| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru> | ||
| 3 | * H.263, MPEG-1, MPEG-2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at> | ||
| 4 | * | ||
| 5 | * This file is part of FFmpeg. | ||
| 6 | * | ||
| 7 | * FFmpeg is free software; you can redistribute it and/or | ||
| 8 | * modify it under the terms of the GNU Lesser General Public | ||
| 9 | * License as published by the Free Software Foundation; either | ||
| 10 | * version 2.1 of the License, or (at your option) any later version. | ||
| 11 | * | ||
| 12 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 15 | * Lesser General Public License for more details. | ||
| 16 | * | ||
| 17 | * You should have received a copy of the GNU Lesser General Public | ||
| 18 | * License along with FFmpeg; if not, write to the Free Software | ||
| 19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 20 | */ | ||
| 21 | |||
| 22 | #include "libavutil/attributes.h" | ||
| 23 | #include "libavutil/cpu.h" | ||
| 24 | #include "libavutil/x86/asm.h" | ||
| 25 | #include "libavutil/x86/cpu.h" | ||
| 26 | #include "libavcodec/avcodec.h" | ||
| 27 | #include "libavcodec/mpegvideo.h" | ||
| 28 | #include "libavcodec/mpegvideodata.h" | ||
| 29 | #include "libavcodec/mpegvideo_unquantize.h" | ||
| 30 | |||
| 31 | #if HAVE_MMX_INLINE | ||
| 32 | |||
| 33 | 194460 | static void dct_unquantize_h263_intra_mmx(MpegEncContext *s, | |
| 34 | int16_t *block, int n, int qscale) | ||
| 35 | { | ||
| 36 | x86_reg level, qmul, qadd, nCoeffs; | ||
| 37 | |||
| 38 | 194460 | qmul = qscale << 1; | |
| 39 | |||
| 40 | av_assert2(s->block_last_index[n]>=0 || s->h263_aic); | ||
| 41 | |||
| 42 |
1/2✓ Branch 0 taken 194460 times.
✗ Branch 1 not taken.
|
194460 | if (!s->h263_aic) { |
| 43 |
2/2✓ Branch 0 taken 129640 times.
✓ Branch 1 taken 64820 times.
|
194460 | if (n < 4) |
| 44 | 129640 | level = block[0] * s->y_dc_scale; | |
| 45 | else | ||
| 46 | 64820 | level = block[0] * s->c_dc_scale; | |
| 47 | 194460 | qadd = (qscale - 1) | 1; | |
| 48 | }else{ | ||
| 49 | ✗ | qadd = 0; | |
| 50 | ✗ | level= block[0]; | |
| 51 | } | ||
| 52 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 194460 times.
|
194460 | if(s->ac_pred) |
| 53 | ✗ | nCoeffs=63; | |
| 54 | else | ||
| 55 | 194460 | nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]; | |
| 56 | |||
| 57 | 194460 | __asm__ volatile( | |
| 58 | "movd %1, %%mm6 \n\t" //qmul | ||
| 59 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 60 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 61 | "movd %2, %%mm5 \n\t" //qadd | ||
| 62 | "pxor %%mm7, %%mm7 \n\t" | ||
| 63 | "packssdw %%mm5, %%mm5 \n\t" | ||
| 64 | "packssdw %%mm5, %%mm5 \n\t" | ||
| 65 | "psubw %%mm5, %%mm7 \n\t" | ||
| 66 | "pxor %%mm4, %%mm4 \n\t" | ||
| 67 | ".p2align 4 \n\t" | ||
| 68 | "1: \n\t" | ||
| 69 | "movq (%0, %3), %%mm0 \n\t" | ||
| 70 | "movq 8(%0, %3), %%mm1 \n\t" | ||
| 71 | |||
| 72 | "pmullw %%mm6, %%mm0 \n\t" | ||
| 73 | "pmullw %%mm6, %%mm1 \n\t" | ||
| 74 | |||
| 75 | "movq (%0, %3), %%mm2 \n\t" | ||
| 76 | "movq 8(%0, %3), %%mm3 \n\t" | ||
| 77 | |||
| 78 | "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 79 | "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 80 | |||
| 81 | "pxor %%mm2, %%mm0 \n\t" | ||
| 82 | "pxor %%mm3, %%mm1 \n\t" | ||
| 83 | |||
| 84 | "paddw %%mm7, %%mm0 \n\t" | ||
| 85 | "paddw %%mm7, %%mm1 \n\t" | ||
| 86 | |||
| 87 | "pxor %%mm0, %%mm2 \n\t" | ||
| 88 | "pxor %%mm1, %%mm3 \n\t" | ||
| 89 | |||
| 90 | "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 91 | "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 92 | |||
| 93 | "pandn %%mm2, %%mm0 \n\t" | ||
| 94 | "pandn %%mm3, %%mm1 \n\t" | ||
| 95 | |||
| 96 | "movq %%mm0, (%0, %3) \n\t" | ||
| 97 | "movq %%mm1, 8(%0, %3) \n\t" | ||
| 98 | |||
| 99 | "add $16, %3 \n\t" | ||
| 100 | "jng 1b \n\t" | ||
| 101 | 194460 | ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs)) | |
| 102 | : "memory" | ||
| 103 | ); | ||
| 104 | 194460 | block[0]= level; | |
| 105 | 194460 | } | |
| 106 | |||
| 107 | |||
| 108 | 49919 | static void dct_unquantize_h263_inter_mmx(MpegEncContext *s, | |
| 109 | int16_t *block, int n, int qscale) | ||
| 110 | { | ||
| 111 | x86_reg qmul, qadd, nCoeffs; | ||
| 112 | |||
| 113 | 49919 | qmul = qscale << 1; | |
| 114 | 49919 | qadd = (qscale - 1) | 1; | |
| 115 | |||
| 116 | av_assert2(s->block_last_index[n]>=0 || s->h263_aic); | ||
| 117 | |||
| 118 | 49919 | nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; | |
| 119 | |||
| 120 | 49919 | __asm__ volatile( | |
| 121 | "movd %1, %%mm6 \n\t" //qmul | ||
| 122 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 123 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 124 | "movd %2, %%mm5 \n\t" //qadd | ||
| 125 | "pxor %%mm7, %%mm7 \n\t" | ||
| 126 | "packssdw %%mm5, %%mm5 \n\t" | ||
| 127 | "packssdw %%mm5, %%mm5 \n\t" | ||
| 128 | "psubw %%mm5, %%mm7 \n\t" | ||
| 129 | "pxor %%mm4, %%mm4 \n\t" | ||
| 130 | ".p2align 4 \n\t" | ||
| 131 | "1: \n\t" | ||
| 132 | "movq (%0, %3), %%mm0 \n\t" | ||
| 133 | "movq 8(%0, %3), %%mm1 \n\t" | ||
| 134 | |||
| 135 | "pmullw %%mm6, %%mm0 \n\t" | ||
| 136 | "pmullw %%mm6, %%mm1 \n\t" | ||
| 137 | |||
| 138 | "movq (%0, %3), %%mm2 \n\t" | ||
| 139 | "movq 8(%0, %3), %%mm3 \n\t" | ||
| 140 | |||
| 141 | "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 142 | "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 143 | |||
| 144 | "pxor %%mm2, %%mm0 \n\t" | ||
| 145 | "pxor %%mm3, %%mm1 \n\t" | ||
| 146 | |||
| 147 | "paddw %%mm7, %%mm0 \n\t" | ||
| 148 | "paddw %%mm7, %%mm1 \n\t" | ||
| 149 | |||
| 150 | "pxor %%mm0, %%mm2 \n\t" | ||
| 151 | "pxor %%mm1, %%mm3 \n\t" | ||
| 152 | |||
| 153 | "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 154 | "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 155 | |||
| 156 | "pandn %%mm2, %%mm0 \n\t" | ||
| 157 | "pandn %%mm3, %%mm1 \n\t" | ||
| 158 | |||
| 159 | "movq %%mm0, (%0, %3) \n\t" | ||
| 160 | "movq %%mm1, 8(%0, %3) \n\t" | ||
| 161 | |||
| 162 | "add $16, %3 \n\t" | ||
| 163 | "jng 1b \n\t" | ||
| 164 | 49919 | ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs)) | |
| 165 | : "memory" | ||
| 166 | ); | ||
| 167 | 49919 | } | |
| 168 | |||
| 169 | ✗ | static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s, | |
| 170 | int16_t *block, int n, int qscale) | ||
| 171 | { | ||
| 172 | x86_reg nCoeffs; | ||
| 173 | const uint16_t *quant_matrix; | ||
| 174 | int block0; | ||
| 175 | |||
| 176 | av_assert2(s->block_last_index[n]>=0); | ||
| 177 | |||
| 178 | ✗ | nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1; | |
| 179 | |||
| 180 | ✗ | if (n < 4) | |
| 181 | ✗ | block0 = block[0] * s->y_dc_scale; | |
| 182 | else | ||
| 183 | ✗ | block0 = block[0] * s->c_dc_scale; | |
| 184 | /* XXX: only MPEG-1 */ | ||
| 185 | ✗ | quant_matrix = s->intra_matrix; | |
| 186 | ✗ | __asm__ volatile( | |
| 187 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 188 | "psrlw $15, %%mm7 \n\t" | ||
| 189 | "movd %2, %%mm6 \n\t" | ||
| 190 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 191 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 192 | "mov %3, %%"FF_REG_a" \n\t" | ||
| 193 | ".p2align 4 \n\t" | ||
| 194 | "1: \n\t" | ||
| 195 | "movq (%0, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 196 | "movq 8(%0, %%"FF_REG_a"), %%mm1\n\t" | ||
| 197 | "movq (%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 198 | "movq 8(%1, %%"FF_REG_a"), %%mm5\n\t" | ||
| 199 | "pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i] | ||
| 200 | "pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i] | ||
| 201 | "pxor %%mm2, %%mm2 \n\t" | ||
| 202 | "pxor %%mm3, %%mm3 \n\t" | ||
| 203 | "pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 204 | "pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 205 | "pxor %%mm2, %%mm0 \n\t" | ||
| 206 | "pxor %%mm3, %%mm1 \n\t" | ||
| 207 | "psubw %%mm2, %%mm0 \n\t" // abs(block[i]) | ||
| 208 | "psubw %%mm3, %%mm1 \n\t" // abs(block[i]) | ||
| 209 | "pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*q | ||
| 210 | "pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*q | ||
| 211 | "pxor %%mm4, %%mm4 \n\t" | ||
| 212 | "pxor %%mm5, %%mm5 \n\t" // FIXME slow | ||
| 213 | "pcmpeqw (%0, %%"FF_REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 214 | "pcmpeqw 8(%0, %%"FF_REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0 | ||
| 215 | "psraw $3, %%mm0 \n\t" | ||
| 216 | "psraw $3, %%mm1 \n\t" | ||
| 217 | "psubw %%mm7, %%mm0 \n\t" | ||
| 218 | "psubw %%mm7, %%mm1 \n\t" | ||
| 219 | "por %%mm7, %%mm0 \n\t" | ||
| 220 | "por %%mm7, %%mm1 \n\t" | ||
| 221 | "pxor %%mm2, %%mm0 \n\t" | ||
| 222 | "pxor %%mm3, %%mm1 \n\t" | ||
| 223 | "psubw %%mm2, %%mm0 \n\t" | ||
| 224 | "psubw %%mm3, %%mm1 \n\t" | ||
| 225 | "pandn %%mm0, %%mm4 \n\t" | ||
| 226 | "pandn %%mm1, %%mm5 \n\t" | ||
| 227 | "movq %%mm4, (%0, %%"FF_REG_a") \n\t" | ||
| 228 | "movq %%mm5, 8(%0, %%"FF_REG_a")\n\t" | ||
| 229 | |||
| 230 | "add $16, %%"FF_REG_a" \n\t" | ||
| 231 | "js 1b \n\t" | ||
| 232 | ✗ | ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs) | |
| 233 | : "%"FF_REG_a, "memory" | ||
| 234 | ); | ||
| 235 | ✗ | block[0]= block0; | |
| 236 | ✗ | } | |
| 237 | |||
| 238 | ✗ | static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s, | |
| 239 | int16_t *block, int n, int qscale) | ||
| 240 | { | ||
| 241 | x86_reg nCoeffs; | ||
| 242 | const uint16_t *quant_matrix; | ||
| 243 | |||
| 244 | av_assert2(s->block_last_index[n]>=0); | ||
| 245 | |||
| 246 | ✗ | nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1; | |
| 247 | |||
| 248 | ✗ | quant_matrix = s->inter_matrix; | |
| 249 | ✗ | __asm__ volatile( | |
| 250 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 251 | "psrlw $15, %%mm7 \n\t" | ||
| 252 | "movd %2, %%mm6 \n\t" | ||
| 253 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 254 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 255 | "mov %3, %%"FF_REG_a" \n\t" | ||
| 256 | ".p2align 4 \n\t" | ||
| 257 | "1: \n\t" | ||
| 258 | "movq (%0, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 259 | "movq 8(%0, %%"FF_REG_a"), %%mm1\n\t" | ||
| 260 | "movq (%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 261 | "movq 8(%1, %%"FF_REG_a"), %%mm5\n\t" | ||
| 262 | "pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i] | ||
| 263 | "pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i] | ||
| 264 | "pxor %%mm2, %%mm2 \n\t" | ||
| 265 | "pxor %%mm3, %%mm3 \n\t" | ||
| 266 | "pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 267 | "pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 268 | "pxor %%mm2, %%mm0 \n\t" | ||
| 269 | "pxor %%mm3, %%mm1 \n\t" | ||
| 270 | "psubw %%mm2, %%mm0 \n\t" // abs(block[i]) | ||
| 271 | "psubw %%mm3, %%mm1 \n\t" // abs(block[i]) | ||
| 272 | "paddw %%mm0, %%mm0 \n\t" // abs(block[i])*2 | ||
| 273 | "paddw %%mm1, %%mm1 \n\t" // abs(block[i])*2 | ||
| 274 | "paddw %%mm7, %%mm0 \n\t" // abs(block[i])*2 + 1 | ||
| 275 | "paddw %%mm7, %%mm1 \n\t" // abs(block[i])*2 + 1 | ||
| 276 | "pmullw %%mm4, %%mm0 \n\t" // (abs(block[i])*2 + 1)*q | ||
| 277 | "pmullw %%mm5, %%mm1 \n\t" // (abs(block[i])*2 + 1)*q | ||
| 278 | "pxor %%mm4, %%mm4 \n\t" | ||
| 279 | "pxor %%mm5, %%mm5 \n\t" // FIXME slow | ||
| 280 | "pcmpeqw (%0, %%"FF_REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 281 | "pcmpeqw 8(%0, %%"FF_REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0 | ||
| 282 | "psraw $4, %%mm0 \n\t" | ||
| 283 | "psraw $4, %%mm1 \n\t" | ||
| 284 | "psubw %%mm7, %%mm0 \n\t" | ||
| 285 | "psubw %%mm7, %%mm1 \n\t" | ||
| 286 | "por %%mm7, %%mm0 \n\t" | ||
| 287 | "por %%mm7, %%mm1 \n\t" | ||
| 288 | "pxor %%mm2, %%mm0 \n\t" | ||
| 289 | "pxor %%mm3, %%mm1 \n\t" | ||
| 290 | "psubw %%mm2, %%mm0 \n\t" | ||
| 291 | "psubw %%mm3, %%mm1 \n\t" | ||
| 292 | "pandn %%mm0, %%mm4 \n\t" | ||
| 293 | "pandn %%mm1, %%mm5 \n\t" | ||
| 294 | "movq %%mm4, (%0, %%"FF_REG_a") \n\t" | ||
| 295 | "movq %%mm5, 8(%0, %%"FF_REG_a")\n\t" | ||
| 296 | |||
| 297 | "add $16, %%"FF_REG_a" \n\t" | ||
| 298 | "js 1b \n\t" | ||
| 299 | ✗ | ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs) | |
| 300 | : "%"FF_REG_a, "memory" | ||
| 301 | ); | ||
| 302 | ✗ | } | |
| 303 | |||
| 304 | ✗ | static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s, | |
| 305 | int16_t *block, int n, int qscale) | ||
| 306 | { | ||
| 307 | x86_reg nCoeffs; | ||
| 308 | const uint16_t *quant_matrix; | ||
| 309 | int block0; | ||
| 310 | |||
| 311 | av_assert2(s->block_last_index[n]>=0); | ||
| 312 | |||
| 313 | ✗ | if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale]; | |
| 314 | ✗ | else qscale <<= 1; | |
| 315 | |||
| 316 | ✗ | nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]; | |
| 317 | |||
| 318 | ✗ | if (n < 4) | |
| 319 | ✗ | block0 = block[0] * s->y_dc_scale; | |
| 320 | else | ||
| 321 | ✗ | block0 = block[0] * s->c_dc_scale; | |
| 322 | ✗ | quant_matrix = s->intra_matrix; | |
| 323 | ✗ | __asm__ volatile( | |
| 324 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 325 | "psrlw $15, %%mm7 \n\t" | ||
| 326 | "movd %2, %%mm6 \n\t" | ||
| 327 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 328 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 329 | "mov %3, %%"FF_REG_a" \n\t" | ||
| 330 | ".p2align 4 \n\t" | ||
| 331 | "1: \n\t" | ||
| 332 | "movq (%0, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 333 | "movq 8(%0, %%"FF_REG_a"), %%mm1\n\t" | ||
| 334 | "movq (%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 335 | "movq 8(%1, %%"FF_REG_a"), %%mm5\n\t" | ||
| 336 | "pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i] | ||
| 337 | "pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i] | ||
| 338 | "pxor %%mm2, %%mm2 \n\t" | ||
| 339 | "pxor %%mm3, %%mm3 \n\t" | ||
| 340 | "pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 341 | "pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 342 | "pxor %%mm2, %%mm0 \n\t" | ||
| 343 | "pxor %%mm3, %%mm1 \n\t" | ||
| 344 | "psubw %%mm2, %%mm0 \n\t" // abs(block[i]) | ||
| 345 | "psubw %%mm3, %%mm1 \n\t" // abs(block[i]) | ||
| 346 | "pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*q | ||
| 347 | "pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*q | ||
| 348 | "pxor %%mm4, %%mm4 \n\t" | ||
| 349 | "pxor %%mm5, %%mm5 \n\t" // FIXME slow | ||
| 350 | "pcmpeqw (%0, %%"FF_REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 351 | "pcmpeqw 8(%0, %%"FF_REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0 | ||
| 352 | "psraw $4, %%mm0 \n\t" | ||
| 353 | "psraw $4, %%mm1 \n\t" | ||
| 354 | "pxor %%mm2, %%mm0 \n\t" | ||
| 355 | "pxor %%mm3, %%mm1 \n\t" | ||
| 356 | "psubw %%mm2, %%mm0 \n\t" | ||
| 357 | "psubw %%mm3, %%mm1 \n\t" | ||
| 358 | "pandn %%mm0, %%mm4 \n\t" | ||
| 359 | "pandn %%mm1, %%mm5 \n\t" | ||
| 360 | "movq %%mm4, (%0, %%"FF_REG_a") \n\t" | ||
| 361 | "movq %%mm5, 8(%0, %%"FF_REG_a")\n\t" | ||
| 362 | |||
| 363 | "add $16, %%"FF_REG_a" \n\t" | ||
| 364 | "jng 1b \n\t" | ||
| 365 | ✗ | ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs) | |
| 366 | : "%"FF_REG_a, "memory" | ||
| 367 | ); | ||
| 368 | ✗ | block[0]= block0; | |
| 369 | //Note, we do not do mismatch control for intra as errors cannot accumulate | ||
| 370 | ✗ | } | |
| 371 | |||
| 372 | 5941 | static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s, | |
| 373 | int16_t *block, int n, int qscale) | ||
| 374 | { | ||
| 375 | x86_reg nCoeffs; | ||
| 376 | const uint16_t *quant_matrix; | ||
| 377 | |||
| 378 | av_assert2(s->block_last_index[n]>=0); | ||
| 379 | |||
| 380 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5941 times.
|
5941 | if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale]; |
| 381 | 5941 | else qscale <<= 1; | |
| 382 | |||
| 383 | 5941 | nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]; | |
| 384 | |||
| 385 | 5941 | quant_matrix = s->inter_matrix; | |
| 386 | 5941 | __asm__ volatile( | |
| 387 | "pcmpeqw %%mm7, %%mm7 \n\t" | ||
| 388 | "psrlq $48, %%mm7 \n\t" | ||
| 389 | "movd %2, %%mm6 \n\t" | ||
| 390 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 391 | "packssdw %%mm6, %%mm6 \n\t" | ||
| 392 | "mov %3, %%"FF_REG_a" \n\t" | ||
| 393 | ".p2align 4 \n\t" | ||
| 394 | "1: \n\t" | ||
| 395 | "movq (%0, %%"FF_REG_a"), %%mm0 \n\t" | ||
| 396 | "movq 8(%0, %%"FF_REG_a"), %%mm1\n\t" | ||
| 397 | "movq (%1, %%"FF_REG_a"), %%mm4 \n\t" | ||
| 398 | "movq 8(%1, %%"FF_REG_a"), %%mm5\n\t" | ||
| 399 | "pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i] | ||
| 400 | "pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i] | ||
| 401 | "pxor %%mm2, %%mm2 \n\t" | ||
| 402 | "pxor %%mm3, %%mm3 \n\t" | ||
| 403 | "pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 404 | "pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 | ||
| 405 | "pxor %%mm2, %%mm0 \n\t" | ||
| 406 | "pxor %%mm3, %%mm1 \n\t" | ||
| 407 | "psubw %%mm2, %%mm0 \n\t" // abs(block[i]) | ||
| 408 | "psubw %%mm3, %%mm1 \n\t" // abs(block[i]) | ||
| 409 | "paddw %%mm0, %%mm0 \n\t" // abs(block[i])*2 | ||
| 410 | "paddw %%mm1, %%mm1 \n\t" // abs(block[i])*2 | ||
| 411 | "pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*2*q | ||
| 412 | "pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*2*q | ||
| 413 | "paddw %%mm4, %%mm0 \n\t" // (abs(block[i])*2 + 1)*q | ||
| 414 | "paddw %%mm5, %%mm1 \n\t" // (abs(block[i])*2 + 1)*q | ||
| 415 | "pxor %%mm4, %%mm4 \n\t" | ||
| 416 | "pxor %%mm5, %%mm5 \n\t" // FIXME slow | ||
| 417 | "pcmpeqw (%0, %%"FF_REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0 | ||
| 418 | "pcmpeqw 8(%0, %%"FF_REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0 | ||
| 419 | "psrlw $5, %%mm0 \n\t" | ||
| 420 | "psrlw $5, %%mm1 \n\t" | ||
| 421 | "pxor %%mm2, %%mm0 \n\t" | ||
| 422 | "pxor %%mm3, %%mm1 \n\t" | ||
| 423 | "psubw %%mm2, %%mm0 \n\t" | ||
| 424 | "psubw %%mm3, %%mm1 \n\t" | ||
| 425 | "pandn %%mm0, %%mm4 \n\t" | ||
| 426 | "pandn %%mm1, %%mm5 \n\t" | ||
| 427 | "pxor %%mm4, %%mm7 \n\t" | ||
| 428 | "pxor %%mm5, %%mm7 \n\t" | ||
| 429 | "movq %%mm4, (%0, %%"FF_REG_a") \n\t" | ||
| 430 | "movq %%mm5, 8(%0, %%"FF_REG_a")\n\t" | ||
| 431 | |||
| 432 | "add $16, %%"FF_REG_a" \n\t" | ||
| 433 | "jng 1b \n\t" | ||
| 434 | "movd 124(%0, %3), %%mm0 \n\t" | ||
| 435 | "movq %%mm7, %%mm6 \n\t" | ||
| 436 | "psrlq $32, %%mm7 \n\t" | ||
| 437 | "pxor %%mm6, %%mm7 \n\t" | ||
| 438 | "movq %%mm7, %%mm6 \n\t" | ||
| 439 | "psrlq $16, %%mm7 \n\t" | ||
| 440 | "pxor %%mm6, %%mm7 \n\t" | ||
| 441 | "pslld $31, %%mm7 \n\t" | ||
| 442 | "psrlq $15, %%mm7 \n\t" | ||
| 443 | "pxor %%mm7, %%mm0 \n\t" | ||
| 444 | "movd %%mm0, 124(%0, %3) \n\t" | ||
| 445 | |||
| 446 | 5941 | ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs) | |
| 447 | : "%"FF_REG_a, "memory" | ||
| 448 | ); | ||
| 449 | 5941 | } | |
| 450 | |||
| 451 | #endif /* HAVE_MMX_INLINE */ | ||
| 452 | |||
| 453 | 683 | av_cold void ff_mpv_unquantize_init_x86(MPVUnquantDSPContext *s, int bitexact) | |
| 454 | { | ||
| 455 | #if HAVE_MMX_INLINE | ||
| 456 | 683 | int cpu_flags = av_get_cpu_flags(); | |
| 457 | |||
| 458 |
2/2✓ Branch 0 taken 86 times.
✓ Branch 1 taken 597 times.
|
683 | if (INLINE_MMX(cpu_flags)) { |
| 459 | 86 | s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx; | |
| 460 | 86 | s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx; | |
| 461 | 86 | s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx; | |
| 462 | 86 | s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx; | |
| 463 |
2/2✓ Branch 0 taken 81 times.
✓ Branch 1 taken 5 times.
|
86 | if (!bitexact) |
| 464 | 81 | s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx; | |
| 465 | 86 | s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx; | |
| 466 | } | ||
| 467 | #endif /* HAVE_MMX_INLINE */ | ||
| 468 | 683 | } | |
| 469 |