ffmpeg / libavcodec / i386 / cavsdsp_mmx.c @ b550bfaa
History | View | Annotate | Download (20.7 KB)
| 1 |
/*
|
|---|---|
| 2 |
* Chinese AVS video (AVS1-P2, JiZhun profile) decoder.
|
| 3 |
* Copyright (c) 2006 Stefan Gehrer <stefan.gehrer@gmx.de>
|
| 4 |
*
|
| 5 |
* MMX optimised DSP functions, based on H.264 optimisations by
|
| 6 |
* Michael Niedermayer and Loren Merritt
|
| 7 |
*
|
| 8 |
* This file is part of FFmpeg.
|
| 9 |
*
|
| 10 |
* FFmpeg is free software; you can redistribute it and/or
|
| 11 |
* modify it under the terms of the GNU Lesser General Public
|
| 12 |
* License as published by the Free Software Foundation; either
|
| 13 |
* version 2.1 of the License, or (at your option) any later version.
|
| 14 |
*
|
| 15 |
* FFmpeg is distributed in the hope that it will be useful,
|
| 16 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 17 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 18 |
* Lesser General Public License for more details.
|
| 19 |
*
|
| 20 |
* You should have received a copy of the GNU Lesser General Public
|
| 21 |
* License along with FFmpeg; if not, write to the Free Software
|
| 22 |
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
| 23 |
*/
|
| 24 |
|
| 25 |
#include "dsputil.h" |
| 26 |
#include "common.h" |
| 27 |
|
| 28 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_4 ) = 0x0004000400040004ULL; |
| 29 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_5 ) = 0x0005000500050005ULL; |
| 30 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_7 ) = 0x0007000700070007ULL; |
| 31 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_42) = 0x002A002A002A002AULL; |
| 32 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_64) = 0x0040004000400040ULL; |
| 33 |
DECLARE_ALIGNED_8(static const uint64_t,ff_pw_96) = 0x0060006000600060ULL; |
| 34 |
|
| 35 |
/*****************************************************************************
|
| 36 |
*
|
| 37 |
* inverse transform
|
| 38 |
*
|
| 39 |
****************************************************************************/
|
| 40 |
|
| 41 |
#define SUMSUB_BA( a, b ) \
|
| 42 |
"paddw "#b", "#a" \n\t"\ |
| 43 |
"paddw "#b", "#b" \n\t"\ |
| 44 |
"psubw "#a", "#b" \n\t" |
| 45 |
|
| 46 |
#define SBUTTERFLY(a,b,t,n)\
|
| 47 |
"movq " #a ", " #t " \n\t" /* abcd */\ |
| 48 |
"punpckl" #n " " #b ", " #a " \n\t" /* aebf */\ |
| 49 |
"punpckh" #n " " #b ", " #t " \n\t" /* cgdh */ |
| 50 |
|
| 51 |
#define TRANSPOSE4(a,b,c,d,t)\
|
| 52 |
SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
|
| 53 |
SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
|
| 54 |
SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
|
| 55 |
SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
|
| 56 |
|
| 57 |
static inline void cavs_idct8_1d(int16_t *block, uint64_t bias) |
| 58 |
{
|
| 59 |
asm volatile( |
| 60 |
"movq 112(%0), %%mm4 \n\t" /* mm4 = src7 */ |
| 61 |
"movq 16(%0), %%mm5 \n\t" /* mm5 = src1 */ |
| 62 |
"movq 80(%0), %%mm2 \n\t" /* mm2 = src5 */ |
| 63 |
"movq 48(%0), %%mm7 \n\t" /* mm7 = src3 */ |
| 64 |
"movq %%mm4, %%mm0 \n\t"
|
| 65 |
"movq %%mm5, %%mm3 \n\t"
|
| 66 |
"movq %%mm2, %%mm6 \n\t"
|
| 67 |
"movq %%mm7, %%mm1 \n\t"
|
| 68 |
|
| 69 |
"paddw %%mm4, %%mm4 \n\t" /* mm4 = 2*src7 */ |
| 70 |
"paddw %%mm3, %%mm3 \n\t" /* mm3 = 2*src1 */ |
| 71 |
"paddw %%mm6, %%mm6 \n\t" /* mm6 = 2*src5 */ |
| 72 |
"paddw %%mm1, %%mm1 \n\t" /* mm1 = 2*src3 */ |
| 73 |
"paddw %%mm4, %%mm0 \n\t" /* mm0 = 3*src7 */ |
| 74 |
"paddw %%mm3, %%mm5 \n\t" /* mm5 = 3*src1 */ |
| 75 |
"paddw %%mm6, %%mm2 \n\t" /* mm2 = 3*src5 */ |
| 76 |
"paddw %%mm1, %%mm7 \n\t" /* mm7 = 3*src3 */ |
| 77 |
"psubw %%mm4, %%mm5 \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */ |
| 78 |
"paddw %%mm6, %%mm7 \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */ |
| 79 |
"psubw %%mm2, %%mm1 \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */ |
| 80 |
"paddw %%mm0, %%mm3 \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */ |
| 81 |
|
| 82 |
"movq %%mm5, %%mm4 \n\t"
|
| 83 |
"movq %%mm7, %%mm6 \n\t"
|
| 84 |
"movq %%mm3, %%mm0 \n\t"
|
| 85 |
"movq %%mm1, %%mm2 \n\t"
|
| 86 |
SUMSUB_BA( %%mm7, %%mm5 ) /* mm7 = a0 + a1 mm5 = a0 - a1 */
|
| 87 |
"paddw %%mm3, %%mm7 \n\t" /* mm7 = a0 + a1 + a3 */ |
| 88 |
"paddw %%mm1, %%mm5 \n\t" /* mm5 = a0 - a1 + a2 */ |
| 89 |
"paddw %%mm7, %%mm7 \n\t"
|
| 90 |
"paddw %%mm5, %%mm5 \n\t"
|
| 91 |
"paddw %%mm6, %%mm7 \n\t" /* mm7 = b4 */ |
| 92 |
"paddw %%mm4, %%mm5 \n\t" /* mm5 = b5 */ |
| 93 |
|
| 94 |
SUMSUB_BA( %%mm1, %%mm3 ) /* mm1 = a3 + a2 mm3 = a3 - a2 */
|
| 95 |
"psubw %%mm1, %%mm4 \n\t" /* mm4 = a0 - a2 - a3 */ |
| 96 |
"movq %%mm4, %%mm1 \n\t" /* mm1 = a0 - a2 - a3 */ |
| 97 |
"psubw %%mm6, %%mm3 \n\t" /* mm3 = a3 - a2 - a1 */ |
| 98 |
"paddw %%mm1, %%mm1 \n\t"
|
| 99 |
"paddw %%mm3, %%mm3 \n\t"
|
| 100 |
"psubw %%mm2, %%mm1 \n\t" /* mm1 = b7 */ |
| 101 |
"paddw %%mm0, %%mm3 \n\t" /* mm3 = b6 */ |
| 102 |
|
| 103 |
"movq 32(%0), %%mm2 \n\t" /* mm2 = src2 */ |
| 104 |
"movq 96(%0), %%mm6 \n\t" /* mm6 = src6 */ |
| 105 |
"movq %%mm2, %%mm4 \n\t"
|
| 106 |
"movq %%mm6, %%mm0 \n\t"
|
| 107 |
"psllw $2, %%mm4 \n\t" /* mm4 = 4*src2 */ |
| 108 |
"psllw $2, %%mm6 \n\t" /* mm6 = 4*src6 */ |
| 109 |
"paddw %%mm4, %%mm2 \n\t" /* mm2 = 5*src2 */ |
| 110 |
"paddw %%mm6, %%mm0 \n\t" /* mm0 = 5*src6 */ |
| 111 |
"paddw %%mm2, %%mm2 \n\t"
|
| 112 |
"paddw %%mm0, %%mm0 \n\t"
|
| 113 |
"psubw %%mm0, %%mm4 \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */ |
| 114 |
"paddw %%mm2, %%mm6 \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */ |
| 115 |
|
| 116 |
"movq (%0), %%mm2 \n\t" /* mm2 = src0 */ |
| 117 |
"movq 64(%0), %%mm0 \n\t" /* mm0 = src4 */ |
| 118 |
SUMSUB_BA( %%mm0, %%mm2 ) /* mm0 = src0+src4 mm2 = src0-src4 */
|
| 119 |
"psllw $3, %%mm0 \n\t"
|
| 120 |
"psllw $3, %%mm2 \n\t"
|
| 121 |
"paddw %1, %%mm0 \n\t" /* add rounding bias */ |
| 122 |
"paddw %1, %%mm2 \n\t" /* add rounding bias */ |
| 123 |
|
| 124 |
SUMSUB_BA( %%mm6, %%mm0 ) /* mm6 = a4 + a6 mm0 = a4 - a6 */
|
| 125 |
SUMSUB_BA( %%mm4, %%mm2 ) /* mm4 = a5 + a7 mm2 = a5 - a7 */
|
| 126 |
SUMSUB_BA( %%mm7, %%mm6 ) /* mm7 = dst0 mm6 = dst7 */
|
| 127 |
SUMSUB_BA( %%mm5, %%mm4 ) /* mm5 = dst1 mm4 = dst6 */
|
| 128 |
SUMSUB_BA( %%mm3, %%mm2 ) /* mm3 = dst2 mm2 = dst5 */
|
| 129 |
SUMSUB_BA( %%mm1, %%mm0 ) /* mm1 = dst3 mm0 = dst4 */
|
| 130 |
:: "r"(block), "m"(bias) |
| 131 |
); |
| 132 |
} |
| 133 |
|
| 134 |
static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride) |
| 135 |
{
|
| 136 |
int i;
|
| 137 |
DECLARE_ALIGNED_8(int16_t, b2[64]);
|
| 138 |
|
| 139 |
for(i=0; i<2; i++){ |
| 140 |
DECLARE_ALIGNED_8(uint64_t, tmp); |
| 141 |
|
| 142 |
cavs_idct8_1d(block+4*i, ff_pw_4);
|
| 143 |
|
| 144 |
asm volatile( |
| 145 |
"psraw $3, %%mm7 \n\t"
|
| 146 |
"psraw $3, %%mm6 \n\t"
|
| 147 |
"psraw $3, %%mm5 \n\t"
|
| 148 |
"psraw $3, %%mm4 \n\t"
|
| 149 |
"psraw $3, %%mm3 \n\t"
|
| 150 |
"psraw $3, %%mm2 \n\t"
|
| 151 |
"psraw $3, %%mm1 \n\t"
|
| 152 |
"psraw $3, %%mm0 \n\t"
|
| 153 |
"movq %%mm7, %0 \n\t"
|
| 154 |
TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 ) |
| 155 |
"movq %%mm0, 8(%1) \n\t"
|
| 156 |
"movq %%mm6, 24(%1) \n\t"
|
| 157 |
"movq %%mm7, 40(%1) \n\t"
|
| 158 |
"movq %%mm4, 56(%1) \n\t"
|
| 159 |
"movq %0, %%mm7 \n\t"
|
| 160 |
TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 ) |
| 161 |
"movq %%mm7, (%1) \n\t"
|
| 162 |
"movq %%mm1, 16(%1) \n\t"
|
| 163 |
"movq %%mm0, 32(%1) \n\t"
|
| 164 |
"movq %%mm3, 48(%1) \n\t"
|
| 165 |
: "=m"(tmp)
|
| 166 |
: "r"(b2+32*i) |
| 167 |
: "memory"
|
| 168 |
); |
| 169 |
} |
| 170 |
|
| 171 |
for(i=0; i<2; i++){ |
| 172 |
cavs_idct8_1d(b2+4*i, ff_pw_64);
|
| 173 |
|
| 174 |
asm volatile( |
| 175 |
"psraw $7, %%mm7 \n\t"
|
| 176 |
"psraw $7, %%mm6 \n\t"
|
| 177 |
"psraw $7, %%mm5 \n\t"
|
| 178 |
"psraw $7, %%mm4 \n\t"
|
| 179 |
"psraw $7, %%mm3 \n\t"
|
| 180 |
"psraw $7, %%mm2 \n\t"
|
| 181 |
"psraw $7, %%mm1 \n\t"
|
| 182 |
"psraw $7, %%mm0 \n\t"
|
| 183 |
"movq %%mm7, (%0) \n\t"
|
| 184 |
"movq %%mm5, 16(%0) \n\t"
|
| 185 |
"movq %%mm3, 32(%0) \n\t"
|
| 186 |
"movq %%mm1, 48(%0) \n\t"
|
| 187 |
"movq %%mm0, 64(%0) \n\t"
|
| 188 |
"movq %%mm2, 80(%0) \n\t"
|
| 189 |
"movq %%mm4, 96(%0) \n\t"
|
| 190 |
"movq %%mm6, 112(%0) \n\t"
|
| 191 |
:: "r"(b2+4*i) |
| 192 |
: "memory"
|
| 193 |
); |
| 194 |
} |
| 195 |
|
| 196 |
add_pixels_clamped_mmx(b2, dst, stride); |
| 197 |
|
| 198 |
/* clear block */
|
| 199 |
asm volatile( |
| 200 |
"pxor %%mm7, %%mm7 \n\t"
|
| 201 |
"movq %%mm7, (%0) \n\t"
|
| 202 |
"movq %%mm7, 8(%0) \n\t"
|
| 203 |
"movq %%mm7, 16(%0) \n\t"
|
| 204 |
"movq %%mm7, 24(%0) \n\t"
|
| 205 |
"movq %%mm7, 32(%0) \n\t"
|
| 206 |
"movq %%mm7, 40(%0) \n\t"
|
| 207 |
"movq %%mm7, 48(%0) \n\t"
|
| 208 |
"movq %%mm7, 56(%0) \n\t"
|
| 209 |
"movq %%mm7, 64(%0) \n\t"
|
| 210 |
"movq %%mm7, 72(%0) \n\t"
|
| 211 |
"movq %%mm7, 80(%0) \n\t"
|
| 212 |
"movq %%mm7, 88(%0) \n\t"
|
| 213 |
"movq %%mm7, 96(%0) \n\t"
|
| 214 |
"movq %%mm7, 104(%0) \n\t"
|
| 215 |
"movq %%mm7, 112(%0) \n\t"
|
| 216 |
"movq %%mm7, 120(%0) \n\t"
|
| 217 |
:: "r" (block)
|
| 218 |
); |
| 219 |
} |
| 220 |
|
| 221 |
/*****************************************************************************
|
| 222 |
*
|
| 223 |
* motion compensation
|
| 224 |
*
|
| 225 |
****************************************************************************/
|
| 226 |
|
| 227 |
/* vertical filter [-1 -2 96 42 -7 0] */
|
| 228 |
#define QPEL_CAVSV1(A,B,C,D,E,F,OP) \
|
| 229 |
"movd (%0), "#F" \n\t"\ |
| 230 |
"movq "#C", %%mm6 \n\t"\ |
| 231 |
"pmullw %5, %%mm6 \n\t"\
|
| 232 |
"movq "#D", %%mm7 \n\t"\ |
| 233 |
"pmullw %6, %%mm7 \n\t"\
|
| 234 |
"psllw $3, "#E" \n\t"\ |
| 235 |
"psubw "#E", %%mm6 \n\t"\ |
| 236 |
"psraw $3, "#E" \n\t"\ |
| 237 |
"paddw %%mm7, %%mm6 \n\t"\
|
| 238 |
"paddw "#E", %%mm6 \n\t"\ |
| 239 |
"paddw "#B", "#B" \n\t"\ |
| 240 |
"pxor %%mm7, %%mm7 \n\t"\
|
| 241 |
"add %2, %0 \n\t"\
|
| 242 |
"punpcklbw %%mm7, "#F" \n\t"\ |
| 243 |
"psubw "#B", %%mm6 \n\t"\ |
| 244 |
"psraw $1, "#B" \n\t"\ |
| 245 |
"psubw "#A", %%mm6 \n\t"\ |
| 246 |
"paddw %4, %%mm6 \n\t"\
|
| 247 |
"psraw $7, %%mm6 \n\t"\
|
| 248 |
"packuswb %%mm6, %%mm6 \n\t"\
|
| 249 |
OP(%%mm6, (%1), A, d) \
|
| 250 |
"add %3, %1 \n\t"
|
| 251 |
|
| 252 |
/* vertical filter [ 0 -1 5 5 -1 0] */
|
| 253 |
#define QPEL_CAVSV2(A,B,C,D,E,F,OP) \
|
| 254 |
"movd (%0), "#F" \n\t"\ |
| 255 |
"movq "#C", %%mm6 \n\t"\ |
| 256 |
"paddw "#D", %%mm6 \n\t"\ |
| 257 |
"pmullw %5, %%mm6 \n\t"\
|
| 258 |
"add %2, %0 \n\t"\
|
| 259 |
"punpcklbw %%mm7, "#F" \n\t"\ |
| 260 |
"psubw "#B", %%mm6 \n\t"\ |
| 261 |
"psubw "#E", %%mm6 \n\t"\ |
| 262 |
"paddw %4, %%mm6 \n\t"\
|
| 263 |
"psraw $3, %%mm6 \n\t"\
|
| 264 |
"packuswb %%mm6, %%mm6 \n\t"\
|
| 265 |
OP(%%mm6, (%1), A, d) \
|
| 266 |
"add %3, %1 \n\t"
|
| 267 |
|
| 268 |
/* vertical filter [ 0 -7 42 96 -2 -1] */
|
| 269 |
#define QPEL_CAVSV3(A,B,C,D,E,F,OP) \
|
| 270 |
"movd (%0), "#F" \n\t"\ |
| 271 |
"movq "#C", %%mm6 \n\t"\ |
| 272 |
"pmullw %6, %%mm6 \n\t"\
|
| 273 |
"movq "#D", %%mm7 \n\t"\ |
| 274 |
"pmullw %5, %%mm7 \n\t"\
|
| 275 |
"psllw $3, "#B" \n\t"\ |
| 276 |
"psubw "#B", %%mm6 \n\t"\ |
| 277 |
"psraw $3, "#B" \n\t"\ |
| 278 |
"paddw %%mm7, %%mm6 \n\t"\
|
| 279 |
"paddw "#B", %%mm6 \n\t"\ |
| 280 |
"paddw "#E", "#E" \n\t"\ |
| 281 |
"pxor %%mm7, %%mm7 \n\t"\
|
| 282 |
"add %2, %0 \n\t"\
|
| 283 |
"punpcklbw %%mm7, "#F" \n\t"\ |
| 284 |
"psubw "#E", %%mm6 \n\t"\ |
| 285 |
"psraw $1, "#E" \n\t"\ |
| 286 |
"psubw "#F", %%mm6 \n\t"\ |
| 287 |
"paddw %4, %%mm6 \n\t"\
|
| 288 |
"psraw $7, %%mm6 \n\t"\
|
| 289 |
"packuswb %%mm6, %%mm6 \n\t"\
|
| 290 |
OP(%%mm6, (%1), A, d) \
|
| 291 |
"add %3, %1 \n\t"
|
| 292 |
|
| 293 |
|
| 294 |
#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
|
| 295 |
int w= 2;\ |
| 296 |
src -= 2*srcStride;\
|
| 297 |
\ |
| 298 |
while(w--){\
|
| 299 |
asm volatile(\ |
| 300 |
"pxor %%mm7, %%mm7 \n\t"\
|
| 301 |
"movd (%0), %%mm0 \n\t"\
|
| 302 |
"add %2, %0 \n\t"\
|
| 303 |
"movd (%0), %%mm1 \n\t"\
|
| 304 |
"add %2, %0 \n\t"\
|
| 305 |
"movd (%0), %%mm2 \n\t"\
|
| 306 |
"add %2, %0 \n\t"\
|
| 307 |
"movd (%0), %%mm3 \n\t"\
|
| 308 |
"add %2, %0 \n\t"\
|
| 309 |
"movd (%0), %%mm4 \n\t"\
|
| 310 |
"add %2, %0 \n\t"\
|
| 311 |
"punpcklbw %%mm7, %%mm0 \n\t"\
|
| 312 |
"punpcklbw %%mm7, %%mm1 \n\t"\
|
| 313 |
"punpcklbw %%mm7, %%mm2 \n\t"\
|
| 314 |
"punpcklbw %%mm7, %%mm3 \n\t"\
|
| 315 |
"punpcklbw %%mm7, %%mm4 \n\t"\
|
| 316 |
VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\ |
| 317 |
VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\ |
| 318 |
VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\ |
| 319 |
VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\ |
| 320 |
VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\ |
| 321 |
VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\ |
| 322 |
VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\ |
| 323 |
VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\ |
| 324 |
\ |
| 325 |
: "+a"(src), "+c"(dst)\ |
| 326 |
: "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\ |
| 327 |
: "memory"\
|
| 328 |
);\ |
| 329 |
if(h==16){\ |
| 330 |
asm volatile(\ |
| 331 |
VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\ |
| 332 |
VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\ |
| 333 |
VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\ |
| 334 |
VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\ |
| 335 |
VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\ |
| 336 |
VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\ |
| 337 |
VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\ |
| 338 |
VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\ |
| 339 |
\ |
| 340 |
: "+a"(src), "+c"(dst)\ |
| 341 |
: "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\ |
| 342 |
: "memory"\
|
| 343 |
);\ |
| 344 |
}\ |
| 345 |
src += 4-(h+5)*srcStride;\ |
| 346 |
dst += 4-h*dstStride;\
|
| 347 |
} |
| 348 |
|
| 349 |
#define QPEL_CAVS(OPNAME, OP, MMX)\
|
| 350 |
static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 351 |
int h=8;\ |
| 352 |
asm volatile(\ |
| 353 |
"pxor %%mm7, %%mm7 \n\t"\
|
| 354 |
"movq %5, %%mm6 \n\t"\
|
| 355 |
"1: \n\t"\
|
| 356 |
"movq (%0), %%mm0 \n\t"\
|
| 357 |
"movq 1(%0), %%mm2 \n\t"\
|
| 358 |
"movq %%mm0, %%mm1 \n\t"\
|
| 359 |
"movq %%mm2, %%mm3 \n\t"\
|
| 360 |
"punpcklbw %%mm7, %%mm0 \n\t"\
|
| 361 |
"punpckhbw %%mm7, %%mm1 \n\t"\
|
| 362 |
"punpcklbw %%mm7, %%mm2 \n\t"\
|
| 363 |
"punpckhbw %%mm7, %%mm3 \n\t"\
|
| 364 |
"paddw %%mm2, %%mm0 \n\t"\
|
| 365 |
"paddw %%mm3, %%mm1 \n\t"\
|
| 366 |
"pmullw %%mm6, %%mm0 \n\t"\
|
| 367 |
"pmullw %%mm6, %%mm1 \n\t"\
|
| 368 |
"movq -1(%0), %%mm2 \n\t"\
|
| 369 |
"movq 2(%0), %%mm4 \n\t"\
|
| 370 |
"movq %%mm2, %%mm3 \n\t"\
|
| 371 |
"movq %%mm4, %%mm5 \n\t"\
|
| 372 |
"punpcklbw %%mm7, %%mm2 \n\t"\
|
| 373 |
"punpckhbw %%mm7, %%mm3 \n\t"\
|
| 374 |
"punpcklbw %%mm7, %%mm4 \n\t"\
|
| 375 |
"punpckhbw %%mm7, %%mm5 \n\t"\
|
| 376 |
"paddw %%mm4, %%mm2 \n\t"\
|
| 377 |
"paddw %%mm3, %%mm5 \n\t"\
|
| 378 |
"psubw %%mm2, %%mm0 \n\t"\
|
| 379 |
"psubw %%mm5, %%mm1 \n\t"\
|
| 380 |
"movq %6, %%mm5 \n\t"\
|
| 381 |
"paddw %%mm5, %%mm0 \n\t"\
|
| 382 |
"paddw %%mm5, %%mm1 \n\t"\
|
| 383 |
"psraw $3, %%mm0 \n\t"\
|
| 384 |
"psraw $3, %%mm1 \n\t"\
|
| 385 |
"packuswb %%mm1, %%mm0 \n\t"\
|
| 386 |
OP(%%mm0, (%1),%%mm5, q) \
|
| 387 |
"add %3, %0 \n\t"\
|
| 388 |
"add %4, %1 \n\t"\
|
| 389 |
"decl %2 \n\t"\
|
| 390 |
" jnz 1b \n\t"\
|
| 391 |
: "+a"(src), "+c"(dst), "+m"(h)\ |
| 392 |
: "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\ |
| 393 |
: "memory"\
|
| 394 |
);\ |
| 395 |
}\ |
| 396 |
\ |
| 397 |
static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
| 398 |
QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ |
| 399 |
}\ |
| 400 |
\ |
| 401 |
static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
| 402 |
QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5) \ |
| 403 |
}\ |
| 404 |
\ |
| 405 |
static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\ |
| 406 |
QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42) \ |
| 407 |
}\ |
| 408 |
\ |
| 409 |
static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 410 |
OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 8);\ |
| 411 |
}\ |
| 412 |
static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 413 |
OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst , src , dstStride, srcStride, 16);\ |
| 414 |
OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ |
| 415 |
}\ |
| 416 |
\ |
| 417 |
static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 418 |
OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 8);\ |
| 419 |
}\ |
| 420 |
static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 421 |
OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst , src , dstStride, srcStride, 16);\ |
| 422 |
OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ |
| 423 |
}\ |
| 424 |
\ |
| 425 |
static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 426 |
OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 8);\ |
| 427 |
}\ |
| 428 |
static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 429 |
OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst , src , dstStride, srcStride, 16);\ |
| 430 |
OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\ |
| 431 |
}\ |
| 432 |
\ |
| 433 |
static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\ |
| 434 |
OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ |
| 435 |
OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ |
| 436 |
src += 8*srcStride;\
|
| 437 |
dst += 8*dstStride;\
|
| 438 |
OPNAME ## cavs_qpel8_h_ ## MMX(dst , src , dstStride, srcStride);\ |
| 439 |
OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\ |
| 440 |
}\ |
| 441 |
|
| 442 |
#define CAVS_MC(OPNAME, SIZE, MMX) \
|
| 443 |
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ |
| 444 |
OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\ |
| 445 |
}\ |
| 446 |
\ |
| 447 |
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ |
| 448 |
OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\ |
| 449 |
}\ |
| 450 |
\ |
| 451 |
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ |
| 452 |
OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\ |
| 453 |
}\ |
| 454 |
\ |
| 455 |
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\ |
| 456 |
OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\ |
| 457 |
}\ |
| 458 |
|
| 459 |
#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" |
| 460 |
#define AVG_3DNOW_OP(a,b,temp, size) \
|
| 461 |
"mov" #size " " #b ", " #temp " \n\t"\ |
| 462 |
"pavgusb " #temp ", " #a " \n\t"\ |
| 463 |
"mov" #size " " #a ", " #b " \n\t" |
| 464 |
#define AVG_MMX2_OP(a,b,temp, size) \
|
| 465 |
"mov" #size " " #b ", " #temp " \n\t"\ |
| 466 |
"pavgb " #temp ", " #a " \n\t"\ |
| 467 |
"mov" #size " " #a ", " #b " \n\t" |
| 468 |
|
| 469 |
QPEL_CAVS(put_, PUT_OP, 3dnow)
|
| 470 |
QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
|
| 471 |
QPEL_CAVS(put_, PUT_OP, mmx2) |
| 472 |
QPEL_CAVS(avg_, AVG_MMX2_OP, mmx2) |
| 473 |
|
| 474 |
CAVS_MC(put_, 8, 3dnow) |
| 475 |
CAVS_MC(put_, 16,3dnow) |
| 476 |
CAVS_MC(avg_, 8, 3dnow) |
| 477 |
CAVS_MC(avg_, 16,3dnow) |
| 478 |
CAVS_MC(put_, 8, mmx2)
|
| 479 |
CAVS_MC(put_, 16,mmx2)
|
| 480 |
CAVS_MC(avg_, 8, mmx2)
|
| 481 |
CAVS_MC(avg_, 16,mmx2)
|
| 482 |
|
| 483 |
void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride); |
| 484 |
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride); |
| 485 |
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride); |
| 486 |
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride); |
| 487 |
|
| 488 |
void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
|
| 489 |
#define dspfunc(PFX, IDX, NUM) \
|
| 490 |
c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \ |
| 491 |
c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \ |
| 492 |
c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \ |
| 493 |
c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \ |
| 494 |
c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \ |
| 495 |
|
| 496 |
dspfunc(put_cavs_qpel, 0, 16); |
| 497 |
dspfunc(put_cavs_qpel, 1, 8); |
| 498 |
dspfunc(avg_cavs_qpel, 0, 16); |
| 499 |
dspfunc(avg_cavs_qpel, 1, 8); |
| 500 |
#undef dspfunc
|
| 501 |
c->cavs_idct8_add = cavs_idct8_add_mmx; |
| 502 |
} |
| 503 |
|
| 504 |
void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
|
| 505 |
#define dspfunc(PFX, IDX, NUM) \
|
| 506 |
c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \ |
| 507 |
c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \ |
| 508 |
c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \ |
| 509 |
c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \ |
| 510 |
c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \ |
| 511 |
|
| 512 |
dspfunc(put_cavs_qpel, 0, 16); |
| 513 |
dspfunc(put_cavs_qpel, 1, 8); |
| 514 |
dspfunc(avg_cavs_qpel, 0, 16); |
| 515 |
dspfunc(avg_cavs_qpel, 1, 8); |
| 516 |
#undef dspfunc
|
| 517 |
c->cavs_idct8_add = cavs_idct8_add_mmx; |
| 518 |
} |