;/*****************************************************************************; *; *  XVID MPEG-4 VIDEO CODEC; *  mmx version - inverse discrete cosine transformation ; *; *  Initial version provided by Intel at AppNote AP-922       ; *  Copyright (C) 1999 Intel Corporation,                                     ; * ; *  Modifications ; *  Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru>; *  Copyright (c) 2000 Michel Lespinasse <walken@zoy.org>; *; *  ported to NASM and some minor changes                                      ; *  Copyright (C) 2001 Peter Ross <pross@xvid.org>; *; *  This file is part of XviD, a free MPEG-4 video encoder/decoder; *; *  XviD is free software; you can redistribute it and/or modify it; *  under the terms of the GNU General Public License as published by; *  the Free Software Foundation; either version 2 of the License, or; *  (at your option) any later version.; *; *  This program is distributed in the hope that it will be useful,; *  but WITHOUT ANY WARRANTY; without even the implied warranty of; *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the; *  GNU General Public License for more details.; *; *  You should have received a copy of the GNU General Public License; *  along with this program; if not, write to the Free Software; *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA; *; *  Under section 8 of the GNU General Public License, the copyright; *  holders of XVID explicitly forbid distribution in the following; *  countries:; *; *    - Japan; *    - United States of America; *; *  Linking XviD statically or dynamically with other modules is making a; *  combined work based on XviD.  Thus, the terms and conditions of the; *  GNU General Public License cover the whole combination.; *; *  As a special exception, the copyright holders of XviD give you; *  permission to link XviD with independent modules that communicate with; *  XviD solely through the VFW1.1 and DShow interfaces, regardless of the; *  license terms of these independent modules, and to copy and distribute; *  the resulting combined work under terms of your choice, provided that; *  every copy of the combined work is accompanied by a complete copy of; *  the source code of XviD (the version of XviD used to produce the; *  combined work), being distributed under the terms of the GNU General; *  Public License plus this exception.  An independent module is a module; *  which is not derived from or based on XviD.; *; *  Note that people who make modified versions of XviD are not obligated; *  to grant this special exception for their modified versions; it is; *  their choice whether to do so.  The GNU General Public License gives; *  permission to release a modified version without this exception; this; *  exception also makes it possible to release a modified version which; *  carries forward this exception.; *; * $Id: idct_mmx.asm,v 1.5 2002/11/16 23:51:58 edgomez Exp $; *; *************************************************************************/;=============================================================================;; These examples contain code fragments for first stage iDCT 8x8; (for rows) and first stage DCT 8x8 (for columns);;=============================================================================bits 32%macro cglobal 1 	%ifdef PREFIX		global _%1 		%define %1 _%1	%else		global %1	%endif%endmacro%define BITS_INV_ACC    5                        ; 4 or 5 for IEEE%define SHIFT_INV_ROW   16 - BITS_INV_ACC%define SHIFT_INV_COL   1 + BITS_INV_ACC%define RND_INV_ROW     1024 * (6 - BITS_INV_ACC) ; 1 << (SHIFT_INV_ROW-1)%define RND_INV_COL     16 * (BITS_INV_ACC - 3)   ; 1 << (SHIFT_INV_COL-1)%define RND_INV_CORR    RND_INV_COL - 1          ; correction -1.0 and round%define BITS_FRW_ACC    3                        ; 2 or 3 for accuracy%define SHIFT_FRW_COL   BITS_FRW_ACC%define SHIFT_FRW_ROW   BITS_FRW_ACC + 17%define RND_FRW_ROW     262144 * (BITS_FRW_ACC - 1)       ; 1 << (SHIFT_FRW_ROW-1)%ifdef FORMAT_COFFsection .data data%elsesection .data data align=16%endifalign 16one_corr        dw            1,            1,            1,            1round_inv_row   dd  RND_INV_ROW,  RND_INV_ROWround_inv_col   dw  RND_INV_COL,  RND_INV_COL,  RND_INV_COL, RND_INV_COLround_inv_corr  dw  RND_INV_CORR, RND_INV_CORR, RND_INV_CORR, RND_INV_CORRround_frw_row   dd  RND_FRW_ROW,  RND_FRW_ROW  tg_1_16       dw  13036,  13036,  13036,  13036    ; tg * (2<<16) + 0.5  tg_2_16       dw  27146,  27146,  27146,  27146    ; tg * (2<<16) + 0.5  tg_3_16       dw -21746, -21746, -21746, -21746    ; tg * (2<<16) + 0.5 cos_4_16       dw -19195, -19195, -19195, -19195    ; cos * (2<<16) + 0.5ocos_4_16       dw  23170,  23170,  23170,  23170    ; cos * (2<<15) + 0.5 otg_3_16       dw  21895, 21895, 21895, 21895       ; tg * (2<<16) + 0.5%if SHIFT_INV_ROW == 12   ; assume SHIFT_INV_ROW == 12rounder_0      dd  65536, 65536rounder_4      dd      0,     0rounder_1      dd   7195,  7195rounder_7      dd   1024,  1024rounder_2      dd   4520,  4520rounder_6      dd   1024,  1024rounder_3      dd   2407,  2407rounder_5      dd    240,   240%elif SHIFT_INV_ROW == 11   ; assume SHIFT_INV_ROW == 11rounder_0      dd  65536, 65536rounder_4      dd      0,     0rounder_1      dd   3597,  3597rounder_7      dd    512,   512rounder_2      dd   2260,  2260rounder_6      dd    512,   512rounder_3      dd   1203,  1203rounder_5      dd    120,   120%else%error invalid _SHIFT_INV_ROW_%endif;=============================================================================;; The first stage iDCT 8x8 - inverse DCTs of rows;;-----------------------------------------------------------------------------; The 8-point inverse DCT direct algorithm;-----------------------------------------------------------------------------;; static const short w[32] = {;	FIX(cos_4_16),  FIX(cos_2_16),  FIX(cos_4_16),  FIX(cos_6_16),;	FIX(cos_4_16),  FIX(cos_6_16), -FIX(cos_4_16), -FIX(cos_2_16),;	FIX(cos_4_16), -FIX(cos_6_16), -FIX(cos_4_16),  FIX(cos_2_16),;	FIX(cos_4_16), -FIX(cos_2_16),  FIX(cos_4_16), -FIX(cos_6_16),;	FIX(cos_1_16),  FIX(cos_3_16),  FIX(cos_5_16),  FIX(cos_7_16),;	FIX(cos_3_16), -FIX(cos_7_16), -FIX(cos_1_16), -FIX(cos_5_16),;	FIX(cos_5_16), -FIX(cos_1_16),  FIX(cos_7_16),  FIX(cos_3_16),;	FIX(cos_7_16), -FIX(cos_5_16),  FIX(cos_3_16), -FIX(cos_1_16) };;; #define DCT_8_INV_ROW(x, y); {; 	int a0, a1, a2, a3, b0, b1, b2, b3;;; 	a0 =x[0]*w[0]+x[2]*w[1]+x[4]*w[2]+x[6]*w[3];; 	a1 =x[0]*w[4]+x[2]*w[5]+x[4]*w[6]+x[6]*w[7];; 	a2 = x[0] * w[ 8] + x[2] * w[ 9] + x[4] * w[10] + x[6] * w[11];; 	a3 = x[0] * w[12] + x[2] * w[13] + x[4] * w[14] + x[6] * w[15];; 	b0 = x[1] * w[16] + x[3] * w[17] + x[5] * w[18] + x[7] * w[19];; 	b1 = x[1] * w[20] + x[3] * w[21] + x[5] * w[22] + x[7] * w[23];; 	b2 = x[1] * w[24] + x[3] * w[25] + x[5] * w[26] + x[7] * w[27];; 	b3 = x[1] * w[28] + x[3] * w[29] + x[5] * w[30] + x[7] * w[31];;; 	y[0] = SHIFT_ROUND ( a0 + b0 );; 	y[1] = SHIFT_ROUND ( a1 + b1 );; 	y[2] = SHIFT_ROUND ( a2 + b2 );; 	y[3] = SHIFT_ROUND ( a3 + b3 );; 	y[4] = SHIFT_ROUND ( a3 - b3 );; 	y[5] = SHIFT_ROUND ( a2 - b2 );; 	y[6] = SHIFT_ROUND ( a1 - b1 );; 	y[7] = SHIFT_ROUND ( a0 - b0 );; };;-----------------------------------------------------------------------------;; In this implementation the outputs of the iDCT-1D are multiplied; 	for rows 0,4 - by cos_4_16,; 	for rows 1,7 - by cos_1_16,; 	for rows 2,6 - by cos_2_16,; 	for rows 3,5 - by cos_3_16; and are shifted to the left for better accuracy;; For the constants used,; 	FIX(float_const) = (short) (float_const * (1<<15) + 0.5);;=============================================================================;=============================================================================; MMX code;=============================================================================; Table for rows 0,4 - constants are multiplied by cos_4_16tab_i_04        dw  16384,  16384,  16384, -16384    ; movq-> w06 w04 w02 w00                dw  21407,   8867,   8867, -21407    ; w07 w05 w03 w01                dw  16384, -16384,  16384,  16384    ; w14 w12 w10 w08                dw  -8867,  21407, -21407,  -8867    ; w15 w13 w11 w09                dw  22725,  12873,  19266, -22725    ; w22 w20 w18 w16                dw  19266,   4520,  -4520, -12873    ; w23 w21 w19 w17                dw  12873,   4520,   4520,  19266    ; w30 w28 w26 w24                dw -22725,  19266, -12873, -22725    ; w31 w29 w27 w25; Table for rows 1,7 - constants are multiplied by cos_1_16tab_i_17        dw  22725,  22725,  22725, -22725    ; movq-> w06 w04 w02 w00                dw  29692,  12299,  12299, -29692    ; w07 w05 w03 w01                dw  22725, -22725,  22725,  22725    ; w14 w12 w10 w08                dw -12299,  29692, -29692, -12299    ; w15 w13 w11 w09                dw  31521,  17855,  26722, -31521    ; w22 w20 w18 w16                dw  26722,   6270,  -6270, -17855    ; w23 w21 w19 w17                dw  17855,   6270,   6270,  26722    ; w30 w28 w26 w24                dw -31521,  26722, -17855, -31521    ; w31 w29 w27 w25; Table for rows 2,6 - constants are multiplied by cos_2_16tab_i_26        dw  21407,  21407,  21407, -21407    ; movq-> w06 w04 w02 w00                dw  27969,  11585,  11585, -27969    ; w07 w05 w03 w01                dw  21407, -21407,  21407,  21407    ; w14 w12 w10 w08                dw -11585,  27969, -27969, -11585    ; w15 w13 w11 w09                dw  29692,  16819,  25172, -29692    ; w22 w20 w18 w16                dw  25172,   5906,  -5906, -16819    ; w23 w21 w19 w17                dw  16819,   5906,   5906,  25172    ; w30 w28 w26 w24                dw -29692,  25172, -16819, -29692    ; w31 w29 w27 w25; Table for rows 3,5 - constants are multiplied by cos_3_16tab_i_35        dw  19266,  19266,  19266, -19266    ; movq-> w06 w04 w02 w00                dw  25172,  10426,  10426, -25172    ; w07 w05 w03 w01                dw  19266, -19266,  19266,  19266    ; w14 w12 w10 w08                dw -10426,  25172, -25172, -10426    ; w15 w13 w11 w09                dw  26722,  15137,  22654, -26722    ; w22 w20 w18 w16                dw  22654,   5315,  -5315, -15137    ; w23 w21 w19 w17                dw  15137,   5315,   5315,  22654    ; w30 w28 w26 w24                dw -26722,  22654, -15137, -26722    ; w31 w29 w27 w25;-----------------------------------------------------------------------------;; DCT_8_INV_ROW_1  INP, OUT, TABLE, ROUNDER;%macro  DCT_8_INV_ROW_1         4        movq mm0, [%1]            ; 0     ; x3 x2 x1 x0        movq mm1, [%1+8]          ; 1     ; x7 x6 x5 x4	movq mm2, mm0 			; 2	; x3 x2 x1 x0        movq mm3, [%3]          ; 3     ; w06 w04 w02 w00	punpcklwd mm0, mm1 			; x5 x1 x4 x0	movq mm5, mm0 			; 5	; x5 x1 x4 x0	punpckldq mm0, mm0 			; x4 x0 x4 x0        movq mm4, [%3+8]        ; 4     ; w07 w05 w03 w01	punpckhwd mm2, mm1		; 1	; x7 x3 x6 x2	pmaddwd mm3, mm0 			; x4*w06+x0*w04 x4*w02+x0*w00	movq mm6, mm2 			; 6 	; x7 x3 x6 x2        movq mm1, [%3+32]       ; 1     ; w22 w20 w18 w16	punpckldq mm2, mm2 			; x6 x2 x6 x2	pmaddwd mm4, mm2 			; x6*w07+x2*w05 x6*w03+x2*w01	punpckhdq mm5, mm5 			; x5 x1 x5 x1        pmaddwd mm0, [%3+16]            ; x4*w14+x0*w12 x4*w10+x0*w08	punpckhdq mm6, mm6 			; x7 x3 x7 x3        movq mm7, [%3+40]       ; 7     ; w23 w21 w19 w17	pmaddwd mm1, mm5 			; x5*w22+x1*w20 x5*w18+x1*w16        paddd mm3, [%4]               ; +%4	pmaddwd mm7, mm6 			; x7*w23+x3*w21 x7*w19+x3*w17        pmaddwd mm2, [%3+24]            ; x6*w15+x2*w13 x6*w11+x2*w09	paddd mm3, mm4 			; 4 	; a1=sum(even1) a0=sum(even0)        pmaddwd mm5, [%3+48]            ; x5*w30+x1*w28 x5*w26+x1*w24	movq mm4, mm3 			; 4 	; a1 a0        pmaddwd mm6, [%3+56]            ; x7*w31+x3*w29 x7*w27+x3*w25	paddd mm1, mm7 			; 7 	; b1=sum(odd1) b0=sum(odd0)        paddd mm0, [%4]               ; +%4	psubd mm3, mm1 				; a1-b1 a0-b0	psrad mm3, SHIFT_INV_ROW 		; y6=a1-b1 y7=a0-b0	paddd mm1, mm4 			; 4 	; a1+b1 a0+b0	paddd mm0, mm2 			; 2 	; a3=sum(even3) a2=sum(even2)	psrad mm1, SHIFT_INV_ROW 		; y1=a1+b1 y0=a0+b0	paddd mm5, mm6 			; 6 	; b3=sum(odd3) b2=sum(odd2)	movq mm4, mm0 			; 4 	; a3 a2	paddd mm0, mm5 				; a3+b3 a2+b2	psubd mm4, mm5 			; 5 	; a3-b3 a2-b2	psrad mm0, SHIFT_INV_ROW 		; y3=a3+b3 y2=a2+b2	psrad mm4, SHIFT_INV_ROW 		; y4=a3-b3 y5=a2-b2	packssdw mm1, mm0 		; 0 	; y3 y2 y1 y0	packssdw mm4, mm3 		; 3 	; y6 y7 y4 y5	movq mm7, mm4 			; 7 	; y6 y7 y4 y5	psrld mm4, 16 				; 0 y6 0 y4	pslld mm7, 16 				; y7 0 y5 0        movq [%2], mm1            ; 1     ; save y3 y2 y1 y0                             		por mm7, mm4 			; 4 	; y7 y6 y5 y4        movq [%2+8], mm7          ; 7     ; save y7 y6 y5 y4%endmacro;=============================================================================; code for Pentium III;=============================================================================; %3 for rows 0,4 - constants are multiplied by cos_4_16tab_i_04_sse    dw 16384, 21407, 16384, 8867 ; movq-> w05 w04 w01 w00                dw 16384, 8867, -16384, -21407 ; w07 w06 w03 w02                dw 16384, -8867, 16384, -21407 ; w13 w12 w09 w08                dw -16384, 21407, 16384, -8867 ; w15 w14 w11 w10                dw 22725, 19266, 19266, -4520 ; w21 w20 w17 w16                dw 12873, 4520, -22725, -12873 ; w23 w22 w19 w18                dw 12873, -22725, 4520, -12873 ; w29 w28 w25 w24                dw 4520, 19266, 19266, -22725 ; w31 w30 w27 w26; %3 for rows 1,7 - constants are multiplied by cos_1_16tab_i_17_sse    dw 22725, 29692, 22725, 12299 ; movq-> w05 w04 w01 w00                dw 22725, 12299, -22725, -29692 ; w07 w06 w03 w02                dw 22725, -12299, 22725, -29692 ; w13 w12 w09 w08                dw -22725, 29692, 22725, -12299 ; w15 w14 w11 w10                dw 31521, 26722, 26722, -6270 ; w21 w20 w17 w16                dw 17855, 6270, -31521, -17855 ; w23 w22 w19 w18                dw 17855, -31521, 6270, -17855 ; w29 w28 w25 w24                dw 6270, 26722, 26722, -31521 ; w31 w30 w27 w26; %3 for rows 2,6 - constants are multiplied by cos_2_16tab_i_26_sse    dw 21407, 27969, 21407, 11585 ; movq-> w05 w04 w01 w00                dw 21407, 11585, -21407, -27969 ; w07 w06 w03 w02                dw 21407, -11585, 21407, -27969 ; w13 w12 w09 w08                dw -21407, 27969, 21407, -11585 ; w15 w14 w11 w10                dw 29692, 25172, 25172, -5906 ; w21 w20 w17 w16