#define T3 43790#define C4 23170    const __m64 _T1 = _mm_set1_pi16(T1);    const __m64 _T2 = _mm_set1_pi16(T2);    const __m64 _T3 = _mm_set1_pi16(T3);    const __m64 _C4 = _mm_set1_pi16(C4);    /* column code adapted from peter gubanov */    /* http://www.elecard.com/peter/idct.shtml */    movq_m2r (_T1, mm0);		/* mm0 = T1 */    movq_m2r (*(col+offset+1*8), mm1);	/* mm1 = x1 */    movq_r2r (mm0, mm2);		/* mm2 = T1 */    movq_m2r (*(col+offset+7*8), mm4);	/* mm4 = x7 */    pmulhw_r2r (mm1, mm0);		/* mm0 = T1*x1 */    movq_m2r (_T3, mm5);		/* mm5 = T3 */    pmulhw_r2r (mm4, mm2);		/* mm2 = T1*x7 */    movq_m2r (*(col+offset+5*8), mm6);	/* mm6 = x5 */    movq_r2r (mm5, mm7);		/* mm7 = T3-1 */    movq_m2r (*(col+offset+3*8), mm3);	/* mm3 = x3 */    psubsw_r2r (mm4, mm0);		/* mm0 = v17 */    movq_m2r (_T2, mm4);		/* mm4 = T2 */    pmulhw_r2r (mm3, mm5);		/* mm5 = (T3-1)*x3 */    paddsw_r2r (mm2, mm1);		/* mm1 = u17 */    pmulhw_r2r (mm6, mm7);		/* mm7 = (T3-1)*x5 */    /* slot */    movq_r2r (mm4, mm2);		/* mm2 = T2 */    paddsw_r2r (mm3, mm5);		/* mm5 = T3*x3 */    pmulhw_m2r (*(col+offset+2*8), mm4);/* mm4 = T2*x2 */    paddsw_r2r (mm6, mm7);		/* mm7 = T3*x5 */    psubsw_r2r (mm6, mm5);		/* mm5 = v35 */    paddsw_r2r (mm3, mm7);		/* mm7 = u35 */    movq_m2r (*(col+offset+6*8), mm3);	/* mm3 = x6 */    movq_r2r (mm0, mm6);		/* mm6 = v17 */    pmulhw_r2r (mm3, mm2);		/* mm2 = T2*x6 */    psubsw_r2r (mm5, mm0);		/* mm0 = b3 */    psubsw_r2r (mm3, mm4);		/* mm4 = v26 */    paddsw_r2r (mm6, mm5);		/* mm5 = v12 */    movq_r2m (mm0, *(col+offset+3*8));	/* save b3 in scratch0 */    movq_r2r (mm1, mm6);		/* mm6 = u17 */    paddsw_m2r (*(col+offset+2*8), mm2);/* mm2 = u26 */    paddsw_r2r (mm7, mm6);		/* mm6 = b0 */    psubsw_r2r (mm7, mm1);		/* mm1 = u12 */    movq_r2r (mm1, mm7);		/* mm7 = u12 */    movq_m2r (*(col+offset+0*8), mm3);	/* mm3 = x0 */    paddsw_r2r (mm5, mm1);		/* mm1 = u12+v12 */    movq_m2r (_C4, mm0);		/* mm0 = C4/2 */    psubsw_r2r (mm5, mm7);		/* mm7 = u12-v12 */    movq_r2m (mm6, *(col+offset+5*8));	/* save b0 in scratch1 */    pmulhw_r2r (mm0, mm1);		/* mm1 = b1/2 */    movq_r2r (mm4, mm6);		/* mm6 = v26 */    pmulhw_r2r (mm0, mm7);		/* mm7 = b2/2 */    movq_m2r (*(col+offset+4*8), mm5);	/* mm5 = x4 */    movq_r2r (mm3, mm0);		/* mm0 = x0 */    psubsw_r2r (mm5, mm3);		/* mm3 = v04 */    paddsw_r2r (mm5, mm0);		/* mm0 = u04 */    paddsw_r2r (mm3, mm4);		/* mm4 = a1 */    movq_r2r (mm0, mm5);		/* mm5 = u04 */    psubsw_r2r (mm6, mm3);		/* mm3 = a2 */    paddsw_r2r (mm2, mm5);		/* mm5 = a0 */    paddsw_r2r (mm1, mm1);		/* mm1 = b1 */    psubsw_r2r (mm2, mm0);		/* mm0 = a3 */    paddsw_r2r (mm7, mm7);		/* mm7 = b2 */    movq_r2r (mm3, mm2);		/* mm2 = a2 */    movq_r2r (mm4, mm6);		/* mm6 = a1 */    paddsw_r2r (mm7, mm3);		/* mm3 = a2+b2 */    psraw_i2r (COL_SHIFT, mm3);		/* mm3 = y2 */    paddsw_r2r (mm1, mm4);		/* mm4 = a1+b1 */    psraw_i2r (COL_SHIFT, mm4);		/* mm4 = y1 */    psubsw_r2r (mm1, mm6);		/* mm6 = a1-b1 */    movq_m2r (*(col+offset+5*8), mm1);	/* mm1 = b0 */    psubsw_r2r (mm7, mm2);		/* mm2 = a2-b2 */    psraw_i2r (COL_SHIFT, mm6);		/* mm6 = y6 */    movq_r2r (mm5, mm7);		/* mm7 = a0 */    movq_r2m (mm4, *(col+offset+1*8));	/* save y1 */    psraw_i2r (COL_SHIFT, mm2);		/* mm2 = y5 */    movq_r2m (mm3, *(col+offset+2*8));	/* save y2 */    paddsw_r2r (mm1, mm5);		/* mm5 = a0+b0 */    movq_m2r (*(col+offset+3*8), mm4);	/* mm4 = b3 */    psubsw_r2r (mm1, mm7);		/* mm7 = a0-b0 */    psraw_i2r (COL_SHIFT, mm5);		/* mm5 = y0 */    movq_r2r (mm0, mm3);		/* mm3 = a3 */    movq_r2m (mm2, *(col+offset+5*8));	/* save y5 */    psubsw_r2r (mm4, mm3);		/* mm3 = a3-b3 */    psraw_i2r (COL_SHIFT, mm7);		/* mm7 = y7 */    paddsw_r2r (mm0, mm4);		/* mm4 = a3+b3 */    movq_r2m (mm5, *(col+offset+0*8));	/* save y0 */    psraw_i2r (COL_SHIFT, mm3);		/* mm3 = y4 */    movq_r2m (mm6, *(col+offset+6*8));	/* save y6 */    psraw_i2r (COL_SHIFT, mm4);		/* mm4 = y3 */    movq_r2m (mm7, *(col+offset+7*8));	/* save y7 */    movq_r2m (mm3, *(col+offset+4*8));	/* save y4 */    movq_r2m (mm4, *(col+offset+3*8));	/* save y3 */}static __align16(const int32_t, rounder0[]) = rounder ((1 << (COL_SHIFT - 1)) - 0.5);static __align16(const int32_t, rounder4[]) = rounder (0);static __align16(const int32_t, rounder1[]) = rounder (1.25683487303);	/* C1*(C1/C4+C1+C7)/2 */static __align16(const int32_t, rounder7[]) = rounder (-0.25);		/* C1*(C7/C4+C7-C1)/2 */static __align16(const int32_t, rounder2[]) = rounder (0.60355339059);	/* C2 * (C6+C2)/2 */static __align16(const int32_t, rounder6[]) = rounder (-0.25);		/* C2 * (C6-C2)/2 */static __align16(const int32_t, rounder3[]) = rounder (0.087788325588);	/* C3*(-C3/C4+C3+C5)/2 */static __align16(const int32_t, rounder5[]) = rounder (-0.441341716183);	/* C3*(-C5/C4+C5-C3)/2 */#define declare_idct(idct,table,idct_row_head,idct_row,idct_row_tail,idct_row_mid)	\static inline void idct (int16_t * const block)				\{									\    __m64 mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7;                              \    static __align16(const int16_t,table04[]) =			\	table (22725, 21407, 19266, 16384, 12873,  8867, 4520);		\    static __align16(const int16_t,table17[]) =			\	table (31521, 29692, 26722, 22725, 17855, 12299, 6270);		\    static __align16(const int16_t,table26[]) =			\	table (29692, 27969, 25172, 21407, 16819, 11585, 5906);		\    static __align16(const int16_t,table35[]) =			\	table (26722, 25172, 22654, 19266, 15137, 10426, 5315);		\									\    idct_row_head (block, 0*8, table04,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table04, rounder0,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 0*8, 4*8, table04,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table04, rounder4,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 4*8, 1*8, table17,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table17, rounder1,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 1*8, 7*8, table17,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table17, rounder7,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 7*8, 2*8, table26,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table26, rounder2,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 2*8, 6*8, table26,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table26, rounder6,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 6*8, 3*8, table35,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table35, rounder3,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_mid (block, 3*8, 5*8, table35,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);				\    idct_row (table35, rounder5,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);					\    idct_row_tail (block, 5*8,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);						\									\    idct_col (block, 0,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);						\    idct_col (block, 4,mm0,mm1,mm2,mm3,mm4,mm5,mm6,mm7);						\}#define COPY_MMX(offset,r0,r1,r2)	\do {					\    movq_m2r (*(block+offset), r0);	\    dest += stride;			\    movq_m2r (*(block+offset+4), r1);	\    movq_r2m (r2, *dest);		\    packuswb_r2r (r1, r0);		\} while (0)static inline void block_copy (int16_t * const block, uint8_t * dest,			       const int stride){    __m64 mm0,mm1,mm2,mm3;    movq_m2r (*(block+0*8), mm0);    movq_m2r (*(block+0*8+4), mm1);    movq_m2r (*(block+1*8), mm2);    packuswb_r2r (mm1, mm0);    movq_m2r (*(block+1*8+4), mm3);    movq_r2m (mm0, *dest);    packuswb_r2r (mm3, mm2);    COPY_MMX (2*8, mm0, mm1, mm2);    COPY_MMX (3*8, mm2, mm3, mm0);    COPY_MMX (4*8, mm0, mm1, mm2);    COPY_MMX (5*8, mm2, mm3, mm0);    COPY_MMX (6*8, mm0, mm1, mm2);    COPY_MMX (7*8, mm2, mm3, mm0);    movq_r2m (mm2, *(dest+stride));}#define ADD_MMX(offset,r1,r2,r3,r4)	\do {					\    movq_m2r (*(dest+2*stride), r1);	\    packuswb_r2r (r4, r3);		\    movq_r2r (r1, r2);			\    dest += stride;			\    movq_r2m (r3, *dest);		\    punpcklbw_r2r (mm0, r1);		\    paddsw_m2r (*(block+offset), r1);	\    punpckhbw_r2r (mm0, r2);		\    paddsw_m2r (*(block+offset+4), r2);	\} while (0)static __forceinline void block_add (int16_t * const block, uint8_t * dest,			      const int stride){    __m64 mm0,mm1,mm2,mm3,mm4;    movq_m2r (*dest, mm1);    pxor_r2r (mm0, mm0);    movq_m2r (*(dest+stride), mm3);    movq_r2r (mm1, mm2);    punpcklbw_r2r (mm0, mm1);    movq_r2r (mm3, mm4);    paddsw_m2r (*(block+0*8), mm1);    punpckhbw_r2r (mm0, mm2);    paddsw_m2r (*(block+0*8+4), mm2);    punpcklbw_r2r (mm0, mm3);    paddsw_m2r (*(block+1*8), mm3);    packuswb_r2r (mm2, mm1);    punpckhbw_r2r (mm0, mm4);    movq_r2m (mm1, *dest);    paddsw_m2r (*(block+1*8+4), mm4);    ADD_MMX (2*8, mm1, mm2, mm3, mm4);    ADD_MMX (3*8, mm3, mm4, mm1, mm2);    ADD_MMX (4*8, mm1, mm2, mm3, mm4);    ADD_MMX (5*8, mm3, mm4, mm1, mm2);    ADD_MMX (6*8, mm1, mm2, mm3, mm4);    ADD_MMX (7*8, mm3, mm4, mm1, mm2);    packuswb_r2r (mm4, mm3);    movq_r2m (mm3, *(dest+stride));}static __forceinline void block_zero (int16_t * const block){    __m64 mm0;    pxor_r2r (mm0, mm0);    movq_r2m (mm0, *(block+0*4));    movq_r2m (mm0, *(block+1*4));    movq_r2m (mm0, *(block+2*4));    movq_r2m (mm0, *(block+3*4));    movq_r2m (mm0, *(block+4*4));    movq_r2m (mm0, *(block+5*4));    movq_r2m (mm0, *(block+6*4));    movq_r2m (mm0, *(block+7*4));    movq_r2m (mm0, *(block+8*4));    movq_r2m (mm0, *(block+9*4));    movq_r2m (mm0, *(block+10*4));    movq_r2m (mm0, *(block+11*4));    movq_r2m (mm0, *(block+12*4));    movq_r2m (mm0, *(block+13*4));    movq_r2m (mm0, *(block+14*4));    movq_r2m (mm0, *(block+15*4));}#define CPU_MMXEXT 0#define CPU_MMX 1#define dup4(reg)			\do {					\    if (cpu != CPU_MMXEXT) {		\	punpcklwd_r2r (reg, reg);	\	punpckldq_r2r (reg, reg);	\    } else				\	pshufw_r2r (reg, reg, 0x00);	\} while (0)static __forceinline void block_add_DC (int16_t * const block, uint8_t * dest,				 const int stride, const int cpu){    __m64 mm0,mm1,mm2,mm3;    movd_v2r ((block[0] + 64) >> 7, mm0);    pxor_r2r (mm1, mm1);    movq_m2r (*dest, mm2);    dup4 (mm0);    psubsw_r2r (mm0, mm1);    packuswb_r2r (mm0, mm0);    paddusb_r2r (mm0, mm2);    packuswb_r2r (mm1, mm1);    movq_m2r (*(dest + stride), mm3);    psubusb_r2r (mm1, mm2);    block[0] = 0;    paddusb_r2r (mm0, mm3);    movq_r2m (mm2, *dest);    psubusb_r2r (mm1, mm3);    movq_m2r (*(dest + 2*stride), mm2);    dest += stride;    movq_r2m (mm3, *dest);    paddusb_r2r (mm0, mm2);    movq_m2r (*(dest + 2*stride), mm3);    psubusb_r2r (mm1, mm2);    dest += stride;    paddusb_r2r (mm0, mm3);    movq_r2m (mm2, *dest);    psubusb_r2r (mm1, mm3);    movq_m2r (*(dest + 2*stride), mm2);    dest += stride;    movq_r2m (mm3, *dest);    paddusb_r2r (mm0, mm2);    movq_m2r (*(dest + 2*stride), mm3);    psubusb_r2r (mm1, mm2);    dest += stride;    paddusb_r2r (mm0, mm3);    movq_r2m (mm2, *dest);    psubusb_r2r (mm1, mm3);    movq_m2r (*(dest + 2*stride), mm2);    dest += stride;    movq_r2m (mm3, *dest);    paddusb_r2r (mm0, mm2);    movq_m2r (*(dest + 2*stride), mm3);    psubusb_r2r (mm1, mm2);    block[63] = 0;    paddusb_r2r (mm0, mm3);    movq_r2m (mm2, *(dest + stride));    psubusb_r2r (mm1, mm3);    movq_r2m (mm3, *(dest + 2*stride));}declare_idct (mmxext_idct, mmxext_table,	      mmxext_row_head, mmxext_row, mmxext_row_tail, mmxext_row_mid)void mpeg2_idct_copy_mmxext (int16_t * const block, uint8_t * const dest,			     const int stride){    mmxext_idct (block);    block_copy (block, dest, stride);    block_zero (block);}void mpeg2_idct_add_mmxext (const int last, int16_t * const block,			    uint8_t * const dest, const int stride){    if (last != 129 || (block[0] & (7 << 4)) == (4 << 4)) {	mmxext_idct (block);	block_add (block, dest, stride);	block_zero (block);    } else	block_add_DC (block, dest, stride, CPU_MMXEXT);}declare_idct (mmx_idct, mmx_table,	      mmx_row_head, mmx_row, mmx_row_tail, mmx_row_mid)void mpeg2_idct_copy_mmx (int16_t * const block, uint8_t * const dest,			  const int stride){    mmx_idct (block);    block_copy (block, dest, stride);    block_zero (block);}void mpeg2_idct_add_mmx (const int last, int16_t * const block,			 uint8_t * const dest, const int stride){    if (last != 129 || (block[0] & (7 << 4)) == (4 << 4)) {	mmx_idct (block);	block_add (block, dest, stride);	block_zero (block);    } else	block_add_DC (block, dest, stride, CPU_MMX);}void mpeg2_idct_mmx_init (void){    extern __align16(uint8_t,mpeg2_scan_norm[64]);    extern __align16(uint8_t,mpeg2_scan_alt[64]);    int i, j;    /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */    for (i = 0; i < 64; i++) {	j = mpeg2_scan_norm[i];	mpeg2_scan_norm[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);	j = mpeg2_scan_alt[i];	mpeg2_scan_alt[i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2);    }}#endif