postprocess_template.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.10平台上编译为嵌入式图形界面操作系统。

/*
    Copyright (C) 2001-2002 Michael Niedermayer ([e-mail])

    This program 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
*/

/**
 * @file postprocess_template.c
 * mmx/mmx2/3dnow postprocess code.
 */


#undef PAVGB
#undef PMINUB
#undef PMAXUB

#ifdef HAVE_MMX2
#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
#elif defined (HAVE_3DNOW)
#define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
#endif

#ifdef HAVE_MMX2
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
#elif defined (HAVE_MMX)
#define PMINUB(b,a,t) \
	"movq " #a ", " #t " \n\t"\
	"psubusb " #b ", " #t " \n\t"\
	"psubb " #t ", " #a " \n\t"
#endif

#ifdef HAVE_MMX2
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
#elif defined (HAVE_MMX)
#define PMAXUB(a,b) \
	"psubusb " #a ", " #b " \n\t"\
	"paddb " #a ", " #b " \n\t"
#endif

//FIXME? |255-0| = 1 (shouldnt be a problem ...)
#ifdef HAVE_MMX
/**
 * Check if the middle 8x8 Block in the given 8x16 block is flat
 */
static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
	int numEq= 0, dcOk;
	src+= stride*4; // src points to begin of the 8x8 Block
asm volatile(
		"movq %0, %%mm7					\n\t" 
		"movq %1, %%mm6					\n\t" 
                : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
                );
                
asm volatile(
		"leal (%2, %3), %%eax				\n\t"
//	0	1	2	3	4	5	6	7	8	9
//	%1	eax	eax+%2	eax+2%2	%1+4%2	ecx	ecx+%2	ecx+2%2	%1+8%2	ecx+4%2

		"movq (%2), %%mm0				\n\t"
		"movq (%%eax), %%mm1				\n\t"
                "movq %%mm0, %%mm3				\n\t"
                "movq %%mm0, %%mm4				\n\t"
                PMAXUB(%%mm1, %%mm4)
                PMINUB(%%mm1, %%mm3, %%mm5)
		"psubb %%mm1, %%mm0				\n\t" // mm0 = differnece
		"paddb %%mm7, %%mm0				\n\t"
		"pcmpgtb %%mm6, %%mm0				\n\t"

		"movq (%%eax,%3), %%mm2				\n\t"
                PMAXUB(%%mm2, %%mm4)
                PMINUB(%%mm2, %%mm3, %%mm5)
		"psubb %%mm2, %%mm1				\n\t"
		"paddb %%mm7, %%mm1				\n\t"
		"pcmpgtb %%mm6, %%mm1				\n\t"
		"paddb %%mm1, %%mm0				\n\t"

		"movq (%%eax, %3, 2), %%mm1			\n\t"
                PMAXUB(%%mm1, %%mm4)
                PMINUB(%%mm1, %%mm3, %%mm5)
		"psubb %%mm1, %%mm2				\n\t"
		"paddb %%mm7, %%mm2				\n\t"
		"pcmpgtb %%mm6, %%mm2				\n\t"
		"paddb %%mm2, %%mm0				\n\t"
		
		"leal (%%eax, %3, 4), %%eax			\n\t"

		"movq (%2, %3, 4), %%mm2			\n\t"
                PMAXUB(%%mm2, %%mm4)
                PMINUB(%%mm2, %%mm3, %%mm5)
		"psubb %%mm2, %%mm1				\n\t"
		"paddb %%mm7, %%mm1				\n\t"
		"pcmpgtb %%mm6, %%mm1				\n\t"
		"paddb %%mm1, %%mm0				\n\t"

		"movq (%%eax), %%mm1				\n\t"
                PMAXUB(%%mm1, %%mm4)
                PMINUB(%%mm1, %%mm3, %%mm5)
		"psubb %%mm1, %%mm2				\n\t"
		"paddb %%mm7, %%mm2				\n\t"
		"pcmpgtb %%mm6, %%mm2				\n\t"
		"paddb %%mm2, %%mm0				\n\t"

		"movq (%%eax, %3), %%mm2			\n\t"
                PMAXUB(%%mm2, %%mm4)
                PMINUB(%%mm2, %%mm3, %%mm5)
		"psubb %%mm2, %%mm1				\n\t"
		"paddb %%mm7, %%mm1				\n\t"
		"pcmpgtb %%mm6, %%mm1				\n\t"
		"paddb %%mm1, %%mm0				\n\t"

		"movq (%%eax, %3, 2), %%mm1			\n\t"
                PMAXUB(%%mm1, %%mm4)
                PMINUB(%%mm1, %%mm3, %%mm5)
		"psubb %%mm1, %%mm2				\n\t"
		"paddb %%mm7, %%mm2				\n\t"
		"pcmpgtb %%mm6, %%mm2				\n\t"
		"paddb %%mm2, %%mm0				\n\t"
		"psubusb %%mm3, %%mm4				\n\t"

		"						\n\t"
#ifdef HAVE_MMX2
		"pxor %%mm7, %%mm7				\n\t"
		"psadbw %%mm7, %%mm0				\n\t"
#else
		"movq %%mm0, %%mm1				\n\t"
		"psrlw $8, %%mm0				\n\t"
		"paddb %%mm1, %%mm0				\n\t"
		"movq %%mm0, %%mm1				\n\t"
		"psrlq $16, %%mm0				\n\t"
		"paddb %%mm1, %%mm0				\n\t"
		"movq %%mm0, %%mm1				\n\t"
		"psrlq $32, %%mm0				\n\t"
		"paddb %%mm1, %%mm0				\n\t"
#endif
                "movq %4, %%mm7					\n\t" // QP,..., QP
		"paddusb %%mm7, %%mm7				\n\t" // 2QP ... 2QP
		"psubusb %%mm7, %%mm4				\n\t" // Diff <= 2QP -> 0
		"packssdw %%mm4, %%mm4				\n\t"
		"movd %%mm0, %0					\n\t"
		"movd %%mm4, %1					\n\t"

		: "=r" (numEq), "=r" (dcOk)
		: "r" (src), "r" (stride), "m" (c->pQPb)
		: "%eax"
		);

	numEq= (-numEq) &0xFF;
	if(numEq > c->ppMode.flatnessThreshold){
            if(dcOk) return 0;
            else     return 1;
        }else{
            return 2;
        }
}
#endif

/**
 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
 */
static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
{
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
	src+= stride*3;
	asm volatile(	//"movv %0 %1 %2\n\t"
		"movq %2, %%mm0			\n\t"  // QP,..., QP
		"pxor %%mm4, %%mm4				\n\t"

		"movq (%0), %%mm6				\n\t"
		"movq (%0, %1), %%mm5				\n\t"
		"movq %%mm5, %%mm1				\n\t"
		"movq %%mm6, %%mm2				\n\t"
		"psubusb %%mm6, %%mm5				\n\t"
		"psubusb %%mm1, %%mm2				\n\t"
		"por %%mm5, %%mm2				\n\t" // ABS Diff of lines
		"psubusb %%mm0, %%mm2				\n\t" // diff <= QP -> 0
		"pcmpeqb %%mm4, %%mm2			\n\t" // diff <= QP -> FF

		"pand %%mm2, %%mm6				\n\t"
		"pandn %%mm1, %%mm2				\n\t"
		"por %%mm2, %%mm6				\n\t"// First Line to Filter

		"movq (%0, %1, 8), %%mm5			\n\t"
		"leal (%0, %1, 4), %%eax			\n\t"
		"leal (%0, %1, 8), %%ecx			\n\t"
		"subl %1, %%ecx					\n\t"
		"addl %1, %0					\n\t" // %0 points to line 1 not 0
		"movq (%0, %1, 8), %%mm7			\n\t"
		"movq %%mm5, %%mm1				\n\t"
		"movq %%mm7, %%mm2				\n\t"
		"psubusb %%mm7, %%mm5				\n\t"
		"psubusb %%mm1, %%mm2				\n\t"
		"por %%mm5, %%mm2				\n\t" // ABS Diff of lines
		"psubusb %%mm0, %%mm2				\n\t" // diff <= QP -> 0
		"pcmpeqb %%mm4, %%mm2			\n\t" // diff <= QP -> FF

		"pand %%mm2, %%mm7				\n\t"
		"pandn %%mm1, %%mm2				\n\t"
		"por %%mm2, %%mm7				\n\t" // First Line to Filter


		// 	1	2	3	4	5	6	7	8
		//	%0	%0+%1	%0+2%1	eax	%0+4%1	eax+2%1	ecx	eax+4%1
		// 6 4 2 2 1 1
		// 6 4 4 2
		// 6 8 2

		"movq (%0, %1), %%mm0				\n\t" //  1
		"movq %%mm0, %%mm1				\n\t" //  1
		PAVGB(%%mm6, %%mm0)				      //1 1	/2
		PAVGB(%%mm6, %%mm0)				      //3 1	/4

		"movq (%0, %1, 4), %%mm2			\n\t" //     1
		"movq %%mm2, %%mm5				\n\t" //     1
		PAVGB((%%eax), %%mm2)				      //    11	/2
		PAVGB((%0, %1, 2), %%mm2)			      //   211	/4
		"movq %%mm2, %%mm3				\n\t" //   211	/4
		"movq (%0), %%mm4				\n\t" // 1
		PAVGB(%%mm4, %%mm3)				      // 4 211	/8
		PAVGB(%%mm0, %%mm3)				      //642211	/16
		"movq %%mm3, (%0)				\n\t" // X
		// mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
		"movq %%mm1, %%mm0				\n\t" //  1
		PAVGB(%%mm6, %%mm0)				      //1 1	/2
		"movq %%mm4, %%mm3				\n\t" // 1
		PAVGB((%0,%1,2), %%mm3)				      // 1 1	/2
		PAVGB((%%eax,%1,2), %%mm5)			      //     11	/2
		PAVGB((%%eax), %%mm5)				      //    211 /4
		PAVGB(%%mm5, %%mm3)				      // 2 2211 /8
		PAVGB(%%mm0, %%mm3)				      //4242211 /16
		"movq %%mm3, (%0,%1)				\n\t" //  X
		// mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
		PAVGB(%%mm4, %%mm6)				      //11	/2
		"movq (%%ecx), %%mm0				\n\t" //       1
		PAVGB((%%eax, %1, 2), %%mm0)			      //      11/2
		"movq %%mm0, %%mm3				\n\t" //      11/2
		PAVGB(%%mm1, %%mm0)				      //  2   11/4
		PAVGB(%%mm6, %%mm0)				      //222   11/8
		PAVGB(%%mm2, %%mm0)				      //22242211/16
		"movq (%0, %1, 2), %%mm2			\n\t" //   1
		"movq %%mm0, (%0, %1, 2)			\n\t" //   X
		// mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
		"movq (%%eax, %1, 4), %%mm0			\n\t" //        1
		PAVGB((%%ecx), %%mm0)				      //       11	/2
		PAVGB(%%mm0, %%mm6)				      //11     11	/4
		PAVGB(%%mm1, %%mm4)				      // 11		/2
		PAVGB(%%mm2, %%mm1)				      //  11		/2
		PAVGB(%%mm1, %%mm6)				      //1122   11	/8
		PAVGB(%%mm5, %%mm6)				      //112242211	/16
		"movq (%%eax), %%mm5				\n\t" //    1
		"movq %%mm6, (%%eax)				\n\t" //    X
		// mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
		"movq (%%eax, %1, 4), %%mm6			\n\t" //        1
		PAVGB(%%mm7, %%mm6)				      //        11	/2
		PAVGB(%%mm4, %%mm6)				      // 11     11	/4
		PAVGB(%%mm3, %%mm6)				      // 11   2211	/8
		PAVGB(%%mm5, %%mm2)				      //   11		/2
		"movq (%0, %1, 4), %%mm4			\n\t" //     1
		PAVGB(%%mm4, %%mm2)				      //   112		/4
		PAVGB(%%mm2, %%mm6)				      // 112242211	/16
		"movq %%mm6, (%0, %1, 4)			\n\t" //     X
		// mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
		PAVGB(%%mm7, %%mm1)				      //  11     2	/4
		PAVGB(%%mm4, %%mm5)				      //    11		/2
		PAVGB(%%mm5, %%mm0)				      //    11 11	/4
		"movq (%%eax, %1, 2), %%mm6			\n\t" //      1
		PAVGB(%%mm6, %%mm1)				      //  11  4  2	/8
		PAVGB(%%mm0, %%mm1)				      //  11224222	/16
		"movq %%mm1, (%%eax, %1, 2)			\n\t" //      X
		// mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
		PAVGB((%%ecx), %%mm2)				      //   112 4	/8
		"movq (%%eax, %1, 4), %%mm0			\n\t" //        1
		PAVGB(%%mm0, %%mm6)				      //      1 1	/2
		PAVGB(%%mm7, %%mm6)				      //      1 12	/4
		PAVGB(%%mm2, %%mm6)				      //   1122424	/4
		"movq %%mm6, (%%ecx)				\n\t" //       X
		// mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
		PAVGB(%%mm7, %%mm5)				      //    11   2	/4
		PAVGB(%%mm7, %%mm5)				      //    11   6	/8

		PAVGB(%%mm3, %%mm0)				      //      112	/4
		PAVGB(%%mm0, %%mm5)				      //    112246	/16
		"movq %%mm5, (%%eax, %1, 4)			\n\t" //        X
		"subl %1, %0					\n\t"

		:
		: "r" (src), "r" (stride), "m" (c->pQPb)
		: "%eax", "%ecx"
	);
#else
	const int l1= stride;
	const int l2= stride + l1;
	const int l3= stride + l2;
	const int l4= stride + l3;
	const int l5= stride + l4;
	const int l6= stride + l5;
	const int l7= stride + l6;
	const int l8= stride + l7;
	const int l9= stride + l8;
	int x;
	src+= stride*3;
	for(x=0; x<BLOCK_SIZE; x++)
	{
		const int first= ABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
		const int last= ABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];

		int sums[9];
		sums[0] = first + src[l1];
		sums[1] = src[l1] + src[l2];
		sums[2] = src[l2] + src[l3];
		sums[3] = src[l3] + src[l4];
		sums[4] = src[l4] + src[l5];
		sums[5] = src[l5] + src[l6];
		sums[6] = src[l6] + src[l7];
		sums[7] = src[l7] + src[l8];
		sums[8] = src[l8] + last;

		src[l1]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
		src[l2]= ((src[l2]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;
		src[l3]= ((src[l3]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;
		src[l4]= ((src[l4]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;
		src[l5]= ((src[l5]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;
		src[l6]= ((src[l6]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;
		src[l7]= (((last + src[l7])<<2) + ((src[l8] + sums[5])<<1) + sums[3] + 8)>>4;
		src[l8]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;

		src++;
	}
#endif
}

#if 0
/**
 * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
 * values are correctly clipped (MMX2)
 * values are wraparound (C)
 * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
	0 8 16 24
	x = 8
	x/2 = 4
	x/8 = 1
	1 12 12 23
 */
static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
{
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
	src+= stride*3;
// FIXME rounding
	asm volatile(
		"pxor %%mm7, %%mm7				\n\t" // 0
		"movq "MANGLE(b80)", %%mm6			\n\t" // MIN_SIGNED_BYTE
		"leal (%0, %1), %%eax				\n\t"
		"leal (%%eax, %1, 4), %%ecx			\n\t"
//	0	1	2	3	4	5	6	7	8	9
//	%0	eax	eax+%1	eax+2%1	%0+4%1	ecx	ecx+%1	ecx+2%1	%0+8%1	ecx+4%1
		"movq "MANGLE(pQPb)", %%mm0			\n\t" // QP,..., QP
		"movq %%mm0, %%mm1				\n\t" // QP,..., QP
		"paddusb "MANGLE(b02)", %%mm0			\n\t"
		"psrlw $2, %%mm0				\n\t"
		"pand "MANGLE(b3F)", %%mm0			\n\t" // QP/4,..., QP/4
		"paddusb %%mm1, %%mm0				\n\t" // QP*1.25 ...
		"movq (%0, %1, 4), %%mm2			\n\t" // line 4
		"movq (%%ecx), %%mm3				\n\t" // line 5
		"movq %%mm2, %%mm4				\n\t" // line 4
		"pcmpeqb %%mm5, %%mm5				\n\t" // -1
		"pxor %%mm2, %%mm5				\n\t" // -line 4 - 1
		PAVGB(%%mm3, %%mm5)
		"paddb %%mm6, %%mm5				\n\t" // (l5-l4)/2
		"psubusb %%mm3, %%mm4				\n\t"
		"psubusb %%mm2, %%mm3				\n\t"
		"por %%mm3, %%mm4				\n\t" // |l4 - l5|
		"psubusb %%mm0, %%mm4				\n\t"
		"pcmpeqb %%mm7, %%mm4				\n\t"
		"pand %%mm4, %%mm5				\n\t" // d/2

//		"paddb %%mm6, %%mm2				\n\t" // line 4 + 0x80
		"paddb %%mm5, %%mm2				\n\t"
//		"psubb %%mm6, %%mm2				\n\t"
		"movq %%mm2, (%0,%1, 4)				\n\t"

		"movq (%%ecx), %%mm2				\n\t"
//		"paddb %%mm6, %%mm2				\n\t" // line 5 + 0x80
		"psubb %%mm5, %%mm2				\n\t"
//		"psubb %%mm6, %%mm2				\n\t"
		"movq %%mm2, (%%ecx)				\n\t"

		"paddb %%mm6, %%mm5				\n\t"
		"psrlw $2, %%mm5				\n\t"
		"pand "MANGLE(b3F)", %%mm5			\n\t"
		"psubb "MANGLE(b20)", %%mm5			\n\t" // (l5-l4)/8

		"movq (%%eax, %1, 2), %%mm2			\n\t"
		"paddb %%mm6, %%mm2				\n\t" // line 3 + 0x80
		"paddsb %%mm5, %%mm2				\n\t"
		"psubb %%mm6, %%mm2				\n\t"
		"movq %%mm2, (%%eax, %1, 2)			\n\t"

		"movq (%%ecx, %1), %%mm2			\n\t"
		"paddb %%mm6, %%mm2				\n\t" // line 6 + 0x80
		"psubsb %%mm5, %%mm2				\n\t"
		"psubb %%mm6, %%mm2				\n\t"
		"movq %%mm2, (%%ecx, %1)			\n\t"

		:
		: "r" (src), "r" (stride)
		: "%eax", "%ecx"
	);
#else
 	const int l1= stride;