[Mplayer-cvslog] CVS: main/libmpeg2 vis.h, NONE, 1.1 motion_comp_vis.c, NONE, 1.1

Jindrich Makovicka CVS syncmail at mplayerhq.hu
Mon Aug 2 13:17:17 CEST 2004


CVS change done by Jindrich Makovicka CVS

Update of /cvsroot/mplayer/main/libmpeg2
In directory mail:/var2/tmp/cvs-serv14723

Added Files:
	vis.h motion_comp_vis.c 
Log Message:
Importing libmpeg2 from mpeg2dec-0.4.0b

--- NEW FILE ---
/*
 * vis.h
 * Copyright (C) 2003 David S. Miller <davem at redhat.com>
 *
 * 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
 */

/* You may be asking why I hard-code the instruction opcodes and don't
 * use the normal VIS assembler mnenomics for the VIS instructions.
 *
 * The reason is that Sun, in their infinite wisdom, decided that a binary
 * using a VIS instruction will cause it to be marked (in the ELF headers)
 * as doing so, and this prevents the OS from loading such binaries if the
 * current cpu doesn't have VIS.  There is no way to easily override this
 * behavior of the assembler that I am aware of.
 *
 * This totally defeats what libmpeg2 is trying to do which is allow a
 * single binary to be created, and then detect the availability of VIS
 * at runtime.
 *
 * I'm not saying that tainting the binary by default is bad, rather I'm
 * saying that not providing a way to override this easily unnecessarily
 * ties people's hands.
 *
 * Thus, we do the opcode encoding by hand and output 32-bit words in
 * the assembler to keep the binary from becoming tainted.
 */

#define vis_opc_base	((0x1 << 31) | (0x36 << 19))
#define vis_opf(X)	((X) << 5)
#define vis_sreg(X)	(X)
#define vis_dreg(X)	(((X)&0x1f)|((X)>>5))
#define vis_rs1_s(X)	(vis_sreg(X) << 14)
#define vis_rs1_d(X)	(vis_dreg(X) << 14)
#define vis_rs2_s(X)	(vis_sreg(X) << 0)
#define vis_rs2_d(X)	(vis_dreg(X) << 0)
#define vis_rd_s(X)	(vis_sreg(X) << 25)
#define vis_rd_d(X)	(vis_dreg(X) << 25)

#define vis_ss2s(opf,rs1,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_s(rs1) | \
                                       vis_rs2_s(rs2) | \
                                       vis_rd_s(rd)))

#define vis_dd2d(opf,rs1,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_d(rs1) | \
                                       vis_rs2_d(rs2) | \
                                       vis_rd_d(rd)))

#define vis_ss2d(opf,rs1,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_s(rs1) | \
                                       vis_rs2_s(rs2) | \
                                       vis_rd_d(rd)))

#define vis_sd2d(opf,rs1,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_s(rs1) | \
                                       vis_rs2_d(rs2) | \
                                       vis_rd_d(rd)))

#define vis_d2s(opf,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs2_d(rs2) | \
                                       vis_rd_s(rd)))

#define vis_s2d(opf,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs2_s(rs2) | \
                                       vis_rd_d(rd)))

#define vis_d12d(opf,rs1,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_d(rs1) | \
                                       vis_rd_d(rd)))

#define vis_d22d(opf,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs2_d(rs2) | \
                                       vis_rd_d(rd)))

#define vis_s12s(opf,rs1,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs1_s(rs1) | \
                                       vis_rd_s(rd)))

#define vis_s22s(opf,rs2,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rs2_s(rs2) | \
                                       vis_rd_s(rd)))

#define vis_s(opf,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rd_s(rd)))

#define vis_d(opf,rd) \
	__asm__ __volatile__ (".word %0" \
			      : : "i" (vis_opc_base | vis_opf(opf) | \
                                       vis_rd_d(rd)))

#define vis_r2m(op,rd,mem) \
	__asm__ __volatile__ (#op "\t%%f" #rd ", [%0]" : : "r" (&(mem)) )

#define vis_r2m_2(op,rd,mem1,mem2) \
	__asm__ __volatile__ (#op "\t%%f" #rd ", [%0 + %1]" : : "r" (mem1), "r" (mem2) )

#define vis_m2r(op,mem,rd) \
	__asm__ __volatile__ (#op "\t[%0], %%f" #rd : : "r" (&(mem)) )

#define vis_m2r_2(op,mem1,mem2,rd) \
	__asm__ __volatile__ (#op "\t[%0 + %1], %%f" #rd : : "r" (mem1), "r" (mem2) )

static inline void vis_set_gsr(unsigned int _val)
{
	register unsigned int val asm("g1");

	val = _val;
	__asm__ __volatile__(".word 0xa7804000"
			     : : "r" (val));
}

#define VIS_GSR_ALIGNADDR_MASK	0x0000007
#define VIS_GSR_ALIGNADDR_SHIFT	0
#define VIS_GSR_SCALEFACT_MASK	0x0000078
#define VIS_GSR_SCALEFACT_SHIFT	3

#define vis_ld32(mem,rs1)		vis_m2r(ld, mem, rs1)
#define vis_ld32_2(mem1,mem2,rs1)	vis_m2r_2(ld, mem1, mem2, rs1)
#define vis_st32(rs1,mem)		vis_r2m(st, rs1, mem)
#define vis_st32_2(rs1,mem1,mem2)	vis_r2m_2(st, rs1, mem1, mem2)
#define vis_ld64(mem,rs1)		vis_m2r(ldd, mem, rs1)
#define vis_ld64_2(mem1,mem2,rs1)	vis_m2r_2(ldd, mem1, mem2, rs1)
#define vis_st64(rs1,mem)		vis_r2m(std, rs1, mem)
#define vis_st64_2(rs1,mem1,mem2)	vis_r2m_2(std, rs1, mem1, mem2)

#define vis_ldblk(mem, rd) \
do {	register void *__mem asm("g1"); \
	__mem = &(mem); \
	__asm__ __volatile__(".word 0xc1985e00 | %1" \
			     : \
			     : "r" (__mem), \
			       "i" (vis_rd_d(rd)) \
			     : "memory"); \
} while (0)

#define vis_stblk(rd, mem) \
do {	register void *__mem asm("g1"); \
	__mem = &(mem); \
	__asm__ __volatile__(".word 0xc1b85e00 | %1" \
			     : \
			     : "r" (__mem), \
			       "i" (vis_rd_d(rd)) \
			     : "memory"); \
} while (0)

#define vis_membar_storestore()	\
	__asm__ __volatile__(".word 0x8143e008" : : : "memory")

#define vis_membar_sync()	\
	__asm__ __volatile__(".word 0x8143e040" : : : "memory")

/* 16 and 32 bit partitioned addition and subtraction.  The normal
 * versions perform 4 16-bit or 2 32-bit additions or subtractions.
 * The 's' versions perform 2 16-bit or 2 32-bit additions or
 * subtractions.
 */

#define vis_padd16(rs1,rs2,rd)		vis_dd2d(0x50, rs1, rs2, rd)
#define vis_padd16s(rs1,rs2,rd)		vis_ss2s(0x51, rs1, rs2, rd)
#define vis_padd32(rs1,rs2,rd)		vis_dd2d(0x52, rs1, rs2, rd)
#define vis_padd32s(rs1,rs2,rd)		vis_ss2s(0x53, rs1, rs2, rd)
#define vis_psub16(rs1,rs2,rd)		vis_dd2d(0x54, rs1, rs2, rd)
#define vis_psub16s(rs1,rs2,rd)		vis_ss2s(0x55, rs1, rs2, rd)
#define vis_psub32(rs1,rs2,rd)		vis_dd2d(0x56, rs1, rs2, rd)
#define vis_psub32s(rs1,rs2,rd)		vis_ss2s(0x57, rs1, rs2, rd)

/* Pixel formatting instructions.  */

#define vis_pack16(rs2,rd)		vis_d2s( 0x3b,      rs2, rd)
#define vis_pack32(rs1,rs2,rd)		vis_dd2d(0x3a, rs1, rs2, rd)
#define vis_packfix(rs2,rd)		vis_d2s( 0x3d,      rs2, rd)
#define vis_expand(rs2,rd)		vis_s2d( 0x4d,      rs2, rd)
#define vis_pmerge(rs1,rs2,rd)		vis_ss2d(0x4b, rs1, rs2, rd)

/* Partitioned multiply instructions.  */

#define vis_mul8x16(rs1,rs2,rd)		vis_sd2d(0x31, rs1, rs2, rd)
#define vis_mul8x16au(rs1,rs2,rd)	vis_ss2d(0x33, rs1, rs2, rd)
#define vis_mul8x16al(rs1,rs2,rd)	vis_ss2d(0x35, rs1, rs2, rd)
#define vis_mul8sux16(rs1,rs2,rd)	vis_dd2d(0x36, rs1, rs2, rd)
#define vis_mul8ulx16(rs1,rs2,rd)	vis_dd2d(0x37, rs1, rs2, rd)
#define vis_muld8sux16(rs1,rs2,rd)	vis_ss2d(0x38, rs1, rs2, rd)
#define vis_muld8ulx16(rs1,rs2,rd)	vis_ss2d(0x39, rs1, rs2, rd)

/* Alignment instructions.  */

static inline void *vis_alignaddr(void *_ptr)
{
	register void *ptr asm("g1");

	ptr = _ptr;

	__asm__ __volatile__(".word %2"
			     : "=&r" (ptr)
			     : "0" (ptr),
			       "i" (vis_opc_base | vis_opf(0x18) |
				    vis_rs1_s(1) |
				    vis_rs2_s(0) |
				    vis_rd_s(1)));

	return ptr;
}

static inline void vis_alignaddr_g0(void *_ptr)
{
	register void *ptr asm("g1");

	ptr = _ptr;

	__asm__ __volatile__(".word %2"
			     : "=&r" (ptr)
			     : "0" (ptr),
			       "i" (vis_opc_base | vis_opf(0x18) |
				    vis_rs1_s(1) |
				    vis_rs2_s(0) |
				    vis_rd_s(0)));
}

static inline void *vis_alignaddrl(void *_ptr)
{
	register void *ptr asm("g1");

	ptr = _ptr;

	__asm__ __volatile__(".word %2"
			     : "=&r" (ptr)
			     : "0" (ptr),
			       "i" (vis_opc_base | vis_opf(0x19) |
				    vis_rs1_s(1) |
				    vis_rs2_s(0) |
				    vis_rd_s(1)));

	return ptr;
}

static inline void vis_alignaddrl_g0(void *_ptr)
{
	register void *ptr asm("g1");

	ptr = _ptr;

	__asm__ __volatile__(".word %2"
			     : "=&r" (ptr)
			     : "0" (ptr),
			       "i" (vis_opc_base | vis_opf(0x19) |
				    vis_rs1_s(1) |
				    vis_rs2_s(0) |
				    vis_rd_s(0)));
}

#define vis_faligndata(rs1,rs2,rd)	vis_dd2d(0x48, rs1, rs2, rd)

/* Logical operate instructions.  */

#define vis_fzero(rd)			vis_d(   0x60,           rd)
#define vis_fzeros(rd)			vis_s(   0x61,           rd)
#define vis_fone(rd)			vis_d(   0x7e,           rd)
#define vis_fones(rd)			vis_s(   0x7f,           rd)
#define vis_src1(rs1,rd)		vis_d12d(0x74, rs1,      rd)
#define vis_src1s(rs1,rd)		vis_s12s(0x75, rs1,      rd)
#define vis_src2(rs2,rd)		vis_d22d(0x78,      rs2, rd)
#define vis_src2s(rs2,rd)		vis_s22s(0x79,      rs2, rd)
#define vis_not1(rs1,rd)		vis_d12d(0x6a, rs1,      rd)
#define vis_not1s(rs1,rd)		vis_s12s(0x6b, rs1,      rd)
#define vis_not2(rs2,rd)		vis_d22d(0x66,      rs2, rd)
#define vis_not2s(rs2,rd)		vis_s22s(0x67,      rs2, rd)
#define vis_or(rs1,rs2,rd)		vis_dd2d(0x7c, rs1, rs2, rd)
#define vis_ors(rs1,rs2,rd)		vis_ss2s(0x7d, rs1, rs2, rd)
#define vis_nor(rs1,rs2,rd)		vis_dd2d(0x62, rs1, rs2, rd)
#define vis_nors(rs1,rs2,rd)		vis_ss2s(0x63, rs1, rs2, rd)
#define vis_and(rs1,rs2,rd)		vis_dd2d(0x70, rs1, rs2, rd)
#define vis_ands(rs1,rs2,rd)		vis_ss2s(0x71, rs1, rs2, rd)
#define vis_nand(rs1,rs2,rd)		vis_dd2d(0x6e, rs1, rs2, rd)
#define vis_nands(rs1,rs2,rd)		vis_ss2s(0x6f, rs1, rs2, rd)
#define vis_xor(rs1,rs2,rd)		vis_dd2d(0x6c, rs1, rs2, rd)
#define vis_xors(rs1,rs2,rd)		vis_ss2s(0x6d, rs1, rs2, rd)
#define vis_xnor(rs1,rs2,rd)		vis_dd2d(0x72, rs1, rs2, rd)
#define vis_xnors(rs1,rs2,rd)		vis_ss2s(0x73, rs1, rs2, rd)
#define vis_ornot1(rs1,rs2,rd)		vis_dd2d(0x7a, rs1, rs2, rd)
#define vis_ornot1s(rs1,rs2,rd)		vis_ss2s(0x7b, rs1, rs2, rd)
#define vis_ornot2(rs1,rs2,rd)		vis_dd2d(0x76, rs1, rs2, rd)
#define vis_ornot2s(rs1,rs2,rd)		vis_ss2s(0x77, rs1, rs2, rd)
#define vis_andnot1(rs1,rs2,rd)		vis_dd2d(0x68, rs1, rs2, rd)
#define vis_andnot1s(rs1,rs2,rd)	vis_ss2s(0x69, rs1, rs2, rd)
#define vis_andnot2(rs1,rs2,rd)		vis_dd2d(0x64, rs1, rs2, rd)
#define vis_andnot2s(rs1,rs2,rd)	vis_ss2s(0x65, rs1, rs2, rd)

/* Pixel component distance.  */

#define vis_pdist(rs1,rs2,rd)		vis_dd2d(0x3e, rs1, rs2, rd)

--- NEW FILE ---
/*
 * motion_comp_vis.c
 * Copyright (C) 2003 David S. Miller <davem at redhat.com>
 *
 * 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 "config.h"

#ifdef ARCH_SPARC

#include <inttypes.h>

#include "mpeg2.h"
#include "attributes.h"
#include "mpeg2_internal.h"
#include "vis.h"

/* The trick used in some of this file is the formula from the MMX
 * motion comp code, which is:
 *
 * (x+y+1)>>1 == (x|y)-((x^y)>>1)
 *
 * This allows us to average 8 bytes at a time in a 64-bit FPU reg.
 * We avoid overflows by masking before we do the shift, and we
 * implement the shift by multiplying by 1/2 using mul8x16.  So in
 * VIS this is (assume 'x' is in f0, 'y' is in f2, a repeating mask
 * of '0xfe' is in f4, a repeating mask of '0x7f' is in f6, and
 * the value 0x80808080 is in f8):
 *
 *	fxor		f0, f2, f10
 *	fand		f10, f4, f10
 *	fmul8x16	f8, f10, f10
 *	fand		f10, f6, f10
 *	for		f0, f2, f12
 *	fpsub16		f12, f10, f10
 */

#define DUP4(x) {x, x, x, x}
#define DUP8(x) {x, x, x, x, x, x, x, x}
static const int16_t constants1[] ATTR_ALIGN(8) = DUP4 (1);
static const int16_t constants2[] ATTR_ALIGN(8) = DUP4 (2);
static const int16_t constants3[] ATTR_ALIGN(8) = DUP4 (3);
static const int16_t constants6[] ATTR_ALIGN(8) = DUP4 (6);
static const int8_t constants_fe[] ATTR_ALIGN(8) = DUP8 (0xfe);
static const int8_t constants_7f[] ATTR_ALIGN(8) = DUP8 (0x7f);
static const int8_t constants128[] ATTR_ALIGN(8) = DUP8 (128);
static const int16_t constants256_512[] ATTR_ALIGN(8) =
	{256, 512, 256, 512};
static const int16_t constants256_1024[] ATTR_ALIGN(8) =
	{256, 1024, 256, 1024};

#define REF_0		0
#define REF_0_1		1
#define REF_2		2
#define REF_2_1		3
#define REF_4		4
#define REF_4_1		5
#define REF_6		6
#define REF_6_1		7
#define REF_S0		8
#define REF_S0_1	9
#define REF_S2		10
#define REF_S2_1	11
#define REF_S4		12
#define REF_S4_1	13
#define REF_S6		14
#define REF_S6_1	15
#define DST_0		16
#define DST_1		17
#define DST_2		18
#define DST_3		19
#define CONST_1		20
#define CONST_2		20
#define CONST_3		20
#define CONST_6		20
#define MASK_fe		20
#define CONST_128	22
#define CONST_256	22
#define CONST_512	22
#define CONST_1024	22
#define TMP0		24
#define TMP1		25
#define TMP2		26
#define TMP3		27
#define TMP4		28
#define TMP5		29
#define ZERO		30
#define MASK_7f		30

#define TMP6		32
#define TMP8		34
#define TMP10		36
#define TMP12		38
#define TMP14		40
#define TMP16		42
#define TMP18		44
#define TMP20		46
#define TMP22		48
#define TMP24		50
#define TMP26		52
#define TMP28		54
#define TMP30		56
#define TMP32		58

static void MC_put_o_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 16 : 0;
	do {	/* 5 cycles */
		vis_ld64(ref[0], TMP0);

		vis_ld64_2(ref, 8, TMP2);

		vis_ld64_2(ref, offset, TMP4);
		ref += stride;

		vis_faligndata(TMP0, TMP2, REF_0);
		vis_st64(REF_0, dest[0]);

		vis_faligndata(TMP2, TMP4, REF_2);
		vis_st64_2(REF_2, dest, 8);
		dest += stride;
	} while (--height);
}

static void MC_put_o_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 8 : 0;
	do {	/* 4 cycles */
		vis_ld64(ref[0], TMP0);

		vis_ld64_2(ref, offset, TMP2);
		ref += stride;

		/* stall */

		vis_faligndata(TMP0, TMP2, REF_0);
		vis_st64(REF_0, dest[0]);
		dest += stride;
	} while (--height);
}


static void MC_avg_o_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int stride_8 = stride + 8;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 16 : 0;

	vis_ld64(ref[0], TMP0);

	vis_ld64(ref[8], TMP2);

	vis_ld64_2(ref, offset, TMP4);

	vis_ld64(dest[0], DST_0);

	vis_ld64(dest[8], DST_2);

	vis_ld64(constants_fe[0], MASK_fe);
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_ld64(constants_7f[0], MASK_7f);
	vis_faligndata(TMP2, TMP4, REF_2);

	vis_ld64(constants128[0], CONST_128);

	ref += stride;
	height = (height >> 1) - 1;

	do {	/* 24 cycles */
		vis_ld64(ref[0], TMP0);
		vis_xor(DST_0, REF_0, TMP6);

		vis_ld64_2(ref, 8, TMP2);
		vis_and(TMP6, MASK_fe, TMP6);

		vis_ld64_2(ref, offset, TMP4);
		ref += stride;
		vis_mul8x16(CONST_128, TMP6, TMP6);
		vis_xor(DST_2, REF_2, TMP8);

		vis_and(TMP8, MASK_fe, TMP8);

		vis_or(DST_0, REF_0, TMP10);
		vis_ld64_2(dest, stride, DST_0);
		vis_mul8x16(CONST_128, TMP8, TMP8);

		vis_or(DST_2, REF_2, TMP12);
		vis_ld64_2(dest, stride_8, DST_2);

		vis_ld64(ref[0], TMP14);
		vis_and(TMP6, MASK_7f, TMP6);

		vis_and(TMP8, MASK_7f, TMP8);

		vis_psub16(TMP10, TMP6, TMP6);
		vis_st64(TMP6, dest[0]);

		vis_psub16(TMP12, TMP8, TMP8);
		vis_st64_2(TMP8, dest, 8);

		dest += stride;
		vis_ld64_2(ref, 8, TMP16);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, offset, TMP18);
		vis_faligndata(TMP2, TMP4, REF_2);
		ref += stride;

		vis_xor(DST_0, REF_0, TMP20);

		vis_and(TMP20, MASK_fe, TMP20);

		vis_xor(DST_2, REF_2, TMP22);
		vis_mul8x16(CONST_128, TMP20, TMP20);

		vis_and(TMP22, MASK_fe, TMP22);

		vis_or(DST_0, REF_0, TMP24);
		vis_mul8x16(CONST_128, TMP22, TMP22);

		vis_or(DST_2, REF_2, TMP26);

		vis_ld64_2(dest, stride, DST_0);
		vis_faligndata(TMP14, TMP16, REF_0);

		vis_ld64_2(dest, stride_8, DST_2);
		vis_faligndata(TMP16, TMP18, REF_2);

		vis_and(TMP20, MASK_7f, TMP20);

		vis_and(TMP22, MASK_7f, TMP22);

		vis_psub16(TMP24, TMP20, TMP20);
		vis_st64(TMP20, dest[0]);

		vis_psub16(TMP26, TMP22, TMP22);
		vis_st64_2(TMP22, dest, 8);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0], TMP0);
	vis_xor(DST_0, REF_0, TMP6);

	vis_ld64_2(ref, 8, TMP2);
	vis_and(TMP6, MASK_fe, TMP6);

	vis_ld64_2(ref, offset, TMP4);
	vis_mul8x16(CONST_128, TMP6, TMP6);
	vis_xor(DST_2, REF_2, TMP8);

	vis_and(TMP8, MASK_fe, TMP8);

	vis_or(DST_0, REF_0, TMP10);
	vis_ld64_2(dest, stride, DST_0);
	vis_mul8x16(CONST_128, TMP8, TMP8);

	vis_or(DST_2, REF_2, TMP12);
	vis_ld64_2(dest, stride_8, DST_2);

	vis_ld64(ref[0], TMP14);
	vis_and(TMP6, MASK_7f, TMP6);

	vis_and(TMP8, MASK_7f, TMP8);

	vis_psub16(TMP10, TMP6, TMP6);
	vis_st64(TMP6, dest[0]);

	vis_psub16(TMP12, TMP8, TMP8);
	vis_st64_2(TMP8, dest, 8);

	dest += stride;
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_faligndata(TMP2, TMP4, REF_2);

	vis_xor(DST_0, REF_0, TMP20);

	vis_and(TMP20, MASK_fe, TMP20);

	vis_xor(DST_2, REF_2, TMP22);
	vis_mul8x16(CONST_128, TMP20, TMP20);

	vis_and(TMP22, MASK_fe, TMP22);

	vis_or(DST_0, REF_0, TMP24);
	vis_mul8x16(CONST_128, TMP22, TMP22);

	vis_or(DST_2, REF_2, TMP26);

	vis_and(TMP20, MASK_7f, TMP20);

	vis_and(TMP22, MASK_7f, TMP22);

	vis_psub16(TMP24, TMP20, TMP20);
	vis_st64(TMP20, dest[0]);

	vis_psub16(TMP26, TMP22, TMP22);
	vis_st64_2(TMP22, dest, 8);
}

static void MC_avg_o_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 8 : 0;

	vis_ld64(ref[0], TMP0);

	vis_ld64_2(ref, offset, TMP2);

	vis_ld64(dest[0], DST_0);

	vis_ld64(constants_fe[0], MASK_fe);

	vis_ld64(constants_7f[0], MASK_7f);
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_ld64(constants128[0], CONST_128);

	ref += stride;
	height = (height >> 1) - 1;

	do {	/* 12 cycles */
		vis_ld64(ref[0], TMP0);
		vis_xor(DST_0, REF_0, TMP4);

		vis_ld64_2(ref, offset, TMP2);
		vis_and(TMP4, MASK_fe, TMP4);

		vis_or(DST_0, REF_0, TMP6);
		vis_ld64_2(dest, stride, DST_0);
		ref += stride;
		vis_mul8x16(CONST_128, TMP4, TMP4);

		vis_ld64(ref[0], TMP12);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, offset, TMP2);
		vis_xor(DST_0, REF_0, TMP0);
		ref += stride;

		vis_and(TMP0, MASK_fe, TMP0);

		vis_and(TMP4, MASK_7f, TMP4);

		vis_psub16(TMP6, TMP4, TMP4);
		vis_st64(TMP4, dest[0]);
		dest += stride;
		vis_mul8x16(CONST_128, TMP0, TMP0);

		vis_or(DST_0, REF_0, TMP6);
		vis_ld64_2(dest, stride, DST_0);

		vis_faligndata(TMP12, TMP2, REF_0);

		vis_and(TMP0, MASK_7f, TMP0);

		vis_psub16(TMP6, TMP0, TMP4);
		vis_st64(TMP4, dest[0]);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0], TMP0);
	vis_xor(DST_0, REF_0, TMP4);

	vis_ld64_2(ref, offset, TMP2);
	vis_and(TMP4, MASK_fe, TMP4);

	vis_or(DST_0, REF_0, TMP6);
	vis_ld64_2(dest, stride, DST_0);
	vis_mul8x16(CONST_128, TMP4, TMP4);

	vis_faligndata(TMP0, TMP2, REF_0);

	vis_xor(DST_0, REF_0, TMP0);

	vis_and(TMP0, MASK_fe, TMP0);

	vis_and(TMP4, MASK_7f, TMP4);

	vis_psub16(TMP6, TMP4, TMP4);
	vis_st64(TMP4, dest[0]);
	dest += stride;
	vis_mul8x16(CONST_128, TMP0, TMP0);

	vis_or(DST_0, REF_0, TMP6);

	vis_and(TMP0, MASK_7f, TMP0);

	vis_psub16(TMP6, TMP0, TMP4);
	vis_st64(TMP4, dest[0]);
}

static void MC_put_x_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;

	ref = vis_alignaddr(ref);

	vis_ld64(ref[0],    TMP0);

	vis_ld64_2(ref, 8,  TMP2);

	vis_ld64_2(ref, 16, TMP4);

	vis_ld64(constants_fe[0], MASK_fe);

	vis_ld64(constants_7f[0], MASK_7f);
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_ld64(constants128[0], CONST_128);
	vis_faligndata(TMP2, TMP4, REF_4);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_2);
		vis_faligndata(TMP2, TMP4, REF_6);
	} else {
		vis_src1(TMP2, REF_2);
		vis_src1(TMP4, REF_6);
	}

	ref += stride;
	height = (height >> 1) - 1;

	do {	/* 34 cycles */
		vis_ld64(ref[0],    TMP0);
		vis_xor(REF_0, REF_2, TMP6);

		vis_ld64_2(ref, 8,  TMP2);
		vis_xor(REF_4, REF_6, TMP8);

		vis_ld64_2(ref, 16, TMP4);
		vis_and(TMP6, MASK_fe, TMP6);
		ref += stride;

		vis_ld64(ref[0],    TMP14);
		vis_mul8x16(CONST_128, TMP6, TMP6);
		vis_and(TMP8, MASK_fe, TMP8);

		vis_ld64_2(ref, 8,  TMP16);
		vis_mul8x16(CONST_128, TMP8, TMP8);
		vis_or(REF_0, REF_2, TMP10);

		vis_ld64_2(ref, 16, TMP18);
		ref += stride;
		vis_or(REF_4, REF_6, TMP12);

		vis_alignaddr_g0((void *)off);

		vis_faligndata(TMP0, TMP2, REF_0);

		vis_faligndata(TMP2, TMP4, REF_4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_2);
			vis_faligndata(TMP2, TMP4, REF_6);
		} else {
			vis_src1(TMP2, REF_2);
			vis_src1(TMP4, REF_6);
		}

		vis_and(TMP6, MASK_7f, TMP6);

		vis_and(TMP8, MASK_7f, TMP8);

		vis_psub16(TMP10, TMP6, TMP6);
		vis_st64(TMP6, dest[0]);

		vis_psub16(TMP12, TMP8, TMP8);
		vis_st64_2(TMP8, dest, 8);
		dest += stride;

		vis_xor(REF_0, REF_2, TMP6);

		vis_xor(REF_4, REF_6, TMP8);

		vis_and(TMP6, MASK_fe, TMP6);

		vis_mul8x16(CONST_128, TMP6, TMP6);
		vis_and(TMP8, MASK_fe, TMP8);

		vis_mul8x16(CONST_128, TMP8, TMP8);
		vis_or(REF_0, REF_2, TMP10);

		vis_or(REF_4, REF_6, TMP12);

		vis_alignaddr_g0((void *)off);

		vis_faligndata(TMP14, TMP16, REF_0);

		vis_faligndata(TMP16, TMP18, REF_4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP14, TMP16, REF_2);
			vis_faligndata(TMP16, TMP18, REF_6);
		} else {
			vis_src1(TMP16, REF_2);
			vis_src1(TMP18, REF_6);
		}

		vis_and(TMP6, MASK_7f, TMP6);

		vis_and(TMP8, MASK_7f, TMP8);

		vis_psub16(TMP10, TMP6, TMP6);
		vis_st64(TMP6, dest[0]);

		vis_psub16(TMP12, TMP8, TMP8);
		vis_st64_2(TMP8, dest, 8);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0],    TMP0);
	vis_xor(REF_0, REF_2, TMP6);

	vis_ld64_2(ref, 8,  TMP2);
	vis_xor(REF_4, REF_6, TMP8);

	vis_ld64_2(ref, 16, TMP4);
	vis_and(TMP6, MASK_fe, TMP6);

	vis_mul8x16(CONST_128, TMP6, TMP6);
	vis_and(TMP8, MASK_fe, TMP8);

	vis_mul8x16(CONST_128, TMP8, TMP8);
	vis_or(REF_0, REF_2, TMP10);

	vis_or(REF_4, REF_6, TMP12);

	vis_alignaddr_g0((void *)off);

	vis_faligndata(TMP0, TMP2, REF_0);

	vis_faligndata(TMP2, TMP4, REF_4);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_2);
		vis_faligndata(TMP2, TMP4, REF_6);
	} else {
		vis_src1(TMP2, REF_2);
		vis_src1(TMP4, REF_6);
	}

	vis_and(TMP6, MASK_7f, TMP6);

	vis_and(TMP8, MASK_7f, TMP8);

	vis_psub16(TMP10, TMP6, TMP6);
	vis_st64(TMP6, dest[0]);

	vis_psub16(TMP12, TMP8, TMP8);
	vis_st64_2(TMP8, dest, 8);
	dest += stride;

	vis_xor(REF_0, REF_2, TMP6);

	vis_xor(REF_4, REF_6, TMP8);

	vis_and(TMP6, MASK_fe, TMP6);

	vis_mul8x16(CONST_128, TMP6, TMP6);
	vis_and(TMP8, MASK_fe, TMP8);

	vis_mul8x16(CONST_128, TMP8, TMP8);
	vis_or(REF_0, REF_2, TMP10);

	vis_or(REF_4, REF_6, TMP12);

	vis_and(TMP6, MASK_7f, TMP6);

	vis_and(TMP8, MASK_7f, TMP8);

	vis_psub16(TMP10, TMP6, TMP6);
	vis_st64(TMP6, dest[0]);

	vis_psub16(TMP12, TMP8, TMP8);
	vis_st64_2(TMP8, dest, 8);
}

static void MC_put_x_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;

	ref = vis_alignaddr(ref);

	vis_ld64(ref[0], TMP0);

	vis_ld64(ref[8], TMP2);

	vis_ld64(constants_fe[0], MASK_fe);

	vis_ld64(constants_7f[0], MASK_7f);

	vis_ld64(constants128[0], CONST_128);
	vis_faligndata(TMP0, TMP2, REF_0);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_2);
	} else {
		vis_src1(TMP2, REF_2);
	}

	ref += stride;
	height = (height >> 1) - 1;

	do {	/* 20 cycles */
		vis_ld64(ref[0], TMP0);
		vis_xor(REF_0, REF_2, TMP4);

		vis_ld64_2(ref, 8, TMP2);
		vis_and(TMP4, MASK_fe, TMP4);
		ref += stride;

		vis_ld64(ref[0], TMP8);
		vis_or(REF_0, REF_2, TMP6);
		vis_mul8x16(CONST_128, TMP4, TMP4);

		vis_alignaddr_g0((void *)off);

		vis_ld64_2(ref, 8, TMP10);
		ref += stride;
		vis_faligndata(TMP0, TMP2, REF_0);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_2);
		} else {
			vis_src1(TMP2, REF_2);
		}

		vis_and(TMP4, MASK_7f, TMP4);

		vis_psub16(TMP6, TMP4, DST_0);
		vis_st64(DST_0, dest[0]);
		dest += stride;

		vis_xor(REF_0, REF_2, TMP12);

		vis_and(TMP12, MASK_fe, TMP12);

		vis_or(REF_0, REF_2, TMP14);
		vis_mul8x16(CONST_128, TMP12, TMP12);

		vis_alignaddr_g0((void *)off);
		vis_faligndata(TMP8, TMP10, REF_0);
		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP8, TMP10, REF_2);
		} else {
			vis_src1(TMP10, REF_2);
		}

		vis_and(TMP12, MASK_7f, TMP12);

		vis_psub16(TMP14, TMP12, DST_0);
		vis_st64(DST_0, dest[0]);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0], TMP0);
	vis_xor(REF_0, REF_2, TMP4);

	vis_ld64_2(ref, 8, TMP2);
	vis_and(TMP4, MASK_fe, TMP4);

	vis_or(REF_0, REF_2, TMP6);
	vis_mul8x16(CONST_128, TMP4, TMP4);

	vis_alignaddr_g0((void *)off);

	vis_faligndata(TMP0, TMP2, REF_0);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_2);
	} else {
		vis_src1(TMP2, REF_2);
	}

	vis_and(TMP4, MASK_7f, TMP4);

	vis_psub16(TMP6, TMP4, DST_0);
	vis_st64(DST_0, dest[0]);
	dest += stride;

	vis_xor(REF_0, REF_2, TMP12);

	vis_and(TMP12, MASK_fe, TMP12);

	vis_or(REF_0, REF_2, TMP14);
	vis_mul8x16(CONST_128, TMP12, TMP12);

	vis_and(TMP12, MASK_7f, TMP12);

	vis_psub16(TMP14, TMP12, DST_0);
	vis_st64(DST_0, dest[0]);
	dest += stride;
}

static void MC_avg_x_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	vis_ld64(constants3[0], CONST_3);
	vis_fzero(ZERO);
	vis_ld64(constants256_512[0], CONST_256);

	ref = vis_alignaddr(ref);
	do {	/* 26 cycles */
		vis_ld64(ref[0], TMP0);

		vis_ld64(ref[8], TMP2);

		vis_alignaddr_g0((void *)off);

		vis_ld64(ref[16], TMP4);

		vis_ld64(dest[0], DST_0);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64(dest[8], DST_2);
		vis_faligndata(TMP2, TMP4, REF_4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_2);
			vis_faligndata(TMP2, TMP4, REF_6);
		} else {
			vis_src1(TMP2, REF_2);
			vis_src1(TMP4, REF_6);
		}

		vis_mul8x16au(REF_0,   CONST_256, TMP0);

		vis_pmerge(ZERO,     REF_2,     TMP4);
		vis_mul8x16au(REF_0_1, CONST_256, TMP2);

		vis_pmerge(ZERO, REF_2_1, TMP6);

		vis_padd16(TMP0, TMP4, TMP0);

		vis_mul8x16al(DST_0,   CONST_512, TMP4);
		vis_padd16(TMP2, TMP6, TMP2);

		vis_mul8x16al(DST_1,   CONST_512, TMP6);

		vis_mul8x16au(REF_6,   CONST_256, TMP12);

		vis_padd16(TMP0, TMP4, TMP0);
		vis_mul8x16au(REF_6_1, CONST_256, TMP14);

		vis_padd16(TMP2, TMP6, TMP2);
		vis_mul8x16au(REF_4,   CONST_256, TMP16);

		vis_padd16(TMP0, CONST_3, TMP8);
		vis_mul8x16au(REF_4_1, CONST_256, TMP18);

		vis_padd16(TMP2, CONST_3, TMP10);
		vis_pack16(TMP8, DST_0);

		vis_pack16(TMP10, DST_1);
		vis_padd16(TMP16, TMP12, TMP0);

		vis_st64(DST_0, dest[0]);
		vis_mul8x16al(DST_2,   CONST_512, TMP4);
		vis_padd16(TMP18, TMP14, TMP2);

		vis_mul8x16al(DST_3,   CONST_512, TMP6);
		vis_padd16(TMP0, CONST_3, TMP0);

		vis_padd16(TMP2, CONST_3, TMP2);

		vis_padd16(TMP0, TMP4, TMP0);

		vis_padd16(TMP2, TMP6, TMP2);
		vis_pack16(TMP0, DST_2);

		vis_pack16(TMP2, DST_3);
		vis_st64(DST_2, dest[8]);

		ref += stride;
		dest += stride;
	} while (--height);
}

static void MC_avg_x_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;
	int stride_times_2 = stride << 1;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	vis_ld64(constants3[0], CONST_3);
	vis_fzero(ZERO);
	vis_ld64(constants256_512[0], CONST_256);

	ref = vis_alignaddr(ref);
	height >>= 2;
	do {	/* 47 cycles */
		vis_ld64(ref[0],   TMP0);

		vis_ld64_2(ref, 8, TMP2);
		ref += stride;

		vis_alignaddr_g0((void *)off);

		vis_ld64(ref[0],   TMP4);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, 8, TMP6);
		ref += stride;

		vis_ld64(ref[0],   TMP8);

		vis_ld64_2(ref, 8, TMP10);
		ref += stride;
		vis_faligndata(TMP4, TMP6, REF_4);

		vis_ld64(ref[0],   TMP12);

		vis_ld64_2(ref, 8, TMP14);
		ref += stride;
		vis_faligndata(TMP8, TMP10, REF_S0);

		vis_faligndata(TMP12, TMP14, REF_S4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);

			vis_ld64(dest[0], DST_0);
			vis_faligndata(TMP0, TMP2, REF_2);

			vis_ld64_2(dest, stride, DST_2);
			vis_faligndata(TMP4, TMP6, REF_6);

			vis_faligndata(TMP8, TMP10, REF_S2);

			vis_faligndata(TMP12, TMP14, REF_S6);
		} else {
			vis_ld64(dest[0], DST_0);
			vis_src1(TMP2, REF_2);

			vis_ld64_2(dest, stride, DST_2);
			vis_src1(TMP6, REF_6);

			vis_src1(TMP10, REF_S2);

			vis_src1(TMP14, REF_S6);
		}

		vis_pmerge(ZERO,     REF_0,     TMP0);
		vis_mul8x16au(REF_0_1, CONST_256, TMP2);

		vis_pmerge(ZERO,     REF_2,     TMP4);
		vis_mul8x16au(REF_2_1, CONST_256, TMP6);

		vis_padd16(TMP0, CONST_3, TMP0);
		vis_mul8x16al(DST_0,   CONST_512, TMP16);

		vis_padd16(TMP2, CONST_3, TMP2);
		vis_mul8x16al(DST_1,   CONST_512, TMP18);

		vis_padd16(TMP0, TMP4, TMP0);
		vis_mul8x16au(REF_4, CONST_256, TMP8);

		vis_padd16(TMP2, TMP6, TMP2);
		vis_mul8x16au(REF_4_1, CONST_256, TMP10);

		vis_padd16(TMP0, TMP16, TMP0);
		vis_mul8x16au(REF_6, CONST_256, TMP12);

		vis_padd16(TMP2, TMP18, TMP2);
		vis_mul8x16au(REF_6_1, CONST_256, TMP14);

		vis_padd16(TMP8, CONST_3, TMP8);
		vis_mul8x16al(DST_2, CONST_512, TMP16);

		vis_padd16(TMP8, TMP12, TMP8);
		vis_mul8x16al(DST_3, CONST_512, TMP18);

		vis_padd16(TMP10, TMP14, TMP10);
		vis_pack16(TMP0, DST_0);

		vis_pack16(TMP2, DST_1);
		vis_st64(DST_0, dest[0]);
		dest += stride;
		vis_padd16(TMP10, CONST_3, TMP10);

		vis_ld64_2(dest, stride, DST_0);
		vis_padd16(TMP8, TMP16, TMP8);

		vis_ld64_2(dest, stride_times_2, TMP4/*DST_2*/);
		vis_padd16(TMP10, TMP18, TMP10);
		vis_pack16(TMP8, DST_2);

		vis_pack16(TMP10, DST_3);
		vis_st64(DST_2, dest[0]);
		dest += stride;

		vis_mul8x16au(REF_S0_1, CONST_256, TMP2);
		vis_pmerge(ZERO,     REF_S0,     TMP0);

		vis_pmerge(ZERO,     REF_S2,     TMP24);
		vis_mul8x16au(REF_S2_1, CONST_256, TMP6);

		vis_padd16(TMP0, CONST_3, TMP0);
		vis_mul8x16au(REF_S4, CONST_256, TMP8);

		vis_padd16(TMP2, CONST_3, TMP2);
		vis_mul8x16au(REF_S4_1, CONST_256, TMP10);

		vis_padd16(TMP0, TMP24, TMP0);
		vis_mul8x16au(REF_S6, CONST_256, TMP12);

		vis_padd16(TMP2, TMP6, TMP2);
		vis_mul8x16au(REF_S6_1, CONST_256, TMP14);

		vis_padd16(TMP8, CONST_3, TMP8);
		vis_mul8x16al(DST_0,   CONST_512, TMP16);

		vis_padd16(TMP10, CONST_3, TMP10);
		vis_mul8x16al(DST_1,   CONST_512, TMP18);

		vis_padd16(TMP8, TMP12, TMP8);
		vis_mul8x16al(TMP4/*DST_2*/, CONST_512, TMP20);

		vis_mul8x16al(TMP5/*DST_3*/, CONST_512, TMP22);
		vis_padd16(TMP0, TMP16, TMP0);

		vis_padd16(TMP2, TMP18, TMP2);
		vis_pack16(TMP0, DST_0);

		vis_padd16(TMP10, TMP14, TMP10);
		vis_pack16(TMP2, DST_1);
		vis_st64(DST_0, dest[0]);
		dest += stride;

		vis_padd16(TMP8, TMP20, TMP8);

		vis_padd16(TMP10, TMP22, TMP10);
		vis_pack16(TMP8, DST_2);

		vis_pack16(TMP10, DST_3);
		vis_st64(DST_2, dest[0]);
		dest += stride;
	} while (--height);
}

static void MC_put_y_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 16 : 0;

	vis_ld64(ref[0], TMP0);

	vis_ld64_2(ref, 8, TMP2);

	vis_ld64_2(ref, offset, TMP4);
	ref += stride;

	vis_ld64(ref[0], TMP6);
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_ld64_2(ref, 8, TMP8);
	vis_faligndata(TMP2, TMP4, REF_4);

	vis_ld64_2(ref, offset, TMP10);
	ref += stride;

	vis_ld64(constants_fe[0], MASK_fe);
	vis_faligndata(TMP6, TMP8, REF_2);

	vis_ld64(constants_7f[0], MASK_7f);
	vis_faligndata(TMP8, TMP10, REF_6);

	vis_ld64(constants128[0], CONST_128);
	height = (height >> 1) - 1;
	do {	/* 24 cycles */
		vis_ld64(ref[0], TMP0);
		vis_xor(REF_0, REF_2, TMP12);

		vis_ld64_2(ref, 8, TMP2);
		vis_xor(REF_4, REF_6, TMP16);

		vis_ld64_2(ref, offset, TMP4);
		ref += stride;
		vis_or(REF_0, REF_2, TMP14);

		vis_ld64(ref[0], TMP6);
		vis_or(REF_4, REF_6, TMP18);

		vis_ld64_2(ref, 8, TMP8);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, offset, TMP10);
		ref += stride;
		vis_faligndata(TMP2, TMP4, REF_4);

		vis_and(TMP12, MASK_fe, TMP12);

		vis_and(TMP16, MASK_fe, TMP16);
		vis_mul8x16(CONST_128, TMP12, TMP12);

		vis_mul8x16(CONST_128, TMP16, TMP16);
		vis_xor(REF_0, REF_2, TMP0);

		vis_xor(REF_4, REF_6, TMP2);

		vis_or(REF_0, REF_2, TMP20);

		vis_and(TMP12, MASK_7f, TMP12);

		vis_and(TMP16, MASK_7f, TMP16);

		vis_psub16(TMP14, TMP12, TMP12);
		vis_st64(TMP12, dest[0]);

		vis_psub16(TMP18, TMP16, TMP16);
		vis_st64_2(TMP16, dest, 8);
		dest += stride;

		vis_or(REF_4, REF_6, TMP18);

		vis_and(TMP0, MASK_fe, TMP0);

		vis_and(TMP2, MASK_fe, TMP2);
		vis_mul8x16(CONST_128, TMP0, TMP0);

		vis_faligndata(TMP6, TMP8, REF_2);
		vis_mul8x16(CONST_128, TMP2, TMP2);

		vis_faligndata(TMP8, TMP10, REF_6);

		vis_and(TMP0, MASK_7f, TMP0);

		vis_and(TMP2, MASK_7f, TMP2);

		vis_psub16(TMP20, TMP0, TMP0);
		vis_st64(TMP0, dest[0]);

		vis_psub16(TMP18, TMP2, TMP2);
		vis_st64_2(TMP2, dest, 8);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0], TMP0);
	vis_xor(REF_0, REF_2, TMP12);

	vis_ld64_2(ref, 8, TMP2);
	vis_xor(REF_4, REF_6, TMP16);

	vis_ld64_2(ref, offset, TMP4);
	vis_or(REF_0, REF_2, TMP14);

	vis_or(REF_4, REF_6, TMP18);

	vis_faligndata(TMP0, TMP2, REF_0);

	vis_faligndata(TMP2, TMP4, REF_4);

	vis_and(TMP12, MASK_fe, TMP12);

	vis_and(TMP16, MASK_fe, TMP16);
	vis_mul8x16(CONST_128, TMP12, TMP12);

	vis_mul8x16(CONST_128, TMP16, TMP16);
	vis_xor(REF_0, REF_2, TMP0);

	vis_xor(REF_4, REF_6, TMP2);

	vis_or(REF_0, REF_2, TMP20);

	vis_and(TMP12, MASK_7f, TMP12);

	vis_and(TMP16, MASK_7f, TMP16);

	vis_psub16(TMP14, TMP12, TMP12);
	vis_st64(TMP12, dest[0]);

	vis_psub16(TMP18, TMP16, TMP16);
	vis_st64_2(TMP16, dest, 8);
	dest += stride;

	vis_or(REF_4, REF_6, TMP18);

	vis_and(TMP0, MASK_fe, TMP0);

	vis_and(TMP2, MASK_fe, TMP2);
	vis_mul8x16(CONST_128, TMP0, TMP0);

	vis_mul8x16(CONST_128, TMP2, TMP2);

	vis_and(TMP0, MASK_7f, TMP0);

	vis_and(TMP2, MASK_7f, TMP2);

	vis_psub16(TMP20, TMP0, TMP0);
	vis_st64(TMP0, dest[0]);

	vis_psub16(TMP18, TMP2, TMP2);
	vis_st64_2(TMP2, dest, 8);
}

static void MC_put_y_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int offset;

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 8 : 0;

	vis_ld64(ref[0], TMP0);

	vis_ld64_2(ref, offset, TMP2);
	ref += stride;

	vis_ld64(ref[0], TMP4);

	vis_ld64_2(ref, offset, TMP6);
	ref += stride;

	vis_ld64(constants_fe[0], MASK_fe);
	vis_faligndata(TMP0, TMP2, REF_0);

	vis_ld64(constants_7f[0], MASK_7f);
	vis_faligndata(TMP4, TMP6, REF_2);

	vis_ld64(constants128[0], CONST_128);
	height = (height >> 1) - 1;
	do {	/* 12 cycles */
		vis_ld64(ref[0], TMP0);
		vis_xor(REF_0, REF_2, TMP4);

		vis_ld64_2(ref, offset, TMP2);
		ref += stride;
		vis_and(TMP4, MASK_fe, TMP4);

		vis_or(REF_0, REF_2, TMP6);
		vis_mul8x16(CONST_128, TMP4, TMP4);

		vis_faligndata(TMP0, TMP2, REF_0);
		vis_ld64(ref[0], TMP0);

		vis_ld64_2(ref, offset, TMP2);
		ref += stride;
		vis_xor(REF_0, REF_2, TMP12);

		vis_and(TMP4, MASK_7f, TMP4);

		vis_and(TMP12, MASK_fe, TMP12);

		vis_mul8x16(CONST_128, TMP12, TMP12);
		vis_or(REF_0, REF_2, TMP14);

		vis_psub16(TMP6, TMP4, DST_0);
		vis_st64(DST_0, dest[0]);
		dest += stride;

		vis_faligndata(TMP0, TMP2, REF_2);

		vis_and(TMP12, MASK_7f, TMP12);

		vis_psub16(TMP14, TMP12, DST_0);
		vis_st64(DST_0, dest[0]);
		dest += stride;
	} while (--height);

	vis_ld64(ref[0], TMP0);
	vis_xor(REF_0, REF_2, TMP4);

	vis_ld64_2(ref, offset, TMP2);
	vis_and(TMP4, MASK_fe, TMP4);

	vis_or(REF_0, REF_2, TMP6);
	vis_mul8x16(CONST_128, TMP4, TMP4);

	vis_faligndata(TMP0, TMP2, REF_0);

	vis_xor(REF_0, REF_2, TMP12);

	vis_and(TMP4, MASK_7f, TMP4);

	vis_and(TMP12, MASK_fe, TMP12);

	vis_mul8x16(CONST_128, TMP12, TMP12);
	vis_or(REF_0, REF_2, TMP14);

	vis_psub16(TMP6, TMP4, DST_0);
	vis_st64(DST_0, dest[0]);
	dest += stride;

	vis_and(TMP12, MASK_7f, TMP12);

	vis_psub16(TMP14, TMP12, DST_0);
	vis_st64(DST_0, dest[0]);
}

static void MC_avg_y_16_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int stride_8 = stride + 8;
	int stride_16;
	int offset;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 16 : 0;

	vis_ld64(ref[ 0], TMP0);
	vis_fzero(ZERO);

	vis_ld64(ref[ 8], TMP2);

	vis_ld64_2(ref, offset, TMP4);
	stride_16 = stride + offset;

	vis_ld64(constants3[0], CONST_3);
	vis_faligndata(TMP0, TMP2, REF_2);

	vis_ld64(constants256_512[0], CONST_256);
	vis_faligndata(TMP2, TMP4, REF_6);
	height >>= 1;

	do {	/* 31 cycles */
		vis_ld64_2(ref, stride, TMP0);
		vis_pmerge(ZERO,       REF_2,     TMP12);
		vis_mul8x16au(REF_2_1, CONST_256, TMP14);

		vis_ld64_2(ref, stride_8, TMP2);
		vis_pmerge(ZERO,       REF_6,     TMP16);
		vis_mul8x16au(REF_6_1, CONST_256, TMP18);

		vis_ld64_2(ref, stride_16, TMP4);
		ref += stride;

		vis_ld64(dest[0], DST_0);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(dest, 8, DST_2);
		vis_faligndata(TMP2, TMP4, REF_4);

		vis_ld64_2(ref, stride, TMP6);
		vis_pmerge(ZERO,     REF_0,     TMP0);
		vis_mul8x16au(REF_0_1, CONST_256, TMP2);

		vis_ld64_2(ref, stride_8, TMP8);
		vis_pmerge(ZERO,     REF_4,     TMP4);

		vis_ld64_2(ref, stride_16, TMP10);
		ref += stride;

		vis_ld64_2(dest, stride, REF_S0/*DST_4*/);
		vis_faligndata(TMP6, TMP8, REF_2);
		vis_mul8x16au(REF_4_1, CONST_256, TMP6);

		vis_ld64_2(dest, stride_8, REF_S2/*DST_6*/);
		vis_faligndata(TMP8, TMP10, REF_6);
		vis_mul8x16al(DST_0,   CONST_512, TMP20);

		vis_padd16(TMP0, CONST_3, TMP0);
		vis_mul8x16al(DST_1,   CONST_512, TMP22);

		vis_padd16(TMP2, CONST_3, TMP2);
		vis_mul8x16al(DST_2,   CONST_512, TMP24);

		vis_padd16(TMP4, CONST_3, TMP4);
		vis_mul8x16al(DST_3,   CONST_512, TMP26);

		vis_padd16(TMP6, CONST_3, TMP6);

		vis_padd16(TMP12, TMP20, TMP12);
		vis_mul8x16al(REF_S0,   CONST_512, TMP20);

		vis_padd16(TMP14, TMP22, TMP14);
		vis_mul8x16al(REF_S0_1, CONST_512, TMP22);

		vis_padd16(TMP16, TMP24, TMP16);
		vis_mul8x16al(REF_S2,   CONST_512, TMP24);

		vis_padd16(TMP18, TMP26, TMP18);
		vis_mul8x16al(REF_S2_1, CONST_512, TMP26);

		vis_padd16(TMP12, TMP0, TMP12);
		vis_mul8x16au(REF_2,   CONST_256, TMP28);

		vis_padd16(TMP14, TMP2, TMP14);
		vis_mul8x16au(REF_2_1, CONST_256, TMP30);

		vis_padd16(TMP16, TMP4, TMP16);
		vis_mul8x16au(REF_6,   CONST_256, REF_S4);

		vis_padd16(TMP18, TMP6, TMP18);
		vis_mul8x16au(REF_6_1, CONST_256, REF_S6);

		vis_pack16(TMP12, DST_0);
		vis_padd16(TMP28, TMP0, TMP12);

		vis_pack16(TMP14, DST_1);
		vis_st64(DST_0, dest[0]);
		vis_padd16(TMP30, TMP2, TMP14);

		vis_pack16(TMP16, DST_2);
		vis_padd16(REF_S4, TMP4, TMP16);

		vis_pack16(TMP18, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;
		vis_padd16(REF_S6, TMP6, TMP18);

		vis_padd16(TMP12, TMP20, TMP12);

		vis_padd16(TMP14, TMP22, TMP14);
		vis_pack16(TMP12, DST_0);

		vis_padd16(TMP16, TMP24, TMP16);
		vis_pack16(TMP14, DST_1);
		vis_st64(DST_0, dest[0]);

		vis_padd16(TMP18, TMP26, TMP18);
		vis_pack16(TMP16, DST_2);

		vis_pack16(TMP18, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;
	} while (--height);
}

static void MC_avg_y_8_vis (uint8_t * dest, const uint8_t * _ref,
			    const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	int stride_8;
	int offset;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);
	offset = (ref != _ref) ? 8 : 0;

	vis_ld64(ref[ 0], TMP0);
	vis_fzero(ZERO);

	vis_ld64_2(ref, offset, TMP2);
	stride_8 = stride + offset;

	vis_ld64(constants3[0], CONST_3);
	vis_faligndata(TMP0, TMP2, REF_2);

	vis_ld64(constants256_512[0], CONST_256);

	height >>= 1;
	do {	/* 20 cycles */
		vis_ld64_2(ref, stride, TMP0);
		vis_pmerge(ZERO,       REF_2,     TMP8);
		vis_mul8x16au(REF_2_1, CONST_256, TMP10);

		vis_ld64_2(ref, stride_8, TMP2);
		ref += stride;

		vis_ld64(dest[0], DST_0);

		vis_ld64_2(dest, stride, DST_2);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, stride, TMP4);
		vis_mul8x16al(DST_0,   CONST_512, TMP16);
		vis_pmerge(ZERO,       REF_0,     TMP12);

		vis_ld64_2(ref, stride_8, TMP6);
		ref += stride;
		vis_mul8x16al(DST_1,   CONST_512, TMP18);
		vis_pmerge(ZERO,       REF_0_1,   TMP14);

		vis_padd16(TMP12, CONST_3, TMP12);
		vis_mul8x16al(DST_2,   CONST_512, TMP24);

		vis_padd16(TMP14, CONST_3, TMP14);
		vis_mul8x16al(DST_3,   CONST_512, TMP26);

		vis_faligndata(TMP4, TMP6, REF_2);

		vis_padd16(TMP8, TMP12, TMP8);

		vis_padd16(TMP10, TMP14, TMP10);
		vis_mul8x16au(REF_2,   CONST_256, TMP20);

		vis_padd16(TMP8, TMP16, TMP0);
		vis_mul8x16au(REF_2_1, CONST_256, TMP22);

		vis_padd16(TMP10, TMP18, TMP2);
		vis_pack16(TMP0, DST_0);

		vis_pack16(TMP2, DST_1);
		vis_st64(DST_0, dest[0]);
		dest += stride;
		vis_padd16(TMP12, TMP20, TMP12);

		vis_padd16(TMP14, TMP22, TMP14);

		vis_padd16(TMP12, TMP24, TMP0);

		vis_padd16(TMP14, TMP26, TMP2);
		vis_pack16(TMP0, DST_2);

		vis_pack16(TMP2, DST_3);
		vis_st64(DST_2, dest[0]);
		dest += stride;
	} while (--height);
}

static void MC_put_xy_16_vis (uint8_t * dest, const uint8_t * _ref,
			      const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;
	int stride_8 = stride + 8;
	int stride_16 = stride + 16;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);

	vis_ld64(ref[ 0], TMP0);
	vis_fzero(ZERO);

	vis_ld64(ref[ 8], TMP2);

	vis_ld64(ref[16], TMP4);

	vis_ld64(constants2[0], CONST_2);
	vis_faligndata(TMP0, TMP2, REF_S0);

	vis_ld64(constants256_512[0], CONST_256);
	vis_faligndata(TMP2, TMP4, REF_S4);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_S2);
		vis_faligndata(TMP2, TMP4, REF_S6);
	} else {
		vis_src1(TMP2, REF_S2);
		vis_src1(TMP4, REF_S6);
	}

	height >>= 1;
	do {
		vis_ld64_2(ref, stride, TMP0);
		vis_mul8x16au(REF_S0, CONST_256, TMP12);
		vis_pmerge(ZERO,      REF_S0_1,  TMP14);

		vis_alignaddr_g0((void *)off);

		vis_ld64_2(ref, stride_8, TMP2);
		vis_mul8x16au(REF_S2, CONST_256, TMP16);
		vis_pmerge(ZERO,      REF_S2_1,  TMP18);

		vis_ld64_2(ref, stride_16, TMP4);
		ref += stride;
		vis_mul8x16au(REF_S4, CONST_256, TMP20);
		vis_pmerge(ZERO,      REF_S4_1,  TMP22);

		vis_ld64_2(ref, stride, TMP6);
		vis_mul8x16au(REF_S6, CONST_256, TMP24);
		vis_pmerge(ZERO,      REF_S6_1,  TMP26);

		vis_ld64_2(ref, stride_8, TMP8);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, stride_16, TMP10);
		ref += stride;
		vis_faligndata(TMP2, TMP4, REF_4);

		vis_faligndata(TMP6, TMP8, REF_S0);

		vis_faligndata(TMP8, TMP10, REF_S4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_2);
			vis_faligndata(TMP2, TMP4, REF_6);
			vis_faligndata(TMP6, TMP8, REF_S2);
			vis_faligndata(TMP8, TMP10, REF_S6);
		} else {
			vis_src1(TMP2, REF_2);
			vis_src1(TMP4, REF_6);
			vis_src1(TMP8, REF_S2);
			vis_src1(TMP10, REF_S6);
		}

		vis_mul8x16au(REF_0, CONST_256, TMP0);
		vis_pmerge(ZERO,      REF_0_1,  TMP2);

		vis_mul8x16au(REF_2, CONST_256, TMP4);
		vis_pmerge(ZERO,      REF_2_1,  TMP6);

		vis_padd16(TMP0, CONST_2, TMP8);
		vis_mul8x16au(REF_4, CONST_256, TMP0);

		vis_padd16(TMP2, CONST_2, TMP10);
		vis_mul8x16au(REF_4_1, CONST_256, TMP2);

		vis_padd16(TMP8, TMP4, TMP8);
		vis_mul8x16au(REF_6, CONST_256, TMP4);

		vis_padd16(TMP10, TMP6, TMP10);
		vis_mul8x16au(REF_6_1, CONST_256, TMP6);

		vis_padd16(TMP12, TMP8, TMP12);

		vis_padd16(TMP14, TMP10, TMP14);

		vis_padd16(TMP12, TMP16, TMP12);

		vis_padd16(TMP14, TMP18, TMP14);
		vis_pack16(TMP12, DST_0);

		vis_pack16(TMP14, DST_1);
		vis_st64(DST_0, dest[0]);
		vis_padd16(TMP0, CONST_2, TMP12);

		vis_mul8x16au(REF_S0, CONST_256, TMP0);
		vis_padd16(TMP2, CONST_2, TMP14);

		vis_mul8x16au(REF_S0_1, CONST_256, TMP2);
		vis_padd16(TMP12, TMP4, TMP12);

		vis_mul8x16au(REF_S2, CONST_256, TMP4);
		vis_padd16(TMP14, TMP6, TMP14);

		vis_mul8x16au(REF_S2_1, CONST_256, TMP6);
		vis_padd16(TMP20, TMP12, TMP20);

		vis_padd16(TMP22, TMP14, TMP22);

		vis_padd16(TMP20, TMP24, TMP20);

		vis_padd16(TMP22, TMP26, TMP22);
		vis_pack16(TMP20, DST_2);

		vis_pack16(TMP22, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;
		vis_padd16(TMP0, TMP4, TMP24);

		vis_mul8x16au(REF_S4, CONST_256, TMP0);
		vis_padd16(TMP2, TMP6, TMP26);

		vis_mul8x16au(REF_S4_1, CONST_256, TMP2);
		vis_padd16(TMP24, TMP8, TMP24);

		vis_padd16(TMP26, TMP10, TMP26);
		vis_pack16(TMP24, DST_0);

		vis_pack16(TMP26, DST_1);
		vis_st64(DST_0, dest[0]);
		vis_pmerge(ZERO, REF_S6, TMP4);

		vis_pmerge(ZERO,      REF_S6_1,  TMP6);

		vis_padd16(TMP0, TMP4, TMP0);

		vis_padd16(TMP2, TMP6, TMP2);

		vis_padd16(TMP0, TMP12, TMP0);

		vis_padd16(TMP2, TMP14, TMP2);
		vis_pack16(TMP0, DST_2);

		vis_pack16(TMP2, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;
	} while (--height);
}

static void MC_put_xy_8_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;
	int stride_8 = stride + 8;

	vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);

	vis_ld64(ref[ 0], TMP0);
	vis_fzero(ZERO);

	vis_ld64(ref[ 8], TMP2);

	vis_ld64(constants2[0], CONST_2);

	vis_ld64(constants256_512[0], CONST_256);
	vis_faligndata(TMP0, TMP2, REF_S0);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_S2);
	} else {
		vis_src1(TMP2, REF_S2);
	}

	height >>= 1;
	do {	/* 26 cycles */
		vis_ld64_2(ref, stride, TMP0);
		vis_mul8x16au(REF_S0,   CONST_256, TMP8);
		vis_pmerge(ZERO,        REF_S2,    TMP12);

		vis_alignaddr_g0((void *)off);

		vis_ld64_2(ref, stride_8, TMP2);
		ref += stride;
		vis_mul8x16au(REF_S0_1, CONST_256, TMP10);
		vis_pmerge(ZERO,        REF_S2_1,  TMP14);

		vis_ld64_2(ref, stride, TMP4);

		vis_ld64_2(ref, stride_8, TMP6);
		ref += stride;
		vis_faligndata(TMP0, TMP2, REF_S4);

		vis_pmerge(ZERO, REF_S4, TMP18);

		vis_pmerge(ZERO, REF_S4_1, TMP20);

		vis_faligndata(TMP4, TMP6, REF_S0);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_S6);
			vis_faligndata(TMP4, TMP6, REF_S2);
		} else {
			vis_src1(TMP2, REF_S6);
			vis_src1(TMP6, REF_S2);
		}

		vis_padd16(TMP18, CONST_2, TMP18);
		vis_mul8x16au(REF_S6,   CONST_256, TMP22);

		vis_padd16(TMP20, CONST_2, TMP20);
		vis_mul8x16au(REF_S6_1, CONST_256, TMP24);

		vis_mul8x16au(REF_S0,   CONST_256, TMP26);
		vis_pmerge(ZERO, REF_S0_1, TMP28);

		vis_mul8x16au(REF_S2,   CONST_256, TMP30);
		vis_padd16(TMP18, TMP22, TMP18);

		vis_mul8x16au(REF_S2_1, CONST_256, TMP32);
		vis_padd16(TMP20, TMP24, TMP20);

		vis_padd16(TMP8,  TMP18, TMP8);

		vis_padd16(TMP10, TMP20, TMP10);

		vis_padd16(TMP8,  TMP12, TMP8);

		vis_padd16(TMP10, TMP14, TMP10);
		vis_pack16(TMP8,  DST_0);

		vis_pack16(TMP10, DST_1);
		vis_st64(DST_0, dest[0]);
		dest += stride;
		vis_padd16(TMP18, TMP26, TMP18);

		vis_padd16(TMP20, TMP28, TMP20);

		vis_padd16(TMP18, TMP30, TMP18);

		vis_padd16(TMP20, TMP32, TMP20);
		vis_pack16(TMP18, DST_2);

		vis_pack16(TMP20, DST_3);
		vis_st64(DST_2, dest[0]);
		dest += stride;
	} while (--height);
}

static void MC_avg_xy_16_vis (uint8_t * dest, const uint8_t * _ref,
			      const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;
	int stride_8 = stride + 8;
	int stride_16 = stride + 16;

	vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);

	vis_ld64(ref[ 0], TMP0);
	vis_fzero(ZERO);

	vis_ld64(ref[ 8], TMP2);

	vis_ld64(ref[16], TMP4);

	vis_ld64(constants6[0], CONST_6);
	vis_faligndata(TMP0, TMP2, REF_S0);

	vis_ld64(constants256_1024[0], CONST_256);
	vis_faligndata(TMP2, TMP4, REF_S4);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_S2);
		vis_faligndata(TMP2, TMP4, REF_S6);
	} else {
		vis_src1(TMP2, REF_S2);
		vis_src1(TMP4, REF_S6);
	}

	height >>= 1;
	do {	/* 55 cycles */
		vis_ld64_2(ref, stride, TMP0);
		vis_mul8x16au(REF_S0, CONST_256, TMP12);
		vis_pmerge(ZERO,      REF_S0_1,  TMP14);

		vis_alignaddr_g0((void *)off);

		vis_ld64_2(ref, stride_8, TMP2);
		vis_mul8x16au(REF_S2, CONST_256, TMP16);
		vis_pmerge(ZERO,      REF_S2_1,  TMP18);

		vis_ld64_2(ref, stride_16, TMP4);
		ref += stride;
		vis_mul8x16au(REF_S4, CONST_256, TMP20);
		vis_pmerge(ZERO,      REF_S4_1,  TMP22);

		vis_ld64_2(ref, stride, TMP6);
		vis_mul8x16au(REF_S6, CONST_256, TMP24);
		vis_pmerge(ZERO,      REF_S6_1,  TMP26);

		vis_ld64_2(ref, stride_8, TMP8);
		vis_faligndata(TMP0, TMP2, REF_0);

		vis_ld64_2(ref, stride_16, TMP10);
		ref += stride;
		vis_faligndata(TMP2, TMP4, REF_4);

		vis_ld64(dest[0], DST_0);
		vis_faligndata(TMP6, TMP8, REF_S0);

		vis_ld64_2(dest, 8, DST_2);
		vis_faligndata(TMP8, TMP10, REF_S4);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_2);
			vis_faligndata(TMP2, TMP4, REF_6);
			vis_faligndata(TMP6, TMP8, REF_S2);
			vis_faligndata(TMP8, TMP10, REF_S6);
		} else {
			vis_src1(TMP2, REF_2);
			vis_src1(TMP4, REF_6);
			vis_src1(TMP8, REF_S2);
			vis_src1(TMP10, REF_S6);
		}

		vis_mul8x16al(DST_0,   CONST_1024, TMP30);
		vis_pmerge(ZERO, REF_0, TMP0);

		vis_mul8x16al(DST_1,   CONST_1024, TMP32);
		vis_pmerge(ZERO,      REF_0_1,  TMP2);

		vis_mul8x16au(REF_2, CONST_256, TMP4);
		vis_pmerge(ZERO,      REF_2_1,  TMP6);

		vis_mul8x16al(DST_2,   CONST_1024, REF_0);
		vis_padd16(TMP0, CONST_6, TMP0);

		vis_mul8x16al(DST_3,   CONST_1024, REF_2);
		vis_padd16(TMP2, CONST_6, TMP2);

		vis_padd16(TMP0, TMP4, TMP0);
		vis_mul8x16au(REF_4, CONST_256, TMP4);

		vis_padd16(TMP2, TMP6, TMP2);
		vis_mul8x16au(REF_4_1, CONST_256, TMP6);

		vis_padd16(TMP12, TMP0, TMP12);
		vis_mul8x16au(REF_6, CONST_256, TMP8);

		vis_padd16(TMP14, TMP2, TMP14);
		vis_mul8x16au(REF_6_1, CONST_256, TMP10);

		vis_padd16(TMP12, TMP16, TMP12);
		vis_mul8x16au(REF_S0, CONST_256, REF_4);

		vis_padd16(TMP14, TMP18, TMP14);
		vis_mul8x16au(REF_S0_1, CONST_256, REF_6);

		vis_padd16(TMP12, TMP30, TMP12);

		vis_padd16(TMP14, TMP32, TMP14);
		vis_pack16(TMP12, DST_0);

		vis_pack16(TMP14, DST_1);
		vis_st64(DST_0, dest[0]);
		vis_padd16(TMP4, CONST_6, TMP4);

		vis_ld64_2(dest, stride, DST_0);
		vis_padd16(TMP6, CONST_6, TMP6);
		vis_mul8x16au(REF_S2, CONST_256, TMP12);

		vis_padd16(TMP4, TMP8, TMP4);
		vis_mul8x16au(REF_S2_1, CONST_256,  TMP14);

		vis_padd16(TMP6, TMP10, TMP6);

		vis_padd16(TMP20, TMP4, TMP20);

		vis_padd16(TMP22, TMP6, TMP22);

		vis_padd16(TMP20, TMP24, TMP20);

		vis_padd16(TMP22, TMP26, TMP22);

		vis_padd16(TMP20, REF_0, TMP20);
		vis_mul8x16au(REF_S4, CONST_256, REF_0);

		vis_padd16(TMP22, REF_2, TMP22);
		vis_pack16(TMP20, DST_2);

		vis_pack16(TMP22, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;

		vis_ld64_2(dest, 8, DST_2);
		vis_mul8x16al(DST_0,   CONST_1024, TMP30);
		vis_pmerge(ZERO,      REF_S4_1,  REF_2);

		vis_mul8x16al(DST_1,   CONST_1024, TMP32);
		vis_padd16(REF_4, TMP0, TMP8);

		vis_mul8x16au(REF_S6, CONST_256, REF_4);
		vis_padd16(REF_6, TMP2, TMP10);

		vis_mul8x16au(REF_S6_1, CONST_256, REF_6);
		vis_padd16(TMP8, TMP12, TMP8);

		vis_padd16(TMP10, TMP14, TMP10);

		vis_padd16(TMP8, TMP30, TMP8);

		vis_padd16(TMP10, TMP32, TMP10);
		vis_pack16(TMP8, DST_0);

		vis_pack16(TMP10, DST_1);
		vis_st64(DST_0, dest[0]);

		vis_padd16(REF_0, TMP4, REF_0);

		vis_mul8x16al(DST_2,   CONST_1024, TMP30);
		vis_padd16(REF_2, TMP6, REF_2);

		vis_mul8x16al(DST_3,   CONST_1024, TMP32);
		vis_padd16(REF_0, REF_4, REF_0);

		vis_padd16(REF_2, REF_6, REF_2);

		vis_padd16(REF_0, TMP30, REF_0);

		/* stall */

		vis_padd16(REF_2, TMP32, REF_2);
		vis_pack16(REF_0, DST_2);

		vis_pack16(REF_2, DST_3);
		vis_st64_2(DST_2, dest, 8);
		dest += stride;
	} while (--height);
}

static void MC_avg_xy_8_vis (uint8_t * dest, const uint8_t * _ref,
			     const int stride, int height)
{
	uint8_t *ref = (uint8_t *) _ref;
	unsigned long off = (unsigned long) ref & 0x7;
	unsigned long off_plus_1 = off + 1;
	int stride_8 = stride + 8;

	vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT);

	ref = vis_alignaddr(ref);

	vis_ld64(ref[0], TMP0);
	vis_fzero(ZERO);

	vis_ld64_2(ref, 8, TMP2);

	vis_ld64(constants6[0], CONST_6);

	vis_ld64(constants256_1024[0], CONST_256);
	vis_faligndata(TMP0, TMP2, REF_S0);

	if (off != 0x7) {
		vis_alignaddr_g0((void *)off_plus_1);
		vis_faligndata(TMP0, TMP2, REF_S2);
	} else {
		vis_src1(TMP2, REF_S2);
	}

	height >>= 1;
	do {	/* 31 cycles */
		vis_ld64_2(ref, stride, TMP0);
		vis_mul8x16au(REF_S0, CONST_256, TMP8);
		vis_pmerge(ZERO,      REF_S0_1,  TMP10);

		vis_ld64_2(ref, stride_8, TMP2);
		ref += stride;
		vis_mul8x16au(REF_S2, CONST_256, TMP12);
		vis_pmerge(ZERO,      REF_S2_1,  TMP14);

		vis_alignaddr_g0((void *)off);

		vis_ld64_2(ref, stride, TMP4);
		vis_faligndata(TMP0, TMP2, REF_S4);

		vis_ld64_2(ref, stride_8, TMP6);
		ref += stride;

		vis_ld64(dest[0], DST_0);
		vis_faligndata(TMP4, TMP6, REF_S0);

		vis_ld64_2(dest, stride, DST_2);

		if (off != 0x7) {
			vis_alignaddr_g0((void *)off_plus_1);
			vis_faligndata(TMP0, TMP2, REF_S6);
			vis_faligndata(TMP4, TMP6, REF_S2);
		} else {
			vis_src1(TMP2, REF_S6);
			vis_src1(TMP6, REF_S2);
		}

		vis_mul8x16al(DST_0,   CONST_1024, TMP30);
		vis_pmerge(ZERO, REF_S4, TMP22);

		vis_mul8x16al(DST_1,   CONST_1024, TMP32);
		vis_pmerge(ZERO,      REF_S4_1,  TMP24);

		vis_mul8x16au(REF_S6, CONST_256, TMP26);
		vis_pmerge(ZERO,      REF_S6_1,  TMP28);

		vis_mul8x16au(REF_S0, CONST_256, REF_S4);
		vis_padd16(TMP22, CONST_6, TMP22);

		vis_mul8x16au(REF_S0_1, CONST_256, REF_S6);
		vis_padd16(TMP24, CONST_6, TMP24);

		vis_mul8x16al(DST_2,   CONST_1024, REF_0);
		vis_padd16(TMP22, TMP26, TMP22);

		vis_mul8x16al(DST_3,   CONST_1024, REF_2);
		vis_padd16(TMP24, TMP28, TMP24);

		vis_mul8x16au(REF_S2, CONST_256, TMP26);
		vis_padd16(TMP8, TMP22, TMP8);

		vis_mul8x16au(REF_S2_1, CONST_256, TMP28);
		vis_padd16(TMP10, TMP24, TMP10);

		vis_padd16(TMP8, TMP12, TMP8);

		vis_padd16(TMP10, TMP14, TMP10);

		vis_padd16(TMP8, TMP30, TMP8);

		vis_padd16(TMP10, TMP32, TMP10);
		vis_pack16(TMP8, DST_0);

		vis_pack16(TMP10, DST_1);
		vis_st64(DST_0, dest[0]);
		dest += stride;

		vis_padd16(REF_S4, TMP22, TMP12);

		vis_padd16(REF_S6, TMP24, TMP14);

		vis_padd16(TMP12, TMP26, TMP12);

		vis_padd16(TMP14, TMP28, TMP14);

		vis_padd16(TMP12, REF_0, TMP12);

		vis_padd16(TMP14, REF_2, TMP14);
		vis_pack16(TMP12, DST_2);

		vis_pack16(TMP14, DST_3);
		vis_st64(DST_2, dest[0]);
		dest += stride;
	} while (--height);
}

MPEG2_MC_EXTERN(vis);

#endif  /* !(ARCH_SPARC) */




More information about the MPlayer-cvslog mailing list