text_fdct_mmx.c

MPEG编解码的源码(包含视频捕捉、音频处理)

#define INP eax		
#define OUT ecx		
#define TABLE ebx	
#define TABLEF ebx  
#define round_frw_row edx
#define x0 INP + 0*16
#define x1 INP + 1*16
#define x2 INP + 2*16
#define x3 INP + 3*16
#define x4 INP + 4*16
#define x5 INP + 5*16
#define x6 INP + 6*16
#define x7 INP + 7*16
#define y0 OUT + 0*16
#define y1 OUT + 1*16
#define y2 OUT + 2*16
#define y3 OUT + 3*16
#define y4 OUT + 4*16
#define y5 OUT + 5*16
#define y6 OUT + 6*16
#define y7 OUT + 7*16
#define BITS_FRW_ACC	3 
#define SHIFT_FRW_COL	BITS_FRW_ACC
#define SHIFT_FRW_ROW	(BITS_FRW_ACC + 17)
#define SHIFT_FRW_ROW_CLIP2	(4)  
#define SHIFT_FRW_ROW_CLIP1	( SHIFT_FRW_ROW - SHIFT_FRW_ROW_CLIP2 )
#define RND_FRW_ROW		(1 << (SHIFT_FRW_ROW-1))
#define RND_FRW_COL		(1 << (SHIFT_FRW_COL-1))
const static __int64 one_corr = 0x0001000100010001;
const static long r_frw_row[2] = {RND_FRW_ROW, RND_FRW_ROW };
const static short tg_all_16[] = {
	13036, 13036, 13036, 13036,		
	27146, 27146, 27146, 27146,		
	-21746, -21746, -21746, -21746,	
	-19195, -19195, -19195, -19195,	
	23170, 23170, 23170, 23170 };	
#define tg_1_16 (TABLEF + 0)
#define tg_2_16 (TABLEF + 8)
#define tg_3_16 (TABLEF + 16)
#define cos_4_16 (TABLEF + 24)
#define ocos_4_16 (TABLEF + 32)
    
static const short tab_frw_01234567[] = {  
    
    16384, 16384, 21407, -8867,     
    16384, 16384, 8867, -21407,     
    16384, -16384, 8867, 21407,     
    -16384, 16384, -21407, -8867,   
    22725, 12873, 19266, -22725,    
    19266, 4520, -4520, -12873,     
    12873, 4520, 4520, 19266,       
    -22725, 19266, -12873, -22725,  
    
    22725, 22725, 29692, -12299,    
    22725, 22725, 12299, -29692,    
    22725, -22725, 12299, 29692,    
    -22725, 22725, -29692, -12299,  
    31521, 17855, 26722, -31521,    
    26722, 6270, -6270, -17855,     
    17855, 6270, 6270, 26722,       
    -31521, 26722, -17855, -31521,  
    
    21407, 21407, 27969, -11585,    
    21407, 21407, 11585, -27969,    
    21407, -21407, 11585, 27969,    
    -21407, 21407, -27969, -11585,  
    29692, 16819, 25172, -29692,    
    25172, 5906, -5906, -16819,     
    16819, 5906, 5906, 25172,       
    -29692, 25172, -16819, -29692,  
    
    19266, 19266, 25172, -10426,    
    19266, 19266, 10426, -25172,    
    19266, -19266, 10426, 25172,    
    -19266, 19266, -25172, -10426,  
    26722, 15137, 22654, -26722,    
    22654, 5315, -5315, -15137,     
    15137, 5315, 5315, 22654,       
    -26722, 22654, -15137, -26722,  
    
    16384, 16384, 21407, -8867,     
    16384, 16384, 8867, -21407,     
    16384, -16384, 8867, 21407,     
    -16384, 16384, -21407, -8867,   
    22725, 12873, 19266, -22725,    
    19266, 4520, -4520, -12873,     
    12873, 4520, 4520, 19266,       
    -22725, 19266, -12873, -22725,  
    
    19266, 19266, 25172, -10426,    
    19266, 19266, 10426, -25172,    
    19266, -19266, 10426, 25172,    
    -19266, 19266, -25172, -10426,  
    26722, 15137, 22654, -26722,    
    22654, 5315, -5315, -15137,     
    15137, 5315, 5315, 22654,       
    -26722, 22654, -15137, -26722,  
    
    21407, 21407, 27969, -11585,    
    21407, 21407, 11585, -27969,    
    21407, -21407, 11585, 27969,    
    -21407, 21407, -27969, -11585,  
    29692, 16819, 25172, -29692,    
    25172, 5906, -5906, -16819,     
    16819, 5906, 5906, 25172,       
    -29692, 25172, -16819, -29692,  
    
    22725, 22725, 29692, -12299,    
    22725, 22725, 12299, -29692,    
    22725, -22725, 12299, 29692,    
    -22725, 22725, -29692, -12299,  
    31521, 17855, 26722, -31521,    
    26722, 6270, -6270, -17855,     
    17855, 6270, 6270, 26722,       
    -31521, 26722, -17855, -31521   
};
void
fdct_mm32( short *blk )
{
    static __int64 xt70[2]; 
    static int a0, a1, a2, a3, b0, b1, b2, b3;
    static short *sptr, *optr, *tf; 
    static short *xt = (short *) &xt70[0];
    static int j;
    
    const static short _tg_1_16   = 13036;  
    const static short _tg_2_16   = 27146;  
    const static short _tg_3_16   =-21746;  
    const static short _cos_4_16  =-19195;  
    const static short _ocos_4_16 = 23170;  
    const static short _one_corr  =     1;  
    static short t0, t1, t2, t3, t4, t5, t6, t7;
    static short tp03, tm03, tp12, tm12, tp65, tm65;
    static short tp465, tm465, tp765, tm765;
    __asm {
    
    
    
    
    
    
    
    
    
    
    
	mov INP, dword ptr [blk];		;
    ;
    lea TABLEF, dword ptr [tg_all_16];
    mov OUT, INP;
    
    
    
	mmx32_fdct_col03: 
    movq mm0, [x1] ; 0 ; x1
     ;
    movq mm1, [x6] ; 1 ; x6
     movq mm2, mm0 ; 2 ; x1
    movq mm3, [x2] ; 3 ; x2
     paddsw mm0, mm1 ; t1 = x[1] + x[6]
    movq mm4, [x5] ; 4 ; x5
     psllw mm0, SHIFT_FRW_COL ; t1
    movq mm5, [x0] ; 5 ; x0
     paddsw mm4, mm3 ; t2 = x[2] + x[5]
    paddsw mm5, [x7] ; t0 = x[0] + x[7]
     psllw mm4, SHIFT_FRW_COL ; t2
    movq mm6, mm0 ; 6 ; t1
     psubsw mm2, mm1 ; 1 ; t6 = x[1] - x[6]
    movq mm1, qword ptr [tg_2_16] ; 1 ; tg_2_16
     psubsw mm0, mm4 ; tm12 = t1 - t2
    movq mm7, [x3] ; 7 ; x3
     pmulhw mm1, mm0 ; tm12*tg_2_16
    paddsw mm7, [x4] ; t3 = x[3] + x[4]
     psllw mm5, SHIFT_FRW_COL ; t0
    paddsw mm6, mm4 ; 4 ; tp12 = t1 + t2
     psllw mm7, SHIFT_FRW_COL ; t3
    movq mm4, mm5 ; 4 ; t0
     psubsw mm5, mm7 ; tm03 = t0 - t3
    paddsw mm1, mm5 ; y2 = tm03 + tm12*tg_2_16
     paddsw mm4, mm7 ; 7 ; tp03 = t0 + t3
    por mm1, qword ptr one_corr ; correction y2 +0.5
     psllw mm2, SHIFT_FRW_COL+1 ; t6
    pmulhw mm5, qword ptr [tg_2_16] ; tm03*tg_2_16
     movq mm7, mm4 ; 7 ; tp03
    psubsw mm3, [x5] ; t5 = x[2] - x[5]
     psubsw mm4, mm6 ; y4 = tp03 - tp12
    movq [y2], mm1 ; 1 ; save y2
     paddsw mm7, mm6 ; 6 ; y0 = tp03 + tp12
     
    movq mm1, [x3] ; 1 ; x3
     psllw mm3, SHIFT_FRW_COL+1 ; t5
    psubsw mm1, [x4] ; t4 = x[3] - x[4]
     movq mm6, mm2 ; 6 ; t6
    
    movq [y4], mm4 ; 4 ; save y4
     paddsw mm2, mm3 ; t6 + t5
    pmulhw mm2, qword ptr [ocos_4_16] ; tp65 = (t6 + t5)*cos_4_16
     psubsw mm6, mm3 ; 3 ; t6 - t5
    pmulhw mm6, qword ptr [ocos_4_16] ; tm65 = (t6 - t5)*cos_4_16
     psubsw mm5, mm0 ; 0 ; y6 = tm03*tg_2_16 - tm12
    por mm5, qword ptr one_corr ; correction y6 +0.5
     psllw mm1, SHIFT_FRW_COL ; t4
    por mm2, qword ptr one_corr ; correction tp65 +0.5
     movq mm4, mm1 ; 4 ; t4
    movq mm3, [x0] ; 3 ; x0
     paddsw mm1, mm6 ; tp465 = t4 + tm65
    psubsw mm3, [x7] ; t7 = x[0] - x[7]
     psubsw mm4, mm6 ; 6 ; tm465 = t4 - tm65
    movq mm0, qword ptr [tg_1_16] ; 0 ; tg_1_16
     psllw mm3, SHIFT_FRW_COL ; t7
    movq mm6, qword ptr [tg_3_16] ; 6 ; tg_3_16
     pmulhw mm0, mm1 ; tp465*tg_1_16
    movq [y0], mm7 ; 7 ; save y0
     pmulhw mm6, mm4 ; tm465*tg_3_16
    movq [y6], mm5 ; 5 ; save y6
     movq mm7, mm3 ; 7 ; t7
    movq mm5, qword ptr [tg_3_16] ; 5 ; tg_3_16
     psubsw mm7, mm2 ; tm765 = t7 - tp65
    paddsw mm3, mm2 ; 2 ; tp765 = t7 + tp65
     pmulhw mm5, mm7 ; tm765*tg_3_16
    paddsw mm0, mm3 ; y1 = tp765 + tp465*tg_1_16
     paddsw mm6, mm4 ; tm465*tg_3_16
    pmulhw mm3, qword ptr [tg_1_16] ; tp765*tg_1_16
     ;
    por mm0, qword ptr one_corr ; correction y1 +0.5
     paddsw mm5, mm7 ; tm765*tg_3_16
    psubsw mm7, mm6 ; 6 ; y3 = tm765 - tm465*tg_3_16
     add INP, 0x08   ; 
    movq [y1], mm0 ; 0 ; save y1
     paddsw mm5, mm4 ; 4 ; y5 = tm765*tg_3_16 + tm465
    movq [y3], mm7 ; 7 ; save y3
     psubsw mm3, mm1 ; 1 ; y7 = tp765*tg_1_16 - tp465
    movq [y5], mm5 ; 5 ; save y5
  mmx32_fdct_col47: 
    movq mm0, [x1] ; 0 ; x1
     ;
    movq [y7], mm3 ; 3 ; save y7 (columns 0-4)
     ;
    movq mm1, [x6] ; 1 ; x6
     movq mm2, mm0 ; 2 ; x1
    movq mm3, [x2] ; 3 ; x2
     paddsw mm0, mm1 ; t1 = x[1] + x[6]
    movq mm4, [x5] ; 4 ; x5
     psllw mm0, SHIFT_FRW_COL ; t1
    movq mm5, [x0] ; 5 ; x0
     paddsw mm4, mm3 ; t2 = x[2] + x[5]
    paddsw mm5, [x7] ; t0 = x[0] + x[7]
     psllw mm4, SHIFT_FRW_COL ; t2
    movq mm6, mm0 ; 6 ; t1
     psubsw mm2, mm1 ; 1 ; t6 = x[1] - x[6]
    movq mm1, qword ptr [tg_2_16] ; 1 ; tg_2_16
     psubsw mm0, mm4 ; tm12 = t1 - t2
    movq mm7, [x3] ; 7 ; x3
     pmulhw mm1, mm0 ; tm12*tg_2_16
    paddsw mm7, [x4] ; t3 = x[3] + x[4]
     psllw mm5, SHIFT_FRW_COL ; t0
    paddsw mm6, mm4 ; 4 ; tp12 = t1 + t2
     psllw mm7, SHIFT_FRW_COL ; t3
    movq mm4, mm5 ; 4 ; t0
     psubsw mm5, mm7 ; tm03 = t0 - t3
    paddsw mm1, mm5 ; y2 = tm03 + tm12*tg_2_16
     paddsw mm4, mm7 ; 7 ; tp03 = t0 + t3
    por mm1, qword ptr one_corr ; correction y2 +0.5
     psllw mm2, SHIFT_FRW_COL+1 ; t6
    pmulhw mm5, qword ptr [tg_2_16] ; tm03*tg_2_16
     movq mm7, mm4 ; 7 ; tp03
    psubsw mm3, [x5] ; t5 = x[2] - x[5]
     psubsw mm4, mm6 ; y4 = tp03 - tp12
    movq [y2+8], mm1 ; 1 ; save y2
     paddsw mm7, mm6 ; 6 ; y0 = tp03 + tp12
     
    movq mm1, [x3] ; 1 ; x3
     psllw mm3, SHIFT_FRW_COL+1 ; t5
    psubsw mm1, [x4] ; t4 = x[3] - x[4]
     movq mm6, mm2 ; 6 ; t6
    
    movq [y4+8], mm4 ; 4 ; save y4
     paddsw mm2, mm3 ; t6 + t5
    pmulhw mm2, qword ptr [ocos_4_16] ; tp65 = (t6 + t5)*cos_4_16
     psubsw mm6, mm3 ; 3 ; t6 - t5
    pmulhw mm6, qword ptr [ocos_4_16] ; tm65 = (t6 - t5)*cos_4_16
     psubsw mm5, mm0 ; 0 ; y6 = tm03*tg_2_16 - tm12
    por mm5, qword ptr one_corr ; correction y6 +0.5
     psllw mm1, SHIFT_FRW_COL ; t4
    por mm2, qword ptr one_corr ; correction tp65 +0.5
     movq mm4, mm1 ; 4 ; t4
    movq mm3, [x0] ; 3 ; x0
     paddsw mm1, mm6 ; tp465 = t4 + tm65
    psubsw mm3, [x7] ; t7 = x[0] - x[7]
     psubsw mm4, mm6 ; 6 ; tm465 = t4 - tm65
    movq mm0, qword ptr [tg_1_16] ; 0 ; tg_1_16
     psllw mm3, SHIFT_FRW_COL ; t7
    movq mm6, qword ptr [tg_3_16] ; 6 ; tg_3_16
     pmulhw mm0, mm1 ; tp465*tg_1_16
    movq [y0+8], mm7 ; 7 ; save y0
     pmulhw mm6, mm4 ; tm465*tg_3_16
    movq [y6+8], mm5 ; 5 ; save y6
     movq mm7, mm3 ; 7 ; t7
    movq mm5, qword ptr [tg_3_16] ; 5 ; tg_3_16
     psubsw mm7, mm2 ; tm765 = t7 - tp65
    paddsw mm3, mm2 ; 2 ; tp765 = t7 + tp65
     pmulhw mm5, mm7 ; tm765*tg_3_16
    paddsw mm0, mm3 ; y1 = tp765 + tp465*tg_1_16
     paddsw mm6, mm4 ; tm465*tg_3_16
    pmulhw mm3, qword ptr [tg_1_16] ; tp765*tg_1_16
     ;
    por mm0, qword ptr one_corr ; correction y1 +0.5
     paddsw mm5, mm7 ; tm765*tg_3_16
    psubsw mm7, mm6 ; 6 ; y3 = tm765 - tm465*tg_3_16
     ;
    movq [y1+8], mm0 ; 0 ; save y1
     paddsw mm5, mm4 ; 4 ; y5 = tm765*tg_3_16 + tm465
    movq [y3+8], mm7 ; 7 ; save y3
     psubsw mm3, mm1 ; 1 ; y7 = tp765*tg_1_16 - tp465
    movq [y5+8], mm5 ; 5 ; save y5
    movq [y7+8], mm3 ; 3 ; save y7
  
  
    
  
  
  
  
  
  
  
  
	mov INP, dword ptr [blk];		;
	 mov edi, 0x08;	
	lea TABLE, dword ptr [tab_frw_01234567]; 
	 mov OUT, INP;
	lea round_frw_row, dword ptr [r_frw_row];
	
  lp_mmx_fdct_row1:
    movd mm5, dword ptr [INP+12]; 
    punpcklwd mm5, dword ptr [INP+8] 
    movq mm2, mm5;     
    psrlq mm5, 32;     
    movq mm0, qword ptr [INP]; 
    punpcklwd mm5, mm2;
    movq mm1, mm0;     
    paddsw mm0, mm5;   
    psubsw mm1, mm5;   
    movq mm2, mm0;     
    
    
    punpcklwd mm0, mm1;
    punpckhwd mm2, mm1;
    movq mm1, mm2;     
    ;
    movq mm2, mm0 ; 2 ; x3 x2 x1 x0
    movq mm3, qword ptr [TABLE] ; 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  [ xt2 xt0 xt2 xt0 ]
    movq mm4, qword ptr [TABLE+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, qword ptr [TABLE+32] ; 1 ; w22 w20 w18 w16
    punpckldq mm2, mm2 ; x6 x2 x6 x2  [ xt3 xt1 xt3 xt1 ]
    pmaddwd mm4, mm2 ; x6*w07+x2*w05 x6*w03+x2*w01
    punpckhdq mm5, mm5 ; x5 x1 x5 x1  [ xt6 xt4 xt6 xt4 ]
    pmaddwd mm0, qword ptr [TABLE+16] ; x4*w14+x0*w12 x4*w10+x0*w08
    punpckhdq mm6, mm6 ; x7 x3 x7 x3  [ xt7 xt5 xt7 xt5 ]
    movq mm7, qword ptr [TABLE+40] ; 7 ; w23 w21 w19 w17
    pmaddwd mm1, mm5 ; x5*w22+x1*w20 x5*w18+x1*w16
    paddd mm3, qword ptr [round_frw_row] ; +rounder (y2,y0)
    pmaddwd mm7, mm6 ; x7*w23+x3*w21 x7*w19+x3*w17
    pmaddwd mm2, qword ptr [TABLE+24] ; x6*w15+x2*w13 x6*w11+x2*w09
    paddd mm3, mm4 ; 4 ; a1=sum(even1) a0=sum(even0) 
    pmaddwd mm5, qword ptr [TABLE+48] ; x5*w30+x1*w28 x5*w26+x1*w24
    ;
    pmaddwd mm6, qword ptr [TABLE+56] ; x7*w31+x3*w29 x7*w27+x3*w25
    paddd mm1, mm7 ; 7 ; b1=sum(odd1) b0=sum(odd0) 
    paddd mm0, qword ptr [round_frw_row] ; +rounder (y6,y4)
    psrad mm3, SHIFT_FRW_ROW_CLIP1 ;
    paddd mm1, qword ptr [round_frw_row] ; +rounder (y3,y1)
    paddd mm0, mm2 ; 2 ; a3=sum(even3) a2=sum(even2) 
    paddd mm5, qword ptr [round_frw_row] ; +rounder (y7,y5)
    psrad mm1, SHIFT_FRW_ROW_CLIP1 ;
    paddd mm5, mm6 ; 6 ; b3=sum(odd3) b2=sum(odd2) 
    psrad mm0, SHIFT_FRW_ROW_CLIP1 ;
    add OUT, 16;  
    psrad mm5, SHIFT_FRW_ROW_CLIP1;
    add INP, 16;  
    packssdw mm3, mm0 ;
    packssdw mm1, mm5 ;
    movq mm6, mm3;    
    punpcklwd mm3, mm1; 
    sub edi, 0x01;   
    
    punpckhwd mm6, mm1; 
    add TABLE,64;  
    psraw mm3, SHIFT_FRW_ROW_CLIP2;  
    psraw mm6, SHIFT_FRW_ROW_CLIP2;  
    movq qword ptr [OUT-16], mm3 ; 1 ; save y3 y2 y1 y0
    movq qword ptr [OUT-8], mm6 ; 7 ; save y7 y6 y5 y4
    cmp edi, 0x00;
    jg lp_mmx_fdct_row1;  
    emms;
    }
    
  
}