1 ;*****************************************************************************
2 ;* dct.asm: h264 encoder library
3 ;*****************************************************************************
4 ;* Copyright (C) 2003 x264 project
5 ;* $Id: dct.asm,v 1.1 2004/06/03 19:27:07 fenrir Exp $
7 ;* Authors: Min Chen <chenm001.163.com> (converted to nasm)
8 ;* Laurent Aimar <fenrir@via.ecp.fr> (initial version)
10 ;* This program is free software; you can redistribute it and/or modify
11 ;* it under the terms of the GNU General Public License as published by
12 ;* the Free Software Foundation; either version 2 of the License, or
13 ;* (at your option) any later version.
15 ;* This program is distributed in the hope that it will be useful,
16 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
17 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 ;* GNU General Public License for more details.
20 ;* You should have received a copy of the GNU General Public License
21 ;* along with this program; if not, write to the Free Software
22 ;* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
23 ;*****************************************************************************
25 ;*****************************************************************************
27 ;* Revision history: *
29 ;* 2004.04.28 portab all 4x4 function to nasm (CM) *
31 ;*****************************************************************************
35 ;=============================================================================
36 ; Macros and other preprocessor constants
37 ;=============================================================================
40 %define GLOBAL wrt rip
58 %macro MMX_LOAD_DIFF_4P 5
66 %macro MMX_SUMSUB_BA 2
72 %macro MMX_SUMSUB_BADC 4
81 %macro MMX_SUMSUB2_AB 3
89 %macro MMX_SUMSUBD2_AB 4
104 %macro SBUTTERFLYdq 3
110 ;-----------------------------------------------------------------------------
111 ; input ABCD output ADTC
112 ;-----------------------------------------------------------------------------
113 %macro MMX_TRANSPOSE 5
114 SBUTTERFLYwd %1, %2, %5
115 SBUTTERFLYwd %3, %4, %2
116 SBUTTERFLYdq %1, %3, %4
117 SBUTTERFLYdq %5, %2, %3
120 %macro MMX_STORE_DIFF_4P 5
133 ;=============================================================================
134 ; Local Data (Read Only)
135 ;=============================================================================
143 ;-----------------------------------------------------------------------------
144 ; Various memory constants (trigonometric values or rounding values)
145 ;-----------------------------------------------------------------------------
148 x264_mmx_1: dw 1, 1, 1, 1
149 x264_mmx_32: dw 32, 32, 32, 32
150 x264_mmx_PPNN: dw 1, 1, -1, -1
151 x264_mmx_PNPN: dw 1, -1, 1, -1
152 x264_mmx_PNNP: dw 1, -1, -1, 1
153 x264_mmx_PPPN: dw 1, 1, 1, -1
154 x264_mmx_PPNP: dw 1, 1, -1, 1
155 x264_mmx_2121: dw 2, 1, 2, 1
156 x264_mmx_p2n2p1p1: dw 2, -2, 1, 1
158 ;=============================================================================
160 ;=============================================================================
164 cglobal x264_dct4x4dc_mmxext
167 ;-----------------------------------------------------------------------------
168 ; void __cdecl dct4x4dc( int16_t d[4][4] )
169 ;-----------------------------------------------------------------------------
170 x264_dct4x4dc_mmxext:
176 MMX_SUMSUB_BADC mm1, mm0, mm3, mm2 ; mm1=s01 mm0=d01 mm3=s23 mm2=d23
177 MMX_SUMSUB_BADC mm3, mm1, mm2, mm0 ; mm3=s01+s23 mm1=s01-s23 mm2=d01+d23 mm0=d01-d23
179 MMX_TRANSPOSE mm3, mm1, mm0, mm2, mm4 ; in: mm3, mm1, mm0, mm2 out: mm3, mm2, mm4, mm0
181 MMX_SUMSUB_BADC mm2, mm3, mm0, mm4 ; mm2=s01 mm3=d01 mm0=s23 mm4=d23
182 MMX_SUMSUB_BADC mm0, mm2, mm4, mm3 ; mm0=s01+s23 mm2=s01-s23 mm4=d01+d23 mm3=d01-d23
184 MMX_TRANSPOSE mm0, mm2, mm3, mm4, mm1 ; in: mm0, mm2, mm3, mm4 out: mm0, mm4, mm1, mm3
186 movq mm6, [x264_mmx_1 GLOBAL]
201 cglobal x264_idct4x4dc_mmxext
204 ;-----------------------------------------------------------------------------
205 ; void __cdecl x264_idct4x4dc_mmxext( int16_t d[4][4] )
206 ;-----------------------------------------------------------------------------
207 x264_idct4x4dc_mmxext:
213 MMX_SUMSUB_BADC mm1, mm0, mm3, mm2 ; mm1=s01 mm0=d01 mm3=s23 mm2=d23
214 MMX_SUMSUB_BADC mm3, mm1, mm2, mm0 ; mm3=s01+s23 mm1=s01-s23 mm2=d01+d23 mm0=d01-d23
216 MMX_TRANSPOSE mm3, mm1, mm0, mm2, mm4 ; in: mm3, mm1, mm0, mm2 out: mm3, mm2, mm4, mm0
218 MMX_SUMSUB_BADC mm2, mm3, mm0, mm4 ; mm2=s01 mm3=d01 mm0=s23 mm4=d23
219 MMX_SUMSUB_BADC mm0, mm2, mm4, mm3 ; mm0=s01+s23 mm2=s01-s23 mm4=d01+d23 mm3=d01-d23
221 MMX_TRANSPOSE mm0, mm2, mm3, mm4, mm1 ; in: mm0, mm2, mm3, mm4 out: mm0, mm4, mm1, mm3
229 cglobal x264_sub4x4_dct_mmxext
232 ;-----------------------------------------------------------------------------
233 ; void __cdecl x264_sub4x4_dct_mmxext( int16_t dct[4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
234 ;-----------------------------------------------------------------------------
235 x264_sub4x4_dct_mmxext:
238 movsxd rbx, edx ; i_pix1
239 ; mov rcx, rcx ; pix2
240 movsxd rdx, r8d ; i_pix2
245 MMX_LOAD_DIFF_4P mm0, mm6, mm7, [rax ], [rcx]
246 MMX_LOAD_DIFF_4P mm1, mm6, mm7, [rax+rbx ], [rcx+rdx]
247 MMX_LOAD_DIFF_4P mm2, mm6, mm7, [rax+rbx*2], [rcx+rdx*2]
250 MMX_LOAD_DIFF_4P mm3, mm6, mm7, [rax+rbx*2], [rcx+rdx*2]
252 MMX_SUMSUB_BADC mm3, mm0, mm2, mm1 ; mm3=s03 mm0=d03 mm2=s12 mm1=d12
254 MMX_SUMSUB_BA mm2, mm3 ; mm2=s03+s12 mm3=s03-s12
255 MMX_SUMSUB2_AB mm0, mm1, mm4 ; mm0=2.d03+d12 mm4=d03-2.d12
257 ; transpose in: mm2, mm0, mm3, mm4, out: mm2, mm4, mm1, mm3
258 MMX_TRANSPOSE mm2, mm0, mm3, mm4, mm1
260 MMX_SUMSUB_BADC mm3, mm2, mm1, mm4 ; mm3=s03 mm2=d03 mm1=s12 mm4=d12
262 MMX_SUMSUB_BA mm1, mm3 ; mm1=s03+s12 mm3=s03-s12
263 MMX_SUMSUB2_AB mm2, mm4, mm0 ; mm2=2.d03+d12 mm0=d03-2.d12
265 ; transpose in: mm1, mm2, mm3, mm0, out: mm1, mm0, mm4, mm3
266 MMX_TRANSPOSE mm1, mm2, mm3, mm0, mm4
268 movq [rdi+ 0], mm1 ; dct
276 cglobal x264_add4x4_idct_mmxext
279 ;-----------------------------------------------------------------------------
280 ; void __cdecl x264_add4x4_idct_mmxext( uint8_t *p_dst, int i_dst, int16_t dct[4][4] )
281 ;-----------------------------------------------------------------------------
282 x264_add4x4_idct_mmxext:
284 movq mm0, [rdx+ 0] ; dct
290 movsxd rcx, esi ; i_dst
293 ; out:mm0, mm1, mm2, mm3
294 MMX_TRANSPOSE mm0, mm4, mm3, mm1, mm2
296 MMX_SUMSUB_BA mm2, mm0 ; mm2=s02 mm0=d02
297 MMX_SUMSUBD2_AB mm1, mm3, mm5, mm4 ; mm1=s13 mm4=d13 ( well 1 + 3>>1 and 1>>1 + 3)
299 MMX_SUMSUB_BADC mm1, mm2, mm4, mm0 ; mm1=s02+s13 mm2=s02-s13 mm4=d02+d13 mm0=d02-d13
301 ; in: mm1, mm4, mm0, mm2 out: mm1, mm2, mm3, mm0
302 MMX_TRANSPOSE mm1, mm4, mm0, mm2, mm3
304 MMX_SUMSUB_BA mm3, mm1 ; mm3=s02 mm1=d02
305 MMX_SUMSUBD2_AB mm2, mm0, mm5, mm4 ; mm2=s13 mm4=d13 ( well 1 + 3>>1 and 1>>1 + 3)
307 MMX_SUMSUB_BADC mm2, mm3, mm4, mm1 ; mm2=s02+s13 mm3=s02-s13 mm4=d02+d13 mm1=d02-d13
310 movq mm6, [x264_mmx_32 GLOBAL]
312 MMX_STORE_DIFF_4P mm2, mm0, mm6, mm7, [rax]
313 MMX_STORE_DIFF_4P mm4, mm0, mm6, mm7, [rax+rcx]
314 MMX_STORE_DIFF_4P mm1, mm0, mm6, mm7, [rax+rcx*2]
315 MMX_STORE_DIFF_4P mm3, mm0, mm6, mm7, [rax+rdx]
321 ; =============================================================================
323 ; =============================================================================
325 ; -----------------------------------------------------------------------------
326 ; input 2x8 unsigned bytes (%5,%6), zero (%7) output: difference (%1,%2)
327 ; -----------------------------------------------------------------------------
328 %macro MMX_LOAD_DIFF_8P 7
341 %macro MMX_LOADSUMSUB 4 ; returns %1=%3+%4, %2=%3-%4
347 %macro MMX_STORE_DIFF_8P 6
358 cglobal x264_pixel_sub_8x8_mmx
359 cglobal x264_xdct8_mmxext
360 cglobal x264_ydct8_mmx
361 cglobal x264_ydct8_sse2
363 cglobal x264_xidct8_mmxext
364 cglobal x264_yidct8_mmx
365 cglobal x264_yidct8_sse2
366 cglobal x264_pixel_add_8x8_mmx
369 ;-----------------------------------------------------------------------------
370 ; void __cdecl x264_pixel_sub_8x8_mmx( int16_t *diff, uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 );
371 ;-----------------------------------------------------------------------------
372 x264_pixel_sub_8x8_mmx:
373 ; mov rdi, rdi ; diff
374 ; mov rsi, rsi ; pix1
375 movsxd rdx, edx ; i_pix1
376 ; mov rcx, rcx ; pix2
377 movsxd r8, r8d ; i_pix2
383 MMX_LOAD_DIFF_8P mm0, mm1, mm2, mm3, [rsi], [rcx], mm7
385 movq [rdi+disp+8], mm1
394 ;-----------------------------------------------------------------------------
395 ; void __cdecl x264_xdct8_mmxext( int16_t dest[8][8] );
396 ;-----------------------------------------------------------------------------
399 movq mm5, [x264_mmx_PPNN GLOBAL]
400 movq mm6, [x264_mmx_PNNP GLOBAL]
401 movq mm4, [x264_mmx_PPPN GLOBAL]
402 movq mm7, [x264_mmx_PPNP GLOBAL]
404 ;-------------------------------------------------------------------------
405 ; horizontal dct ( compute 1 row at a time -> 8 loops )
406 ;-------------------------------------------------------------------------
412 movq mm1, [rdi+disp+8]
414 pshufw mm2, mm1, 00011011b
416 paddw mm0, mm2 ; (low)s07/s16/d25/s34(high)
417 psubw mm1, mm2 ; (low)d07/d16/d25/d34(high)
419 pshufw mm2, mm0, 00011011b ; (low)s34/s25/s16/s07(high)
420 pmullw mm0, mm5 ; (low)s07/s16/-s25/-s34(high)
421 paddw mm0, mm2 ; (low)a0/a1/a3/a2(high)
424 psraw mm1, 1 ; (low)d07/d16/d25/d34(high) (x>>1)
425 pshufw mm2, mm3, 10110001b ; (low)d16/d07/d34/d25(high)
426 paddw mm1, mm3 ; (low)d07/d16/d25/d34(high) (x+(x>>1))
427 pshufw mm3, mm2, 00011011b ; (low)d25/d34/d07/d16(high)
428 pmullw mm2, mm5 ; (low)d16/d07/-d34/-d25(high)
429 pmullw mm1, mm6 ; (low)d07/-d16/-d25/d34(high) (x+(x>>1))
431 paddw mm1, mm3 ; (low)a4/a6/a5/a7(high)
434 pshufw mm2, mm0, 11001001b ; (low)a1/a3/a0/a2(high)
435 pshufw mm0, mm0, 10011100b ; (low)a0/a2/a1/a3(high)
436 pmullw mm2, [x264_mmx_2121 GLOBAL]
437 pmullw mm0, mm5 ; (low)a0/a2/-a1/-a3(high)
438 psraw mm2, 1 ; (low)a1/a3>>1/a0/a2>>1(high)
439 paddw mm0, mm2 ; (low)dst0/dst2/dst4/dst6(high)
441 pshufw mm1, mm1, 00100111b ; (low)a7/a6/a5/a4(high)
442 pshufw mm2, mm1, 00011011b ; (low)a4/a5/a6/a7(high)
443 psraw mm1, 2 ; (low)a7>>2/a6>>2/a5>>2/a4>>2(high)
444 pmullw mm2, mm4 ; (low)a4/a5/a6/-a7(high)
445 pmullw mm1, mm7 ; (low)a7>>2/a6>>2/-a5>>2/a4>>2(high)
446 paddw mm1, mm2 ; (low)dst1/dst3/dst5/dst7(high)
449 punpcklwd mm0, mm1 ; (low)dst0/dst1/dst2/dst3(high)
450 punpckhwd mm2, mm1 ; (low)dst4/dst5/dst6/dst7(high)
453 movq [rdi+disp+8], mm2
461 ;-----------------------------------------------------------------------------
462 ; void __cdecl x264_ydct8_mmx( int16_t dest[8][8] );
463 ;-----------------------------------------------------------------------------
466 ;-------------------------------------------------------------------------
467 ; vertical dct ( compute 4 columns at a time -> 2 loops )
468 ;-------------------------------------------------------------------------
473 MMX_LOADSUMSUB mm2, mm3, [rdi+disp+0*16], [rdi+disp+7*16] ; mm2 = s07, mm3 = d07
474 MMX_LOADSUMSUB mm1, mm5, [rdi+disp+1*16], [rdi+disp+6*16] ; mm1 = s16, mm5 = d16
475 MMX_LOADSUMSUB mm0, mm6, [rdi+disp+2*16], [rdi+disp+5*16] ; mm0 = s25, mm6 = d25
476 MMX_LOADSUMSUB mm4, mm7, [rdi+disp+3*16], [rdi+disp+4*16] ; mm4 = s34, mm7 = d34
478 MMX_SUMSUB_BA mm4, mm2 ; mm4 = a0, mm2 = a2
479 MMX_SUMSUB_BA mm0, mm1 ; mm0 = a1, mm1 = a3
480 MMX_SUMSUB_BA mm0, mm4 ; mm0 = dst0, mm1 = dst4
482 movq [rdi+disp+0*16], mm0
483 movq [rdi+disp+4*16], mm4
487 paddw mm0, mm2 ; a2 + (a3>>1)
489 psubw mm2, mm1 ; (a2>>1) - a3
491 movq [rdi+disp+2*16], mm0
492 movq [rdi+disp+6*16], mm2
496 paddw mm0, mm6 ; d25+(d25>>1)
498 psubw mm1, mm7 ; a5 = d07-d34-(d25+(d25>>1))
503 paddw mm0, mm5 ; d16+(d16>>1)
505 paddw mm2, mm7 ; a6 = d07+d34-(d16+(d16>>1))
510 paddw mm0, mm3 ; d07+(d07>>1)
512 paddw mm0, mm6 ; a4 = d16+d25+(d07+(d07>>1))
516 paddw mm3, mm7 ; d34+(d34>>1)
518 psubw mm3, mm6 ; a7 = d16-d25+(d34+(d34>>1))
522 paddw mm7, mm0 ; a4 + (a7>>2)
526 paddw mm6, mm1 ; a5 + (a6>>2)
530 psubw mm0, mm3 ; (a4>>2) - a7
531 psubw mm2, mm1 ; a6 - (a5>>2)
533 movq [rdi+disp+1*16], mm7
534 movq [rdi+disp+3*16], mm6
535 movq [rdi+disp+5*16], mm2
536 movq [rdi+disp+7*16], mm0
544 ;-----------------------------------------------------------------------------
545 ; void __cdecl x264_xidct8_mmxext( int16_t dest[8][8] );
546 ;-----------------------------------------------------------------------------
549 movq mm4, [x264_mmx_PPNN GLOBAL]
550 movq mm5, [x264_mmx_PNPN GLOBAL]
551 movq mm6, [x264_mmx_PPNP GLOBAL]
552 movq mm7, [x264_mmx_PPPN GLOBAL]
554 ;-------------------------------------------------------------------------
555 ; horizontal idct ( compute 1 row at a time -> 8 loops )
556 ;-------------------------------------------------------------------------
561 pshufw mm0, [rdi+disp], 11011000b ; (low)d0,d2,d1,d3(high)
562 pshufw mm2, [rdi+disp+8], 11011000b ; (low)d4,d6,d5,d7(high)
564 punpcklwd mm0, mm2 ; (low)d0,d4,d2,d6(high)
565 punpckhwd mm1, mm2 ; (low)d1,d5,d3,d7(high)
567 pshufw mm2, mm0, 10110001b ; (low)d4,d0,d6,d2(high)
568 pmullw mm0, [x264_mmx_p2n2p1p1 GLOBAL]
569 ; (low)2*d0,-2*d4,d2,d6(high)
570 pmullw mm2, mm6 ; (low)d4,d0,-d6,d2(high)
571 psraw mm0, 1 ; (low)d0,-d4,d2>>1,d6>>1(high)
572 paddw mm0, mm2 ; (low)e0,e2,e4,e6(high)
574 movq mm3, mm1 ; (low)d1,d5,d3,d7(high)
575 psraw mm1, 1 ; (low)d1>>1,d5>>1,d3>>1,d7>>1(high)
576 pshufw mm2, mm3, 10110001b ; (low)d5,d1,d7,d3(high)
577 paddw mm1, mm3 ; (low)d1+(d1>>1),d5+(d5>>1),d3+(d3>>1),d7+(d7>>1)(high)
578 pshufw mm3, mm2, 00011011b ; (low)d3,d7,d1,d5(high)
579 pmullw mm1, mm4 ; (low)d1+(d1>>1),d5+(d5>>1),-d3-(d3>>1),-d7-(d7>>1)(high)
580 pmullw mm2, mm5 ; (low)d5,-d1,d7,-d3(high)
582 paddw mm1, mm2 ; (low)e7,e5,e3,e1(high)
584 pshufw mm2, mm0, 00011011b ; (low)e6,e4,e2,e0(high)
585 pmullw mm0, mm4 ; (low)e0,e2,-e4,-e6(high)
586 pshufw mm3, mm1, 00011011b ; (low)e1,e3,e5,e7(high)
587 psraw mm1, 2 ; (low)e7>>2,e5>>2,e3>>2,e1>>2(high)
588 pmullw mm3, mm6 ; (low)e1,e3,-e5,e7(high)
589 pmullw mm1, mm7 ; (low)e7>>2,e5>>2,e3>>2,-e1>>2(high)
590 paddw mm0, mm2 ; (low)f0,f2,f4,f6(high)
591 paddw mm1, mm3 ; (low)f1,f3,f5,f7(high)
593 pshufw mm3, mm0, 00011011b ; (low)f6,f4,f2,f0(high)
594 pshufw mm2, mm1, 00011011b ; (low)f7,f5,f3,f1(high)
599 movq [rdi+disp+8], mm3
607 ;-----------------------------------------------------------------------------
608 ; void __cdecl x264_yidct8_mmx( int16_t dest[8][8] );
609 ;-----------------------------------------------------------------------------
612 ;-------------------------------------------------------------------------
613 ; vertical idct ( compute 4 columns at a time -> 2 loops )
614 ;-------------------------------------------------------------------------
619 movq mm1, [rdi+disp+1*16] ; mm1 = d1
620 movq mm3, [rdi+disp+3*16] ; mm3 = d3
621 movq mm5, [rdi+disp+5*16] ; mm5 = d5
622 movq mm7, [rdi+disp+7*16] ; mm7 = d7
629 psubw mm0, mm3 ; mm0 = e1
636 paddw mm2, mm1 ; mm2 = e3
642 psubw mm4, mm1 ; mm4 = e5
648 paddw mm6, mm3 ; mm6 = e7
658 paddw mm1, mm6 ; mm1 = f1
659 paddw mm3, mm2 ; mm3 = f3
660 psubw mm5, mm4 ; mm5 = f5
661 psubw mm7, mm0 ; mm7 = f7
663 movq mm2, [rdi+disp+2*16] ; mm2 = d2
664 movq mm6, [rdi+disp+6*16] ; mm6 = d6
669 psubw mm4, mm0 ; mm4 = a4
670 paddw mm6, mm2 ; mm6 = a6
672 movq mm2, [rdi+disp+0*16] ; mm2 = d0
673 movq mm0, [rdi+disp+4*16] ; mm0 = d4
674 MMX_SUMSUB_BA mm0, mm2 ; mm0 = a0, mm2 = a2
676 MMX_SUMSUB_BA mm6, mm0 ; mm6 = f0, mm0 = f6
677 MMX_SUMSUB_BA mm4, mm2 ; mm4 = f2, mm2 = f4
679 MMX_SUMSUB_BA mm7, mm6 ; mm7 = g0, mm6 = g7
680 MMX_SUMSUB_BA mm5, mm4 ; mm5 = g1, mm4 = g6
681 MMX_SUMSUB_BA mm3, mm2 ; mm3 = g2, mm2 = g5
682 MMX_SUMSUB_BA mm1, mm0 ; mm1 = g3, mm0 = g4
693 movq [rdi+disp+0*16], mm7
694 movq [rdi+disp+1*16], mm5
695 movq [rdi+disp+2*16], mm3
696 movq [rdi+disp+3*16], mm1
697 movq [rdi+disp+4*16], mm0
698 movq [rdi+disp+5*16], mm2
699 movq [rdi+disp+6*16], mm4
700 movq [rdi+disp+7*16], mm6
708 ;-----------------------------------------------------------------------------
709 ; void __cdecl x264_pixel_add_8x8_mmx( unit8_t *dst, int i_dst, int16_t src[8][8] );
710 ;-----------------------------------------------------------------------------
711 x264_pixel_add_8x8_mmx:
713 movsxd rsi, esi ; i_dst
720 MMX_STORE_DIFF_8P mm0, mm1, [rdi], [rdx+disp], [rdx+disp+8], mm7