1 ;***************************************************************************** 2 ;* x86inc.asm 3 ;***************************************************************************** 4 ;* Copyright (C) 2005-2011 x264 project 5 ;* 6 ;* Authors: Loren Merritt <lorenm (a] u.washington.edu> 7 ;* Anton Mitrofanov <BugMaster (a] narod.ru> 8 ;* Jason Garrett-Glaser <darkshikari (a] gmail.com> 9 ;* 10 ;* Permission to use, copy, modify, and/or distribute this software for any 11 ;* purpose with or without fee is hereby granted, provided that the above 12 ;* copyright notice and this permission notice appear in all copies. 13 ;* 14 ;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 15 ;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 16 ;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 17 ;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 18 ;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 19 ;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 20 ;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 21 ;***************************************************************************** 22 23 ; This is a header file for the x264ASM assembly language, which uses 24 ; NASM/YASM syntax combined with a large number of macros to provide easy 25 ; abstraction between different calling conventions (x86_32, win64, linux64). 26 ; It also has various other useful features to simplify writing the kind of 27 ; DSP functions that are most often used in x264. 28 29 ; Unlike the rest of x264, this file is available under an ISC license, as it 30 ; has significant usefulness outside of x264 and we want it to be available 31 ; to the largest audience possible. Of course, if you modify it for your own 32 ; purposes to add a new feature, we strongly encourage contributing a patch 33 ; as this feature might be useful for others as well. Send patches or ideas 34 ; to x264-devel (a] videolan.org . 35 36 %ifndef THIRD_PARTY_X86INC_X86INC_ASM_ 37 %define THIRD_PARTY_X86INC_X86INC_ASM_ 38 39 ; TODO(wolenetz): Consider either updating this customized version to base from 40 ; a more recent original, or switching to using third_party/ffmpeg's version of 41 ; this abstraction layer. See http://crbug.com/175029 42 43 %define program_name ff 44 45 %ifdef ARCH_X86_64 46 %ifidn __OUTPUT_FORMAT__,win32 47 %define WIN64 48 %elifidn __OUTPUT_FORMAT__,win64 49 %define WIN64 50 %else 51 %define UNIX64 52 %endif 53 %endif 54 55 %ifdef PREFIX 56 %define mangle(x) _ %+ x 57 %else 58 %define mangle(x) x 59 %endif 60 61 ; FIXME: All of the 64bit asm functions that take a stride as an argument 62 ; via register, assume that the high dword of that register is filled with 0. 63 ; This is true in practice (since we never do any 64bit arithmetic on strides, 64 ; and x264's strides are all positive), but is not guaranteed by the ABI. 65 66 ; Name of the .rodata section. 67 ; Kludge: Something on OS X fails to align .rodata even given an align attribute, 68 ; so use a different read-only section. 69 %ifdef CHROMIUM 70 %macro SECTION_RODATA 0-1 16 71 %ifidn __OUTPUT_FORMAT__,macho64 72 SECTION .text align=%1 73 %elifidn __OUTPUT_FORMAT__,macho 74 SECTION .text align=%1 75 fakegot: 76 %elifidn __OUTPUT_FORMAT__,aout 77 section .text 78 %else 79 SECTION .rodata align=%1 80 %endif 81 %endmacro 82 %else 83 %macro SECTION_RODATA 0-1 16 84 %ifidn __OUTPUT_FORMAT__,aout 85 section .text 86 %else 87 SECTION .rodata align=%1 88 %endif 89 %endmacro 90 %endif 91 92 ; aout does not support align= 93 %macro SECTION_TEXT 0-1 16 94 %ifidn __OUTPUT_FORMAT__,aout 95 SECTION .text 96 %else 97 SECTION .text align=%1 98 %endif 99 %endmacro 100 101 %ifdef WIN64 102 %define PIC 103 %elifndef ARCH_X86_64 104 ; For chromium we may build PIC code even for 32 bits system. 105 %ifndef CHROMIUM 106 ; x86_32 doesn't require PIC. 107 ; Some distros prefer shared objects to be PIC, but nothing breaks if 108 ; the code contains a few textrels, so we'll skip that complexity. 109 %undef PIC 110 %endif 111 %endif 112 %ifdef PIC 113 default rel 114 %endif 115 116 ; Macros to eliminate most code duplication between x86_32 and x86_64: 117 ; Currently this works only for leaf functions which load all their arguments 118 ; into registers at the start, and make no other use of the stack. Luckily that 119 ; covers most of x264's asm. 120 121 ; PROLOGUE: 122 ; %1 = number of arguments. loads them from stack if needed. 123 ; %2 = number of registers used. pushes callee-saved regs if needed. 124 ; %3 = number of xmm registers used. pushes callee-saved xmm regs if needed. 125 ; %4 = list of names to define to registers 126 ; PROLOGUE can also be invoked by adding the same options to cglobal 127 128 ; e.g. 129 ; cglobal foo, 2,3,0, dst, src, tmp 130 ; declares a function (foo), taking two args (dst and src) and one local variable (tmp) 131 132 ; TODO Some functions can use some args directly from the stack. If they're the 133 ; last args then you can just not declare them, but if they're in the middle 134 ; we need more flexible macro. 135 136 ; RET: 137 ; Pops anything that was pushed by PROLOGUE 138 139 ; REP_RET: 140 ; Same, but if it doesn't pop anything it becomes a 2-byte ret, for athlons 141 ; which are slow when a normal ret follows a branch. 142 143 ; registers: 144 ; rN and rNq are the native-size register holding function argument N 145 ; rNd, rNw, rNb are dword, word, and byte size 146 ; rNm is the original location of arg N (a register or on the stack), dword 147 ; rNmp is native size 148 149 %macro DECLARE_REG 6 150 %define r%1q %2 151 %define r%1d %3 152 %define r%1w %4 153 %define r%1b %5 154 %define r%1m %6 155 %ifid %6 ; i.e. it's a register 156 %define r%1mp %2 157 %elifdef ARCH_X86_64 ; memory 158 %define r%1mp qword %6 159 %else 160 %define r%1mp dword %6 161 %endif 162 %define r%1 %2 163 %endmacro 164 165 %macro DECLARE_REG_SIZE 2 166 %define r%1q r%1 167 %define e%1q r%1 168 %define r%1d e%1 169 %define e%1d e%1 170 %define r%1w %1 171 %define e%1w %1 172 %define r%1b %2 173 %define e%1b %2 174 %ifndef ARCH_X86_64 175 %define r%1 e%1 176 %endif 177 %endmacro 178 179 DECLARE_REG_SIZE ax, al 180 DECLARE_REG_SIZE bx, bl 181 DECLARE_REG_SIZE cx, cl 182 DECLARE_REG_SIZE dx, dl 183 DECLARE_REG_SIZE si, sil 184 DECLARE_REG_SIZE di, dil 185 DECLARE_REG_SIZE bp, bpl 186 187 ; t# defines for when per-arch register allocation is more complex than just function arguments 188 189 %macro DECLARE_REG_TMP 1-* 190 %assign %%i 0 191 %rep %0 192 CAT_XDEFINE t, %%i, r%1 193 %assign %%i %%i+1 194 %rotate 1 195 %endrep 196 %endmacro 197 198 %macro DECLARE_REG_TMP_SIZE 0-* 199 %rep %0 200 %define t%1q t%1 %+ q 201 %define t%1d t%1 %+ d 202 %define t%1w t%1 %+ w 203 %define t%1b t%1 %+ b 204 %rotate 1 205 %endrep 206 %endmacro 207 208 DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9 209 210 %ifdef ARCH_X86_64 211 %define gprsize 8 212 %else 213 %define gprsize 4 214 %endif 215 216 %macro PUSH 1 217 push %1 218 %assign stack_offset stack_offset+gprsize 219 %endmacro 220 221 %macro POP 1 222 pop %1 223 %assign stack_offset stack_offset-gprsize 224 %endmacro 225 226 %macro SUB 2 227 sub %1, %2 228 %ifidn %1, rsp 229 %assign stack_offset stack_offset+(%2) 230 %endif 231 %endmacro 232 233 %macro ADD 2 234 add %1, %2 235 %ifidn %1, rsp 236 %assign stack_offset stack_offset-(%2) 237 %endif 238 %endmacro 239 240 %macro movifnidn 2 241 %ifnidn %1, %2 242 mov %1, %2 243 %endif 244 %endmacro 245 246 %macro movsxdifnidn 2 247 %ifnidn %1, %2 248 movsxd %1, %2 249 %endif 250 %endmacro 251 252 %macro ASSERT 1 253 %if (%1) == 0 254 %error assert failed 255 %endif 256 %endmacro 257 258 %macro DEFINE_ARGS 0-* 259 %ifdef n_arg_names 260 %assign %%i 0 261 %rep n_arg_names 262 CAT_UNDEF arg_name %+ %%i, q 263 CAT_UNDEF arg_name %+ %%i, d 264 CAT_UNDEF arg_name %+ %%i, w 265 CAT_UNDEF arg_name %+ %%i, b 266 CAT_UNDEF arg_name %+ %%i, m 267 CAT_UNDEF arg_name, %%i 268 %assign %%i %%i+1 269 %endrep 270 %endif 271 272 %assign %%i 0 273 %rep %0 274 %xdefine %1q r %+ %%i %+ q 275 %xdefine %1d r %+ %%i %+ d 276 %xdefine %1w r %+ %%i %+ w 277 %xdefine %1b r %+ %%i %+ b 278 %xdefine %1m r %+ %%i %+ m 279 CAT_XDEFINE arg_name, %%i, %1 280 %assign %%i %%i+1 281 %rotate 1 282 %endrep 283 %assign n_arg_names %%i 284 %endmacro 285 286 %ifdef WIN64 ; Windows x64 ;================================================= 287 288 DECLARE_REG 0, rcx, ecx, cx, cl, ecx 289 DECLARE_REG 1, rdx, edx, dx, dl, edx 290 DECLARE_REG 2, r8, r8d, r8w, r8b, r8d 291 DECLARE_REG 3, r9, r9d, r9w, r9b, r9d 292 DECLARE_REG 4, rdi, edi, di, dil, [rsp + stack_offset + 40] 293 DECLARE_REG 5, rsi, esi, si, sil, [rsp + stack_offset + 48] 294 DECLARE_REG 6, rax, eax, ax, al, [rsp + stack_offset + 56] 295 %define r7m [rsp + stack_offset + 64] 296 %define r8m [rsp + stack_offset + 72] 297 298 %macro LOAD_IF_USED 2 ; reg_id, number_of_args 299 %if %1 < %2 300 mov r%1, [rsp + stack_offset + 8 + %1*8] 301 %endif 302 %endmacro 303 304 %macro PROLOGUE 2-4+ 0 ; #args, #regs, #xmm_regs, arg_names... 305 ASSERT %2 >= %1 306 %assign regs_used %2 307 ASSERT regs_used <= 7 308 %if regs_used > 4 309 push r4 310 push r5 311 %assign stack_offset stack_offset+16 312 %endif 313 WIN64_SPILL_XMM %3 314 LOAD_IF_USED 4, %1 315 LOAD_IF_USED 5, %1 316 LOAD_IF_USED 6, %1 317 DEFINE_ARGS %4 318 %endmacro 319 320 %macro WIN64_SPILL_XMM 1 321 %assign xmm_regs_used %1 322 ASSERT xmm_regs_used <= 16 323 %if xmm_regs_used > 6 324 sub rsp, (xmm_regs_used-6)*16+16 325 %assign stack_offset stack_offset+(xmm_regs_used-6)*16+16 326 %assign %%i xmm_regs_used 327 %rep (xmm_regs_used-6) 328 %assign %%i %%i-1 329 movdqa [rsp + (%%i-6)*16+8], xmm %+ %%i 330 %endrep 331 %endif 332 %endmacro 333 334 %macro WIN64_RESTORE_XMM_INTERNAL 1 335 %if xmm_regs_used > 6 336 %assign %%i xmm_regs_used 337 %rep (xmm_regs_used-6) 338 %assign %%i %%i-1 339 movdqa xmm %+ %%i, [%1 + (%%i-6)*16+8] 340 %endrep 341 add %1, (xmm_regs_used-6)*16+16 342 %endif 343 %endmacro 344 345 %macro WIN64_RESTORE_XMM 1 346 WIN64_RESTORE_XMM_INTERNAL %1 347 %assign stack_offset stack_offset-(xmm_regs_used-6)*16+16 348 %assign xmm_regs_used 0 349 %endmacro 350 351 %macro RET 0 352 WIN64_RESTORE_XMM_INTERNAL rsp 353 %if regs_used > 4 354 pop r5 355 pop r4 356 %endif 357 ret 358 %endmacro 359 360 %macro REP_RET 0 361 %if regs_used > 4 || xmm_regs_used > 6 362 RET 363 %else 364 rep ret 365 %endif 366 %endmacro 367 368 %elifdef ARCH_X86_64 ; *nix x64 ;============================================= 369 370 DECLARE_REG 0, rdi, edi, di, dil, edi 371 DECLARE_REG 1, rsi, esi, si, sil, esi 372 DECLARE_REG 2, rdx, edx, dx, dl, edx 373 DECLARE_REG 3, rcx, ecx, cx, cl, ecx 374 DECLARE_REG 4, r8, r8d, r8w, r8b, r8d 375 DECLARE_REG 5, r9, r9d, r9w, r9b, r9d 376 DECLARE_REG 6, rax, eax, ax, al, [rsp + stack_offset + 8] 377 %define r7m [rsp + stack_offset + 16] 378 %define r8m [rsp + stack_offset + 24] 379 380 %macro LOAD_IF_USED 2 ; reg_id, number_of_args 381 %if %1 < %2 382 mov r%1, [rsp - 40 + %1*8] 383 %endif 384 %endmacro 385 386 %macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names... 387 ASSERT %2 >= %1 388 ASSERT %2 <= 7 389 LOAD_IF_USED 6, %1 390 DEFINE_ARGS %4 391 %endmacro 392 393 %macro RET 0 394 ret 395 %endmacro 396 397 %macro REP_RET 0 398 rep ret 399 %endmacro 400 401 %else ; X86_32 ;============================================================== 402 403 ; Begin chromium edits 404 %ifdef CHROMIUM 405 ; Change the order of registers so we can get the lower 8-bit or the 5th and 6th 406 ; arguments. 407 DECLARE_REG 0, esi, esi, si, null, [esp + stack_offset + 4] 408 DECLARE_REG 1, edi, edi, di, null, [esp + stack_offset + 8] 409 DECLARE_REG 2, ecx, ecx, cx, cl, [esp + stack_offset + 12] 410 DECLARE_REG 3, edx, edx, dx, dl, [esp + stack_offset + 16] 411 DECLARE_REG 4, eax, eax, ax, al, [esp + stack_offset + 20] 412 DECLARE_REG 5, ebx, ebx, bx, bl, [esp + stack_offset + 24] 413 %else 414 DECLARE_REG 0, eax, eax, ax, al, [esp + stack_offset + 4] 415 DECLARE_REG 1, ecx, ecx, cx, cl, [esp + stack_offset + 8] 416 DECLARE_REG 2, edx, edx, dx, dl, [esp + stack_offset + 12] 417 DECLARE_REG 3, ebx, ebx, bx, bl, [esp + stack_offset + 16] 418 DECLARE_REG 4, esi, esi, si, null, [esp + stack_offset + 20] 419 DECLARE_REG 5, edi, edi, di, null, [esp + stack_offset + 24] 420 %endif 421 ; End chromium edits 422 DECLARE_REG 6, ebp, ebp, bp, null, [esp + stack_offset + 28] 423 %define r7m [esp + stack_offset + 32] 424 %define r8m [esp + stack_offset + 36] 425 %define rsp esp 426 427 %macro PUSH_IF_USED 1 ; reg_id 428 %if %1 < regs_used 429 push r%1 430 %assign stack_offset stack_offset+4 431 %endif 432 %endmacro 433 434 %macro POP_IF_USED 1 ; reg_id 435 %if %1 < regs_used 436 pop r%1 437 %endif 438 %endmacro 439 440 %macro LOAD_IF_USED 2 ; reg_id, number_of_args 441 %if %1 < %2 442 mov r%1, [esp + stack_offset + 4 + %1*4] 443 %endif 444 %endmacro 445 446 %macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names... 447 ASSERT %2 >= %1 448 %assign regs_used %2 449 ASSERT regs_used <= 7 450 %ifdef CHROMIUM 451 PUSH_IF_USED 0 452 PUSH_IF_USED 1 453 PUSH_IF_USED 5 454 %else 455 PUSH_IF_USED 3 456 PUSH_IF_USED 4 457 PUSH_IF_USED 5 458 %endif 459 PUSH_IF_USED 6 460 LOAD_IF_USED 0, %1 461 LOAD_IF_USED 1, %1 462 LOAD_IF_USED 2, %1 463 LOAD_IF_USED 3, %1 464 LOAD_IF_USED 4, %1 465 LOAD_IF_USED 5, %1 466 LOAD_IF_USED 6, %1 467 DEFINE_ARGS %4 468 %endmacro 469 470 %macro RET 0 471 POP_IF_USED 6 472 %ifdef CHROMIUM 473 POP_IF_USED 5 474 POP_IF_USED 1 475 POP_IF_USED 0 476 %else 477 POP_IF_USED 5 478 POP_IF_USED 4 479 POP_IF_USED 3 480 %endif 481 ret 482 %endmacro 483 484 %macro REP_RET 0 485 %if regs_used > 3 486 RET 487 %else 488 rep ret 489 %endif 490 %endmacro 491 492 %endif ;====================================================================== 493 494 %ifndef WIN64 495 %macro WIN64_SPILL_XMM 1 496 %endmacro 497 %macro WIN64_RESTORE_XMM 1 498 %endmacro 499 %endif 500 501 502 503 ;============================================================================= 504 ; arch-independent part 505 ;============================================================================= 506 507 %assign function_align 16 508 509 ; Symbol prefix for C linkage 510 %macro cglobal 1-2+ 511 %xdefine %1 mangle(program_name %+ _ %+ %1) 512 %xdefine %1.skip_prologue %1 %+ .skip_prologue 513 %ifidn __OUTPUT_FORMAT__,elf 514 global %1:function hidden 515 %else 516 global %1 517 %endif 518 align function_align 519 %1: 520 RESET_MM_PERMUTATION ; not really needed, but makes disassembly somewhat nicer 521 %assign stack_offset 0 522 %if %0 > 1 523 PROLOGUE %2 524 %endif 525 %endmacro 526 527 %macro cextern 1 528 %xdefine %1 mangle(program_name %+ _ %+ %1) 529 extern %1 530 %endmacro 531 532 ;like cextern, but without the prefix 533 %macro cextern_naked 1 534 %xdefine %1 mangle(%1) 535 extern %1 536 %endmacro 537 538 %macro const 2+ 539 %xdefine %1 mangle(program_name %+ _ %+ %1) 540 global %1 541 %1: %2 542 %endmacro 543 544 ; This is needed for ELF, otherwise the GNU linker assumes the stack is 545 ; executable by default. 546 %ifidn __OUTPUT_FORMAT__,elf 547 SECTION .note.GNU-stack noalloc noexec nowrite progbits 548 %endif 549 550 ; merge mmx and sse* 551 552 %macro CAT_XDEFINE 3 553 %xdefine %1%2 %3 554 %endmacro 555 556 %macro CAT_UNDEF 2 557 %undef %1%2 558 %endmacro 559 560 %macro INIT_MMX 0 561 %assign avx_enabled 0 562 %define RESET_MM_PERMUTATION INIT_MMX 563 %define mmsize 8 564 %define num_mmregs 8 565 %define mova movq 566 %define movu movq 567 %define movh movd 568 %define movnta movntq 569 %assign %%i 0 570 %rep 8 571 CAT_XDEFINE m, %%i, mm %+ %%i 572 CAT_XDEFINE nmm, %%i, %%i 573 %assign %%i %%i+1 574 %endrep 575 %rep 8 576 CAT_UNDEF m, %%i 577 CAT_UNDEF nmm, %%i 578 %assign %%i %%i+1 579 %endrep 580 %endmacro 581 582 %macro INIT_XMM 0 583 %assign avx_enabled 0 584 %define RESET_MM_PERMUTATION INIT_XMM 585 %define mmsize 16 586 %define num_mmregs 8 587 %ifdef ARCH_X86_64 588 %define num_mmregs 16 589 %endif 590 %define mova movdqa 591 %define movu movdqu 592 %define movh movq 593 %define movnta movntdq 594 %assign %%i 0 595 %rep num_mmregs 596 CAT_XDEFINE m, %%i, xmm %+ %%i 597 CAT_XDEFINE nxmm, %%i, %%i 598 %assign %%i %%i+1 599 %endrep 600 %endmacro 601 602 %macro INIT_AVX 0 603 INIT_XMM 604 %assign avx_enabled 1 605 %define PALIGNR PALIGNR_SSSE3 606 %define RESET_MM_PERMUTATION INIT_AVX 607 %endmacro 608 609 %macro INIT_YMM 0 610 %assign avx_enabled 1 611 %define RESET_MM_PERMUTATION INIT_YMM 612 %define mmsize 32 613 %define num_mmregs 8 614 %ifdef ARCH_X86_64 615 %define num_mmregs 16 616 %endif 617 %define mova vmovaps 618 %define movu vmovups 619 %assign %%i 0 620 %rep num_mmregs 621 CAT_XDEFINE m, %%i, ymm %+ %%i 622 CAT_XDEFINE nymm, %%i, %%i 623 %assign %%i %%i+1 624 %endrep 625 %endmacro 626 627 INIT_MMX 628 629 ; I often want to use macros that permute their arguments. e.g. there's no 630 ; efficient way to implement butterfly or transpose or dct without swapping some 631 ; arguments. 632 ; 633 ; I would like to not have to manually keep track of the permutations: 634 ; If I insert a permutation in the middle of a function, it should automatically 635 ; change everything that follows. For more complex macros I may also have multiple 636 ; implementations, e.g. the SSE2 and SSSE3 versions may have different permutations. 637 ; 638 ; Hence these macros. Insert a PERMUTE or some SWAPs at the end of a macro that 639 ; permutes its arguments. It's equivalent to exchanging the contents of the 640 ; registers, except that this way you exchange the register names instead, so it 641 ; doesn't cost any cycles. 642 643 %macro PERMUTE 2-* ; takes a list of pairs to swap 644 %rep %0/2 645 %xdefine tmp%2 m%2 646 %xdefine ntmp%2 nm%2 647 %rotate 2 648 %endrep 649 %rep %0/2 650 %xdefine m%1 tmp%2 651 %xdefine nm%1 ntmp%2 652 %undef tmp%2 653 %undef ntmp%2 654 %rotate 2 655 %endrep 656 %endmacro 657 658 %macro SWAP 2-* ; swaps a single chain (sometimes more concise than pairs) 659 %rep %0-1 660 %ifdef m%1 661 %xdefine tmp m%1 662 %xdefine m%1 m%2 663 %xdefine m%2 tmp 664 CAT_XDEFINE n, m%1, %1 665 CAT_XDEFINE n, m%2, %2 666 %else 667 ; If we were called as "SWAP m0,m1" rather than "SWAP 0,1" infer the original numbers here. 668 ; Be careful using this mode in nested macros though, as in some cases there may be 669 ; other copies of m# that have already been dereferenced and don't get updated correctly. 670 %xdefine %%n1 n %+ %1 671 %xdefine %%n2 n %+ %2 672 %xdefine tmp m %+ %%n1 673 CAT_XDEFINE m, %%n1, m %+ %%n2 674 CAT_XDEFINE m, %%n2, tmp 675 CAT_XDEFINE n, m %+ %%n1, %%n1 676 CAT_XDEFINE n, m %+ %%n2, %%n2 677 %endif 678 %undef tmp 679 %rotate 1 680 %endrep 681 %endmacro 682 683 ; If SAVE_MM_PERMUTATION is placed at the end of a function and given the 684 ; function name, then any later calls to that function will automatically 685 ; load the permutation, so values can be returned in mmregs. 686 %macro SAVE_MM_PERMUTATION 1 ; name to save as 687 %assign %%i 0 688 %rep num_mmregs 689 CAT_XDEFINE %1_m, %%i, m %+ %%i 690 %assign %%i %%i+1 691 %endrep 692 %endmacro 693 694 %macro LOAD_MM_PERMUTATION 1 ; name to load from 695 %assign %%i 0 696 %rep num_mmregs 697 CAT_XDEFINE m, %%i, %1_m %+ %%i 698 CAT_XDEFINE n, m %+ %%i, %%i 699 %assign %%i %%i+1 700 %endrep 701 %endmacro 702 703 %macro call 1 704 call %1 705 %ifdef %1_m0 706 LOAD_MM_PERMUTATION %1 707 %endif 708 %endmacro 709 710 ; Substitutions that reduce instruction size but are functionally equivalent 711 %macro add 2 712 %ifnum %2 713 %if %2==128 714 sub %1, -128 715 %else 716 add %1, %2 717 %endif 718 %else 719 add %1, %2 720 %endif 721 %endmacro 722 723 %macro sub 2 724 %ifnum %2 725 %if %2==128 726 add %1, -128 727 %else 728 sub %1, %2 729 %endif 730 %else 731 sub %1, %2 732 %endif 733 %endmacro 734 735 ;============================================================================= 736 ; AVX abstraction layer 737 ;============================================================================= 738 739 %assign i 0 740 %rep 16 741 %if i < 8 742 CAT_XDEFINE sizeofmm, i, 8 743 %endif 744 CAT_XDEFINE sizeofxmm, i, 16 745 CAT_XDEFINE sizeofymm, i, 32 746 %assign i i+1 747 %endrep 748 %undef i 749 750 ;%1 == instruction 751 ;%2 == 1 if float, 0 if int 752 ;%3 == 0 if 3-operand (xmm, xmm, xmm), 1 if 4-operand (xmm, xmm, xmm, imm) 753 ;%4 == number of operands given 754 ;%5+: operands 755 %macro RUN_AVX_INSTR 6-7+ 756 %if sizeof%5==32 757 v%1 %5, %6, %7 758 %else 759 %if sizeof%5==8 760 %define %%regmov movq 761 %elif %2 762 %define %%regmov movaps 763 %else 764 %define %%regmov movdqa 765 %endif 766 767 %if %4>=3+%3 768 %ifnidn %5, %6 769 %if avx_enabled && sizeof%5==16 770 v%1 %5, %6, %7 771 %else 772 %%regmov %5, %6 773 %1 %5, %7 774 %endif 775 %else 776 %1 %5, %7 777 %endif 778 %elif %3 779 %1 %5, %6, %7 780 %else 781 %1 %5, %6 782 %endif 783 %endif 784 %endmacro 785 786 ;%1 == instruction 787 ;%2 == 1 if float, 0 if int 788 ;%3 == 0 if 3-operand (xmm, xmm, xmm), 1 if 4-operand (xmm, xmm, xmm, imm) 789 %macro AVX_INSTR 3 790 %macro %1 2-8 fnord, fnord, fnord, %1, %2, %3 791 %ifidn %3, fnord 792 RUN_AVX_INSTR %6, %7, %8, 2, %1, %2 793 %elifidn %4, fnord 794 RUN_AVX_INSTR %6, %7, %8, 3, %1, %2, %3 795 %elifidn %5, fnord 796 RUN_AVX_INSTR %6, %7, %8, 4, %1, %2, %3, %4 797 %else 798 RUN_AVX_INSTR %6, %7, %8, 5, %1, %2, %3, %4, %5 799 %endif 800 %endmacro 801 %endmacro 802 803 AVX_INSTR addpd, 1, 0 804 AVX_INSTR addps, 1, 0 805 AVX_INSTR addsd, 1, 0 806 AVX_INSTR addss, 1, 0 807 AVX_INSTR addsubpd, 1, 0 808 AVX_INSTR addsubps, 1, 0 809 AVX_INSTR andpd, 1, 0 810 AVX_INSTR andps, 1, 0 811 AVX_INSTR andnpd, 1, 0 812 AVX_INSTR andnps, 1, 0 813 AVX_INSTR blendpd, 1, 0 814 AVX_INSTR blendps, 1, 0 815 AVX_INSTR blendvpd, 1, 0 816 AVX_INSTR blendvps, 1, 0 817 AVX_INSTR cmppd, 1, 0 818 AVX_INSTR cmpps, 1, 0 819 AVX_INSTR cmpsd, 1, 0 820 AVX_INSTR cmpss, 1, 0 821 AVX_INSTR divpd, 1, 0 822 AVX_INSTR divps, 1, 0 823 AVX_INSTR divsd, 1, 0 824 AVX_INSTR divss, 1, 0 825 AVX_INSTR dppd, 1, 0 826 AVX_INSTR dpps, 1, 0 827 AVX_INSTR haddpd, 1, 0 828 AVX_INSTR haddps, 1, 0 829 AVX_INSTR hsubpd, 1, 0 830 AVX_INSTR hsubps, 1, 0 831 AVX_INSTR maxpd, 1, 0 832 AVX_INSTR maxps, 1, 0 833 AVX_INSTR maxsd, 1, 0 834 AVX_INSTR maxss, 1, 0 835 AVX_INSTR minpd, 1, 0 836 AVX_INSTR minps, 1, 0 837 AVX_INSTR minsd, 1, 0 838 AVX_INSTR minss, 1, 0 839 AVX_INSTR mpsadbw, 0, 1 840 AVX_INSTR mulpd, 1, 0 841 AVX_INSTR mulps, 1, 0 842 AVX_INSTR mulsd, 1, 0 843 AVX_INSTR mulss, 1, 0 844 AVX_INSTR orpd, 1, 0 845 AVX_INSTR orps, 1, 0 846 AVX_INSTR packsswb, 0, 0 847 AVX_INSTR packssdw, 0, 0 848 AVX_INSTR packuswb, 0, 0 849 AVX_INSTR packusdw, 0, 0 850 AVX_INSTR paddb, 0, 0 851 AVX_INSTR paddw, 0, 0 852 AVX_INSTR paddd, 0, 0 853 AVX_INSTR paddq, 0, 0 854 AVX_INSTR paddsb, 0, 0 855 AVX_INSTR paddsw, 0, 0 856 AVX_INSTR paddusb, 0, 0 857 AVX_INSTR paddusw, 0, 0 858 AVX_INSTR palignr, 0, 1 859 AVX_INSTR pand, 0, 0 860 AVX_INSTR pandn, 0, 0 861 AVX_INSTR pavgb, 0, 0 862 AVX_INSTR pavgw, 0, 0 863 AVX_INSTR pblendvb, 0, 0 864 AVX_INSTR pblendw, 0, 1 865 AVX_INSTR pcmpestri, 0, 0 866 AVX_INSTR pcmpestrm, 0, 0 867 AVX_INSTR pcmpistri, 0, 0 868 AVX_INSTR pcmpistrm, 0, 0 869 AVX_INSTR pcmpeqb, 0, 0 870 AVX_INSTR pcmpeqw, 0, 0 871 AVX_INSTR pcmpeqd, 0, 0 872 AVX_INSTR pcmpeqq, 0, 0 873 AVX_INSTR pcmpgtb, 0, 0 874 AVX_INSTR pcmpgtw, 0, 0 875 AVX_INSTR pcmpgtd, 0, 0 876 AVX_INSTR pcmpgtq, 0, 0 877 AVX_INSTR phaddw, 0, 0 878 AVX_INSTR phaddd, 0, 0 879 AVX_INSTR phaddsw, 0, 0 880 AVX_INSTR phsubw, 0, 0 881 AVX_INSTR phsubd, 0, 0 882 AVX_INSTR phsubsw, 0, 0 883 AVX_INSTR pmaddwd, 0, 0 884 AVX_INSTR pmaddubsw, 0, 0 885 AVX_INSTR pmaxsb, 0, 0 886 AVX_INSTR pmaxsw, 0, 0 887 AVX_INSTR pmaxsd, 0, 0 888 AVX_INSTR pmaxub, 0, 0 889 AVX_INSTR pmaxuw, 0, 0 890 AVX_INSTR pmaxud, 0, 0 891 AVX_INSTR pminsb, 0, 0 892 AVX_INSTR pminsw, 0, 0 893 AVX_INSTR pminsd, 0, 0 894 AVX_INSTR pminub, 0, 0 895 AVX_INSTR pminuw, 0, 0 896 AVX_INSTR pminud, 0, 0 897 AVX_INSTR pmulhuw, 0, 0 898 AVX_INSTR pmulhrsw, 0, 0 899 AVX_INSTR pmulhw, 0, 0 900 AVX_INSTR pmullw, 0, 0 901 AVX_INSTR pmulld, 0, 0 902 AVX_INSTR pmuludq, 0, 0 903 AVX_INSTR pmuldq, 0, 0 904 AVX_INSTR por, 0, 0 905 AVX_INSTR psadbw, 0, 0 906 AVX_INSTR pshufb, 0, 0 907 AVX_INSTR psignb, 0, 0 908 AVX_INSTR psignw, 0, 0 909 AVX_INSTR psignd, 0, 0 910 AVX_INSTR psllw, 0, 0 911 AVX_INSTR pslld, 0, 0 912 AVX_INSTR psllq, 0, 0 913 AVX_INSTR pslldq, 0, 0 914 AVX_INSTR psraw, 0, 0 915 AVX_INSTR psrad, 0, 0 916 AVX_INSTR psrlw, 0, 0 917 AVX_INSTR psrld, 0, 0 918 AVX_INSTR psrlq, 0, 0 919 AVX_INSTR psrldq, 0, 0 920 AVX_INSTR psubb, 0, 0 921 AVX_INSTR psubw, 0, 0 922 AVX_INSTR psubd, 0, 0 923 AVX_INSTR psubq, 0, 0 924 AVX_INSTR psubsb, 0, 0 925 AVX_INSTR psubsw, 0, 0 926 AVX_INSTR psubusb, 0, 0 927 AVX_INSTR psubusw, 0, 0 928 AVX_INSTR punpckhbw, 0, 0 929 AVX_INSTR punpckhwd, 0, 0 930 AVX_INSTR punpckhdq, 0, 0 931 AVX_INSTR punpckhqdq, 0, 0 932 AVX_INSTR punpcklbw, 0, 0 933 AVX_INSTR punpcklwd, 0, 0 934 AVX_INSTR punpckldq, 0, 0 935 AVX_INSTR punpcklqdq, 0, 0 936 AVX_INSTR pxor, 0, 0 937 AVX_INSTR shufps, 0, 1 938 AVX_INSTR subpd, 1, 0 939 AVX_INSTR subps, 1, 0 940 AVX_INSTR subsd, 1, 0 941 AVX_INSTR subss, 1, 0 942 AVX_INSTR unpckhpd, 1, 0 943 AVX_INSTR unpckhps, 1, 0 944 AVX_INSTR unpcklpd, 1, 0 945 AVX_INSTR unpcklps, 1, 0 946 AVX_INSTR xorpd, 1, 0 947 AVX_INSTR xorps, 1, 0 948 949 ; 3DNow instructions, for sharing code between AVX, SSE and 3DN 950 AVX_INSTR pfadd, 1, 0 951 AVX_INSTR pfsub, 1, 0 952 AVX_INSTR pfmul, 1, 0 953 954 ;============================================================================= 955 ; Chromium extensions 956 ;============================================================================= 957 958 %ifdef CHROMIUM 959 ; Always build PIC code on Mac or Linux for Chromium. 960 %ifdef MACHO 961 %define PIC 962 %endif 963 %ifdef ELF 964 %define PIC 965 %endif 966 967 ; 968 ; LOAD_SYM %1 (reg), %2 (sym) 969 ; Copies the address to a local symbol to the specified register. 970 ; 971 972 %macro LOAD_SYM 2 973 974 %ifdef PIC 975 call %%geteip 976 add %1, %2 - $ 977 jmp %%end 978 %%geteip: 979 mov %1, [rsp] 980 ret 981 %%end: 982 983 %else 984 lea %1, [%2] 985 %endif 986 987 %endmacro 988 989 ; 990 ; MOVq %1 (xmm), %2 (reg) 991 ; MOVq %1 (reg), %2 (xmm) 992 ; Copies a general-purpose register to an XMM register, and vice versa. 993 ; 994 %macro MOVq 2 995 %if gprsize == 8 996 movq %1, %2 997 %else 998 movd %1, %2 999 %endif 1000 %endmacro 1001 1002 %endif ; CHROMIUM 1003 1004 %endif ; THIRD_PARTY_X86INC_X86INC_ASM_ 1005