motion_comp_vis.c

mpeg2 decoder filter过滤服务端

/*
 * motion_comp_vis.c
 * Copyright (C) 2003 David S. Miller <[e-mail]>
 *
 * 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);