Mercurial > libavcodec.hg
annotate x86/vp8dsp-init.c @ 12340:2d15f62f4f8a libavcodec
VP8: move zeroing of luma DC block into the WHT
Lets us do the zeroing in asm instead of C.
Also makes it consistent with the way the regular iDCT code does it.
author | darkshikari |
---|---|
date | Mon, 02 Aug 2010 20:18:09 +0000 |
parents | 435319d67bd8 |
children | 3fc4c625b6f3 |
rev | line source |
---|---|
11975 | 1 /* |
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> | |
5 * | |
6 * This file is part of FFmpeg. | |
7 * | |
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. | |
12 * | |
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. | |
17 * | |
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 | |
21 */ | |
22 | |
23 #include "libavutil/x86_cpu.h" | |
24 #include "libavcodec/vp8dsp.h" | |
25 | |
11991 | 26 #if HAVE_YASM |
27 | |
11975 | 28 /* |
29 * MC functions | |
30 */ | |
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); | |
43 | |
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); | |
56 | |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
57 extern void ff_put_vp8_epel4_h4_ssse3 (uint8_t *dst, int dststride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
58 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
59 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
60 extern void ff_put_vp8_epel4_h6_ssse3 (uint8_t *dst, int dststride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
61 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
62 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
63 extern void ff_put_vp8_epel4_v4_ssse3 (uint8_t *dst, int dststride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
64 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
65 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
66 extern void ff_put_vp8_epel4_v6_ssse3 (uint8_t *dst, int dststride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
67 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
68 int height, int mx, int my); |
11975 | 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); | |
81 | |
11991 | 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); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
88 extern void ff_put_vp8_bilinear4_h_ssse3 (uint8_t *dst, int dststride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
89 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
90 int height, int mx, int my); |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
91 extern void ff_put_vp8_bilinear8_h_ssse3 (uint8_t *dst, int dststride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
92 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
93 int height, int mx, int my); |
11991 | 94 |
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); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
101 extern void ff_put_vp8_bilinear4_v_ssse3 (uint8_t *dst, int dststride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
102 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
103 int height, int mx, int my); |
11991 | 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); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
107 |
11991 | 108 |
11992 | 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); | |
118 | |
11991 | 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) \ | |
11975 | 123 { \ |
11991 | 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); \ | |
11975 | 128 } |
11991 | 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) \ | |
11975 | 133 { \ |
11991 | 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); \ | |
11975 | 138 } |
139 | |
11991 | 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) | |
11975 | 150 |
11991 | 151 TAP_W16(sse2, epel, h6) |
152 TAP_W16(sse2, epel, v6) | |
153 TAP_W16(sse2, bilinear, h) | |
154 TAP_W16(sse2, bilinear, v) | |
11975 | 155 |
11991 | 156 TAP_W16(ssse3, epel, h6) |
157 TAP_W16(ssse3, epel, v6) | |
158 TAP_W16(ssse3, bilinear, h) | |
159 TAP_W16(ssse3, bilinear, v) | |
11975 | 160 |
161 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \ | |
11991 | 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) \ | |
11975 | 165 { \ |
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); \ | |
11991 | 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); \ | |
11975 | 173 } |
174 | |
175 #define HVTAPMMX(x, y) \ | |
176 HVTAP(mmxext, 8, x, y, 4, 8) \ | |
177 HVTAP(mmxext, 8, x, y, 8, 16) | |
178 | |
179 HVTAPMMX(4, 4) | |
180 HVTAPMMX(4, 6) | |
181 HVTAPMMX(6, 4) | |
182 HVTAPMMX(6, 6) | |
183 HVTAP(mmxext, 8, 6, 6, 16, 16) | |
184 | |
185 #define HVTAPSSE2(x, y, w) \ | |
186 HVTAP(sse2, 16, x, y, w, 16) \ | |
187 HVTAP(ssse3, 16, x, y, w, 16) | |
188 | |
189 HVTAPSSE2(4, 4, 8) | |
190 HVTAPSSE2(4, 6, 8) | |
191 HVTAPSSE2(6, 4, 8) | |
192 HVTAPSSE2(6, 6, 8) | |
193 HVTAPSSE2(6, 6, 16) | |
11991 | 194 |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
195 HVTAP(ssse3, 16, 4, 4, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
196 HVTAP(ssse3, 16, 4, 6, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
197 HVTAP(ssse3, 16, 6, 4, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
198 HVTAP(ssse3, 16, 6, 6, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
199 |
11991 | 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) \ | |
204 { \ | |
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); \ | |
210 } | |
211 | |
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) | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
217 HVBILIN(ssse3, 8, 4, 8) |
11991 | 218 HVBILIN(ssse3, 8, 8, 16) |
219 HVBILIN(ssse3, 8, 16, 16) | |
11975 | 220 |
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); | |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
223 extern void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, DCTELEM block[4][16], int stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
224 extern void ff_vp8_idct_dc_add4y_sse2(uint8_t *dst, DCTELEM block[4][16], int stride); |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
225 extern void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, DCTELEM block[2][16], int stride); |
12209 | 226 extern void ff_vp8_luma_dc_wht_mmx(DCTELEM block[4][4][16], DCTELEM dc[16]); |
12340
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12334
diff
changeset
|
227 extern void ff_vp8_luma_dc_wht_sse(DCTELEM block[4][4][16], DCTELEM dc[16]); |
12013 | 228 extern void ff_vp8_idct_add_mmx(uint8_t *dst, DCTELEM block[16], int stride); |
12235
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12227
diff
changeset
|
229 extern void ff_vp8_idct_add_sse(uint8_t *dst, DCTELEM block[16], int stride); |
12086
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
230 |
12210 | 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); | |
12204
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
250 |
12210 | 251 DECLARE_LOOP_FILTER(mmx) |
252 DECLARE_LOOP_FILTER(mmxext) | |
253 DECLARE_LOOP_FILTER(sse2) | |
254 DECLARE_LOOP_FILTER(ssse3) | |
12227
d07e6037846d
Use pextrw for SSE4 mbedge filter result writing, speedup 5-10cycles on
rbultje
parents:
12214
diff
changeset
|
255 DECLARE_LOOP_FILTER(sse4) |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
256 |
11991 | 257 #endif |
258 | |
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 | |
263 | |
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) | |
271 | |
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 | |
281 | |
11975 | 282 |
283 av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c) | |
284 { | |
285 mm_flags = mm_support(); | |
286 | |
11976 | 287 #if HAVE_YASM |
11975 | 288 if (mm_flags & FF_MM_MMX) { |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
289 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmx; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
290 c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_mmx; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
291 c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmx; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
292 c->vp8_idct_add = ff_vp8_idct_add_mmx; |
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
293 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmx; |
11992 | 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; | |
12086
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
298 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
299 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx; |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
300 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx; |
12168
b246b214c2e9
VP8 H/V inner loopfilter MMX/MMXEXT/SSE2 optimizations.
rbultje
parents:
12086
diff
changeset
|
301 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
302 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
303 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx; |
12204
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
304 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmx; |
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
305 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmx; |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
306 |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
307 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_mmx; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
308 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_mmx; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
309 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_mmx; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
310 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmx; |
11975 | 311 } |
312 | |
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. */ | |
11993 | 315 if (mm_flags & FF_MM_MMX2) { |
11991 | 316 VP8_LUMA_MC_FUNC(0, 16, mmxext); |
317 VP8_MC_FUNC(1, 8, mmxext); | |
12042 | 318 VP8_MC_FUNC(2, 4, mmxext); |
11991 | 319 VP8_BILINEAR_MC_FUNC(0, 16, mmxext); |
320 VP8_BILINEAR_MC_FUNC(1, 8, mmxext); | |
12042 | 321 VP8_BILINEAR_MC_FUNC(2, 4, mmxext); |
12086
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
322 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
323 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmxext; |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
324 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmxext; |
12168
b246b214c2e9
VP8 H/V inner loopfilter MMX/MMXEXT/SSE2 optimizations.
rbultje
parents:
12086
diff
changeset
|
325 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
326 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext; |
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
327 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext; |
12204
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
328 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmxext; |
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
329 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmxext; |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
330 |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
331 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_mmxext; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
332 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_mmxext; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
333 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_mmxext; |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
334 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmxext; |
11975 | 335 } |
336 | |
11992 | 337 if (mm_flags & FF_MM_SSE) { |
12235
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12227
diff
changeset
|
338 c->vp8_idct_add = ff_vp8_idct_add_sse; |
12340
2d15f62f4f8a
VP8: move zeroing of luma DC block into the WHT
darkshikari
parents:
12334
diff
changeset
|
339 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_sse; |
11992 | 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; | |
342 } | |
343 | |
12197
fbf4d5b1b664
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
rbultje
parents:
12196
diff
changeset
|
344 if (mm_flags & (FF_MM_SSE2|FF_MM_SSE2SLOW)) { |
11991 | 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); | |
12086
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
349 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
350 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2; |
12168
b246b214c2e9
VP8 H/V inner loopfilter MMX/MMXEXT/SSE2 optimizations.
rbultje
parents:
12086
diff
changeset
|
351 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
352 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2; |
12204
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
353 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_sse2; |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
354 |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
355 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_sse2; |
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
356 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_sse2; |
12197
fbf4d5b1b664
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
rbultje
parents:
12196
diff
changeset
|
357 } |
fbf4d5b1b664
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
rbultje
parents:
12196
diff
changeset
|
358 |
fbf4d5b1b664
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
rbultje
parents:
12196
diff
changeset
|
359 if (mm_flags & FF_MM_SSE2) { |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
360 c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_sse2; |
12238 | 361 |
12334
435319d67bd8
Use word-writing instead of dword-writing (with two cached but otherwise
rbultje
parents:
12241
diff
changeset
|
362 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2; |
435319d67bd8
Use word-writing instead of dword-writing (with two cached but otherwise
rbultje
parents:
12241
diff
changeset
|
363 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
364 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2; |
12204
563339ea87aa
Chroma (width=8) inner loopfilter MMX/MMX2/SSE2 for VP8 decoder.
rbultje
parents:
12198
diff
changeset
|
365 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_sse2; |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
366 |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
367 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_sse2; |
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
368 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse2; |
11975 | 369 } |
370 | |
371 if (mm_flags & FF_MM_SSSE3) { | |
11991 | 372 VP8_LUMA_MC_FUNC(0, 16, ssse3); |
373 VP8_MC_FUNC(1, 8, ssse3); | |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
374 VP8_MC_FUNC(2, 4, ssse3); |
11991 | 375 VP8_BILINEAR_MC_FUNC(0, 16, ssse3); |
376 VP8_BILINEAR_MC_FUNC(1, 8, ssse3); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
377 VP8_BILINEAR_MC_FUNC(2, 4, ssse3); |
12210 | 378 |
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; | |
381 | |
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; | |
386 | |
387 c->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16y_mbedge_ssse3; | |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
388 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_ssse3; |
12210 | 389 c->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mbedge_ssse3; |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
390 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_ssse3; |
11975 | 391 } |
392 | |
393 if (mm_flags & FF_MM_SSE4) { | |
394 c->vp8_idct_dc_add = ff_vp8_idct_dc_add_sse4; | |
12227
d07e6037846d
Use pextrw for SSE4 mbedge filter result writing, speedup 5-10cycles on
rbultje
parents:
12214
diff
changeset
|
395 |
12334
435319d67bd8
Use word-writing instead of dword-writing (with two cached but otherwise
rbultje
parents:
12241
diff
changeset
|
396 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse4; |
12227
d07e6037846d
Use pextrw for SSE4 mbedge filter result writing, speedup 5-10cycles on
rbultje
parents:
12214
diff
changeset
|
397 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_sse4; |
d07e6037846d
Use pextrw for SSE4 mbedge filter result writing, speedup 5-10cycles on
rbultje
parents:
12214
diff
changeset
|
398 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse4; |
11975 | 399 } |
11976 | 400 #endif |
11975 | 401 } |