VirtualBox

source: vbox/trunk/src/libs/ffmpeg-20060710/libavcodec/i386/dsputil_h264_template_mmx.c@ 5776

Last change on this file since 5776 was 5776, checked in by vboxsync, 17 years ago

ffmpeg: exported to OSE
File size: 11.1 KB
Line 
1/*
2 * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>,
3 * Loren Merritt
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20/**
21 * MMX optimized version of (put|avg)_h264_chroma_mc8.
22 * H264_CHROMA_MC8_TMPL must be defined to the desired function name
23 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg
24 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function
25 */
26static void H264_CHROMA_MC8_TMPL(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
27{
28 DECLARE_ALIGNED_8(uint64_t, AA);
29 DECLARE_ALIGNED_8(uint64_t, DD);
30 int i;
31
32 if(y==0 && x==0) {
33 /* no filter needed */
34 H264_CHROMA_MC8_MV0(dst, src, stride, h);
35 return;
36 }
37
38 assert(x<8 && y<8 && x>=0 && y>=0);
39
40 if(y==0 || x==0)
41 {
42 /* 1 dimensional filter only */
43 const int dxy = x ? 1 : stride;
44
45 asm volatile(
46 "movd %0, %%mm5\n\t"
47 "movq %1, %%mm4\n\t"
48 "punpcklwd %%mm5, %%mm5\n\t"
49 "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */
50 "movq %%mm4, %%mm6\n\t"
51 "pxor %%mm7, %%mm7\n\t"
52 "psubw %%mm5, %%mm4\n\t" /* mm4 = A = 8-x */
53 "psrlw $1, %%mm6\n\t" /* mm6 = 4 */
54 :: "rm"(x+y), "m"(ff_pw_8));
55
56 for(i=0; i<h; i++) {
57 asm volatile(
58 /* mm0 = src[0..7], mm1 = src[1..8] */
59 "movq %0, %%mm0\n\t"
60 "movq %1, %%mm2\n\t"
61 :: "m"(src[0]), "m"(src[dxy]));
62
63 asm volatile(
64 /* [mm0,mm1] = A * src[0..7] */
65 /* [mm2,mm3] = B * src[1..8] */
66 "movq %%mm0, %%mm1\n\t"
67 "movq %%mm2, %%mm3\n\t"
68 "punpcklbw %%mm7, %%mm0\n\t"
69 "punpckhbw %%mm7, %%mm1\n\t"
70 "punpcklbw %%mm7, %%mm2\n\t"
71 "punpckhbw %%mm7, %%mm3\n\t"
72 "pmullw %%mm4, %%mm0\n\t"
73 "pmullw %%mm4, %%mm1\n\t"
74 "pmullw %%mm5, %%mm2\n\t"
75 "pmullw %%mm5, %%mm3\n\t"
76
77 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + 4) >> 3 */
78 "paddw %%mm6, %%mm0\n\t"
79 "paddw %%mm6, %%mm1\n\t"
80 "paddw %%mm2, %%mm0\n\t"
81 "paddw %%mm3, %%mm1\n\t"
82 "psrlw $3, %%mm0\n\t"
83 "psrlw $3, %%mm1\n\t"
84 "packuswb %%mm1, %%mm0\n\t"
85 H264_CHROMA_OP(%0, %%mm0)
86 "movq %%mm0, %0\n\t"
87 : "=m" (dst[0]));
88
89 src += stride;
90 dst += stride;
91 }
92 return;
93 }
94
95 /* general case, bilinear */
96 asm volatile("movd %2, %%mm4\n\t"
97 "movd %3, %%mm6\n\t"
98 "punpcklwd %%mm4, %%mm4\n\t"
99 "punpcklwd %%mm6, %%mm6\n\t"
100 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
101 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
102 "movq %%mm4, %%mm5\n\t"
103 "pmullw %%mm6, %%mm4\n\t" /* mm4 = x * y */
104 "psllw $3, %%mm5\n\t"
105 "psllw $3, %%mm6\n\t"
106 "movq %%mm5, %%mm7\n\t"
107 "paddw %%mm6, %%mm7\n\t"
108 "movq %%mm4, %1\n\t" /* DD = x * y */
109 "psubw %%mm4, %%mm5\n\t" /* mm5 = B = 8x - xy */
110 "psubw %%mm4, %%mm6\n\t" /* mm6 = C = 8y - xy */
111 "paddw %4, %%mm4\n\t"
112 "psubw %%mm7, %%mm4\n\t" /* mm4 = A = xy - (8x+8y) + 64 */
113 "pxor %%mm7, %%mm7\n\t"
114 "movq %%mm4, %0\n\t"
115 : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64));
116
117 asm volatile(
118 /* mm0 = src[0..7], mm1 = src[1..8] */
119 "movq %0, %%mm0\n\t"
120 "movq %1, %%mm1\n\t"
121 : : "m" (src[0]), "m" (src[1]));
122
123 for(i=0; i<h; i++) {
124 src += stride;
125
126 asm volatile(
127 /* mm2 = A * src[0..3] + B * src[1..4] */
128 /* mm3 = A * src[4..7] + B * src[5..8] */
129 "movq %%mm0, %%mm2\n\t"
130 "movq %%mm1, %%mm3\n\t"
131 "punpckhbw %%mm7, %%mm0\n\t"
132 "punpcklbw %%mm7, %%mm1\n\t"
133 "punpcklbw %%mm7, %%mm2\n\t"
134 "punpckhbw %%mm7, %%mm3\n\t"
135 "pmullw %0, %%mm0\n\t"
136 "pmullw %0, %%mm2\n\t"
137 "pmullw %%mm5, %%mm1\n\t"
138 "pmullw %%mm5, %%mm3\n\t"
139 "paddw %%mm1, %%mm2\n\t"
140 "paddw %%mm0, %%mm3\n\t"
141 : : "m" (AA));
142
143 asm volatile(
144 /* [mm2,mm3] += C * src[0..7] */
145 "movq %0, %%mm0\n\t"
146 "movq %%mm0, %%mm1\n\t"
147 "punpcklbw %%mm7, %%mm0\n\t"
148 "punpckhbw %%mm7, %%mm1\n\t"
149 "pmullw %%mm6, %%mm0\n\t"
150 "pmullw %%mm6, %%mm1\n\t"
151 "paddw %%mm0, %%mm2\n\t"
152 "paddw %%mm1, %%mm3\n\t"
153 : : "m" (src[0]));
154
155 asm volatile(
156 /* [mm2,mm3] += D * src[1..8] */
157 "movq %1, %%mm1\n\t"
158 "movq %%mm1, %%mm0\n\t"
159 "movq %%mm1, %%mm4\n\t"
160 "punpcklbw %%mm7, %%mm0\n\t"
161 "punpckhbw %%mm7, %%mm4\n\t"
162 "pmullw %2, %%mm0\n\t"
163 "pmullw %2, %%mm4\n\t"
164 "paddw %%mm0, %%mm2\n\t"
165 "paddw %%mm4, %%mm3\n\t"
166 "movq %0, %%mm0\n\t"
167 : : "m" (src[0]), "m" (src[1]), "m" (DD));
168
169 asm volatile(
170 /* dst[0..7] = ([mm2,mm3] + 32) >> 6 */
171 "paddw %1, %%mm2\n\t"
172 "paddw %1, %%mm3\n\t"
173 "psrlw $6, %%mm2\n\t"
174 "psrlw $6, %%mm3\n\t"
175 "packuswb %%mm3, %%mm2\n\t"
176 H264_CHROMA_OP(%0, %%mm2)
177 "movq %%mm2, %0\n\t"
178 : "=m" (dst[0]) : "m" (ff_pw_32));
179 dst+= stride;
180 }
181}
182
183static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
184{
185 DECLARE_ALIGNED_8(uint64_t, AA);
186 DECLARE_ALIGNED_8(uint64_t, DD);
187 int i;
188
189 /* no special case for mv=(0,0) in 4x*, since it's much less common than in 8x*.
190 * could still save a few cycles, but maybe not worth the complexity. */
191
192 assert(x<8 && y<8 && x>=0 && y>=0);
193
194 asm volatile("movd %2, %%mm4\n\t"
195 "movd %3, %%mm6\n\t"
196 "punpcklwd %%mm4, %%mm4\n\t"
197 "punpcklwd %%mm6, %%mm6\n\t"
198 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */
199 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */
200 "movq %%mm4, %%mm5\n\t"
201 "pmullw %%mm6, %%mm4\n\t" /* mm4 = x * y */
202 "psllw $3, %%mm5\n\t"
203 "psllw $3, %%mm6\n\t"
204 "movq %%mm5, %%mm7\n\t"
205 "paddw %%mm6, %%mm7\n\t"
206 "movq %%mm4, %1\n\t" /* DD = x * y */
207 "psubw %%mm4, %%mm5\n\t" /* mm5 = B = 8x - xy */
208 "psubw %%mm4, %%mm6\n\t" /* mm6 = C = 8y - xy */
209 "paddw %4, %%mm4\n\t"
210 "psubw %%mm7, %%mm4\n\t" /* mm4 = A = xy - (8x+8y) + 64 */
211 "pxor %%mm7, %%mm7\n\t"
212 "movq %%mm4, %0\n\t"
213 : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64));
214
215 asm volatile(
216 /* mm0 = src[0..3], mm1 = src[1..4] */
217 "movd %0, %%mm0\n\t"
218 "movd %1, %%mm1\n\t"
219 "punpcklbw %%mm7, %%mm0\n\t"
220 "punpcklbw %%mm7, %%mm1\n\t"
221 : : "m" (src[0]), "m" (src[1]));
222
223 for(i=0; i<h; i++) {
224 asm volatile(
225 /* mm2 = A * src[0..3] + B * src[1..4] */
226 "movq %%mm0, %%mm2\n\t"
227 "pmullw %0, %%mm2\n\t"
228 "pmullw %%mm5, %%mm1\n\t"
229 "paddw %%mm1, %%mm2\n\t"
230 : : "m" (AA));
231
232 src += stride;
233 asm volatile(
234 /* mm0 = src[0..3], mm1 = src[1..4] */
235 "movd %0, %%mm0\n\t"
236 "movd %1, %%mm1\n\t"
237 "punpcklbw %%mm7, %%mm0\n\t"
238 "punpcklbw %%mm7, %%mm1\n\t"
239 : : "m" (src[0]), "m" (src[1]));
240
241 asm volatile(
242 /* mm2 += C * src[0..3] + D * src[1..4] */
243 "movq %%mm0, %%mm3\n\t"
244 "movq %%mm1, %%mm4\n\t"
245 "pmullw %%mm6, %%mm3\n\t"
246 "pmullw %0, %%mm4\n\t"
247 "paddw %%mm3, %%mm2\n\t"
248 "paddw %%mm4, %%mm2\n\t"
249 : : "m" (DD));
250
251 asm volatile(
252 /* dst[0..3] = pack((mm2 + 32) >> 6) */
253 "paddw %1, %%mm2\n\t"
254 "psrlw $6, %%mm2\n\t"
255 "packuswb %%mm7, %%mm2\n\t"
256 H264_CHROMA_OP4(%0, %%mm2, %%mm3)
257 "movd %%mm2, %0\n\t"
258 : "=m" (dst[0]) : "m" (ff_pw_32));
259 dst += stride;
260 }
261}
262
263#ifdef H264_CHROMA_MC2_TMPL
264static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
265{
266 int CD=((1<<16)-1)*x*y + 8*y;
267 int AB=((8<<16)-8)*x + 64 - CD;
268 int i;
269
270 asm volatile(
271 /* mm5 = {A,B,A,B} */
272 /* mm6 = {C,D,C,D} */
273 "movd %0, %%mm5\n\t"
274 "movd %1, %%mm6\n\t"
275 "punpckldq %%mm5, %%mm5\n\t"
276 "punpckldq %%mm6, %%mm6\n\t"
277 "pxor %%mm7, %%mm7\n\t"
278 :: "r"(AB), "r"(CD));
279
280 asm volatile(
281 /* mm0 = src[0,1,1,2] */
282 "movd %0, %%mm0\n\t"
283 "punpcklbw %%mm7, %%mm0\n\t"
284 "pshufw $0x94, %%mm0, %%mm0\n\t"
285 :: "m"(src[0]));
286
287 for(i=0; i<h; i++) {
288 asm volatile(
289 /* mm1 = A * src[0,1] + B * src[1,2] */
290 "movq %%mm0, %%mm1\n\t"
291 "pmaddwd %%mm5, %%mm1\n\t"
292 ::);
293
294 src += stride;
295 asm volatile(
296 /* mm0 = src[0,1,1,2] */
297 "movd %0, %%mm0\n\t"
298 "punpcklbw %%mm7, %%mm0\n\t"
299 "pshufw $0x94, %%mm0, %%mm0\n\t"
300 :: "m"(src[0]));
301
302 asm volatile(
303 /* mm1 += C * src[0,1] + D * src[1,2] */
304 "movq %%mm0, %%mm2\n\t"
305 "pmaddwd %%mm6, %%mm2\n\t"
306 "paddw %%mm2, %%mm1\n\t"
307 ::);
308
309 asm volatile(
310 /* dst[0,1] = pack((mm1 + 32) >> 6) */
311 "paddw %1, %%mm1\n\t"
312 "psrlw $6, %%mm1\n\t"
313 "packssdw %%mm7, %%mm1\n\t"
314 "packuswb %%mm7, %%mm1\n\t"
315 /* writes garbage to the right of dst.
316 * ok because partitions are processed from left to right. */
317 H264_CHROMA_OP4(%0, %%mm1, %%mm3)
318 "movd %%mm1, %0\n\t"
319 : "=m" (dst[0]) : "m" (ff_pw_32));
320 dst += stride;
321 }
322}
323#endif
324
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette