1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
|
;*****************************************************************************
;* x86inc.asm
;*****************************************************************************
;* Copyright (C) 2005-2008 x264 project
;*
;* Authors: Loren Merritt <lorenm@u.washington.edu>
;* Anton Mitrofanov <BugMaster@narod.ru>
;*
;* Permission to use, copy, modify, and/or distribute this software for any
;* purpose with or without fee is hereby granted, provided that the above
;* copyright notice and this permission notice appear in all copies.
;*
;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
;*****************************************************************************
; This is a header file for the x264ASM assembly language, which uses
; NASM/YASM syntax combined with a large number of macros to provide easy
; abstraction between different calling conventions (x86_32, win64, linux64).
; It also has various other useful features to simplify writing the kind of
; DSP functions that are most often used in x264.
; Unlike the rest of x264, this file is available under an ISC license, as it
; has significant usefulness outside of x264 and we want it to be available
; to the largest audience possible. Of course, if you modify it for your own
; purposes to add a new feature, we strongly encourage contributing a patch
; as this feature might be useful for others as well. Send patches or ideas
; to x264-devel@videolan.org .
%define program_name ff
%ifdef ARCH_X86_64
%ifidn __OUTPUT_FORMAT__,win32
%define WIN64
%else
%define UNIX64
%endif
%endif
%ifdef PREFIX
%define mangle(x) _ %+ x
%else
%define mangle(x) x
%endif
; FIXME: All of the 64bit asm functions that take a stride as an argument
; via register, assume that the high dword of that register is filled with 0.
; This is true in practice (since we never do any 64bit arithmetic on strides,
; and x264's strides are all positive), but is not guaranteed by the ABI.
; Name of the .rodata section.
; Kludge: Something on OS X fails to align .rodata even given an align attribute,
; so use a different read-only section.
%macro SECTION_RODATA 0-1 16
%ifidn __OUTPUT_FORMAT__,macho64
SECTION .text align=%1
%elifidn __OUTPUT_FORMAT__,macho
SECTION .text align=%1
fakegot:
%else
SECTION .rodata align=%1
%endif
%endmacro
%ifdef WIN64
%define PIC
%elifndef ARCH_X86_64
; x86_32 doesn't require PIC.
; Some distros prefer shared objects to be PIC, but nothing breaks if
; the code contains a few textrels, so we'll skip that complexity.
%undef PIC
%endif
%ifdef PIC
default rel
%endif
; Macros to eliminate most code duplication between x86_32 and x86_64:
; Currently this works only for leaf functions which load all their arguments
; into registers at the start, and make no other use of the stack. Luckily that
; covers most of x264's asm.
; PROLOGUE:
; %1 = number of arguments. loads them from stack if needed.
; %2 = number of registers used. pushes callee-saved regs if needed.
; %3 = number of xmm registers used. pushes callee-saved xmm regs if needed.
; %4 = list of names to define to registers
; PROLOGUE can also be invoked by adding the same options to cglobal
; e.g.
; cglobal foo, 2,3,0, dst, src, tmp
; declares a function (foo), taking two args (dst and src) and one local variable (tmp)
; TODO Some functions can use some args directly from the stack. If they're the
; last args then you can just not declare them, but if they're in the middle
; we need more flexible macro.
; RET:
; Pops anything that was pushed by PROLOGUE
; REP_RET:
; Same, but if it doesn't pop anything it becomes a 2-byte ret, for athlons
; which are slow when a normal ret follows a branch.
; registers:
; rN and rNq are the native-size register holding function argument N
; rNd, rNw, rNb are dword, word, and byte size
; rNm is the original location of arg N (a register or on the stack), dword
; rNmp is native size
%macro DECLARE_REG 6
%define r%1q %2
%define r%1d %3
%define r%1w %4
%define r%1b %5
%define r%1m %6
%ifid %6 ; i.e. it's a register
%define r%1mp %2
%elifdef ARCH_X86_64 ; memory
%define r%1mp qword %6
%else
%define r%1mp dword %6
%endif
%define r%1 %2
%endmacro
%macro DECLARE_REG_SIZE 2
%define r%1q r%1
%define e%1q r%1
%define r%1d e%1
%define e%1d e%1
%define r%1w %1
%define e%1w %1
%define r%1b %2
%define e%1b %2
%ifndef ARCH_X86_64
%define r%1 e%1
%endif
%endmacro
DECLARE_REG_SIZE ax, al
DECLARE_REG_SIZE bx, bl
DECLARE_REG_SIZE cx, cl
DECLARE_REG_SIZE dx, dl
DECLARE_REG_SIZE si, sil
DECLARE_REG_SIZE di, dil
DECLARE_REG_SIZE bp, bpl
; t# defines for when per-arch register allocation is more complex than just function arguments
%macro DECLARE_REG_TMP 1-*
%assign %%i 0
%rep %0
CAT_XDEFINE t, %%i, r%1
%assign %%i %%i+1
%rotate 1
%endrep
%endmacro
%macro DECLARE_REG_TMP_SIZE 0-*
%rep %0
%define t%1q t%1 %+ q
%define t%1d t%1 %+ d
%define t%1w t%1 %+ w
%define t%1b t%1 %+ b
%rotate 1
%endrep
%endmacro
DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9
%ifdef ARCH_X86_64
%define gprsize 8
%else
%define gprsize 4
%endif
%macro PUSH 1
push %1
%assign stack_offset stack_offset+gprsize
%endmacro
%macro POP 1
pop %1
%assign stack_offset stack_offset-gprsize
%endmacro
%macro SUB 2
sub %1, %2
%ifidn %1, rsp
%assign stack_offset stack_offset+(%2)
%endif
%endmacro
%macro ADD 2
add %1, %2
%ifidn %1, rsp
%assign stack_offset stack_offset-(%2)
%endif
%endmacro
%macro movifnidn 2
%ifnidn %1, %2
mov %1, %2
%endif
%endmacro
%macro movsxdifnidn 2
%ifnidn %1, %2
movsxd %1, %2
%endif
%endmacro
%macro ASSERT 1
%if (%1) == 0
%error assert failed
%endif
%endmacro
%macro DEFINE_ARGS 0-*
%ifdef n_arg_names
%assign %%i 0
%rep n_arg_names
CAT_UNDEF arg_name %+ %%i, q
CAT_UNDEF arg_name %+ %%i, d
CAT_UNDEF arg_name %+ %%i, w
CAT_UNDEF arg_name %+ %%i, b
CAT_UNDEF arg_name %+ %%i, m
CAT_UNDEF arg_name, %%i
%assign %%i %%i+1
%endrep
%endif
%assign %%i 0
%rep %0
%xdefine %1q r %+ %%i %+ q
%xdefine %1d r %+ %%i %+ d
%xdefine %1w r %+ %%i %+ w
%xdefine %1b r %+ %%i %+ b
%xdefine %1m r %+ %%i %+ m
CAT_XDEFINE arg_name, %%i, %1
%assign %%i %%i+1
%rotate 1
%endrep
%assign n_arg_names %%i
%endmacro
%ifdef WIN64 ; Windows x64 ;=================================================
DECLARE_REG 0, rcx, ecx, cx, cl, ecx
DECLARE_REG 1, rdx, edx, dx, dl, edx
DECLARE_REG 2, r8, r8d, r8w, r8b, r8d
DECLARE_REG 3, r9, r9d, r9w, r9b, r9d
DECLARE_REG 4, rdi, edi, di, dil, [rsp + stack_offset + 40]
DECLARE_REG 5, rsi, esi, si, sil, [rsp + stack_offset + 48]
DECLARE_REG 6, rax, eax, ax, al, [rsp + stack_offset + 56]
%define r7m [rsp + stack_offset + 64]
%define r8m [rsp + stack_offset + 72]
%macro LOAD_IF_USED 2 ; reg_id, number_of_args
%if %1 < %2
mov r%1, [rsp + stack_offset + 8 + %1*8]
%endif
%endmacro
%macro PROLOGUE 2-4+ 0 ; #args, #regs, #xmm_regs, arg_names...
ASSERT %2 >= %1
%assign regs_used %2
ASSERT regs_used <= 7
%if regs_used > 4
push r4
push r5
%assign stack_offset stack_offset+16
%endif
WIN64_SPILL_XMM %3
LOAD_IF_USED 4, %1
LOAD_IF_USED 5, %1
LOAD_IF_USED 6, %1
DEFINE_ARGS %4
%endmacro
%macro WIN64_SPILL_XMM 1
%assign xmm_regs_used %1
ASSERT xmm_regs_used <= 16
%if xmm_regs_used > 6
sub rsp, (xmm_regs_used-6)*16+16
%assign stack_offset stack_offset+(xmm_regs_used-6)*16+16
%assign %%i xmm_regs_used
%rep (xmm_regs_used-6)
%assign %%i %%i-1
movdqa [rsp + (%%i-6)*16+8], xmm %+ %%i
%endrep
%endif
%endmacro
%macro WIN64_RESTORE_XMM_INTERNAL 1
%if xmm_regs_used > 6
%assign %%i xmm_regs_used
%rep (xmm_regs_used-6)
%assign %%i %%i-1
movdqa xmm %+ %%i, [%1 + (%%i-6)*16+8]
%endrep
add %1, (xmm_regs_used-6)*16+16
%endif
%endmacro
%macro WIN64_RESTORE_XMM 1
WIN64_RESTORE_XMM_INTERNAL %1
%assign stack_offset stack_offset-(xmm_regs_used-6)*16+16
%assign xmm_regs_used 0
%endmacro
%macro RET 0
WIN64_RESTORE_XMM_INTERNAL rsp
%if regs_used > 4
pop r5
pop r4
%endif
ret
%endmacro
%macro REP_RET 0
%if regs_used > 4 || xmm_regs_used > 6
RET
%else
rep ret
%endif
%endmacro
%elifdef ARCH_X86_64 ; *nix x64 ;=============================================
DECLARE_REG 0, rdi, edi, di, dil, edi
DECLARE_REG 1, rsi, esi, si, sil, esi
DECLARE_REG 2, rdx, edx, dx, dl, edx
DECLARE_REG 3, rcx, ecx, cx, cl, ecx
DECLARE_REG 4, r8, r8d, r8w, r8b, r8d
DECLARE_REG 5, r9, r9d, r9w, r9b, r9d
DECLARE_REG 6, rax, eax, ax, al, [rsp + stack_offset + 8]
%define r7m [rsp + stack_offset + 16]
%define r8m [rsp + stack_offset + 24]
%macro LOAD_IF_USED 2 ; reg_id, number_of_args
%if %1 < %2
mov r%1, [rsp - 40 + %1*8]
%endif
%endmacro
%macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names...
ASSERT %2 >= %1
ASSERT %2 <= 7
LOAD_IF_USED 6, %1
DEFINE_ARGS %4
%endmacro
%macro RET 0
ret
%endmacro
%macro REP_RET 0
rep ret
%endmacro
%else ; X86_32 ;==============================================================
DECLARE_REG 0, eax, eax, ax, al, [esp + stack_offset + 4]
DECLARE_REG 1, ecx, ecx, cx, cl, [esp + stack_offset + 8]
DECLARE_REG 2, edx, edx, dx, dl, [esp + stack_offset + 12]
DECLARE_REG 3, ebx, ebx, bx, bl, [esp + stack_offset + 16]
DECLARE_REG 4, esi, esi, si, null, [esp + stack_offset + 20]
DECLARE_REG 5, edi, edi, di, null, [esp + stack_offset + 24]
DECLARE_REG 6, ebp, ebp, bp, null, [esp + stack_offset + 28]
%define r7m [esp + stack_offset + 32]
%define r8m [esp + stack_offset + 36]
%define rsp esp
%macro PUSH_IF_USED 1 ; reg_id
%if %1 < regs_used
push r%1
%assign stack_offset stack_offset+4
%endif
%endmacro
%macro POP_IF_USED 1 ; reg_id
%if %1 < regs_used
pop r%1
%endif
%endmacro
%macro LOAD_IF_USED 2 ; reg_id, number_of_args
%if %1 < %2
mov r%1, [esp + stack_offset + 4 + %1*4]
%endif
%endmacro
%macro PROLOGUE 2-4+ ; #args, #regs, #xmm_regs, arg_names...
ASSERT %2 >= %1
%assign regs_used %2
ASSERT regs_used <= 7
PUSH_IF_USED 3
PUSH_IF_USED 4
PUSH_IF_USED 5
PUSH_IF_USED 6
LOAD_IF_USED 0, %1
LOAD_IF_USED 1, %1
LOAD_IF_USED 2, %1
LOAD_IF_USED 3, %1
LOAD_IF_USED 4, %1
LOAD_IF_USED 5, %1
LOAD_IF_USED 6, %1
DEFINE_ARGS %4
%endmacro
%macro RET 0
POP_IF_USED 6
POP_IF_USED 5
POP_IF_USED 4
POP_IF_USED 3
ret
%endmacro
%macro REP_RET 0
%if regs_used > 3
RET
%else
rep ret
%endif
%endmacro
%endif ;======================================================================
%ifndef WIN64
%macro WIN64_SPILL_XMM 1
%endmacro
%macro WIN64_RESTORE_XMM 1
%endmacro
%endif
;=============================================================================
; arch-independent part
;=============================================================================
%assign function_align 16
; Symbol prefix for C linkage
%macro cglobal 1-2+
%xdefine %1 mangle(program_name %+ _ %+ %1)
%xdefine %1.skip_prologue %1 %+ .skip_prologue
%ifidn __OUTPUT_FORMAT__,elf
global %1:function hidden
%else
global %1
%endif
align function_align
%1:
RESET_MM_PERMUTATION ; not really needed, but makes disassembly somewhat nicer
%assign stack_offset 0
%if %0 > 1
PROLOGUE %2
%endif
%endmacro
%macro cextern 1
%xdefine %1 mangle(program_name %+ _ %+ %1)
extern %1
%endmacro
;like cextern, but without the prefix
%macro cextern_naked 1
%xdefine %1 mangle(%1)
extern %1
%endmacro
%macro const 2+
%xdefine %1 mangle(program_name %+ _ %+ %1)
global %1
%1: %2
%endmacro
; This is needed for ELF, otherwise the GNU linker assumes the stack is
; executable by default.
%ifidn __OUTPUT_FORMAT__,elf
SECTION .note.GNU-stack noalloc noexec nowrite progbits
%endif
; merge mmx and sse*
%macro CAT_XDEFINE 3
%xdefine %1%2 %3
%endmacro
%macro CAT_UNDEF 2
%undef %1%2
%endmacro
%macro INIT_MMX 0
%define RESET_MM_PERMUTATION INIT_MMX
%define mmsize 8
%define num_mmregs 8
%define mova movq
%define movu movq
%define movh movd
%define movnta movntq
%assign %%i 0
%rep 8
CAT_XDEFINE m, %%i, mm %+ %%i
CAT_XDEFINE nmm, %%i, %%i
%assign %%i %%i+1
%endrep
%rep 8
CAT_UNDEF m, %%i
CAT_UNDEF nmm, %%i
%assign %%i %%i+1
%endrep
%endmacro
%macro INIT_XMM 0
%define RESET_MM_PERMUTATION INIT_XMM
%define mmsize 16
%define num_mmregs 8
%ifdef ARCH_X86_64
%define num_mmregs 16
%endif
%define mova movdqa
%define movu movdqu
%define movh movq
%define movnta movntdq
%assign %%i 0
%rep num_mmregs
CAT_XDEFINE m, %%i, xmm %+ %%i
CAT_XDEFINE nxmm, %%i, %%i
%assign %%i %%i+1
%endrep
%endmacro
INIT_MMX
; I often want to use macros that permute their arguments. e.g. there's no
; efficient way to implement butterfly or transpose or dct without swapping some
; arguments.
;
; I would like to not have to manually keep track of the permutations:
; If I insert a permutation in the middle of a function, it should automatically
; change everything that follows. For more complex macros I may also have multiple
; implementations, e.g. the SSE2 and SSSE3 versions may have different permutations.
;
; Hence these macros. Insert a PERMUTE or some SWAPs at the end of a macro that
; permutes its arguments. It's equivalent to exchanging the contents of the
; registers, except that this way you exchange the register names instead, so it
; doesn't cost any cycles.
%macro PERMUTE 2-* ; takes a list of pairs to swap
%rep %0/2
%xdefine tmp%2 m%2
%xdefine ntmp%2 nm%2
%rotate 2
%endrep
%rep %0/2
%xdefine m%1 tmp%2
%xdefine nm%1 ntmp%2
%undef tmp%2
%undef ntmp%2
%rotate 2
%endrep
%endmacro
%macro SWAP 2-* ; swaps a single chain (sometimes more concise than pairs)
%rep %0-1
%ifdef m%1
%xdefine tmp m%1
%xdefine m%1 m%2
%xdefine m%2 tmp
CAT_XDEFINE n, m%1, %1
CAT_XDEFINE n, m%2, %2
%else
; If we were called as "SWAP m0,m1" rather than "SWAP 0,1" infer the original numbers here.
; Be careful using this mode in nested macros though, as in some cases there may be
; other copies of m# that have already been dereferenced and don't get updated correctly.
%xdefine %%n1 n %+ %1
%xdefine %%n2 n %+ %2
%xdefine tmp m %+ %%n1
CAT_XDEFINE m, %%n1, m %+ %%n2
CAT_XDEFINE m, %%n2, tmp
CAT_XDEFINE n, m %+ %%n1, %%n1
CAT_XDEFINE n, m %+ %%n2, %%n2
%endif
%undef tmp
%rotate 1
%endrep
%endmacro
; If SAVE_MM_PERMUTATION is placed at the end of a function and given the
; function name, then any later calls to that function will automatically
; load the permutation, so values can be returned in mmregs.
%macro SAVE_MM_PERMUTATION 1 ; name to save as
%assign %%i 0
%rep num_mmregs
CAT_XDEFINE %1_m, %%i, m %+ %%i
%assign %%i %%i+1
%endrep
%endmacro
%macro LOAD_MM_PERMUTATION 1 ; name to load from
%assign %%i 0
%rep num_mmregs
CAT_XDEFINE m, %%i, %1_m %+ %%i
CAT_XDEFINE n, m %+ %%i, %%i
%assign %%i %%i+1
%endrep
%endmacro
%macro call 1
call %1
%ifdef %1_m0
LOAD_MM_PERMUTATION %1
%endif
%endmacro
; Substitutions that reduce instruction size but are functionally equivalent
%macro add 2
%ifnum %2
%if %2==128
sub %1, -128
%else
add %1, %2
%endif
%else
add %1, %2
%endif
%endmacro
%macro sub 2
%ifnum %2
%if %2==128
add %1, -128
%else
sub %1, %2
%endif
%else
sub %1, %2
%endif
%endmacro
|