Parent Directory
| 
Revision Log
Revision 1795 - (view) (download)
| 1 : | edgomez | 1382 | ;/************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - 3dne Quantization/Dequantization - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2002-2003 Jaan Kalda | ||
| 7 : | ; * | ||
| 8 : | Isibaar | 1795 | ; * This program is free software ; you can r_EDIstribute it and/or modify |
| 9 : | edgomez | 1382 | ; * it under the terms of the GNU General Public License as published by |
| 10 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 11 : | ; * (at your option) any later version. | ||
| 12 : | ; * | ||
| 13 : | ; * This program is distributed in the hope that it will be useful, | ||
| 14 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 15 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 16 : | ; * GNU General Public License for more details. | ||
| 17 : | ; * | ||
| 18 : | ; * You should have received a copy of the GNU General Public License | ||
| 19 : | ; * along with this program ; if not, write to the Free Software | ||
| 20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | ; * | ||
| 22 : | Isibaar | 1795 | ; * $Id: quantize_h263_3dne.asm,v 1.9 2008-11-26 01:04:34 Isibaar Exp $ |
| 23 : | edgomez | 1382 | ; * |
| 24 : | ; *************************************************************************/ | ||
| 25 : | ; | ||
| 26 : | ; these 3dne functions are compatible with iSSE, but are optimized specifically for | ||
| 27 : | ; K7 pipelines | ||
| 28 : | |||
| 29 : | ; enable dequant saturate [-2048,2047], test purposes only. | ||
| 30 : | %define SATURATE | ||
| 31 : | |||
| 32 : | Isibaar | 1795 | %include "nasm.inc" |
| 33 : | edgomez | 1382 | |
| 34 : | ;============================================================================= | ||
| 35 : | ; Local data | ||
| 36 : | ;============================================================================= | ||
| 37 : | |||
| 38 : | Isibaar | 1795 | DATA |
| 39 : | edgomez | 1382 | |
| 40 : | Isibaar | 1795 | align SECTION_ALIGN |
| 41 : | edgomez | 1382 | int_div: |
| 42 : | dd 0 | ||
| 43 : | %assign i 1 | ||
| 44 : | %rep 255 | ||
| 45 : | dd (1 << 16) / (i) + 1 | ||
| 46 : | %assign i i+1 | ||
| 47 : | %endrep | ||
| 48 : | |||
| 49 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 50 : | edgomez | 1382 | plus_one: |
| 51 : | times 8 dw 1 | ||
| 52 : | |||
| 53 : | ;----------------------------------------------------------------------------- | ||
| 54 : | ; subtract by Q/2 table | ||
| 55 : | ;----------------------------------------------------------------------------- | ||
| 56 : | |||
| 57 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 58 : | edgomez | 1382 | mmx_sub: |
| 59 : | %assign i 1 | ||
| 60 : | %rep 31 | ||
| 61 : | times 4 dw i / 2 | ||
| 62 : | %assign i i+1 | ||
| 63 : | %endrep | ||
| 64 : | |||
| 65 : | |||
| 66 : | ;----------------------------------------------------------------------------- | ||
| 67 : | ; | ||
| 68 : | ; divide by 2Q table | ||
| 69 : | ; | ||
| 70 : | ; use a shift of 16 to take full advantage of _pmulhw_ | ||
| 71 : | ; for q=1, _pmulhw_ will overflow so it is treated seperately | ||
| 72 : | ; (3dnow2 provides _pmulhuw_ which wont cause overflow) | ||
| 73 : | ; | ||
| 74 : | ;----------------------------------------------------------------------------- | ||
| 75 : | |||
| 76 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 77 : | edgomez | 1382 | mmx_div: |
| 78 : | %assign i 1 | ||
| 79 : | %rep 31 | ||
| 80 : | times 4 dw (1 << 16) / (i * 2) + 1 | ||
| 81 : | %assign i i+1 | ||
| 82 : | %endrep | ||
| 83 : | |||
| 84 : | ;----------------------------------------------------------------------------- | ||
| 85 : | ; add by (odd(Q) ? Q : Q - 1) table | ||
| 86 : | ;----------------------------------------------------------------------------- | ||
| 87 : | |||
| 88 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 89 : | edgomez | 1382 | mmx_add: |
| 90 : | %assign i 1 | ||
| 91 : | %rep 31 | ||
| 92 : | %if i % 2 != 0 | ||
| 93 : | times 4 dw i | ||
| 94 : | %else | ||
| 95 : | times 4 dw i - 1 | ||
| 96 : | %endif | ||
| 97 : | %assign i i+1 | ||
| 98 : | %endrep | ||
| 99 : | |||
| 100 : | ;----------------------------------------------------------------------------- | ||
| 101 : | ; multiple by 2Q table | ||
| 102 : | ;----------------------------------------------------------------------------- | ||
| 103 : | |||
| 104 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 105 : | edgomez | 1382 | mmx_mul: |
| 106 : | %assign i 1 | ||
| 107 : | %rep 31 | ||
| 108 : | times 4 dw i * 2 | ||
| 109 : | %assign i i+1 | ||
| 110 : | %endrep | ||
| 111 : | |||
| 112 : | ;----------------------------------------------------------------------------- | ||
| 113 : | ; saturation limits | ||
| 114 : | ;----------------------------------------------------------------------------- | ||
| 115 : | |||
| 116 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 117 : | edgomez | 1382 | mmx_32768_minus_2048: |
| 118 : | times 4 dw (32768-2048) | ||
| 119 : | mmx_32767_minus_2047: | ||
| 120 : | times 4 dw (32767-2047) | ||
| 121 : | |||
| 122 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 123 : | edgomez | 1382 | mmx_2047: |
| 124 : | times 4 dw 2047 | ||
| 125 : | |||
| 126 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 127 : | edgomez | 1382 | mmzero: |
| 128 : | dd 0, 0 | ||
| 129 : | int2047: | ||
| 130 : | dd 2047 | ||
| 131 : | int_2048: | ||
| 132 : | dd -2048 | ||
| 133 : | |||
| 134 : | ;============================================================================= | ||
| 135 : | ; Code | ||
| 136 : | ;============================================================================= | ||
| 137 : | |||
| 138 : | Isibaar | 1795 | SECTION .rotext align=SECTION_ALIGN |
| 139 : | edgomez | 1382 | |
| 140 : | |||
| 141 : | ;----------------------------------------------------------------------------- | ||
| 142 : | ; | ||
| 143 : | ; uint32_t quant_h263_intra_3dne(int16_t * coeff, | ||
| 144 : | ; const int16_t const * data, | ||
| 145 : | ; const uint32_t quant, | ||
| 146 : | ; const uint32_t dcscalar, | ||
| 147 : | ; const uint16_t *mpeg_matrices); | ||
| 148 : | ; | ||
| 149 : | ;----------------------------------------------------------------------------- | ||
| 150 : | ;This is Athlon-optimized code (ca 70 clk per call) | ||
| 151 : | |||
| 152 : | %macro quant_intra1 1 | ||
| 153 : | psubw mm1, mm0 ;A3 | ||
| 154 : | psubw mm3, mm2 ;B3 | ||
| 155 : | %if (%1) | ||
| 156 : | psubw mm5, mm4 ;C8 | ||
| 157 : | psubw mm7, mm6 ;D8 | ||
| 158 : | %endif | ||
| 159 : | |||
| 160 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 161 : | movq mm4, [_ECX + %1 * 32 +16] ;C1 | ||
| 162 : | edgomez | 1382 | pmaxsw mm1, mm0 ;A4 |
| 163 : | Isibaar | 1795 | movq mm6, [_ECX + %1 * 32 +24] ;D1 |
| 164 : | edgomez | 1382 | pmaxsw mm3, mm2 ;B4 |
| 165 : | |||
| 166 : | |||
| 167 : | psraw mm0, 15 ;A5 | ||
| 168 : | psraw mm2, 15 ;B5 | ||
| 169 : | %if (%1) | ||
| 170 : | Isibaar | 1795 | movq [_EDX + %1 * 32 + 16-32], mm5 ;C9 |
| 171 : | movq [_EDX + %1 * 32 + 24-32], mm7 ;D9 | ||
| 172 : | edgomez | 1382 | %endif |
| 173 : | |||
| 174 : | psrlw mm1, 1 ;A6 | ||
| 175 : | psrlw mm3, 1 ;B6 | ||
| 176 : | Isibaar | 1795 | movq mm5, [_EBX] ;C2 |
| 177 : | movq mm7, [_EBX] ;D2 | ||
| 178 : | edgomez | 1382 | |
| 179 : | pxor mm1, mm0 ;A7 | ||
| 180 : | pxor mm3, mm2 ;B7 | ||
| 181 : | |||
| 182 : | psubw mm5, mm4 ;C3 | ||
| 183 : | psubw mm7, mm6 ;D3 | ||
| 184 : | psubw mm1, mm0 ;A8 | ||
| 185 : | psubw mm3, mm2 ;B8 | ||
| 186 : | |||
| 187 : | %if (%1 == 0) | ||
| 188 : | Isibaar | 1795 | push _EBP |
| 189 : | movq mm0, [_ECX + %1 * 32 +32] | ||
| 190 : | edgomez | 1382 | %elif (%1 < 3) |
| 191 : | Isibaar | 1795 | movq mm0, [_ECX + %1 * 32 +32] ;A1 |
| 192 : | edgomez | 1382 | %endif |
| 193 : | pmaxsw mm5, mm4 ;C4 | ||
| 194 : | %if (%1 < 3) | ||
| 195 : | Isibaar | 1795 | movq mm2, [_ECX + %1 * 32 +8+32] ;B1 |
| 196 : | edgomez | 1382 | %else |
| 197 : | Isibaar | 1795 | cmp _ESP, _ESP |
| 198 : | edgomez | 1382 | %endif |
| 199 : | pmaxsw mm7, mm6 ;D4 | ||
| 200 : | |||
| 201 : | psraw mm4, 15 ;C5 | ||
| 202 : | psraw mm6, 15 ;D5 | ||
| 203 : | Isibaar | 1795 | movq [byte _EDX + %1 * 32], mm1 ;A9 |
| 204 : | movq [_EDX + %1 * 32+8], mm3 ;B9 | ||
| 205 : | edgomez | 1382 | |
| 206 : | |||
| 207 : | psrlw mm5, 1 ;C6 | ||
| 208 : | psrlw mm7, 1 ;D6 | ||
| 209 : | %if (%1 < 3) | ||
| 210 : | Isibaar | 1795 | movq mm1, [_EBX] ;A2 |
| 211 : | movq mm3, [_EBX] ;B2 | ||
| 212 : | edgomez | 1382 | %endif |
| 213 : | %if (%1 == 3) | ||
| 214 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 215 : | lea r9, [int_div] | ||
| 216 : | imul eax, dword [r9+4*_EDI] | ||
| 217 : | %else | ||
| 218 : | imul _EAX, [int_div+4*_EDI] | ||
| 219 : | edgomez | 1382 | %endif |
| 220 : | Isibaar | 1795 | %endif |
| 221 : | edgomez | 1382 | pxor mm5, mm4 ;C7 |
| 222 : | pxor mm7, mm6 ;D7 | ||
| 223 : | %endm | ||
| 224 : | |||
| 225 : | |||
| 226 : | %macro quant_intra 1 | ||
| 227 : | ; Rules for athlon: | ||
| 228 : | ; 1) schedule latencies | ||
| 229 : | ; 2) add/mul and load/store in 2:1 proportion | ||
| 230 : | ; 3) avoid spliting >3byte instructions over 8byte boundaries | ||
| 231 : | |||
| 232 : | psubw mm1, mm0 ;A3 | ||
| 233 : | psubw mm3, mm2 ;B3 | ||
| 234 : | %if (%1) | ||
| 235 : | psubw mm5, mm4 ;C8 | ||
| 236 : | psubw mm7, mm6 ;D8 | ||
| 237 : | %endif | ||
| 238 : | |||
| 239 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 240 : | movq mm4, [_ECX + %1 * 32 +16] ;C1 | ||
| 241 : | edgomez | 1382 | pmaxsw mm1, mm0 ;A4 |
| 242 : | Isibaar | 1795 | movq mm6, [_ECX + %1 * 32 +24] ;D1 |
| 243 : | edgomez | 1382 | pmaxsw mm3, mm2 ;B4 |
| 244 : | |||
| 245 : | |||
| 246 : | psraw mm0, 15 ;A5 | ||
| 247 : | psraw mm2, 15 ;B5 | ||
| 248 : | %if (%1) | ||
| 249 : | Isibaar | 1795 | movq [_EDX + %1 * 32 + 16-32], mm5 ;C9 |
| 250 : | movq [_EDX + %1 * 32 + 24-32], mm7 ;D9 | ||
| 251 : | edgomez | 1382 | %endif |
| 252 : | |||
| 253 : | Isibaar | 1795 | pmulhw mm1, [_ESI] ;A6 |
| 254 : | pmulhw mm3, [_ESI] ;B6 | ||
| 255 : | movq mm5, [_EBX] ;C2 | ||
| 256 : | movq mm7, [_EBX] ;D2 | ||
| 257 : | edgomez | 1382 | |
| 258 : | nop | ||
| 259 : | nop | ||
| 260 : | pxor mm1, mm0 ;A7 | ||
| 261 : | pxor mm3, mm2 ;B7 | ||
| 262 : | |||
| 263 : | psubw mm5, mm4 ;C3 | ||
| 264 : | psubw mm7, mm6 ;D3 | ||
| 265 : | psubw mm1, mm0 ;A8 | ||
| 266 : | psubw mm3, mm2 ;B8 | ||
| 267 : | |||
| 268 : | |||
| 269 : | %if (%1 < 3) | ||
| 270 : | Isibaar | 1795 | movq mm0, [_ECX + %1 * 32 +32] ;A1 |
| 271 : | edgomez | 1382 | %endif |
| 272 : | pmaxsw mm5, mm4 ;C4 | ||
| 273 : | %if (%1 < 3) | ||
| 274 : | Isibaar | 1795 | movq mm2, [_ECX + %1 * 32 +8+32] ;B1 |
| 275 : | edgomez | 1382 | %else |
| 276 : | Isibaar | 1795 | cmp _ESP, _ESP |
| 277 : | edgomez | 1382 | %endif |
| 278 : | pmaxsw mm7,mm6 ;D4 | ||
| 279 : | |||
| 280 : | psraw mm4, 15 ;C5 | ||
| 281 : | psraw mm6, 15 ;D5 | ||
| 282 : | Isibaar | 1795 | movq [byte _EDX + %1 * 32], mm1 ;A9 |
| 283 : | movq [_EDX + %1 * 32+8], mm3 ;B9 | ||
| 284 : | edgomez | 1382 | |
| 285 : | |||
| 286 : | Isibaar | 1795 | pmulhw mm5, [_ESI] ;C6 |
| 287 : | pmulhw mm7, [_ESI] ;D6 | ||
| 288 : | edgomez | 1382 | %if (%1 < 3) |
| 289 : | Isibaar | 1795 | movq mm1, [_EBX] ;A2 |
| 290 : | movq mm3, [_EBX] ;B2 | ||
| 291 : | edgomez | 1382 | %endif |
| 292 : | %if (%1 == 0) | ||
| 293 : | Isibaar | 1795 | push _EBP |
| 294 : | edgomez | 1382 | %elif (%1 < 3) |
| 295 : | nop | ||
| 296 : | %endif | ||
| 297 : | nop | ||
| 298 : | %if (%1 == 3) | ||
| 299 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 300 : | lea r9, [int_div] | ||
| 301 : | imul eax, dword [r9+4*_EDI] | ||
| 302 : | %else | ||
| 303 : | imul _EAX, [int_div+4*_EDI] | ||
| 304 : | edgomez | 1382 | %endif |
| 305 : | Isibaar | 1795 | %endif |
| 306 : | edgomez | 1382 | pxor mm5, mm4 ;C7 |
| 307 : | pxor mm7, mm6 ;D7 | ||
| 308 : | %endmacro | ||
| 309 : | |||
| 310 : | |||
| 311 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 312 : | edgomez | 1382 | cglobal quant_h263_intra_3dne |
| 313 : | quant_h263_intra_3dne: | ||
| 314 : | |||
| 315 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 316 : | mov TMP0, [_ESP] | ||
| 317 : | add _ESP, PTR_SIZE | ||
| 318 : | %ifndef WINDOWS | ||
| 319 : | push prm6 | ||
| 320 : | push prm5 | ||
| 321 : | %endif | ||
| 322 : | push prm4 | ||
| 323 : | push prm3 | ||
| 324 : | push prm2 | ||
| 325 : | push prm1 | ||
| 326 : | sub _ESP, PTR_SIZE | ||
| 327 : | mov [_ESP], TMP0 | ||
| 328 : | %endif | ||
| 329 : | |||
| 330 : | mov _EAX, [_ESP + 3*PTR_SIZE] ; quant | ||
| 331 : | mov _ECX, [_ESP + 2*PTR_SIZE] ; data | ||
| 332 : | mov _EDX, [_ESP + 1*PTR_SIZE] ; coeff | ||
| 333 : | edgomez | 1382 | cmp al, 1 |
| 334 : | pxor mm1, mm1 | ||
| 335 : | pxor mm3, mm3 | ||
| 336 : | Isibaar | 1795 | movq mm0, [_ECX] ; mm0 = [1st] |
| 337 : | movq mm2, [_ECX + 8] | ||
| 338 : | push _ESI | ||
| 339 : | %ifdef ARCH_IS_X86_64 | ||
| 340 : | lea _ESI, [mmx_div] | ||
| 341 : | lea _ESI, [_ESI + _EAX*8 - 8] | ||
| 342 : | %else | ||
| 343 : | lea _ESI, [mmx_div + _EAX*8 - 8] | ||
| 344 : | %endif | ||
| 345 : | edgomez | 1382 | |
| 346 : | Isibaar | 1795 | push _EBX |
| 347 : | mov _EBX, mmzero | ||
| 348 : | push _EDI | ||
| 349 : | edgomez | 1382 | jz near .q1loop |
| 350 : | |||
| 351 : | quant_intra 0 | ||
| 352 : | Isibaar | 1795 | mov _EBP, [_ESP + (4+4)*PTR_SIZE] ; dcscalar |
| 353 : | ; NB -- there are 3 pushes in the function preambule and one more | ||
| 354 : | ; in "quant_intra 0", thus an added offset of 16 bytes | ||
| 355 : | movsx _EAX, word [byte _ECX] ; DC | ||
| 356 : | edgomez | 1382 | |
| 357 : | quant_intra 1 | ||
| 358 : | Isibaar | 1795 | mov _EDI, _EAX |
| 359 : | sar _EDI, 31 ; sign(DC) | ||
| 360 : | shr _EBP, byte 1 ; _EBP = dcscalar/2 | ||
| 361 : | edgomez | 1382 | |
| 362 : | quant_intra 2 | ||
| 363 : | Isibaar | 1795 | sub _EAX, _EDI ; DC (+1) |
| 364 : | xor _EBP, _EDI ; sign(DC) dcscalar /2 (-1) | ||
| 365 : | mov _EDI, [_ESP + (4+4)*PTR_SIZE] ; dscalar | ||
| 366 : | lea _EAX, [byte _EAX + _EBP] ; DC + sign(DC) dcscalar/2 | ||
| 367 : | mov _EBP, [byte _ESP] | ||
| 368 : | edgomez | 1382 | |
| 369 : | quant_intra 3 | ||
| 370 : | psubw mm5, mm4 ;C8 | ||
| 371 : | Isibaar | 1795 | mov _ESI, [_ESP + 3*PTR_SIZE] ; pop back the register value |
| 372 : | mov _EDI, [_ESP + 1*PTR_SIZE] ; pop back the register value | ||
| 373 : | sar _EAX, 16 | ||
| 374 : | lea _EBX, [byte _EAX + 1] ; workaround for _EAX < 0 | ||
| 375 : | cmovs _EAX, _EBX ; conditionnaly move the corrected value | ||
| 376 : | mov [_EDX], ax ; coeff[0] = ax | ||
| 377 : | mov _EBX, [_ESP + 2*PTR_SIZE] ; pop back the register value | ||
| 378 : | add _ESP, byte 4*PTR_SIZE ; "quant_intra 0" pushed _EBP, but we don't restore that one, just correct the stack offset by 16 | ||
| 379 : | edgomez | 1382 | psubw mm7, mm6 ;D8 |
| 380 : | Isibaar | 1795 | movq [_EDX + 3 * 32 + 16], mm5 ;C9 |
| 381 : | movq [_EDX + 3 * 32 + 24], mm7 ;D9 | ||
| 382 : | edgomez | 1382 | |
| 383 : | Isibaar | 1795 | xor _EAX, _EAX |
| 384 : | |||
| 385 : | %ifdef ARCH_IS_X86_64 | ||
| 386 : | mov TMP0, [_ESP] | ||
| 387 : | %ifndef WINDOWS | ||
| 388 : | add _ESP, 6*PTR_SIZE | ||
| 389 : | %else | ||
| 390 : | add _ESP, 4*PTR_SIZE | ||
| 391 : | %endif | ||
| 392 : | mov [_ESP], TMP0 | ||
| 393 : | %endif | ||
| 394 : | |||
| 395 : | edgomez | 1382 | ret |
| 396 : | |||
| 397 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 398 : | edgomez | 1382 | |
| 399 : | Isibaar | 1793 | .q1loop: |
| 400 : | edgomez | 1382 | quant_intra1 0 |
| 401 : | Isibaar | 1795 | mov _EBP, [_ESP + (4+4)*PTR_SIZE] ; dcscalar |
| 402 : | movsx _EAX, word [byte _ECX] ; DC | ||
| 403 : | edgomez | 1382 | |
| 404 : | quant_intra1 1 | ||
| 405 : | Isibaar | 1795 | mov _EDI, _EAX |
| 406 : | sar _EDI, 31 ; sign(DC) | ||
| 407 : | shr _EBP, byte 1 ; _EBP = dcscalar /2 | ||
| 408 : | edgomez | 1382 | |
| 409 : | quant_intra1 2 | ||
| 410 : | Isibaar | 1795 | sub _EAX, _EDI ; DC (+1) |
| 411 : | xor _EBP, _EDI ; sign(DC) dcscalar /2 (-1) | ||
| 412 : | mov _EDI, [_ESP + (4+4)*PTR_SIZE] ; dcscalar | ||
| 413 : | lea _EAX, [byte _EAX + _EBP] ; DC + sign(DC) dcscalar /2 | ||
| 414 : | mov _EBP, [byte _ESP] | ||
| 415 : | edgomez | 1382 | |
| 416 : | quant_intra1 3 | ||
| 417 : | psubw mm5, mm4 ;C8 | ||
| 418 : | Isibaar | 1795 | mov _ESI, [_ESP + 3*PTR_SIZE] ; pop back the register value |
| 419 : | mov _EDI, [_ESP + 1*PTR_SIZE] ; pop back the register value | ||
| 420 : | sar _EAX, 16 | ||
| 421 : | lea _EBX, [byte _EAX + 1] ; workaround for _EAX < 0 | ||
| 422 : | cmovs _EAX, _EBX ; conditionnaly move the corrected value | ||
| 423 : | mov [_EDX], ax ; coeff[0] = ax | ||
| 424 : | mov _EBX, [_ESP + 2*PTR_SIZE] ; pop back the register value | ||
| 425 : | add _ESP, byte 4*PTR_SIZE ; "quant_intra 0" pushed _EBP, but we don't restore that one, just correct the stack offset by 16 | ||
| 426 : | edgomez | 1382 | psubw mm7, mm6 ;D8 |
| 427 : | Isibaar | 1795 | movq [_EDX + 3 * 32 + 16], mm5 ;C9 |
| 428 : | movq [_EDX + 3 * 32 + 24], mm7 ;D9 | ||
| 429 : | edgomez | 1382 | |
| 430 : | Isibaar | 1795 | xor _EAX, _EAX |
| 431 : | |||
| 432 : | %ifdef ARCH_IS_X86_64 | ||
| 433 : | mov TMP0, [_ESP] | ||
| 434 : | %ifndef WINDOWS | ||
| 435 : | add _ESP, 6*PTR_SIZE | ||
| 436 : | %else | ||
| 437 : | add _ESP, 4*PTR_SIZE | ||
| 438 : | %endif | ||
| 439 : | mov [_ESP], TMP0 | ||
| 440 : | %endif | ||
| 441 : | |||
| 442 : | edgomez | 1382 | ret |
| 443 : | Isibaar | 1793 | ENDFUNC |
| 444 : | edgomez | 1382 | |
| 445 : | |||
| 446 : | ;----------------------------------------------------------------------------- | ||
| 447 : | ; | ||
| 448 : | ; uint32_t quant_h263_inter_3dne(int16_t * coeff, | ||
| 449 : | ; const int16_t const * data, | ||
| 450 : | ; const uint32_t quant, | ||
| 451 : | ; const uint16_t *mpeg_matrices); | ||
| 452 : | ; | ||
| 453 : | ;----------------------------------------------------------------------------- | ||
| 454 : | ;This is Athlon-optimized code (ca 90 clk per call) | ||
| 455 : | ;Optimized by Jaan, 30 Nov 2002 | ||
| 456 : | |||
| 457 : | |||
| 458 : | %macro quantinter 1 | ||
| 459 : | Isibaar | 1795 | movq mm1, [_EAX] ;A2 |
| 460 : | edgomez | 1382 | psraw mm3, 15 ;B6 |
| 461 : | %if (%1) | ||
| 462 : | psubw mm2, mm6 ;C10 | ||
| 463 : | %endif | ||
| 464 : | psubw mm1, mm0 ;A3 | ||
| 465 : | pmulhw mm4, mm7 ;B7 | ||
| 466 : | Isibaar | 1795 | movq mm6, [_ECX + %1*24+16] ;C1 |
| 467 : | edgomez | 1382 | pmaxsw mm1, mm0 ;A4 |
| 468 : | paddw mm5, mm4 ;B8 | ||
| 469 : | %if (%1) | ||
| 470 : | Isibaar | 1795 | movq [_EDX + %1*24+16-24], mm2 ;C11 |
| 471 : | edgomez | 1382 | %endif |
| 472 : | Isibaar | 1795 | psubusw mm1, [_EBX] ;A5 mm0 -= sub (unsigned, dont go < 0) |
| 473 : | edgomez | 1382 | pxor mm4, mm3 ;B9 |
| 474 : | Isibaar | 1795 | movq mm2, [_EAX] ;C2 |
| 475 : | edgomez | 1382 | psraw mm0, 15 ;A6 |
| 476 : | psubw mm4, mm3 ;B10 | ||
| 477 : | psubw mm2, mm6 ;C3 | ||
| 478 : | pmulhw mm1, mm7 ;A7 mm0 = (mm0 / 2Q) >> 24 | ||
| 479 : | Isibaar | 1795 | movq mm3, [_ECX + %1*24+8] ;B1 |
| 480 : | edgomez | 1382 | pmaxsw mm2, mm6 ;C4 |
| 481 : | paddw mm5, mm1 ;A8 sum += mm0 | ||
| 482 : | %if (%1) | ||
| 483 : | Isibaar | 1795 | movq [_EDX + %1*24+8-24], mm4 ;B11 |
| 484 : | edgomez | 1382 | %else |
| 485 : | Isibaar | 1795 | movq [_EDX + 120], mm4 ;B11 |
| 486 : | edgomez | 1382 | %endif |
| 487 : | Isibaar | 1795 | psubusw mm2, [_EBX] ;C5 |
| 488 : | edgomez | 1382 | pxor mm1, mm0 ;A9 mm0 *= sign(mm0) |
| 489 : | Isibaar | 1795 | movq mm4, [_EAX] ;B2 |
| 490 : | edgomez | 1382 | psraw mm6, 15 ;C6 |
| 491 : | psubw mm1, mm0 ;A10 undisplace | ||
| 492 : | psubw mm4, mm3 ;B3 | ||
| 493 : | pmulhw mm2, mm7 ;C7 | ||
| 494 : | Isibaar | 1795 | movq mm0, [_ECX + %1*24+24] ;A1 mm0 = [1st] |
| 495 : | edgomez | 1382 | pmaxsw mm4, mm3 ;B4 |
| 496 : | paddw mm5, mm2 ;C8 | ||
| 497 : | Isibaar | 1795 | movq [byte _EDX + %1*24], mm1 ;A11 |
| 498 : | psubusw mm4, [_EBX] ;B5 | ||
| 499 : | edgomez | 1382 | pxor mm2, mm6 ;C9 |
| 500 : | %endmacro | ||
| 501 : | |||
| 502 : | %macro quantinter1 1 | ||
| 503 : | Isibaar | 1795 | movq mm0, [byte _ECX + %1*16] ;mm0 = [1st] |
| 504 : | movq mm3, [_ECX + %1*16+8] ; | ||
| 505 : | movq mm1, [_EAX] | ||
| 506 : | movq mm4, [_EAX] | ||
| 507 : | edgomez | 1382 | psubw mm1, mm0 |
| 508 : | psubw mm4, mm3 | ||
| 509 : | pmaxsw mm1, mm0 | ||
| 510 : | pmaxsw mm4, mm3 | ||
| 511 : | psubusw mm1, mm6 ; mm0 -= sub (unsigned, dont go < 0) | ||
| 512 : | psubusw mm4, mm6 ; | ||
| 513 : | psraw mm0, 15 | ||
| 514 : | psraw mm3, 15 | ||
| 515 : | psrlw mm1, 1 ; mm0 = (mm0 / 2Q) >> 16 | ||
| 516 : | psrlw mm4, 1 ; | ||
| 517 : | paddw mm5, mm1 ; sum += mm0 | ||
| 518 : | pxor mm1, mm0 ; mm0 *= sign(mm0) | ||
| 519 : | paddw mm5, mm4 | ||
| 520 : | pxor mm4, mm3 ; | ||
| 521 : | psubw mm1, mm0 ; undisplace | ||
| 522 : | psubw mm4, mm3 | ||
| 523 : | Isibaar | 1795 | cmp _ESP, _ESP |
| 524 : | movq [byte _EDX + %1*16], mm1 | ||
| 525 : | movq [_EDX + %1*16+8], mm4 | ||
| 526 : | edgomez | 1382 | %endmacro |
| 527 : | |||
| 528 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 529 : | edgomez | 1382 | cglobal quant_h263_inter_3dne |
| 530 : | quant_h263_inter_3dne: | ||
| 531 : | |||
| 532 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 533 : | mov TMP0, [_ESP] | ||
| 534 : | add _ESP, PTR_SIZE | ||
| 535 : | %ifndef WINDOWS | ||
| 536 : | push prm6 | ||
| 537 : | push prm5 | ||
| 538 : | %endif | ||
| 539 : | push prm4 | ||
| 540 : | push prm3 | ||
| 541 : | push prm2 | ||
| 542 : | push prm1 | ||
| 543 : | sub _ESP, PTR_SIZE | ||
| 544 : | mov [_ESP], TMP0 | ||
| 545 : | %endif | ||
| 546 : | |||
| 547 : | mov _EDX, [_ESP + 1*PTR_SIZE] ; coeff | ||
| 548 : | mov _ECX, [_ESP + 2*PTR_SIZE] ; data | ||
| 549 : | mov _EAX, [_ESP + 3*PTR_SIZE] ; quant | ||
| 550 : | push _EBX | ||
| 551 : | |||
| 552 : | edgomez | 1382 | pxor mm5, mm5 ; sum |
| 553 : | nop | ||
| 554 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 555 : | lea _EBX, [mmx_div] | ||
| 556 : | movq mm7, [_EBX + _EAX * 8 - 8] | ||
| 557 : | lea _EBX, [mmx_sub] | ||
| 558 : | lea _EBX, [_EBX + _EAX * 8 - 8] | ||
| 559 : | %else | ||
| 560 : | lea _EBX,[mmx_sub + _EAX * 8 - 8] ; sub | ||
| 561 : | movq mm7, [mmx_div + _EAX * 8 - 8] ; divider | ||
| 562 : | %endif | ||
| 563 : | edgomez | 1382 | |
| 564 : | cmp al, 1 | ||
| 565 : | Isibaar | 1795 | lea _EAX, [mmzero] |
| 566 : | edgomez | 1382 | jz near .q1loop |
| 567 : | Isibaar | 1795 | cmp _ESP, _ESP |
| 568 : | ALIGN SECTION_ALIGN | ||
| 569 : | movq mm3, [_ECX + 120] ;B1 | ||
| 570 : | edgomez | 1382 | pxor mm4, mm4 ;B2 |
| 571 : | psubw mm4, mm3 ;B3 | ||
| 572 : | Isibaar | 1795 | movq mm0, [_ECX] ;A1 mm0 = [1st] |
| 573 : | edgomez | 1382 | pmaxsw mm4, mm3 ;B4 |
| 574 : | Isibaar | 1795 | psubusw mm4, [_EBX] ;B5 |
| 575 : | edgomez | 1382 | |
| 576 : | quantinter 0 | ||
| 577 : | quantinter 1 | ||
| 578 : | quantinter 2 | ||
| 579 : | quantinter 3 | ||
| 580 : | quantinter 4 | ||
| 581 : | |||
| 582 : | psraw mm3, 15 ;B6 | ||
| 583 : | psubw mm2, mm6 ;C10 | ||
| 584 : | pmulhw mm4, mm7 ;B7 | ||
| 585 : | paddw mm5, mm4 ;B8 | ||
| 586 : | pxor mm4, mm3 ;B9 | ||
| 587 : | psubw mm4, mm3 ;B10 | ||
| 588 : | Isibaar | 1795 | movq [_EDX + 4*24+16], mm2 ;C11 |
| 589 : | pop _EBX | ||
| 590 : | movq [_EDX + 4*24+8], mm4 ;B11 | ||
| 591 : | edgomez | 1382 | pmaddwd mm5, [plus_one] |
| 592 : | movq mm0, mm5 | ||
| 593 : | punpckhdq mm5, mm5 | ||
| 594 : | paddd mm0, mm5 | ||
| 595 : | movd eax, mm0 ; return sum | ||
| 596 : | |||
| 597 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 598 : | mov TMP0, [_ESP] | ||
| 599 : | %ifndef WINDOWS | ||
| 600 : | add _ESP, 6*PTR_SIZE | ||
| 601 : | %else | ||
| 602 : | add _ESP, 4*PTR_SIZE | ||
| 603 : | %endif | ||
| 604 : | mov [_ESP], TMP0 | ||
| 605 : | %endif | ||
| 606 : | |||
| 607 : | edgomez | 1382 | ret |
| 608 : | |||
| 609 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 610 : | Isibaar | 1793 | .q1loop: |
| 611 : | Isibaar | 1795 | movq mm6, [byte _EBX] |
| 612 : | edgomez | 1382 | |
| 613 : | quantinter1 0 | ||
| 614 : | quantinter1 1 | ||
| 615 : | quantinter1 2 | ||
| 616 : | quantinter1 3 | ||
| 617 : | quantinter1 4 | ||
| 618 : | quantinter1 5 | ||
| 619 : | quantinter1 6 | ||
| 620 : | quantinter1 7 | ||
| 621 : | |||
| 622 : | pmaddwd mm5, [plus_one] | ||
| 623 : | movq mm0, mm5 | ||
| 624 : | psrlq mm5, 32 | ||
| 625 : | paddd mm0, mm5 | ||
| 626 : | movd eax, mm0 ; return sum | ||
| 627 : | |||
| 628 : | Isibaar | 1795 | pop _EBX |
| 629 : | edgomez | 1382 | |
| 630 : | Isibaar | 1795 | %ifdef ARCH_IS_X86_64 |
| 631 : | mov TMP0, [_ESP] | ||
| 632 : | %ifndef WINDOWS | ||
| 633 : | add _ESP, 6*PTR_SIZE | ||
| 634 : | %else | ||
| 635 : | add _ESP, 4*PTR_SIZE | ||
| 636 : | %endif | ||
| 637 : | mov [_ESP], TMP0 | ||
| 638 : | %endif | ||
| 639 : | |||
| 640 : | edgomez | 1382 | ret |
| 641 : | Isibaar | 1793 | ENDFUNC |
| 642 : | edgomez | 1382 | |
| 643 : | Isibaar | 1795 | |
| 644 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
| 645 : | ; | ||
| 646 : | ; uint32_t dequant_h263_intra_3dne(int16_t *data, | ||
| 647 : | ; const int16_t const *coeff, | ||
| 648 : | ; const uint32_t quant, | ||
| 649 : | ; const uint32_t dcscalar, | ||
| 650 : | ; const uint16_t *mpeg_matrices); | ||
| 651 : | ; | ||
| 652 : | ;----------------------------------------------------------------------------- | ||
| 653 : | |||
| 654 : | ; this is the same as dequant_inter_3dne, except that we're | ||
| 655 : | ; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) | ||
| 656 : | |||
| 657 : | ;This is Athlon-optimized code (ca 106 clk per call) | ||
| 658 : | |||
| 659 : | %macro dequant 1 | ||
| 660 : | Isibaar | 1795 | movq mm1, [_ECX+%1*24] ; c = coeff[i] ;A2 |
| 661 : | edgomez | 1382 | psubw mm0, mm1 ;-c ;A3 (1st dep) |
| 662 : | %if (%1) | ||
| 663 : | paddw mm4, mm6 ;C11 mm6 free (4th+) | ||
| 664 : | %endif | ||
| 665 : | pmaxsw mm0, mm1 ;|c| ;A4 (2nd) | ||
| 666 : | %if (%1) | ||
| 667 : | Isibaar | 1795 | mov _EBP, _EBP |
| 668 : | pminsw mm4, [_EBX] ;C12 saturates to +2047 (5th+) later | ||
| 669 : | edgomez | 1382 | %endif |
| 670 : | Isibaar | 1795 | movq mm6, [_ESI] ;0 ;A5 mm6 in use |
| 671 : | pandn mm7, [_EAX] ;B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 672 : | edgomez | 1382 | %if (%1) |
| 673 : | pxor mm5, mm4 ;C13 (6th+) 1later | ||
| 674 : | %endif | ||
| 675 : | Isibaar | 1795 | movq mm4, [_ESI] ;C1 ;0 |
| 676 : | mov _ESP, _ESP | ||
| 677 : | pcmpeqw mm6, [_ECX+%1*24] ;A6 (c ==0) ? -1 : 0 (1st) | ||
| 678 : | ALIGN SECTION_ALIGN | ||
| 679 : | edgomez | 1382 | psraw mm1, 15 ; sign(c) ;A7 (2nd) |
| 680 : | %if (%1) | ||
| 681 : | Isibaar | 1795 | movq [_EDX+%1*24+16-24], mm5 ; C14 (7th) 2later |
| 682 : | edgomez | 1382 | %endif |
| 683 : | paddw mm7, mm3 ;B10 offset +negate back (3rd) | ||
| 684 : | Isibaar | 1795 | pmullw mm0, [_EDI] ;*= 2Q ;A8 (3rd+) |
| 685 : | edgomez | 1382 | paddw mm2, mm7 ;B11 mm7 free (4th+) |
| 686 : | Isibaar | 1795 | lea _EBP, [byte _EBP] |
| 687 : | movq mm5, [_ECX+%1*24+16] ;C2 ; c = coeff[i] | ||
| 688 : | edgomez | 1382 | psubw mm4, mm5 ;-c ;C3 (1st dep) |
| 689 : | Isibaar | 1795 | pandn mm6, [_EAX] ;A9 offset = isZero ? 0 : quant_add (2nd) |
| 690 : | pminsw mm2, [_EBX] ;B12 saturates to +2047 (5th+) | ||
| 691 : | edgomez | 1382 | pxor mm3, mm2 ;B13 (6th+) |
| 692 : | Isibaar | 1795 | movq mm2, [byte _ESI] ;B1 ;0 |
| 693 : | edgomez | 1382 | %if (%1) |
| 694 : | Isibaar | 1795 | movq [_EDX+%1*24+8-24], mm3 ;B14 (7th) |
| 695 : | edgomez | 1382 | %else |
| 696 : | Isibaar | 1795 | movq [_EDX+120], mm3 |
| 697 : | edgomez | 1382 | %endif |
| 698 : | pmaxsw mm4, mm5 ;|c| ;C4 (2nd) | ||
| 699 : | paddw mm6, mm1 ;A10 offset +negate back (3rd) | ||
| 700 : | Isibaar | 1795 | movq mm3, [_ECX+%1*24 + 8] ;B2 ; c = coeff[i] |
| 701 : | edgomez | 1382 | psubw mm2, mm3 ;-c ;B3 (1st dep) |
| 702 : | paddw mm0, mm6 ;A11 mm6 free (4th+) | ||
| 703 : | Isibaar | 1795 | movq mm6, [byte _ESI] ;0 ;C5 mm6 in use |
| 704 : | pcmpeqw mm6, [_ECX+%1*24+16] ;C6 (c ==0) ? -1 : 0 (1st) | ||
| 705 : | pminsw mm0, [_EBX] ;A12 saturates to +2047 (5th+) | ||
| 706 : | edgomez | 1382 | pmaxsw mm2, mm3 ;|c| ;B4 (2nd) |
| 707 : | pxor mm1, mm0 ;A13 (6th+) | ||
| 708 : | Isibaar | 1795 | pmullw mm4, [_EDI] ;*= 2Q ;C8 (3rd+) |
| 709 : | edgomez | 1382 | psraw mm5, 15 ; sign(c) ;C7 (2nd) |
| 710 : | Isibaar | 1795 | movq mm7, [byte _ESI] ;0 ;B5 mm7 in use |
| 711 : | pcmpeqw mm7, [_ECX+%1*24 + 8] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 712 : | edgomez | 1382 | %if (%1 < 4) |
| 713 : | Isibaar | 1795 | movq mm0, [byte _ESI] ;A1 ;0 |
| 714 : | edgomez | 1382 | %endif |
| 715 : | Isibaar | 1795 | pandn mm6, [byte _EAX] ;C9 offset = isZero ? 0 : quant_add (2nd) |
| 716 : | edgomez | 1382 | psraw mm3, 15 ;sign(c) ;B7 (2nd) |
| 717 : | Isibaar | 1795 | movq [byte _EDX+%1*24], mm1 ;A14 (7th) |
| 718 : | edgomez | 1382 | paddw mm6, mm5 ;C10 offset +negate back (3rd) |
| 719 : | Isibaar | 1795 | pmullw mm2, [_EDI] ;*= 2Q ;B8 (3rd+) |
| 720 : | mov _ESP, _ESP | ||
| 721 : | edgomez | 1382 | %endmacro |
| 722 : | |||
| 723 : | |||
| 724 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 725 : | edgomez | 1382 | cglobal dequant_h263_intra_3dne |
| 726 : | dequant_h263_intra_3dne: | ||
| 727 : | Isibaar | 1795 | |
| 728 : | %ifdef ARCH_IS_X86_64 | ||
| 729 : | mov TMP0, [_ESP] | ||
| 730 : | add _ESP, PTR_SIZE | ||
| 731 : | %ifndef WINDOWS | ||
| 732 : | push prm6 | ||
| 733 : | push prm5 | ||
| 734 : | %endif | ||
| 735 : | push prm4 | ||
| 736 : | push prm3 | ||
| 737 : | push prm2 | ||
| 738 : | push prm1 | ||
| 739 : | sub _ESP, PTR_SIZE | ||
| 740 : | mov [_ESP], TMP0 | ||
| 741 : | %endif | ||
| 742 : | |||
| 743 : | mov _ECX, [_ESP+ 2*PTR_SIZE] ; coeff | ||
| 744 : | mov _EAX, [_ESP+ 3*PTR_SIZE] ; quant | ||
| 745 : | edgomez | 1382 | pxor mm0, mm0 |
| 746 : | pxor mm2, mm2 | ||
| 747 : | Isibaar | 1795 | push _EDI |
| 748 : | push _EBX | ||
| 749 : | %ifdef ARCH_IS_X86_64 | ||
| 750 : | lea _EDI, [mmx_mul] | ||
| 751 : | lea _EDI, [_EDI + _EAX*8 - 8] ; 2*quant | ||
| 752 : | %else | ||
| 753 : | lea _EDI, [mmx_mul + _EAX*8 - 8] ; 2*quant | ||
| 754 : | %endif | ||
| 755 : | push _EBP | ||
| 756 : | mov _EBX, mmx_2047 | ||
| 757 : | movsx _EBP, word [_ECX] | ||
| 758 : | %ifdef ARCH_IS_X86_64 | ||
| 759 : | lea r9, [mmx_add] | ||
| 760 : | lea _EAX, [r9 + _EAX*8 - 8] ; quant or quant-1 | ||
| 761 : | %else | ||
| 762 : | lea _EAX, [mmx_add + _EAX*8 - 8] ; quant or quant-1 | ||
| 763 : | %endif | ||
| 764 : | push _ESI | ||
| 765 : | mov _ESI, mmzero | ||
| 766 : | edgomez | 1382 | pxor mm7, mm7 |
| 767 : | Isibaar | 1795 | movq mm3, [_ECX+120] ;B2 ; c = coeff[i] |
| 768 : | pcmpeqw mm7, [_ECX+120] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 769 : | edgomez | 1382 | |
| 770 : | Isibaar | 1795 | imul _EBP, [_ESP+(4+4)*PTR_SIZE] ; dcscalar |
| 771 : | edgomez | 1382 | psubw mm2, mm3 ;-c ;B3 (1st dep) |
| 772 : | pmaxsw mm2, mm3 ;|c| ;B4 (2nd) | ||
| 773 : | Isibaar | 1795 | pmullw mm2, [_EDI] ;*= 2Q ;B8 (3rd+) |
| 774 : | edgomez | 1382 | psraw mm3, 15 ; sign(c) ;B7 (2nd) |
| 775 : | Isibaar | 1795 | mov _EDX, [_ESP+ (1+4)*PTR_SIZE] ; data |
| 776 : | edgomez | 1382 | |
| 777 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 778 : | edgomez | 1382 | dequant 0 |
| 779 : | |||
| 780 : | Isibaar | 1795 | cmp _EBP, -2048 |
| 781 : | mov _ESP, _ESP | ||
| 782 : | edgomez | 1382 | |
| 783 : | dequant 1 | ||
| 784 : | |||
| 785 : | Isibaar | 1795 | cmovl _EBP, [int_2048] |
| 786 : | edgomez | 1382 | nop |
| 787 : | |||
| 788 : | dequant 2 | ||
| 789 : | |||
| 790 : | Isibaar | 1795 | cmp _EBP, 2047 |
| 791 : | mov _ESP, _ESP | ||
| 792 : | edgomez | 1382 | |
| 793 : | dequant 3 | ||
| 794 : | |||
| 795 : | Isibaar | 1795 | cmovg _EBP, [int2047] |
| 796 : | edgomez | 1382 | nop |
| 797 : | |||
| 798 : | dequant 4 | ||
| 799 : | |||
| 800 : | paddw mm4, mm6 ;C11 mm6 free (4th+) | ||
| 801 : | Isibaar | 1795 | pminsw mm4, [_EBX] ;C12 saturates to +2047 (5th+) |
| 802 : | pandn mm7, [_EAX] ;B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 803 : | mov _EAX, _EBP | ||
| 804 : | mov _ESI, [_ESP] | ||
| 805 : | mov _EBP, [_ESP+PTR_SIZE] | ||
| 806 : | edgomez | 1382 | pxor mm5, mm4 ;C13 (6th+) |
| 807 : | paddw mm7, mm3 ;B10 offset +negate back (3rd) | ||
| 808 : | Isibaar | 1795 | movq [_EDX+4*24+16], mm5 ;C14 (7th) |
| 809 : | edgomez | 1382 | paddw mm2, mm7 ;B11 mm7 free (4th+) |
| 810 : | Isibaar | 1795 | pminsw mm2, [_EBX] ;B12 saturates to +2047 (5th+) |
| 811 : | mov _EBX, [_ESP+2*PTR_SIZE] | ||
| 812 : | mov _EDI, [_ESP+3*PTR_SIZE] | ||
| 813 : | add _ESP, byte 4*PTR_SIZE | ||
| 814 : | edgomez | 1382 | pxor mm3, mm2 ;B13 (6th+) |
| 815 : | Isibaar | 1795 | movq [_EDX+4*24+8], mm3 ;B14 (7th) |
| 816 : | mov [_EDX], ax | ||
| 817 : | edgomez | 1382 | |
| 818 : | Isibaar | 1795 | xor _EAX, _EAX |
| 819 : | |||
| 820 : | %ifdef ARCH_IS_X86_64 | ||
| 821 : | mov TMP0, [_ESP] | ||
| 822 : | %ifndef WINDOWS | ||
| 823 : | add _ESP, 6*PTR_SIZE | ||
| 824 : | %else | ||
| 825 : | add _ESP, 4*PTR_SIZE | ||
| 826 : | %endif | ||
| 827 : | mov [_ESP], TMP0 | ||
| 828 : | %endif | ||
| 829 : | |||
| 830 : | edgomez | 1382 | ret |
| 831 : | Isibaar | 1793 | ENDFUNC |
| 832 : | edgomez | 1382 | |
| 833 : | Isibaar | 1795 | |
| 834 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
| 835 : | ; | ||
| 836 : | ; uint32_t dequant_h263_inter_3dne(int16_t * data, | ||
| 837 : | ; const int16_t * const coeff, | ||
| 838 : | ; const uint32_t quant, | ||
| 839 : | ; const uint16_t *mpeg_matrices); | ||
| 840 : | ; | ||
| 841 : | ;----------------------------------------------------------------------------- | ||
| 842 : | |||
| 843 : | ; this is the same as dequant_inter_3dne, | ||
| 844 : | ; except that we're saturating using 'pminsw' (saves 2 cycles/loop) | ||
| 845 : | ; This is Athlon-optimized code (ca 100 clk per call) | ||
| 846 : | |||
| 847 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 848 : | edgomez | 1382 | cglobal dequant_h263_inter_3dne |
| 849 : | dequant_h263_inter_3dne: | ||
| 850 : | Isibaar | 1795 | |
| 851 : | %ifdef ARCH_IS_X86_64 | ||
| 852 : | mov TMP0, [_ESP] | ||
| 853 : | add _ESP, PTR_SIZE | ||
| 854 : | %ifndef WINDOWS | ||
| 855 : | push prm6 | ||
| 856 : | push prm5 | ||
| 857 : | %endif | ||
| 858 : | push prm4 | ||
| 859 : | push prm3 | ||
| 860 : | push prm2 | ||
| 861 : | push prm1 | ||
| 862 : | sub _ESP, PTR_SIZE | ||
| 863 : | mov [_ESP], TMP0 | ||
| 864 : | %endif | ||
| 865 : | |||
| 866 : | mov _ECX, [_ESP+ 2*PTR_SIZE] ; coeff | ||
| 867 : | mov _EAX, [_ESP+ 3*PTR_SIZE] ; quant | ||
| 868 : | edgomez | 1382 | pxor mm0, mm0 |
| 869 : | pxor mm2, mm2 | ||
| 870 : | Isibaar | 1795 | push _EDI |
| 871 : | push _EBX | ||
| 872 : | push _ESI | ||
| 873 : | %ifdef ARCH_IS_X86_64 | ||
| 874 : | lea _EDI, [mmx_mul] | ||
| 875 : | lea _EDI, [_EDI + _EAX*8 - 8] ; 2*quant | ||
| 876 : | %else | ||
| 877 : | lea _EDI, [mmx_mul + _EAX*8 - 8] ; 2*quant | ||
| 878 : | %endif | ||
| 879 : | mov _EBX, mmx_2047 | ||
| 880 : | edgomez | 1382 | pxor mm7, mm7 |
| 881 : | Isibaar | 1795 | movq mm3, [_ECX+120] ;B2 ; c = coeff[i] |
| 882 : | pcmpeqw mm7, [_ECX+120] ;B6 (c ==0) ? -1 : 0 (1st) | ||
| 883 : | %ifdef ARCH_IS_X86_64 | ||
| 884 : | lea r9, [mmx_add] | ||
| 885 : | lea _EAX, [r9 + _EAX*8 - 8] ; quant or quant-1 | ||
| 886 : | %else | ||
| 887 : | lea _EAX, [mmx_add + _EAX*8 - 8] ; quant or quant-1 | ||
| 888 : | %endif | ||
| 889 : | edgomez | 1382 | psubw mm2, mm3 ;-c ;B3 (1st dep) |
| 890 : | Isibaar | 1795 | mov _ESI, mmzero |
| 891 : | edgomez | 1382 | pmaxsw mm2, mm3 ;|c| ;B4 (2nd) |
| 892 : | Isibaar | 1795 | pmullw mm2, [_EDI] ;*= 2Q ;B8 (3rd+) |
| 893 : | edgomez | 1382 | psraw mm3, 15 ; sign(c) ;B7 (2nd) |
| 894 : | Isibaar | 1795 | mov _EDX, [_ESP+ (1+3)*PTR_SIZE] ; data |
| 895 : | edgomez | 1382 | |
| 896 : | Isibaar | 1795 | ALIGN SECTION_ALIGN |
| 897 : | edgomez | 1382 | |
| 898 : | dequant 0 | ||
| 899 : | dequant 1 | ||
| 900 : | dequant 2 | ||
| 901 : | dequant 3 | ||
| 902 : | dequant 4 | ||
| 903 : | |||
| 904 : | paddw mm4, mm6 ;C11 mm6 free (4th+) | ||
| 905 : | Isibaar | 1795 | pminsw mm4, [_EBX] ;C12 saturates to +2047 (5th+) |
| 906 : | pandn mm7, [_EAX] ;B9 offset = isZero ? 0 : quant_add (2nd) | ||
| 907 : | mov _ESI, [_ESP] | ||
| 908 : | edgomez | 1382 | pxor mm5, mm4 ;C13 (6th+) |
| 909 : | paddw mm7, mm3 ;B10 offset +negate back (3rd) | ||
| 910 : | Isibaar | 1795 | movq [_EDX+4*24+16], mm5 ;C14 (7th) |
| 911 : | edgomez | 1382 | paddw mm2, mm7 ;B11 mm7 free (4th+) |
| 912 : | Isibaar | 1795 | pminsw mm2, [_EBX] ;B12 saturates to +2047 (5th+) |
| 913 : | mov _EBX, [_ESP+PTR_SIZE] | ||
| 914 : | mov _EDI, [_ESP+2*PTR_SIZE] | ||
| 915 : | add _ESP, byte 3*PTR_SIZE | ||
| 916 : | edgomez | 1382 | pxor mm3, mm2 ;B13 (6th+) |
| 917 : | Isibaar | 1795 | movq [_EDX+4*24+8], mm3 ;B14 (7th) |
| 918 : | edgomez | 1382 | |
| 919 : | Isibaar | 1795 | xor _EAX, _EAX |
| 920 : | |||
| 921 : | %ifdef ARCH_IS_X86_64 | ||
| 922 : | mov TMP0, [_ESP] | ||
| 923 : | %ifndef WINDOWS | ||
| 924 : | add _ESP, 6*PTR_SIZE | ||
| 925 : | %else | ||
| 926 : | add _ESP, 4*PTR_SIZE | ||
| 927 : | %endif | ||
| 928 : | mov [_ESP], TMP0 | ||
| 929 : | %endif | ||
| 930 : | |||
| 931 : | edgomez | 1382 | ret |
| 932 : | Isibaar | 1793 | ENDFUNC |
| 933 : | edgomez | 1540 | |
| 934 : | Isibaar | 1790 | %ifidn __OUTPUT_FORMAT__,elf |
| 935 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 936 : | %endif | ||
| 937 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |