// Simplified transform when only in[0], in[1] and in[4] are non-zero
static void TransformAC3(const int16_t* in, uint8_t* dst) {
const int a = in[0] + 4;
const int c4 = MUL(in[4], kC2);
const int d4 = MUL(in[4], kC1);
const int c1 = MUL(in[1], kC2);
const int d1 = MUL(in[1], kC1);
STORE2(0, a + d4, d1, c1);
STORE2(1, a + c4, d1, c1);
STORE2(2, a - c4, d1, c1);
STORE2(3, a - d4, d1, c1);
}
/*
* NAME: make_grammar()
* DESCRIPTION: create a pre-processed grammar string
*/
static string *make_grammar(rule *rgxlist, rule *strlist, rule *estrlist, rule *prodlist,
int nrgx, int nstr, int nestr, int nprod, long size)
{
int start, prod1;
string *gram;
char *p, *q;
rule *rl, *r;
rulesym *rs;
int n;
gram = str_new((char *) NULL, size);
/* header */
p = gram->text;
*p++ = GRAM_VERSION; /* version number */
STORE2(p, -1); p += 2; /* whitespace rule */
STORE2(p, -1); p += 2; /* nomatch rule */
STORE2(p, nrgx); p += 4; /* # regular expression rules */
STORE2(p, nstr); p += 2; /* # string rules */
STORE2(p, nestr); p += 2; /* # escaped string rules */
nprod++; /* +1 for start rule */
STORE2(p, nprod); /* # production rules */
n = nrgx + nstr + nestr + nprod;
prod1 = nrgx + nstr + nestr + 1;
q = p + 4 + ((n + nstr) << 1);
p = gram->text + size;
/* determine production rule offsets */
for (rl = prodlist; rl != (rule *) NULL; rl = rl->next) {
size -= (rl->num << 1) + rl->len + 2;
p -= (rl->num << 1) + rl->len + 2;
q -= 2; STORE2(q, size);
STORE2(p, rl->num);
rl->num = --n;
rl->len = size;
}
/* start rule offset */
size -= 6;
p -= 6;
q -= 2; STORE2(q, size);
--n;
start = size;
/* deal with strings */
for (rl = estrlist; rl != (rule *) NULL; rl = rl->next) {
size -= rl->symb->len + 1;
p -= rl->symb->len + 1;
q -= 2; STORE2(q, size);
rl->num = --n;
*p = rl->symb->len;
memcpy(p + 1, rl->symb->text, rl->symb->len);
}
for (rl = strlist; rl != (rule *) NULL; rl = rl->next) {
*--q = rl->symb->len;
q -= 3; STORE3(q, rl->len);
rl->num = --n;
}
/* deal with regexps */
nrgx = 0;
for (rl = rgxlist; rl != (rule *) NULL; rl = rl->next) {
size -= rl->num + rl->len + 2;
q -= 2; STORE2(q, size);
p = gram->text + size;
STORE2(p, rl->num);
rl->num = --n;
p += 2;
for (r = rl; r != (rule *) NULL; r = r->alt) {
if (r->u.rgx != (string *) NULL) {
*p++ = r->u.rgx->len;
memcpy(p, r->u.rgx->text, r->u.rgx->len);
p += r->u.rgx->len;
nrgx++;
} else {
/* nomatch */
STORE2(gram->text + 3, n);
}
}
if (rl->symb->len == 10 && strcmp(rl->symb->text, "whitespace") == 0) {
p = gram->text + 1;
STORE2(p, n);
}
}
p = gram->text + 7;
STORE2(p, nrgx); /* total regular expressions */
/* fill in production rules */
nprod = 1;
for (rl = prodlist; rl != (rule *) NULL; rl = rl->next) {
q = gram->text + rl->len + 2;
for (r = rl; r != (rule *) NULL; r = r->alt) {
p = q + 2;
n = 0;
for (rs = r->u.syms; rs != (rulesym *) NULL; rs = rs->next) {
STORE2(p, rs->rule->num); p += 2;
n++;
}
if (r->func != (string *) NULL) {
memcpy(p, r->func->text, r->func->len + 1);
p += r->func->len + 1;
}
*q++ = n;
*q = p - q - 1;
q = p;
nprod++;
}
}
/* start rule */
p = gram->text + start;
*p++ = 0;
*p++ = 1;
*p++ = 1;
*p++ = 2;
*p++ = prod1 >> 8;
*p = prod1;
p = gram->text + 15;
STORE2(p, nprod);
return gram;
}
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);
}
