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_epel4_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_epel4_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_epel4_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_epel4_v6_ssse3 (uint8_t *dst, int dststride,
67 uint8_t *src, int srcstride,
68 int height, int mx, int my);
69 extern void ff_put_vp8_epel8_h4_ssse3 (uint8_t *dst, int dststride,
70 uint8_t *src, int srcstride,
71 int height, int mx, int my);
72 extern void ff_put_vp8_epel8_h6_ssse3 (uint8_t *dst, int dststride,
73 uint8_t *src, int srcstride,
74 int height, int mx, int my);
75 extern void ff_put_vp8_epel8_v4_ssse3 (uint8_t *dst, int dststride,
76 uint8_t *src, int srcstride,
77 int height, int mx, int my);
78 extern void ff_put_vp8_epel8_v6_ssse3 (uint8_t *dst, int dststride,
79 uint8_t *src, int srcstride,
80 int height, int mx, int my);
82 extern void ff_put_vp8_bilinear4_h_mmxext(uint8_t *dst, int dststride,
83 uint8_t *src, int srcstride,
84 int height, int mx, int my);
85 extern void ff_put_vp8_bilinear8_h_sse2 (uint8_t *dst, int dststride,
86 uint8_t *src, int srcstride,
87 int height, int mx, int my);
88 extern void ff_put_vp8_bilinear4_h_ssse3 (uint8_t *dst, int dststride,
89 uint8_t *src, int srcstride,
90 int height, int mx, int my);
91 extern void ff_put_vp8_bilinear8_h_ssse3 (uint8_t *dst, int dststride,
92 uint8_t *src, int srcstride,
93 int height, int mx, int my);
95 extern void ff_put_vp8_bilinear4_v_mmxext(uint8_t *dst, int dststride,
96 uint8_t *src, int srcstride,
97 int height, int mx, int my);
98 extern void ff_put_vp8_bilinear8_v_sse2 (uint8_t *dst, int dststride,
99 uint8_t *src, int srcstride,
100 int height, int mx, int my);
101 extern void ff_put_vp8_bilinear4_v_ssse3 (uint8_t *dst, int dststride,
102 uint8_t *src, int srcstride,
103 int height, int mx, int my);
104 extern void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, int dststride,
105 uint8_t *src, int srcstride,
106 int height, int mx, int my);
109 extern void ff_put_vp8_pixels8_mmx (uint8_t *dst, int dststride,
110 uint8_t *src, int srcstride,
111 int height, int mx, int my);
112 extern void ff_put_vp8_pixels16_mmx(uint8_t *dst, int dststride,
113 uint8_t *src, int srcstride,
114 int height, int mx, int my);
115 extern void ff_put_vp8_pixels16_sse(uint8_t *dst, int dststride,
116 uint8_t *src, int srcstride,
117 int height, int mx, int my);
119 #define TAP_W16(OPT, FILTERTYPE, TAPTYPE) \
120 static void ff_put_vp8_ ## FILTERTYPE ## 16_ ## TAPTYPE ## _ ## OPT( \
121 uint8_t *dst, int dststride, uint8_t *src, \
122 int srcstride, int height, int mx, int my) \
124 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
125 dst, dststride, src, srcstride, height, mx, my); \
126 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
127 dst + 8, dststride, src + 8, srcstride, height, mx, my); \
129 #define TAP_W8(OPT, FILTERTYPE, TAPTYPE) \
130 static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
131 uint8_t *dst, int dststride, uint8_t *src, \
132 int srcstride, int height, int mx, int my) \
134 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
135 dst, dststride, src, srcstride, height, mx, my); \
136 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
137 dst + 4, dststride, src + 4, srcstride, height, mx, my); \
140 TAP_W8 (mmxext, epel, h4)
141 TAP_W8 (mmxext, epel, h6)
142 TAP_W16(mmxext, epel, h6)
143 TAP_W8 (mmxext, epel, v4)
144 TAP_W8 (mmxext, epel, v6)
145 TAP_W16(mmxext, epel, v6)
146 TAP_W8 (mmxext, bilinear, h)
147 TAP_W16(mmxext, bilinear, h)
148 TAP_W8 (mmxext, bilinear, v)
149 TAP_W16(mmxext, bilinear, v)
151 TAP_W16(sse2, epel, h6)
152 TAP_W16(sse2, epel, v6)
153 TAP_W16(sse2, bilinear, h)
154 TAP_W16(sse2, bilinear, v)
156 TAP_W16(ssse3, epel, h6)
157 TAP_W16(ssse3, epel, v6)
158 TAP_W16(ssse3, bilinear, h)
159 TAP_W16(ssse3, bilinear, v)
161 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \
162 static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT( \
163 uint8_t *dst, int dststride, uint8_t *src, \
164 int srcstride, int height, int mx, int my) \
166 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \
167 uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \
168 src -= srcstride * (TAPNUMY / 2 - 1); \
169 ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT( \
170 tmp, SIZE, src, srcstride, height + TAPNUMY - 1, mx, my); \
171 ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT( \
172 dst, dststride, tmpptr, SIZE, height, mx, my); \
175 #define HVTAPMMX(x, y) \
176 HVTAP(mmxext, 8, x, y, 4, 8) \
177 HVTAP(mmxext, 8, x, y, 8, 16)
183 HVTAP(mmxext, 8, 6, 6, 16, 16)
185 #define HVTAPSSE2(x, y, w) \
186 HVTAP(sse2, 16, x, y, w, 16) \
187 HVTAP(ssse3, 16, x, y, w, 16)
195 HVTAP(ssse3, 16, 4, 4, 4, 8)
196 HVTAP(ssse3, 16, 4, 6, 4, 8)
197 HVTAP(ssse3, 16, 6, 4, 4, 8)
198 HVTAP(ssse3, 16, 6, 6, 4, 8)
200 #define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \
201 static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \
202 uint8_t *dst, int dststride, uint8_t *src, \
203 int srcstride, int height, int mx, int my) \
205 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + 2)]; \
206 ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT( \
207 tmp, SIZE, src, srcstride, height + 1, mx, my); \
208 ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT( \
209 dst, dststride, tmp, SIZE, height, mx, my); \
212 HVBILIN(mmxext, 8, 4, 8)
213 HVBILIN(mmxext, 8, 8, 16)
214 HVBILIN(mmxext, 8, 16, 16)
215 HVBILIN(sse2, 8, 8, 16)
216 HVBILIN(sse2, 8, 16, 16)
217 HVBILIN(ssse3, 8, 4, 8)
218 HVBILIN(ssse3, 8, 8, 16)
219 HVBILIN(ssse3, 8, 16, 16)
221 extern void ff_vp8_idct_dc_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
222 extern void ff_vp8_idct_dc_add_sse4(uint8_t *dst, DCTELEM block[16], int stride);
223 extern void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, DCTELEM block[4][16], int stride);
224 extern void ff_vp8_idct_dc_add4y_sse2(uint8_t *dst, DCTELEM block[4][16], int stride);
225 extern void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, DCTELEM block[2][16], int stride);
226 extern void ff_vp8_luma_dc_wht_mmx(DCTELEM block[4][4][16], DCTELEM dc[16]);
227 extern void ff_vp8_luma_dc_wht_sse(DCTELEM block[4][4][16], DCTELEM dc[16]);
228 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride);
229 extern void ff_vp8_idct_add_sse(uint8_t *dst, DCTELEM block[16], int stride);
231 #define DECLARE_LOOP_FILTER(NAME)\
232 extern void ff_vp8_v_loop_filter_simple_ ## NAME(uint8_t *dst, int stride, int flim);\
233 extern void ff_vp8_h_loop_filter_simple_ ## NAME(uint8_t *dst, int stride, int flim);\
234 extern void ff_vp8_v_loop_filter16y_inner_ ## NAME (uint8_t *dst, int stride,\
235 int e, int i, int hvt);\
236 extern void ff_vp8_h_loop_filter16y_inner_ ## NAME (uint8_t *dst, int stride,\
237 int e, int i, int hvt);\
238 extern void ff_vp8_v_loop_filter8uv_inner_ ## NAME (uint8_t *dstU, uint8_t *dstV,\
239 int s, int e, int i, int hvt);\
240 extern void ff_vp8_h_loop_filter8uv_inner_ ## NAME (uint8_t *dstU, uint8_t *dstV,\
241 int s, int e, int i, int hvt);\
242 extern void ff_vp8_v_loop_filter16y_mbedge_ ## NAME(uint8_t *dst, int stride,\
243 int e, int i, int hvt);\
244 extern void ff_vp8_h_loop_filter16y_mbedge_ ## NAME(uint8_t *dst, int stride,\
245 int e, int i, int hvt);\
246 extern void ff_vp8_v_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU, uint8_t *dstV,\
247 int s, int e, int i, int hvt);\
248 extern void ff_vp8_h_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU, uint8_t *dstV,\
249 int s, int e, int i, int hvt);
251 DECLARE_LOOP_FILTER(mmx)
252 DECLARE_LOOP_FILTER(mmxext)
253 DECLARE_LOOP_FILTER(sse2)
254 DECLARE_LOOP_FILTER(ssse3)
255 DECLARE_LOOP_FILTER(sse4)
259 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
260 c->put_vp8_epel_pixels_tab[IDX][0][2] = ff_put_vp8_epel ## SIZE ## _h6_ ## OPT; \
261 c->put_vp8_epel_pixels_tab[IDX][2][0] = ff_put_vp8_epel ## SIZE ## _v6_ ## OPT; \
262 c->put_vp8_epel_pixels_tab[IDX][2][2] = ff_put_vp8_epel ## SIZE ## _h6v6_ ## OPT
264 #define VP8_MC_FUNC(IDX, SIZE, OPT) \
265 c->put_vp8_epel_pixels_tab[IDX][0][1] = ff_put_vp8_epel ## SIZE ## _h4_ ## OPT; \
266 c->put_vp8_epel_pixels_tab[IDX][1][0] = ff_put_vp8_epel ## SIZE ## _v4_ ## OPT; \
267 c->put_vp8_epel_pixels_tab[IDX][1][1] = ff_put_vp8_epel ## SIZE ## _h4v4_ ## OPT; \
268 c->put_vp8_epel_pixels_tab[IDX][1][2] = ff_put_vp8_epel ## SIZE ## _h6v4_ ## OPT; \
269 c->put_vp8_epel_pixels_tab[IDX][2][1] = ff_put_vp8_epel ## SIZE ## _h4v6_ ## OPT; \
270 VP8_LUMA_MC_FUNC(IDX, SIZE, OPT)
272 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE, OPT) \
273 c->put_vp8_bilinear_pixels_tab[IDX][0][1] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
274 c->put_vp8_bilinear_pixels_tab[IDX][0][2] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
275 c->put_vp8_bilinear_pixels_tab[IDX][1][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
276 c->put_vp8_bilinear_pixels_tab[IDX][1][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
277 c->put_vp8_bilinear_pixels_tab[IDX][1][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
278 c->put_vp8_bilinear_pixels_tab[IDX][2][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
279 c->put_vp8_bilinear_pixels_tab[IDX][2][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
280 c->put_vp8_bilinear_pixels_tab[IDX][2][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT
283 av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c)
285 mm_flags = mm_support();
288 if (mm_flags & FF_MM_MMX) {
289 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmx;
290 c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_mmx;
291 c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmx;
292 c->vp8_idct_add = ff_vp8_idct_add_mmx;
293 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmx;
294 c->put_vp8_epel_pixels_tab[0][0][0] =
295 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
296 c->put_vp8_epel_pixels_tab[1][0][0] =
297 c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
299 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx;
300 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx;
302 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx;
303 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx;
304 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmx;
305 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmx;
307 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_mmx;
308 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_mmx;
309 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_mmx;
310 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmx;
313 /* note that 4-tap width=16 functions are missing because w=16
314 * is only used for luma, and luma is always a copy or sixtap. */
315 if (mm_flags & FF_MM_MMX2) {
316 VP8_LUMA_MC_FUNC(0, 16, mmxext);
317 VP8_MC_FUNC(1, 8, mmxext);
318 VP8_MC_FUNC(2, 4, mmxext);
319 VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
320 VP8_BILINEAR_MC_FUNC(1, 8, mmxext);
321 VP8_BILINEAR_MC_FUNC(2, 4, mmxext);
323 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmxext;
324 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmxext;
326 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext;
327 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext;
328 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmxext;
329 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmxext;
331 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_mmxext;
332 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_mmxext;
333 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_mmxext;
334 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmxext;
337 if (mm_flags & FF_MM_SSE) {
338 c->vp8_idct_add = ff_vp8_idct_add_sse;
339 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_sse;
340 c->put_vp8_epel_pixels_tab[0][0][0] =
341 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
344 if (mm_flags & (FF_MM_SSE2|FF_MM_SSE2SLOW)) {
345 VP8_LUMA_MC_FUNC(0, 16, sse2);
346 VP8_MC_FUNC(1, 8, sse2);
347 VP8_BILINEAR_MC_FUNC(0, 16, sse2);
348 VP8_BILINEAR_MC_FUNC(1, 8, sse2);
350 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2;
352 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2;
353 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_sse2;
355 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_sse2;
356 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_sse2;
359 if (mm_flags & FF_MM_SSE2) {
360 c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_sse2;
362 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2;
364 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2;
365 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_sse2;
367 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_sse2;
368 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse2;
371 if (mm_flags & FF_MM_SSSE3) {
372 VP8_LUMA_MC_FUNC(0, 16, ssse3);
373 VP8_MC_FUNC(1, 8, ssse3);
374 VP8_MC_FUNC(2, 4, ssse3);
375 VP8_BILINEAR_MC_FUNC(0, 16, ssse3);
376 VP8_BILINEAR_MC_FUNC(1, 8, ssse3);
377 VP8_BILINEAR_MC_FUNC(2, 4, ssse3);
379 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_ssse3;
380 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_ssse3;
382 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_ssse3;
383 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_ssse3;
384 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_ssse3;
385 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_ssse3;
387 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_ssse3;
388 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_ssse3;
389 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_ssse3;
390 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_ssse3;
393 if (mm_flags & FF_MM_SSE4) {
394 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_sse4;
396 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse4;
397 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_sse4;
398 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse4;