Mercurial > libavcodec.hg
comparison x86/dsputil_h264_template_mmx.c @ 8430:7768bdfd4f7b libavcodec
Rename libavcodec/i386/ --> libavcodec/x86/.
It contains optimizations that are not specific to i386 and
libavutil uses this naming scheme already.
author | diego |
---|---|
date | Mon, 22 Dec 2008 09:12:42 +0000 |
parents | |
children | cc64e1343397 |
comparison
equal
deleted
inserted
replaced
8429:b3ecaba81501 | 8430:7768bdfd4f7b |
---|---|
1 /* | |
2 * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>, | |
3 * Loren Merritt | |
4 * | |
5 * This file is part of FFmpeg. | |
6 * | |
7 * FFmpeg is free software; you can redistribute it and/or | |
8 * modify it under the terms of the GNU Lesser General Public | |
9 * License as published by the Free Software Foundation; either | |
10 * version 2.1 of the License, or (at your option) any later version. | |
11 * | |
12 * FFmpeg is distributed in the hope that it will be useful, | |
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
18 * License along with FFmpeg; if not, write to the Free Software | |
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 */ | |
21 | |
22 /** | |
23 * MMX optimized version of (put|avg)_h264_chroma_mc8. | |
24 * H264_CHROMA_MC8_TMPL must be defined to the desired function name | |
25 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg | |
26 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function | |
27 */ | |
28 static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y, int rnd) | |
29 { | |
30 const uint64_t *rnd_reg; | |
31 DECLARE_ALIGNED_8(uint64_t, AA); | |
32 DECLARE_ALIGNED_8(uint64_t, DD); | |
33 int i; | |
34 | |
35 if(y==0 && x==0) { | |
36 /* no filter needed */ | |
37 H264_CHROMA_MC8_MV0(dst, src, stride, h); | |
38 return; | |
39 } | |
40 | |
41 assert(x<8 && y<8 && x>=0 && y>=0); | |
42 | |
43 if(y==0 || x==0) | |
44 { | |
45 /* 1 dimensional filter only */ | |
46 const int dxy = x ? 1 : stride; | |
47 | |
48 rnd_reg = rnd ? &ff_pw_4 : &ff_pw_3; | |
49 | |
50 __asm__ volatile( | |
51 "movd %0, %%mm5\n\t" | |
52 "movq %1, %%mm4\n\t" | |
53 "movq %2, %%mm6\n\t" /* mm6 = rnd */ | |
54 "punpcklwd %%mm5, %%mm5\n\t" | |
55 "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */ | |
56 "pxor %%mm7, %%mm7\n\t" | |
57 "psubw %%mm5, %%mm4\n\t" /* mm4 = A = 8-x */ | |
58 :: "rm"(x+y), "m"(ff_pw_8), "m"(*rnd_reg)); | |
59 | |
60 for(i=0; i<h; i++) { | |
61 __asm__ volatile( | |
62 /* mm0 = src[0..7], mm1 = src[1..8] */ | |
63 "movq %0, %%mm0\n\t" | |
64 "movq %1, %%mm2\n\t" | |
65 :: "m"(src[0]), "m"(src[dxy])); | |
66 | |
67 __asm__ volatile( | |
68 /* [mm0,mm1] = A * src[0..7] */ | |
69 /* [mm2,mm3] = B * src[1..8] */ | |
70 "movq %%mm0, %%mm1\n\t" | |
71 "movq %%mm2, %%mm3\n\t" | |
72 "punpcklbw %%mm7, %%mm0\n\t" | |
73 "punpckhbw %%mm7, %%mm1\n\t" | |
74 "punpcklbw %%mm7, %%mm2\n\t" | |
75 "punpckhbw %%mm7, %%mm3\n\t" | |
76 "pmullw %%mm4, %%mm0\n\t" | |
77 "pmullw %%mm4, %%mm1\n\t" | |
78 "pmullw %%mm5, %%mm2\n\t" | |
79 "pmullw %%mm5, %%mm3\n\t" | |
80 | |
81 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + 4) >> 3 */ | |
82 "paddw %%mm6, %%mm0\n\t" | |
83 "paddw %%mm6, %%mm1\n\t" | |
84 "paddw %%mm2, %%mm0\n\t" | |
85 "paddw %%mm3, %%mm1\n\t" | |
86 "psrlw $3, %%mm0\n\t" | |
87 "psrlw $3, %%mm1\n\t" | |
88 "packuswb %%mm1, %%mm0\n\t" | |
89 H264_CHROMA_OP(%0, %%mm0) | |
90 "movq %%mm0, %0\n\t" | |
91 : "=m" (dst[0])); | |
92 | |
93 src += stride; | |
94 dst += stride; | |
95 } | |
96 return; | |
97 } | |
98 | |
99 /* general case, bilinear */ | |
100 rnd_reg = rnd ? &ff_pw_32.a : &ff_pw_28.a; | |
101 __asm__ volatile("movd %2, %%mm4\n\t" | |
102 "movd %3, %%mm6\n\t" | |
103 "punpcklwd %%mm4, %%mm4\n\t" | |
104 "punpcklwd %%mm6, %%mm6\n\t" | |
105 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */ | |
106 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */ | |
107 "movq %%mm4, %%mm5\n\t" | |
108 "pmullw %%mm6, %%mm4\n\t" /* mm4 = x * y */ | |
109 "psllw $3, %%mm5\n\t" | |
110 "psllw $3, %%mm6\n\t" | |
111 "movq %%mm5, %%mm7\n\t" | |
112 "paddw %%mm6, %%mm7\n\t" | |
113 "movq %%mm4, %1\n\t" /* DD = x * y */ | |
114 "psubw %%mm4, %%mm5\n\t" /* mm5 = B = 8x - xy */ | |
115 "psubw %%mm4, %%mm6\n\t" /* mm6 = C = 8y - xy */ | |
116 "paddw %4, %%mm4\n\t" | |
117 "psubw %%mm7, %%mm4\n\t" /* mm4 = A = xy - (8x+8y) + 64 */ | |
118 "pxor %%mm7, %%mm7\n\t" | |
119 "movq %%mm4, %0\n\t" | |
120 : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64)); | |
121 | |
122 __asm__ volatile( | |
123 /* mm0 = src[0..7], mm1 = src[1..8] */ | |
124 "movq %0, %%mm0\n\t" | |
125 "movq %1, %%mm1\n\t" | |
126 : : "m" (src[0]), "m" (src[1])); | |
127 | |
128 for(i=0; i<h; i++) { | |
129 src += stride; | |
130 | |
131 __asm__ volatile( | |
132 /* mm2 = A * src[0..3] + B * src[1..4] */ | |
133 /* mm3 = A * src[4..7] + B * src[5..8] */ | |
134 "movq %%mm0, %%mm2\n\t" | |
135 "movq %%mm1, %%mm3\n\t" | |
136 "punpckhbw %%mm7, %%mm0\n\t" | |
137 "punpcklbw %%mm7, %%mm1\n\t" | |
138 "punpcklbw %%mm7, %%mm2\n\t" | |
139 "punpckhbw %%mm7, %%mm3\n\t" | |
140 "pmullw %0, %%mm0\n\t" | |
141 "pmullw %0, %%mm2\n\t" | |
142 "pmullw %%mm5, %%mm1\n\t" | |
143 "pmullw %%mm5, %%mm3\n\t" | |
144 "paddw %%mm1, %%mm2\n\t" | |
145 "paddw %%mm0, %%mm3\n\t" | |
146 : : "m" (AA)); | |
147 | |
148 __asm__ volatile( | |
149 /* [mm2,mm3] += C * src[0..7] */ | |
150 "movq %0, %%mm0\n\t" | |
151 "movq %%mm0, %%mm1\n\t" | |
152 "punpcklbw %%mm7, %%mm0\n\t" | |
153 "punpckhbw %%mm7, %%mm1\n\t" | |
154 "pmullw %%mm6, %%mm0\n\t" | |
155 "pmullw %%mm6, %%mm1\n\t" | |
156 "paddw %%mm0, %%mm2\n\t" | |
157 "paddw %%mm1, %%mm3\n\t" | |
158 : : "m" (src[0])); | |
159 | |
160 __asm__ volatile( | |
161 /* [mm2,mm3] += D * src[1..8] */ | |
162 "movq %1, %%mm1\n\t" | |
163 "movq %%mm1, %%mm0\n\t" | |
164 "movq %%mm1, %%mm4\n\t" | |
165 "punpcklbw %%mm7, %%mm0\n\t" | |
166 "punpckhbw %%mm7, %%mm4\n\t" | |
167 "pmullw %2, %%mm0\n\t" | |
168 "pmullw %2, %%mm4\n\t" | |
169 "paddw %%mm0, %%mm2\n\t" | |
170 "paddw %%mm4, %%mm3\n\t" | |
171 "movq %0, %%mm0\n\t" | |
172 : : "m" (src[0]), "m" (src[1]), "m" (DD)); | |
173 | |
174 __asm__ volatile( | |
175 /* dst[0..7] = ([mm2,mm3] + 32) >> 6 */ | |
176 "paddw %1, %%mm2\n\t" | |
177 "paddw %1, %%mm3\n\t" | |
178 "psrlw $6, %%mm2\n\t" | |
179 "psrlw $6, %%mm3\n\t" | |
180 "packuswb %%mm3, %%mm2\n\t" | |
181 H264_CHROMA_OP(%0, %%mm2) | |
182 "movq %%mm2, %0\n\t" | |
183 : "=m" (dst[0]) : "m" (*rnd_reg)); | |
184 dst+= stride; | |
185 } | |
186 } | |
187 | |
188 static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) | |
189 { | |
190 __asm__ volatile( | |
191 "pxor %%mm7, %%mm7 \n\t" | |
192 "movd %5, %%mm2 \n\t" | |
193 "movd %6, %%mm3 \n\t" | |
194 "movq "MANGLE(ff_pw_8)", %%mm4\n\t" | |
195 "movq "MANGLE(ff_pw_8)", %%mm5\n\t" | |
196 "punpcklwd %%mm2, %%mm2 \n\t" | |
197 "punpcklwd %%mm3, %%mm3 \n\t" | |
198 "punpcklwd %%mm2, %%mm2 \n\t" | |
199 "punpcklwd %%mm3, %%mm3 \n\t" | |
200 "psubw %%mm2, %%mm4 \n\t" | |
201 "psubw %%mm3, %%mm5 \n\t" | |
202 | |
203 "movd (%1), %%mm0 \n\t" | |
204 "movd 1(%1), %%mm6 \n\t" | |
205 "add %3, %1 \n\t" | |
206 "punpcklbw %%mm7, %%mm0 \n\t" | |
207 "punpcklbw %%mm7, %%mm6 \n\t" | |
208 "pmullw %%mm4, %%mm0 \n\t" | |
209 "pmullw %%mm2, %%mm6 \n\t" | |
210 "paddw %%mm0, %%mm6 \n\t" | |
211 | |
212 "1: \n\t" | |
213 "movd (%1), %%mm0 \n\t" | |
214 "movd 1(%1), %%mm1 \n\t" | |
215 "add %3, %1 \n\t" | |
216 "punpcklbw %%mm7, %%mm0 \n\t" | |
217 "punpcklbw %%mm7, %%mm1 \n\t" | |
218 "pmullw %%mm4, %%mm0 \n\t" | |
219 "pmullw %%mm2, %%mm1 \n\t" | |
220 "paddw %%mm0, %%mm1 \n\t" | |
221 "movq %%mm1, %%mm0 \n\t" | |
222 "pmullw %%mm5, %%mm6 \n\t" | |
223 "pmullw %%mm3, %%mm1 \n\t" | |
224 "paddw %4, %%mm6 \n\t" | |
225 "paddw %%mm6, %%mm1 \n\t" | |
226 "psrlw $6, %%mm1 \n\t" | |
227 "packuswb %%mm1, %%mm1 \n\t" | |
228 H264_CHROMA_OP4((%0), %%mm1, %%mm6) | |
229 "movd %%mm1, (%0) \n\t" | |
230 "add %3, %0 \n\t" | |
231 "movd (%1), %%mm6 \n\t" | |
232 "movd 1(%1), %%mm1 \n\t" | |
233 "add %3, %1 \n\t" | |
234 "punpcklbw %%mm7, %%mm6 \n\t" | |
235 "punpcklbw %%mm7, %%mm1 \n\t" | |
236 "pmullw %%mm4, %%mm6 \n\t" | |
237 "pmullw %%mm2, %%mm1 \n\t" | |
238 "paddw %%mm6, %%mm1 \n\t" | |
239 "movq %%mm1, %%mm6 \n\t" | |
240 "pmullw %%mm5, %%mm0 \n\t" | |
241 "pmullw %%mm3, %%mm1 \n\t" | |
242 "paddw %4, %%mm0 \n\t" | |
243 "paddw %%mm0, %%mm1 \n\t" | |
244 "psrlw $6, %%mm1 \n\t" | |
245 "packuswb %%mm1, %%mm1 \n\t" | |
246 H264_CHROMA_OP4((%0), %%mm1, %%mm0) | |
247 "movd %%mm1, (%0) \n\t" | |
248 "add %3, %0 \n\t" | |
249 "sub $2, %2 \n\t" | |
250 "jnz 1b \n\t" | |
251 : "+r"(dst), "+r"(src), "+r"(h) | |
252 : "r"((x86_reg)stride), "m"(ff_pw_32), "m"(x), "m"(y) | |
253 ); | |
254 } | |
255 | |
256 #ifdef H264_CHROMA_MC2_TMPL | |
257 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) | |
258 { | |
259 int tmp = ((1<<16)-1)*x + 8; | |
260 int CD= tmp*y; | |
261 int AB= (tmp<<3) - CD; | |
262 __asm__ volatile( | |
263 /* mm5 = {A,B,A,B} */ | |
264 /* mm6 = {C,D,C,D} */ | |
265 "movd %0, %%mm5\n\t" | |
266 "movd %1, %%mm6\n\t" | |
267 "punpckldq %%mm5, %%mm5\n\t" | |
268 "punpckldq %%mm6, %%mm6\n\t" | |
269 "pxor %%mm7, %%mm7\n\t" | |
270 /* mm0 = src[0,1,1,2] */ | |
271 "movd %2, %%mm2\n\t" | |
272 "punpcklbw %%mm7, %%mm2\n\t" | |
273 "pshufw $0x94, %%mm2, %%mm2\n\t" | |
274 :: "r"(AB), "r"(CD), "m"(src[0])); | |
275 | |
276 | |
277 __asm__ volatile( | |
278 "1:\n\t" | |
279 "add %4, %1\n\t" | |
280 /* mm1 = A * src[0,1] + B * src[1,2] */ | |
281 "movq %%mm2, %%mm1\n\t" | |
282 "pmaddwd %%mm5, %%mm1\n\t" | |
283 /* mm0 = src[0,1,1,2] */ | |
284 "movd (%1), %%mm0\n\t" | |
285 "punpcklbw %%mm7, %%mm0\n\t" | |
286 "pshufw $0x94, %%mm0, %%mm0\n\t" | |
287 /* mm1 += C * src[0,1] + D * src[1,2] */ | |
288 "movq %%mm0, %%mm2\n\t" | |
289 "pmaddwd %%mm6, %%mm0\n\t" | |
290 "paddw %3, %%mm1\n\t" | |
291 "paddw %%mm0, %%mm1\n\t" | |
292 /* dst[0,1] = pack((mm1 + 32) >> 6) */ | |
293 "psrlw $6, %%mm1\n\t" | |
294 "packssdw %%mm7, %%mm1\n\t" | |
295 "packuswb %%mm7, %%mm1\n\t" | |
296 H264_CHROMA_OP4((%0), %%mm1, %%mm3) | |
297 "movd %%mm1, %%esi\n\t" | |
298 "movw %%si, (%0)\n\t" | |
299 "add %4, %0\n\t" | |
300 "sub $1, %2\n\t" | |
301 "jnz 1b\n\t" | |
302 : "+r" (dst), "+r"(src), "+r"(h) | |
303 : "m" (ff_pw_32), "r"((x86_reg)stride) | |
304 : "%esi"); | |
305 | |
306 } | |
307 #endif | |
308 |