mlib_status CONV_FUNC(MxN)(mlib_image *dst,
const mlib_image *src,
const mlib_s32 *kernel,
mlib_s32 m,
mlib_s32 n,
mlib_s32 dm,
mlib_s32 dn,
mlib_s32 scale,
mlib_s32 cmask)
{
FTYPE buff[BUFF_SIZE], *buffs_arr[2*(MAX_N + 1)];
FTYPE **buffs = buffs_arr, *buffd;
FTYPE akernel[256], *k = akernel, fscale = DSCALE;
mlib_s32 mn, l, off, kw, bsize, buff_ind;
mlib_s32 d0, d1;
FTYPE k0, k1, k2, k3, k4, k5, k6;
FTYPE p0, p1, p2, p3, p4, p5, p6, p7;
d64_2x32 dd;
DEF_VARS(DTYPE);
mlib_s32 chan2;
mlib_s32 *buffo, *buffi;
mlib_status status = MLIB_SUCCESS;
GET_SRC_DST_PARAMETERS(DTYPE);
if (scale > 30) {
fscale *= 1.0/(1 << 30);
scale -= 30;
}
fscale /= (1 << scale);
mn = m*n;
if (mn > 256) {
k = mlib_malloc(mn*sizeof(mlib_d64));
if (k == NULL) return MLIB_FAILURE;
}
for (i = 0; i < mn; i++) {
k[i] = kernel[i]*fscale;
}
if (m == 1) {
status = mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
FREE_AND_RETURN_STATUS;
}
bsize = (n + 3)*wid;
if ((bsize > BUFF_SIZE) || (n > MAX_N)) {
pbuff = mlib_malloc(sizeof(FTYPE)*bsize + sizeof(FTYPE *)*2*(n + 1));
if (pbuff == NULL) {
status = MLIB_FAILURE;
FREE_AND_RETURN_STATUS;
}
buffs = (FTYPE **)(pbuff + bsize);
}
for (l = 0; l < (n + 1); l++) buffs[l] = pbuff + l*wid;
for (l = 0; l < (n + 1); l++) buffs[l + (n + 1)] = buffs[l];
buffd = buffs[n] + wid;
buffo = (mlib_s32*)(buffd + wid);
buffi = buffo + (wid &~ 1);
chan1 = nchannel;
chan2 = chan1 + chan1;
wid -= (m - 1);
hgt -= (n - 1);
adr_dst += dn*dll + dm*nchannel;
for (c = 0; c < nchannel; c++) {
if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
sl = adr_src + c;
dl = adr_dst + c;
for (l = 0; l < n; l++) {
FTYPE *buff = buffs[l];
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i < wid + (m - 1); i++) {
buff[i] = (FTYPE)sl[i*chan1];
}
sl += sll;
}
buff_ind = 0;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i < wid; i++) buffd[i] = 0.0;
for (j = 0; j < hgt; j++) {
FTYPE **buffc = buffs + buff_ind;
FTYPE *buffn = buffc[n];
FTYPE *pk = k;
for (l = 0; l < n; l++) {
FTYPE *buff_l = buffc[l];
for (off = 0; off < m;) {
FTYPE *buff = buff_l + off;
kw = m - off;
if (kw > 2*MAX_KER) kw = MAX_KER; else
if (kw > MAX_KER) kw = kw/2;
off += kw;
sp = sl;
dp = dl;
p2 = buff[0]; p3 = buff[1]; p4 = buff[2];
p5 = buff[3]; p6 = buff[4]; p7 = buff[5];
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
pk += kw;
if (kw == 7) {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
p6 = buff[i + 6]; p7 = buff[i + 7];
buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; p5 = p7;
p6 = buff[i + 6]; p7 = buff[i + 7];
LOAD_BUFF(buffi);
dd.d64 = *(FTYPE *)(buffi + i);
buffn[i ] = (FTYPE)dd.i32s.i0;
buffn[i + 1] = (FTYPE)dd.i32s.i1;
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 6) {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
p5 = buff[i + 5]; p6 = buff[i + 6];
buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6;
p5 = buff[i + 5]; p6 = buff[i + 6];
buffn[i ] = (FTYPE)sp[0];
buffn[i + 1] = (FTYPE)sp[chan1];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 5) {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5;
p4 = buff[i + 4]; p5 = buff[i + 5];
buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4; p3 = p5;
p4 = buff[i + 4]; p5 = buff[i + 5];
buffn[i ] = (FTYPE)sp[0];
buffn[i + 1] = (FTYPE)sp[chan1];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 4) {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4;
p3 = buff[i + 3]; p4 = buff[i + 4];
buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4;
p3 = buff[i + 3]; p4 = buff[i + 4];
buffn[i ] = (FTYPE)sp[0];
buffn[i + 1] = (FTYPE)sp[chan1];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 3) {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3;
p2 = buff[i + 2]; p3 = buff[i + 3];
buffd[i ] += p0*k0 + p1*k1 + p2*k2;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3;
p2 = buff[i + 2]; p3 = buff[i + 3];
buffn[i ] = (FTYPE)sp[0];
buffn[i + 1] = (FTYPE)sp[chan1];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
} else /*if (kw == 2)*/ {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2;
p1 = buff[i + 1]; p2 = buff[i + 2];
buffd[i ] += p0*k0 + p1*k1;
buffd[i + 1] += p1*k0 + p2*k1;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2;
p1 = buff[i + 1]; p2 = buff[i + 2];
buffn[i ] = (FTYPE)sp[0];
buffn[i + 1] = (FTYPE)sp[chan1];
d0 = D2I(p0*k0 + p1*k1 + buffd[i ]);
d1 = D2I(p1*k0 + p2*k1 + buffd[i + 1]);
dp[0 ] = FROM_S32(d0);
dp[chan1] = FROM_S32(d1);
buffd[i ] = 0.0;
buffd[i + 1] = 0.0;
sp += chan2;
dp += chan2;
}
}
}
}
}
/* last pixels */
for (; i < wid; i++) {
FTYPE *pk = k, s = 0;
mlib_s32 x, d0;
for (l = 0; l < n; l++) {
FTYPE *buff = buffc[l] + i;
for (x = 0; x < m; x++) s += buff[x] * (*pk++);
}
d0 = D2I(s);
dp[0] = FROM_S32(d0);
buffn[i] = (FTYPE)sp[0];
sp += chan1;
dp += chan1;
}
for (l = 0; l < (m - 1); l++) buffn[wid + l] = sp[l*chan1];
/* next line */
sl += sll;
dl += dll;
buff_ind++;
if (buff_ind >= n + 1) buff_ind = 0;
}
}
FREE_AND_RETURN_STATUS;
}
mlib_status
mlib_c_conv2x2ext_u8(
mlib_image *dst,
const mlib_image *src,
mlib_s32 dx_l,
mlib_s32 dx_r,
mlib_s32 dy_t,
mlib_s32 dy_b,
const mlib_s32 *kern,
mlib_s32 scalef_expon,
mlib_s32 cmask)
{
mlib_d64 buff_arr[4 * BUFF_LINE];
mlib_s32 *pbuff = (mlib_s32 *)buff_arr, *buff0, *buff1, *buff2, *buffT;
DTYPE *adr_src, *sl, *sp, *sl1;
DTYPE *adr_dst, *dl, *dp;
mlib_d64 k0, k1, k2, k3, scalef = 1.0;
mlib_d64 p00, p01, p02, p10, p11, p12;
mlib_s32 wid, hgt, sll, dll, wid1;
mlib_s32 nchannel, chan1, chan2;
mlib_s32 i, j, c, swid;
LOAD_KERNEL_INTO_DOUBLE();
GET_SRC_DST_PARAMETERS(DTYPE);
vis_write_gsr(23 << 3);
swid = wid + D_KER;
wid1 = (swid + 1) & ~1;
if (wid1 > BUFF_LINE) {
pbuff = __mlib_malloc(4 * sizeof (mlib_s32) * wid1);
if (pbuff == NULL)
return (MLIB_FAILURE);
}
buff0 = pbuff;
buff1 = buff0 + wid1;
buff2 = buff1 + wid1;
chan1 = nchannel;
chan2 = chan1 + chan1;
swid -= dx_r;
for (c = 0; c < nchannel; c++) {
if (!(cmask & (1 << (nchannel - 1 - c))))
continue;
sl = adr_src + c;
dl = adr_dst + c;
if ((hgt - dy_b) > 0)
sl1 = sl + sll;
else
sl1 = sl;
#pragma pipeloop(0)
for (i = 0; i < swid; i++) {
buff0[i - 1] = (mlib_s32)sl[i * chan1];
buff1[i - 1] = (mlib_s32)sl1[i * chan1];
}
if (dx_r != 0) {
buff0[swid - 1] = buff0[swid - 2];
buff1[swid - 1] = buff1[swid - 2];
}
if ((hgt - dy_b) > 1)
sl = sl1 + sll;
else
sl = sl1;
for (j = 0; j < hgt; j++) {
sp = sl;
dp = dl;
buff2[-1] = (mlib_s32)sp[0];
sp += chan1;
p02 = buff0[-1];
p12 = buff1[-1];
#pragma pipeloop(0)
for (i = 0; i <= (wid - 2); i += 2) {
d64_2x32 sd0, sd1;
d64_2x32 dd0, dd1;
p00 = p02;
p10 = p12;
sd0.d64 = *(TYPE_64BIT *) (buff0 + i);
sd1.d64 = *(TYPE_64BIT *) (buff1 + i);
p01 = (mlib_d64)sd0.i32s.i0;
p02 = (mlib_d64)sd0.i32s.i1;
p11 = (mlib_d64)sd1.i32s.i0;
p12 = (mlib_d64)sd1.i32s.i1;
LOAD_BUFF(buff2);
dd0.i32s.i0 =
CLAMP_S32(p00 * k0 + p01 * k1 + p10 * k2 +
p11 * k3);
dd0.i32s.i1 =
CLAMP_S32(p01 * k0 + p02 * k1 + p11 * k2 +
p12 * k3);
dd1.d64 = vis_fpack32(dd1.d64, dd0.d64);
STORE2(dd1.i32s.i0, dd1.i32s.i1);
sp += chan2;
dp += chan2;
}
for (; i < wid; i++) {
d64_2x32 dd0, dd1;
p00 = buff0[i - 1];
p10 = buff1[i - 1];
p01 = buff0[i];
p11 = buff1[i];
buff2[i] = (mlib_s32)sp[0];
dd0.i32s.i1 =
CLAMP_S32(p00 * k0 + p01 * k1 + p10 * k2 +
p11 * k3);
dd1.d64 = vis_fpack32(dd1.d64, dd0.d64);
dp[0] = dd1.i32s.i1;
sp += chan1;
dp += chan1;
}
if (dx_r != 0)
buff2[swid - 1] = buff2[swid - 2];
if (j < hgt - dy_b - 2)
sl += sll;
dl += dll;
buffT = buff0;
buff0 = buff1;
buff1 = buff2;
buff2 = buffT;
}
}
if (pbuff != (mlib_s32 *)buff_arr)
__mlib_free(pbuff);
return (MLIB_SUCCESS);
}
