Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__H264__App

annotate libavcodec/x86/h264dsp_mmx.c @ 2:897f711a7157

find changesets by author, revision, files, or words in the commit message
rearrange to work with autoconf
author Nina Engelhardt <nengel@mailbox.tu-berlin.de>
date Tue, 25 Sep 2012 15:55:33 +0200
parents
children
rev   line source
nengel@2 1 /*
nengel@2 2 * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
nengel@2 3 *
nengel@2 4 * This file is part of FFmpeg.
nengel@2 5 *
nengel@2 6 * FFmpeg is free software; you can redistribute it and/or
nengel@2 7 * modify it under the terms of the GNU Lesser General Public
nengel@2 8 * License as published by the Free Software Foundation; either
nengel@2 9 * version 2.1 of the License, or (at your option) any later version.
nengel@2 10 *
nengel@2 11 * FFmpeg is distributed in the hope that it will be useful,
nengel@2 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
nengel@2 14 * Lesser General Public License for more details.
nengel@2 15 *
nengel@2 16 * You should have received a copy of the GNU Lesser General Public
nengel@2 17 * License along with FFmpeg; if not, write to the Free Software
nengel@2 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2 19 */
nengel@2 20
nengel@2 21 #include "dsputil_mmx.h"
nengel@2 22
nengel@2 23 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1 ) = 0x0103010301030103ULL;
nengel@2 24 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_7_3 ) = 0x0307030703070307ULL;
nengel@2 25
nengel@2 26 /***********************************/
nengel@2 27 /* IDCT */
nengel@2 28
nengel@2 29 #define SUMSUB_BADC( a, b, c, d ) \
nengel@2 30 "paddw "#b", "#a" \n\t"\
nengel@2 31 "paddw "#d", "#c" \n\t"\
nengel@2 32 "paddw "#b", "#b" \n\t"\
nengel@2 33 "paddw "#d", "#d" \n\t"\
nengel@2 34 "psubw "#a", "#b" \n\t"\
nengel@2 35 "psubw "#c", "#d" \n\t"
nengel@2 36
nengel@2 37 #define SUMSUBD2_AB( a, b, t ) \
nengel@2 38 "movq "#b", "#t" \n\t"\
nengel@2 39 "psraw $1 , "#b" \n\t"\
nengel@2 40 "paddw "#a", "#b" \n\t"\
nengel@2 41 "psraw $1 , "#a" \n\t"\
nengel@2 42 "psubw "#t", "#a" \n\t"
nengel@2 43
nengel@2 44 #define IDCT4_1D( s02, s13, d02, d13, t ) \
nengel@2 45 SUMSUB_BA ( s02, d02 )\
nengel@2 46 SUMSUBD2_AB( s13, d13, t )\
nengel@2 47 SUMSUB_BADC( d13, s02, s13, d02 )
nengel@2 48
nengel@2 49 #define STORE_DIFF_4P( p, t, z ) \
nengel@2 50 "psraw $6, "#p" \n\t"\
nengel@2 51 "movd (%0), "#t" \n\t"\
nengel@2 52 "punpcklbw "#z", "#t" \n\t"\
nengel@2 53 "paddsw "#t", "#p" \n\t"\
nengel@2 54 "packuswb "#z", "#p" \n\t"\
nengel@2 55 "movd "#p", (%0) \n\t"
nengel@2 56
nengel@2 57 static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
nengel@2 58 {
nengel@2 59 /* Load dct coeffs */
nengel@2 60 __asm__ volatile(
nengel@2 61 "movq (%0), %%mm0 \n\t"
nengel@2 62 "movq 8(%0), %%mm1 \n\t"
nengel@2 63 "movq 16(%0), %%mm2 \n\t"
nengel@2 64 "movq 24(%0), %%mm3 \n\t"
nengel@2 65 :: "r"(block) );
nengel@2 66
nengel@2 67 __asm__ volatile(
nengel@2 68 /* mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13 */
nengel@2 69 IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )
nengel@2 70
nengel@2 71 "movq %0, %%mm6 \n\t"
nengel@2 72 /* in: 1,4,0,2 out: 1,2,3,0 */
nengel@2 73 TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )
nengel@2 74
nengel@2 75 "paddw %%mm6, %%mm3 \n\t"
nengel@2 76
nengel@2 77 /* mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13 */
nengel@2 78 IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )
nengel@2 79
nengel@2 80 "pxor %%mm7, %%mm7 \n\t"
nengel@2 81 :: "m"(ff_pw_32));
nengel@2 82
nengel@2 83 __asm__ volatile(
nengel@2 84 STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
nengel@2 85 "add %1, %0 \n\t"
nengel@2 86 STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
nengel@2 87 "add %1, %0 \n\t"
nengel@2 88 STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)
nengel@2 89 "add %1, %0 \n\t"
nengel@2 90 STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)
nengel@2 91 : "+r"(dst)
nengel@2 92 : "r" ((x86_reg)stride)
nengel@2 93 );
nengel@2 94 }
nengel@2 95
nengel@2 96 static inline void h264_idct8_1d(int16_t *block)
nengel@2 97 {
nengel@2 98 __asm__ volatile(
nengel@2 99 "movq 112(%0), %%mm7 \n\t"
nengel@2 100 "movq 80(%0), %%mm0 \n\t"
nengel@2 101 "movq 48(%0), %%mm3 \n\t"
nengel@2 102 "movq 16(%0), %%mm5 \n\t"
nengel@2 103
nengel@2 104 "movq %%mm0, %%mm4 \n\t"
nengel@2 105 "movq %%mm5, %%mm1 \n\t"
nengel@2 106 "psraw $1, %%mm4 \n\t"
nengel@2 107 "psraw $1, %%mm1 \n\t"
nengel@2 108 "paddw %%mm0, %%mm4 \n\t"
nengel@2 109 "paddw %%mm5, %%mm1 \n\t"
nengel@2 110 "paddw %%mm7, %%mm4 \n\t"
nengel@2 111 "paddw %%mm0, %%mm1 \n\t"
nengel@2 112 "psubw %%mm5, %%mm4 \n\t"
nengel@2 113 "paddw %%mm3, %%mm1 \n\t"
nengel@2 114
nengel@2 115 "psubw %%mm3, %%mm5 \n\t"
nengel@2 116 "psubw %%mm3, %%mm0 \n\t"
nengel@2 117 "paddw %%mm7, %%mm5 \n\t"
nengel@2 118 "psubw %%mm7, %%mm0 \n\t"
nengel@2 119 "psraw $1, %%mm3 \n\t"
nengel@2 120 "psraw $1, %%mm7 \n\t"
nengel@2 121 "psubw %%mm3, %%mm5 \n\t"
nengel@2 122 "psubw %%mm7, %%mm0 \n\t"
nengel@2 123
nengel@2 124 "movq %%mm4, %%mm3 \n\t"
nengel@2 125 "movq %%mm1, %%mm7 \n\t"
nengel@2 126 "psraw $2, %%mm1 \n\t"
nengel@2 127 "psraw $2, %%mm3 \n\t"
nengel@2 128 "paddw %%mm5, %%mm3 \n\t"
nengel@2 129 "psraw $2, %%mm5 \n\t"
nengel@2 130 "paddw %%mm0, %%mm1 \n\t"
nengel@2 131 "psraw $2, %%mm0 \n\t"
nengel@2 132 "psubw %%mm4, %%mm5 \n\t"
nengel@2 133 "psubw %%mm0, %%mm7 \n\t"
nengel@2 134
nengel@2 135 "movq 32(%0), %%mm2 \n\t"
nengel@2 136 "movq 96(%0), %%mm6 \n\t"
nengel@2 137 "movq %%mm2, %%mm4 \n\t"
nengel@2 138 "movq %%mm6, %%mm0 \n\t"
nengel@2 139 "psraw $1, %%mm4 \n\t"
nengel@2 140 "psraw $1, %%mm6 \n\t"
nengel@2 141 "psubw %%mm0, %%mm4 \n\t"
nengel@2 142 "paddw %%mm2, %%mm6 \n\t"
nengel@2 143
nengel@2 144 "movq (%0), %%mm2 \n\t"
nengel@2 145 "movq 64(%0), %%mm0 \n\t"
nengel@2 146 SUMSUB_BA( %%mm0, %%mm2 )
nengel@2 147 SUMSUB_BA( %%mm6, %%mm0 )
nengel@2 148 SUMSUB_BA( %%mm4, %%mm2 )
nengel@2 149 SUMSUB_BA( %%mm7, %%mm6 )
nengel@2 150 SUMSUB_BA( %%mm5, %%mm4 )
nengel@2 151 SUMSUB_BA( %%mm3, %%mm2 )
nengel@2 152 SUMSUB_BA( %%mm1, %%mm0 )
nengel@2 153 :: "r"(block)
nengel@2 154 );
nengel@2 155 }
nengel@2 156
nengel@2 157 static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
nengel@2 158 {
nengel@2 159 int i;
nengel@2 160 DECLARE_ALIGNED(8, int16_t, b2)[64];
nengel@2 161
nengel@2 162 block[0] += 32;
nengel@2 163
nengel@2 164 for(i=0; i<2; i++){
nengel@2 165 DECLARE_ALIGNED(8, uint64_t, tmp);
nengel@2 166
nengel@2 167 h264_idct8_1d(block+4*i);
nengel@2 168
nengel@2 169 __asm__ volatile(
nengel@2 170 "movq %%mm7, %0 \n\t"
nengel@2 171 TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
nengel@2 172 "movq %%mm0, 8(%1) \n\t"
nengel@2 173 "movq %%mm6, 24(%1) \n\t"
nengel@2 174 "movq %%mm7, 40(%1) \n\t"
nengel@2 175 "movq %%mm4, 56(%1) \n\t"
nengel@2 176 "movq %0, %%mm7 \n\t"
nengel@2 177 TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )
nengel@2 178 "movq %%mm7, (%1) \n\t"
nengel@2 179 "movq %%mm1, 16(%1) \n\t"
nengel@2 180 "movq %%mm0, 32(%1) \n\t"
nengel@2 181 "movq %%mm3, 48(%1) \n\t"
nengel@2 182 : "=m"(tmp)
nengel@2 183 : "r"(b2+32*i)
nengel@2 184 : "memory"
nengel@2 185 );
nengel@2 186 }
nengel@2 187
nengel@2 188 for(i=0; i<2; i++){
nengel@2 189 h264_idct8_1d(b2+4*i);
nengel@2 190
nengel@2 191 __asm__ volatile(
nengel@2 192 "psraw $6, %%mm7 \n\t"
nengel@2 193 "psraw $6, %%mm6 \n\t"
nengel@2 194 "psraw $6, %%mm5 \n\t"
nengel@2 195 "psraw $6, %%mm4 \n\t"
nengel@2 196 "psraw $6, %%mm3 \n\t"
nengel@2 197 "psraw $6, %%mm2 \n\t"
nengel@2 198 "psraw $6, %%mm1 \n\t"
nengel@2 199 "psraw $6, %%mm0 \n\t"
nengel@2 200
nengel@2 201 "movq %%mm7, (%0) \n\t"
nengel@2 202 "movq %%mm5, 16(%0) \n\t"
nengel@2 203 "movq %%mm3, 32(%0) \n\t"
nengel@2 204 "movq %%mm1, 48(%0) \n\t"
nengel@2 205 "movq %%mm0, 64(%0) \n\t"
nengel@2 206 "movq %%mm2, 80(%0) \n\t"
nengel@2 207 "movq %%mm4, 96(%0) \n\t"
nengel@2 208 "movq %%mm6, 112(%0) \n\t"
nengel@2 209 :: "r"(b2+4*i)
nengel@2 210 : "memory"
nengel@2 211 );
nengel@2 212 }
nengel@2 213
nengel@2 214 add_pixels_clamped_mmx(b2, dst, stride);
nengel@2 215 }
nengel@2 216
nengel@2 217 #define STORE_DIFF_8P( p, d, t, z )\
nengel@2 218 "movq "#d", "#t" \n"\
nengel@2 219 "psraw $6, "#p" \n"\
nengel@2 220 "punpcklbw "#z", "#t" \n"\
nengel@2 221 "paddsw "#t", "#p" \n"\
nengel@2 222 "packuswb "#p", "#p" \n"\
nengel@2 223 "movq "#p", "#d" \n"
nengel@2 224
nengel@2 225 #define H264_IDCT8_1D_SSE2(a,b,c,d,e,f,g,h)\
nengel@2 226 "movdqa "#c", "#a" \n"\
nengel@2 227 "movdqa "#g", "#e" \n"\
nengel@2 228 "psraw $1, "#c" \n"\
nengel@2 229 "psraw $1, "#g" \n"\
nengel@2 230 "psubw "#e", "#c" \n"\
nengel@2 231 "paddw "#a", "#g" \n"\
nengel@2 232 "movdqa "#b", "#e" \n"\
nengel@2 233 "psraw $1, "#e" \n"\
nengel@2 234 "paddw "#b", "#e" \n"\
nengel@2 235 "paddw "#d", "#e" \n"\
nengel@2 236 "paddw "#f", "#e" \n"\
nengel@2 237 "movdqa "#f", "#a" \n"\
nengel@2 238 "psraw $1, "#a" \n"\
nengel@2 239 "paddw "#f", "#a" \n"\
nengel@2 240 "paddw "#h", "#a" \n"\
nengel@2 241 "psubw "#b", "#a" \n"\
nengel@2 242 "psubw "#d", "#b" \n"\
nengel@2 243 "psubw "#d", "#f" \n"\
nengel@2 244 "paddw "#h", "#b" \n"\
nengel@2 245 "psubw "#h", "#f" \n"\
nengel@2 246 "psraw $1, "#d" \n"\
nengel@2 247 "psraw $1, "#h" \n"\
nengel@2 248 "psubw "#d", "#b" \n"\
nengel@2 249 "psubw "#h", "#f" \n"\
nengel@2 250 "movdqa "#e", "#d" \n"\
nengel@2 251 "movdqa "#a", "#h" \n"\
nengel@2 252 "psraw $2, "#d" \n"\
nengel@2 253 "psraw $2, "#h" \n"\
nengel@2 254 "paddw "#f", "#d" \n"\
nengel@2 255 "paddw "#b", "#h" \n"\
nengel@2 256 "psraw $2, "#f" \n"\
nengel@2 257 "psraw $2, "#b" \n"\
nengel@2 258 "psubw "#f", "#e" \n"\
nengel@2 259 "psubw "#a", "#b" \n"\
nengel@2 260 "movdqa 0x00(%1), "#a" \n"\
nengel@2 261 "movdqa 0x40(%1), "#f" \n"\
nengel@2 262 SUMSUB_BA(f, a)\
nengel@2 263 SUMSUB_BA(g, f)\
nengel@2 264 SUMSUB_BA(c, a)\
nengel@2 265 SUMSUB_BA(e, g)\
nengel@2 266 SUMSUB_BA(b, c)\
nengel@2 267 SUMSUB_BA(h, a)\
nengel@2 268 SUMSUB_BA(d, f)
nengel@2 269
nengel@2 270 static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
nengel@2 271 {
nengel@2 272 __asm__ volatile(
nengel@2 273 "movdqa 0x10(%1), %%xmm1 \n"
nengel@2 274 "movdqa 0x20(%1), %%xmm2 \n"
nengel@2 275 "movdqa 0x30(%1), %%xmm3 \n"
nengel@2 276 "movdqa 0x50(%1), %%xmm5 \n"
nengel@2 277 "movdqa 0x60(%1), %%xmm6 \n"
nengel@2 278 "movdqa 0x70(%1), %%xmm7 \n"
nengel@2 279 H264_IDCT8_1D_SSE2(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
nengel@2 280 TRANSPOSE8(%%xmm4, %%xmm1, %%xmm7, %%xmm3, %%xmm5, %%xmm0, %%xmm2, %%xmm6, (%1))
nengel@2 281 "paddw %4, %%xmm4 \n"
nengel@2 282 "movdqa %%xmm4, 0x00(%1) \n"
nengel@2 283 "movdqa %%xmm2, 0x40(%1) \n"
nengel@2 284 H264_IDCT8_1D_SSE2(%%xmm4, %%xmm0, %%xmm6, %%xmm3, %%xmm2, %%xmm5, %%xmm7, %%xmm1)
nengel@2 285 "movdqa %%xmm6, 0x60(%1) \n"
nengel@2 286 "movdqa %%xmm7, 0x70(%1) \n"
nengel@2 287 "pxor %%xmm7, %%xmm7 \n"
nengel@2 288 STORE_DIFF_8P(%%xmm2, (%0), %%xmm6, %%xmm7)
nengel@2 289 STORE_DIFF_8P(%%xmm0, (%0,%2), %%xmm6, %%xmm7)
nengel@2 290 STORE_DIFF_8P(%%xmm1, (%0,%2,2), %%xmm6, %%xmm7)
nengel@2 291 STORE_DIFF_8P(%%xmm3, (%0,%3), %%xmm6, %%xmm7)
nengel@2 292 "lea (%0,%2,4), %0 \n"
nengel@2 293 STORE_DIFF_8P(%%xmm5, (%0), %%xmm6, %%xmm7)
nengel@2 294 STORE_DIFF_8P(%%xmm4, (%0,%2), %%xmm6, %%xmm7)
nengel@2 295 "movdqa 0x60(%1), %%xmm0 \n"
nengel@2 296 "movdqa 0x70(%1), %%xmm1 \n"
nengel@2 297 STORE_DIFF_8P(%%xmm0, (%0,%2,2), %%xmm6, %%xmm7)
nengel@2 298 STORE_DIFF_8P(%%xmm1, (%0,%3), %%xmm6, %%xmm7)
nengel@2 299 :"+r"(dst)
nengel@2 300 :"r"(block), "r"((x86_reg)stride), "r"((x86_reg)3L*stride), "m"(ff_pw_32)
nengel@2 301 );
nengel@2 302 }
nengel@2 303
nengel@2 304 static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
nengel@2 305 {
nengel@2 306 int dc = (block[0] + 32) >> 6;
nengel@2 307 __asm__ volatile(
nengel@2 308 "movd %0, %%mm0 \n\t"
nengel@2 309 "pshufw $0, %%mm0, %%mm0 \n\t"
nengel@2 310 "pxor %%mm1, %%mm1 \n\t"
nengel@2 311 "psubw %%mm0, %%mm1 \n\t"
nengel@2 312 "packuswb %%mm0, %%mm0 \n\t"
nengel@2 313 "packuswb %%mm1, %%mm1 \n\t"
nengel@2 314 ::"r"(dc)
nengel@2 315 );
nengel@2 316 __asm__ volatile(
nengel@2 317 "movd %0, %%mm2 \n\t"
nengel@2 318 "movd %1, %%mm3 \n\t"
nengel@2 319 "movd %2, %%mm4 \n\t"
nengel@2 320 "movd %3, %%mm5 \n\t"
nengel@2 321 "paddusb %%mm0, %%mm2 \n\t"
nengel@2 322 "paddusb %%mm0, %%mm3 \n\t"
nengel@2 323 "paddusb %%mm0, %%mm4 \n\t"
nengel@2 324 "paddusb %%mm0, %%mm5 \n\t"
nengel@2 325 "psubusb %%mm1, %%mm2 \n\t"
nengel@2 326 "psubusb %%mm1, %%mm3 \n\t"
nengel@2 327 "psubusb %%mm1, %%mm4 \n\t"
nengel@2 328 "psubusb %%mm1, %%mm5 \n\t"
nengel@2 329 "movd %%mm2, %0 \n\t"
nengel@2 330 "movd %%mm3, %1 \n\t"
nengel@2 331 "movd %%mm4, %2 \n\t"
nengel@2 332 "movd %%mm5, %3 \n\t"
nengel@2 333 :"+m"(*(uint32_t*)(dst+0*stride)),
nengel@2 334 "+m"(*(uint32_t*)(dst+1*stride)),
nengel@2 335 "+m"(*(uint32_t*)(dst+2*stride)),
nengel@2 336 "+m"(*(uint32_t*)(dst+3*stride))
nengel@2 337 );
nengel@2 338 }
nengel@2 339
nengel@2 340 static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
nengel@2 341 {
nengel@2 342 int dc = (block[0] + 32) >> 6;
nengel@2 343 int y;
nengel@2 344 __asm__ volatile(
nengel@2 345 "movd %0, %%mm0 \n\t"
nengel@2 346 "pshufw $0, %%mm0, %%mm0 \n\t"
nengel@2 347 "pxor %%mm1, %%mm1 \n\t"
nengel@2 348 "psubw %%mm0, %%mm1 \n\t"
nengel@2 349 "packuswb %%mm0, %%mm0 \n\t"
nengel@2 350 "packuswb %%mm1, %%mm1 \n\t"
nengel@2 351 ::"r"(dc)
nengel@2 352 );
nengel@2 353 for(y=2; y--; dst += 4*stride){
nengel@2 354 __asm__ volatile(
nengel@2 355 "movq %0, %%mm2 \n\t"
nengel@2 356 "movq %1, %%mm3 \n\t"
nengel@2 357 "movq %2, %%mm4 \n\t"
nengel@2 358 "movq %3, %%mm5 \n\t"
nengel@2 359 "paddusb %%mm0, %%mm2 \n\t"
nengel@2 360 "paddusb %%mm0, %%mm3 \n\t"
nengel@2 361 "paddusb %%mm0, %%mm4 \n\t"
nengel@2 362 "paddusb %%mm0, %%mm5 \n\t"
nengel@2 363 "psubusb %%mm1, %%mm2 \n\t"
nengel@2 364 "psubusb %%mm1, %%mm3 \n\t"
nengel@2 365 "psubusb %%mm1, %%mm4 \n\t"
nengel@2 366 "psubusb %%mm1, %%mm5 \n\t"
nengel@2 367 "movq %%mm2, %0 \n\t"
nengel@2 368 "movq %%mm3, %1 \n\t"
nengel@2 369 "movq %%mm4, %2 \n\t"
nengel@2 370 "movq %%mm5, %3 \n\t"
nengel@2 371 :"+m"(*(uint64_t*)(dst+0*stride)),
nengel@2 372 "+m"(*(uint64_t*)(dst+1*stride)),
nengel@2 373 "+m"(*(uint64_t*)(dst+2*stride)),
nengel@2 374 "+m"(*(uint64_t*)(dst+3*stride))
nengel@2 375 );
nengel@2 376 }
nengel@2 377 }
nengel@2 378
nengel@2 379 //FIXME this table is a duplicate from h264data.h, and will be removed once the tables from, h264 have been split
nengel@2 380 static const uint8_t scan8[16 + 2*4]={
nengel@2 381 4+1*8, 5+1*8, 4+2*8, 5+2*8,
nengel@2 382 6+1*8, 7+1*8, 6+2*8, 7+2*8,
nengel@2 383 4+3*8, 5+3*8, 4+4*8, 5+4*8,
nengel@2 384 6+3*8, 7+3*8, 6+4*8, 7+4*8,
nengel@2 385 1+1*8, 2+1*8,
nengel@2 386 1+2*8, 2+2*8,
nengel@2 387 1+4*8, 2+4*8,
nengel@2 388 1+5*8, 2+5*8,
nengel@2 389 };
nengel@2 390
nengel@2 391 static void ff_h264_idct_add16_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 392 int i;
nengel@2 393 for(i=0; i<16; i++){
nengel@2 394 int nnz = nnzc[ scan8[i] ];
nengel@2 395 if(nnz){
nengel@2 396 if(nnz==1 && block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
nengel@2 397 else ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
nengel@2 398 }
nengel@2 399 }
nengel@2 400 }
nengel@2 401
nengel@2 402 static void ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 403 int i;
nengel@2 404 for(i=0; i<16; i++){
nengel@2 405 if(nnzc[ scan8[i] ]) ff_h264_idct_add_mmx (dst + block_offset[i], block + i*16, stride);
nengel@2 406 else if(block[i*16]) ff_h264_idct_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
nengel@2 407 }
nengel@2 408 }
nengel@2 409
nengel@2 410 static void ff_h264_idct8_add4_mmx2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 411 int i;
nengel@2 412 for(i=0; i<16; i+=4){
nengel@2 413 int nnz = nnzc[ scan8[i] ];
nengel@2 414 if(nnz){
nengel@2 415 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
nengel@2 416 else ff_h264_idct8_add_mmx (dst + block_offset[i], block + i*16, stride);
nengel@2 417 }
nengel@2 418 }
nengel@2 419 }
nengel@2 420
nengel@2 421 static void ff_h264_idct8_add4_sse2(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 422 int i;
nengel@2 423 for(i=0; i<16; i+=4){
nengel@2 424 int nnz = nnzc[ scan8[i] ];
nengel@2 425 if(nnz){
nengel@2 426 if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_mmx2(dst + block_offset[i], block + i*16, stride);
nengel@2 427 else ff_h264_idct8_add_sse2 (dst + block_offset[i], block + i*16, stride);
nengel@2 428 }
nengel@2 429 }
nengel@2 430 }
nengel@2 431
nengel@2 432 static void ff_h264_idct_add8_mmx2(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2 433 int i;
nengel@2 434 for(i=16; i<16+8; i++){
nengel@2 435 if(nnzc[ scan8[i] ])
nengel@2 436 ff_h264_idct_add_mmx (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2 437 else if(block[i*16])
nengel@2 438 ff_h264_idct_dc_add_mmx2(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2 439 }
nengel@2 440 }
nengel@2 441
nengel@2 442 /***********************************/
nengel@2 443 /* deblocking */
nengel@2 444
nengel@2 445 // out: o = |x-y|>a
nengel@2 446 // clobbers: t
nengel@2 447 #define DIFF_GT_MMX(x,y,a,o,t)\
nengel@2 448 "movq "#y", "#t" \n\t"\
nengel@2 449 "movq "#x", "#o" \n\t"\
nengel@2 450 "psubusb "#x", "#t" \n\t"\
nengel@2 451 "psubusb "#y", "#o" \n\t"\
nengel@2 452 "por "#t", "#o" \n\t"\
nengel@2 453 "psubusb "#a", "#o" \n\t"
nengel@2 454
nengel@2 455 // out: o = |x-y|>a
nengel@2 456 // clobbers: t
nengel@2 457 #define DIFF_GT2_MMX(x,y,a,o,t)\
nengel@2 458 "movq "#y", "#t" \n\t"\
nengel@2 459 "movq "#x", "#o" \n\t"\
nengel@2 460 "psubusb "#x", "#t" \n\t"\
nengel@2 461 "psubusb "#y", "#o" \n\t"\
nengel@2 462 "psubusb "#a", "#t" \n\t"\
nengel@2 463 "psubusb "#a", "#o" \n\t"\
nengel@2 464 "pcmpeqb "#t", "#o" \n\t"\
nengel@2 465
nengel@2 466 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
nengel@2 467 // out: mm5=beta-1, mm7=mask
nengel@2 468 // clobbers: mm4,mm6
nengel@2 469 #define H264_DEBLOCK_MASK(alpha1, beta1) \
nengel@2 470 "pshufw $0, "#alpha1", %%mm4 \n\t"\
nengel@2 471 "pshufw $0, "#beta1 ", %%mm5 \n\t"\
nengel@2 472 "packuswb %%mm4, %%mm4 \n\t"\
nengel@2 473 "packuswb %%mm5, %%mm5 \n\t"\
nengel@2 474 DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
nengel@2 475 DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
nengel@2 476 "por %%mm4, %%mm7 \n\t"\
nengel@2 477 DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
nengel@2 478 "por %%mm4, %%mm7 \n\t"\
nengel@2 479 "pxor %%mm6, %%mm6 \n\t"\
nengel@2 480 "pcmpeqb %%mm6, %%mm7 \n\t"
nengel@2 481
nengel@2 482 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
nengel@2 483 // out: mm1=p0' mm2=q0'
nengel@2 484 // clobbers: mm0,3-6
nengel@2 485 #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
nengel@2 486 "movq %%mm1 , %%mm5 \n\t"\
nengel@2 487 "pxor %%mm2 , %%mm5 \n\t" /* p0^q0*/\
nengel@2 488 "pand "#pb_01" , %%mm5 \n\t" /* (p0^q0)&1*/\
nengel@2 489 "pcmpeqb %%mm4 , %%mm4 \n\t"\
nengel@2 490 "pxor %%mm4 , %%mm3 \n\t"\
nengel@2 491 "pavgb %%mm0 , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\
nengel@2 492 "pavgb "MANGLE(ff_pb_3)" , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\
nengel@2 493 "pxor %%mm1 , %%mm4 \n\t"\
nengel@2 494 "pavgb %%mm2 , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\
nengel@2 495 "pavgb %%mm5 , %%mm3 \n\t"\
nengel@2 496 "paddusb %%mm4 , %%mm3 \n\t" /* d+128+33*/\
nengel@2 497 "movq "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\
nengel@2 498 "psubusb %%mm3 , %%mm6 \n\t"\
nengel@2 499 "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\
nengel@2 500 "pminub %%mm7 , %%mm6 \n\t"\
nengel@2 501 "pminub %%mm7 , %%mm3 \n\t"\
nengel@2 502 "psubusb %%mm6 , %%mm1 \n\t"\
nengel@2 503 "psubusb %%mm3 , %%mm2 \n\t"\
nengel@2 504 "paddusb %%mm3 , %%mm1 \n\t"\
nengel@2 505 "paddusb %%mm6 , %%mm2 \n\t"
nengel@2 506
nengel@2 507 // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
nengel@2 508 // out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
nengel@2 509 // clobbers: q2, tmp, tc0
nengel@2 510 #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
nengel@2 511 "movq %%mm1, "#tmp" \n\t"\
nengel@2 512 "pavgb %%mm2, "#tmp" \n\t"\
nengel@2 513 "pavgb "#tmp", "#q2" \n\t" /* avg(p2,avg(p0,q0)) */\
nengel@2 514 "pxor "q2addr", "#tmp" \n\t"\
nengel@2 515 "pand %9, "#tmp" \n\t" /* (p2^avg(p0,q0))&1 */\
nengel@2 516 "psubusb "#tmp", "#q2" \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
nengel@2 517 "movq "#p1", "#tmp" \n\t"\
nengel@2 518 "psubusb "#tc0", "#tmp" \n\t"\
nengel@2 519 "paddusb "#p1", "#tc0" \n\t"\
nengel@2 520 "pmaxub "#tmp", "#q2" \n\t"\
nengel@2 521 "pminub "#tc0", "#q2" \n\t"\
nengel@2 522 "movq "#q2", "q1addr" \n\t"
nengel@2 523
nengel@2 524 static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
nengel@2 525 {
nengel@2 526 DECLARE_ALIGNED(8, uint64_t, tmp0)[2];
nengel@2 527
nengel@2 528 __asm__ volatile(
nengel@2 529 "movq (%2,%4), %%mm0 \n\t" //p1
nengel@2 530 "movq (%2,%4,2), %%mm1 \n\t" //p0
nengel@2 531 "movq (%3), %%mm2 \n\t" //q0
nengel@2 532 "movq (%3,%4), %%mm3 \n\t" //q1
nengel@2 533 H264_DEBLOCK_MASK(%7, %8)
nengel@2 534
nengel@2 535 "movd %6, %%mm4 \n\t"
nengel@2 536 "punpcklbw %%mm4, %%mm4 \n\t"
nengel@2 537 "punpcklwd %%mm4, %%mm4 \n\t"
nengel@2 538 "pcmpeqb %%mm3, %%mm3 \n\t"
nengel@2 539 "movq %%mm4, %%mm6 \n\t"
nengel@2 540 "pcmpgtb %%mm3, %%mm4 \n\t"
nengel@2 541 "movq %%mm6, %1 \n\t"
nengel@2 542 "pand %%mm4, %%mm7 \n\t"
nengel@2 543 "movq %%mm7, %0 \n\t"
nengel@2 544
nengel@2 545 /* filter p1 */
nengel@2 546 "movq (%2), %%mm3 \n\t" //p2
nengel@2 547 DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
nengel@2 548 "pand %%mm7, %%mm6 \n\t" // mask & |p2-p0|<beta
nengel@2 549 "pand %1, %%mm7 \n\t" // mask & tc0
nengel@2 550 "movq %%mm7, %%mm4 \n\t"
nengel@2 551 "psubb %%mm6, %%mm7 \n\t"
nengel@2 552 "pand %%mm4, %%mm6 \n\t" // mask & |p2-p0|<beta & tc0
nengel@2 553 H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%2)", "(%2,%4)", %%mm6, %%mm4)
nengel@2 554
nengel@2 555 /* filter q1 */
nengel@2 556 "movq (%3,%4,2), %%mm4 \n\t" //q2
nengel@2 557 DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
nengel@2 558 "pand %0, %%mm6 \n\t"
nengel@2 559 "movq %1, %%mm5 \n\t" // can be merged with the and below but is slower then
nengel@2 560 "pand %%mm6, %%mm5 \n\t"
nengel@2 561 "psubb %%mm6, %%mm7 \n\t"
nengel@2 562 "movq (%3,%4), %%mm3 \n\t"
nengel@2 563 H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%3,%4,2)", "(%3,%4)", %%mm5, %%mm6)
nengel@2 564
nengel@2 565 /* filter p0, q0 */
nengel@2 566 H264_DEBLOCK_P0_Q0(%9, unused)
nengel@2 567 "movq %%mm1, (%2,%4,2) \n\t"
nengel@2 568 "movq %%mm2, (%3) \n\t"
nengel@2 569
nengel@2 570 : "=m"(tmp0[0]), "=m"(tmp0[1])
nengel@2 571 : "r"(pix-3*stride), "r"(pix), "r"((x86_reg)stride),
nengel@2 572 "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
nengel@2 573 "m"(ff_bone)
nengel@2 574 );
nengel@2 575 }
nengel@2 576
nengel@2 577 static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
nengel@2 578 {
nengel@2 579 if((tc0[0] & tc0[1]) >= 0)
nengel@2 580 h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
nengel@2 581 if((tc0[2] & tc0[3]) >= 0)
nengel@2 582 h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
nengel@2 583 }
nengel@2 584 static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
nengel@2 585 {
nengel@2 586 //FIXME: could cut some load/stores by merging transpose with filter
nengel@2 587 // also, it only needs to transpose 6x8
nengel@2 588 DECLARE_ALIGNED(8, uint8_t, trans)[8*8];
nengel@2 589 int i;
nengel@2 590 for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
nengel@2 591 if((tc0[0] & tc0[1]) < 0)
nengel@2 592 continue;
nengel@2 593 transpose4x4(trans, pix-4, 8, stride);
nengel@2 594 transpose4x4(trans +4*8, pix, 8, stride);
nengel@2 595 transpose4x4(trans+4, pix-4+4*stride, 8, stride);
nengel@2 596 transpose4x4(trans+4+4*8, pix +4*stride, 8, stride);
nengel@2 597 h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
nengel@2 598 transpose4x4(pix-2, trans +2*8, stride, 8);
nengel@2 599 transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
nengel@2 600 }
nengel@2 601 }
nengel@2 602
nengel@2 603 static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
nengel@2 604 {
nengel@2 605 __asm__ volatile(
nengel@2 606 "movq (%0), %%mm0 \n\t" //p1
nengel@2 607 "movq (%0,%2), %%mm1 \n\t" //p0
nengel@2 608 "movq (%1), %%mm2 \n\t" //q0
nengel@2 609 "movq (%1,%2), %%mm3 \n\t" //q1
nengel@2 610 H264_DEBLOCK_MASK(%4, %5)
nengel@2 611 "movd %3, %%mm6 \n\t"
nengel@2 612 "punpcklbw %%mm6, %%mm6 \n\t"
nengel@2 613 "pand %%mm6, %%mm7 \n\t" // mm7 = tc&mask
nengel@2 614 H264_DEBLOCK_P0_Q0(%6, %7)
nengel@2 615 "movq %%mm1, (%0,%2) \n\t"
nengel@2 616 "movq %%mm2, (%1) \n\t"
nengel@2 617
nengel@2 618 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
nengel@2 619 "r"(*(uint32_t*)tc0),
nengel@2 620 "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
nengel@2 621 );
nengel@2 622 }
nengel@2 623
nengel@2 624 static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
nengel@2 625 {
nengel@2 626 h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
nengel@2 627 }
nengel@2 628
nengel@2 629 static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
nengel@2 630 {
nengel@2 631 //FIXME: could cut some load/stores by merging transpose with filter
nengel@2 632 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
nengel@2 633 transpose4x4(trans, pix-2, 8, stride);
nengel@2 634 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
nengel@2 635 h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
nengel@2 636 transpose4x4(pix-2, trans, stride, 8);
nengel@2 637 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
nengel@2 638 }
nengel@2 639
nengel@2 640 // p0 = (p0 + q1 + 2*p1 + 2) >> 2
nengel@2 641 #define H264_FILTER_CHROMA4(p0, p1, q1, one) \
nengel@2 642 "movq "#p0", %%mm4 \n\t"\
nengel@2 643 "pxor "#q1", %%mm4 \n\t"\
nengel@2 644 "pand "#one", %%mm4 \n\t" /* mm4 = (p0^q1)&1 */\
nengel@2 645 "pavgb "#q1", "#p0" \n\t"\
nengel@2 646 "psubusb %%mm4, "#p0" \n\t"\
nengel@2 647 "pavgb "#p1", "#p0" \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
nengel@2 648
nengel@2 649 static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
nengel@2 650 {
nengel@2 651 __asm__ volatile(
nengel@2 652 "movq (%0), %%mm0 \n\t"
nengel@2 653 "movq (%0,%2), %%mm1 \n\t"
nengel@2 654 "movq (%1), %%mm2 \n\t"
nengel@2 655 "movq (%1,%2), %%mm3 \n\t"
nengel@2 656 H264_DEBLOCK_MASK(%3, %4)
nengel@2 657 "movq %%mm1, %%mm5 \n\t"
nengel@2 658 "movq %%mm2, %%mm6 \n\t"
nengel@2 659 H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
nengel@2 660 H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
nengel@2 661 "psubb %%mm5, %%mm1 \n\t"
nengel@2 662 "psubb %%mm6, %%mm2 \n\t"
nengel@2 663 "pand %%mm7, %%mm1 \n\t"
nengel@2 664 "pand %%mm7, %%mm2 \n\t"
nengel@2 665 "paddb %%mm5, %%mm1 \n\t"
nengel@2 666 "paddb %%mm6, %%mm2 \n\t"
nengel@2 667 "movq %%mm1, (%0,%2) \n\t"
nengel@2 668 "movq %%mm2, (%1) \n\t"
nengel@2 669 :: "r"(pix-2*stride), "r"(pix), "r"((x86_reg)stride),
nengel@2 670 "m"(alpha1), "m"(beta1), "m"(ff_bone)
nengel@2 671 );
nengel@2 672 }
nengel@2 673
nengel@2 674 static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
nengel@2 675 {
nengel@2 676 h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
nengel@2 677 }
nengel@2 678
nengel@2 679 static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
nengel@2 680 {
nengel@2 681 //FIXME: could cut some load/stores by merging transpose with filter
nengel@2 682 DECLARE_ALIGNED(8, uint8_t, trans)[8*4];
nengel@2 683 transpose4x4(trans, pix-2, 8, stride);
nengel@2 684 transpose4x4(trans+4, pix-2+4*stride, 8, stride);
nengel@2 685 h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
nengel@2 686 transpose4x4(pix-2, trans, stride, 8);
nengel@2 687 transpose4x4(pix-2+4*stride, trans+4, stride, 8);
nengel@2 688 }
nengel@2 689
nengel@2 690 static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
nengel@2 691 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
nengel@2 692 int dir;
nengel@2 693 __asm__ volatile(
nengel@2 694 "movq %0, %%mm7 \n"
nengel@2 695 "movq %1, %%mm6 \n"
nengel@2 696 ::"m"(ff_pb_1), "m"(ff_pb_3)
nengel@2 697 );
nengel@2 698 if(field)
nengel@2 699 __asm__ volatile(
nengel@2 700 "movq %0, %%mm6 \n"
nengel@2 701 ::"m"(ff_pb_3_1)
nengel@2 702 );
nengel@2 703 __asm__ volatile(
nengel@2 704 "movq %%mm6, %%mm5 \n"
nengel@2 705 "paddb %%mm5, %%mm5 \n"
nengel@2 706 :);
nengel@2 707
nengel@2 708 // could do a special case for dir==0 && edges==1, but it only reduces the
nengel@2 709 // average filter time by 1.2%
nengel@2 710 for( dir=1; dir>=0; dir-- ) {
nengel@2 711 const x86_reg d_idx = dir ? -8 : -1;
nengel@2 712 const int mask_mv = dir ? mask_mv1 : mask_mv0;
nengel@2 713 DECLARE_ALIGNED(8, const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
nengel@2 714 int b_idx, edge;
nengel@2 715 for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
nengel@2 716 __asm__ volatile(
nengel@2 717 "pand %0, %%mm0 \n\t"
nengel@2 718 ::"m"(mask_dir)
nengel@2 719 );
nengel@2 720 if(!(mask_mv & edge)) {
nengel@2 721 if(bidir) {
nengel@2 722 __asm__ volatile(
nengel@2 723 "movd (%1,%0), %%mm2 \n"
nengel@2 724 "punpckldq 40(%1,%0), %%mm2 \n" // { ref0[bn], ref1[bn] }
nengel@2 725 "pshufw $0x44, (%1), %%mm0 \n" // { ref0[b], ref0[b] }
nengel@2 726 "pshufw $0x44, 40(%1), %%mm1 \n" // { ref1[b], ref1[b] }
nengel@2 727 "pshufw $0x4E, %%mm2, %%mm3 \n"
nengel@2 728 "psubb %%mm2, %%mm0 \n" // { ref0[b]!=ref0[bn], ref0[b]!=ref1[bn] }
nengel@2 729 "psubb %%mm3, %%mm1 \n" // { ref1[b]!=ref1[bn], ref1[b]!=ref0[bn] }
nengel@2 730 "1: \n"
nengel@2 731 "por %%mm1, %%mm0 \n"
nengel@2 732 "movq (%2,%0,4), %%mm1 \n"
nengel@2 733 "movq 8(%2,%0,4), %%mm2 \n"
nengel@2 734 "movq %%mm1, %%mm3 \n"
nengel@2 735 "movq %%mm2, %%mm4 \n"
nengel@2 736 "psubw (%2), %%mm1 \n"
nengel@2 737 "psubw 8(%2), %%mm2 \n"
nengel@2 738 "psubw 160(%2), %%mm3 \n"
nengel@2 739 "psubw 168(%2), %%mm4 \n"
nengel@2 740 "packsswb %%mm2, %%mm1 \n"
nengel@2 741 "packsswb %%mm4, %%mm3 \n"
nengel@2 742 "paddb %%mm6, %%mm1 \n"
nengel@2 743 "paddb %%mm6, %%mm3 \n"
nengel@2 744 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
nengel@2 745 "psubusb %%mm5, %%mm3 \n"
nengel@2 746 "packsswb %%mm3, %%mm1 \n"
nengel@2 747 "add $40, %0 \n"
nengel@2 748 "cmp $40, %0 \n"
nengel@2 749 "jl 1b \n"
nengel@2 750 "sub $80, %0 \n"
nengel@2 751 "pshufw $0x4E, %%mm1, %%mm1 \n"
nengel@2 752 "por %%mm1, %%mm0 \n"
nengel@2 753 "pshufw $0x4E, %%mm0, %%mm1 \n"
nengel@2 754 "pminub %%mm1, %%mm0 \n"
nengel@2 755 ::"r"(d_idx),
nengel@2 756 "r"(ref[0]+b_idx),
nengel@2 757 "r"(mv[0]+b_idx)
nengel@2 758 );
nengel@2 759 } else {
nengel@2 760 __asm__ volatile(
nengel@2 761 "movd (%1), %%mm0 \n"
nengel@2 762 "psubb (%1,%0), %%mm0 \n" // ref[b] != ref[bn]
nengel@2 763 "movq (%2), %%mm1 \n"
nengel@2 764 "movq 8(%2), %%mm2 \n"
nengel@2 765 "psubw (%2,%0,4), %%mm1 \n"
nengel@2 766 "psubw 8(%2,%0,4), %%mm2 \n"
nengel@2 767 "packsswb %%mm2, %%mm1 \n"
nengel@2 768 "paddb %%mm6, %%mm1 \n"
nengel@2 769 "psubusb %%mm5, %%mm1 \n" // abs(mv[b] - mv[bn]) >= limit
nengel@2 770 "packsswb %%mm1, %%mm1 \n"
nengel@2 771 "por %%mm1, %%mm0 \n"
nengel@2 772 ::"r"(d_idx),
nengel@2 773 "r"(ref[0]+b_idx),
nengel@2 774 "r"(mv[0]+b_idx)
nengel@2 775 );
nengel@2 776 }
nengel@2 777 }
nengel@2 778 __asm__ volatile(
nengel@2 779 "movd %0, %%mm1 \n"
nengel@2 780 "por %1, %%mm1 \n" // nnz[b] || nnz[bn]
nengel@2 781 ::"m"(nnz[b_idx]),
nengel@2 782 "m"(nnz[b_idx+d_idx])
nengel@2 783 );
nengel@2 784 __asm__ volatile(
nengel@2 785 "pminub %%mm7, %%mm1 \n"
nengel@2 786 "pminub %%mm7, %%mm0 \n"
nengel@2 787 "psllw $1, %%mm1 \n"
nengel@2 788 "pxor %%mm2, %%mm2 \n"
nengel@2 789 "pmaxub %%mm0, %%mm1 \n"
nengel@2 790 "punpcklbw %%mm2, %%mm1 \n"
nengel@2 791 "movq %%mm1, %0 \n"
nengel@2 792 :"=m"(*bS[dir][edge])
nengel@2 793 ::"memory"
nengel@2 794 );
nengel@2 795 }
nengel@2 796 edges = 4;
nengel@2 797 step = 1;
nengel@2 798 }
nengel@2 799 __asm__ volatile(
nengel@2 800 "movq (%0), %%mm0 \n\t"
nengel@2 801 "movq 8(%0), %%mm1 \n\t"
nengel@2 802 "movq 16(%0), %%mm2 \n\t"
nengel@2 803 "movq 24(%0), %%mm3 \n\t"
nengel@2 804 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
nengel@2 805 "movq %%mm0, (%0) \n\t"
nengel@2 806 "movq %%mm3, 8(%0) \n\t"
nengel@2 807 "movq %%mm4, 16(%0) \n\t"
nengel@2 808 "movq %%mm2, 24(%0) \n\t"
nengel@2 809 ::"r"(bS[0])
nengel@2 810 :"memory"
nengel@2 811 );
nengel@2 812 }
nengel@2 813
nengel@2 814 /***********************************/
nengel@2 815 /* motion compensation */
nengel@2 816
nengel@2 817 #define QPEL_H264V_MM(A,B,C,D,E,F,OP,T,Z,d,q)\
nengel@2 818 "mov"#q" "#C", "#T" \n\t"\
nengel@2 819 "mov"#d" (%0), "#F" \n\t"\
nengel@2 820 "paddw "#D", "#T" \n\t"\
nengel@2 821 "psllw $2, "#T" \n\t"\
nengel@2 822 "psubw "#B", "#T" \n\t"\
nengel@2 823 "psubw "#E", "#T" \n\t"\
nengel@2 824 "punpcklbw "#Z", "#F" \n\t"\
nengel@2 825 "pmullw %4, "#T" \n\t"\
nengel@2 826 "paddw %5, "#A" \n\t"\
nengel@2 827 "add %2, %0 \n\t"\
nengel@2 828 "paddw "#F", "#A" \n\t"\
nengel@2 829 "paddw "#A", "#T" \n\t"\
nengel@2 830 "psraw $5, "#T" \n\t"\
nengel@2 831 "packuswb "#T", "#T" \n\t"\
nengel@2 832 OP(T, (%1), A, d)\
nengel@2 833 "add %3, %1 \n\t"
nengel@2 834
nengel@2 835 #define QPEL_H264HV_MM(A,B,C,D,E,F,OF,T,Z,d,q)\
nengel@2 836 "mov"#q" "#C", "#T" \n\t"\
nengel@2 837 "mov"#d" (%0), "#F" \n\t"\
nengel@2 838 "paddw "#D", "#T" \n\t"\
nengel@2 839 "psllw $2, "#T" \n\t"\
nengel@2 840 "paddw %4, "#A" \n\t"\
nengel@2 841 "psubw "#B", "#T" \n\t"\
nengel@2 842 "psubw "#E", "#T" \n\t"\
nengel@2 843 "punpcklbw "#Z", "#F" \n\t"\
nengel@2 844 "pmullw %3, "#T" \n\t"\
nengel@2 845 "paddw "#F", "#A" \n\t"\
nengel@2 846 "add %2, %0 \n\t"\
nengel@2 847 "paddw "#A", "#T" \n\t"\
nengel@2 848 "mov"#q" "#T", "#OF"(%1) \n\t"
nengel@2 849
nengel@2 850 #define QPEL_H264V(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%mm6,%%mm7,d,q)
nengel@2 851 #define QPEL_H264HV(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%mm6,%%mm7,d,q)
nengel@2 852 #define QPEL_H264V_XMM(A,B,C,D,E,F,OP) QPEL_H264V_MM(A,B,C,D,E,F,OP,%%xmm6,%%xmm7,q,dqa)
nengel@2 853 #define QPEL_H264HV_XMM(A,B,C,D,E,F,OF) QPEL_H264HV_MM(A,B,C,D,E,F,OF,%%xmm6,%%xmm7,q,dqa)
nengel@2 854
nengel@2 855
nengel@2 856 #define QPEL_H264(OPNAME, OP, MMX)\
nengel@2 857 \
nengel@2 858 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
nengel@2 859 int h=8;\
nengel@2 860 __asm__ volatile(\
nengel@2 861 "pxor %%mm7, %%mm7 \n\t"\
nengel@2 862 "movq %0, %%mm6 \n\t"\
nengel@2 863 :: "m"(ff_pw_5)\
nengel@2 864 );\
nengel@2 865 do{\
nengel@2 866 __asm__ volatile(\
nengel@2 867 "movq (%0), %%mm0 \n\t"\
nengel@2 868 "movq 1(%0), %%mm2 \n\t"\
nengel@2 869 "movq %%mm0, %%mm1 \n\t"\
nengel@2 870 "movq %%mm2, %%mm3 \n\t"\
nengel@2 871 "punpcklbw %%mm7, %%mm0 \n\t"\
nengel@2 872 "punpckhbw %%mm7, %%mm1 \n\t"\
nengel@2 873 "punpcklbw %%mm7, %%mm2 \n\t"\
nengel@2 874 "punpckhbw %%mm7, %%mm3 \n\t"\
nengel@2 875 "paddw %%mm2, %%mm0 \n\t"\
nengel@2 876 "paddw %%mm3, %%mm1 \n\t"\
nengel@2 877 "psllw $2, %%mm0 \n\t"\
nengel@2 878 "psllw $2, %%mm1 \n\t"\
nengel@2 879 "movq -1(%0), %%mm2 \n\t"\
nengel@2 880 "movq 2(%0), %%mm4 \n\t"\
nengel@2 881 "movq %%mm2, %%mm3 \n\t"\
nengel@2 882 "movq %%mm4, %%mm5 \n\t"\
nengel@2 883 "punpcklbw %%mm7, %%mm2 \n\t"\
nengel@2 884 "punpckhbw %%mm7, %%mm3 \n\t"\
nengel@2 885 "punpcklbw %%mm7, %%mm4 \n\t"\
nengel@2 886 "punpckhbw %%mm7, %%mm5 \n\t"\
nengel@2 887 "paddw %%mm4, %%mm2 \n\t"\
nengel@2 888 "paddw %%mm3, %%mm5 \n\t"\
nengel@2 889 "psubw %%mm2, %%mm0 \n\t"\
nengel@2 890 "psubw %%mm5, %%mm1 \n\t"\
nengel@2 891 "pmullw %%mm6, %%mm0 \n\t"\
nengel@2 892 "pmullw %%mm6, %%mm1 \n\t"\
nengel@2 893 "movd -2(%0), %%mm2 \n\t"\
nengel@2 894 "movd 7(%0), %%mm5 \n\t"\
nengel@2 895 "punpcklbw %%mm7, %%mm2 \n\t"\
nengel@2 896 "punpcklbw %%mm7, %%mm5 \n\t"\
nengel@2 897 "paddw %%mm3, %%mm2 \n\t"\
nengel@2 898 "paddw %%mm5, %%mm4 \n\t"\
nengel@2 899 "movq %5, %%mm5 \n\t"\
nengel@2 900 "paddw %%mm5, %%mm2 \n\t"\
nengel@2 901 "paddw %%mm5, %%mm4 \n\t"\
nengel@2 902 "paddw %%mm2, %%mm0 \n\t"\
nengel@2 903 "paddw %%mm4, %%mm1 \n\t"\
nengel@2 904 "psraw $5, %%mm0 \n\t"\
nengel@2 905 "psraw $5, %%mm1 \n\t"\
nengel@2 906 "movq (%2), %%mm4 \n\t"\
nengel@2 907 "packuswb %%mm1, %%mm0 \n\t"\
nengel@2 908 PAVGB" %%mm4, %%mm0 \n\t"\
nengel@2 909 OP(%%mm0, (%1),%%mm5, q)\
nengel@2 910 "add %4, %0 \n\t"\
nengel@2 911 "add %4, %1 \n\t"\
nengel@2 912 "add %3, %2 \n\t"\
nengel@2 913 : "+a"(src), "+c"(dst), "+d"(src2)\
nengel@2 914 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
nengel@2 915 "m"(ff_pw_16)\
nengel@2 916 : "memory"\
nengel@2 917 );\
nengel@2 918 }while(--h);\
nengel@2 919 }\
nengel@2 920 \
nengel@2 921 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
nengel@2 922 int w = size>>4;\
nengel@2 923 do{\
nengel@2 924 int h = size;\
nengel@2 925 __asm__ volatile(\
nengel@2 926 "1: \n\t"\
nengel@2 927 "movq (%0), %%mm0 \n\t"\
nengel@2 928 "movq 8(%0), %%mm3 \n\t"\
nengel@2 929 "movq 2(%0), %%mm1 \n\t"\
nengel@2 930 "movq 10(%0), %%mm4 \n\t"\
nengel@2 931 "paddw %%mm4, %%mm0 \n\t"\
nengel@2 932 "paddw %%mm3, %%mm1 \n\t"\
nengel@2 933 "paddw 18(%0), %%mm3 \n\t"\
nengel@2 934 "paddw 16(%0), %%mm4 \n\t"\
nengel@2 935 "movq 4(%0), %%mm2 \n\t"\
nengel@2 936 "movq 12(%0), %%mm5 \n\t"\
nengel@2 937 "paddw 6(%0), %%mm2 \n\t"\
nengel@2 938 "paddw 14(%0), %%mm5 \n\t"\
nengel@2 939 "psubw %%mm1, %%mm0 \n\t"\
nengel@2 940 "psubw %%mm4, %%mm3 \n\t"\
nengel@2 941 "psraw $2, %%mm0 \n\t"\
nengel@2 942 "psraw $2, %%mm3 \n\t"\
nengel@2 943 "psubw %%mm1, %%mm0 \n\t"\
nengel@2 944 "psubw %%mm4, %%mm3 \n\t"\
nengel@2 945 "paddsw %%mm2, %%mm0 \n\t"\
nengel@2 946 "paddsw %%mm5, %%mm3 \n\t"\
nengel@2 947 "psraw $2, %%mm0 \n\t"\
nengel@2 948 "psraw $2, %%mm3 \n\t"\
nengel@2 949 "paddw %%mm2, %%mm0 \n\t"\
nengel@2 950 "paddw %%mm5, %%mm3 \n\t"\
nengel@2 951 "psraw $6, %%mm0 \n\t"\
nengel@2 952 "psraw $6, %%mm3 \n\t"\
nengel@2 953 "packuswb %%mm3, %%mm0 \n\t"\
nengel@2 954 OP(%%mm0, (%1),%%mm7, q)\
nengel@2 955 "add $48, %0 \n\t"\
nengel@2 956 "add %3, %1 \n\t"\
nengel@2 957 "decl %2 \n\t"\
nengel@2 958 " jnz 1b \n\t"\
nengel@2 959 : "+a"(tmp), "+c"(dst), "+g"(h)\
nengel@2 960 : "S"((x86_reg)dstStride)\
nengel@2 961 : "memory"\
nengel@2 962 );\
nengel@2 963 tmp += 8 - size*24;\
nengel@2 964 dst += 8 - size*dstStride;\
nengel@2 965 }while(w--);\
nengel@2 966 }\
nengel@2 967 \
nengel@2 968 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
nengel@2 969 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
nengel@2 970 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
nengel@2 971 src += 8*dstStride;\
nengel@2 972 dst += 8*dstStride;\
nengel@2 973 src2 += 8*src2Stride;\
nengel@2 974 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
nengel@2 975 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
nengel@2 976 }\
nengel@2 977 static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
nengel@2 978 {\
nengel@2 979 do{\
nengel@2 980 __asm__ volatile(\
nengel@2 981 "movq (%1), %%mm0 \n\t"\
nengel@2 982 "movq 8(%1), %%mm1 \n\t"\
nengel@2 983 "movq 48(%1), %%mm2 \n\t"\
nengel@2 984 "movq 8+48(%1), %%mm3 \n\t"\
nengel@2 985 "psraw $5, %%mm0 \n\t"\
nengel@2 986 "psraw $5, %%mm1 \n\t"\
nengel@2 987 "psraw $5, %%mm2 \n\t"\
nengel@2 988 "psraw $5, %%mm3 \n\t"\
nengel@2 989 "packuswb %%mm1, %%mm0 \n\t"\
nengel@2 990 "packuswb %%mm3, %%mm2 \n\t"\
nengel@2 991 PAVGB" (%0), %%mm0 \n\t"\
nengel@2 992 PAVGB" (%0,%3), %%mm2 \n\t"\
nengel@2 993 OP(%%mm0, (%2), %%mm5, q)\
nengel@2 994 OP(%%mm2, (%2,%4), %%mm5, q)\
nengel@2 995 ::"a"(src8), "c"(src16), "d"(dst),\
nengel@2 996 "r"((x86_reg)src8Stride), "r"((x86_reg)dstStride)\
nengel@2 997 :"memory");\
nengel@2 998 src8 += 2L*src8Stride;\
nengel@2 999 src16 += 48;\
nengel@2 1000 dst += 2L*dstStride;\
nengel@2 1001 }while(h-=2);\
nengel@2 1002 }\
nengel@2 1003 static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
nengel@2 1004 {\
nengel@2 1005 OPNAME ## pixels8_l2_shift5_ ## MMX(dst , src16 , src8 , dstStride, src8Stride, h);\
nengel@2 1006 OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
nengel@2 1007 }\
nengel@2 1008
nengel@2 1009
nengel@2 1010 #if ARCH_X86_64
nengel@2 1011 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
nengel@2 1012 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
nengel@2 1013 int h=16;\
nengel@2 1014 __asm__ volatile(\
nengel@2 1015 "pxor %%xmm15, %%xmm15 \n\t"\
nengel@2 1016 "movdqa %6, %%xmm14 \n\t"\
nengel@2 1017 "movdqa %7, %%xmm13 \n\t"\
nengel@2 1018 "1: \n\t"\
nengel@2 1019 "lddqu 6(%0), %%xmm1 \n\t"\
nengel@2 1020 "lddqu -2(%0), %%xmm7 \n\t"\
nengel@2 1021 "movdqa %%xmm1, %%xmm0 \n\t"\
nengel@2 1022 "punpckhbw %%xmm15, %%xmm1 \n\t"\
nengel@2 1023 "punpcklbw %%xmm15, %%xmm0 \n\t"\
nengel@2 1024 "punpcklbw %%xmm15, %%xmm7 \n\t"\
nengel@2 1025 "movdqa %%xmm1, %%xmm2 \n\t"\
nengel@2 1026 "movdqa %%xmm0, %%xmm6 \n\t"\
nengel@2 1027 "movdqa %%xmm1, %%xmm3 \n\t"\
nengel@2 1028 "movdqa %%xmm0, %%xmm8 \n\t"\
nengel@2 1029 "movdqa %%xmm1, %%xmm4 \n\t"\
nengel@2 1030 "movdqa %%xmm0, %%xmm9 \n\t"\
nengel@2 1031 "movdqa %%xmm0, %%xmm12 \n\t"\
nengel@2 1032 "movdqa %%xmm1, %%xmm11 \n\t"\
nengel@2 1033 "palignr $10,%%xmm0, %%xmm11\n\t"\
nengel@2 1034 "palignr $10,%%xmm7, %%xmm12\n\t"\
nengel@2 1035 "palignr $2, %%xmm0, %%xmm4 \n\t"\
nengel@2 1036 "palignr $2, %%xmm7, %%xmm9 \n\t"\
nengel@2 1037 "palignr $4, %%xmm0, %%xmm3 \n\t"\
nengel@2 1038 "palignr $4, %%xmm7, %%xmm8 \n\t"\
nengel@2 1039 "palignr $6, %%xmm0, %%xmm2 \n\t"\
nengel@2 1040 "palignr $6, %%xmm7, %%xmm6 \n\t"\
nengel@2 1041 "paddw %%xmm0 ,%%xmm11 \n\t"\
nengel@2 1042 "palignr $8, %%xmm0, %%xmm1 \n\t"\
nengel@2 1043 "palignr $8, %%xmm7, %%xmm0 \n\t"\
nengel@2 1044 "paddw %%xmm12,%%xmm7 \n\t"\
nengel@2 1045 "paddw %%xmm3, %%xmm2 \n\t"\
nengel@2 1046 "paddw %%xmm8, %%xmm6 \n\t"\
nengel@2 1047 "paddw %%xmm4, %%xmm1 \n\t"\
nengel@2 1048 "paddw %%xmm9, %%xmm0 \n\t"\
nengel@2 1049 "psllw $2, %%xmm2 \n\t"\
nengel@2 1050 "psllw $2, %%xmm6 \n\t"\
nengel@2 1051 "psubw %%xmm1, %%xmm2 \n\t"\
nengel@2 1052 "psubw %%xmm0, %%xmm6 \n\t"\
nengel@2 1053 "paddw %%xmm13,%%xmm11 \n\t"\
nengel@2 1054 "paddw %%xmm13,%%xmm7 \n\t"\
nengel@2 1055 "pmullw %%xmm14,%%xmm2 \n\t"\
nengel@2 1056 "pmullw %%xmm14,%%xmm6 \n\t"\
nengel@2 1057 "lddqu (%2), %%xmm3 \n\t"\
nengel@2 1058 "paddw %%xmm11,%%xmm2 \n\t"\
nengel@2 1059 "paddw %%xmm7, %%xmm6 \n\t"\
nengel@2 1060 "psraw $5, %%xmm2 \n\t"\
nengel@2 1061 "psraw $5, %%xmm6 \n\t"\
nengel@2 1062 "packuswb %%xmm2,%%xmm6 \n\t"\
nengel@2 1063 "pavgb %%xmm3, %%xmm6 \n\t"\
nengel@2 1064 OP(%%xmm6, (%1), %%xmm4, dqa)\
nengel@2 1065 "add %5, %0 \n\t"\
nengel@2 1066 "add %5, %1 \n\t"\
nengel@2 1067 "add %4, %2 \n\t"\
nengel@2 1068 "decl %3 \n\t"\
nengel@2 1069 "jg 1b \n\t"\
nengel@2 1070 : "+a"(src), "+c"(dst), "+d"(src2), "+g"(h)\
nengel@2 1071 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
nengel@2 1072 "m"(ff_pw_5), "m"(ff_pw_16)\
nengel@2 1073 : "memory"\
nengel@2 1074 );\
nengel@2 1075 }
nengel@2 1076 #else // ARCH_X86_64
nengel@2 1077 #define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
nengel@2 1078 static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
nengel@2 1079 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
nengel@2 1080 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
nengel@2 1081 src += 8*dstStride;\
nengel@2 1082 dst += 8*dstStride;\
nengel@2 1083 src2 += 8*src2Stride;\
nengel@2 1084 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst , src , src2 , dstStride, src2Stride);\
nengel@2 1085 OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
nengel@2 1086 }
nengel@2 1087 #endif // ARCH_X86_64
nengel@2 1088
nengel@2 1089 #define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
nengel@2 1090 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
nengel@2 1091 int h=8;\
nengel@2 1092 __asm__ volatile(\
nengel@2 1093 "pxor %%xmm7, %%xmm7 \n\t"\
nengel@2 1094 "movdqa %0, %%xmm6 \n\t"\
nengel@2 1095 :: "m"(ff_pw_5)\
nengel@2 1096 );\
nengel@2 1097 do{\
nengel@2 1098 __asm__ volatile(\
nengel@2 1099 "lddqu -2(%0), %%xmm1 \n\t"\
nengel@2 1100 "movdqa %%xmm1, %%xmm0 \n\t"\
nengel@2 1101 "punpckhbw %%xmm7, %%xmm1 \n\t"\
nengel@2 1102 "punpcklbw %%xmm7, %%xmm0 \n\t"\
nengel@2 1103 "movdqa %%xmm1, %%xmm2 \n\t"\
nengel@2 1104 "movdqa %%xmm1, %%xmm3 \n\t"\
nengel@2 1105 "movdqa %%xmm1, %%xmm4 \n\t"\
nengel@2 1106 "movdqa %%xmm1, %%xmm5 \n\t"\
nengel@2 1107 "palignr $2, %%xmm0, %%xmm4 \n\t"\
nengel@2 1108 "palignr $4, %%xmm0, %%xmm3 \n\t"\
nengel@2 1109 "palignr $6, %%xmm0, %%xmm2 \n\t"\
nengel@2 1110 "palignr $8, %%xmm0, %%xmm1 \n\t"\
nengel@2 1111 "palignr $10,%%xmm0, %%xmm5 \n\t"\
nengel@2 1112 "paddw %%xmm5, %%xmm0 \n\t"\
nengel@2 1113 "paddw %%xmm3, %%xmm2 \n\t"\
nengel@2 1114 "paddw %%xmm4, %%xmm1 \n\t"\
nengel@2 1115 "psllw $2, %%xmm2 \n\t"\
nengel@2 1116 "movq (%2), %%xmm3 \n\t"\
nengel@2 1117 "psubw %%xmm1, %%xmm2 \n\t"\
nengel@2 1118 "paddw %5, %%xmm0 \n\t"\
nengel@2 1119 "pmullw %%xmm6, %%xmm2 \n\t"\
nengel@2 1120 "paddw %%xmm0, %%xmm2 \n\t"\
nengel@2 1121 "psraw $5, %%xmm2 \n\t"\
nengel@2 1122 "packuswb %%xmm2, %%xmm2 \n\t"\
nengel@2 1123 "pavgb %%xmm3, %%xmm2 \n\t"\
nengel@2 1124 OP(%%xmm2, (%1), %%xmm4, q)\
nengel@2 1125 "add %4, %0 \n\t"\
nengel@2 1126 "add %4, %1 \n\t"\
nengel@2 1127 "add %3, %2 \n\t"\
nengel@2 1128 : "+a"(src), "+c"(dst), "+d"(src2)\
nengel@2 1129 : "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\
nengel@2 1130 "m"(ff_pw_16)\
nengel@2 1131 : "memory"\
nengel@2 1132 );\
nengel@2 1133 }while(--h);\
nengel@2 1134 }\
nengel@2 1135 QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
nengel@2 1136 \
nengel@2 1137 static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
nengel@2 1138 int h=8;\
nengel@2 1139 __asm__ volatile(\
nengel@2 1140 "pxor %%xmm7, %%xmm7 \n\t"\
nengel@2 1141 "movdqa %5, %%xmm6 \n\t"\
nengel@2 1142 "1: \n\t"\
nengel@2 1143 "lddqu -2(%0), %%xmm1 \n\t"\
nengel@2 1144 "movdqa %%xmm1, %%xmm0 \n\t"\
nengel@2 1145 "punpckhbw %%xmm7, %%xmm1 \n\t"\
nengel@2 1146 "punpcklbw %%xmm7, %%xmm0 \n\t"\
nengel@2 1147 "movdqa %%xmm1, %%xmm2 \n\t"\
nengel@2 1148 "movdqa %%xmm1, %%xmm3 \n\t"\
nengel@2 1149 "movdqa %%xmm1, %%xmm4 \n\t"\
nengel@2 1150 "movdqa %%xmm1, %%xmm5 \n\t"\
nengel@2 1151 "palignr $2, %%xmm0, %%xmm4 \n\t"\
nengel@2 1152 "palignr $4, %%xmm0, %%xmm3 \n\t"\
nengel@2 1153 "palignr $6, %%xmm0, %%xmm2 \n\t"\
nengel@2 1154 "palignr $8, %%xmm0, %%xmm1 \n\t"\
nengel@2 1155 "palignr $10,%%xmm0, %%xmm5 \n\t"\
nengel@2 1156 "paddw %%xmm5, %%xmm0 \n\t"\
nengel@2 1157 "paddw %%xmm3, %%xmm2 \n\t"\
nengel@2 1158 "paddw %%xmm4, %%xmm1 \n\t"\
nengel@2 1159 "psllw $2, %%xmm2 \n\t"\
nengel@2 1160 "psubw %%xmm1, %%xmm2 \n\t"\
nengel@2 1161 "paddw %6, %%xmm0 \n\t"\
nengel@2 1162 "pmullw %%xmm6, %%xmm2 \n\t"\
nengel@2 1163 "paddw %%xmm0, %%xmm2 \n\t"\
nengel@2 1164 "psraw $5, %%xmm2 \n\t"\
nengel@2 1165 "packuswb %%xmm2, %%xmm2 \n\t"\
nengel@2 1166 OP(%%xmm2, (%1), %%xmm4, q)\
nengel@2 1167 "add %3, %0 \n\t"\
nengel@2 1168 "add %4, %1 \n\t"\
nengel@2 1169 "decl %2 \n\t"\
nengel@2 1170 " jnz 1b \n\t"\
nengel@2 1171 : "+a"(src), "+c"(dst), "+g"(h)\
nengel@2 1172 : "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\
nengel@2 1173 "m"(ff_pw_5), "m"(ff_pw_16)\
nengel@2 1174 : "memory"\
nengel@2 1175 );\
nengel@2 1176 }\
nengel@2 1177 static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
nengel@2 1178 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
nengel@2 1179 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
nengel@2 1180 src += 8*srcStride;\
nengel@2 1181 dst += 8*dstStride;\
nengel@2 1182 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst , src , dstStride, srcStride);\
nengel@2 1183 OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
nengel@2 1184 }\
nengel@2 1185
nengel@2 1186 #define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
nengel@2 1187 static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
nengel@2 1188 src -= 2*srcStride;\
nengel@2 1189 \
nengel@2 1190 __asm__ volatile(\
nengel@2 1191 "pxor %%xmm7, %%xmm7 \n\t"\
nengel@2 1192 "movq (%0), %%xmm0 \n\t"\
nengel@2 1193 "add %2, %0 \n\t"\
nengel@2 1194 "movq (%0), %%xmm1 \n\t"\
nengel@2 1195 "add %2, %0 \n\t"\
nengel@2 1196 "movq (%0), %%xmm2 \n\t"\
nengel@2 1197 "add %2, %0 \n\t"\
nengel@2 1198 "movq (%0), %%xmm3 \n\t"\
nengel@2 1199 "add %2, %0 \n\t"\
nengel@2 1200 "movq (%0), %%xmm4 \n\t"\
nengel@2 1201 "add %2, %0 \n\t"\
nengel@2 1202 "punpcklbw %%xmm7, %%xmm0 \n\t"\
nengel@2 1203 "punpcklbw %%xmm7, %%xmm1 \n\t"\
nengel@2 1204 "punpcklbw %%xmm7, %%xmm2 \n\t"\
nengel@2 1205 "punpcklbw %%xmm7, %%xmm3 \n\t"\
nengel@2 1206 "punpcklbw %%xmm7, %%xmm4 \n\t"\
nengel@2 1207 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
nengel@2 1208 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
nengel@2 1209 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
nengel@2 1210 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
nengel@2 1211 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
nengel@2 1212 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
nengel@2 1213 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
nengel@2 1214 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
nengel@2 1215 \
nengel@2 1216 : "+a"(src), "+c"(dst)\
nengel@2 1217 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
nengel@2 1218 : "memory"\
nengel@2 1219 );\
nengel@2 1220 if(h==16){\
nengel@2 1221 __asm__ volatile(\
nengel@2 1222 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
nengel@2 1223 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
nengel@2 1224 QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
nengel@2 1225 QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
nengel@2 1226 QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\
nengel@2 1227 QPEL_H264V_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, OP)\
nengel@2 1228 QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
nengel@2 1229 QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
nengel@2 1230 \
nengel@2 1231 : "+a"(src), "+c"(dst)\
nengel@2 1232 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
nengel@2 1233 : "memory"\
nengel@2 1234 );\
nengel@2 1235 }\
nengel@2 1236 }\
nengel@2 1237 static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
nengel@2 1238 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 8);\
nengel@2 1239 }\
nengel@2 1240 static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
nengel@2 1241 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst , src , dstStride, srcStride, 16);\
nengel@2 1242 OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
nengel@2 1243 }
nengel@2 1244
nengel@2 1245 static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
nengel@2 1246 int w = (size+8)>>3;
nengel@2 1247 src -= 2*srcStride+2;
nengel@2 1248 while(w--){
nengel@2 1249 __asm__ volatile(
nengel@2 1250 "pxor %%xmm7, %%xmm7 \n\t"
nengel@2 1251 "movq (%0), %%xmm0 \n\t"
nengel@2 1252 "add %2, %0 \n\t"
nengel@2 1253 "movq (%0), %%xmm1 \n\t"
nengel@2 1254 "add %2, %0 \n\t"
nengel@2 1255 "movq (%0), %%xmm2 \n\t"
nengel@2 1256 "add %2, %0 \n\t"
nengel@2 1257 "movq (%0), %%xmm3 \n\t"
nengel@2 1258 "add %2, %0 \n\t"
nengel@2 1259 "movq (%0), %%xmm4 \n\t"
nengel@2 1260 "add %2, %0 \n\t"
nengel@2 1261 "punpcklbw %%xmm7, %%xmm0 \n\t"
nengel@2 1262 "punpcklbw %%xmm7, %%xmm1 \n\t"
nengel@2 1263 "punpcklbw %%xmm7, %%xmm2 \n\t"
nengel@2 1264 "punpcklbw %%xmm7, %%xmm3 \n\t"
nengel@2 1265 "punpcklbw %%xmm7, %%xmm4 \n\t"
nengel@2 1266 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 0*48)
nengel@2 1267 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 1*48)
nengel@2 1268 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 2*48)
nengel@2 1269 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 3*48)
nengel@2 1270 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 4*48)
nengel@2 1271 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 5*48)
nengel@2 1272 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 6*48)
nengel@2 1273 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 7*48)
nengel@2 1274 : "+a"(src)
nengel@2 1275 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
nengel@2 1276 : "memory"
nengel@2 1277 );
nengel@2 1278 if(size==16){
nengel@2 1279 __asm__ volatile(
nengel@2 1280 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 8*48)
nengel@2 1281 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 9*48)
nengel@2 1282 QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
nengel@2 1283 QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
nengel@2 1284 QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)
nengel@2 1285 QPEL_H264HV_XMM(%%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, 13*48)
nengel@2 1286 QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, 14*48)
nengel@2 1287 QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, 15*48)
nengel@2 1288 : "+a"(src)
nengel@2 1289 : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
nengel@2 1290 : "memory"
nengel@2 1291 );
nengel@2 1292 }
nengel@2 1293 tmp += 8;
nengel@2 1294 src += 8 - (size+5)*srcStride;
nengel@2 1295 }
nengel@2 1296 }
nengel@2 1297
nengel@2 1298 #define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
nengel@2 1299 static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
nengel@2 1300 int h = size;\
nengel@2 1301 if(size == 16){\
nengel@2 1302 __asm__ volatile(\
nengel@2 1303 "1: \n\t"\
nengel@2 1304 "movdqa 32(%0), %%xmm4 \n\t"\
nengel@2 1305 "movdqa 16(%0), %%xmm5 \n\t"\
nengel@2 1306 "movdqa (%0), %%xmm7 \n\t"\
nengel@2 1307 "movdqa %%xmm4, %%xmm3 \n\t"\
nengel@2 1308 "movdqa %%xmm4, %%xmm2 \n\t"\
nengel@2 1309 "movdqa %%xmm4, %%xmm1 \n\t"\
nengel@2 1310 "movdqa %%xmm4, %%xmm0 \n\t"\
nengel@2 1311 "palignr $10, %%xmm5, %%xmm0 \n\t"\
nengel@2 1312 "palignr $8, %%xmm5, %%xmm1 \n\t"\
nengel@2 1313 "palignr $6, %%xmm5, %%xmm2 \n\t"\
nengel@2 1314 "palignr $4, %%xmm5, %%xmm3 \n\t"\
nengel@2 1315 "palignr $2, %%xmm5, %%xmm4 \n\t"\
nengel@2 1316 "paddw %%xmm5, %%xmm0 \n\t"\
nengel@2 1317 "paddw %%xmm4, %%xmm1 \n\t"\
nengel@2 1318 "paddw %%xmm3, %%xmm2 \n\t"\
nengel@2 1319 "movdqa %%xmm5, %%xmm6 \n\t"\
nengel@2 1320 "movdqa %%xmm5, %%xmm4 \n\t"\
nengel@2 1321 "movdqa %%xmm5, %%xmm3 \n\t"\
nengel@2 1322 "palignr $8, %%xmm7, %%xmm4 \n\t"\
nengel@2 1323 "palignr $2, %%xmm7, %%xmm6 \n\t"\
nengel@2 1324 "palignr $10, %%xmm7, %%xmm3 \n\t"\
nengel@2 1325 "paddw %%xmm6, %%xmm4 \n\t"\
nengel@2 1326 "movdqa %%xmm5, %%xmm6 \n\t"\
nengel@2 1327 "palignr $6, %%xmm7, %%xmm5 \n\t"\
nengel@2 1328 "palignr $4, %%xmm7, %%xmm6 \n\t"\
nengel@2 1329 "paddw %%xmm7, %%xmm3 \n\t"\
nengel@2 1330 "paddw %%xmm6, %%xmm5 \n\t"\
nengel@2 1331 \
nengel@2 1332 "psubw %%xmm1, %%xmm0 \n\t"\
nengel@2 1333 "psubw %%xmm4, %%xmm3 \n\t"\
nengel@2 1334 "psraw $2, %%xmm0 \n\t"\
nengel@2 1335 "psraw $2, %%xmm3 \n\t"\
nengel@2 1336 "psubw %%xmm1, %%xmm0 \n\t"\
nengel@2 1337 "psubw %%xmm4, %%xmm3 \n\t"\
nengel@2 1338 "paddw %%xmm2, %%xmm0 \n\t"\
nengel@2 1339 "paddw %%xmm5, %%xmm3 \n\t"\
nengel@2 1340 "psraw $2, %%xmm0 \n\t"\
nengel@2 1341 "psraw $2, %%xmm3 \n\t"\
nengel@2 1342 "paddw %%xmm2, %%xmm0 \n\t"\
nengel@2 1343 "paddw %%xmm5, %%xmm3 \n\t"\
nengel@2 1344 "psraw $6, %%xmm0 \n\t"\
nengel@2 1345 "psraw $6, %%xmm3 \n\t"\
nengel@2 1346 "packuswb %%xmm0, %%xmm3 \n\t"\
nengel@2 1347 OP(%%xmm3, (%1), %%xmm7, dqa)\
nengel@2 1348 "add $48, %0 \n\t"\
nengel@2 1349 "add %3, %1 \n\t"\
nengel@2 1350 "decl %2 \n\t"\
nengel@2 1351 " jnz 1b \n\t"\
nengel@2 1352 : "+a"(tmp), "+c"(dst), "+g"(h)\
nengel@2 1353 : "S"((x86_reg)dstStride)\
nengel@2 1354 : "memory"\
nengel@2 1355 );\
nengel@2 1356 }else{\
nengel@2 1357 __asm__ volatile(\
nengel@2 1358 "1: \n\t"\
nengel@2 1359 "movdqa 16(%0), %%xmm1 \n\t"\
nengel@2 1360 "movdqa (%0), %%xmm0 \n\t"\
nengel@2 1361 "movdqa %%xmm1, %%xmm2 \n\t"\
nengel@2 1362 "movdqa %%xmm1, %%xmm3 \n\t"\
nengel@2 1363 "movdqa %%xmm1, %%xmm4 \n\t"\
nengel@2 1364 "movdqa %%xmm1, %%xmm5 \n\t"\
nengel@2 1365 "palignr $10, %%xmm0, %%xmm5 \n\t"\
nengel@2 1366 "palignr $8, %%xmm0, %%xmm4 \n\t"\
nengel@2 1367 "palignr $6, %%xmm0, %%xmm3 \n\t"\
nengel@2 1368 "palignr $4, %%xmm0, %%xmm2 \n\t"\
nengel@2 1369 "palignr $2, %%xmm0, %%xmm1 \n\t"\
nengel@2 1370 "paddw %%xmm5, %%xmm0 \n\t"\
nengel@2 1371 "paddw %%xmm4, %%xmm1 \n\t"\
nengel@2 1372 "paddw %%xmm3, %%xmm2 \n\t"\
nengel@2 1373 "psubw %%xmm1, %%xmm0 \n\t"\
nengel@2 1374 "psraw $2, %%xmm0 \n\t"\
nengel@2 1375 "psubw %%xmm1, %%xmm0 \n\t"\
nengel@2 1376 "paddw %%xmm2, %%xmm0 \n\t"\
nengel@2 1377 "psraw $2, %%xmm0 \n\t"\
nengel@2 1378 "paddw %%xmm2, %%xmm0 \n\t"\
nengel@2 1379 "psraw $6, %%xmm0 \n\t"\
nengel@2 1380 "packuswb %%xmm0, %%xmm0 \n\t"\
nengel@2 1381 OP(%%xmm0, (%1), %%xmm7, q)\
nengel@2 1382 "add $48, %0 \n\t"\
nengel@2 1383 "add %3, %1 \n\t"\
nengel@2 1384 "decl %2 \n\t"\
nengel@2 1385 " jnz 1b \n\t"\
nengel@2 1386 : "+a"(tmp), "+c"(dst), "+g"(h)\
nengel@2 1387 : "S"((x86_reg)dstStride)\
nengel@2 1388 : "memory"\
nengel@2 1389 );\
nengel@2 1390 }\
nengel@2 1391 }
nengel@2 1392
nengel@2 1393 #define QPEL_H264_HV_XMM(OPNAME, OP, MMX)\
nengel@2 1394 static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\
nengel@2 1395 put_h264_qpel8or16_hv1_lowpass_sse2(tmp, src, tmpStride, srcStride, size);\
nengel@2 1396 OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(dst, tmp, dstStride, tmpStride, size);\
nengel@2 1397 }\
nengel@2 1398 static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
nengel@2 1399 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 8);\
nengel@2 1400 }\
nengel@2 1401 static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
nengel@2 1402 OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst, tmp, src, dstStride, tmpStride, srcStride, 16);\
nengel@2 1403 }\
nengel@2 1404
nengel@2 1405 #define put_pixels8_l2_sse2 put_pixels8_l2_mmx2
nengel@2 1406 #define avg_pixels8_l2_sse2 avg_pixels8_l2_mmx2
nengel@2 1407 #define put_pixels16_l2_sse2 put_pixels16_l2_mmx2
nengel@2 1408 #define avg_pixels16_l2_sse2 avg_pixels16_l2_mmx2
nengel@2 1409 #define put_pixels8_l2_ssse3 put_pixels8_l2_mmx2
nengel@2 1410 #define avg_pixels8_l2_ssse3 avg_pixels8_l2_mmx2
nengel@2 1411 #define put_pixels16_l2_ssse3 put_pixels16_l2_mmx2
nengel@2 1412 #define avg_pixels16_l2_ssse3 avg_pixels16_l2_mmx2
nengel@2 1413
nengel@2 1414 #define put_pixels8_l2_shift5_sse2 put_pixels8_l2_shift5_mmx2
nengel@2 1415 #define avg_pixels8_l2_shift5_sse2 avg_pixels8_l2_shift5_mmx2
nengel@2 1416 #define put_pixels16_l2_shift5_sse2 put_pixels16_l2_shift5_mmx2
nengel@2 1417 #define avg_pixels16_l2_shift5_sse2 avg_pixels16_l2_shift5_mmx2
nengel@2 1418 #define put_pixels8_l2_shift5_ssse3 put_pixels8_l2_shift5_mmx2
nengel@2 1419 #define avg_pixels8_l2_shift5_ssse3 avg_pixels8_l2_shift5_mmx2
nengel@2 1420 #define put_pixels16_l2_shift5_ssse3 put_pixels16_l2_shift5_mmx2
nengel@2 1421 #define avg_pixels16_l2_shift5_ssse3 avg_pixels16_l2_shift5_mmx2
nengel@2 1422
nengel@2 1423 #define put_h264_qpel8_h_lowpass_l2_sse2 put_h264_qpel8_h_lowpass_l2_mmx2
nengel@2 1424 #define avg_h264_qpel8_h_lowpass_l2_sse2 avg_h264_qpel8_h_lowpass_l2_mmx2
nengel@2 1425 #define put_h264_qpel16_h_lowpass_l2_sse2 put_h264_qpel16_h_lowpass_l2_mmx2
nengel@2 1426 #define avg_h264_qpel16_h_lowpass_l2_sse2 avg_h264_qpel16_h_lowpass_l2_mmx2
nengel@2 1427
nengel@2 1428 #define put_h264_qpel8_v_lowpass_ssse3 put_h264_qpel8_v_lowpass_sse2
nengel@2 1429 #define avg_h264_qpel8_v_lowpass_ssse3 avg_h264_qpel8_v_lowpass_sse2
nengel@2 1430 #define put_h264_qpel16_v_lowpass_ssse3 put_h264_qpel16_v_lowpass_sse2
nengel@2 1431 #define avg_h264_qpel16_v_lowpass_ssse3 avg_h264_qpel16_v_lowpass_sse2
nengel@2 1432
nengel@2 1433 #define put_h264_qpel8or16_hv2_lowpass_sse2 put_h264_qpel8or16_hv2_lowpass_mmx2
nengel@2 1434 #define avg_h264_qpel8or16_hv2_lowpass_sse2 avg_h264_qpel8or16_hv2_lowpass_mmx2
nengel@2 1435
nengel@2 1436 #define H264_MC(OPNAME, SIZE, MMX, ALIGN) \
nengel@2 1437 H264_MC_C(OPNAME, SIZE, MMX, ALIGN)\
nengel@2 1438 H264_MC_V(OPNAME, SIZE, MMX, ALIGN)\
nengel@2 1439 H264_MC_H(OPNAME, SIZE, MMX, ALIGN)\
nengel@2 1440 H264_MC_HV(OPNAME, SIZE, MMX, ALIGN)\
nengel@2 1441
nengel@2 1442 // static void put_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
nengel@2 1443 // put_pixels16_sse2(dst, src, stride, 16);
nengel@2 1444 // }
nengel@2 1445 // static void avg_h264_qpel16_mc00_sse2 (uint8_t *dst, uint8_t *src, int stride){
nengel@2 1446 // avg_pixels16_sse2(dst, src, stride, 16);
nengel@2 1447 // }
nengel@2 1448 #define put_h264_qpel8_mc00_sse2 put_h264_qpel8_mc00_mmx2
nengel@2 1449 #define avg_h264_qpel8_mc00_sse2 avg_h264_qpel8_mc00_mmx2
nengel@2 1450
nengel@2 1451 #define H264_MC_C(OPNAME, SIZE, MMX, ALIGN) \
nengel@2 1452 static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1453 OPNAME ## pixels ## SIZE ## _ ## MMX(dst, src, stride, SIZE);\
nengel@2 1454 }\
nengel@2 1455
nengel@2 1456 #define H264_MC_H(OPNAME, SIZE, MMX, ALIGN) \
nengel@2 1457 static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1458 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
nengel@2 1459 }\
nengel@2 1460 \
nengel@2 1461 static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1462 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
nengel@2 1463 }\
nengel@2 1464 \
nengel@2 1465 static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1466 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
nengel@2 1467 }\
nengel@2 1468
nengel@2 1469 #define H264_MC_V(OPNAME, SIZE, MMX, ALIGN) \
nengel@2 1470 static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1471 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1472 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
nengel@2 1473 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, temp, stride, stride, SIZE);\
nengel@2 1474 }\
nengel@2 1475 \
nengel@2 1476 static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1477 OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
nengel@2 1478 }\
nengel@2 1479 \
nengel@2 1480 static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1481 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1482 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
nengel@2 1483 OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, temp, stride, stride, SIZE);\
nengel@2 1484 }\
nengel@2 1485
nengel@2 1486 #define H264_MC_HV(OPNAME, SIZE, MMX, ALIGN) \
nengel@2 1487 static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1488 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1489 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
nengel@2 1490 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
nengel@2 1491 }\
nengel@2 1492 \
nengel@2 1493 static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1494 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1495 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
nengel@2 1496 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, temp, stride, SIZE);\
nengel@2 1497 }\
nengel@2 1498 \
nengel@2 1499 static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1500 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1501 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src, SIZE, stride);\
nengel@2 1502 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
nengel@2 1503 }\
nengel@2 1504 \
nengel@2 1505 static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1506 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*SIZE];\
nengel@2 1507 put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(temp, src+1, SIZE, stride);\
nengel@2 1508 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, temp, stride, SIZE);\
nengel@2 1509 }\
nengel@2 1510 \
nengel@2 1511 static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1512 DECLARE_ALIGNED(ALIGN, uint16_t, temp)[SIZE*(SIZE<8?12:24)];\
nengel@2 1513 OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, temp, src, stride, SIZE, stride);\
nengel@2 1514 }\
nengel@2 1515 \
nengel@2 1516 static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1517 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
nengel@2 1518 uint8_t * const halfHV= temp;\
nengel@2 1519 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
nengel@2 1520 assert(((int)temp & 7) == 0);\
nengel@2 1521 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
nengel@2 1522 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
nengel@2 1523 }\
nengel@2 1524 \
nengel@2 1525 static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1526 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
nengel@2 1527 uint8_t * const halfHV= temp;\
nengel@2 1528 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
nengel@2 1529 assert(((int)temp & 7) == 0);\
nengel@2 1530 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
nengel@2 1531 OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
nengel@2 1532 }\
nengel@2 1533 \
nengel@2 1534 static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1535 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
nengel@2 1536 uint8_t * const halfHV= temp;\
nengel@2 1537 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
nengel@2 1538 assert(((int)temp & 7) == 0);\
nengel@2 1539 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
nengel@2 1540 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
nengel@2 1541 }\
nengel@2 1542 \
nengel@2 1543 static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
nengel@2 1544 DECLARE_ALIGNED(ALIGN, uint8_t, temp)[SIZE*(SIZE<8?12:24)*2 + SIZE*SIZE];\
nengel@2 1545 uint8_t * const halfHV= temp;\
nengel@2 1546 int16_t * const halfV= (int16_t*)(temp + SIZE*SIZE);\
nengel@2 1547 assert(((int)temp & 7) == 0);\
nengel@2 1548 put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
nengel@2 1549 OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
nengel@2 1550 }\
nengel@2 1551
nengel@2 1552 #define H264_MC_4816(MMX)\
nengel@2 1553 H264_MC(put_, 4, MMX, 8)\
nengel@2 1554 H264_MC(put_, 8, MMX, 8)\
nengel@2 1555 H264_MC(put_, 16,MMX, 8)\
nengel@2 1556 H264_MC(avg_, 4, MMX, 8)\
nengel@2 1557 H264_MC(avg_, 8, MMX, 8)\
nengel@2 1558 H264_MC(avg_, 16,MMX, 8)\
nengel@2 1559
nengel@2 1560 #define H264_MC_816(QPEL, XMM)\
nengel@2 1561 QPEL(put_, 8, XMM, 16)\
nengel@2 1562 QPEL(put_, 16,XMM, 16)\
nengel@2 1563 QPEL(avg_, 8, XMM, 16)\
nengel@2 1564 QPEL(avg_, 16,XMM, 16)\
nengel@2 1565
nengel@2 1566
nengel@2 1567 #define AVG_3DNOW_OP(a,b,temp, size) \
nengel@2 1568 "mov" #size " " #b ", " #temp " \n\t"\
nengel@2 1569 "pavgusb " #temp ", " #a " \n\t"\
nengel@2 1570 "mov" #size " " #a ", " #b " \n\t"
nengel@2 1571 #define AVG_MMX2_OP(a,b,temp, size) \
nengel@2 1572 "mov" #size " " #b ", " #temp " \n\t"\
nengel@2 1573 "pavgb " #temp ", " #a " \n\t"\
nengel@2 1574 "mov" #size " " #a ", " #b " \n\t"
nengel@2 1575
nengel@2 1576 ///this does not get detected correctly, uncomment on AMD machine
nengel@2 1577 #ifdef HAVE_AMD3DNOW
nengel@2 1578 #define PAVGB "pavgusb"
nengel@2 1579 //QPEL_H264(put_, PUT_OP, 3dnow)
nengel@2 1580 //QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
nengel@2 1581 #undef PAVGB
nengel@2 1582 #endif
nengel@2 1583
nengel@2 1584 #define PAVGB "pavgb"
nengel@2 1585 QPEL_H264(put_, PUT_OP, mmx2)
nengel@2 1586 QPEL_H264(avg_, AVG_MMX2_OP, mmx2)
nengel@2 1587 QPEL_H264_V_XMM(put_, PUT_OP, sse2)
nengel@2 1588 QPEL_H264_V_XMM(avg_, AVG_MMX2_OP, sse2)
nengel@2 1589 QPEL_H264_HV_XMM(put_, PUT_OP, sse2)
nengel@2 1590 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, sse2)
nengel@2 1591 #if HAVE_SSSE3
nengel@2 1592 QPEL_H264_H_XMM(put_, PUT_OP, ssse3)
nengel@2 1593 QPEL_H264_H_XMM(avg_, AVG_MMX2_OP, ssse3)
nengel@2 1594 QPEL_H264_HV2_XMM(put_, PUT_OP, ssse3)
nengel@2 1595 QPEL_H264_HV2_XMM(avg_, AVG_MMX2_OP, ssse3)
nengel@2 1596 QPEL_H264_HV_XMM(put_, PUT_OP, ssse3)
nengel@2 1597 QPEL_H264_HV_XMM(avg_, AVG_MMX2_OP, ssse3)
nengel@2 1598 #endif
nengel@2 1599 #undef PAVGB
nengel@2 1600
nengel@2 1601 H264_MC_816(H264_MC_V, sse2)
nengel@2 1602 H264_MC_816(H264_MC_HV, sse2)
nengel@2 1603 #if HAVE_SSSE3
nengel@2 1604 H264_MC_816(H264_MC_H, ssse3)
nengel@2 1605 H264_MC_816(H264_MC_HV, ssse3)
nengel@2 1606 #endif
nengel@2 1607
nengel@2 1608 /* rnd interleaved with rnd div 8, use p+1 to access rnd div 8 */
nengel@2 1609 DECLARE_ALIGNED(8, static const uint64_t, h264_rnd_reg)[4] = {
nengel@2 1610 0x0020002000200020ULL, 0x0004000400040004ULL, 0x001C001C001C001CULL, 0x0003000300030003ULL
nengel@2 1611 };
nengel@2 1612
nengel@2 1613 #if HAVE_SSSE3
nengel@2 1614 #define AVG_OP(X)
nengel@2 1615 #undef H264_CHROMA_MC8_TMPL
nengel@2 1616 #undef H264_CHROMA_MC4_TMPL
nengel@2 1617 #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_ssse3
nengel@2 1618 #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_ssse3
nengel@2 1619 #define H264_CHROMA_MC8_MV0 put_pixels8_mmx
nengel@2 1620 #include "dsputil_h264_template_ssse3.c"
nengel@2 1621 static void put_h264_chroma_mc8_ssse3_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
nengel@2 1622 {
nengel@2 1623 put_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
nengel@2 1624 }
nengel@2 1625
nengel@2 1626 #undef AVG_OP
nengel@2 1627 #undef H264_CHROMA_MC8_TMPL
nengel@2 1628 #undef H264_CHROMA_MC4_TMPL
nengel@2 1629 #undef H264_CHROMA_MC8_MV0
nengel@2 1630 #define AVG_OP(X) X
nengel@2 1631 #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_ssse3
nengel@2 1632 #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_ssse3
nengel@2 1633 #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
nengel@2 1634 #include "dsputil_h264_template_ssse3.c"
nengel@2 1635 static void avg_h264_chroma_mc8_ssse3_rnd(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y)
nengel@2 1636 {
nengel@2 1637 avg_h264_chroma_mc8_ssse3(dst, src, stride, h, x, y, 1);
nengel@2 1638 }
nengel@2 1639 #undef AVG_OP
nengel@2 1640 #undef H264_CHROMA_MC8_TMPL
nengel@2 1641 #undef H264_CHROMA_MC4_TMPL
nengel@2 1642 #undef H264_CHROMA_MC8_MV0
nengel@2 1643 #endif
nengel@2 1644
nengel@2 1645 /***********************************/
nengel@2 1646 /* weighted prediction */
nengel@2 1647
nengel@2 1648 static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
nengel@2 1649 {
nengel@2 1650 int x, y;
nengel@2 1651 offset <<= log2_denom;
nengel@2 1652 offset += (1 << log2_denom) >> 1;
nengel@2 1653 __asm__ volatile(
nengel@2 1654 "movd %0, %%mm4 \n\t"
nengel@2 1655 "movd %1, %%mm5 \n\t"
nengel@2 1656 "movd %2, %%mm6 \n\t"
nengel@2 1657 "pshufw $0, %%mm4, %%mm4 \n\t"
nengel@2 1658 "pshufw $0, %%mm5, %%mm5 \n\t"
nengel@2 1659 "pxor %%mm7, %%mm7 \n\t"
nengel@2 1660 :: "g"(weight), "g"(offset), "g"(log2_denom)
nengel@2 1661 );
nengel@2 1662 for(y=0; y<h; y+=2){
nengel@2 1663 for(x=0; x<w; x+=4){
nengel@2 1664 __asm__ volatile(
nengel@2 1665 "movd %0, %%mm0 \n\t"
nengel@2 1666 "movd %1, %%mm1 \n\t"
nengel@2 1667 "punpcklbw %%mm7, %%mm0 \n\t"
nengel@2 1668 "punpcklbw %%mm7, %%mm1 \n\t"
nengel@2 1669 "pmullw %%mm4, %%mm0 \n\t"
nengel@2 1670 "pmullw %%mm4, %%mm1 \n\t"
nengel@2 1671 "paddsw %%mm5, %%mm0 \n\t"
nengel@2 1672 "paddsw %%mm5, %%mm1 \n\t"
nengel@2 1673 "psraw %%mm6, %%mm0 \n\t"
nengel@2 1674 "psraw %%mm6, %%mm1 \n\t"
nengel@2 1675 "packuswb %%mm7, %%mm0 \n\t"
nengel@2 1676 "packuswb %%mm7, %%mm1 \n\t"
nengel@2 1677 "movd %%mm0, %0 \n\t"
nengel@2 1678 "movd %%mm1, %1 \n\t"
nengel@2 1679 : "+m"(*(uint32_t*)(dst+x)),
nengel@2 1680 "+m"(*(uint32_t*)(dst+x+stride))
nengel@2 1681 );
nengel@2 1682 }
nengel@2 1683 dst += 2*stride;
nengel@2 1684 }
nengel@2 1685 }
nengel@2 1686
nengel@2 1687 static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset, int w, int h)
nengel@2 1688 {
nengel@2 1689 int x, y;
nengel@2 1690 offset = ((offset + 1) | 1) << log2_denom;
nengel@2 1691 __asm__ volatile(
nengel@2 1692 "movd %0, %%mm3 \n\t"
nengel@2 1693 "movd %1, %%mm4 \n\t"
nengel@2 1694 "movd %2, %%mm5 \n\t"
nengel@2 1695 "movd %3, %%mm6 \n\t"
nengel@2 1696 "pshufw $0, %%mm3, %%mm3 \n\t"
nengel@2 1697 "pshufw $0, %%mm4, %%mm4 \n\t"
nengel@2 1698 "pshufw $0, %%mm5, %%mm5 \n\t"
nengel@2 1699 "pxor %%mm7, %%mm7 \n\t"
nengel@2 1700 :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
nengel@2 1701 );
nengel@2 1702 for(y=0; y<h; y++){
nengel@2 1703 for(x=0; x<w; x+=4){
nengel@2 1704 __asm__ volatile(
nengel@2 1705 "movd %0, %%mm0 \n\t"
nengel@2 1706 "movd %1, %%mm1 \n\t"
nengel@2 1707 "punpcklbw %%mm7, %%mm0 \n\t"
nengel@2 1708 "punpcklbw %%mm7, %%mm1 \n\t"
nengel@2 1709 "pmullw %%mm3, %%mm0 \n\t"
nengel@2 1710 "pmullw %%mm4, %%mm1 \n\t"
nengel@2 1711 "paddsw %%mm1, %%mm0 \n\t"
nengel@2 1712 "paddsw %%mm5, %%mm0 \n\t"
nengel@2 1713 "psraw %%mm6, %%mm0 \n\t"
nengel@2 1714 "packuswb %%mm0, %%mm0 \n\t"
nengel@2 1715 "movd %%mm0, %0 \n\t"
nengel@2 1716 : "+m"(*(uint32_t*)(dst+x))
nengel@2 1717 : "m"(*(uint32_t*)(src+x))
nengel@2 1718 );
nengel@2 1719 }
nengel@2 1720 src += stride;
nengel@2 1721 dst += stride;
nengel@2 1722 }
nengel@2 1723 }
nengel@2 1724
nengel@2 1725 #define H264_WEIGHT(W,H) \
nengel@2 1726 static void ff_h264_biweight_ ## W ## x ## H ## _mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
nengel@2 1727 ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
nengel@2 1728 } \
nengel@2 1729 static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
nengel@2 1730 ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
nengel@2 1731 }
nengel@2 1732
nengel@2 1733 H264_WEIGHT(16,16)
nengel@2 1734 H264_WEIGHT(16, 8)
nengel@2 1735 H264_WEIGHT( 8,16)
nengel@2 1736 H264_WEIGHT( 8, 8)
nengel@2 1737 H264_WEIGHT( 8, 4)
nengel@2 1738 H264_WEIGHT( 4, 8)
nengel@2 1739 H264_WEIGHT( 4, 4)
nengel@2 1740 H264_WEIGHT( 4, 2)
nengel@2 1741