Mercurial > libavcodec.hg
annotate i386/dsputil_h264_template_mmx.c @ 4541:1c8bc533da57 libavcodec
factorize more calls to fill_rectangle()
author | michael |
---|---|
date | Sun, 18 Feb 2007 01:37:50 +0000 |
parents | e64290e64366 |
children | d0c94a0d43ca |
rev | line source |
---|---|
2732 | 1 /* |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
2 * Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>, |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
3 * Loren Merritt |
2732 | 4 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
2732 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
2732 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
2732 | 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 | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3218
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2922
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
2732 | 20 */ |
21 | |
22 /** | |
23 * MMX optimized version of (put|avg)_h264_chroma_mc8. | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
24 * H264_CHROMA_MC8_TMPL must be defined to the desired function name |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
25 * H264_CHROMA_OP must be defined to empty for put and pavgb/pavgusb for avg |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
26 * H264_CHROMA_MC8_MV0 must be defined to a (put|avg)_pixels8 function |
2732 | 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) | |
29 { | |
3089 | 30 DECLARE_ALIGNED_8(uint64_t, AA); |
31 DECLARE_ALIGNED_8(uint64_t, DD); | |
2732 | 32 int i; |
33 | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
34 if(y==0 && x==0) { |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
35 /* no filter needed */ |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
36 H264_CHROMA_MC8_MV0(dst, src, stride, h); |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
37 return; |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
38 } |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
39 |
2732 | 40 assert(x<8 && y<8 && x>=0 && y>=0); |
41 | |
3218 | 42 if(y==0 || x==0) |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
43 { |
3218 | 44 /* 1 dimensional filter only */ |
45 const int dxy = x ? 1 : stride; | |
46 | |
47 asm volatile( | |
48 "movd %0, %%mm5\n\t" | |
49 "movq %1, %%mm4\n\t" | |
50 "punpcklwd %%mm5, %%mm5\n\t" | |
51 "punpckldq %%mm5, %%mm5\n\t" /* mm5 = B = x */ | |
52 "movq %%mm4, %%mm6\n\t" | |
53 "pxor %%mm7, %%mm7\n\t" | |
54 "psubw %%mm5, %%mm4\n\t" /* mm4 = A = 8-x */ | |
55 "psrlw $1, %%mm6\n\t" /* mm6 = 4 */ | |
56 :: "rm"(x+y), "m"(ff_pw_8)); | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
57 |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
58 for(i=0; i<h; i++) { |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
59 asm volatile( |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
60 /* mm0 = src[0..7], mm1 = src[1..8] */ |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
61 "movq %0, %%mm0\n\t" |
3218 | 62 "movq %1, %%mm2\n\t" |
63 :: "m"(src[0]), "m"(src[dxy])); | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
64 |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
65 asm volatile( |
3218 | 66 /* [mm0,mm1] = A * src[0..7] */ |
67 /* [mm2,mm3] = B * src[1..8] */ | |
68 "movq %%mm0, %%mm1\n\t" | |
69 "movq %%mm2, %%mm3\n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
70 "punpcklbw %%mm7, %%mm0\n\t" |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
71 "punpckhbw %%mm7, %%mm1\n\t" |
3218 | 72 "punpcklbw %%mm7, %%mm2\n\t" |
73 "punpckhbw %%mm7, %%mm3\n\t" | |
74 "pmullw %%mm4, %%mm0\n\t" | |
75 "pmullw %%mm4, %%mm1\n\t" | |
76 "pmullw %%mm5, %%mm2\n\t" | |
77 "pmullw %%mm5, %%mm3\n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
78 |
3218 | 79 /* dst[0..7] = (A * src[0..7] + B * src[1..8] + 4) >> 3 */ |
80 "paddw %%mm6, %%mm0\n\t" | |
81 "paddw %%mm6, %%mm1\n\t" | |
82 "paddw %%mm2, %%mm0\n\t" | |
83 "paddw %%mm3, %%mm1\n\t" | |
84 "psrlw $3, %%mm0\n\t" | |
85 "psrlw $3, %%mm1\n\t" | |
86 "packuswb %%mm1, %%mm0\n\t" | |
87 H264_CHROMA_OP(%0, %%mm0) | |
88 "movq %%mm0, %0\n\t" | |
89 : "=m" (dst[0])); | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
90 |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
91 src += stride; |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
92 dst += stride; |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
93 } |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
94 return; |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
95 } |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
96 |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
97 /* general case, bilinear */ |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
98 asm volatile("movd %2, %%mm4\n\t" |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
99 "movd %3, %%mm6\n\t" |
2732 | 100 "punpcklwd %%mm4, %%mm4\n\t" |
101 "punpcklwd %%mm6, %%mm6\n\t" | |
102 "punpckldq %%mm4, %%mm4\n\t" /* mm4 = x words */ | |
103 "punpckldq %%mm6, %%mm6\n\t" /* mm6 = y words */ | |
104 "movq %%mm4, %%mm5\n\t" | |
105 "pmullw %%mm6, %%mm4\n\t" /* mm4 = x * y */ | |
106 "psllw $3, %%mm5\n\t" | |
107 "psllw $3, %%mm6\n\t" | |
108 "movq %%mm5, %%mm7\n\t" | |
109 "paddw %%mm6, %%mm7\n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
110 "movq %%mm4, %1\n\t" /* DD = x * y */ |
2732 | 111 "psubw %%mm4, %%mm5\n\t" /* mm5 = B = 8x - xy */ |
112 "psubw %%mm4, %%mm6\n\t" /* mm6 = C = 8y - xy */ | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
113 "paddw %4, %%mm4\n\t" |
2732 | 114 "psubw %%mm7, %%mm4\n\t" /* mm4 = A = xy - (8x+8y) + 64 */ |
115 "pxor %%mm7, %%mm7\n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
116 "movq %%mm4, %0\n\t" |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
117 : "=m" (AA), "=m" (DD) : "rm" (x), "rm" (y), "m" (ff_pw_64)); |
2732 | 118 |
119 asm volatile( | |
120 /* mm0 = src[0..7], mm1 = src[1..8] */ | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
121 "movq %0, %%mm0\n\t" |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
122 "movq %1, %%mm1\n\t" |
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
123 : : "m" (src[0]), "m" (src[1])); |
2732 | 124 |
125 for(i=0; i<h; i++) { | |
3218 | 126 src += stride; |
2732 | 127 |
3218 | 128 asm volatile( |
129 /* mm2 = A * src[0..3] + B * src[1..4] */ | |
130 /* mm3 = A * src[4..7] + B * src[5..8] */ | |
131 "movq %%mm0, %%mm2\n\t" | |
132 "movq %%mm1, %%mm3\n\t" | |
133 "punpckhbw %%mm7, %%mm0\n\t" | |
134 "punpcklbw %%mm7, %%mm1\n\t" | |
135 "punpcklbw %%mm7, %%mm2\n\t" | |
136 "punpckhbw %%mm7, %%mm3\n\t" | |
137 "pmullw %0, %%mm0\n\t" | |
138 "pmullw %0, %%mm2\n\t" | |
2732 | 139 "pmullw %%mm5, %%mm1\n\t" |
3218 | 140 "pmullw %%mm5, %%mm3\n\t" |
141 "paddw %%mm1, %%mm2\n\t" | |
142 "paddw %%mm0, %%mm3\n\t" | |
2732 | 143 : : "m" (AA)); |
144 | |
145 asm volatile( | |
3218 | 146 /* [mm2,mm3] += C * src[0..7] */ |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
147 "movq %0, %%mm0\n\t" |
3218 | 148 "movq %%mm0, %%mm1\n\t" |
149 "punpcklbw %%mm7, %%mm0\n\t" | |
150 "punpckhbw %%mm7, %%mm1\n\t" | |
151 "pmullw %%mm6, %%mm0\n\t" | |
152 "pmullw %%mm6, %%mm1\n\t" | |
153 "paddw %%mm0, %%mm2\n\t" | |
154 "paddw %%mm1, %%mm3\n\t" | |
155 : : "m" (src[0])); | |
2732 | 156 |
157 asm volatile( | |
3218 | 158 /* [mm2,mm3] += D * src[1..8] */ |
159 "movq %1, %%mm1\n\t" | |
160 "movq %%mm1, %%mm0\n\t" | |
161 "movq %%mm1, %%mm4\n\t" | |
162 "punpcklbw %%mm7, %%mm0\n\t" | |
163 "punpckhbw %%mm7, %%mm4\n\t" | |
164 "pmullw %2, %%mm0\n\t" | |
165 "pmullw %2, %%mm4\n\t" | |
166 "paddw %%mm0, %%mm2\n\t" | |
167 "paddw %%mm4, %%mm3\n\t" | |
168 "movq %0, %%mm0\n\t" | |
169 : : "m" (src[0]), "m" (src[1]), "m" (DD)); | |
2732 | 170 |
171 asm volatile( | |
3218 | 172 /* dst[0..7] = ([mm2,mm3] + 32) >> 6 */ |
2732 | 173 "paddw %1, %%mm2\n\t" |
174 "paddw %1, %%mm3\n\t" | |
175 "psrlw $6, %%mm2\n\t" | |
176 "psrlw $6, %%mm3\n\t" | |
177 "packuswb %%mm3, %%mm2\n\t" | |
178 H264_CHROMA_OP(%0, %%mm2) | |
179 "movq %%mm2, %0\n\t" | |
2754 | 180 : "=m" (dst[0]) : "m" (ff_pw_32)); |
2732 | 181 dst+= stride; |
182 } | |
183 } | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
184 |
3213 | 185 static void H264_CHROMA_MC4_TMPL(uint8_t *dst/*align 4*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y) |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
186 { |
4538 | 187 asm volatile( |
188 "pxor %%mm7, %%mm7 \n\t" | |
189 "movd %5, %%mm2 \n\t" | |
190 "movd %6, %%mm3 \n\t" | |
191 "movq %7, %%mm4 \n\t" | |
192 "movq %7, %%mm5 \n\t" | |
193 "punpcklwd %%mm2, %%mm2 \n\t" | |
194 "punpcklwd %%mm3, %%mm3 \n\t" | |
195 "punpcklwd %%mm2, %%mm2 \n\t" | |
196 "punpcklwd %%mm3, %%mm3 \n\t" | |
197 "psubw %%mm2, %%mm4 \n\t" | |
198 "psubw %%mm3, %%mm5 \n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
199 |
4538 | 200 "movd (%1), %%mm0 \n\t" |
201 "movd 1(%1), %%mm6 \n\t" | |
202 "add %3, %1 \n\t" | |
203 "punpcklbw %%mm7, %%mm0 \n\t" | |
204 "punpcklbw %%mm7, %%mm6 \n\t" | |
205 "pmullw %%mm4, %%mm0 \n\t" | |
206 "pmullw %%mm2, %%mm6 \n\t" | |
207 "paddw %%mm0, %%mm6 \n\t" | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
208 |
4538 | 209 "1: \n\t" |
210 "movd (%1), %%mm0 \n\t" | |
211 "movd 1(%1), %%mm1 \n\t" | |
212 "add %3, %1 \n\t" | |
213 "punpcklbw %%mm7, %%mm0 \n\t" | |
214 "punpcklbw %%mm7, %%mm1 \n\t" | |
215 "pmullw %%mm4, %%mm0 \n\t" | |
216 "pmullw %%mm2, %%mm1 \n\t" | |
217 "paddw %%mm0, %%mm1 \n\t" | |
218 "movq %%mm1, %%mm0 \n\t" | |
219 "pmullw %%mm5, %%mm6 \n\t" | |
220 "pmullw %%mm3, %%mm1 \n\t" | |
221 "paddw %%mm6, %%mm1 \n\t" | |
222 "paddw %4, %%mm1 \n\t" | |
223 "psrlw $6, %%mm1 \n\t" | |
224 "packuswb %%mm1, %%mm1 \n\t" | |
225 H264_CHROMA_OP4((%0), %%mm1, %%mm6) | |
226 "movd %%mm1, (%0) \n\t" | |
227 "add %3, %0 \n\t" | |
228 "movd (%1), %%mm6 \n\t" | |
229 "movd 1(%1), %%mm1 \n\t" | |
230 "add %3, %1 \n\t" | |
231 "punpcklbw %%mm7, %%mm6 \n\t" | |
232 "punpcklbw %%mm7, %%mm1 \n\t" | |
233 "pmullw %%mm4, %%mm6 \n\t" | |
234 "pmullw %%mm2, %%mm1 \n\t" | |
235 "paddw %%mm6, %%mm1 \n\t" | |
236 "movq %%mm1, %%mm6 \n\t" | |
237 "pmullw %%mm5, %%mm0 \n\t" | |
238 "pmullw %%mm3, %%mm1 \n\t" | |
239 "paddw %%mm0, %%mm1 \n\t" | |
240 "paddw %4, %%mm1 \n\t" | |
241 "psrlw $6, %%mm1 \n\t" | |
242 "packuswb %%mm1, %%mm1 \n\t" | |
243 H264_CHROMA_OP4((%0), %%mm1, %%mm0) | |
244 "movd %%mm1, (%0) \n\t" | |
245 "add %3, %0 \n\t" | |
246 "sub $2, %2 \n\t" | |
247 "jnz 1b \n\t" | |
248 : "+r"(dst), "+r"(src), "+r"(h) | |
249 : "r"(stride), "m"(ff_pw_32), "m"(x), "m"(y), "m"(ff_pw_8) | |
250 ); | |
2922
d772011258ec
faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents:
2754
diff
changeset
|
251 } |
3213 | 252 |
253 #ifdef H264_CHROMA_MC2_TMPL | |
4508 | 254 static void H264_CHROMA_MC2_TMPL(uint8_t *dst/*align 2*/, uint8_t *src/*align 1*/, long stride, int h, int x, int y) |
3213 | 255 { |
4505
cb5628800a62
factor out common subexprssion (gcc of course is too stupid to do this ...)
michael
parents:
4504
diff
changeset
|
256 int tmp = ((1<<16)-1)*x + 8; |
cb5628800a62
factor out common subexprssion (gcc of course is too stupid to do this ...)
michael
parents:
4504
diff
changeset
|
257 int CD= tmp*y; |
cb5628800a62
factor out common subexprssion (gcc of course is too stupid to do this ...)
michael
parents:
4504
diff
changeset
|
258 int AB= (tmp<<3) - CD; |
3213 | 259 asm volatile( |
260 /* mm5 = {A,B,A,B} */ | |
261 /* mm6 = {C,D,C,D} */ | |
262 "movd %0, %%mm5\n\t" | |
263 "movd %1, %%mm6\n\t" | |
264 "punpckldq %%mm5, %%mm5\n\t" | |
265 "punpckldq %%mm6, %%mm6\n\t" | |
266 "pxor %%mm7, %%mm7\n\t" | |
4504 | 267 /* mm0 = src[0,1,1,2] */ |
268 "movd %2, %%mm0\n\t" | |
269 "punpcklbw %%mm7, %%mm0\n\t" | |
270 "pshufw $0x94, %%mm0, %%mm0\n\t" | |
271 :: "r"(AB), "r"(CD), "m"(src[0])); | |
272 | |
3213 | 273 |
274 asm volatile( | |
4504 | 275 "1:\n\t" |
4509 | 276 "add %4, %1\n\t" |
4504 | 277 /* mm1 = A * src[0,1] + B * src[1,2] */ |
278 "movq %%mm0, %%mm1\n\t" | |
279 "pmaddwd %%mm5, %%mm1\n\t" | |
3213 | 280 /* mm0 = src[0,1,1,2] */ |
4504 | 281 "movd (%1), %%mm0\n\t" |
3213 | 282 "punpcklbw %%mm7, %%mm0\n\t" |
283 "pshufw $0x94, %%mm0, %%mm0\n\t" | |
4504 | 284 /* mm1 += C * src[0,1] + D * src[1,2] */ |
285 "movq %%mm0, %%mm2\n\t" | |
286 "pmaddwd %%mm6, %%mm2\n\t" | |
287 "paddw %%mm2, %%mm1\n\t" | |
288 /* dst[0,1] = pack((mm1 + 32) >> 6) */ | |
289 "paddw %3, %%mm1\n\t" | |
290 "psrlw $6, %%mm1\n\t" | |
291 "packssdw %%mm7, %%mm1\n\t" | |
292 "packuswb %%mm7, %%mm1\n\t" | |
293 /* writes garbage to the right of dst. | |
294 * ok because partitions are processed from left to right. */ | |
295 H264_CHROMA_OP4((%0), %%mm1, %%mm3) | |
296 "movd %%mm1, (%0)\n\t" | |
4508 | 297 "add %4, %0\n\t" |
298 "sub $1, %2\n\t" | |
4504 | 299 "jnz 1b\n\t" |
300 : "+r" (dst), "+r"(src), "+r"(h) : "m" (ff_pw_32), "r"(stride)); | |
3213 | 301 |
302 } | |
303 #endif | |
304 |