2 * VP8 DSP functions x86-optimized
3 * Copyright (c) 2010 Ronald S. Bultje <rsbultje@gmail.com>
4 * Copyright (c) 2010 Jason Garrett-Glaser <darkshikari@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 #include "libavutil/x86_cpu.h"
24 #include "libavcodec/vp8dsp.h"
31 extern void ff_put_vp8_epel4_h4_mmxext(uint8_t *dst, int dststride,
32 uint8_t *src, int srcstride,
33 int height, int mx, int my);
34 extern void ff_put_vp8_epel4_h6_mmxext(uint8_t *dst, int dststride,
35 uint8_t *src, int srcstride,
36 int height, int mx, int my);
37 extern void ff_put_vp8_epel4_v4_mmxext(uint8_t *dst, int dststride,
38 uint8_t *src, int srcstride,
39 int height, int mx, int my);
40 extern void ff_put_vp8_epel4_v6_mmxext(uint8_t *dst, int dststride,
41 uint8_t *src, int srcstride,
42 int height, int mx, int my);
44 extern void ff_put_vp8_epel8_h4_sse2 (uint8_t *dst, int dststride,
45 uint8_t *src, int srcstride,
46 int height, int mx, int my);
47 extern void ff_put_vp8_epel8_h6_sse2 (uint8_t *dst, int dststride,
48 uint8_t *src, int srcstride,
49 int height, int mx, int my);
50 extern void ff_put_vp8_epel8_v4_sse2 (uint8_t *dst, int dststride,
51 uint8_t *src, int srcstride,
52 int height, int mx, int my);
53 extern void ff_put_vp8_epel8_v6_sse2 (uint8_t *dst, int dststride,
54 uint8_t *src, int srcstride,
55 int height, int mx, int my);
57 extern void ff_put_vp8_epel8_h4_ssse3 (uint8_t *dst, int dststride,
58 uint8_t *src, int srcstride,
59 int height, int mx, int my);
60 extern void ff_put_vp8_epel8_h6_ssse3 (uint8_t *dst, int dststride,
61 uint8_t *src, int srcstride,
62 int height, int mx, int my);
63 extern void ff_put_vp8_epel8_v4_ssse3 (uint8_t *dst, int dststride,
64 uint8_t *src, int srcstride,
65 int height, int mx, int my);
66 extern void ff_put_vp8_epel8_v6_ssse3 (uint8_t *dst, int dststride,
67 uint8_t *src, int srcstride,
68 int height, int mx, int my);
70 extern void ff_put_vp8_bilinear4_h_mmxext(uint8_t *dst, int dststride,
71 uint8_t *src, int srcstride,
72 int height, int mx, int my);
73 extern void ff_put_vp8_bilinear8_h_sse2 (uint8_t *dst, int dststride,
74 uint8_t *src, int srcstride,
75 int height, int mx, int my);
77 extern void ff_put_vp8_bilinear4_v_mmxext(uint8_t *dst, int dststride,
78 uint8_t *src, int srcstride,
79 int height, int mx, int my);
80 extern void ff_put_vp8_bilinear8_v_sse2 (uint8_t *dst, int dststride,
81 uint8_t *src, int srcstride,
82 int height, int mx, int my);
83 extern void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, int dststride,
84 uint8_t *src, int srcstride,
85 int height, int mx, int my);
86 extern void ff_put_vp8_bilinear8_h_ssse3 (uint8_t *dst, int dststride,
87 uint8_t *src, int srcstride,
88 int height, int mx, int my);
90 extern void ff_put_vp8_pixels8_mmx (uint8_t *dst, int dststride,
91 uint8_t *src, int srcstride,
92 int height, int mx, int my);
93 extern void ff_put_vp8_pixels16_mmx(uint8_t *dst, int dststride,
94 uint8_t *src, int srcstride,
95 int height, int mx, int my);
96 extern void ff_put_vp8_pixels16_sse(uint8_t *dst, int dststride,
97 uint8_t *src, int srcstride,
98 int height, int mx, int my);
100 #define TAP_W16(OPT, FILTERTYPE, TAPTYPE) \
101 static void ff_put_vp8_ ## FILTERTYPE ## 16_ ## TAPTYPE ## _ ## OPT( \
102 uint8_t *dst, int dststride, uint8_t *src, \
103 int srcstride, int height, int mx, int my) \
105 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
106 dst, dststride, src, srcstride, height, mx, my); \
107 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
108 dst + 8, dststride, src + 8, srcstride, height, mx, my); \
110 #define TAP_W8(OPT, FILTERTYPE, TAPTYPE) \
111 static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
112 uint8_t *dst, int dststride, uint8_t *src, \
113 int srcstride, int height, int mx, int my) \
115 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
116 dst, dststride, src, srcstride, height, mx, my); \
117 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
118 dst + 4, dststride, src + 4, srcstride, height, mx, my); \
121 TAP_W8 (mmxext, epel, h4)
122 TAP_W8 (mmxext, epel, h6)
123 TAP_W16(mmxext, epel, h6)
124 TAP_W8 (mmxext, epel, v4)
125 TAP_W8 (mmxext, epel, v6)
126 TAP_W16(mmxext, epel, v6)
127 TAP_W8 (mmxext, bilinear, h)
128 TAP_W16(mmxext, bilinear, h)
129 TAP_W8 (mmxext, bilinear, v)
130 TAP_W16(mmxext, bilinear, v)
132 TAP_W16(sse2, epel, h6)
133 TAP_W16(sse2, epel, v6)
134 TAP_W16(sse2, bilinear, h)
135 TAP_W16(sse2, bilinear, v)
137 TAP_W16(ssse3, epel, h6)
138 TAP_W16(ssse3, epel, v6)
139 TAP_W16(ssse3, bilinear, h)
140 TAP_W16(ssse3, bilinear, v)
142 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \
143 static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT( \
144 uint8_t *dst, int dststride, uint8_t *src, \
145 int srcstride, int height, int mx, int my) \
147 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \
148 uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \
149 src -= srcstride * (TAPNUMY / 2 - 1); \
150 ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT( \
151 tmp, SIZE, src, srcstride, height + TAPNUMY - 1, mx, my); \
152 ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT( \
153 dst, dststride, tmpptr, SIZE, height, mx, my); \
156 #define HVTAPMMX(x, y) \
157 HVTAP(mmxext, 8, x, y, 4, 8) \
158 HVTAP(mmxext, 8, x, y, 8, 16)
164 HVTAP(mmxext, 8, 6, 6, 16, 16)
166 #define HVTAPSSE2(x, y, w) \
167 HVTAP(sse2, 16, x, y, w, 16) \
168 HVTAP(ssse3, 16, x, y, w, 16)
176 #define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \
177 static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \
178 uint8_t *dst, int dststride, uint8_t *src, \
179 int srcstride, int height, int mx, int my) \
181 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + 2)]; \
182 ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT( \
183 tmp, SIZE, src, srcstride, height + 1, mx, my); \
184 ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT( \
185 dst, dststride, tmp, SIZE, height, mx, my); \
188 HVBILIN(mmxext, 8, 4, 8)
189 HVBILIN(mmxext, 8, 8, 16)
190 HVBILIN(mmxext, 8, 16, 16)
191 HVBILIN(sse2, 8, 8, 16)
192 HVBILIN(sse2, 8, 16, 16)
193 HVBILIN(ssse3, 8, 8, 16)
194 HVBILIN(ssse3, 8, 16, 16)
196 extern void ff_vp8_idct_dc_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
197 extern void ff_vp8_idct_dc_add_sse4(uint8_t *dst, DCTELEM block[16], int stride);
198 extern void ff_vp8_luma_dc_wht_mmxext(DCTELEM block[4][4][16], DCTELEM dc[16]);
199 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
202 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
203 c->put_vp8_epel_pixels_tab[IDX][0][2] = ff_put_vp8_epel ## SIZE ## _h6_ ## OPT; \
204 c->put_vp8_epel_pixels_tab[IDX][2][0] = ff_put_vp8_epel ## SIZE ## _v6_ ## OPT; \
205 c->put_vp8_epel_pixels_tab[IDX][2][2] = ff_put_vp8_epel ## SIZE ## _h6v6_ ## OPT
207 #define VP8_MC_FUNC(IDX, SIZE, OPT) \
208 c->put_vp8_epel_pixels_tab[IDX][0][1] = ff_put_vp8_epel ## SIZE ## _h4_ ## OPT; \
209 c->put_vp8_epel_pixels_tab[IDX][1][0] = ff_put_vp8_epel ## SIZE ## _v4_ ## OPT; \
210 c->put_vp8_epel_pixels_tab[IDX][1][1] = ff_put_vp8_epel ## SIZE ## _h4v4_ ## OPT; \
211 c->put_vp8_epel_pixels_tab[IDX][1][2] = ff_put_vp8_epel ## SIZE ## _h6v4_ ## OPT; \
212 c->put_vp8_epel_pixels_tab[IDX][2][1] = ff_put_vp8_epel ## SIZE ## _h4v6_ ## OPT; \
213 VP8_LUMA_MC_FUNC(IDX, SIZE, OPT)
215 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE, OPT) \
216 c->put_vp8_bilinear_pixels_tab[IDX][0][1] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
217 c->put_vp8_bilinear_pixels_tab[IDX][0][2] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
218 c->put_vp8_bilinear_pixels_tab[IDX][1][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
219 c->put_vp8_bilinear_pixels_tab[IDX][1][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
220 c->put_vp8_bilinear_pixels_tab[IDX][1][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
221 c->put_vp8_bilinear_pixels_tab[IDX][2][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
222 c->put_vp8_bilinear_pixels_tab[IDX][2][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
223 c->put_vp8_bilinear_pixels_tab[IDX][2][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT
226 av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
228 mm_flags = mm_support();
231 if (mm_flags & FF_MM_MMX) {
232 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmx;
233 c->vp8_idct_add = ff_vp8_idct_add_mmx;
234 c->put_vp8_epel_pixels_tab[0][0][0] =
235 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
236 c->put_vp8_epel_pixels_tab[1][0][0] =
237 c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
240 /* note that 4-tap width=16 functions are missing because w=16
241 * is only used for luma, and luma is always a copy or sixtap. */
242 if (mm_flags & FF_MM_MMX2) {
243 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmxext;
244 VP8_LUMA_MC_FUNC(0, 16, mmxext);
245 VP8_MC_FUNC(1, 8, mmxext);
246 VP8_MC_FUNC(1, 4, mmxext);
247 VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
248 VP8_BILINEAR_MC_FUNC(1, 8, mmxext);
249 VP8_BILINEAR_MC_FUNC(1, 4, mmxext);
252 if (mm_flags & FF_MM_SSE) {
253 c->put_vp8_epel_pixels_tab[0][0][0] =
254 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
257 if (mm_flags & FF_MM_SSE2) {
258 VP8_LUMA_MC_FUNC(0, 16, sse2);
259 VP8_MC_FUNC(1, 8, sse2);
260 VP8_BILINEAR_MC_FUNC(0, 16, sse2);
261 VP8_BILINEAR_MC_FUNC(1, 8, sse2);
264 if (mm_flags & FF_MM_SSSE3) {
265 VP8_LUMA_MC_FUNC(0, 16, ssse3);
266 VP8_MC_FUNC(1, 8, ssse3);
267 VP8_BILINEAR_MC_FUNC(0, 16, ssse3);
268 VP8_BILINEAR_MC_FUNC(1, 8, ssse3);
271 if (mm_flags & FF_MM_SSE4) {
272 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_sse4;