Mercurial > libavcodec.hg
annotate x86/vp8dsp-init.c @ 12497:c5ffa8b81f9c libavcodec
Move sse16_sse2() from inline asm to yasm. It is one of the functions causing
Win64/FATE issues.
author | rbultje |
---|---|
date | Fri, 17 Sep 2010 01:44:17 +0000 |
parents | 9fef0a8ddd63 |
children |
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 | |
12475
9fef0a8ddd63
Move mm_support() from libavcodec to libavutil, make it a public
stefano
parents:
12456
diff
changeset
|
23 #include "libavutil/cpu.h" |
11975 | 24 #include "libavutil/x86_cpu.h" |
25 #include "libavcodec/vp8dsp.h" | |
26 | |
11991 | 27 #if HAVE_YASM |
28 | |
11975 | 29 /* |
30 * MC functions | |
31 */ | |
32 extern void ff_put_vp8_epel4_h4_mmxext(uint8_t *dst, int dststride, | |
33 uint8_t *src, int srcstride, | |
34 int height, int mx, int my); | |
35 extern void ff_put_vp8_epel4_h6_mmxext(uint8_t *dst, int dststride, | |
36 uint8_t *src, int srcstride, | |
37 int height, int mx, int my); | |
38 extern void ff_put_vp8_epel4_v4_mmxext(uint8_t *dst, int dststride, | |
39 uint8_t *src, int srcstride, | |
40 int height, int mx, int my); | |
41 extern void ff_put_vp8_epel4_v6_mmxext(uint8_t *dst, int dststride, | |
42 uint8_t *src, int srcstride, | |
43 int height, int mx, int my); | |
44 | |
45 extern void ff_put_vp8_epel8_h4_sse2 (uint8_t *dst, int dststride, | |
46 uint8_t *src, int srcstride, | |
47 int height, int mx, int my); | |
48 extern void ff_put_vp8_epel8_h6_sse2 (uint8_t *dst, int dststride, | |
49 uint8_t *src, int srcstride, | |
50 int height, int mx, int my); | |
51 extern void ff_put_vp8_epel8_v4_sse2 (uint8_t *dst, int dststride, | |
52 uint8_t *src, int srcstride, | |
53 int height, int mx, int my); | |
54 extern void ff_put_vp8_epel8_v6_sse2 (uint8_t *dst, int dststride, | |
55 uint8_t *src, int srcstride, | |
56 int height, int mx, int my); | |
57 | |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
58 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
|
59 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
60 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
61 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
|
62 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
63 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
64 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
|
65 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
66 int height, int mx, int my); |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
67 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
|
68 uint8_t *src, int srcstride, |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
69 int height, int mx, int my); |
11975 | 70 extern void ff_put_vp8_epel8_h4_ssse3 (uint8_t *dst, int dststride, |
71 uint8_t *src, int srcstride, | |
72 int height, int mx, int my); | |
73 extern void ff_put_vp8_epel8_h6_ssse3 (uint8_t *dst, int dststride, | |
74 uint8_t *src, int srcstride, | |
75 int height, int mx, int my); | |
76 extern void ff_put_vp8_epel8_v4_ssse3 (uint8_t *dst, int dststride, | |
77 uint8_t *src, int srcstride, | |
78 int height, int mx, int my); | |
79 extern void ff_put_vp8_epel8_v6_ssse3 (uint8_t *dst, int dststride, | |
80 uint8_t *src, int srcstride, | |
81 int height, int mx, int my); | |
82 | |
11991 | 83 extern void ff_put_vp8_bilinear4_h_mmxext(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_sse2 (uint8_t *dst, int dststride, | |
87 uint8_t *src, int srcstride, | |
88 int height, int mx, int my); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
89 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
|
90 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
91 int height, int mx, int my); |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
92 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
|
93 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
94 int height, int mx, int my); |
11991 | 95 |
96 extern void ff_put_vp8_bilinear4_v_mmxext(uint8_t *dst, int dststride, | |
97 uint8_t *src, int srcstride, | |
98 int height, int mx, int my); | |
99 extern void ff_put_vp8_bilinear8_v_sse2 (uint8_t *dst, int dststride, | |
100 uint8_t *src, int srcstride, | |
101 int height, int mx, int my); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
102 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
|
103 uint8_t *src, int srcstride, |
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
104 int height, int mx, int my); |
11991 | 105 extern void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, int dststride, |
106 uint8_t *src, int srcstride, | |
107 int height, int mx, int my); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
108 |
11991 | 109 |
11992 | 110 extern void ff_put_vp8_pixels8_mmx (uint8_t *dst, int dststride, |
111 uint8_t *src, int srcstride, | |
112 int height, int mx, int my); | |
113 extern void ff_put_vp8_pixels16_mmx(uint8_t *dst, int dststride, | |
114 uint8_t *src, int srcstride, | |
115 int height, int mx, int my); | |
116 extern void ff_put_vp8_pixels16_sse(uint8_t *dst, int dststride, | |
117 uint8_t *src, int srcstride, | |
118 int height, int mx, int my); | |
119 | |
11991 | 120 #define TAP_W16(OPT, FILTERTYPE, TAPTYPE) \ |
121 static void ff_put_vp8_ ## FILTERTYPE ## 16_ ## TAPTYPE ## _ ## OPT( \ | |
122 uint8_t *dst, int dststride, uint8_t *src, \ | |
123 int srcstride, int height, int mx, int my) \ | |
11975 | 124 { \ |
11991 | 125 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \ |
126 dst, dststride, src, srcstride, height, mx, my); \ | |
127 ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \ | |
128 dst + 8, dststride, src + 8, srcstride, height, mx, my); \ | |
11975 | 129 } |
11991 | 130 #define TAP_W8(OPT, FILTERTYPE, TAPTYPE) \ |
131 static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \ | |
132 uint8_t *dst, int dststride, uint8_t *src, \ | |
133 int srcstride, int height, int mx, int my) \ | |
11975 | 134 { \ |
11991 | 135 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \ |
136 dst, dststride, src, srcstride, height, mx, my); \ | |
137 ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \ | |
138 dst + 4, dststride, src + 4, srcstride, height, mx, my); \ | |
11975 | 139 } |
140 | |
11991 | 141 TAP_W8 (mmxext, epel, h4) |
142 TAP_W8 (mmxext, epel, h6) | |
143 TAP_W16(mmxext, epel, h6) | |
144 TAP_W8 (mmxext, epel, v4) | |
145 TAP_W8 (mmxext, epel, v6) | |
146 TAP_W16(mmxext, epel, v6) | |
147 TAP_W8 (mmxext, bilinear, h) | |
148 TAP_W16(mmxext, bilinear, h) | |
149 TAP_W8 (mmxext, bilinear, v) | |
150 TAP_W16(mmxext, bilinear, v) | |
11975 | 151 |
11991 | 152 TAP_W16(sse2, epel, h6) |
153 TAP_W16(sse2, epel, v6) | |
154 TAP_W16(sse2, bilinear, h) | |
155 TAP_W16(sse2, bilinear, v) | |
11975 | 156 |
11991 | 157 TAP_W16(ssse3, epel, h6) |
158 TAP_W16(ssse3, epel, v6) | |
159 TAP_W16(ssse3, bilinear, h) | |
160 TAP_W16(ssse3, bilinear, v) | |
11975 | 161 |
162 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \ | |
11991 | 163 static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT( \ |
164 uint8_t *dst, int dststride, uint8_t *src, \ | |
165 int srcstride, int height, int mx, int my) \ | |
11975 | 166 { \ |
167 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \ | |
168 uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \ | |
169 src -= srcstride * (TAPNUMY / 2 - 1); \ | |
11991 | 170 ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT( \ |
171 tmp, SIZE, src, srcstride, height + TAPNUMY - 1, mx, my); \ | |
172 ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT( \ | |
173 dst, dststride, tmpptr, SIZE, height, mx, my); \ | |
11975 | 174 } |
175 | |
176 #define HVTAPMMX(x, y) \ | |
177 HVTAP(mmxext, 8, x, y, 4, 8) \ | |
178 HVTAP(mmxext, 8, x, y, 8, 16) | |
179 | |
180 HVTAPMMX(4, 4) | |
181 HVTAPMMX(4, 6) | |
182 HVTAPMMX(6, 4) | |
183 HVTAPMMX(6, 6) | |
184 HVTAP(mmxext, 8, 6, 6, 16, 16) | |
185 | |
186 #define HVTAPSSE2(x, y, w) \ | |
187 HVTAP(sse2, 16, x, y, w, 16) \ | |
188 HVTAP(ssse3, 16, x, y, w, 16) | |
189 | |
190 HVTAPSSE2(4, 4, 8) | |
191 HVTAPSSE2(4, 6, 8) | |
192 HVTAPSSE2(6, 4, 8) | |
193 HVTAPSSE2(6, 6, 8) | |
194 HVTAPSSE2(6, 6, 16) | |
11991 | 195 |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
196 HVTAP(ssse3, 16, 4, 4, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
197 HVTAP(ssse3, 16, 4, 6, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
198 HVTAP(ssse3, 16, 6, 4, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
199 HVTAP(ssse3, 16, 6, 6, 4, 8) |
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
200 |
11991 | 201 #define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \ |
202 static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \ | |
203 uint8_t *dst, int dststride, uint8_t *src, \ | |
204 int srcstride, int height, int mx, int my) \ | |
205 { \ | |
206 DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + 2)]; \ | |
207 ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT( \ | |
208 tmp, SIZE, src, srcstride, height + 1, mx, my); \ | |
209 ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT( \ | |
210 dst, dststride, tmp, SIZE, height, mx, my); \ | |
211 } | |
212 | |
213 HVBILIN(mmxext, 8, 4, 8) | |
214 HVBILIN(mmxext, 8, 8, 16) | |
215 HVBILIN(mmxext, 8, 16, 16) | |
216 HVBILIN(sse2, 8, 8, 16) | |
217 HVBILIN(sse2, 8, 16, 16) | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
218 HVBILIN(ssse3, 8, 4, 8) |
11991 | 219 HVBILIN(ssse3, 8, 8, 16) |
220 HVBILIN(ssse3, 8, 16, 16) | |
11975 | 221 |
222 extern void ff_vp8_idct_dc_add_mmx(uint8_t *dst, DCTELEM block[16], int stride); | |
223 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
|
224 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
|
225 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
|
226 extern void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, DCTELEM block[2][16], int stride); |
12209 | 227 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
|
228 extern void ff_vp8_luma_dc_wht_sse(DCTELEM block[4][4][16], DCTELEM dc[16]); |
12013 | 229 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
|
230 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
|
231 |
12210 | 232 #define DECLARE_LOOP_FILTER(NAME)\ |
233 extern void ff_vp8_v_loop_filter_simple_ ## NAME(uint8_t *dst, int stride, int flim);\ | |
234 extern void ff_vp8_h_loop_filter_simple_ ## NAME(uint8_t *dst, int stride, int flim);\ | |
235 extern void ff_vp8_v_loop_filter16y_inner_ ## NAME (uint8_t *dst, int stride,\ | |
236 int e, int i, int hvt);\ | |
237 extern void ff_vp8_h_loop_filter16y_inner_ ## NAME (uint8_t *dst, int stride,\ | |
238 int e, int i, int hvt);\ | |
239 extern void ff_vp8_v_loop_filter8uv_inner_ ## NAME (uint8_t *dstU, uint8_t *dstV,\ | |
240 int s, int e, int i, int hvt);\ | |
241 extern void ff_vp8_h_loop_filter8uv_inner_ ## NAME (uint8_t *dstU, uint8_t *dstV,\ | |
242 int s, int e, int i, int hvt);\ | |
243 extern void ff_vp8_v_loop_filter16y_mbedge_ ## NAME(uint8_t *dst, int stride,\ | |
244 int e, int i, int hvt);\ | |
245 extern void ff_vp8_h_loop_filter16y_mbedge_ ## NAME(uint8_t *dst, int stride,\ | |
246 int e, int i, int hvt);\ | |
247 extern void ff_vp8_v_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU, uint8_t *dstV,\ | |
248 int s, int e, int i, int hvt);\ | |
249 extern void ff_vp8_h_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU, uint8_t *dstV,\ | |
250 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
|
251 |
12210 | 252 DECLARE_LOOP_FILTER(mmx) |
253 DECLARE_LOOP_FILTER(mmxext) | |
254 DECLARE_LOOP_FILTER(sse2) | |
255 DECLARE_LOOP_FILTER(ssse3) | |
12227
d07e6037846d
Use pextrw for SSE4 mbedge filter result writing, speedup 5-10cycles on
rbultje
parents:
12214
diff
changeset
|
256 DECLARE_LOOP_FILTER(sse4) |
12205
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
257 |
11991 | 258 #endif |
259 | |
260 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \ | |
261 c->put_vp8_epel_pixels_tab[IDX][0][2] = ff_put_vp8_epel ## SIZE ## _h6_ ## OPT; \ | |
262 c->put_vp8_epel_pixels_tab[IDX][2][0] = ff_put_vp8_epel ## SIZE ## _v6_ ## OPT; \ | |
263 c->put_vp8_epel_pixels_tab[IDX][2][2] = ff_put_vp8_epel ## SIZE ## _h6v6_ ## OPT | |
264 | |
265 #define VP8_MC_FUNC(IDX, SIZE, OPT) \ | |
266 c->put_vp8_epel_pixels_tab[IDX][0][1] = ff_put_vp8_epel ## SIZE ## _h4_ ## OPT; \ | |
267 c->put_vp8_epel_pixels_tab[IDX][1][0] = ff_put_vp8_epel ## SIZE ## _v4_ ## OPT; \ | |
268 c->put_vp8_epel_pixels_tab[IDX][1][1] = ff_put_vp8_epel ## SIZE ## _h4v4_ ## OPT; \ | |
269 c->put_vp8_epel_pixels_tab[IDX][1][2] = ff_put_vp8_epel ## SIZE ## _h6v4_ ## OPT; \ | |
270 c->put_vp8_epel_pixels_tab[IDX][2][1] = ff_put_vp8_epel ## SIZE ## _h4v6_ ## OPT; \ | |
271 VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) | |
272 | |
273 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE, OPT) \ | |
274 c->put_vp8_bilinear_pixels_tab[IDX][0][1] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \ | |
275 c->put_vp8_bilinear_pixels_tab[IDX][0][2] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \ | |
276 c->put_vp8_bilinear_pixels_tab[IDX][1][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \ | |
277 c->put_vp8_bilinear_pixels_tab[IDX][1][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \ | |
278 c->put_vp8_bilinear_pixels_tab[IDX][1][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \ | |
279 c->put_vp8_bilinear_pixels_tab[IDX][2][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \ | |
280 c->put_vp8_bilinear_pixels_tab[IDX][2][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \ | |
281 c->put_vp8_bilinear_pixels_tab[IDX][2][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT | |
282 | |
11975 | 283 |
284 av_cold void ff_vp8dsp_init_x86(VP8DSPContext* c) | |
285 { | |
12475
9fef0a8ddd63
Move mm_support() from libavcodec to libavutil, make it a public
stefano
parents:
12456
diff
changeset
|
286 int mm_flags = av_get_cpu_flags(); |
11975 | 287 |
11976 | 288 #if HAVE_YASM |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
289 if (mm_flags & AV_CPU_FLAG_MMX) { |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
290 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
|
291 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
|
292 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
|
293 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
|
294 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmx; |
11992 | 295 c->put_vp8_epel_pixels_tab[0][0][0] = |
296 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx; | |
297 c->put_vp8_epel_pixels_tab[1][0][0] = | |
298 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
|
299 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
300 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
|
301 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
|
302 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
303 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
|
304 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
|
305 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
|
306 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
|
307 |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
308 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
|
309 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
|
310 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
|
311 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmx; |
11975 | 312 } |
313 | |
314 /* note that 4-tap width=16 functions are missing because w=16 | |
315 * is only used for luma, and luma is always a copy or sixtap. */ | |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
316 if (mm_flags & AV_CPU_FLAG_MMX2) { |
11991 | 317 VP8_LUMA_MC_FUNC(0, 16, mmxext); |
318 VP8_MC_FUNC(1, 8, mmxext); | |
12042 | 319 VP8_MC_FUNC(2, 4, mmxext); |
11991 | 320 VP8_BILINEAR_MC_FUNC(0, 16, mmxext); |
321 VP8_BILINEAR_MC_FUNC(1, 8, mmxext); | |
12042 | 322 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
|
323 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
324 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
|
325 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
|
326 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
327 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
|
328 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
|
329 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
|
330 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
|
331 |
d38e8565ba05
VP8 MBedge loopfilter MMX/MMX2/SSE2 functions for both luma (width=16)
rbultje
parents:
12204
diff
changeset
|
332 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
|
333 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
|
334 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
|
335 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_mmxext; |
11975 | 336 } |
337 | |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
338 if (mm_flags & AV_CPU_FLAG_SSE) { |
12235
e08d65897115
VP8: clear DCT blocks in iDCT instead of using clear_blocks.
darkshikari
parents:
12227
diff
changeset
|
339 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
|
340 c->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_sse; |
11992 | 341 c->put_vp8_epel_pixels_tab[0][0][0] = |
342 c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse; | |
343 } | |
344 | |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
345 if (mm_flags & (AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW)) { |
11991 | 346 VP8_LUMA_MC_FUNC(0, 16, sse2); |
347 VP8_MC_FUNC(1, 8, sse2); | |
348 VP8_BILINEAR_MC_FUNC(0, 16, sse2); | |
349 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
|
350 |
d780ae746855
Simple H/V loopfilter for VP8 in MMX, MMX2 and SSE2 (yay for yasm macros).
rbultje
parents:
12082
diff
changeset
|
351 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
|
352 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
353 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
|
354 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
|
355 |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
356 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
|
357 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
|
358 } |
fbf4d5b1b664
Remove FF_MM_SSE2/3 flags for CPUs where this is generally not faster than
rbultje
parents:
12196
diff
changeset
|
359 |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
360 if (mm_flags & AV_CPU_FLAG_SSE2) { |
12241
c7f6ddcc5c01
VP8: optimize DC-only chroma case in the same way as luma.
darkshikari
parents:
12238
diff
changeset
|
361 c->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_sse2; |
12238 | 362 |
12334
435319d67bd8
Use word-writing instead of dword-writing (with two cached but otherwise
rbultje
parents:
12241
diff
changeset
|
363 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
|
364 |
12194
80b142c2e9f7
Change function prototypes for width=8 inner and mbedge loopfilter functions
rbultje
parents:
12168
diff
changeset
|
365 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
|
366 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
|
367 |
12214
657d353cd515
Fix and enable horizontal >=SSE2 mbedge loopfilter.
rbultje
parents:
12210
diff
changeset
|
368 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
|
369 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse2; |
11975 | 370 } |
371 | |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
372 if (mm_flags & AV_CPU_FLAG_SSSE3) { |
11991 | 373 VP8_LUMA_MC_FUNC(0, 16, ssse3); |
374 VP8_MC_FUNC(1, 8, ssse3); | |
12054
b8f80fe02861
SSSE3 versions of width4 VP8 6-tap MC functions
darkshikari
parents:
12042
diff
changeset
|
375 VP8_MC_FUNC(2, 4, ssse3); |
11991 | 376 VP8_BILINEAR_MC_FUNC(0, 16, ssse3); |
377 VP8_BILINEAR_MC_FUNC(1, 8, ssse3); | |
12082
8527154f6e81
SSSE3 versions of vp8 width4 bilinear MC functions
darkshikari
parents:
12054
diff
changeset
|
378 VP8_BILINEAR_MC_FUNC(2, 4, ssse3); |
12210 | 379 |
380 c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_ssse3; | |
381 c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_ssse3; | |
382 | |
383 c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_ssse3; | |
384 c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_ssse3; | |
385 c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_ssse3; | |
386 c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_ssse3; | |
387 | |
388 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
|
389 c->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16y_mbedge_ssse3; |
12210 | 390 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
|
391 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_ssse3; |
11975 | 392 } |
393 | |
12456
a5ddb39627fd
Rename FF_MM_ symbols related to CPU features flags as AV_CPU_FLAG_
stefano
parents:
12414
diff
changeset
|
394 if (mm_flags & AV_CPU_FLAG_SSE4) { |
11975 | 395 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
|
396 |
12334
435319d67bd8
Use word-writing instead of dword-writing (with two cached but otherwise
rbultje
parents:
12241
diff
changeset
|
397 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
|
398 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
|
399 c->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mbedge_sse4; |
11975 | 400 } |
11976 | 401 #endif |
11975 | 402 } |