/* * idct_mmx.c * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org> * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca> * * This file is part of mpeg2dec, a free MPEG-2 video stream decoder. * See http://libmpeg2.sourceforge.net/ for updates. * * mpeg2dec 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. * * mpeg2dec 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 */#include <stddef.h>#include <inttypes.h>#include "config.h"#ifdef ARCH_X86#pragma warning(disable: 4305 4309 4799 4127 4701)#include "mpeg2.h"#include "attributes.h"#include "mpeg2_internal.h"#include "mmx.h"#ifdef __cplusplususing namespace csimd;#endif#define ROW_SHIFT 15#define COL_SHIFT 6#define round(bias) ((int)(((bias)+0.5) * (1<<ROW_SHIFT)))#define rounder(bias) {round (bias), round (bias)}#if 0/* C row IDCT - its just here to document the MMXEXT and MMX versions */static inline void idct_row (int16_t * row, int offset,			     int16_t * table, int32_t * rounder){    int C1, C2, C3, C4, C5, C6, C7;    int a0, a1, a2, a3, b0, b1, b2, b3;    row += offset;    C1 = table[1];    C2 = table[2];    C3 = table[3];    C4 = table[4];    C5 = table[5];    C6 = table[6];    C7 = table[7];    a0 = C4*row[0] + C2*row[2] + C4*row[4] + C6*row[6] + *rounder;    a1 = C4*row[0] + C6*row[2] - C4*row[4] - C2*row[6] + *rounder;    a2 = C4*row[0] - C6*row[2] - C4*row[4] + C2*row[6] + *rounder;    a3 = C4*row[0] - C2*row[2] + C4*row[4] - C6*row[6] + *rounder;    b0 = C1*row[1] + C3*row[3] + C5*row[5] + C7*row[7];    b1 = C3*row[1] - C7*row[3] - C1*row[5] - C5*row[7];    b2 = C5*row[1] - C1*row[3] + C7*row[5] + C3*row[7];    b3 = C7*row[1] - C5*row[3] + C3*row[5] - C1*row[7];    row[0] = (a0 + b0) >> ROW_SHIFT;    row[1] = (a1 + b1) >> ROW_SHIFT;    row[2] = (a2 + b2) >> ROW_SHIFT;    row[3] = (a3 + b3) >> ROW_SHIFT;    row[4] = (a3 - b3) >> ROW_SHIFT;    row[5] = (a2 - b2) >> ROW_SHIFT;    row[6] = (a1 - b1) >> ROW_SHIFT;    row[7] = (a0 - b0) >> ROW_SHIFT;}#endif/* MMXEXT row IDCT */#define mmxext_table(c1,c2,c3,c4,c5,c6,c7)	{  c4,  c2, -c4, -c2,	\						   c4,  c6,  c4,  c6,	\						   c1,  c3, -c1, -c5,	\						   c5,  c7,  c3, -c7,	\						   c4, -c6,  c4, -c6,	\						  -c4,  c2,  c4, -c2,	\						   c5, -c1,  c3, -c1,	\						   c7,  c3,  c7, -c5 }static __forceinline void mmxext_row_head (int16_t * const row, const int offset,				    const int16_t * const table,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    movq_m2r (*(row+offset), mm2);	/* mm2 = x6 x4 x2 x0 */    movq_m2r (*(row+offset+4), mm5);	/* mm5 = x7 x5 x3 x1 */    movq_r2r (mm2, mm0);		/* mm0 = x6 x4 x2 x0 */    movq_m2r (*table, mm3);		/* mm3 = -C2 -C4 C2 C4 */    movq_r2r (mm5, mm6);		/* mm6 = x7 x5 x3 x1 */    movq_m2r (*(table+4), mm4);		/* mm4 = C6 C4 C6 C4 */    pmaddwd_r2r (mm0, mm3);		/* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */    pshufw_r2r (mm2, mm2, 0x4e);	/* mm2 = x2 x0 x6 x4 */}static __forceinline void mmxext_row (const int16_t * const table,			       const int32_t * const rounder,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    movq_m2r (*(table+8), mm1);		/* mm1 = -C5 -C1 C3 C1 */    pmaddwd_r2r (mm2, mm4);		/* mm4 = C4*x0+C6*x2 C4*x4+C6*x6 */    pmaddwd_m2r (*(table+16), mm0);	/* mm0 = C4*x4-C6*x6 C4*x0-C6*x2 */    pshufw_r2r (mm6, mm6, 0x4e);	/* mm6 = x3 x1 x7 x5 */    movq_m2r (*(table+12), mm7);	/* mm7 = -C7 C3 C7 C5 */    pmaddwd_r2r (mm5, mm1);		/* mm1 = -C1*x5-C5*x7 C1*x1+C3*x3 */    paddd_m2r (*rounder, mm3);		/* mm3 += rounder */    pmaddwd_r2r (mm6, mm7);		/* mm7 = C3*x1-C7*x3 C5*x5+C7*x7 */    pmaddwd_m2r (*(table+20), mm2);	/* mm2 = C4*x0-C2*x2 -C4*x4+C2*x6 */    paddd_r2r (mm4, mm3);		/* mm3 = a1 a0 + rounder */    pmaddwd_m2r (*(table+24), mm5);	/* mm5 = C3*x5-C1*x7 C5*x1-C1*x3 */    movq_r2r (mm3, mm4);		/* mm4 = a1 a0 + rounder */    pmaddwd_m2r (*(table+28), mm6);	/* mm6 = C7*x1-C5*x3 C7*x5+C3*x7 */    paddd_r2r (mm7, mm1);		/* mm1 = b1 b0 */    paddd_m2r (*rounder, mm0);		/* mm0 += rounder */    psubd_r2r (mm1, mm3);		/* mm3 = a1-b1 a0-b0 + rounder */    psrad_i2r (ROW_SHIFT, mm3);		/* mm3 = y6 y7 */    paddd_r2r (mm4, mm1);		/* mm1 = a1+b1 a0+b0 + rounder */    paddd_r2r (mm2, mm0);		/* mm0 = a3 a2 + rounder */    psrad_i2r (ROW_SHIFT, mm1);		/* mm1 = y1 y0 */    paddd_r2r (mm6, mm5);		/* mm5 = b3 b2 */    movq_r2r (mm0, mm4);		/* mm4 = a3 a2 + rounder */    paddd_r2r (mm5, mm0);		/* mm0 = a3+b3 a2+b2 + rounder */    psubd_r2r (mm5, mm4);		/* mm4 = a3-b3 a2-b2 + rounder */}static __forceinline void mmxext_row_tail (int16_t * const row, const int store,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    psrad_i2r (ROW_SHIFT, mm0);		/* mm0 = y3 y2 */    psrad_i2r (ROW_SHIFT, mm4);		/* mm4 = y4 y5 */    packssdw_r2r (mm0, mm1);		/* mm1 = y3 y2 y1 y0 */    packssdw_r2r (mm3, mm4);		/* mm4 = y6 y7 y4 y5 */    movq_r2m (mm1, *(row+store));	/* save y3 y2 y1 y0 */    pshufw_r2r (mm4, mm4, 0xb1);	/* mm4 = y7 y6 y5 y4 */    /* slot */    movq_r2m (mm4, *(row+store+4));	/* save y7 y6 y5 y4 */}static __forceinline void mmxext_row_mid (int16_t * const row, const int store,				   const int offset,				   const int16_t * const table,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    movq_m2r (*(row+offset), mm2);	/* mm2 = x6 x4 x2 x0 */    psrad_i2r (ROW_SHIFT, mm0);		/* mm0 = y3 y2 */    movq_m2r (*(row+offset+4), mm5);	/* mm5 = x7 x5 x3 x1 */    psrad_i2r (ROW_SHIFT, mm4);		/* mm4 = y4 y5 */    packssdw_r2r (mm0, mm1);		/* mm1 = y3 y2 y1 y0 */    movq_r2r (mm5, mm6);		/* mm6 = x7 x5 x3 x1 */    packssdw_r2r (mm3, mm4);		/* mm4 = y6 y7 y4 y5 */    movq_r2r (mm2, mm0);		/* mm0 = x6 x4 x2 x0 */    movq_r2m (mm1, *(row+store));	/* save y3 y2 y1 y0 */    pshufw_r2r (mm4, mm4, 0xb1);	/* mm4 = y7 y6 y5 y4 */    movq_m2r (*table, mm3);		/* mm3 = -C2 -C4 C2 C4 */    movq_r2m (mm4, *(row+store+4));	/* save y7 y6 y5 y4 */    pmaddwd_r2r (mm0, mm3);		/* mm3 = -C4*x4-C2*x6 C4*x0+C2*x2 */    movq_m2r (*(table+4), mm4);		/* mm4 = C6 C4 C6 C4 */    pshufw_r2r (mm2, mm2, 0x4e);	/* mm2 = x2 x0 x6 x4 */}/* MMX row IDCT */#define mmx_table(c1,c2,c3,c4,c5,c6,c7)	{  c4,  c2,  c4,  c6,	\					   c4,  c6, -c4, -c2,	\					   c1,  c3,  c3, -c7,	\					   c5,  c7, -c1, -c5,	\					   c4, -c6,  c4, -c2,	\					  -c4,  c2,  c4, -c6,	\					   c5, -c1,  c7, -c5,	\					   c7,  c3,  c3, -c1 }static __forceinline void mmx_row_head (int16_t * const row, const int offset,				 const int16_t * const table,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    movq_m2r (*(row+offset), mm2);	/* mm2 = x6 x4 x2 x0 */    movq_m2r (*(row+offset+4), mm5);	/* mm5 = x7 x5 x3 x1 */    movq_r2r (mm2, mm0);		/* mm0 = x6 x4 x2 x0 */    movq_m2r (*table, mm3);		/* mm3 = C6 C4 C2 C4 */    movq_r2r (mm5, mm6);		/* mm6 = x7 x5 x3 x1 */    punpckldq_r2r (mm0, mm0);		/* mm0 = x2 x0 x2 x0 */    movq_m2r (*(table+4), mm4);		/* mm4 = -C2 -C4 C6 C4 */    pmaddwd_r2r (mm0, mm3);		/* mm3 = C4*x0+C6*x2 C4*x0+C2*x2 */    movq_m2r (*(table+8), mm1);		/* mm1 = -C7 C3 C3 C1 */    punpckhdq_r2r (mm2, mm2);		/* mm2 = x6 x4 x6 x4 */}static __forceinline void mmx_row (const int16_t * const table,			    const int32_t * const rounder,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    pmaddwd_r2r (mm2, mm4);		/* mm4 = -C4*x4-C2*x6 C4*x4+C6*x6 */    punpckldq_r2r (mm5, mm5);		/* mm5 = x3 x1 x3 x1 */    pmaddwd_m2r (*(table+16), mm0);	/* mm0 = C4*x0-C2*x2 C4*x0-C6*x2 */    punpckhdq_r2r (mm6, mm6);		/* mm6 = x7 x5 x7 x5 */    movq_m2r (*(table+12), mm7);	/* mm7 = -C5 -C1 C7 C5 */    pmaddwd_r2r (mm5, mm1);		/* mm1 = C3*x1-C7*x3 C1*x1+C3*x3 */    paddd_m2r (*rounder, mm3);		/* mm3 += rounder */    pmaddwd_r2r (mm6, mm7);		/* mm7 = -C1*x5-C5*x7 C5*x5+C7*x7 */    pmaddwd_m2r (*(table+20), mm2);	/* mm2 = C4*x4-C6*x6 -C4*x4+C2*x6 */    paddd_r2r (mm4, mm3);		/* mm3 = a1 a0 + rounder */    pmaddwd_m2r (*(table+24), mm5);	/* mm5 = C7*x1-C5*x3 C5*x1-C1*x3 */    movq_r2r (mm3, mm4);		/* mm4 = a1 a0 + rounder */    pmaddwd_m2r (*(table+28), mm6);	/* mm6 = C3*x5-C1*x7 C7*x5+C3*x7 */    paddd_r2r (mm7, mm1);		/* mm1 = b1 b0 */    paddd_m2r (*rounder, mm0);		/* mm0 += rounder */    psubd_r2r (mm1, mm3);		/* mm3 = a1-b1 a0-b0 + rounder */    psrad_i2r (ROW_SHIFT, mm3);		/* mm3 = y6 y7 */    paddd_r2r (mm4, mm1);		/* mm1 = a1+b1 a0+b0 + rounder */    paddd_r2r (mm2, mm0);		/* mm0 = a3 a2 + rounder */    psrad_i2r (ROW_SHIFT, mm1);		/* mm1 = y1 y0 */    paddd_r2r (mm6, mm5);		/* mm5 = b3 b2 */    movq_r2r (mm0, mm7);		/* mm7 = a3 a2 + rounder */    paddd_r2r (mm5, mm0);		/* mm0 = a3+b3 a2+b2 + rounder */    psubd_r2r (mm5, mm7);		/* mm7 = a3-b3 a2-b2 + rounder */}static __forceinline void mmx_row_tail (int16_t * const row, const int store,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    psrad_i2r (ROW_SHIFT, mm0);		/* mm0 = y3 y2 */    psrad_i2r (ROW_SHIFT, mm7);		/* mm7 = y4 y5 */    packssdw_r2r (mm0, mm1);		/* mm1 = y3 y2 y1 y0 */    packssdw_r2r (mm3, mm7);		/* mm7 = y6 y7 y4 y5 */    movq_r2m (mm1, *(row+store));	/* save y3 y2 y1 y0 */    movq_r2r (mm7, mm4);		/* mm4 = y6 y7 y4 y5 */    pslld_i2r (16, mm7);		/* mm7 = y7 0 y5 0 */    psrld_i2r (16, mm4);		/* mm4 = 0 y6 0 y4 */    por_r2r (mm4, mm7);			/* mm7 = y7 y6 y5 y4 */    /* slot */    movq_r2m (mm7, *(row+store+4));	/* save y7 y6 y5 y4 */}static __forceinline void mmx_row_mid (int16_t * const row, const int store,				const int offset, const int16_t * const table,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){    movq_m2r (*(row+offset), mm2);	/* mm2 = x6 x4 x2 x0 */    psrad_i2r (ROW_SHIFT, mm0);		/* mm0 = y3 y2 */    movq_m2r (*(row+offset+4), mm5);	/* mm5 = x7 x5 x3 x1 */    psrad_i2r (ROW_SHIFT, mm7);		/* mm7 = y4 y5 */    packssdw_r2r (mm0, mm1);		/* mm1 = y3 y2 y1 y0 */    movq_r2r (mm5, mm6);		/* mm6 = x7 x5 x3 x1 */    packssdw_r2r (mm3, mm7);		/* mm7 = y6 y7 y4 y5 */    movq_r2r (mm2, mm0);		/* mm0 = x6 x4 x2 x0 */    movq_r2m (mm1, *(row+store));	/* save y3 y2 y1 y0 */    movq_r2r (mm7, mm1);		/* mm1 = y6 y7 y4 y5 */    punpckldq_r2r (mm0, mm0);		/* mm0 = x2 x0 x2 x0 */    psrld_i2r (16, mm7);		/* mm7 = 0 y6 0 y4 */    movq_m2r (*table, mm3);		/* mm3 = C6 C4 C2 C4 */    pslld_i2r (16, mm1);		/* mm1 = y7 0 y5 0 */    movq_m2r (*(table+4), mm4);		/* mm4 = -C2 -C4 C6 C4 */    por_r2r (mm1, mm7);			/* mm7 = y7 y6 y5 y4 */    movq_m2r (*(table+8), mm1);		/* mm1 = -C7 C3 C3 C1 */    punpckhdq_r2r (mm2, mm2);		/* mm2 = x6 x4 x6 x4 */    movq_r2m (mm7, *(row+store+4));	/* save y7 y6 y5 y4 */    pmaddwd_r2r (mm0, mm3);		/* mm3 = C4*x0+C6*x2 C4*x0+C2*x2 */}#if 0/* C column IDCT - its just here to document the MMXEXT and MMX versions */static inline void idct_col (int16_t * col, int offset){/* multiplication - as implemented on mmx */#define F(c,x) (((c) * (x)) >> 16)/* saturation - it helps us handle torture test cases */#define S(x) (((x)>32767) ? 32767 : ((x)<-32768) ? -32768 : (x))    int16_t x0, x1, x2, x3, x4, x5, x6, x7;    int16_t y0, y1, y2, y3, y4, y5, y6, y7;    int16_t a0, a1, a2, a3, b0, b1, b2, b3;    int16_t u04, v04, u26, v26, u17, v17, u35, v35, u12, v12;    col += offset;    x0 = col[0*8];    x1 = col[1*8];    x2 = col[2*8];    x3 = col[3*8];    x4 = col[4*8];    x5 = col[5*8];    x6 = col[6*8];    x7 = col[7*8];    u04 = S (x0 + x4);    v04 = S (x0 - x4);    u26 = S (F (T2, x6) + x2);    v26 = S (F (T2, x2) - x6);    a0 = S (u04 + u26);    a1 = S (v04 + v26);    a2 = S (v04 - v26);    a3 = S (u04 - u26);    u17 = S (F (T1, x7) + x1);    v17 = S (F (T1, x1) - x7);    u35 = S (F (T3, x5) + x3);    v35 = S (F (T3, x3) - x5);    b0 = S (u17 + u35);    b3 = S (v17 - v35);    u12 = S (u17 - u35);    v12 = S (v17 + v35);    u12 = S (2 * F (C4, u12));    v12 = S (2 * F (C4, v12));    b1 = S (u12 + v12);    b2 = S (u12 - v12);    y0 = S (a0 + b0) >> COL_SHIFT;    y1 = S (a1 + b1) >> COL_SHIFT;    y2 = S (a2 + b2) >> COL_SHIFT;    y3 = S (a3 + b3) >> COL_SHIFT;    y4 = S (a3 - b3) >> COL_SHIFT;    y5 = S (a2 - b2) >> COL_SHIFT;    y6 = S (a1 - b1) >> COL_SHIFT;    y7 = S (a0 - b0) >> COL_SHIFT;    col[0*8] = y0;    col[1*8] = y1;    col[2*8] = y2;    col[3*8] = y3;    col[4*8] = y4;    col[5*8] = y5;    col[6*8] = y6;    col[7*8] = y7;}#endif/* MMX column IDCT */static __forceinline void idct_col (int16_t * const col, const int offset,__m64 &mm0,__m64 &mm1,__m64 &mm2,__m64 &mm3,__m64 &mm4,__m64 &mm5,__m64 &mm6,__m64 &mm7){#define T1 13036#define T2 27146