Attachment #291523 for bug #389323




#define rounder(bias) {round (bias), round (bias)}
#define rounder_sse2(bias) {round (bias), round (bias), round (bias), round (bias)}


#if 0
/* C row IDCT - it is just here to document the MMXEXT and MMX versions */
static inline void idct_row (int16_t * row, int offset,

    /* Almost identical to mmxext version:  */
    /* just do both 4x8 columns in paraller */

    static/* const*/ short t1_vector[] ATTR_ALIGN(16) = {T1,T1,T1,T1,T1,T1,T1,T1};
    static/* const*/ short t2_vector[] ATTR_ALIGN(16) = {T2,T2,T2,T2,T2,T2,T2,T2};
    static/* const*/ short t3_vector[] ATTR_ALIGN(16) = {T3,T3,T3,T3,T3,T3,T3,T3};
    static/* const*/ short c4_vector[] ATTR_ALIGN(16) = {C4,C4,C4,C4,C4,C4,C4,C4};

#if defined(__x86_64__)


/* MMX column IDCT */
static inline void idct_col (int16_t * const col, const int offset)
{
    static/* const*/ short t1_vector[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
    static/* const*/ short t2_vector[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
    static/* const*/ short t3_vector[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
    static/* const*/ short c4_vector[] ATTR_ALIGN(8) = {C4,C4,C4,C4};

    /* column code adapted from peter gubanov */
    /* http://www.elecard.com/peter/idct.shtml */

}


static/* const*/ int32_t rounder0[] ATTR_ALIGN(8) =
    rounder ((1 << (COL_SHIFT - 1)) - 0.5);
static/* const*/ int32_t rounder4[] ATTR_ALIGN(8) = rounder (0);
static/* const*/ int32_t rounder1[] ATTR_ALIGN(8) =
    rounder (1.25683487303);	/* C1*(C1/C4+C1+C7)/2 */
static/* const*/ int32_t rounder7[] ATTR_ALIGN(8) =
    rounder (-0.25);		/* C1*(C7/C4+C7-C1)/2 */
static/* const*/ int32_t rounder2[] ATTR_ALIGN(8) =
    rounder (0.60355339059);	/* C2 * (C6+C2)/2 */
static/* const*/ int32_t rounder6[] ATTR_ALIGN(8) =
    rounder (-0.25);		/* C2 * (C6-C2)/2 */
static/* const*/ int32_t rounder3[] ATTR_ALIGN(8) =
    rounder (0.087788325588);	/* C3*(-C3/C4+C3+C5)/2 */
static/* const*/ int32_t rounder5[] ATTR_ALIGN(8) =
    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)				\
{									\
    static/* const*/ int16_t table04[] ATTR_ALIGN(16) =			\
	table (22725, 21407, 19266, 16384, 12873,  8867, 4520);		\
    static/* const*/ int16_t table17[] ATTR_ALIGN(16) =			\
	table (31521, 29692, 26722, 22725, 17855, 12299, 6270);		\
    static/* const*/ int16_t table26[] ATTR_ALIGN(16) =			\
	table (29692, 27969, 25172, 21407, 16819, 11585, 5906);		\
    static/* const*/ int16_t table35[] ATTR_ALIGN(16) =			\
	table (26722, 25172, 22654, 19266, 15137, 10426, 5315);		\
									\
    idct_row_head (block, 0*8, table04);				\


static inline void sse2_idct (int16_t * const block)
{
    static/* const*/ int16_t table04[] ATTR_ALIGN(16) =
	sse2_table (22725, 21407, 19266, 16384, 12873,  8867, 4520);
    static/* const*/ int16_t table17[] ATTR_ALIGN(16) =
	sse2_table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
    static/* const*/ int16_t table26[] ATTR_ALIGN(16) =
	sse2_table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
    static/* const*/ int16_t table35[] ATTR_ALIGN(16) =
	sse2_table (26722, 25172, 22654, 19266, 15137, 10426, 5315);

    static/* const*/ int32_t rounder0_128[] ATTR_ALIGN(16) =
	rounder_sse2 ((1 << (COL_SHIFT - 1)) - 0.5);
    static/* const*/ int32_t rounder4_128[] ATTR_ALIGN(16) = rounder_sse2 (0);
    static/* const*/ int32_t rounder1_128[] ATTR_ALIGN(16) =
	rounder_sse2 (1.25683487303);	/* C1*(C1/C4+C1+C7)/2 */
    static/* const*/ int32_t rounder7_128[] ATTR_ALIGN(16) =
	rounder_sse2 (-0.25);		/* C1*(C7/C4+C7-C1)/2 */
    static/* const*/ int32_t rounder2_128[] ATTR_ALIGN(16) =
	rounder_sse2 (0.60355339059);	/* C2 * (C6+C2)/2 */
    static/* const*/ int32_t rounder6_128[] ATTR_ALIGN(16) =
	rounder_sse2 (-0.25);		/* C2 * (C6-C2)/2 */
    static/* const*/ int32_t rounder3_128[] ATTR_ALIGN(16) =
	rounder_sse2 (0.087788325588);	/* C3*(-C3/C4+C3+C5)/2 */
    static/* const*/ int32_t rounder5_128[] ATTR_ALIGN(16) =
	rounder_sse2 (-0.441341716183);	/* C3*(-C5/C4+C5-C3)/2 */

#if defined(__x86_64__)

Return to bug 389323

Lines 39-45 Link Here

(-)libmpeg2-0.5.1/libmpeg2/idct_mmx.c.orig (-33 / +32 lines)
39	#define rounder(bias) {round (bias), round (bias)}	39	#define rounder(bias) {round (bias), round (bias)}
40	#define rounder_sse2(bias) {round (bias), round (bias), round (bias), round (bias)}	40	#define rounder_sse2(bias) {round (bias), round (bias), round (bias), round (bias)}
41		41
42
43	#if 0	42	#if 0
44	/* C row IDCT - it is just here to document the MMXEXT and MMX versions */	43	/* C row IDCT - it is just here to document the MMXEXT and MMX versions */
45	static inline void idct_row (int16_t * row, int offset,	44	static inline void idct_row (int16_t * row, int offset,
Lines 464-473 static inline void sse2_idct_col (int16_ Link Here
464	/* Almost identical to mmxext version: */	463	/* Almost identical to mmxext version: */
465	/* just do both 4x8 columns in paraller */	464	/* just do both 4x8 columns in paraller */
466		465
467	static const short t1_vector[] ATTR_ALIGN(16) = {T1,T1,T1,T1,T1,T1,T1,T1};	466	static/* const*/ short t1_vector[] ATTR_ALIGN(16) = {T1,T1,T1,T1,T1,T1,T1,T1};
468	static const short t2_vector[] ATTR_ALIGN(16) = {T2,T2,T2,T2,T2,T2,T2,T2};	467	static/* const*/ short t2_vector[] ATTR_ALIGN(16) = {T2,T2,T2,T2,T2,T2,T2,T2};
469	static const short t3_vector[] ATTR_ALIGN(16) = {T3,T3,T3,T3,T3,T3,T3,T3};	468	static/* const*/ short t3_vector[] ATTR_ALIGN(16) = {T3,T3,T3,T3,T3,T3,T3,T3};
470	static const short c4_vector[] ATTR_ALIGN(16) = {C4,C4,C4,C4,C4,C4,C4,C4};	469	static/* const*/ short c4_vector[] ATTR_ALIGN(16) = {C4,C4,C4,C4,C4,C4,C4,C4};
471		470
472	#if defined(__x86_64__)	471	#if defined(__x86_64__)
473		472
Lines 710-719 static inline void sse2_idct_col (int16_ Link Here
710	/* MMX column IDCT */	709	/* MMX column IDCT */
711	static inline void idct_col (int16_t * const col, const int offset)	710	static inline void idct_col (int16_t * const col, const int offset)
712	{	711	{
713	static const short t1_vector[] ATTR_ALIGN(8) = {T1,T1,T1,T1};	712	static/* const*/ short t1_vector[] ATTR_ALIGN(8) = {T1,T1,T1,T1};
714	static const short t2_vector[] ATTR_ALIGN(8) = {T2,T2,T2,T2};	713	static/* const*/ short t2_vector[] ATTR_ALIGN(8) = {T2,T2,T2,T2};
715	static const short t3_vector[] ATTR_ALIGN(8) = {T3,T3,T3,T3};	714	static/* const*/ short t3_vector[] ATTR_ALIGN(8) = {T3,T3,T3,T3};
716	static const short c4_vector[] ATTR_ALIGN(8) = {C4,C4,C4,C4};	715	static/* const*/ short c4_vector[] ATTR_ALIGN(8) = {C4,C4,C4,C4};
717		716
718	/* column code adapted from peter gubanov */	717	/* column code adapted from peter gubanov */
719	/* http://www.elecard.com/peter/idct.shtml */	718	/* http://www.elecard.com/peter/idct.shtml */
Lines 847-879 static inline void idct_col (int16_t * c Link Here
847	}	846	}
848		847
849		848
850	static const int32_t rounder0[] ATTR_ALIGN(8) =	849	static/* const*/ int32_t rounder0[] ATTR_ALIGN(8) =
851	rounder ((1 << (COL_SHIFT - 1)) - 0.5);	850	rounder ((1 << (COL_SHIFT - 1)) - 0.5);
852	static const int32_t rounder4[] ATTR_ALIGN(8) = rounder (0);	851	static/* const*/ int32_t rounder4[] ATTR_ALIGN(8) = rounder (0);
853	static const int32_t rounder1[] ATTR_ALIGN(8) =	852	static/* const*/ int32_t rounder1[] ATTR_ALIGN(8) =
854	rounder (1.25683487303); /* C1(C1/C4+C1+C7)/2 /	853	rounder (1.25683487303); /* C1(C1/C4+C1+C7)/2 /
855	static const int32_t rounder7[] ATTR_ALIGN(8) =	854	static/* const*/ int32_t rounder7[] ATTR_ALIGN(8) =
856	rounder (-0.25); /* C1(C7/C4+C7-C1)/2 /	855	rounder (-0.25); /* C1(C7/C4+C7-C1)/2 /
857	static const int32_t rounder2[] ATTR_ALIGN(8) =	856	static/* const*/ int32_t rounder2[] ATTR_ALIGN(8) =
858	rounder (0.60355339059); /* C2 * (C6+C2)/2 */	857	rounder (0.60355339059); /* C2 * (C6+C2)/2 */
859	static const int32_t rounder6[] ATTR_ALIGN(8) =	858	static/* const*/ int32_t rounder6[] ATTR_ALIGN(8) =
860	rounder (-0.25); /* C2 * (C6-C2)/2 */	859	rounder (-0.25); /* C2 * (C6-C2)/2 */
861	static const int32_t rounder3[] ATTR_ALIGN(8) =	860	static/* const*/ int32_t rounder3[] ATTR_ALIGN(8) =
862	rounder (0.087788325588); /* C3(-C3/C4+C3+C5)/2 /	861	rounder (0.087788325588); /* C3(-C3/C4+C3+C5)/2 /
863	static const int32_t rounder5[] ATTR_ALIGN(8) =	862	static/* const*/ int32_t rounder5[] ATTR_ALIGN(8) =
864	rounder (-0.441341716183); /* C3(-C5/C4+C5-C3)/2 /	863	rounder (-0.441341716183); /* C3(-C5/C4+C5-C3)/2 /
865		864
866		865
867	#define declare_idct(idct,table,idct_row_head,idct_row,idct_row_tail,idct_row_mid) \	866	#define declare_idct(idct,table,idct_row_head,idct_row,idct_row_tail,idct_row_mid) \
868	static inline void idct (int16_t * const block) \	867	static inline void idct (int16_t * const block) \
869	{ \	868	{ \
870	static const int16_t table04[] ATTR_ALIGN(16) = \	869	static/* const*/ int16_t table04[] ATTR_ALIGN(16) = \
871	table (22725, 21407, 19266, 16384, 12873, 8867, 4520); \	870	table (22725, 21407, 19266, 16384, 12873, 8867, 4520); \
872	static const int16_t table17[] ATTR_ALIGN(16) = \	871	static/* const*/ int16_t table17[] ATTR_ALIGN(16) = \
873	table (31521, 29692, 26722, 22725, 17855, 12299, 6270); \	872	table (31521, 29692, 26722, 22725, 17855, 12299, 6270); \
874	static const int16_t table26[] ATTR_ALIGN(16) = \	873	static/* const*/ int16_t table26[] ATTR_ALIGN(16) = \
875	table (29692, 27969, 25172, 21407, 16819, 11585, 5906); \	874	table (29692, 27969, 25172, 21407, 16819, 11585, 5906); \
876	static const int16_t table35[] ATTR_ALIGN(16) = \	875	static/* const*/ int16_t table35[] ATTR_ALIGN(16) = \
877	table (26722, 25172, 22654, 19266, 15137, 10426, 5315); \	876	table (26722, 25172, 22654, 19266, 15137, 10426, 5315); \
878	\	877	\
879	idct_row_head (block, 0*8, table04); \	878	idct_row_head (block, 0*8, table04); \
Lines 900-928 static inline void idct (int16_t * const Link Here
900		899
901	static inline void sse2_idct (int16_t * const block)	900	static inline void sse2_idct (int16_t * const block)
902	{	901	{
903	static const int16_t table04[] ATTR_ALIGN(16) =	902	static/* const*/ int16_t table04[] ATTR_ALIGN(16) =
904	sse2_table (22725, 21407, 19266, 16384, 12873, 8867, 4520);	903	sse2_table (22725, 21407, 19266, 16384, 12873, 8867, 4520);
905	static const int16_t table17[] ATTR_ALIGN(16) =	904	static/* const*/ int16_t table17[] ATTR_ALIGN(16) =
906	sse2_table (31521, 29692, 26722, 22725, 17855, 12299, 6270);	905	sse2_table (31521, 29692, 26722, 22725, 17855, 12299, 6270);
907	static const int16_t table26[] ATTR_ALIGN(16) =	906	static/* const*/ int16_t table26[] ATTR_ALIGN(16) =
908	sse2_table (29692, 27969, 25172, 21407, 16819, 11585, 5906);	907	sse2_table (29692, 27969, 25172, 21407, 16819, 11585, 5906);
909	static const int16_t table35[] ATTR_ALIGN(16) =	908	static/* const*/ int16_t table35[] ATTR_ALIGN(16) =
910	sse2_table (26722, 25172, 22654, 19266, 15137, 10426, 5315);	909	sse2_table (26722, 25172, 22654, 19266, 15137, 10426, 5315);
911		910
912	static const int32_t rounder0_128[] ATTR_ALIGN(16) =	911	static/* const*/ int32_t rounder0_128[] ATTR_ALIGN(16) =
913	rounder_sse2 ((1 << (COL_SHIFT - 1)) - 0.5);	912	rounder_sse2 ((1 << (COL_SHIFT - 1)) - 0.5);
914	static const int32_t rounder4_128[] ATTR_ALIGN(16) = rounder_sse2 (0);	913	static/* const*/ int32_t rounder4_128[] ATTR_ALIGN(16) = rounder_sse2 (0);
915	static const int32_t rounder1_128[] ATTR_ALIGN(16) =	914	static/* const*/ int32_t rounder1_128[] ATTR_ALIGN(16) =
916	rounder_sse2 (1.25683487303); /* C1(C1/C4+C1+C7)/2 /	915	rounder_sse2 (1.25683487303); /* C1(C1/C4+C1+C7)/2 /
917	static const int32_t rounder7_128[] ATTR_ALIGN(16) =	916	static/* const*/ int32_t rounder7_128[] ATTR_ALIGN(16) =
918	rounder_sse2 (-0.25); /* C1(C7/C4+C7-C1)/2 /	917	rounder_sse2 (-0.25); /* C1(C7/C4+C7-C1)/2 /
919	static const int32_t rounder2_128[] ATTR_ALIGN(16) =	918	static/* const*/ int32_t rounder2_128[] ATTR_ALIGN(16) =
920	rounder_sse2 (0.60355339059); /* C2 * (C6+C2)/2 */	919	rounder_sse2 (0.60355339059); /* C2 * (C6+C2)/2 */
921	static const int32_t rounder6_128[] ATTR_ALIGN(16) =	920	static/* const*/ int32_t rounder6_128[] ATTR_ALIGN(16) =
922	rounder_sse2 (-0.25); /* C2 * (C6-C2)/2 */	921	rounder_sse2 (-0.25); /* C2 * (C6-C2)/2 */
923	static const int32_t rounder3_128[] ATTR_ALIGN(16) =	922	static/* const*/ int32_t rounder3_128[] ATTR_ALIGN(16) =
924	rounder_sse2 (0.087788325588); /* C3(-C3/C4+C3+C5)/2 /	923	rounder_sse2 (0.087788325588); /* C3(-C3/C4+C3+C5)/2 /
925	static const int32_t rounder5_128[] ATTR_ALIGN(16) =	924	static/* const*/ int32_t rounder5_128[] ATTR_ALIGN(16) =
926	rounder_sse2 (-0.441341716183); /* C3(-C5/C4+C5-C3)/2 /	925	rounder_sse2 (-0.441341716183); /* C3(-C5/C4+C5-C3)/2 /
927		926
928	#if defined(__x86_64__)	927	#if defined(__x86_64__)