t_u32 mul_add(t_u32 *z, t_u32 *x, t_u32 *y, int n, t_u32 a)
{
if (a < (1 << A_BITS)) {
if (n == 8) {
DEF_VARS(8);
t_s32 c = 0;
MUL_U32_S64_8(0);
ADD_S64_U32_8(0);
return c;
} else if (n == 16) {
DEF_VARS(16);
t_s32 c = 0;
MUL_U32_S64_8(0);
MUL_U32_S64_8(4);
ADD_S64_U32_8(0);
ADD_S64_U32_8(8);
return c;
} else {
DEF_VARS(BUFF_SIZE);
t_s32 i, c = 0;
#pragma pipeloop(0)
for (i = 0; i < (n+1)/2; i ++) {
MUL_U32_S64_2(i);
}
#pragma pipeloop(0)
for (i = 0; i < n; i ++) {
ADD_S64_U32(i);
}
return c;
}
} else {
if (n == 8) {
DEF_VARS(2*8);
t_d64 d0, d1, db;
t_u32 uc = 0;
da = (t_d64)(a & A_MASK);
db = (t_d64)(a >> A_BITS);
MUL_U32_S64_D_8(0);
ADD_S64_U32_D_8(0);
return uc;
} else if (n == 16) {
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;
}
static mlib_status mlib_ImageConv1xN(mlib_image *dst,
const mlib_image *src,
const mlib_d64 *k,
mlib_s32 n,
mlib_s32 dn,
mlib_s32 cmask)
{
FTYPE buff[BUFF_SIZE];
mlib_s32 off, kh;
mlib_s32 d0, d1;
const FTYPE *pk;
FTYPE k0, k1, k2, k3;
FTYPE p0, p1, p2, p3, p4;
DEF_VARS(DTYPE);
DTYPE *sl_c, *dl_c, *sl0;
mlib_s32 l, hsize, max_hsize;
GET_SRC_DST_PARAMETERS(DTYPE);
hgt -= (n - 1);
adr_dst += dn*dll;
max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll;
if (!max_hsize) max_hsize = 1;
if (max_hsize > BUFF_SIZE) {
pbuff = mlib_malloc(sizeof(FTYPE)*max_hsize);
}
chan1 = nchannel;
sl_c = adr_src;
dl_c = adr_dst;
for (l = 0; l < hgt; l += hsize) {
hsize = hgt - l;
if (hsize > max_hsize) hsize = max_hsize;
for (c = 0; c < nchannel; c++) {
if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
sl = sl_c + c;
dl = dl_c + c;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j < hsize; j++) pbuff[j] = 0.0;
for (i = 0; i < wid; i++) {
sl0 = sl;
for (off = 0; off < (n - 4); off += 4) {
pk = k + off;
sp = sl0;
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
sp += 3*sll;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j < hsize; j += 2) {
p0 = p2; p1 = p3; p2 = p4;
p3 = sp[0];
p4 = sp[sll];
pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
sp += 2*sll;
}
sl0 += 4*sll;
}
pk = k + off;
sp = sl0;
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll];
dp = dl;
kh = n - off;
if (kh == 4) {
sp += 3*sll;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j <= (hsize - 2); j += 2) {
p0 = p2; p1 = p3; p2 = p4;
p3 = sp[0];
p4 = sp[sll];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1]);
dp[0 ] = FROM_S32(d0);
dp[dll] = FROM_S32(d1);
pbuff[j] = 0;
pbuff[j + 1] = 0;
sp += 2*sll;
dp += 2*dll;
}
if (j < hsize) {
p0 = p2; p1 = p3; p2 = p4;
p3 = sp[0];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]);
pbuff[j] = 0;
dp[0] = FROM_S32(d0);
}
} else if (kh == 3) {
sp += 2*sll;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j <= (hsize - 2); j += 2) {
p0 = p2; p1 = p3;
p2 = sp[0];
p3 = sp[sll];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]);
d1 = D2I(p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1]);
dp[0 ] = FROM_S32(d0);
dp[dll] = FROM_S32(d1);
pbuff[j] = 0;
pbuff[j + 1] = 0;
sp += 2*sll;
dp += 2*dll;
}
if (j < hsize) {
p0 = p2; p1 = p3;
p2 = sp[0];
d0 = D2I(p0*k0 + p1*k1 + p2*k2 + pbuff[j]);
pbuff[j] = 0;
dp[0] = FROM_S32(d0);
}
} else if (kh == 2) {
sp += sll;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j <= (hsize - 2); j += 2) {
p0 = p2;
p1 = sp[0];
p2 = sp[sll];
d0 = D2I(p0*k0 + p1*k1 + pbuff[j]);
d1 = D2I(p1*k0 + p2*k1 + pbuff[j + 1]);
dp[0 ] = FROM_S32(d0);
dp[dll] = FROM_S32(d1);
pbuff[j] = 0;
pbuff[j + 1] = 0;
sp += 2*sll;
dp += 2*dll;
}
if (j < hsize) {
p0 = p2;
p1 = sp[0];
d0 = D2I(p0*k0 + p1*k1 + pbuff[j]);
pbuff[j] = 0;
dp[0] = FROM_S32(d0);
}
} else /* if (kh == 1) */ {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = 0; j < hsize; j++) {
p0 = sp[0];
d0 = D2I(p0*k0 + pbuff[j]);
dp[0] = FROM_S32(d0);
pbuff[j] = 0;
sp += sll;
dp += dll;
}
}
sl += chan1;
dl += chan1;
}
}
sl_c += max_hsize*sll;
dl_c += max_hsize*dll;
}
if (pbuff != buff) mlib_free(pbuff);
return MLIB_SUCCESS;
}
mlib_status CONV_FUNC(MxN)(mlib_image *dst,
const mlib_image *src,
const mlib_d64 *ker,
mlib_s32 m,
mlib_s32 n,
mlib_s32 dm,
mlib_s32 dn,
mlib_s32 cmask)
{
DTYPE k0, k1, k2, k3, k4, k5, k6, *sp;
DTYPE p0, p1, p2, p3, p4, p5, p6, p7;
mlib_s32 l, off, kw;
DEF_VARS(DTYPE);
mlib_s32 chan2 = chan1 + chan1;
mlib_s32 chan3 = chan1 + chan2;
#ifdef TYPE_DOUBLE
const mlib_d64 *k = ker;
#else
mlib_f32 k_arr[MAX_NM], *k = k_arr;
if (n*m > MAX_NM) {
k = mlib_malloc(n*m*sizeof(mlib_f32));
if (k == NULL) return MLIB_FAILURE;
}
for (i = 0; i < n*m; i++) k[i] = (mlib_f32)ker[i];
#endif /* TYPE_DOUBLE */
if (m == 1) return mlib_ImageConv1xN(dst, src, k, n, dn, cmask);
wid -= (m - 1);
hgt -= (n - 1);
adr_dst += dn*dll + dm*chan1;
for (c = 0; c < chan1; c++) {
if (!(cmask & (1 << (chan1 - 1 - c)))) continue;
sl = adr_src + c;
dl = adr_dst + c;
for (j = 0; j < hgt; j++) {
const DTYPE *pk = k;
for (l = 0; l < n; l++) {
DTYPE *sp0 = sl + l*sll;
for (off = 0; off < m; off += kw, pk += kw, sp0 += chan1) {
kw = m - off;
if (kw > 2*MAX_KER) kw = MAX_KER; else
if (kw > MAX_KER) kw = kw/2;
p2 = sp0[0]; p3 = sp0[chan1]; p4 = sp0[chan2];
sp0 += chan3;
p5 = sp0[0]; p6 = sp0[chan1]; p7 = sp0[chan2];
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
dp = dl;
if (kw == 7) {
sp = sp0 += chan3;
if (pk == k) {
#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 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
sp += chan2;
dp += chan2;
}
} 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 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6;
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 6) {
sp = sp0 += chan2;
if (pk == k) {
#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 = sp[0]; p6 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
sp += chan2;
dp += chan2;
}
} 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 = sp[0]; p6 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5;
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 5) {
sp = sp0 += chan1;
if (pk == k) {
#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 = sp[0]; p5 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
sp += chan2;
dp += chan2;
}
} 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 = sp[0]; p5 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4;
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 4) {
sp = sp0;
if (pk == k) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3; p2 = p4;
p3 = sp[0]; p4 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3;
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3;
sp += chan2;
dp += chan2;
}
} 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 = sp[0]; p4 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
sp += chan2;
dp += chan2;
}
}
} else if (kw == 3) {
sp = sp0 -= chan1;
if (pk == k) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3;
p2 = sp[0]; p3 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1 + p2*k2;
dp[chan1] = p1*k0 + p2*k1 + p3*k2;
sp += chan2;
dp += chan2;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2; p1 = p3;
p2 = sp[0]; p3 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1 + p2*k2;
dp[chan1] += p1*k0 + p2*k1 + p3*k2;
sp += chan2;
dp += chan2;
}
}
} else { /* kw == 2 */
sp = sp0 -= chan2;
if (pk == k) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2;
p1 = sp[0]; p2 = sp[chan1];
dp[0 ] = p0*k0 + p1*k1;
dp[chan1] = p1*k0 + p2*k1;
sp += chan2;
dp += chan2;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = p2;
p1 = sp[0]; p2 = sp[chan1];
dp[0 ] += p0*k0 + p1*k1;
dp[chan1] += p1*k0 + p2*k1;
sp += chan2;
dp += chan2;
}
}
}
}
}
/* last pixels */
if (wid & 1) {
DTYPE *sp0 = sl + i*chan1, s = 0;
const DTYPE *pk = k;
mlib_s32 x;
for (l = 0; l < n; l++) {
DTYPE *sp = sp0 + l*sll;
for (x = 0; x < m; x++) s += sp[x*chan1] * (*pk++);
}
dp[0] = s;
}
/* next line */
sl += sll;
dl += dll;
}
}
#ifndef TYPE_DOUBLE
if (k != k_arr) mlib_free(k);
#endif /* TYPE_DOUBLE */
return MLIB_SUCCESS;
}
