pred_comp.c

Motion JPEG编解码器源代码

		} else /* w == 16 */ {
		    /* {{{ */
		    perm = vec_lvsl(0, s);
		    perm1 = vec_splat_u8(1);
		    perm1 = vec_add(perm, perm1);

		    lS = vec_ld(0, (unsigned char*)s);
		    lS1 = vec_ld(16, (unsigned char*)s);
		    lS0 = vec_perm(lS, lS1, perm);
		    lS1 = vec_perm(lS, lS1, perm1);
		    s += lx;

		    for (i = 0; i < h; i++) {
			lS = vec_ld(0, (unsigned char*)s);
			lS3 = vec_ld(16, (unsigned char*)s);
			lS2 = vec_perm(lS, lS3, perm);
			lS3 = vec_perm(lS, lS3, perm1);


			/* (unsigned short[]) s[0-7] */
			vu8(s0H) = vec_mergeh(zero, lS0);

			/* (unsigned short[]) s[8-15] */
			vu8(s0L) = vec_mergel(zero, lS0);

			/* (unsigned short[]) s[1-8] */
			vu8(s1H) = vec_mergeh(zero, lS1);

			/* (unsigned short[]) s[9-16] */
			vu8(s1L) = vec_mergel(zero, lS1);

			/* (unsigned short[]) s+lx[0-7] */
			vu8(s2H) = vec_mergeh(zero, lS2);

			/* (unsigned short[]) s+lx[8-15] */
			vu8(s2L) = vec_mergel(zero, lS2);

			/* (unsigned short[]) s+lx[1-8] */
			vu8(s3H) = vec_mergeh(zero, lS3);

			/* (unsigned short[]) s+lx[9-16] */
			vu8(s3L) = vec_mergel(zero, lS3);



			/* s[i] + s[i+1] */
			s0H = vec_add(s0H, s1H);
			s0L = vec_add(s0L, s1L);

			/* s[i+lx] + s[i+lx+1] */
			s2H = vec_add(s2H, s3H);
			s2L = vec_add(s2L, s3L);

			/* (s[i]+s[i+1]) + (s[i+lx]+s[i+lx+1]) */
			s0H = vec_add(s0H, s2H);
			s0L = vec_add(s0L, s2L);

			/* (s[i]+s[i+1]+s[i+lx]+s[i+lx+1]) + 2 */
			s0H = vec_add(s0H, two);
			s0L = vec_add(s0L, two);

			/* (s[i]+s[i+1]+s[i+lx]+s[i+lx+1]+2) >> 2 */
			s0H = vec_sra(s0H, two);
			s0L = vec_sra(s0L, two);


			/* (unsigned char[]) d[0-7], d[8-15] */
			vu8(dH) = vec_packsu(s0H, s0L);
			vec_st(vu8(dH), 0, (unsigned char*)d);

			lS0 = lS2;
			lS1 = lS3;

			s += lx;
			d += lx;
		    }
		    /* }}} */
		}
#if 0
		for (j=0; j<h; j++)
		{
		    for (i=0; i<w; i++)
			d[i] = (unsigned int)(s[i]+s[i+1]+s[i+lx]+s[i+lx+1]+2)>>2;
		    s+= lx;
		    d+= lx;
		}
#endif
		/* }}} */
	    }
	    /* }}} */
	} else /* !yh */ {
	    /* (xh && !yh) {{{ */
	    if (addflag) {
		/* {{{ */
		zero = vec_splat_u8(0);
		one = vec_splat_u16(1);

		if (w == 8) {
		    /* {{{ */
		    eight = vec_splat_u8(8);

		    perm = vec_lvsl(0, (unsigned char*)s);

		    if (((unsigned long)d & 0xf) == 0) {
			for (i = 0; i < h; i++) {
			    lS0 = vec_ld(0, (unsigned char*)s);
			    lS1 = vec_ld(16, (unsigned char*)s);
			    lS0 = vec_perm(lS0, lS1, perm);
			    lS1 = vec_slo(lS0, eight);

			    lD = vec_ld(0, (unsigned char*)d);

			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s[1-8] */
			    vu8(s1H) = vec_mergeh(zero, lS1);

			    /* (unsigned short[]) d[0-7] */
			    vu8(dH) = vec_mergeh(zero, lD);

			    /* (unsigned short[]) d[8-15] */
			    vu8(dL) = vec_mergel(zero, lD);

			    /* s[i] + s[i+1] */
			    s0H = vec_add(s0H, s1H);

			    /* (s[i]+s[i+1]) + 1 */
			    s0H = vec_add(s0H, one);

			    /* (s[i]+s[i+1]+1) >> 1 */
			    s0H = vec_sra(s0H, one);

			    /* d[i] + ((s[i]+s[i+1]+1)>>1) */
			    dH = vec_add(dH, s0H);

			    /* (d[i]+((s[i]+s[i+1]+1)>>1)) + 1 */
			    dH = vec_add(dH, one);

			    /* ((d[i]+((s[i]+s[i+1]+1)>>1))+1) >> 1 */
			    dH = vec_sra(dH, one);

			    /* (unsigned char[]) d[0-7], d[8-15] */
			    vu8(dH) = vec_packsu(dH, dL);
			    vec_st(vu8(dH), 0, (unsigned char*)d);

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    } else {
			/* {{{ */
			/* d is offset 8 bytes, work on dL (low bytes).
			 */
			for (i = 0; i < h; i++) {
			    lS0 = vec_ld(0, (unsigned char*)s);
			    lS1 = vec_ld(16, (unsigned char*)s);
			    lS0 = vec_perm(lS0, lS1, perm);
			    lS1 = vec_slo(lS0, eight);

			    lD = vec_ld(0, (unsigned char*)d);

			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s[1-8] */
			    vu8(s1H) = vec_mergeh(zero, lS1);

			    /* (unsigned short[]) d[8-15] */
			    vu8(dL) = vec_mergel(zero, lD);

			    /* (unsigned short[]) d[0-7] */
			    vu8(dH) = vec_mergeh(zero, lD);

			    /* s[i] + s[i+1] */
			    s0H = vec_add(s0H, s1H);

			    /* (s[i]+s[i+1]) + 1 */
			    s0H = vec_add(s0H, one);

			    /* (s[i]+s[i+1]+1) >> 1 */
			    s0H = vec_sra(s0H, one);

			    /* d[i] + ((s[i]+s[i+1]+1)>>1) */
			    dL = vec_add(dL, s0H);

			    /* (d[i]+((s[i]+s[i+1]+1)>>1)) + 1 */
			    dL = vec_add(dL, one);

			    /* ((d[i]+((s[i]+s[i+1]+1)>>1))+1) >> 1 */
			    dL = vec_sra(dL, one);

			    /* (unsigned char[]) d[0-7], d[8-15] */
			    vu8(dH) = vec_packsu(dH, dL);
			    vec_st(vu8(dH), 0, (unsigned char*)d);

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    }
		} else /* w == 16 */ {
		    /* {{{ */
		    perm = vec_lvsl(0, s);
		    perm1 = vec_splat_u8(1);
		    perm1 = vec_add(perm, perm1);

		    for (i = 0; i < h; i++) {
			lS = vec_ld(0, (unsigned char*)s);
			lS1 = vec_ld(16, (unsigned char*)s);
			lS0 = vec_perm(lS, lS1, perm);
			lS1 = vec_perm(lS, lS1, perm1);

			lD = vec_ld(0, (unsigned char*)d);

			/* (unsigned short[]) s[0-7] */
			vu8(s0H) = vec_mergeh(zero, lS0);

			/* (unsigned short[]) s[8-15] */
			vu8(s0L) = vec_mergel(zero, lS0);

			/* (unsigned short[]) s[1-8] */
			vu8(s1H) = vec_mergeh(zero, lS1);

			/* (unsigned short[]) s[9-16] */
			vu8(s1L) = vec_mergel(zero, lS1);

			/* (unsigned short[]) d[0-7] */
			vu8(dH) = vec_mergeh(zero, lD);

			/* (unsigned short[]) d[8-15] */
			vu8(dL) = vec_mergel(zero, lD);


			/* s[i] + s[i+1] */
			s0H = vec_add(s0H, s1H);
			s0L = vec_add(s0L, s1L);

			/* (s[i]+s[i+1]) + 1 */
			s0H = vec_add(s0H, one);
			s0L = vec_add(s0L, one);

			/* (s[i]+s[i+1]+1) >> 1 */
			s0H = vec_sra(s0H, one);
			s0L = vec_sra(s0L, one);

			/* d[i] + ((s[i]+s[i+1]+1)>>1) */
			dH = vec_add(dH, s0H);
			dL = vec_add(dL, s0L);

			/* (d[i]+((s[i]+s[i+1]+1)>>1)) + 1 */
			dH = vec_add(dH, one);
			dL = vec_add(dL, one);

			/* ((d[i]+((s[i]+s[i+1]+1)>>1))+1) >> 1 */
			dH = vec_sra(dH, one);
			dL = vec_sra(dL, one);

			/* (unsigned char[]) d[0-7], d[8-15] */
			vu8(dH) = vec_packsu(dH, dL);
			vec_st(vu8(dH), 0, (unsigned char*)d);

			s += lx;
			d += lx;
		    }
#if 0
		    for (j=0; j<h; j++)
		    {
			for (i=0; i<w; i++)
			    d[i] = (d[i] +
				    ((unsigned int)(s[i]+s[i+1]+1)>>1)+1)>>1;
			s += lx;
			d += lx;
		    }
#endif
		    /* }}} */
		}
		/* }}} */
	    } else /* !addflag */ {
		/* {{{ */
		zero = vec_splat_u8(0);
		one = vec_splat_u16(1);

		if (w == 8) {
		    /* {{{ */
		    eight = vec_splat_u8(8);

		    perm = vec_lvsl(0, (unsigned char*)s);

		    if (((unsigned long)d & 0xf) == 0) {
			/* {{{ */
			for (i = 0; i < h; i++) {
			    lS0 = vec_ld(0, (unsigned char*)s);
			    lS1 = vec_ld(16, (unsigned char*)s);
			    lS0 = vec_perm(lS0, lS1, perm);
			    lS1 = vec_slo(lS0, eight);

			    lD = vec_ld(0, (unsigned char*)d);


			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s[1-8] */
			    vu8(s1H) = vec_mergeh(zero, lS1);

			    /* (unsigned short[]) d[8-15] */
			    vu8(dL) = vec_mergel(zero, lD);


			    /* s[i] + s[i+1] */
			    s0H = vec_add(s0H, s1H);

			    /* (s[i]+s[i+1]) + 1 */
			    s0H = vec_add(s0H, one);

			    /* (s[i]+s[i+1]+1) >> 1 */
			    s0H = vec_sra(s0H, one);


			    /* (unsigned char[]) d[0-7], d[8-15] */
			    vu8(dH) = vec_packsu(s0H, dL);
			    vec_st(vu8(dH), 0, (unsigned char*)d);

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    } else {
			/* {{{ */
			/* d is offset 8 bytes, work on dL (low bytes).
			 */
			for (i = 0; i < h; i++) {
			    lS0 = vec_ld(0, (unsigned char*)s);
			    lS1 = vec_ld(16, (unsigned char*)s);
			    lS0 = vec_perm(lS0, lS1, perm);
			    lS1 = vec_slo(lS0, eight);

			    lD = vec_ld(0, (unsigned char*)d);

			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s[1-8] */
			    vu8(s1H) = vec_mergeh(zero, lS1);

			    /* (unsigned short[]) d[0-7] */
			    vu8(dH) = vec_mergeh(zero, lD);


			    /* s[i] + s[i+1] */
			    s0H = vec_add(s0H, s1H);

			    /* (s[i]+s[i+1]) + 1 */
			    s0H = vec_add(s0H, one);

			    /* (s[i]+s[i+1]+1) >> 1 */
			    s0H = vec_sra(s0H, one);


			    /* (unsigned char[]) d[0-7], d[8-15] */
			    vu8(dH) = vec_packsu(dH, s0H);
			    vec_st(vu8(dH), 0, (unsigned char*)d);

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    }
		    /* }}} */
		} else /* w == 16 */ {
		    /* {{{ */
		    perm = vec_lvsl(0, s);
		    perm1 = vec_splat_u8(1);
		    perm1 = vec_add(perm, perm1);

		    for (i = 0; i < h; i++) {
			lS = vec_ld(0, (unsigned char*)s);
			lS1 = vec_ld(16, (unsigned char*)s);
			lS0 = vec_perm(lS, lS1, perm);
			lS1 = vec_perm(lS, lS1, perm1);


			/* (unsigned short[]) s[0-7] */
			vu8(s0H) = vec_mergeh(zero, lS0);

			/* (unsigned short[]) s[8-15] */
			vu8(s0L) = vec_mergel(zero, lS0);

			/* (unsigned short[]) s[1-8] */
			vu8(s1H) = vec_mergeh(zero, lS1);

			/* (unsigned short[]) s[9-16] */
			vu8(s1L) = vec_mergel(zero, lS1);


			/* s[i] + s[i+1] */
			s0H = vec_add(s0H, s1H);
			s0L = vec_add(s0L, s1L);

			/* (s[i]+s[i+1]) + 1 */
			s0H = vec_add(s0H, one);
			s0L = vec_add(s0L, one);

			/* (s[i]+s[i+1]+1) >> 1 */
			s0H = vec_sra(s0H, one);
			s0L = vec_sra(s0L, one);


			/* (unsigned char[]) d[0-7], d[8-15] */
			vu8(dH) = vec_packsu(s0H, s0L);
			vec_st(vu8(dH), 0, (unsigned char*)d);

			s += lx;
			d += lx;
		    }
		    /* }}} */
		}
#if 0
		for (j=0; j<h; j++)
		{
		    for (i=0; i<w; i++)
			d[i] = (unsigned int)(s[i]+s[i+1]+1)>>1;
		    s+= lx;
		    d+= lx;
		}
#endif
		/* }}} */
	    }
	    /* }}} */
	}
    } else /* !xh */ {
	if  (yh) {
	    /* (!xh && yh) {{{ */
	    if (addflag) {
		/* {{{ */
		zero = vec_splat_u8(0);
		one = vec_splat_u16(1);

		if (w == 8) {
		    /* {{{ */

		    perm = vec_lvsl(0, (unsigned char*)s);

		    if (((unsigned long)d & 0xf) == 0) {
			/* {{{ */
			lS0 = vec_ld(0, (unsigned char*)s);
			lS1 = vec_ld(16, (unsigned char*)s);
			lS0 = vec_perm(lS0, lS1, perm);
			s += lx;

			for (i = 0; i < h; i++) {
			    lS2 = vec_ld(0, (unsigned char*)s);
			    lS3 = vec_ld(16, (unsigned char*)s);
			    lS2 = vec_perm(lS2, lS3, perm);

			    lD = vec_ld(0, (unsigned char*)d);


			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s+lx[0-7] */
			    vu8(s2H) = vec_mergeh(zero, lS2);

			    /* (unsigned short[]) d[0-7] */
			    vu8(dH) = vec_mergeh(zero, lD);

			    /* (unsigned short[]) d[8-15] */
			    vu8(dL) = vec_mergel(zero, lD);


			    /* s[i] + s[i+lx] */
			    s0H = vec_add(s0H, s2H);

			    /* (s[i]+s[i+lx]) + 1 */
			    s0H = vec_add(s0H, one);

			    /* (s[i]+s[i+lx]+1) >> 1 */
			    s0H = vec_sra(s0H, one);

			    /* d[i] + ((s[i]+s[i+lx]+1)>>1) */
			    dH = vec_add(dH, s0H);

			    /* (d[i]+((s[i]+s[i+lx]+1)>>1)) + 1 */
			    dH = vec_add(dH, one);

			    /* ((d[i]+((s[i]+s[i+lx]+1)>>1))+1) >> 1 */
			    dH = vec_sra(dH, one);

			    /* (unsigned char[]) d[0-7], d[8-15] */
			    vu8(dH) = vec_packsu(dH, dL);
			    vec_st(vu8(dH), 0, (unsigned char*)d);

			    lS0 = lS2;

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    } else {
			/* {{{ */
			/* d is offset 8 bytes, work on dL (low bytes).
			 */
			lS0 = vec_ld(0, (unsigned char*)s);
			lS1 = vec_ld(16, (unsigned char*)s);
			lS0 = vec_perm(lS0, lS1, perm);
			s += lx;

			for (i = 0; i < h; i++) {
			    lS2 = vec_ld(0, (unsigned char*)s);
			    lS3 = vec_ld(16, (unsigned char*)s);
			    lS2 = vec_perm(lS2, lS3, perm);

			    lD = vec_ld(0, (unsigned char*)d);


			    /* (unsigned short[]) s[0-7] */
			    vu8(s0H) = vec_mergeh(zero, lS0);

			    /* (unsigned short[]) s+lx[0-7] */
			    vu8(s2H) = vec_mergeh(zero, lS2);

			    /* (unsigned short[]) d[0-7] */
			    vu8(dH) = vec_mergeh(zero, lD);

			    /* (unsigned short[]) d[8-15] */
			    vu8(dL) = vec_mergel(zero, lD);


			    /* s[i] + s[i+lx] */
			    s0H = vec_add(s0H, s2H);

			    /* (s[i]+s[i+lx]) + 1 */
			    s0H = vec_add(s0H, one);