pred_comp.c

Motion JPEG编解码器源代码

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

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

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

			    /* ((d[i]+((s[i]+s[i+lx]+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);

			    lS0 = lS2;

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    }
		    /* }}} */
		} else /* w == 16 */ {
		    /* {{{ */

		    perm = vec_lvsl(0, s);

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

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

			/* (unsigned short[]) s+lx[8-15] */
			vu8(s2L) = vec_mergel(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);
			s0L = vec_add(s0L, s2L);

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

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

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

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

			/* ((d[i]+((s[i]+s[i+lx]+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);

			lS0 = lS2;

			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+lx]+1)>>1)+1)>>1;
		    s+= lx;
		    d+= lx;
		}
#endif
		/* }}} */
	    } else /* !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[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);


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


			    /* 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);


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

			    lS0 = lS2;

			    s += lx;
			    d += lx;
			}
			/* }}} */
		    }
		    /* }}} */
		} else /* w == 16 */ {
		    /* {{{ */

		    perm = vec_lvsl(0, s);

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


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

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


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

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

			/* (s[i]+s[i+lx]+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);

			lS0 = lS2;

			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+lx]+1)>>1;
		    s+= lx;
		    d+= lx;
		}
#endif
		/* }}} */
	    }
	    /* }}} */
	} else /* !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) {
			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);

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

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

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

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


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

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

			    /* (d[i]+s[i]+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);

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

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

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

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


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

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

			    /* (d[i]+s[i]+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);

		    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);

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

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


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

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

			/* (d[i]+s[i]+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] = (unsigned int)(d[i]+s[i]+1)>>1;
		    s+= lx;
		    d+= lx;
		}
#endif
		/* }}} */
	    } else /* !addflag */ {
		/* {{{ */
		if (w == 8) {
		    /* {{{ */
		    /* w / 4, int size copying instead of byte size */
		    for (j = 0; j < h; j++) {
			*((unsigned int*)d)   = *((unsigned int*)s);
			*((unsigned int*)d+1) = *((unsigned int*)s+1);
			s += lx;
			d += lx;
		    }
		    /* }}} */
		} else /* w == 16 */ {
		    /* {{{ */
		    perm = vec_lvsl(0, s);

		    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);

			vec_st(vu8(lS0), 0, (unsigned char*)d);

			s += lx;
			d += lx;
		    }
		    /* }}} */
		}
		/* }}} */
	    }
	    /* }}} */
	}
    }

    AMBER_STOP;
}

#if ALTIVEC_TEST_FUNCTION(pred_comp) /* {{{ */
#  ifdef ALTIVEC_VERIFY

void pred_comp_altivec_verify(PRED_COMP_PDECL)
{
    int i, j;
    uint8_t *s, *d;
    unsigned long checksum1, checksum2;
    /* dst is used as input and output, it must be saved/restored */
    uint8_t dstcpy[16][16];
    uint8_t dc;

    /* save dst */
    d = dst + lx*y + x;
    for (j = 0; j < h; j++) {
	for (i = 0; i < w; i++) {
	    dstcpy[j][i] = d[i];
	}
	d += lx;
    }

    pred_comp_altivec(PRED_COMP_ARGS);
    d = dst + lx*y + x;
    for (checksum1 = j = 0; j < h; j++) {
	for (i = 0; i < w; i++) {
	    checksum1 += d[i];
	}
	d += lx;
    }

    /* restore/swap dst & dstcpy */
    d = dst + lx*y + x;
    for (j = 0; j < h; j++) {
	for (i = 0; i < w; i++) {
	    dc = d[i];
	    d[i] = dstcpy[j][i];
	    dstcpy[j][i] = dc;
	}
	d += lx;
    }

    ALTIVEC_TEST_WITH(pred_comp)(PRED_COMP_ARGS);
    d = dst + lx*y + x;
    for (checksum2 = j = 0; j < h; j++) {
	for (i = 0; i < w; i++) {
	    checksum2 += d[i];
	}
	d += lx;
    }

    if (checksum1 != checksum2) {
	s = src + lx*(y+(dy>>1)) + (x+(dx>>1));
	d = dst + lx*y + x;

	mjpeg_debug("pred_comp(" PRED_COMP_PFMT ") s=0x%X, d=0x%x",
	    PRED_COMP_ARGS, s, d);
	mjpeg_debug("pred_comp: checksums differ %d != %d",
	    checksum1, checksum2);

	/* d = dst + lx*y + x; */
	for (j = 0; j < h; j++) {
	    for (i = 0; i < w; i++) {
		if (dstcpy[j][i] != d[i]) {
		    mjpeg_debug("pred_comp: dst[%d][%d] %d != %d",
			j, i, dstcpy[j][i], d[i]);
		}
	    }
	    d += lx;
	}
    }
}

#  else
ALTIVEC_TEST(pred_comp, void, (PRED_COMP_PDECL),
    PRED_COMP_PFMT, PRED_COMP_ARGS);
#  endif
#endif /* }}} */
/* vim:set foldmethod=marker foldlevel=0: */