel1(4, idx, idxh, idxv, 4, dir, type, opt)
211 #define init_subpel3(idx, type, opt) \
212 init_subpel2(idx, 1, 1, hv, type, opt); \
213 init_subpel2(idx, 0, 1, v, type, opt); \
214 init_subpel2(idx, 1, 0, h, type, opt)
216 if (EXTERNAL_MMX(cpu_flags)) {
217 init_fpel(4, 0, 4, put, mmx);
218 init_fpel(3, 0, 8, put, mmx);
221 if (EXTERNAL_SSE(cpu_flags)) {
222 init_fpel(2, 0, 16, put, sse);
223 init_fpel(1, 0, 32, put, sse);
224 init_fpel(0, 0, 64, put, sse);
225 init_fpel(4, 1, 4, avg, sse);
226 init_fpel(3, 1, 8, avg, sse);
229 if (EXTERNAL_SSE2(cpu_flags)) {
230 init_fpel(2, 1, 16, avg, sse2);
231 init_fpel(1, 1, 32, avg, sse2);
232 init_fpel(0, 1, 64, avg, sse2);
235 if (EXTERNAL_SSSE3(cpu_flags)) {
236 init_subpel3(0, put, ssse3);
237 init_subpel3(1, avg, ssse3);
245 #endif /* HAVE_YASM */