void *mlib_ImageCreateRowTable(mlib_image *img)
{
mlib_u8 **rtable, *tline;
mlib_s32 i, im_height, im_stride;
if (img == NULL) return NULL;
if (img -> state) return img -> state;
im_height = mlib_ImageGetHeight(img);
im_stride = mlib_ImageGetStride(img);
tline = mlib_ImageGetData(img);
rtable = mlib_malloc((3 + im_height)*sizeof(mlib_u8 *));
if (rtable == NULL || tline == NULL) return NULL;
rtable[0] = 0;
rtable[1] = (mlib_u8*)((void **)rtable + 1);
rtable[2 + im_height] = (mlib_u8*)((void **)rtable + 1);
for (i = 0; i < im_height; i++) {
rtable[i+2] = tline;
tline += im_stride;
}
img -> state = ((void **)rtable + 2);
return img -> state;
}
mlib_image *mlib_ImageCreateStruct(mlib_type type,
mlib_s32 channels,
mlib_s32 width,
mlib_s32 height,
mlib_s32 stride,
const void *data)
{
mlib_image *image;
if (stride <= 0) {
return NULL;
}
image = (mlib_image *)mlib_malloc(sizeof(mlib_image));
if (image == NULL) {
return NULL;
}
if (mlib_ImageSet(image, type, channels, width, height, stride, data) == NULL) {
mlib_free(image);
image = NULL;
}
return image;
}
mlib_status
mlib_ImageMinFilter7x7_S16(
void *dst,
void *src,
mlib_s32 dlb,
mlib_s32 slb,
mlib_s32 wid,
mlib_s32 hgt)
#endif /* MAX_FILTER */
{
mlib_u8 *pbuff, *buff0, *buff1, *buff2, *buff3, *buff4, *buff5, *buffT;
mlib_u8 *sl, *sp0, *sp1, *sp2, *sp3, *sp4, *sp5, *sp6, *sp7, *dl;
__m64 *dp0, *dp1;
__m64 aa, bb, cc, dd, ee, ff, r0, r1;
__m64 g0, g1, g2, g3, g4, g5, g6, gg;
__m64 h0, h1, h2, h3, h4, h5, h6, hh;
__m64 e_mask;
mlib_s32 i, j, wid8, tail;
wid = (wid - KSIZE1) * SSIZE;
wid8 = (wid + 7) & ~7;
pbuff = mlib_malloc(KSIZE1 * wid8);
buff0 = pbuff;
buff1 = buff0 + wid8;
buff2 = buff1 + wid8;
buff3 = buff2 + wid8;
buff4 = buff3 + wid8;
buff5 = buff4 + wid8;
sl = (mlib_u8 *)src;
dl = (mlib_u8 *)dst + (KSIZE1 / 2) * (dlb + SSIZE);
tail = wid & 7;
e_mask = ((__m64 *) mlib_mask64_arr)[tail];
for (j = 0; j < 3; j++) {
sp0 = buff4;
sp1 = buff5;
sp6 = sl;
sp7 = sl + slb;
sl += 2 * slb;
for (i = 0; i < wid; i += 8) {
g0 = *(__m64 *) sp6;
g1 = *(__m64 *) (sp6 + SSIZE);
g2 = *(__m64 *) (sp6 + 2 * SSIZE);
g3 = *(__m64 *) (sp6 + 3 * SSIZE);
g4 = *(__m64 *) (sp6 + 4 * SSIZE);
g5 = *(__m64 *) (sp6 + 5 * SSIZE);
g6 = *(__m64 *) (sp6 + 6 * SSIZE);
h0 = *(__m64 *) sp7;
h1 = *(__m64 *) (sp7 + SSIZE);
h2 = *(__m64 *) (sp7 + 2 * SSIZE);
h3 = *(__m64 *) (sp7 + 3 * SSIZE);
h4 = *(__m64 *) (sp7 + 4 * SSIZE);
h5 = *(__m64 *) (sp7 + 5 * SSIZE);
h6 = *(__m64 *) (sp7 + 6 * SSIZE);
gg = C_COMP(g0, g1);
hh = C_COMP(h0, h1);
g2 = C_COMP(g2, g3);
h2 = C_COMP(h2, h3);
g4 = C_COMP(g4, g5);
h4 = C_COMP(h4, h5);
gg = C_COMP(gg, g2);
hh = C_COMP(hh, h2);
gg = C_COMP(gg, g4);
hh = C_COMP(hh, h4);
gg = C_COMP(gg, g6);
hh = C_COMP(hh, h6);
*(__m64 *) sp0 = gg;
*(__m64 *) sp1 = hh;
sp0 += 8;
sp1 += 8;
sp6 += 8;
sp7 += 8;
}
if (j < 2) {
buffT = buff0;
buff0 = buff2;
buff2 = buff4;
buff4 = buffT;
buffT = buff1;
buff1 = buff3;
buff3 = buff5;
buff5 = buffT;
}
}
for (j = 0; j <= (hgt - KSIZE1 - 2); j += 2) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff0;
sp1 = buff1;
sp2 = buff2;
sp3 = buff3;
sp4 = buff4;
sp5 = buff5;
sp6 = sl;
sp7 = sl + slb;
/*
* line0: aa
* line1: bb
* line2: cc
* line3: dd
* line4: ee
* line5: ff
* line4: g0 g1 g2 g3 g4 g5 g6
* line5: h0 h1 h2 h3 h4 h5 h6
*/
for (i = 0; i <= wid - 8; i += 8) {
g0 = *(__m64 *) sp6;
g1 = *(__m64 *) (sp6 + SSIZE);
g2 = *(__m64 *) (sp6 + 2 * SSIZE);
g3 = *(__m64 *) (sp6 + 3 * SSIZE);
g4 = *(__m64 *) (sp6 + 4 * SSIZE);
g5 = *(__m64 *) (sp6 + 5 * SSIZE);
g6 = *(__m64 *) (sp6 + 6 * SSIZE);
h0 = *(__m64 *) sp7;
h1 = *(__m64 *) (sp7 + SSIZE);
h2 = *(__m64 *) (sp7 + 2 * SSIZE);
h3 = *(__m64 *) (sp7 + 3 * SSIZE);
h4 = *(__m64 *) (sp7 + 4 * SSIZE);
h5 = *(__m64 *) (sp7 + 5 * SSIZE);
h6 = *(__m64 *) (sp7 + 6 * SSIZE);
gg = C_COMP(g0, g1);
hh = C_COMP(h0, h1);
g2 = C_COMP(g2, g3);
h2 = C_COMP(h2, h3);
g4 = C_COMP(g4, g5);
h4 = C_COMP(h4, h5);
gg = C_COMP(gg, g2);
hh = C_COMP(hh, h2);
gg = C_COMP(gg, g4);
hh = C_COMP(hh, h4);
gg = C_COMP(gg, g6);
hh = C_COMP(hh, h6);
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
ee = *(__m64 *) sp4;
ff = *(__m64 *) sp5;
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
ff = C_COMP(ff, gg);
bb = C_COMP(bb, dd);
bb = C_COMP(bb, ff);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, hh);
*(__m64 *) sp0 = gg;
*(__m64 *) sp1 = hh;
(*dp0++) = r0;
(*dp1++) = r1;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
sp4 += 8;
sp5 += 8;
sp6 += 8;
sp7 += 8;
}
if (tail) {
g0 = *(__m64 *) sp6;
g1 = *(__m64 *) (sp6 + SSIZE);
g2 = *(__m64 *) (sp6 + 2 * SSIZE);
g3 = *(__m64 *) (sp6 + 3 * SSIZE);
g4 = *(__m64 *) (sp6 + 4 * SSIZE);
g5 = *(__m64 *) (sp6 + 5 * SSIZE);
g6 = *(__m64 *) (sp6 + 6 * SSIZE);
h0 = *(__m64 *) sp7;
h1 = *(__m64 *) (sp7 + SSIZE);
h2 = *(__m64 *) (sp7 + 2 * SSIZE);
h3 = *(__m64 *) (sp7 + 3 * SSIZE);
h4 = *(__m64 *) (sp7 + 4 * SSIZE);
h5 = *(__m64 *) (sp7 + 5 * SSIZE);
h6 = *(__m64 *) (sp7 + 6 * SSIZE);
gg = C_COMP(g0, g1);
hh = C_COMP(h0, h1);
g2 = C_COMP(g2, g3);
h2 = C_COMP(h2, h3);
g4 = C_COMP(g4, g5);
h4 = C_COMP(h4, h5);
gg = C_COMP(gg, g2);
hh = C_COMP(hh, h2);
gg = C_COMP(gg, g4);
hh = C_COMP(hh, h4);
gg = C_COMP(gg, g6);
hh = C_COMP(hh, h6);
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
ee = *(__m64 *) sp4;
ff = *(__m64 *) sp5;
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
ff = C_COMP(ff, gg);
bb = C_COMP(bb, dd);
bb = C_COMP(bb, ff);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, hh);
*(__m64 *) sp0 = gg;
*(__m64 *) sp1 = hh;
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
*dp1 =
_mm_or_si64(_mm_and_si64(e_mask, r1),
_mm_andnot_si64(e_mask, *dp1));
}
buffT = buff0;
buff0 = buff2;
buff2 = buff4;
buff4 = buffT;
buffT = buff1;
buff1 = buff3;
buff3 = buff5;
buff5 = buffT;
sl += 2 * slb;
dl += 2 * dlb;
}
/* last line */
if (j == (hgt - KSIZE1 - 1)) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff0;
sp1 = buff1;
sp2 = buff2;
sp3 = buff3;
sp4 = buff4;
sp5 = buff5;
sp6 = sl;
for (i = 0; i <= wid - 8; i += 8) {
g0 = *(__m64 *) sp6;
g1 = *(__m64 *) (sp6 + SSIZE);
g2 = *(__m64 *) (sp6 + 2 * SSIZE);
g3 = *(__m64 *) (sp6 + 3 * SSIZE);
g4 = *(__m64 *) (sp6 + 4 * SSIZE);
g5 = *(__m64 *) (sp6 + 5 * SSIZE);
g6 = *(__m64 *) (sp6 + 6 * SSIZE);
gg = C_COMP(g0, g1);
g2 = C_COMP(g2, g3);
g4 = C_COMP(g4, g5);
gg = C_COMP(gg, g2);
gg = C_COMP(gg, g4);
gg = C_COMP(gg, g6);
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
ee = *(__m64 *) sp4;
ff = *(__m64 *) sp5;
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
ff = C_COMP(ff, gg);
bb = C_COMP(bb, dd);
bb = C_COMP(bb, ff);
r0 = C_COMP(aa, bb);
(*dp0++) = r0;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
sp4 += 8;
sp5 += 8;
sp6 += 8;
}
if (tail) {
g0 = *(__m64 *) sp6;
g1 = *(__m64 *) (sp6 + SSIZE);
g2 = *(__m64 *) (sp6 + 2 * SSIZE);
g3 = *(__m64 *) (sp6 + 3 * SSIZE);
g4 = *(__m64 *) (sp6 + 4 * SSIZE);
g5 = *(__m64 *) (sp6 + 5 * SSIZE);
g6 = *(__m64 *) (sp6 + 6 * SSIZE);
gg = C_COMP(g0, g1);
g2 = C_COMP(g2, g3);
g4 = C_COMP(g4, g5);
gg = C_COMP(gg, g2);
gg = C_COMP(gg, g4);
gg = C_COMP(gg, g6);
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
ee = *(__m64 *) sp4;
ff = *(__m64 *) sp5;
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
ff = C_COMP(ff, gg);
bb = C_COMP(bb, dd);
bb = C_COMP(bb, ff);
r0 = C_COMP(aa, bb);
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
}
}
_mm_empty();
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
mlib_status
mlib_m_conv5x5_u16nw_2(
mlib_image *dst,
mlib_image *src,
mlib_s32 *kern,
mlib_s32 scalef_expon)
{
__m64 *pbuff, *buff_arr[20], **pbuff_arr = buff_arr;
__m64 *buff0, *buff1, *buff2, *buff3;
GET_SRC_DST_PARAMETERS(mlib_s16);
__m64 ker[5][5];
__m64 d0, d1, d2, aa, bb, rr, tmpa, tmpb, ker_off, mask8000;
__m64 prev0h, prev1h, prev2h, prev3h, sum0h, sum1h, sum2h, sum3h, sum4h,
tmph;
__m64 prev0l, prev1l, prev2l, prev3l, sum0l, sum1l, sum2l, sum3l, sum4l,
tmpl;
__m64 *sp, *dp;
mlib_s32 shift, ind, ker_sum = 0;
mlib_s32 row, wid4, i, j;
width -= 4;
height -= 4;
width *= NCHAN;
dl += 2 * (dll + NCHAN);
wid4 = (width + 7) / 4;
pbuff = mlib_malloc(sizeof (__m64) * 20 * wid4);
GET_KERN();
for (i = 0; i < 10; i++) {
buff_arr[i] = pbuff + i * 2 * wid4;
}
ind = 0;
for (j = 1; j <= 4; j++) {
buff0 = buff_arr[ind];
buff1 = buff_arr[ind + 1];
buff2 = buff_arr[ind + 2];
buff3 = buff_arr[ind + 3];
sp = (__m64 *) sl;
d1 = (*sp++);
d1 = _mm_xor_si64(d1, mask8000);
d2 = (*sp++);
d2 = _mm_xor_si64(d2, mask8000);
for (i = 0; i < wid4; i++) {
PREP_5x5();
}
sl += sll;
ind += j;
}
for (row = 0; row < height; row++) {
sp = (__m64 *) sl;
dp = (__m64 *) dl;
buff0 = pbuff_arr[0];
buff1 = pbuff_arr[2];
buff2 = pbuff_arr[5];
buff3 = pbuff_arr[9];
d1 = (*sp++);
d1 = _mm_xor_si64(d1, mask8000);
d2 = (*sp++);
d2 = _mm_xor_si64(d2, mask8000);
for (i = 0; i < width / 4; i++) {
CONV_5x5(hi, i);
dp[i] = rr;
}
if (width & 3) {
__m64 mask =
((__m64 *) mlib_mask64_arr)[2 * (width & 3)];
CONV_5x5(hi, i);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, rr),
_mm_andnot_si64(mask, dp[i]));
}
ind = (pbuff_arr == buff_arr) ? 10 : -10;
pbuff_arr[ind + 0] = pbuff_arr[1];
pbuff_arr[ind + 1] = pbuff_arr[3];
pbuff_arr[ind + 2] = pbuff_arr[4];
pbuff_arr[ind + 3] = pbuff_arr[6];
pbuff_arr[ind + 4] = pbuff_arr[7];
pbuff_arr[ind + 5] = pbuff_arr[8];
pbuff_arr[ind + 6] = pbuff_arr[0];
pbuff_arr[ind + 7] = pbuff_arr[2];
pbuff_arr[ind + 8] = pbuff_arr[5];
pbuff_arr[ind + 9] = pbuff_arr[9];
pbuff_arr += ind;
sl += sll;
dl += dll;
}
_mm_empty();
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
mlib_status
mlib_m_sconv5x5_8nw_2(
mlib_image *dst,
mlib_image *src,
mlib_s32 *hkernel,
mlib_s32 *vkernel,
mlib_s32 scalef_expon)
{
__m64 *pbuff, *buff_arr[5];
__m64 *buff0, *buff1, *buff2, *buff3, *buff4, *buffT;
GET_SRC_DST_PARAMETERS(mlib_u8);
__m64 hker0, hker1, hker2, hker3, hker4;
__m64 vker0, vker1, vker2, vker3, vker4;
__m64 s0, d0, d1, d2, prev0;
__m64 sum0, sum1, sum2, sum3, sum4, aa, bb, res_hi, res_lo;
__m64 zero = _m_zero;
mlib_s32 shift, ind;
mlib_s32 *sp;
mlib_s32 row, wid4, i, j;
width -= 4;
height -= 4;
width *= NCHAN;
dl += 2 * (dll + NCHAN);
wid4 = 2 * ((width + 7) / 8);
pbuff = mlib_malloc(sizeof (__m64) * 5 * wid4);
GET_KERN();
for (i = 0; i < 5; i++) {
buff_arr[i] = pbuff + i * wid4;
}
for (j = 0; j < 4; j++) {
buff4 = buff_arr[j];
sp = (mlib_s32 *)sl;
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d1, lo);
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d2, lo);
for (i = 0; i < wid4; i++) {
*(mlib_s32 *)&s0 = sp[i];
PREP_5x5(lo, i);
}
sl += sll;
ind++;
}
buff0 = buff_arr[0];
buff1 = buff_arr[1];
buff2 = buff_arr[2];
buff3 = buff_arr[3];
buff4 = buff_arr[4];
for (row = 0; row < height; row++) {
__m64 *sp = (__m64 *) sl;
__m64 *dp = (__m64 *) dl;
s0 = (*sp++);
UNPACK_SRC(d1, lo);
UNPACK_SRC(d2, hi);
for (i = 0; i < width / 8; i++) {
s0 = sp[i];
CONV_5x5(lo, 2 * i);
CONV_5x5(hi, 2 * i + 1);
dp[i] = _mm_packs_pu16(res_lo, res_hi);
}
if (width & 7) {
__m64 mask = ((__m64 *) mlib_mask64_arr)[width & 7];
s0 = sp[i];
CONV_5x5(lo, 2 * i);
CONV_5x5(hi, 2 * i + 1);
res_hi = _mm_packs_pu16(res_lo, res_hi);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, res_hi),
_mm_andnot_si64(mask, dp[i]));
}
buffT = buff0;
buff0 = buff1;
buff1 = buff2;
buff2 = buff3;
buff3 = buff4;
buff4 = buffT;
sl += sll;
dl += dll;
}
_mm_empty();
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
mlib_status mlib_AffineEdges(mlib_affine_param *param,
const mlib_image *dst,
const mlib_image *src,
void *buff_lcl,
mlib_s32 buff_size,
mlib_s32 kw,
mlib_s32 kh,
mlib_s32 kw1,
mlib_s32 kh1,
mlib_edge edge,
const mlib_d64 *mtx,
mlib_s32 shiftx,
mlib_s32 shifty)
{
mlib_u8 *buff = buff_lcl;
mlib_u8 **lineAddr = param->lineAddr;
mlib_s32 srcWidth, dstWidth, srcHeight, dstHeight, srcYStride, dstYStride;
mlib_s32 *leftEdges, *rightEdges, *xStarts, *yStarts, bsize0, bsize1 = 0;
mlib_u8 *srcData, *dstData;
mlib_u8 *paddings;
void *warp_tbl = NULL;
mlib_s32 yStart = 0, yFinish = -1, dX, dY;
mlib_d64 xClip, yClip, wClip, hClip;
mlib_d64 delta = 0.;
mlib_d64 minX, minY, maxX, maxY;
mlib_d64 coords[4][2];
mlib_d64 a = mtx[0], b = mtx[1], tx = mtx[2], c = mtx[3], d = mtx[4], ty = mtx[5];
mlib_d64 a2, b2, tx2, c2, d2, ty2;
mlib_d64 dx, dy, div;
mlib_s32 sdx, sdy;
mlib_d64 dTop;
mlib_d64 val0;
mlib_s32 top, bot;
mlib_s32 topIdx, max_xsize = 0;
mlib_s32 i, j, t;
srcData = mlib_ImageGetData(src);
dstData = mlib_ImageGetData(dst);
srcWidth = mlib_ImageGetWidth(src);
srcHeight = mlib_ImageGetHeight(src);
dstWidth = mlib_ImageGetWidth(dst);
dstHeight = mlib_ImageGetHeight(dst);
srcYStride = mlib_ImageGetStride(src);
dstYStride = mlib_ImageGetStride(dst);
paddings = mlib_ImageGetPaddings(src);
if (srcWidth >= (1 << 15) || srcHeight >= (1 << 15)) {
return MLIB_FAILURE;
}
div = a * d - b * c;
if (div == 0.0) {
return MLIB_FAILURE;
}
bsize0 = (dstHeight * sizeof(mlib_s32) + 7) & ~7;
if (lineAddr == NULL) {
bsize1 = ((srcHeight + 4 * kh) * sizeof(mlib_u8 *) + 7) & ~7;
}
param->buff_malloc = NULL;
if ((4 * bsize0 + bsize1) > buff_size) {
buff = param->buff_malloc = mlib_malloc(4 * bsize0 + bsize1);
if (buff == NULL)
return MLIB_FAILURE;
}
leftEdges = (mlib_s32 *) (buff);
rightEdges = (mlib_s32 *) (buff += bsize0);
xStarts = (mlib_s32 *) (buff += bsize0);
yStarts = (mlib_s32 *) (buff += bsize0);
if (lineAddr == NULL) {
mlib_u8 *srcLinePtr = srcData;
lineAddr = (mlib_u8 **) (buff += bsize0);
for (i = 0; i < 2 * kh; i++)
lineAddr[i] = srcLinePtr;
lineAddr += 2 * kh;
for (i = 0; i < srcHeight - 1; i++) {
lineAddr[i] = srcLinePtr;
srcLinePtr += srcYStride;
}
for (i = srcHeight - 1; i < srcHeight + 2 * kh; i++)
lineAddr[i] = srcLinePtr;
}
if ((mlib_s32) edge < 0) { /* process edges */
minX = 0;
minY = 0;
maxX = srcWidth;
maxY = srcHeight;
}
else {
if (kw > 1)
delta = -0.5; /* for MLIB_NEAREST filter delta = 0. */
minX = (kw1 - delta);
minY = (kh1 - delta);
maxX = srcWidth - ((kw - 1) - (kw1 - delta));
maxY = srcHeight - ((kh - 1) - (kh1 - delta));
if (edge == MLIB_EDGE_SRC_PADDED) {
if (minX < paddings[0])
minX = paddings[0];
if (minY < paddings[1])
minY = paddings[1];
if (maxX > (srcWidth - paddings[2]))
maxX = srcWidth - paddings[2];
if (maxY > (srcHeight - paddings[3]))
maxY = srcHeight - paddings[3];
}
}
xClip = minX;
yClip = minY;
wClip = maxX;
hClip = maxY;
/*
* STORE_PARAM(param, src);
* STORE_PARAM(param, dst);
*/
param->src = (void *)src;
param->dst = (void *)dst;
STORE_PARAM(param, lineAddr);
STORE_PARAM(param, dstData);
STORE_PARAM(param, srcYStride);
STORE_PARAM(param, dstYStride);
STORE_PARAM(param, leftEdges);
STORE_PARAM(param, rightEdges);
STORE_PARAM(param, xStarts);
STORE_PARAM(param, yStarts);
STORE_PARAM(param, max_xsize);
STORE_PARAM(param, yStart);
STORE_PARAM(param, yFinish);
STORE_PARAM(param, warp_tbl);
if ((xClip >= wClip) || (yClip >= hClip)) {
return MLIB_SUCCESS;
}
a2 = d;
b2 = -b;
tx2 = (-d * tx + b * ty);
c2 = -c;
d2 = a;
ty2 = (c * tx - a * ty);
dx = a2;
dy = c2;
tx -= 0.5;
ty -= 0.5;
coords[0][0] = xClip * a + yClip * b + tx;
coords[0][1] = xClip * c + yClip * d + ty;
coords[2][0] = wClip * a + hClip * b + tx;
coords[2][1] = wClip * c + hClip * d + ty;
if (div > 0) {
coords[1][0] = wClip * a + yClip * b + tx;
coords[1][1] = wClip * c + yClip * d + ty;
coords[3][0] = xClip * a + hClip * b + tx;
coords[3][1] = xClip * c + hClip * d + ty;
}
else {
coords[3][0] = wClip * a + yClip * b + tx;
coords[3][1] = wClip * c + yClip * d + ty;
coords[1][0] = xClip * a + hClip * b + tx;
coords[1][1] = xClip * c + hClip * d + ty;
}
topIdx = 0;
for (i = 1; i < 4; i++) {
if (coords[i][1] < coords[topIdx][1])
topIdx = i;
}
dTop = coords[topIdx][1];
val0 = dTop;
SAT32(top);
bot = -1;
if (top >= dstHeight) {
return MLIB_SUCCESS;
}
if (dTop >= 0.0) {
mlib_d64 xLeft, xRight, x;
mlib_s32 nextIdx;
if (dTop == top) {
xLeft = coords[topIdx][0];
xRight = coords[topIdx][0];
nextIdx = (topIdx + 1) & 0x3;
if (dTop == coords[nextIdx][1]) {
x = coords[nextIdx][0];
xLeft = (xLeft <= x) ? xLeft : x;
xRight = (xRight >= x) ? xRight : x;
}
nextIdx = (topIdx - 1) & 0x3;
if (dTop == coords[nextIdx][1]) {
x = coords[nextIdx][0];
xLeft = (xLeft <= x) ? xLeft : x;
xRight = (xRight >= x) ? xRight : x;
}
val0 = xLeft;
SAT32(t);
leftEdges[top] = (t >= xLeft) ? t : ++t;
if (xLeft >= MLIB_S32_MAX)
leftEdges[top] = MLIB_S32_MAX;
val0 = xRight;
SAT32(rightEdges[top]);
}
else
top++;
}
else
top = 0;
for (i = 0; i < 2; i++) {
mlib_d64 dY1 = coords[(topIdx - i) & 0x3][1];
mlib_d64 dX1 = coords[(topIdx - i) & 0x3][0];
mlib_d64 dY2 = coords[(topIdx - i - 1) & 0x3][1];
mlib_d64 dX2 = coords[(topIdx - i - 1) & 0x3][0];
mlib_d64 x = dX1, slope = (dX2 - dX1) / (dY2 - dY1);
mlib_s32 y1;
mlib_s32 y2;
if (dY1 == dY2)
continue;
if (dY1 < 0.0)
y1 = 0;
else {
val0 = dY1 + 1;
SAT32(y1);
}
val0 = dY2;
SAT32(y2);
if (y2 >= dstHeight)
y2 = (mlib_s32) (dstHeight - 1);
x += slope * (y1 - dY1);
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = y1; j <= y2; j++) {
val0 = x;
SAT32(t);
leftEdges[j] = (t >= x) ? t : ++t;
if (x >= MLIB_S32_MAX)
leftEdges[j] = MLIB_S32_MAX;
x += slope;
}
}
for (i = 0; i < 2; i++) {
mlib_d64 dY1 = coords[(topIdx + i) & 0x3][1];
mlib_d64 dX1 = coords[(topIdx + i) & 0x3][0];
mlib_d64 dY2 = coords[(topIdx + i + 1) & 0x3][1];
mlib_d64 dX2 = coords[(topIdx + i + 1) & 0x3][0];
mlib_d64 x = dX1, slope = (dX2 - dX1) / (dY2 - dY1);
mlib_s32 y1;
mlib_s32 y2;
if (dY1 == dY2)
continue;
if (dY1 < 0.0)
y1 = 0;
else {
val0 = dY1 + 1;
SAT32(y1);
}
val0 = dY2;
SAT32(y2);
if (y2 >= dstHeight)
y2 = (mlib_s32) (dstHeight - 1);
x += slope * (y1 - dY1);
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (j = y1; j <= y2; j++) {
val0 = x;
SAT32(rightEdges[j]);
x += slope;
}
bot = y2;
}
{
mlib_d64 dxCl = xClip * div;
mlib_d64 dyCl = yClip * div;
mlib_d64 dwCl = wClip * div;
mlib_d64 dhCl = hClip * div;
mlib_s32 xCl = (mlib_s32) (xClip + delta);
mlib_s32 yCl = (mlib_s32) (yClip + delta);
mlib_s32 wCl = (mlib_s32) (wClip + delta);
mlib_s32 hCl = (mlib_s32) (hClip + delta);
/*
* mlib_s32 xCl = (mlib_s32)(xClip + delta);
* mlib_s32 yCl = (mlib_s32)(yClip + delta);
* mlib_s32 wCl = (mlib_s32)(wClip);
* mlib_s32 hCl = (mlib_s32)(hClip);
*/
if (edge == MLIB_EDGE_SRC_PADDED) {
xCl = kw1;
yCl = kh1;
wCl = (mlib_s32) (srcWidth - ((kw - 1) - kw1));
hCl = (mlib_s32) (srcHeight - ((kh - 1) - kh1));
}
div = 1.0 / div;
sdx = (mlib_s32) (a2 * div * (1 << shiftx));
sdy = (mlib_s32) (c2 * div * (1 << shifty));
if (div > 0) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = top; i <= bot; i++) {
mlib_s32 xLeft = leftEdges[i];
mlib_s32 xRight = rightEdges[i];
mlib_s32 xs, ys, x_e, y_e, x_s, y_s;
mlib_d64 dxs, dys, dxe, dye;
mlib_d64 xl, ii, xr;
xLeft = (xLeft < 0) ? 0 : xLeft;
xRight = (xRight >= dstWidth) ? (mlib_s32) (dstWidth - 1) : xRight;
xl = xLeft + 0.5;
ii = i + 0.5;
xr = xRight + 0.5;
dxs = xl * a2 + ii * b2 + tx2;
dys = xl * c2 + ii * d2 + ty2;
if ((dxs < dxCl) || (dxs >= dwCl) || (dys < dyCl) || (dys >= dhCl)) {
dxs += dx;
dys += dy;
xLeft++;
if ((dxs < dxCl) || (dxs >= dwCl) || (dys < dyCl) || (dys >= dhCl))
xRight = -1;
}
dxe = xr * a2 + ii * b2 + tx2;
dye = xr * c2 + ii * d2 + ty2;
if ((dxe < dxCl) || (dxe >= dwCl) || (dye < dyCl) || (dye >= dhCl)) {
dxe -= dx;
dye -= dy;
xRight--;
if ((dxe < dxCl) || (dxe >= dwCl) || (dye < dyCl) || (dye >= dhCl))
xRight = -1;
}
xs = (mlib_s32) ((dxs * div + delta) * (1 << shiftx));
x_s = xs >> shiftx;
ys = (mlib_s32) ((dys * div + delta) * (1 << shifty));
y_s = ys >> shifty;
if (x_s < xCl)
xs = (xCl << shiftx);
else if (x_s >= wCl)
xs = ((wCl << shiftx) - 1);
if (y_s < yCl)
ys = (yCl << shifty);
else if (y_s >= hCl)
ys = ((hCl << shifty) - 1);
if (xRight >= xLeft) {
x_e = ((xRight - xLeft) * sdx + xs) >> shiftx;
y_e = ((xRight - xLeft) * sdy + ys) >> shifty;
if ((x_e < xCl) || (x_e >= wCl)) {
if (sdx > 0)
sdx -= 1;
else
sdx += 1;
}
if ((y_e < yCl) || (y_e >= hCl)) {
if (sdy > 0)
sdy -= 1;
else
sdy += 1;
}
}
leftEdges[i] = xLeft;
rightEdges[i] = xRight;
xStarts[i] = xs;
yStarts[i] = ys;
if ((xRight - xLeft + 1) > max_xsize)
max_xsize = (xRight - xLeft + 1);
}
}
else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = top; i <= bot; i++) {
mlib_status
mlib_m_conv5x5_8nw_4(
mlib_image *dst,
mlib_image *src,
mlib_s32 *kern,
mlib_s32 scalef_expon)
{
__m64 *pbuff, *buff_arr[20], **pbuff_arr = buff_arr;
__m64 *buff0, *buff1, *buff2, *buff3;
GET_SRC_DST_PARAMETERS(mlib_u8);
__m64 ker[5][5];
__m64 s0, d0, d1, d2, d3, d4, prev0, prev1, prev2, prev3, aa, bb, cc;
__m64 sum0, sum1, sum2, sum3, sum4, res_hi, res_lo;
__m64 zero = _m_zero;
mlib_s32 shift, ind;
mlib_s32 *sp;
mlib_s32 row, wid4, i, j;
width -= (KSIZE - 1);
height -= (KSIZE - 1);
width *= NCHAN;
dl += ((KSIZE - 1) / 2) * (dll + NCHAN);
wid4 = (width + 7) / 4;
pbuff = mlib_malloc(sizeof (__m64) * 10 * wid4);
GET_KERN();
for (i = 0; i < 10; i++) {
buff_arr[i] = pbuff + i * wid4;
}
ind = 0;
for (j = 1; j <= 4; j++) {
buff0 = buff_arr[ind];
buff1 = buff_arr[ind + 1];
buff2 = buff_arr[ind + 2];
buff3 = buff_arr[ind + 3];
sp = (mlib_s32 *)sl;
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d1, lo);
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d2, lo);
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d3, lo);
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d4, lo);
for (i = 0; i < wid4; i++) {
*(mlib_s32 *)&s0 = sp[i];
PREP_5x5();
}
sl += sll;
ind += j;
}
for (row = 0; row < height; row++) {
__m64 *sp = (__m64 *) sl;
__m64 *dp = (__m64 *) dl;
buff0 = pbuff_arr[0];
buff1 = pbuff_arr[2];
buff2 = pbuff_arr[5];
buff3 = pbuff_arr[9];
s0 = (*sp++);
UNPACK_SRC(d1, lo);
UNPACK_SRC(d2, hi);
s0 = (*sp++);
UNPACK_SRC(d3, lo);
UNPACK_SRC(d4, hi);
for (i = 0; i < width / 8; i++) {
s0 = sp[i];
CONV_5x5(lo, 2 * i);
CONV_5x5(hi, 2 * i + 1);
dp[i] = _mm_packs_pu16(res_lo, res_hi);
}
if (width & 7) {
__m64 mask;
mask = ((__m64 *) mlib_mask64_arr)[width & 7];
s0 = sp[i];
CONV_5x5(lo, 2 * i);
CONV_5x5(hi, 2 * i + 1);
res_hi = _mm_packs_pu16(res_lo, res_hi);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, res_hi),
_mm_andnot_si64(mask, dp[i]));
}
ind = (pbuff_arr == buff_arr) ? 10 : -10;
pbuff_arr[ind + 0] = pbuff_arr[1];
pbuff_arr[ind + 1] = pbuff_arr[3];
pbuff_arr[ind + 2] = pbuff_arr[4];
pbuff_arr[ind + 3] = pbuff_arr[6];
pbuff_arr[ind + 4] = pbuff_arr[7];
pbuff_arr[ind + 5] = pbuff_arr[8];
pbuff_arr[ind + 6] = pbuff_arr[0];
pbuff_arr[ind + 7] = pbuff_arr[2];
pbuff_arr[ind + 8] = pbuff_arr[5];
pbuff_arr[ind + 9] = pbuff_arr[9];
pbuff_arr += ind;
sl += sll;
dl += dll;
}
_mm_empty();
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;
}
mlib_status CONV_FUNC_I(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)
{
mlib_s32 buff[BUFF_SIZE], *buffd = buff;
mlib_s32 l, off, kw;
mlib_s32 d0, d1, shift1, shift2;
mlib_s32 k0, k1, k2, k3, k4, k5, k6;
mlib_s32 p0, p1, p2, p3, p4, p5, p6, p7;
DTYPE *adr_src, *sl, *sp = NULL;
DTYPE *adr_dst, *dl, *dp = NULL;
mlib_s32 wid, hgt, sll, dll;
mlib_s32 nchannel, chan1;
mlib_s32 i, j, c;
mlib_s32 chan2;
mlib_s32 k_locl[MAX_N*MAX_N], *k = k_locl;
GET_SRC_DST_PARAMETERS(DTYPE);
#if IMG_TYPE != 1
shift1 = 16;
#else
shift1 = 8;
#endif /* IMG_TYPE != 1 */
shift2 = scale - shift1;
chan1 = nchannel;
chan2 = chan1 + chan1;
wid -= (m - 1);
hgt -= (n - 1);
adr_dst += dn*dll + dm*nchannel;
if (wid > BUFF_SIZE) {
buffd = mlib_malloc(sizeof(mlib_s32)*wid);
if (buffd == NULL) return MLIB_FAILURE;
}
if (m*n > MAX_N*MAX_N) {
k = mlib_malloc(sizeof(mlib_s32)*(m*n));
if (k == NULL) {
if (buffd != buff) mlib_free(buffd);
return MLIB_FAILURE;
}
}
for (i = 0; i < m*n; i++) {
k[i] = kernel[i] >> shift1;
}
for (c = 0; c < nchannel; c++) {
if (!(cmask & (1 << (nchannel - 1 - c)))) continue;
sl = adr_src + c;
dl = adr_dst + c;
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i < wid; i++) buffd[i] = 0;
for (j = 0; j < hgt; j++) {
mlib_s32 *pk = k;
for (l = 0; l < n; l++) {
DTYPE *sp0 = sl + l*sll;
for (off = 0; off < m;) {
sp = sp0 + off*chan1;
dp = dl;
kw = m - off;
if (kw > 2*MAX_KER) kw = MAX_KER; else
if (kw > MAX_KER) kw = kw/2;
off += kw;
p2 = sp[0]; p3 = sp[chan1]; p4 = sp[chan2];
p5 = sp[chan2 + chan1]; p6 = sp[chan2 + chan2]; p7 = sp[5*chan1];
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3];
k4 = pk[4]; k5 = pk[5]; k6 = pk[6];
pk += kw;
sp += (kw - 1)*chan1;
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 = sp[0];
p7 = sp[chan1];
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;
sp += 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 = p7;
p6 = sp[0];
p7 = sp[chan1];
d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 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 = sp[0];
p6 = sp[chan1];
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;
sp += 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];
d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 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 = sp[0];
p5 = sp[chan1];
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;
sp += 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];
d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 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 = sp[0];
p4 = sp[chan1];
buffd[i ] += p0*k0 + p1*k1 + p2*k2 + p3*k3;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3;
sp += 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];
d0 = (p0*k0 + p1*k1 + p2*k2 + p3*k3 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + p3*k2 + p4*k3 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 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 = sp[0];
p3 = sp[chan1];
buffd[i ] += p0*k0 + p1*k1 + p2*k2;
buffd[i + 1] += p1*k0 + p2*k1 + p3*k2;
sp += 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];
d0 = (p0*k0 + p1*k1 + p2*k2 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + p3*k2 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 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 = sp[0];
p2 = sp[chan1];
buffd[i ] += p0*k0 + p1*k1;
buffd[i + 1] += p1*k0 + p2*k1;
sp += 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];
d0 = (p0*k0 + p1*k1 + buffd[i ]);
d1 = (p1*k0 + p2*k1 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 0;
sp += chan2;
dp += chan2;
}
}
} else /*if (kw == 1)*/ {
if (l < (n - 1) || off < m) {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = sp[0];
p1 = sp[chan1];
buffd[i ] += p0*k0;
buffd[i + 1] += p1*k0;
sp += chan2;
}
} else {
#ifdef __SUNPRO_C
#pragma pipeloop(0)
#endif /* __SUNPRO_C */
for (i = 0; i <= (wid - 2); i += 2) {
p0 = sp[0];
p1 = sp[chan1];
d0 = (p0*k0 + buffd[i ]);
d1 = (p1*k0 + buffd[i + 1]);
STORE_RES(dp[0 ], d0);
STORE_RES(dp[chan1], d1);
buffd[i ] = 0;
buffd[i + 1] = 0;
sp += chan2;
dp += chan2;
}
}
}
}
}
/* last pixels */
for (; i < wid; i++) {
mlib_s32 *pk = k, s = 0;
mlib_s32 x;
for (l = 0; l < n; l++) {
sp = sl + l*sll + i*chan1;
for (x = 0; x < m; x++) {
s += sp[0] * pk[0];
sp += chan1;
pk ++;
}
}
STORE_RES(dp[0], s);
sp += chan1;
dp += chan1;
}
sl += sll;
dl += dll;
}
}
if (buffd != buff) mlib_free(buffd);
if (k != k_locl) mlib_free(k);
return MLIB_SUCCESS;
}
mlib_status mlib_ImageAffineEdgeExtend_BL(mlib_affine_param *param,
mlib_affine_param *param_e,
const void *colormap)
{
GET_EDGE_PARAMS();
mlib_d64 scale = 1.0 / (mlib_d64) MLIB_PREC;
mlib_s32 xDelta, yDelta, xFlag, yFlag;
mlib_d64 t, u, pix0;
mlib_d64 a00, a01, a10, a11;
if (colormap != NULL) {
mlib_s32 max_xsize = param_e->max_xsize;
mlib_type ltype = mlib_ImageGetLutType(colormap);
mlib_d64 *plut = (mlib_d64 *) mlib_ImageGetLutDoubleData(colormap);
void *buff;
channels = mlib_ImageGetLutChannels(colormap);
plut -= channels * mlib_ImageGetLutOffset(colormap);
if (max_xsize == 0) {
return MLIB_SUCCESS;
}
if (ltype == MLIB_BYTE) {
buff = mlib_malloc(channels * max_xsize);
}
else {
buff = mlib_malloc(channels * max_xsize * sizeof(mlib_s16));
}
if (buff == NULL)
return MLIB_FAILURE;
switch (ltype) {
case MLIB_BYTE:
switch (type) {
case MLIB_BYTE:
MLIB_PROCESS_EDGES(MLIB_EDGE_INDEX_u8i, mlib_u8);
break;
case MLIB_SHORT:
srcStride >>= 1;
MLIB_PROCESS_EDGES(MLIB_EDGE_INDEX_u8i, mlib_s16);
break;
}
break;
case MLIB_SHORT:
switch (type) {
case MLIB_BYTE:
MLIB_PROCESS_EDGES(MLIB_EDGE_INDEX_s16i, mlib_u8);
break;
case MLIB_SHORT:
srcStride >>= 1;
MLIB_PROCESS_EDGES(MLIB_EDGE_INDEX_s16i, mlib_s16);
break;
}
break;
}
mlib_free(buff);
return MLIB_SUCCESS;
}
mlib_status
mlib_ImageMinFilter5x5_U8(
void *dst,
void *src,
mlib_s32 dlb,
mlib_s32 slb,
mlib_s32 wid,
mlib_s32 hgt)
#endif /* MAX_FILTER */
{
mlib_u8 *pbuff, *buff0, *buff1, *buff2, *buff3, *buffT;
mlib_u8 *sl, *sp0, *sp1, *sp2, *sp3, *sp4, *sp5, *dl;
__m64 *dp0, *dp1;
__m64 aa, bb, cc, dd, e0, e1, e2, e3, e4, ee, f0, f1, f2, f3, f4, ff,
r0, r1;
__m64 e_mask, mask8080;
mlib_s32 i, j, wid8, tail;
wid = (wid - KSIZE1) * SSIZE;
wid8 = (wid + 7) & ~7;
pbuff = mlib_malloc(4 * wid8);
buff0 = pbuff;
buff1 = buff0 + wid8;
buff2 = buff1 + wid8;
buff3 = buff2 + wid8;
sl = (mlib_u8 *)src;
dl = (mlib_u8 *)dst + 2 * (dlb + SSIZE);
tail = wid & 7;
e_mask = ((__m64 *) mlib_mask64_arr)[tail];
mask8080 = mmx_from_int_dup(0x80808080);
for (j = 0; j < 2; j++) {
sp0 = buff0;
sp1 = buff1;
sp4 = sl;
sp5 = sl + slb;
sl += 2 * slb;
for (i = 0; i < wid; i += 8) {
e0 = *(__m64 *) sp4;
e1 = *(__m64 *) (sp4 + SSIZE);
e2 = *(__m64 *) (sp4 + 2 * SSIZE);
e3 = *(__m64 *) (sp4 + 3 * SSIZE);
e4 = *(__m64 *) (sp4 + 4 * SSIZE);
f0 = *(__m64 *) sp5;
f1 = *(__m64 *) (sp5 + SSIZE);
f2 = *(__m64 *) (sp5 + 2 * SSIZE);
f3 = *(__m64 *) (sp5 + 3 * SSIZE);
f4 = *(__m64 *) (sp5 + 4 * SSIZE);
ee = C_COMP(e0, e1);
ff = C_COMP(f0, f1);
e2 = C_COMP(e2, e3);
f2 = C_COMP(f2, f3);
ee = C_COMP(ee, e4);
ff = C_COMP(ff, f4);
ee = C_COMP(ee, e2);
ff = C_COMP(ff, f2);
*(__m64 *) sp0 = ee;
*(__m64 *) sp1 = ff;
sp0 += 8;
sp1 += 8;
sp4 += 8;
sp5 += 8;
}
buffT = buff0;
buff0 = buff2;
buff2 = buffT;
buffT = buff1;
buff1 = buff3;
buff3 = buffT;
}
for (j = 0; j <= (hgt - KSIZE1 - 2); j += 2) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff0;
sp1 = buff1;
sp2 = buff2;
sp3 = buff3;
sp4 = sl;
sp5 = sl + slb;
/*
* line0: aa
* line1: bb
* line2: cc
* line3: dd
* line4: e0 e1 e2 e3 e4
* line5: f0 f1 f2 f3 f4
*/
for (i = 0; i <= wid - 8; i += 8) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
e0 = *(__m64 *) sp4;
e1 = *(__m64 *) (sp4 + SSIZE);
e2 = *(__m64 *) (sp4 + 2 * SSIZE);
e3 = *(__m64 *) (sp4 + 3 * SSIZE);
e4 = *(__m64 *) (sp4 + 4 * SSIZE);
f0 = *(__m64 *) sp5;
f1 = *(__m64 *) (sp5 + SSIZE);
f2 = *(__m64 *) (sp5 + 2 * SSIZE);
f3 = *(__m64 *) (sp5 + 3 * SSIZE);
f4 = *(__m64 *) (sp5 + 4 * SSIZE);
ee = C_COMP(e0, e1);
ff = C_COMP(f0, f1);
e2 = C_COMP(e2, e3);
f2 = C_COMP(f2, f3);
ee = C_COMP(ee, e4);
ff = C_COMP(ff, f4);
ee = C_COMP(ee, e2);
ff = C_COMP(ff, f2);
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
bb = C_COMP(bb, dd);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, ff);
*(__m64 *) sp0 = ee;
*(__m64 *) sp1 = ff;
(*dp0++) = r0;
(*dp1++) = r1;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
sp4 += 8;
sp5 += 8;
}
if (tail) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
e0 = *(__m64 *) sp4;
e1 = *(__m64 *) (sp4 + SSIZE);
e2 = *(__m64 *) (sp4 + 2 * SSIZE);
e3 = *(__m64 *) (sp4 + 3 * SSIZE);
e4 = *(__m64 *) (sp4 + 4 * SSIZE);
f0 = *(__m64 *) sp5;
f1 = *(__m64 *) (sp5 + SSIZE);
f2 = *(__m64 *) (sp5 + 2 * SSIZE);
f3 = *(__m64 *) (sp5 + 3 * SSIZE);
f4 = *(__m64 *) (sp5 + 4 * SSIZE);
ee = C_COMP(e0, e1);
ff = C_COMP(f0, f1);
e2 = C_COMP(e2, e3);
f2 = C_COMP(f2, f3);
ee = C_COMP(ee, e4);
ff = C_COMP(ff, f4);
ee = C_COMP(ee, e2);
ff = C_COMP(ff, f2);
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
bb = C_COMP(bb, dd);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, ff);
*(__m64 *) sp0 = ee;
*(__m64 *) sp1 = ff;
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
*dp1 =
_mm_or_si64(_mm_and_si64(e_mask, r1),
_mm_andnot_si64(e_mask, *dp1));
}
buffT = buff0;
buff0 = buff2;
buff2 = buffT;
buffT = buff1;
buff1 = buff3;
buff3 = buffT;
sl += 2 * slb;
dl += 2 * dlb;
}
/* last line */
if (j == (hgt - KSIZE1 - 1)) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff0;
sp1 = buff1;
sp2 = buff2;
sp3 = buff3;
sp4 = sl;
for (i = 0; i <= wid - 8; i += 8) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
e0 = *(__m64 *) sp4;
e1 = *(__m64 *) (sp4 + SSIZE);
e2 = *(__m64 *) (sp4 + 2 * SSIZE);
e3 = *(__m64 *) (sp4 + 3 * SSIZE);
e4 = *(__m64 *) (sp4 + 4 * SSIZE);
ee = C_COMP(e0, e1);
e2 = C_COMP(e2, e3);
ee = C_COMP(ee, e4);
ee = C_COMP(ee, e2);
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
bb = C_COMP(bb, dd);
r0 = C_COMP(aa, bb);
(*dp0++) = r0;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
sp4 += 8;
}
if (tail) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
cc = *(__m64 *) sp2;
dd = *(__m64 *) sp3;
e0 = *(__m64 *) sp4;
e1 = *(__m64 *) (sp4 + SSIZE);
e2 = *(__m64 *) (sp4 + 2 * SSIZE);
e3 = *(__m64 *) (sp4 + 3 * SSIZE);
e4 = *(__m64 *) (sp4 + 4 * SSIZE);
ee = C_COMP(e0, e1);
e2 = C_COMP(e2, e3);
ee = C_COMP(ee, e4);
ee = C_COMP(ee, e2);
bb = C_COMP(bb, cc);
dd = C_COMP(dd, ee);
bb = C_COMP(bb, dd);
r0 = C_COMP(aa, bb);
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
}
}
_mm_empty();
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
mlib_status mlib_convMxN_8nw_mask(mlib_image *dst,
const mlib_image *src,
mlib_s32 m,
mlib_s32 n,
mlib_s32 dm,
mlib_s32 dn,
const mlib_s32 *kern,
mlib_s32 scale,
mlib_s32 cmask)
{
mlib_d64 *buffs_local[3 * (MAX_N + 1)], **buffs = buffs_local, **buff;
mlib_d64 *buff0, *buff1, *buff2, *buff3, *buffn, *buffd, *buffe;
mlib_d64 s00, s01, s10, s11, s20, s21, s30, s31, s0, s1, s2, s3;
mlib_d64 d00, d01, d10, d11, d20, d21, d30, d31;
mlib_d64 dd, d0, d1;
mlib_s32 ik, jk, ik_last, jk_size, coff, off, doff;
mlib_u8 *sl, *sp, *dl;
mlib_s32 hgt = mlib_ImageGetHeight(src);
mlib_s32 wid = mlib_ImageGetWidth(src);
mlib_s32 sll = mlib_ImageGetStride(src);
mlib_s32 dll = mlib_ImageGetStride(dst);
mlib_u8 *adr_src = (mlib_u8 *) mlib_ImageGetData(src);
mlib_u8 *adr_dst = (mlib_u8 *) mlib_ImageGetData(dst);
mlib_s32 ssize, xsize, dsize, esize, buff_ind;
mlib_d64 *pbuff, *dp;
mlib_f32 *karr = (mlib_f32 *) kern;
mlib_s32 gsr_scale = (31 - scale) << 3;
mlib_d64 drnd = vis_to_double_dup(mlib_round_8[31 - scale]);
mlib_s32 i, j, l, chan, testchan;
mlib_s32 nchan = mlib_ImageGetChannels(dst);
void (*p_proc_load) (const mlib_u8 *, mlib_u8 *, mlib_s32, mlib_s32);
void (*p_proc_store) (const mlib_u8 *, mlib_u8 *, mlib_s32, mlib_s32);
if (n > MAX_N) {
buffs = mlib_malloc(3 * (n + 1) * sizeof(mlib_d64 *));
if (buffs == NULL)
return MLIB_FAILURE;
}
buff = buffs + 2 * (n + 1);
adr_dst += dn * dll + dm * nchan;
ssize = wid;
dsize = (ssize + 7) / 8;
esize = dsize + 4;
pbuff = mlib_malloc((n + 4) * esize * sizeof(mlib_d64));
if (pbuff == NULL) {
if (buffs != buffs_local)
mlib_free(buffs);
return MLIB_FAILURE;
}
for (i = 0; i < (n + 1); i++)
buffs[i] = pbuff + i * esize;
for (i = 0; i < (n + 1); i++)
buffs[(n + 1) + i] = buffs[i];
buffd = buffs[n] + esize;
buffe = buffd + 2 * esize;
hgt -= (n - 1);
xsize = ssize - (m - 1);
vis_write_gsr(gsr_scale + 7);
if (nchan == 2) {
p_proc_load = &mlib_v_ImageChannelExtract_U8_21_D1;
p_proc_store = &mlib_v_ImageChannelInsert_U8_12_D1;
}
else if (nchan == 3) {
p_proc_load = &mlib_v_ImageChannelExtract_U8_31_D1;
p_proc_store = &mlib_v_ImageChannelInsert_U8_13_D1;
}
else {
p_proc_load = &mlib_v_ImageChannelExtract_U8_41_D1;
p_proc_store = &mlib_v_ImageChannelInsert_U8_14_D1;
}
testchan = 1;
for (chan = 0; chan < nchan; chan++) {
buff_ind = 0;
sl = adr_src;
dl = adr_dst;
if ((cmask & testchan) == 0) {
testchan <<= 1;
continue;
}
for (l = 0; l < n; l++) {
mlib_d64 *buffn = buffs[l];
sp = sl + l * sll;
(*p_proc_load) ((mlib_u8 *) sp, (mlib_u8 *) buffn, ssize, testchan);
}
/* init buffer */
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
buffd[2 * i] = drnd;
buffd[2 * i + 1] = drnd;
}
for (j = 0; j < hgt; j++) {
mlib_d64 **buffc = buffs + buff_ind;
mlib_f32 *pk = karr, k0, k1, k2, k3;
sp = sl + n * sll;
for (l = 0; l < n; l++) {
buff[l] = buffc[l];
}
buffn = buffc[n];
(*p_proc_load) ((mlib_u8 *) sp, (mlib_u8 *) buffn, ssize, testchan);
ik_last = (m - 1);
for (jk = 0; jk < n; jk += jk_size) {
jk_size = n - jk;
if (jk_size >= 6)
jk_size = 4;
if (jk_size == 5)
jk_size = 3;
coff = 0;
if (jk_size == 1) {
for (ik = 0; ik < m; ik++, coff++) {
if (!jk && ik == ik_last)
continue;
k0 = pk[ik];
doff = coff / 8;
buff0 = buff[jk] + doff;
off = coff & 7;
vis_write_gsr(gsr_scale + off);
s01 = buff0[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s01 = buff0[i + 1];
s0 = vis_faligndata(s00, s01);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d00, d0);
d1 = vis_fpadd16(d01, d1);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
pk += m;
}
else if (jk_size == 2) {
for (ik = 0; ik < m; ik++, coff++) {
if (!jk && ik == ik_last)
continue;
k0 = pk[ik];
k1 = pk[ik + m];
doff = coff / 8;
buff0 = buff[jk] + doff;
buff1 = buff[jk + 1] + doff;
off = coff & 7;
vis_write_gsr(gsr_scale + off);
s01 = buff0[0];
s11 = buff1[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s10 = s11;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d00, d0);
d0 = vis_fpadd16(d10, d0);
d1 = vis_fpadd16(d01, d1);
d1 = vis_fpadd16(d11, d1);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
pk += 2 * m;
}
else if (jk_size == 3) {
for (ik = 0; ik < m; ik++, coff++) {
if (!jk && ik == ik_last)
continue;
k0 = pk[ik];
k1 = pk[ik + m];
k2 = pk[ik + 2 * m];
doff = coff / 8;
buff0 = buff[jk] + doff;
buff1 = buff[jk + 1] + doff;
buff2 = buff[jk + 2] + doff;
off = coff & 7;
vis_write_gsr(gsr_scale + off);
if (off == 0) {
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s0 = buff0[i];
s1 = buff1[i];
s2 = buff2[i];
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d00 = vis_fpadd16(d00, d10);
d0 = vis_fpadd16(d20, d0);
d0 = vis_fpadd16(d00, d0);
d01 = vis_fpadd16(d01, d11);
d1 = vis_fpadd16(d21, d1);
d1 = vis_fpadd16(d01, d1);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
else if (off == 4) {
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s00 = s01;
s10 = s11;
s20 = s21;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
d00 = vis_fmul8x16au(vis_read_lo(s00), k0);
d01 = vis_fmul8x16au(vis_read_hi(s01), k0);
d10 = vis_fmul8x16au(vis_read_lo(s10), k1);
d11 = vis_fmul8x16au(vis_read_hi(s11), k1);
d20 = vis_fmul8x16au(vis_read_lo(s20), k2);
d21 = vis_fmul8x16au(vis_read_hi(s21), k2);
d00 = vis_fpadd16(d00, d10);
d0 = vis_fpadd16(d20, d0);
d0 = vis_fpadd16(d00, d0);
d01 = vis_fpadd16(d01, d11);
d1 = vis_fpadd16(d21, d1);
d1 = vis_fpadd16(d01, d1);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
else {
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s00 = s01;
s10 = s11;
s20 = s21;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
s2 = vis_faligndata(s20, s21);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d00 = vis_fpadd16(d00, d10);
d0 = vis_fpadd16(d20, d0);
d0 = vis_fpadd16(d00, d0);
d01 = vis_fpadd16(d01, d11);
d1 = vis_fpadd16(d21, d1);
d1 = vis_fpadd16(d01, d1);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
}
pk += 3 * m;
}
else { /* jk_size == 4 */
for (ik = 0; ik < m; ik++, coff++) {
if (!jk && ik == ik_last)
continue;
k0 = pk[ik];
k1 = pk[ik + m];
k2 = pk[ik + 2 * m];
k3 = pk[ik + 3 * m];
doff = coff / 8;
buff0 = buff[jk] + doff;
buff1 = buff[jk + 1] + doff;
buff2 = buff[jk + 2] + doff;
buff3 = buff[jk + 3] + doff;
off = coff & 7;
vis_write_gsr(gsr_scale + off);
if (off == 0) {
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s0 = buff0[i];
s1 = buff1[i];
s2 = buff2[i];
s3 = buff3[i];
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d30 = vis_fmul8x16au(vis_read_hi(s3), k3);
d31 = vis_fmul8x16au(vis_read_lo(s3), k3);
d00 = vis_fpadd16(d00, d10);
d20 = vis_fpadd16(d20, d30);
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d20);
d01 = vis_fpadd16(d01, d11);
d21 = vis_fpadd16(d21, d31);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d21);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
else if (off == 4) {
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
s31 = buff3[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s00 = s01;
s10 = s11;
s20 = s21;
s30 = s31;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
s31 = buff3[i + 1];
d00 = vis_fmul8x16au(vis_read_lo(s00), k0);
d01 = vis_fmul8x16au(vis_read_hi(s01), k0);
d10 = vis_fmul8x16au(vis_read_lo(s10), k1);
d11 = vis_fmul8x16au(vis_read_hi(s11), k1);
d20 = vis_fmul8x16au(vis_read_lo(s20), k2);
d21 = vis_fmul8x16au(vis_read_hi(s21), k2);
d30 = vis_fmul8x16au(vis_read_lo(s30), k3);
d31 = vis_fmul8x16au(vis_read_hi(s31), k3);
d00 = vis_fpadd16(d00, d10);
d20 = vis_fpadd16(d20, d30);
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d20);
d01 = vis_fpadd16(d01, d11);
d21 = vis_fpadd16(d21, d31);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d21);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
else {
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
s31 = buff3[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
s00 = s01;
s10 = s11;
s20 = s21;
s30 = s31;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
s31 = buff3[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
s2 = vis_faligndata(s20, s21);
s3 = vis_faligndata(s30, s31);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d30 = vis_fmul8x16au(vis_read_hi(s3), k3);
d31 = vis_fmul8x16au(vis_read_lo(s3), k3);
d00 = vis_fpadd16(d00, d10);
d20 = vis_fpadd16(d20, d30);
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d20);
d01 = vis_fpadd16(d01, d11);
d21 = vis_fpadd16(d21, d31);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d21);
buffd[2 * i] = d0;
buffd[2 * i + 1] = d1;
}
}
}
pk += 4 * m;
}
}
/*****************************************
*****************************************
** Final iteration **
*****************************************
*****************************************/
jk_size = n;
if (jk_size >= 6)
jk_size = 4;
if (jk_size == 5)
jk_size = 3;
k0 = karr[ik_last];
k1 = karr[ik_last + m];
k2 = karr[ik_last + 2 * m];
k3 = karr[ik_last + 3 * m];
off = ik_last;
doff = off / 8;
off &= 7;
buff0 = buff[0] + doff;
buff1 = buff[1] + doff;
buff2 = buff[2] + doff;
buff3 = buff[3] + doff;
vis_write_gsr(gsr_scale + off);
if (jk_size == 1) {
dp = buffe;
s01 = buff0[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s01 = buff0[i + 1];
s0 = vis_faligndata(s00, s01);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d0, d00);
d1 = vis_fpadd16(d1, d01);
dd = vis_fpack16_pair(d0, d1);
dp[i] = dd;
buffd[2 * i] = drnd;
buffd[2 * i + 1] = drnd;
}
}
else if (jk_size == 2) {
dp = buffe;
s01 = buff0[0];
s11 = buff1[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s10 = s11;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d10);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d11);
dd = vis_fpack16_pair(d0, d1);
dp[i] = dd;
buffd[2 * i] = drnd;
buffd[2 * i + 1] = drnd;
}
}
else if (jk_size == 3) {
dp = buffe;
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s10 = s11;
s20 = s21;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
s2 = vis_faligndata(s20, s21);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d10);
d0 = vis_fpadd16(d0, d20);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d11);
d1 = vis_fpadd16(d1, d21);
dd = vis_fpack16_pair(d0, d1);
dp[i] = dd;
buffd[2 * i] = drnd;
buffd[2 * i + 1] = drnd;
}
}
else { /* if (jk_size == 4) */
dp = buffe;
s01 = buff0[0];
s11 = buff1[0];
s21 = buff2[0];
s31 = buff3[0];
#pragma pipeloop(0)
for (i = 0; i < (xsize + 7) / 8; i++) {
s00 = s01;
s10 = s11;
s20 = s21;
s30 = s31;
s01 = buff0[i + 1];
s11 = buff1[i + 1];
s21 = buff2[i + 1];
s31 = buff3[i + 1];
s0 = vis_faligndata(s00, s01);
s1 = vis_faligndata(s10, s11);
s2 = vis_faligndata(s20, s21);
s3 = vis_faligndata(s30, s31);
d00 = vis_fmul8x16au(vis_read_hi(s0), k0);
d01 = vis_fmul8x16au(vis_read_lo(s0), k0);
d10 = vis_fmul8x16au(vis_read_hi(s1), k1);
d11 = vis_fmul8x16au(vis_read_lo(s1), k1);
d20 = vis_fmul8x16au(vis_read_hi(s2), k2);
d21 = vis_fmul8x16au(vis_read_lo(s2), k2);
d30 = vis_fmul8x16au(vis_read_hi(s3), k3);
d31 = vis_fmul8x16au(vis_read_lo(s3), k3);
d0 = buffd[2 * i];
d1 = buffd[2 * i + 1];
d0 = vis_fpadd16(d0, d00);
d0 = vis_fpadd16(d0, d10);
d0 = vis_fpadd16(d0, d20);
d0 = vis_fpadd16(d0, d30);
d1 = vis_fpadd16(d1, d01);
d1 = vis_fpadd16(d1, d11);
d1 = vis_fpadd16(d1, d21);
d1 = vis_fpadd16(d1, d31);
dd = vis_fpack16_pair(d0, d1);
dp[i] = dd;
buffd[2 * i] = drnd;
buffd[2 * i + 1] = drnd;
}
}
(*p_proc_store) ((mlib_u8 *) buffe, (mlib_u8 *) dl, xsize, testchan);
sl += sll;
dl += dll;
buff_ind++;
if (buff_ind >= (n + 1))
buff_ind = 0;
}
testchan <<= 1;
}
mlib_free(pbuff);
if (buffs != buffs_local)
mlib_free(buffs);
return MLIB_SUCCESS;
}
mlib_status
mlib_m_ImageInitInterpTableAffine_S16(
mlib_interp_table * table,
mlib_s32 nchan)
{
mlib_s32 width, height, width_bits, height_bits, vis_width_bits,
vis_height_bits;
mlib_s32 subsampleBitsH, subsampleBitsV;
mlib_s32 i, j, c, scale, num_copy, num_copy_old;
mlib_s32 isum;
mlib_s32 max_scale, min_scale, scaleh, scalev;
mlib_s32 norm_scale_v, norm_scale_h;
mlib_d64 dscale, *dataH, *dataV;
mlib_d64 **ptr_tablex, *tablex, *tablex_old, *tabley;
mlib_d64 max, d;
mlib_d64 sumh, sumv, normh, normv;
if (!table)
return (MLIB_FAILURE);
if (table->shift_vis_affine < 0)
return (MLIB_FAILURE);
if (nchan == 1) {
num_copy = 1;
ptr_tablex = &(table->dataH_s16_1);
} else if (nchan == 2) {
num_copy = 2;
ptr_tablex = &(table->dataH_s16_3);
} else if (nchan == 3 || nchan == 4) {
num_copy = 4;
ptr_tablex = &(table->dataH_s16_4);
} else
return (MLIB_FAILURE);
if (*ptr_tablex != NULL && table->dataV_s16_1 != NULL)
return (MLIB_SUCCESS);
dataH = mlib_ImageGetInterpDoubleDataH(table);
dataV = mlib_ImageGetInterpDoubleDataV(table);
if (!dataH || !dataV)
return (MLIB_FAILURE);
width = mlib_ImageGetInterpWidth(table);
height = mlib_ImageGetInterpHeight(table);
width_bits = mlib_ImageGetInterpWidthBits(table);
height_bits = mlib_ImageGetInterpHeightBits(table);
vis_width_bits = table->vis_width_bits;
vis_height_bits = table->vis_height_bits;
subsampleBitsH = mlib_ImageGetInterpSubsampleBitsH(table);
subsampleBitsV = mlib_ImageGetInterpSubsampleBitsV(table);
if (table->dataV_s16_1 != NULL) {
if (table->dataH_s16_1 != NULL) {
tablex_old = table->dataH_s16_1;
num_copy_old = 1;
} else if (table->dataH_s16_3 != NULL) {
tablex_old = table->dataH_s16_3;
num_copy_old = 3;
} else {
tablex_old = table->dataH_s16_4;
num_copy_old = 4;
}
tablex =
mlib_malloc(num_copy * (1 << subsampleBitsH) *
(1 << vis_width_bits) * sizeof (mlib_s16));
if (tablex == NULL)
return (MLIB_FAILURE);
for (j = 0; j < ((width + 1) & ~1); j++) {
mlib_s16 *tbl = (mlib_s16 *)tablex + j * num_copy;
mlib_s16 *tbl_old =
(mlib_s16 *)tablex_old + j * num_copy_old;
for (i = 0; i < (1 << subsampleBitsH); i++) {
mlib_s16 v =
tbl_old[num_copy_old *
(i << vis_width_bits)];
for (c = 0; c < num_copy; c++) {
tbl[num_copy * (i << vis_width_bits) +
c] = v;
}
}
}
*ptr_tablex = tablex;
return (MLIB_SUCCESS);
}
sumv = 0;
max = 0;
for (i = 0; i < (1 << subsampleBitsV); i++) {
mlib_d64 s = 0;
mlib_s32 ind = (i << height_bits);
for (j = 0; j < height; j++) {
d = mlib_fabs(dataV[j + ind]);
s += d;
max = (max > d) ? max : d;
}
sumv = (sumv > s) ? sumv : s;
}
/* all fhkernels = 0 */
if (sumv == 0) {
dscale = 0;
/* X table */
tablex =
mlib_malloc(num_copy * (1 << subsampleBitsH) *
(1 << vis_width_bits) * sizeof (mlib_s16));
if (tablex == NULL)
return (MLIB_FAILURE);
INIT_TABLE_16(tablex, (1 << subsampleBitsH), width, width_bits,
vis_width_bits, dataH);
if ((dataH == dataV) && num_copy == 4)
tabley = tablex;
else {
num_copy = 4;
tabley =
mlib_malloc(num_copy * (1 << subsampleBitsV) *
(1 << vis_height_bits) * sizeof (mlib_s16));
if (tabley == NULL) {
mlib_free(tablex);
return (MLIB_FAILURE);
}
INIT_TABLE_16(tabley, (1 << subsampleBitsV), height,
height_bits, vis_height_bits, dataV);
*ptr_tablex = tablex;
table->dataV_s16_1 = tabley;
/* Store shift */
table->shift_vis_affine = 43;
return (MLIB_SUCCESS);
}
}
normv = 32767.0 / (32768.0 * sumv);
scalev = mlib_ilogb(sumv * normv);
isum = mlib_ilogb(max * normv);
/* all elements must be in the range -32768, 32767 */
if (scalev == isum)
norm_scale_v = 14;
/* but sumv may be in the range -65576, 65575 */
else
norm_scale_v = 15;
min_scale = 25;
max_scale = 40;
normh = 32768.0 * sumv / 32767;
if (dataH != dataV) {
sumh = 0;
max = 0;
for (i = 0; i < (1 << subsampleBitsH); i++) {
mlib_d64 s = 0;
mlib_s32 ind = (i << width_bits);
for (j = 0; j < width; j++) {
d = mlib_fabs(dataH[j + ind]);
s += d;
max = (max > d) ? max : d;
}
sumh = (sumh > s) ? sumh : s;
}
} else
sumh = sumv;
isum = mlib_ilogb(max * normh);
scaleh = mlib_ilogb(sumh * normh);
/* all elements must be in the range -32768, 32767 */
if (scaleh == isum)
norm_scale_h = 14;
/* but sumh may be in the range -65576, 65575 */
else
norm_scale_h = 15;
scale = norm_scale_v + norm_scale_h - (scaleh + scalev);
if (scale < min_scale) {
table->shift_vis_affine = -1;
/* koeff. are so large */
return (MLIB_FAILURE);
}
if (scale > max_scale) {
scaleh += (scale - max_scale + 1) >> 1;
scalev += (scale - max_scale) >> 1;
scale = max_scale;
}
mlib_status
mlib_m_conv3x3_16nw_4(
mlib_image *dst,
const mlib_image *src,
const mlib_s32 *kern,
mlib_s32 scalef_expon)
{
__m64 buff_loc[6 * BUFF_LINE], *pbuff = buff_loc;
__m64 *buff0, *buff1, *buff2, *buffT;
GET_SRC_DST_PARAMETERS(mlib_s16);
__m64 ker1, ker2, ker3, ker4, ker5, ker6, ker7, ker8, ker9;
__m64 d0, d1, d2, rr, tmpa, tmpb;
__m64 prev0h, prev1h, sum0h, sum1h, sum2h, tmph;
__m64 prev0l, prev1l, sum0l, sum1l, sum2l, tmpl;
__m64 *sp, *dp;
mlib_s32 shift;
mlib_s32 row, wid4, i, j;
width -= 2;
height -= 2;
width *= NCHAN;
dl += dll + NCHAN;
wid4 = (width + 3) / 4;
if (wid4 > BUFF_LINE) {
pbuff = mlib_malloc(sizeof (__m64) * 6 * wid4);
}
GET_KERN();
buff0 = pbuff;
buff1 = buff0 + 2 * wid4;
buff2 = buff1 + 2 * wid4;
for (j = 0; j < 2; j++) {
sp = (__m64 *) sl;
d1 = (*sp++);
d2 = (*sp++);
for (i = 0; i < wid4; i++) {
PREP_3x3(i);
}
sl += sll;
if (j == 0) {
buffT = buff1;
buff1 = buff0;
buff0 = buffT;
}
}
for (row = 0; row < height; row++) {
sp = (__m64 *) sl;
dp = (__m64 *) dl;
d1 = (*sp++);
d2 = (*sp++);
for (i = 0; i < width / 4; i++) {
CONV_3x3(i);
dp[i] = rr;
}
if (width & 3) {
__m64 mask =
((__m64 *) mlib_mask64_arr)[2 * (width & 3)];
CONV_3x3(i);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, rr),
_mm_andnot_si64(mask, dp[i]));
}
buffT = buff1;
buff1 = buff0;
buff0 = buffT;
sl += sll;
dl += dll;
}
_mm_empty();
if (pbuff != buff_loc)
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
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_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_image *mlib_ImageCreate(mlib_type type,
mlib_s32 channels,
mlib_s32 width,
mlib_s32 height)
{
mlib_image *image;
mlib_s32 wb; /* width in bytes */
void *data;
/* sanity check */
if (width <= 0 || height <= 0 || channels < 1 || channels > 4) {
return NULL;
};
switch (type) {
case MLIB_DOUBLE:
wb = width * channels * 8;
break;
case MLIB_FLOAT:
case MLIB_INT:
wb = width * channels * 4;
break;
case MLIB_USHORT:
case MLIB_SHORT:
wb = width * channels * 2;
break;
case MLIB_BYTE:
wb = width * channels;
break;
case MLIB_BIT:
wb = (width * channels + 7) / 8;
break;
default:
return NULL;
}
data = mlib_malloc(wb * height);
if (data == NULL) {
return NULL;
}
image = (mlib_image *)mlib_malloc(sizeof(mlib_image));
if (image == NULL) {
mlib_free(data);
return NULL;
};
image -> type = type;
image -> channels = channels;
image -> width = width;
image -> height = height;
image -> stride = wb;
image -> data = data;
image -> flags = ((width & 0xf) << 8); /* set width field */
image -> flags |= ((height & 0xf) << 12); /* set height field */
image -> flags |= ((wb & 0xf) << 16); /* set stride field */
image -> flags |= (mlib_addr)data & 0xff;
image -> format = MLIB_FORMAT_UNKNOWN;
image -> paddings[0] = 0;
image -> paddings[1] = 0;
image -> paddings[2] = 0;
image -> paddings[3] = 0;
image -> bitoffset = 0;
if ((type == MLIB_BIT) && (wb * 8 != width * channels)) {
image -> flags |= MLIB_IMAGE_ONEDVECTOR; /* not 1-d vector */
}
image -> flags &= MLIB_IMAGE_ATTRIBUTESET;
image -> state = NULL;
return image;
}
/* *********************************************************** */
mlib_status
mlib_m_sconv3x3_8nw_1(
mlib_image *dst,
mlib_image *src,
mlib_s32 *hkernel,
mlib_s32 *vkernel,
mlib_s32 scalef_expon)
{
__m64 buff_loc[3 * BUFF_LINE], *pbuff = buff_loc;
__m64 *buff0, *buff1, *buffT;
GET_SRC_DST_PARAMETERS(mlib_u8);
__m64 hker0, hker1, hker2, vker0, vker1, vker2;
__m64 s0, d0, d1, sum0, sum1, sum2, aa, bb, res_hi, res_lo;
__m64 zero = _m_zero;
mlib_s32 shift;
mlib_s32 *sp;
mlib_s32 row, wid4, i, j;
width -= 2;
height -= 2;
dl += dll + 1;
wid4 = (width + 7) / 4;
if (wid4 > BUFF_LINE) {
pbuff = mlib_malloc(sizeof (__m64) * 3 * wid4);
}
GET_KERN();
buff0 = pbuff;
buff1 = buff0 + wid4;
for (j = 0; j < 2; j++) {
sp = (mlib_s32 *)sl;
*(mlib_s32 *)&s0 = (*sp++);
UNPACK_SRC(d1, lo);
for (i = 0; i < wid4; i++) {
*(mlib_s32 *)&s0 = sp[i];
PREP_3x3_1ch(lo, i);
}
sl += sll;
buffT = buff1;
buff1 = buff0;
buff0 = buffT;
}
for (row = 0; row < height; row++) {
__m64 *sp = (__m64 *) sl;
__m64 *dp = (__m64 *) dl;
s0 = (*sp++);
UNPACK_SRC(d1, lo);
for (i = 0; i < width / 8; i++) {
CONV_3x3_1ch(hi, 2 * i);
s0 = sp[i];
CONV_3x3_1ch(lo, 2 * i + 1);
dp[i] = _mm_packs_pu16(res_hi, res_lo);
}
if (width & 7) {
__m64 mask;
mask = ((__m64 *) mlib_mask64_arr)[width & 7];
CONV_3x3_1ch(hi, 2 * i);
s0 = sp[i];
CONV_3x3_1ch(lo, 2 * i + 1);
res_hi = _mm_packs_pu16(res_hi, res_lo);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, res_hi),
_mm_andnot_si64(mask, dp[i]));
}
buffT = buff1;
buff1 = buff0;
buff0 = buffT;
sl += sll;
dl += dll;
}
_mm_empty();
if (pbuff != buff_loc)
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
mlib_status
mlib_ImageMinFilter3x3_S16(
void *dst,
void *src,
mlib_s32 dlb,
mlib_s32 slb,
mlib_s32 wid,
mlib_s32 hgt)
#endif /* MAX_FILTER */
{
mlib_u8 *buff, *buff1;
mlib_u8 *sl, *sp0, *sp1, *sp2, *sp3, *dl;
__m64 *dp0, *dp1;
__m64 aa, bb, c0, c1, c2, cc, d0, d1, d2, dd, r0, r1;
__m64 e_mask;
mlib_s32 i, j, wid8, tail;
wid = (wid - 2) * SSIZE;
wid8 = (wid + 7) & ~7;
buff = mlib_malloc(2 * wid8);
buff1 = buff + wid8;
sl = (mlib_u8 *)src;
/* dst ptrs skip top j and left col */
dl = (mlib_u8 *)dst + dlb + SSIZE;
tail = wid & 7;
e_mask = ((__m64 *) mlib_mask64_arr)[tail];
sp0 = buff;
sp1 = buff1;
sp2 = sl;
sp3 = sp2 + slb;
sl += 2 * slb;
for (i = 0; i < wid; i += 8) {
c0 = *(__m64 *) sp2;
c1 = *(__m64 *) (sp2 + SSIZE);
c2 = *(__m64 *) (sp2 + 2 * SSIZE);
d0 = *(__m64 *) sp3;
d1 = *(__m64 *) (sp3 + SSIZE);
d2 = *(__m64 *) (sp3 + 2 * SSIZE);
cc = C_COMP(c0, c1);
dd = C_COMP(d0, d1);
cc = C_COMP(cc, c2);
dd = C_COMP(dd, d2);
*(__m64 *) sp0 = cc;
*(__m64 *) sp1 = dd;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
}
for (j = 0; j <= (hgt - 2 - 2); j += 2) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff;
sp1 = buff1;
sp2 = sl;
sp3 = sp2 + slb;
/*
* line0: aa
* line1: bb
* line2: c0 c1 c2
* line3: d0 d1 d2
*/
for (i = 0; i <= wid - 8; i += 8) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
c0 = *(__m64 *) sp2;
c1 = *(__m64 *) (sp2 + SSIZE);
c2 = *(__m64 *) (sp2 + 2 * SSIZE);
d0 = *(__m64 *) sp3;
d1 = *(__m64 *) (sp3 + SSIZE);
d2 = *(__m64 *) (sp3 + 2 * SSIZE);
cc = C_COMP(c0, c1);
dd = C_COMP(d0, d1);
cc = C_COMP(cc, c2);
dd = C_COMP(dd, d2);
bb = C_COMP(bb, cc);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, dd);
*(__m64 *) sp0 = cc;
*(__m64 *) sp1 = dd;
(*dp0++) = r0;
(*dp1++) = r1;
sp0 += 8;
sp1 += 8;
sp2 += 8;
sp3 += 8;
}
if (tail) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
c0 = *(__m64 *) sp2;
c1 = *(__m64 *) (sp2 + SSIZE);
c2 = *(__m64 *) (sp2 + 2 * SSIZE);
d0 = *(__m64 *) sp3;
d1 = *(__m64 *) (sp3 + SSIZE);
d2 = *(__m64 *) (sp3 + 2 * SSIZE);
cc = C_COMP(c0, c1);
dd = C_COMP(d0, d1);
cc = C_COMP(cc, c2);
dd = C_COMP(dd, d2);
bb = C_COMP(bb, cc);
r0 = C_COMP(aa, bb);
r1 = C_COMP(bb, dd);
*(__m64 *) sp0 = cc;
*(__m64 *) sp1 = dd;
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
*dp1 =
_mm_or_si64(_mm_and_si64(e_mask, r1),
_mm_andnot_si64(e_mask, *dp1));
}
sl += 2 * slb;
dl += 2 * dlb;
}
/* last line */
if (j == (hgt - 3)) {
dp0 = (void *)dl;
dp1 = (void *)(dl + dlb);
sp0 = buff;
sp1 = buff1;
sp2 = sl;
for (i = 0; i <= wid - 8; i += 8) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
c0 = *(__m64 *) sp2;
c1 = *(__m64 *) (sp2 + SSIZE);
c2 = *(__m64 *) (sp2 + 2 * SSIZE);
cc = C_COMP(c0, c1);
cc = C_COMP(cc, c2);
r0 = C_COMP(aa, bb);
r0 = C_COMP(r0, cc);
(*dp0++) = r0;
sp0 += 8;
sp1 += 8;
sp2 += 8;
}
if (tail) {
aa = *(__m64 *) sp0;
bb = *(__m64 *) sp1;
c0 = *(__m64 *) sp2;
c1 = *(__m64 *) (sp2 + SSIZE);
c2 = *(__m64 *) (sp2 + 2 * SSIZE);
c1 = C_COMP(c0, c1);
cc = C_COMP(c1, c2);
r0 = C_COMP(aa, bb);
r0 = C_COMP(r0, cc);
*dp0 =
_mm_or_si64(_mm_and_si64(e_mask, r0),
_mm_andnot_si64(e_mask, *dp0));
}
}
_mm_empty();
mlib_free(buff);
return (MLIB_SUCCESS);
}