void
mlib_v_ImageNot_na(
mlib_u8 *sa,
mlib_u8 *da,
mlib_s32 size)
{
/* end points in dst */
mlib_u8 *dend;
/* 8-byte aligned start points in dst */
mlib_d64 *dp;
/* 8-byte aligned start point in src */
mlib_d64 *sp;
/* 8-byte source data */
mlib_d64 s0, s1;
/* offset of address in dst */
mlib_s32 j;
/* edge mask */
mlib_s32 emask;
/* prepare the destination addresses */
dp = (mlib_d64 *)((mlib_addr)da & (~7));
j = (mlib_addr)dp - (mlib_addr)da;
dend = da + size - 1;
/* prepare the source address */
sp = (mlib_d64 *)vis_alignaddr(sa, j);
/* generate edge mask for the start point */
emask = vis_edge8(da, dend);
s1 = vis_ld_d64_nf(sp);
if (emask != 0xff) {
s0 = s1;
s1 = vis_ld_d64_nf(sp + 1);
s0 = vis_faligndata(s0, s1);
vis_pst_8(vis_fnot(s0), dp++, emask);
sp++;
j += 8;
}
#pragma pipeloop(0)
for (; j <= (size - 8); j += 8) {
s0 = s1;
s1 = vis_ld_d64_nf(sp + 1);
(*dp++) = vis_fnot(vis_faligndata(s0, s1));
sp++;
}
if (j < size) {
s0 = vis_faligndata(s1, vis_ld_d64_nf(sp + 1));
emask = vis_edge8(dp, dend);
vis_pst_8(vis_fnot(s0), dp, emask);
}
}
void
mlib_ImageLineXor8000(
const mlib_u8 *src,
mlib_u8 *dst,
mlib_s32 size)
{
mlib_u8 *dend;
mlib_d64 *dptr;
mlib_d64 *sptr;
mlib_d64 s0, s1;
mlib_d64 mask8000 = vis_to_double_dup(0x80008000);
mlib_s32 j;
mlib_s32 emask;
/* prepare the destination addresses */
dptr = (mlib_d64 *)((mlib_addr)dst & (~7));
j = (mlib_addr)dptr - (mlib_addr)dst;
dend = (mlib_u8 *)dst + size - 1;
/* prepare the source address */
sptr = (mlib_d64 *)VIS_ALIGNADDR(src, j);
/* generate edge mask for the start point */
emask = vis_edge8(dst, dend);
s1 = vis_ld_d64_nf(sptr);
if (emask != 0xff) {
s0 = s1;
s1 = vis_ld_d64_nf(sptr + 1);
s0 = vis_fxor(vis_faligndata(s0, s1), mask8000);
vis_pst_8(s0, dptr++, emask);
sptr++;
j += 8;
}
#pragma pipeloop(0)
for (; j <= (size - 16); j += 8) {
s0 = s1;
s1 = sptr[1];
(*dptr++) = vis_fxor(vis_faligndata(s0, s1), mask8000);
sptr++;
}
if (j <= (size - 8)) {
s0 = s1;
s1 = vis_ld_d64_nf(sptr + 1);
(*dptr++) = vis_fxor(vis_faligndata(s0, s1), mask8000);
sptr++;
j += 8;
}
if (j < size) {
s0 = vis_fxor(vis_faligndata(s1, vis_ld_d64_nf(sptr + 1)),
mask8000);
emask = vis_edge8(dptr, dend);
vis_pst_8(s0, dptr, emask);
}
}
mlib_status
__mlib_VideoUpSample420_Nearest(
mlib_u8 *dst0,
mlib_u8 *dst1,
const mlib_u8 *src,
mlib_s32 n)
{
mlib_d64 *sp = (mlib_d64 *)src;
mlib_d64 *dp0 = (mlib_d64 *)dst0;
mlib_d64 *dp1 = (mlib_d64 *)dst1;
mlib_u8 *dend0 = dst0 + 2 * n - 1;
mlib_d64 sa, da;
mlib_s32 emask, i;
if (n <= 0)
return (MLIB_FAILURE);
#pragma pipeloop(0)
for (i = 0; i <= (n - 8); i += 8) {
sa = *sp;
*dp0 = *dp1 = vis_fpmerge(vis_read_hi(sa), vis_read_hi(sa));
*(dp0 + 1) = *(dp1 + 1) =
vis_fpmerge(vis_read_lo(sa), vis_read_lo(sa));
sp++;
dp0 += 2;
dp1 += 2;
}
if (i < n) {
sa = vis_ld_d64_nf(sp);
da = vis_fpmerge(vis_read_hi(sa), vis_read_hi(sa));
emask = vis_edge8(dp0, dend0);
vis_pst_8(da, dp0, emask);
vis_pst_8(da, dp1, emask);
i += 4;
dp0++;
dp1++;
if (i < n) {
da = vis_fpmerge(vis_read_lo(sa), vis_read_lo(sa));
emask = vis_edge8(dp0, dend0);
vis_pst_8(da, dp0, emask);
vis_pst_8(da, dp1, emask);
}
}
return (MLIB_SUCCESS);
}
mlib_status
mlib_v_conv3x3_8nw_4(
mlib_image *dst,
const mlib_image *src,
const mlib_s32 *kernel,
mlib_s32 scalef_expon,
mlib_s32 cmask)
{
/* pointers to dst row */
mlib_u8 *da, *d_a;
/* pointers to src, dst data */
mlib_u8 *adr_dst, *adr_src, *dend;
/* pointers to src rows */
mlib_u8 *sa, *sa1, *sa2;
/* pointers to rows in interm. src buf */
mlib_d64 *buff_src, *sbuf1, *sbuf2, *prow;
/* pointers to rows in interm. src buf */
mlib_d64 *sbuf3;
/* pointer to row in interm. dst buf */
mlib_d64 *dbuf;
/* mlib_d64 pointers to rows in interm. src buf */
mlib_d64 *s1, *s2, *s3;
/* mlib_d64 pointer to row in interm. dst buf */
mlib_d64 *ddst;
/* data */
mlib_d64 d1, d2, d_1, d_2, d21, d22;
/* data */
mlib_d64 d3, d_3, d23;
mlib_f32 k1k2, k3k4, k5k6, k7k8, k9k9;
/* src, dst and interm. buf. strides */
mlib_s32 dlb, slb, buf_slb;
mlib_s32 dh, dw;
mlib_d64 out0, out1;
mlib_d64 tmp0, tmp1, rnd;
mlib_d64 *dsa, *dp;
mlib_d64 sd0, sd1, sd00;
mlib_s32 emask, cmask1;
mlib_s32 rval, gsr_scale, i, j;
gsr_scale = 31 - scalef_expon;
vis_write_gsr((gsr_scale << 3));
rval = mlib_round_8[gsr_scale];
rnd = vis_freg_pair(vis_to_float(rval), vis_to_float(rval));
cmask = ((cmask & 0xf) << 4) + (cmask & 0xf);
cmask = (cmask << 8) + (cmask);
GET_SRC_DST_PARAMETERS();
LOAD_KERNEL_INTO_FLOAT();
buf_slb = (4 * dw + 24) >> 3;
PREPARE_INTERM_BUFFERS();
dw -= 2;
dw *= 4;
dh -= 2;
sa = adr_src;
sa1 = sa + slb;
sa2 = sa1 + slb;
d_a = adr_dst + dlb + 4;
/* load interm. src buff */
PREPARE_TO_LOAD_LINE(sbuf2, sa);
#pragma pipeloop(0)
LOAD_LINE_INTO_BUFFER(8);
/* load interm. src buff */
PREPARE_TO_LOAD_LINE(sbuf3, sa1);
#pragma pipeloop(0)
LOAD_LINE_INTO_BUFFER(8);
#pragma pipeloop(0)
for (j = 0; j < dh; j++) {
LOOP_INI();
PREPARE_TO_LOAD_LINE(sbuf3, sa2);
#pragma pipeloop(0)
LOAD_LINE_INTO_BUFFER(8);
vis_alignaddr(s1, 4);
d1 = *s1;
d2 = *s2;
d3 = *s3;
#pragma pipeloop(0)
for (i = 0; i < dw; i += 8) {
d_1 = *(s1 + 1);
d_2 = *(s2 + 1);
d_3 = *(s3 + 1);
out0 = out1 = rnd;
CONV_AU(d1, k1k2);
CONV_AL(d2, k3k4);
CONV_AU(d3, k7k8);
d21 = vis_faligndata(d1, d_1);
d22 = vis_faligndata(d2, d_2);
d23 = vis_faligndata(d3, d_3);
CONV_AL(d21, k1k2);
CONV_AU(d22, k5k6);
CONV_AL(d23, k7k8);
CONV_AU(d_1, k3k4);
CONV_AL(d_2, k5k6);
CONV_AU(d_3, k9k9);
(*ddst++) = vis_fpack16_pair(out0, out1);
d1 = d_1;
d2 = d_2;
d3 = d_3;
s1++;
s2++;
s3++;
}
ddst = dbuf;
/* prepare the destination addresses */
dp = (mlib_d64 *)((mlib_addr)da & (~7));
i = (mlib_addr)dp - (mlib_addr)da;
cmask1 = cmask >> (-i);
ddst = vis_alignaddr(ddst, i);
/* generate edge mask for the start point */
emask = vis_edge8(da, dend);
sd1 = ddst[0];
if (emask != 0xff) {
sd0 = sd1;
sd1 = ddst[1];
sd0 = vis_faligndata(sd0, sd1);
vis_pst_8(sd0, dp++, emask & cmask1);
ddst++;
i += 8;
}
#pragma pipeloop(0)
for (; i <= (dw - 8); i += 8) {
sd0 = sd1;
sd1 = ddst[1];
sd00 = vis_faligndata(sd0, sd1);
vis_pst_8(sd00, dp++, cmask1);
ddst++;
}
if (i < dw) {
sd0 = vis_faligndata(sd1, ddst[1]);
emask = vis_edge8(dp, dend);
vis_pst_8(sd0, dp, emask & cmask1);
}
sa2 = sa2 + slb;
d_a += dlb;
}
__mlib_free(buff_src);
return (MLIB_SUCCESS);
}
static void
mlib_VectorDotProd_U8C_al_x(
mlib_d64 *z,
const void *x,
const void *y,
mlib_s32 n)
/* The case of even address of vector x */
{
mlib_u8 *pxend, *px = (mlib_u8 *)x, *py = (mlib_u8 *)y;
mlib_d64 sum_r = 0.0, sum_i = 0.0;
mlib_d64 *dpx, *dpy, *dpxend;
mlib_d64 dx, dy, dy0, dy1;
mlib_d64 dx_r, dy_r, dy_i;
mlib_d64 d_iih, d_iil, d_irh, d_irl, d_rih, d_ril, d_rrh, d_rrl;
mlib_d64 d_ih, d_il, d_rh, d_rl;
mlib_d64 ds_r, ds_i, ds1_r, ds1_i;
mlib_d64 lb_mask = vis_to_double_dup(0x00FF00FF);
mlib_d64 edge[2];
mlib_f32 fsum;
mlib_s32 d_left;
mlib_s32 emask, off;
mlib_d64 done = vis_to_double_dup(0x1000100);
edge[0] = edge[1] = 0;
dpx = (mlib_d64 *)((mlib_addr)px & (~7));
off = (mlib_addr)dpx - (mlib_addr)px;
dpy = vis_alignaddr((void *)py, off);
pxend = px + n + n - 1;
dpxend = (mlib_d64 *)((mlib_addr)pxend & (~7));
emask = vis_edge8(px, pxend);
vis_pst_8(dpx[0], edge, emask);
dx = edge[0];
dy = vis_ld_d64_nf(dpy);
if (((((mlib_addr)px) ^ ((mlib_addr)py)) & 7) == 0) {
while ((mlib_addr)dpx < (mlib_addr)dpxend) {
d_left = dpxend - dpx;
if (d_left > MAX_LOOP)
d_left = MAX_LOOP;
ds_i = ds_r = ds1_i = ds1_r = 0.0;
for (; d_left > 0; d_left--) {
DPROD_U8C;
SUM_U8C;
dx = dpx[1];
dy = dpy[1];
dpx++;
dpy++;
}
ds_i = vis_fpadd32(ds_i, ds1_i);
ds_r = vis_fpadd32(ds_r, ds1_r);
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
} else {
dy1 = vis_ld_d64_nf(dpy+1);
dy = vis_faligndata(dy, dy1);
while ((mlib_addr)dpx < (mlib_addr)dpxend) {
d_left = dpxend - dpx;
if (d_left > MAX_LOOP)
d_left = MAX_LOOP;
ds_i = ds_r = ds1_i = ds1_r = 0.0;
for (; d_left > 0; d_left--) {
DPROD_U8C;
SUM_U8C;
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy+2);
dx = vis_ld_d64_nf(dpx+1);
dy = vis_faligndata(dy0, dy1);
dpx++;
dpy++;
}
ds_i = vis_fpadd32(ds_i, ds1_i);
ds_r = vis_fpadd32(ds_r, ds1_r);
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
}
if ((mlib_addr)dpx <= (mlib_addr)pxend) {
emask = vis_edge8(dpx, pxend);
vis_pst_8(dx, edge + 1, emask);
dx = edge[1];
DPROD_U8C;
SUM_U8C_TAIL;
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
z[0] = sum_r;
z[1] = sum_i;
#undef MAX_LOOP
}
src = da[0];
da[0] = (src & (~mask)) | (sa[0] & mask);
da++;
sa++;
size = size - 8 + offset;
b_size = size >> 3; /* size in bytes */
/* prepare the destination addresses */
dp = (mlib_d64 *) ((mlib_addr) da & (~7));
j = (mlib_addr) dp - (mlib_addr) da;
dend = da + b_size - 1;
/* prepare the source address */
sp = (mlib_d64 *) VIS_ALIGNADDR(sa, j);
/* generate edge mask for the start point */
emask = vis_edge8(da, dend);
s1 = vis_ld_d64_nf(sp);
if (emask != 0xff) {
s0 = s1;
s1 = vis_ld_d64_nf(sp+1);
s0 = vis_faligndata(s0, s1);
vis_pst_8(s0, dp++, emask);
sp++;
j += 8;
}
#pragma pipeloop(0)
for (; j <= (b_size - 8); j += 8) {
s0 = s1;
s1 = vis_ld_d64_nf(sp+1);
mlib_status
__mlib_VectorNorm_S8_Sat(
mlib_d64 *z,
const mlib_s8 *x,
mlib_s32 n)
{
mlib_s8 *pxend, *px = (mlib_s8 *)x;
mlib_d64 *dpx, *dpxend;
mlib_d64 sum = 0.0;
mlib_d64 dx, dr1, dr2, dr3, dr4, dr5, dr6, dr7, dr8;
mlib_d64 ds1, ds2;
mlib_d64 edge[2];
mlib_d64 fzero = vis_fzero();
mlib_f32 f4ones = vis_to_float(0x01010101);
mlib_f32 fsum;
mlib_s32 d_left;
mlib_s32 emask;
if (n <= 0)
return (MLIB_FAILURE);
edge[0] = edge[1] = 0;
dpx = (mlib_d64 *)((mlib_addr)px & (~7));
pxend = px + n - 1;
dpxend = (mlib_d64 *)((mlib_addr)pxend & (~7));
emask = vis_edge8(px, pxend);
vis_pst_8(dpx[0], edge, emask);
dx = edge[0];
while ((mlib_addr)dpx < (mlib_addr)dpxend) {
d_left = dpxend - dpx;
if (d_left > MAX_LOOP)
d_left = MAX_LOOP;
ds1 = ds2 = 0.0;
for (; d_left > 0; d_left--) {
NORM_S8;
SUM_S8;
dpx++;
dx = dpx[0];
}
fsum = vis_read_hi(ds1);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds1);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds2);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds2);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
}
if ((mlib_addr)dpx <= (mlib_addr)pxend) {
emask = vis_edge8(dpx, pxend);
vis_pst_8(dx, edge + 1, emask);
dx = edge[1];
NORM_S8;
ds1 = vis_fpadd32(dr5, dr6);
ds2 = vis_fpadd32(dr7, dr8);
fsum = vis_read_hi(ds1);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds1);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds2);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds2);
sum += (mlib_d64)*((mlib_s32 *)&fsum);
}
z[0] = mlib_sqrt(sum / 256.0);
return (MLIB_SUCCESS);
#undef MAX_LOOP
}
mlib_status
__mlib_VectorSubS_U8_U8_Sat(
mlib_u8 *z,
const mlib_u8 *x,
const mlib_u8 *c,
mlib_s32 n)
{
/* edge masks */
mlib_s32 emask;
/* offset of address alignment in destination */
mlib_s32 off;
mlib_u8 *pzend;
mlib_d64 *dpx, *dpz, *dpzend;
mlib_d64 dx, dx0, dx1, dr0, dr1, dr;
mlib_u16 cc = *((mlib_u8 *)c);
/* prepare the scaling factors */
mlib_d64 dc = vis_to_double_dup((cc << 4) | (cc << 20));
if (n <= 0)
return (MLIB_FAILURE);
/* initialize GSR scale factor */
vis_write_gsr(3 << 3);
pzend = (mlib_u8 *)z + n - 1;
dpzend = (mlib_d64 *)((mlib_addr)pzend & (~7));
dpz = (mlib_d64 *)((mlib_addr)z & (~7));
off = (mlib_addr)dpz - (mlib_addr)z;
/*
* prepare the source address
*/
dpx = (mlib_d64 *)vis_alignaddr((void *)x, off);
dx0 = vis_ld_d64_nf(dpx);
dpx++;
dx1 = vis_ld_d64_nf(dpx);
dpx++;
/*
* generate edge mask for the start bytes
*/
emask = vis_edge8(z, pzend);
dx = vis_faligndata(dx0, dx1);
SUBS_U8_SAT;
/* store first bytes of result */
vis_pst_8(dr, dpz, emask);
dpz++;
dx0 = dx1;
#pragma pipeloop(0)
for (; (mlib_addr)dpz < (mlib_addr)dpzend; ) {
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUBS_U8_SAT;
(*dpz++) = dr;
dx0 = dx1;
}
if ((mlib_addr)dpz <= (mlib_addr)pzend) {
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUBS_U8_SAT;
/* prepare edge mask for the last bytes */
emask = vis_edge8(dpz, pzend);
/* store last bytes of result */
vis_pst_8(dr, dpz, emask);
}
return (MLIB_SUCCESS);
}
void
mlib_v_VideoColorYUV2RGB444_all_align(
mlib_u8 *rgb,
const mlib_u8 *y,
const mlib_u8 *u,
const mlib_u8 *v,
mlib_s32 size)
{
mlib_u8 *dend;
mlib_f32 *sf0, *sf1, *sf2, *pfd, fzero = vis_fzeros();
mlib_s32 i, n, m, emask;
mlib_d64 *buff2, pbuff_arr2[BUFF_SIZE + 4];
mlib_d64 tmp_arr64[2];
mlib_d64 k01 = vis_to_double_dup(0x0000f375);
mlib_d64 k02 = vis_to_double_dup(0x3317e5fa);
mlib_d64 k11 = vis_to_double_dup(0xf3754097);
mlib_d64 k12 = vis_to_double_dup(0xe5fa0000);
mlib_d64 k21 = vis_to_double_dup(0x40970000);
mlib_d64 k22 = vis_to_double_dup(0x00003317);
mlib_d64 c_0 = vis_to_double_dup(0xe42010f4);
mlib_d64 c_1 = vis_to_double_dup(0x10f4dd60);
mlib_d64 c_2 = vis_to_double_dup(0xdd60e420);
mlib_d64 k_0 = vis_to_double_dup(0x25432543);
do {
/* loop on buffer size */
if (size > 2 * BUFF_SIZE) {
n = 2 * BUFF_SIZE;
} else {
n = size;
}
m = n >> 2;
buff2 = pbuff_arr2;
sf0 = (mlib_f32 *)y;
sf1 = (mlib_f32 *)u;
sf2 = (mlib_f32 *)v;
dend = rgb + 3 * n - 1;
pfd = (mlib_f32 *)rgb;
#pragma pipeloop(0)
for (i = 0; i < m; i++) {
mlib_d64 s00, s01, s02, s10, s11, s12, s20, s21, s22,
s_0;
mlib_f32 x0, x1, x2;
mlib_d64 d_0235, d_xx14, d_23xx, d_0145;
x0 = (*sf0++);
x1 = (*sf1++);
x2 = (*sf2++);
s_0 = vis_fmul8x16(x0, k_0);
s01 = vis_fmul8x16(x1, k01);
s11 = vis_fmul8x16(x1, k11);
s21 = vis_fmul8x16(x1, k21);
s02 = vis_fmul8x16(x2, k02);
s12 = vis_fmul8x16(x2, k12);
s22 = vis_fmul8x16(x2, k22);
s00 = vis_fpadd16(s_0, s01);
s10 = vis_fpadd16(s_0, s11);
s20 = vis_fpadd16(s_0, s21);
s02 = vis_fpadd16(s02, c_0);
s12 = vis_fpadd16(s12, c_1);
s22 = vis_fpadd16(s22, c_2);
s00 = vis_fpadd16(s00, s02);
s10 = vis_fpadd16(s10, s12);
s20 = vis_fpadd16(s20, s22);
d_0235 = vis_fpmerge(vis_fpack16(s00),
vis_fpack16(s10));
d_xx14 = vis_freg_pair(fzero, vis_fpack16(s20));
/*
* merge buff values to 3-channel array
*/
d_23xx = vis_faligndata(d_0235, d_0235);
d_0145 = vis_bshuffle(d_0235, d_xx14);
pfd[0] = vis_read_hi(d_0145);
pfd[1] = vis_read_hi(d_23xx);
pfd[2] = vis_read_lo(d_0145);
buff2 += 2;
pfd += 3;
}
if ((mlib_u8 *)pfd <= dend) {
mlib_d64 d_0235, d_xx14, d_23xx, d_0145;
mlib_f32 *tmp_arr32 = (mlib_f32 *)tmp_arr64;
mlib_d64 s00, s01, s02, s10, s11, s12, s20, s21, s22,
s_0;
mlib_f32 x0, x1, x2;
x0 = (*sf0++);
x1 = (*sf1++);
x2 = (*sf2++);
s_0 = vis_fmul8x16(x0, k_0);
s01 = vis_fmul8x16(x1, k01);
s11 = vis_fmul8x16(x1, k11);
s21 = vis_fmul8x16(x1, k21);
s02 = vis_fmul8x16(x2, k02);
s12 = vis_fmul8x16(x2, k12);
s22 = vis_fmul8x16(x2, k22);
s00 = vis_fpadd16(s_0, s01);
s10 = vis_fpadd16(s_0, s11);
s20 = vis_fpadd16(s_0, s21);
s02 = vis_fpadd16(s02, c_0);
s12 = vis_fpadd16(s12, c_1);
s22 = vis_fpadd16(s22, c_2);
s00 = vis_fpadd16(s00, s02);
s10 = vis_fpadd16(s10, s12);
s20 = vis_fpadd16(s20, s22);
d_0235 = vis_fpmerge(vis_fpack16(s00),
vis_fpack16(s10));
d_xx14 = vis_freg_pair(fzero, vis_fpack16(s20));
d_23xx = vis_faligndata(d_0235, d_0235);
d_0145 = vis_bshuffle(d_0235, d_xx14);
emask = vis_edge8(pfd, dend);
if ((mlib_addr)pfd & 7) {
pfd--;
tmp_arr32++;
}
tmp_arr32[0] = vis_read_hi(d_0145);
tmp_arr32[1] = vis_read_hi(d_23xx);
tmp_arr32[2] = vis_read_lo(d_0145);
vis_pst_8(tmp_arr64[0], pfd, emask);
pfd += 2;
emask = vis_edge8(pfd, dend);
if ((mlib_u8 *)pfd <= dend)
vis_pst_8(tmp_arr64[1], pfd, emask);
}
y += n;
u += n;
v += n;
rgb += 3 * n;
size -= n;
} while (size);
}
mlib_status
__mlib_VideoColorJFIFYCC2RGB444(
mlib_u8 *rgb,
const mlib_u8 *y,
const mlib_u8 *cb,
const mlib_u8 *cr,
mlib_s32 size)
{
mlib_u8 *dend;
mlib_f32 *sf0, *sf1, *sf2, *pfd;
mlib_f32 fzero = vis_fzeros();
mlib_s32 i, n, m, emask;
mlib_d64 tmp_arr64[2];
mlib_d64 k01 = vis_to_double_dup(0x0000f4fd);
mlib_d64 k02 = vis_to_double_dup(0x2cdde926);
mlib_d64 k11 = vis_to_double_dup(0xf4fd38b4);
mlib_d64 k12 = vis_to_double_dup(0xe9260000);
mlib_d64 k21 = vis_to_double_dup(0x38b40000);
mlib_d64 k22 = vis_to_double_dup(0x00002cdd);
mlib_d64 c_0 = vis_to_double_dup(0xe9a110ff);
mlib_d64 c_1 = vis_to_double_dup(0x10ffe3b6);
mlib_d64 c_2 = vis_to_double_dup(0xe3b6e9a1);
mlib_d64 k_0 = vis_to_double_dup(0x20002000);
if (size <= 0)
return (MLIB_FAILURE);
vis_write_gsr((2 << 3) + 2);
vis_write_bmask(0x0489AB37, 0);
do {
/* loop on buffer size */
if (size > 2 * BUFF_SIZE) {
n = 2 * BUFF_SIZE;
} else {
n = size;
}
m = (n - 1) >> 2;
sf0 = (mlib_f32 *)y;
sf1 = (mlib_f32 *)cb;
sf2 = (mlib_f32 *)cr;
dend = rgb + 3 * n - 1;
pfd = (mlib_f32 *)rgb;
#pragma pipeloop(0)
#pragma unroll(4)
for (i = 0; i < m; i++) {
mlib_d64 s00, s01, s02, s10, s11, s12, s20, s21, s22,
s_0;
mlib_d64 d_0235, d_0145;
mlib_f32 x0, x1, x2;
x0 = (*sf0++);
x1 = (*sf1++);
x2 = (*sf2++);
s_0 = vis_fmul8x16(x0, k_0);
s01 = vis_fmul8x16(x1, k01);
s11 = vis_fmul8x16(x1, k11);
s21 = vis_fmul8x16(x1, k21);
s02 = vis_fmul8x16(x2, k02);
s12 = vis_fmul8x16(x2, k12);
s22 = vis_fmul8x16(x2, k22);
s00 = vis_fpadd16(s_0, s01);
s10 = vis_fpadd16(s_0, s11);
s20 = vis_fpadd16(s_0, s21);
s02 = vis_fpadd16(s02, c_0);
s12 = vis_fpadd16(s12, c_1);
s22 = vis_fpadd16(s22, c_2);
s00 = vis_fpadd16(s00, s02);
s10 = vis_fpadd16(s10, s12);
s20 = vis_fpadd16(s20, s22);
d_0235 = vis_fpack16_pair(s00, s10);
s20 = vis_freg_pair(vis_fpack16(s20), fzero);
d_0145 = vis_bshuffle(d_0235, s20);
d_0235 = vis_fpack32(d_0235, d_0235);
d_0235 = vis_fpmerge(vis_read_hi(d_0235),
vis_read_lo(d_0235));
pfd[0] = vis_read_hi(d_0145);
pfd[1] = vis_read_hi(d_0235);
pfd[2] = vis_read_lo(d_0145);
pfd += 3;
}
/*
* last pixels
*/
if ((mlib_u8 *)pfd <= dend) {
mlib_d64 s00, s01, s02, s10, s11, s12, s20, s21, s22,
s_0;
mlib_d64 d_0235, d_xx14, d_0145;
mlib_f32 x0, x1, x2;
mlib_f32 *tmp_arr32 = (mlib_f32 *)tmp_arr64;
x0 = *sf0;
x1 = *sf1;
x2 = *sf2;
s_0 = vis_fmul8x16(x0, k_0);
s01 = vis_fmul8x16(x1, k01);
s11 = vis_fmul8x16(x1, k11);
s21 = vis_fmul8x16(x1, k21);
s02 = vis_fmul8x16(x2, k02);
s12 = vis_fmul8x16(x2, k12);
s22 = vis_fmul8x16(x2, k22);
s00 = vis_fpadd16(s_0, s01);
s10 = vis_fpadd16(s_0, s11);
s20 = vis_fpadd16(s_0, s21);
s02 = vis_fpadd16(s02, c_0);
s12 = vis_fpadd16(s12, c_1);
s22 = vis_fpadd16(s22, c_2);
s00 = vis_fpadd16(s00, s02);
s10 = vis_fpadd16(s10, s12);
s20 = vis_fpadd16(s20, s22);
d_0235 = vis_fpack16_pair(s00, s10);
d_xx14 = vis_freg_pair(vis_fpack16(s20), fzero);
d_0145 = vis_bshuffle(d_0235, d_xx14);
d_0235 = vis_fpack32(d_0235, d_0235);
d_0235 = vis_fpmerge(vis_read_hi(d_0235),
vis_read_lo(d_0235));
emask = vis_edge8(pfd, dend);
if ((mlib_addr)pfd & 7) {
pfd--;
tmp_arr32++;
}
tmp_arr32[0] = vis_read_hi(d_0145);
tmp_arr32[1] = vis_read_hi(d_0235);
tmp_arr32[2] = vis_read_lo(d_0145);
vis_pst_8(tmp_arr64[0], pfd, emask);
pfd += 2;
emask = vis_edge8(pfd, dend);
if ((mlib_u8 *)pfd <= dend)
vis_pst_8(tmp_arr64[1], pfd, emask);
}
y += n;
cb += n;
cr += n;
rgb += 3 * n;
size -= n;
} while (size);
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VideoColorARGB2JFIFYCC422(
mlib_u8 *y,
mlib_u8 *cb,
mlib_u8 *cr,
const mlib_u8 *argb,
mlib_s32 n)
{
mlib_d64 *sp = (mlib_d64 *)argb, *py = (mlib_d64 *)y;
mlib_f32 *pcb = (mlib_f32 *)cb, *pcr = (mlib_f32 *)cr;
mlib_u8 *yend = y + n, *cbend = cb + (n >> 1);
mlib_d64 sd01, sd23, sd45, sd67, sd04, sd26, sd15, sd37;
mlib_d64 dh0, dh1, dl0, dl1, z0, z1;
mlib_s32 i;
mlib_f32 k11 = vis_to_float((mlib_s32)(K11 * 8192));
mlib_f32 k12 = vis_to_float((mlib_s32)(K12 * 8192));
mlib_f32 k13 = vis_to_float((mlib_s32)(K13 * 8192));
mlib_f32 k21 = vis_to_float((mlib_s32)(K21 * 4096));
mlib_f32 k22 = vis_to_float((mlib_s32)(K22 * 4096));
mlib_f32 k23 = vis_to_float((mlib_s32)(K23 * 4096));
mlib_f32 k31 = vis_to_float((mlib_s32)(K31 * 4096));
mlib_f32 k32 = vis_to_float((mlib_s32)(K32 * 4096));
mlib_f32 k33 = vis_to_float((mlib_s32)(K33 * 4096));
mlib_d64 off128 = vis_to_double_dup(0x10101010);
mlib_d64 off0 = vis_to_double_dup(0x00100010);
if (n <= 0)
return (MLIB_FAILURE);
vis_write_gsr(2 << 3);
n = n >> 3;
#pragma pipeloop(0)
for (i = 0; i < n; i++) {
sd01 = (*sp++);
sd23 = (*sp++);
sd45 = (*sp++);
sd67 = (*sp++);
CHANNELSEPARATE_U8_422(sd01, sd23, sd45, sd67, dh0, dh1, dl0,
dl1);
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k11, k12, k13, off0, z0, z1);
z1 = vis_fpadd16(z1, off0);
py[0] = vis_fpmerge(vis_fpack16(z0), vis_fpack16(z1));
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k21, k22, k23, off128, z0, z1);
pcb[0] = vis_fpack16(vis_fpadd16(z0, z1));
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k31, k32, k33, off128, z0, z1);
pcr[0] = vis_fpack16(vis_fpadd16(z0, z1));
py++;
pcb++;
pcr++;
}
if ((mlib_u8 *)pcb < cbend) {
mlib_d64 yd;
mlib_f32 cbf, crf;
mlib_s32 ymask, cmask;
sd01 = (*sp++);
sd23 = vis_ld_d64_nf(sp); sp++;
sd45 = vis_ld_d64_nf(sp); sp++;
sd67 = vis_ld_d64_nf(sp);
CHANNELSEPARATE_U8_422(sd01, sd23, sd45, sd67, dh0, dh1, dl0,
dl1);
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k11, k12, k13, off0, z0, z1);
z1 = vis_fpadd16(z1, off0);
yd = vis_fpmerge(vis_fpack16(z0), vis_fpack16(z1));
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k21, k22, k23, off128, z0, z1);
cbf = vis_fpack16(vis_fpadd16(z0, z1));
CHANNELWEIGHT_U8_2p(vis_read_lo(dh0), vis_read_hi(dh1),
vis_read_lo(dh1), vis_read_lo(dl0), vis_read_hi(dl1),
vis_read_lo(dl1), k31, k32, k33, off128, z0, z1);
crf = vis_fpack16(vis_fpadd16(z0, z1));
ymask = vis_edge8(py, yend - 1);
vis_pst_8(yd, py, ymask);
cmask = vis_edge8(pcb, cbend - 1);
if (cmask & 0xf0) {
vis_pst_8(vis_freg_pair(cbf, vis_fzeros()), pcb, cmask);
vis_pst_8(vis_freg_pair(crf, vis_fzeros()), pcr, cmask);
} else {
vis_pst_8(vis_freg_pair(vis_fzeros(), cbf), pcb - 1,
cmask);
vis_pst_8(vis_freg_pair(vis_fzeros(), crf), pcr - 1,
cmask);
}
}
return (MLIB_SUCCESS);
}
void
mlib_v_ImageLookUp_S16_U8_124_D1(
const mlib_s16 *src,
mlib_u8 *dst,
mlib_s32 xsize,
const mlib_u8 *table0,
const mlib_u8 *table1,
const mlib_u8 *table2,
const mlib_u8 *table3)
{
/* pointer to source data */
mlib_s16 *sp;
/* source data */
mlib_s32 s0, s1, s2, s3;
/* source data */
mlib_s32 s4, s5, s6, s7;
/* pointer to start of destination */
mlib_u8 *dl;
/* pointer to end of destination */
mlib_u8 *dend;
/* aligned pointer to destination */
mlib_d64 *dp;
/* destination data */
mlib_d64 t0, t1, t2;
/* destination data */
mlib_d64 t3, t4, t5;
/* destination data */
mlib_d64 t6, t7, acc0;
/* edge mask */
mlib_s32 emask;
/* loop variable */
mlib_s32 i, num;
/* destination data */
mlib_d64 acc1;
dl = dst;
dp = (mlib_d64 *)dl;
dend = dl + xsize - 1;
sp = (void *)src;
vis_alignaddr((void *)0, 7);
if (xsize >= 8) {
s0 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
s4 = sp[4];
s5 = sp[5];
s6 = sp[6];
s7 = sp[7];
sp += 8;
vis_write_bmask(0x012389ab, 0);
#pragma pipeloop(0)
for (i = 0; i <= xsize - 16; i += 8, sp += 8) {
t7 = VIS_LD_U8_I(table3, s7);
t6 = VIS_LD_U8_I(table2, s6);
t5 = VIS_LD_U8_I(table1, s5);
t4 = VIS_LD_U8_I(table0, s4);
t3 = VIS_LD_U8_I(table3, s3);
t2 = VIS_LD_U8_I(table2, s2);
t1 = VIS_LD_U8_I(table1, s1);
t0 = VIS_LD_U8_I(table0, s0);
acc1 = vis_faligndata(t7, acc1);
acc1 = vis_faligndata(t6, acc1);
acc1 = vis_faligndata(t5, acc1);
acc1 = vis_faligndata(t4, acc1);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
s4 = sp[4];
s5 = sp[5];
s6 = sp[6];
s7 = sp[7];
(*dp++) = vis_bshuffle(acc0, acc1);
}
t7 = VIS_LD_U8_I(table3, s7);
t6 = VIS_LD_U8_I(table2, s6);
t5 = VIS_LD_U8_I(table1, s5);
t4 = VIS_LD_U8_I(table0, s4);
t3 = VIS_LD_U8_I(table3, s3);
t2 = VIS_LD_U8_I(table2, s2);
t1 = VIS_LD_U8_I(table1, s1);
t0 = VIS_LD_U8_I(table0, s0);
acc1 = vis_faligndata(t7, acc1);
acc1 = vis_faligndata(t6, acc1);
acc1 = vis_faligndata(t5, acc1);
acc1 = vis_faligndata(t4, acc1);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
(*dp++) = vis_bshuffle(acc0, acc1);
}
if ((mlib_addr)dp <= (mlib_addr)dend) {
num = (mlib_addr)dend - (mlib_addr)dp;
sp += num;
num++;
if ((num & 3) == 1) {
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc0 = vis_faligndata(t0, acc0);
num--;
} else if ((num & 3) == 2) {
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table1, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc0 = vis_faligndata(t0, acc0);
num -= 2;
} else if ((num & 3) == 3) {
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table2, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table1, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc0 = vis_faligndata(t0, acc0);
num -= 3;
}
if (num != 0) {
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table3, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table2, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table1, s0);
acc0 = vis_faligndata(t0, acc0);
s0 = (mlib_s32)*sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc0 = vis_faligndata(t0, acc0);
}
emask = vis_edge8(dp, dend);
vis_pst_8(acc0, dp, emask);
}
}
static mlib_status
mlib_v_VideoColorYUV2ABGR422_nonalign(
mlib_u8 *abgr,
const mlib_u8 *y,
const mlib_u8 *u,
const mlib_u8 *v,
mlib_s32 width,
mlib_s32 height,
mlib_s32 abgr_stride,
mlib_s32 y_stride,
mlib_s32 uv_stride)
{
/* pointers to src address */
mlib_u8 *sp2, *sp3, *sl2, *sl3;
/* pointers to src address */
mlib_u8 *sp1, *sl1;
/* pointers to dst address */
mlib_u8 *dp, *dl, *dend;
/* all. pointer to y */
mlib_d64 *spy;
/* all. pointer to dst */
mlib_d64 *dpp;
/* u, v data */
mlib_f32 fu0, fu1, fv0, fv1;
/* y data */
mlib_d64 dy0, dy1, dy3;
mlib_d64 du, dv;
/* (1.1644, 1.5966)*8192 */
mlib_f32 k12 = vis_to_float(0x25433317);
/* (-.3920, -.8132)*8192 */
mlib_f32 k34 = vis_to_float(0xf375e5fa);
/* 2.0184*8192 */
mlib_f32 k5 = vis_to_float(0x1004097);
mlib_d64 k_222_9952 = vis_to_double(0x1be01be0, 0x1be01be0);
mlib_d64 k_135_6352 = vis_to_double(0x10f410f4, 0x10f410f4);
mlib_d64 k_276_9856 = vis_to_double(0x22a022a0, 0x22a022a0);
mlib_d64 u_3920_hi, u_20184_hi, v_15966_hi, v_8132_hi;
mlib_d64 u_3920_lo, u_20184_lo, v_15966_lo, v_8132_lo;
mlib_d64 y_11644_hi, y_11644_lo;
mlib_d64 r_hi, r_lo, g_hi, g_lo, b_hi, b_lo;
mlib_d64 temp_r_hi, temp_r_lo, temp_g_hi, temp_g_lo, temp_b_hi,
temp_b_lo;
mlib_f32 red_hi, red_lo, green_hi, green_lo, blue_hi, blue_lo;
mlib_d64 blue_red_hi, x_green_hi, blue_red_lo, x_green_lo;
mlib_d64 dd, dd0, dd1;
/* loop variable */
mlib_s32 i, j;
/* alpha_ch. is not written */
mlib_s32 emask = 0x7777;
mlib_s32 emask1;
mlib_s32 off;
mlib_f32 *dfu, *dfv;
mlib_d64 du0, du1, dv0, dv1;
mlib_s32 off2, off3;
mlib_s32 inc;
/*
* initialize GSR scale factor
*/
vis_write_gsr(2 << 3);
sp1 = sl1 = (mlib_u8 *)y;
sp2 = sl2 = (mlib_u8 *)u;
sp3 = sl3 = (mlib_u8 *)v;
dl = dp = (mlib_u8 *)abgr;
/*
* row loop
*/
for (j = 0; j < height; j++) {
spy = (mlib_d64 *)vis_alignaddr(sp1, 0);
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
dfu = (mlib_f32 *)((mlib_addr)sp2 & ~3);
off2 = (sp2 - (mlib_u8 *)dfu) * 2;
dfv = (mlib_f32 *)((mlib_addr)sp3 & ~3);
off3 = (sp3 - (mlib_u8 *)dfv) * 2;
dend = dp + width * 4 - 1;
emask1 = vis_edge8(dp, dend);
i = dp - (mlib_u8 *)dpp;
emask >>= i;
inc = (emask1 != 0xff);
emask1 &= emask;
off = 8 - i;
vis_alignaddr((void *)off2, 0);
fu0 = vis_ld_f32_nf(dfu); dfu++;
fu1 = vis_ld_f32_nf(dfu); dfu++;
du0 = vis_fpmerge(fu0, fu0);
du1 = vis_fpmerge(fu1, fu1);
du = vis_faligndata(du0, du1);
du0 = du1;
vis_alignaddr((void *)off3, 0);
fv0 = vis_ld_f32_nf(dfv); dfv++;
fv1 = vis_ld_f32_nf(dfv); dfv++;
dv0 = vis_fpmerge(fv0, fv0);
dv1 = vis_fpmerge(fv1, fv1);
dv = vis_faligndata(dv0, dv1);
dv0 = dv1;
/* U*(-0.3920); */
u_3920_hi = vis_fmul8x16au(vis_read_hi(du), k34);
/* V*(-0.8132); */
v_8132_hi = vis_fmul8x16al(vis_read_hi(dv), k34);
/* U*(-0.3920); */
u_3920_lo = vis_fmul8x16au(vis_read_lo(du), k34);
/* V*(-0.8132); */
v_8132_lo = vis_fmul8x16al(vis_read_lo(dv), k34);
vis_alignaddr(sp1, 0);
dy0 = vis_ld_d64_nf(spy); spy++;
dy3 = vis_ld_d64_nf(spy); spy++;
dy1 = vis_faligndata(dy0, dy3);
dy0 = dy3;
/*
* 16-pixel column loop
*/
#pragma pipeloop(0)
for (i = 0; i <= width - 8; i += 8) {
/* U*2.0184 */
u_20184_hi = vis_fmul8x16al(vis_read_hi(du), k5);
g_hi = vis_fpadd16(u_3920_hi, v_8132_hi);
u_20184_lo = vis_fmul8x16al(vis_read_lo(du), k5);
g_hi = vis_fpadd16(g_hi, k_135_6352);
/* V*1.5966 */
v_15966_hi = vis_fmul8x16al(vis_read_hi(dv), k12);
g_lo = vis_fpadd16(u_3920_lo, v_8132_lo);
v_15966_lo = vis_fmul8x16al(vis_read_lo(dv), k12);
g_lo = vis_fpadd16(g_lo, k_135_6352);
/* Y*1.1644 */
y_11644_hi = vis_fmul8x16au(vis_read_hi(dy1), k12);
b_hi = vis_fpsub16(u_20184_hi, k_276_9856);
/* Y*1.1644 */
y_11644_lo = vis_fmul8x16au(vis_read_lo(dy1), k12);
b_lo = vis_fpsub16(u_20184_lo, k_276_9856);
r_hi = vis_fpsub16(v_15966_hi, k_222_9952);
r_lo = vis_fpsub16(v_15966_lo, k_222_9952);
temp_g_hi = vis_fpadd16(g_hi, y_11644_hi);
temp_b_hi = vis_fpadd16(b_hi, y_11644_hi);
green_hi = vis_fpack16(temp_g_hi);
temp_r_hi = vis_fpadd16(r_hi, y_11644_hi);
blue_hi = vis_fpack16(temp_b_hi);
temp_g_lo = vis_fpadd16(g_lo, y_11644_lo);
red_hi = vis_fpack16(temp_r_hi);
temp_b_lo = vis_fpadd16(b_lo, y_11644_lo);
vis_alignaddr((void *)off2, 0);
fu1 = vis_ld_f32_nf(dfu); dfu++;
du1 = vis_fpmerge(fu1, fu1);
du = vis_faligndata(du0, du1);
du0 = du1;
green_lo = vis_fpack16(temp_g_lo);
temp_r_lo = vis_fpadd16(r_lo, y_11644_lo);
blue_lo = vis_fpack16(temp_b_lo);
x_green_hi = vis_fmul8x16au(green_hi, k5);
red_lo = vis_fpack16(temp_r_lo);
blue_red_hi = vis_fpmerge(blue_hi, red_hi);
x_green_lo = vis_fmul8x16au(green_lo, k5);
blue_red_lo = vis_fpmerge(blue_lo, red_lo);
vis_alignaddr((void *)off3, 0);
fv1 = vis_ld_f32_nf(dfv); dfv++;
dv1 = vis_fpmerge(fv1, fv1);
dv = vis_faligndata(dv0, dv1);
dv0 = dv1;
vis_alignaddr((void *)off, 0);
/* U*(-0.3920); */
u_3920_hi = vis_fmul8x16au(vis_read_hi(du), k34);
dd1 = vis_fpmerge(vis_read_hi(x_green_hi),
vis_read_hi(blue_red_hi));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
dpp += inc;
inc = 1;
/* V*(-0.8132); */
v_8132_hi = vis_fmul8x16al(vis_read_hi(dv), k34);
dd0 = vis_fpmerge(vis_read_lo(x_green_hi),
vis_read_lo(blue_red_hi));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
u_3920_lo = vis_fmul8x16au(vis_read_lo(du), k34);
dd1 = vis_fpmerge(vis_read_hi(x_green_lo),
vis_read_hi(blue_red_lo));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
v_8132_lo = vis_fmul8x16al(vis_read_lo(dv), k34);
dd0 = vis_fpmerge(vis_read_lo(x_green_lo),
vis_read_lo(blue_red_lo));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
vis_alignaddr(sp1, 0);
dy3 = vis_ld_d64_nf(spy); spy++;
dy1 = vis_faligndata(dy0, dy3);
dy0 = dy3;
emask1 = emask;
}
if (i < width) {
vis_alignaddr((void *)off, 0);
/* U*2.0184 */
u_20184_hi = vis_fmul8x16al(vis_read_hi(du), k5);
g_hi = vis_fpadd16(u_3920_hi, v_8132_hi);
u_20184_lo = vis_fmul8x16al(vis_read_lo(du), k5);
g_hi = vis_fpadd16(g_hi, k_135_6352);
/* V*1.5966 */
v_15966_hi = vis_fmul8x16al(vis_read_hi(dv), k12);
g_lo = vis_fpadd16(u_3920_lo, v_8132_lo);
v_15966_lo = vis_fmul8x16al(vis_read_lo(dv), k12);
g_lo = vis_fpadd16(g_lo, k_135_6352);
/* Y*1.1644 */
y_11644_hi = vis_fmul8x16au(vis_read_hi(dy1), k12);
b_hi = vis_fpsub16(u_20184_hi, k_276_9856);
/* Y*1.1644 */
y_11644_lo = vis_fmul8x16au(vis_read_lo(dy1), k12);
b_lo = vis_fpsub16(u_20184_lo, k_276_9856);
r_hi = vis_fpsub16(v_15966_hi, k_222_9952);
r_lo = vis_fpsub16(v_15966_lo, k_222_9952);
temp_g_hi = vis_fpadd16(g_hi, y_11644_hi);
temp_b_hi = vis_fpadd16(b_hi, y_11644_hi);
green_hi = vis_fpack16(temp_g_hi);
temp_r_hi = vis_fpadd16(r_hi, y_11644_hi);
blue_hi = vis_fpack16(temp_b_hi);
temp_g_lo = vis_fpadd16(g_lo, y_11644_lo);
red_hi = vis_fpack16(temp_r_hi);
temp_b_lo = vis_fpadd16(b_lo, y_11644_lo);
green_lo = vis_fpack16(temp_g_lo);
temp_r_lo = vis_fpadd16(r_lo, y_11644_lo);
blue_lo = vis_fpack16(temp_b_lo);
x_green_hi = vis_fmul8x16au(green_hi, k5);
red_lo = vis_fpack16(temp_r_lo);
blue_red_hi = vis_fpmerge(blue_hi, red_hi);
x_green_lo = vis_fmul8x16au(green_lo, k5);
blue_red_lo = vis_fpmerge(blue_lo, red_lo);
dd1 = vis_fpmerge(vis_read_hi(x_green_hi),
vis_read_hi(blue_red_hi));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
dd0 = dd1;
dpp += inc;
i += 2;
if (i < width) {
dd1 = vis_fpmerge(vis_read_lo(x_green_hi),
vis_read_lo(blue_red_hi));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = dd1;
i += 2;
if (i < width) {
dd1 = vis_fpmerge(vis_read_hi
(x_green_lo),
vis_read_hi(blue_red_lo));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = dd1;
}
}
}
vis_alignaddr((void *)off, 0);
emask1 = vis_edge8(dpp, dend);
emask1 &= emask;
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
sp1 = sl1 = sl1 + y_stride;
sp2 = sl2 = sl2 + uv_stride;
sp3 = sl3 = sl3 + uv_stride;
dl = dp = dl + abgr_stride;
emask = 0x7777;
}
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorSub_U8_U8_Mod(
mlib_u8 *z,
const mlib_u8 *x,
const mlib_u8 *y,
mlib_s32 n)
{
/* 8-byte aligned start point in destination */
mlib_d64 *dpz;
/* 8-byte aligned start point in source */
mlib_d64 *dpx, *dpy;
/* source data */
mlib_d64 dx, dy, dx0;
mlib_d64 dx1, dy0, dy1;
/* destination data */
mlib_d64 dz;
/* intermediate result */
mlib_d64 dh, dl;
mlib_d64 dxl, dyl;
/* end point of a line in destination */
mlib_u8 *pzend;
/* start point of a line in source */
mlib_u8 *px, *py;
/* offset of address alignment in destination */
mlib_s32 off;
/* edge mask */
mlib_s32 emask;
mlib_u8 *pzend16;
mlib_s32 sr1, sr2, sr3;
mlib_s32 x8, x12, y8, y12;
mlib_s32 mask = 0x7f7f7f7f;
mlib_u8 *pz;
mlib_s32 n16;
mlib_s32 nrest;
mlib_s32 len = n, i;
/* rest and leng in terms of 8 bytes. */
mlib_s32 rest_8, even_8;
mlib_d64 mask_control = vis_to_double_dup(0xff00ff00);
if (n <= 0)
return (MLIB_FAILURE);
px = (mlib_u8 *)x;
py = (mlib_u8 *)y;
pz = (mlib_u8 *)z;
/*
* prepare the destination address
*/
pzend = pz + n - 1;
/*
* check for 64-bit aligned special case
*/
if ((((mlib_addr)x | (mlib_addr)y | (mlib_addr)z) & 7) == 0) {
/*
* We can process source and destination vectors by 16 bytes.
*/
dpx = (mlib_d64 *)x;
dx = vis_ld_d64_nf(dpx);
dpy = (mlib_d64 *)y;
dy = vis_ld_d64_nf(dpy);
dpz = (mlib_d64 *)z;
n16 = n & (~0xf);
pzend16 = pz + n16;
#pragma pipeloop(0)
while ((mlib_addr)pz < (mlib_addr)pzend16) {
x8 = *((mlib_s32 *)(px + 8));
y8 = *((mlib_s32 *)(py + 8));
sr1 = x8 ^ ~y8;
sr2 = (x8 | ~mask) - (y8 & mask);
sr3 = (sr1 & ~mask) ^ sr2;
*((mlib_s32 *)(pz + 8)) = sr3;
x12 = *((mlib_s32 *)(px + 12));
y12 = *((mlib_s32 *)(py + 12));
sr1 = x12 ^ ~y12;
sr2 = (x12 | ~mask) - (y12 & mask);
sr3 = (sr1 & ~mask) ^ sr2;
*((mlib_s32 *)(pz + 12)) = sr3;
SUB_S8_MOD;
/* store 8 bytes of result */
*((mlib_d64 *)pz) = dz;
dx = vis_ld_d64_nf(px + 16);
dy = vis_ld_d64_nf(py + 16);
px += 16;
py += 16;
pz += 16;
}
dpz = (mlib_d64 *)pzend16;
nrest = n - n16;
if (nrest >= 8) {
SUB_S8_MOD;
dpz[0] = dz;
px += 8;
py += 8;
dpz++;
nrest -= 8;
}
if (nrest > 0) {
dx = *((mlib_d64 *)px);
dy = *((mlib_d64 *)py);
SUB_S8_MOD;
emask = vis_edge8(dpz, pzend);
vis_pst_8(dz, dpz, emask);
}
} else {
/*
* General case.
*/
dpz = (mlib_d64 *)((mlib_addr)z & (~7));
off = (mlib_addr)dpz - (mlib_addr)z;
/*
* generate edge mask for the start point
*/
emask = vis_edge8(pz, pzend);
/*
* prepare the source address
*/
if (off) {
dpy = (mlib_d64 *)vis_alignaddr(py, off);
dy0 = vis_ld_d64_nf(dpy);
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
dpx = (mlib_d64 *)vis_alignaddr(px, off);
dx0 = vis_ld_d64_nf(dpx);
dx1 = vis_ld_d64_nf(dpx + 1);
dx = vis_faligndata(dx0, dx1);
SUB_S8_MOD;
/*
* store first bytes of result
*/
vis_pst_8(dz, dpz, emask);
px += (8 + off);
py += (8 + off);
len -= (8 + off);
dpz++;
if (len <= 0)
return (MLIB_SUCCESS);
}
even_8 = len >> 3;
rest_8 = len & 0x7;
/*
* Now try to analyze source "x" and "y" addresses.
*/
if ((!((mlib_addr)px & 7)) && (!((mlib_addr)py & 7))) {
/*
* Both addresses are 8-byte aligned. No vis_alignaddr
* and vis_faligndata at all.
*/
dpx = (mlib_d64 *)px;
dpy = (mlib_d64 *)py;
dx = vis_ld_d64_nf(dpx);
dpx++;
dy = vis_ld_d64_nf(dpy);
dpy++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dx1 = vis_ld_d64_nf(dpx);
dy1 = vis_ld_d64_nf(dpy);
SUB_S8_MOD;
dx = dx1;
dy = dy1;
/*
* store 8 bytes of result
*/
dpz[0] = dz;
dpx++;
dpy++;
dpz++;
}
dx1 = dx;
dy1 = dy;
} else if ((!((mlib_addr)px & 7))) {
/*
* First ("x") address is 8-byte aligned. vis_alignaddr
* and vis_faligndata only for "y".
*/
dpx = (mlib_d64 *)px;
dpy = vis_alignaddr(py, 0);
dy0 = vis_ld_d64_nf(dpy);
dpy++;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
dx = vis_ld_d64_nf(dpx);
dpx++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
SUB_S8_MOD;
dx = vis_ld_d64_nf(dpx);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
dpx++;
dpy++;
}
dx1 = dx;
dy1 = dy0;
} else if ((!((mlib_addr)py & 7))) {
/*
* Second ("y") address is 8-byte aligned. vis_alignaddr
* and vis_faligndata only for "x".
*/
dpy = (mlib_d64 *)py;
dpx = vis_alignaddr(px, 0);
dx1 = vis_ld_d64_nf(dpx);
dpx++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dy = (*dpy++);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUB_S8_MOD;
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
}
dy1 = vis_ld_d64_nf(dpy);
dpy++;
} else if (((mlib_addr)px & 7) == ((mlib_addr)py & 7)) {
/*
* Both ("x" and "y") address are identically aligned.
* There are 1 vis_alignaddr and 2 vis_faligndata(s) in the loop.
*/
dpx = vis_alignaddr(px, 0);
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dpy = vis_alignaddr(py, 0);
dy1 = vis_ld_d64_nf(dpy);
dpy++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy);
dpy++;
dy = vis_faligndata(dy0, dy1);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUB_S8_MOD;
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
}
} else {
/*
* Both ("x" and "y") address are arbitrary aligned.
* 2 vis_alignaddr(s) and 2 vis_faligndata(s) in the loop.
*/
dpx = vis_alignaddr(px, 0);
dx0 = vis_ld_d64_nf(dpx);
dpx++;
dx1 = vis_ld_d64_nf(dpx);
dx = vis_faligndata(dx0, dx1);
dpy = vis_alignaddr(py, 0);
dy0 = vis_ld_d64_nf(dpy);
dpy++;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
SUB_S8_MOD;
vis_alignaddr(py, 0);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
vis_alignaddr(px, 0);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx + 1);
dx = vis_faligndata(dx0, dx1);
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
dpy++;
dpx++;
}
dx1 = dx0;
dy1 = dy0;
}
if (!rest_8)
return (MLIB_SUCCESS);
vis_alignaddr(px, 0);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dx = vis_faligndata(dx0, dx1);
vis_alignaddr(py, 0);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
SUB_S8_MOD;
/*
* prepare edge mask for the last bytes
*/
emask = vis_edge8((void *)(rest_8), pzend);
/* store last bytes of result */
vis_pst_8(dz, dpz, ~emask);
}
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorSub_S8_S8_Sat(
mlib_s8 *z,
const mlib_s8 *x,
const mlib_s8 *y,
mlib_s32 n)
{
mlib_d64 *dpz, *dpx, *dpy;
mlib_d64 dx, dy, dz, dx0, dx1, dy0, dy1;
mlib_d64 dxh, dxl, dyh, dyl, dzh, dzl;
mlib_d64 dh, dl;
mlib_s8 *pz = z, *px, *py, *pzend;
/* offset of address alignment in destination */
mlib_s32 off;
mlib_s32 len = n, i;
/* rest and leng in terms of 8 bytes. */
mlib_s32 rest_8, even_8;
/* edge masks */
mlib_s32 emask;
mlib_d64 displacement = vis_to_double_dup(0x8000800);
mlib_d64 restore = vis_to_double_dup(0x80808080);
mlib_f32 fmul = vis_to_float(0x1000);
if (n <= 0)
return (MLIB_FAILURE);
px = (mlib_s8 *)x;
py = (mlib_s8 *)y;
/* initialize GSR scale factor */
vis_write_gsr(3 << 3);
dpz = (mlib_d64 *)((mlib_addr)z & (~7));
off = (mlib_addr)dpz - (mlib_addr)z;
pzend = pz + n - 1;
/*
* generate edge mask for the start point
*/
emask = vis_edge8(pz, pzend);
/*
* prepare the source address
*/
if (off) {
dpy = (mlib_d64 *)vis_alignaddr(py, off);
dy0 = vis_ld_d64_nf(dpy);
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
dpx = (mlib_d64 *)vis_alignaddr(px, off);
dx0 = vis_ld_d64_nf(dpx);
dx1 = vis_ld_d64_nf(dpx + 1);
dx = vis_faligndata(dx0, dx1);
SUB_S8_SAT;
/*
* store first bytes of result
*/
vis_pst_8(dz, dpz, emask);
px += (8 + off);
py += (8 + off);
len -= (8 + off);
dpz++;
if (len <= 0)
return (MLIB_SUCCESS);
}
even_8 = len >> 3;
rest_8 = len & 0x7;
/*
* Now try to analyze source "x" and "y" addresses.
*/
if ((!((mlib_addr)px & 7)) && (!((mlib_addr)py & 7))) {
/*
* Both addresses are 8-byte aligned. No vis_alignaddr
* and vis_faligndata at all.
*/
dpx = (mlib_d64 *)px;
dpy = (mlib_d64 *)py;
dx = vis_ld_d64_nf(dpx);
dpx++;
dy = vis_ld_d64_nf(dpy);
dpy++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dx1 = vis_ld_d64_nf(dpx);
dy1 = vis_ld_d64_nf(dpy);
SUB_S8_SAT;
dx = dx1;
dy = dy1;
/*
* store 8 bytes of result
*/
dpz[0] = dz;
dpx++;
dpy++;
dpz++;
}
dx1 = dx;
dy1 = dy;
} else if ((!((mlib_addr)px & 7))) {
/*
* First ("x") address is 8-byte aligned. vis_alignaddr
* and vis_faligndata only for "y".
*/
dpx = (mlib_d64 *)px;
dpy = vis_alignaddr(py, 0);
dy0 = vis_ld_d64_nf(dpy);
dpy++;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
dx = vis_ld_d64_nf(dpx);
dpx++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
SUB_S8_SAT;
dx = vis_ld_d64_nf(dpx);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
dpx++;
dpy++;
}
dx1 = dx;
dy1 = dy0;
} else if ((!((mlib_addr)py & 7))) {
/*
* Second ("y") address is 8-byte aligned. vis_alignaddr
* and vis_faligndata only for "x".
*/
dpy = (mlib_d64 *)py;
dpx = vis_alignaddr(px, 0);
dx1 = vis_ld_d64_nf(dpx);
dpx++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dy = (*dpy++);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUB_S8_SAT;
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
}
dy1 = vis_ld_d64_nf(dpy);
dpy++;
} else if (((mlib_addr)px & 7) == ((mlib_addr)py & 7)) {
/*
* Both ("x" and "y") address are identically aligned.
* There are 1 vis_alignaddr and 2 vis_faligndata(s) in the loop.
*/
dpx = vis_alignaddr(px, 0);
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dpy = vis_alignaddr(py, 0);
dy1 = vis_ld_d64_nf(dpy);
dpy++;
#pragma pipeloop(0)
for (i = 0; i < even_8; i++) {
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy);
dpy++;
dy = vis_faligndata(dy0, dy1);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dpx++;
dx = vis_faligndata(dx0, dx1);
SUB_S8_SAT;
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
}
} else {
/*
* Both ("x" and "y") address are arbitrary aligned.
* 2 vis_alignaddr(s) and 2 vis_faligndata(s) in the loop.
*/
dpx = vis_alignaddr(px, 0);
dx0 = vis_ld_d64_nf(dpx);
dpx++;
dx1 = vis_ld_d64_nf(dpx);
dx = vis_faligndata(dx0, dx1);
dpy = vis_alignaddr(py, 0);
dy0 = vis_ld_d64_nf(dpy);
dpy++;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
/* #pragma pipeloop(0) */
for (i = 0; i < even_8; i++) {
SUB_S8_SAT;
vis_alignaddr(py, 0);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy + 1);
dy = vis_faligndata(dy0, dy1);
vis_alignaddr(px, 0);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx + 1);
dx = vis_faligndata(dx0, dx1);
/*
* store 8 bytes of result
*/
(*dpz++) = dz;
dpy++;
dpx++;
}
dx1 = dx0;
dy1 = dy0;
}
if (!rest_8)
return (MLIB_SUCCESS);
vis_alignaddr(px, 0);
dx0 = dx1;
dx1 = vis_ld_d64_nf(dpx);
dx = vis_faligndata(dx0, dx1);
vis_alignaddr(py, 0);
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy);
dy = vis_faligndata(dy0, dy1);
SUB_S8_SAT;
/*
* prepare edge mask for the last bytes
*/
emask = vis_edge8((void *)(rest_8), pzend);
/* store last bytes of result */
vis_pst_8(dz, dpz, ~emask);
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VideoUpSample420(
mlib_u8 *dst0,
mlib_u8 *dst1,
const mlib_u8 *src0,
const mlib_u8 *src1,
const mlib_u8 *src2,
mlib_s32 n)
{
mlib_u8 *dend0 = dst0 + 2 * n - 1;
mlib_d64 *dp0 = (mlib_d64 *)dst0;
mlib_d64 *dp1 = (mlib_d64 *)dst1;
mlib_d64 *sp0 = (mlib_d64 *)src0;
mlib_d64 *sp1 = (mlib_d64 *)src1;
mlib_d64 *sp2 = (mlib_d64 *)src2;
mlib_d64 d00, d01, d10, d11, d20, d21;
mlib_d64 thiscolsum0_hi, thiscolsum0_lo, lastcolsum0_hi, lastcolsum0_lo;
mlib_d64 shiftcolsum0_hi, shiftcolsum0_lo;
mlib_d64 thiscolsum1_hi, thiscolsum1_lo, lastcolsum1_hi, lastcolsum1_lo;
mlib_d64 shiftcolsum1_hi, shiftcolsum1_lo;
mlib_d64 acc0, acc1, acc2, acc3, acc4, acc5, acc6, acc7;
mlib_d64 ac0, ac1, ac2, ac3, ac4, ac5, ac6, ac7;
mlib_d64 data0, data1, data2, data3, tmp0, tmp1;
mlib_f32 fone = vis_to_float(0x4000000);
mlib_f32 fthree = vis_to_float(0xC000000);
mlib_f32 fone1 = vis_to_float(0x40404040);
mlib_f32 fthree1 = vis_to_float(0xC0C0C0C0);
mlib_d64 dseven = vis_to_double_dup(0x70007);
mlib_d64 deight = vis_to_double_dup(0x80008);
mlib_s32 i, emask;
if (n <= 0)
return (MLIB_FAILURE);
vis_write_gsr((3 << 3) + 2);
d00 = vis_ld_d64_nf(sp0);
d10 = vis_ld_d64_nf(sp1);
d20 = vis_ld_d64_nf(sp2);
sp0++;
sp1++;
sp2++;
lastcolsum0_hi = vis_fmul8x16au(vis_read_hi(d00), fone);
lastcolsum0_lo = vis_fmul8x16au(vis_read_lo(d00), fone);
lastcolsum1_hi = vis_fmul8x16au(vis_read_hi(d20), fone);
lastcolsum1_lo = vis_fmul8x16au(vis_read_lo(d20), fone);
tmp0 = vis_fmul8x16au(vis_read_hi(d10), fthree);
tmp1 = vis_fmul8x16au(vis_read_lo(d10), fthree);
lastcolsum0_hi = vis_fpadd16(lastcolsum0_hi, tmp0);
lastcolsum0_lo = vis_fpadd16(lastcolsum0_lo, tmp1);
lastcolsum1_hi = vis_fpadd16(lastcolsum1_hi, tmp0);
lastcolsum1_lo = vis_fpadd16(lastcolsum1_lo, tmp1);
#pragma pipeloop(0)
for (i = 0; i < n - 8; i += 8) {
d01 = *sp0;
d11 = *sp1;
d21 = *sp2;
sp0++;
sp1++;
sp2++;
thiscolsum0_hi = vis_fmul8x16au(vis_read_hi(d01), fone);
thiscolsum0_lo = vis_fmul8x16au(vis_read_lo(d01), fone);
thiscolsum1_hi = vis_fmul8x16au(vis_read_hi(d21), fone);
thiscolsum1_lo = vis_fmul8x16au(vis_read_lo(d21), fone);
tmp0 = vis_fmul8x16au(vis_read_hi(d11), fthree);
tmp1 = vis_fmul8x16au(vis_read_lo(d11), fthree);
thiscolsum0_hi = vis_fpadd16(thiscolsum0_hi, tmp0);
thiscolsum0_lo = vis_fpadd16(thiscolsum0_lo, tmp1);
thiscolsum1_hi = vis_fpadd16(thiscolsum1_hi, tmp0);
thiscolsum1_lo = vis_fpadd16(thiscolsum1_lo, tmp1);
acc0 = vis_fmul8x16(fone1, lastcolsum0_hi);
acc1 = vis_fmul8x16(fone1, lastcolsum0_lo);
acc2 = vis_fmul8x16(fthree1, lastcolsum0_hi);
acc3 = vis_fmul8x16(fthree1, lastcolsum0_lo);
acc4 = vis_fmul8x16(fone1, lastcolsum1_hi);
acc5 = vis_fmul8x16(fone1, lastcolsum1_lo);
acc6 = vis_fmul8x16(fthree1, lastcolsum1_hi);
acc7 = vis_fmul8x16(fthree1, lastcolsum1_lo);
shiftcolsum0_hi =
vis_faligndata(lastcolsum0_hi, lastcolsum0_lo);
shiftcolsum0_lo =
vis_faligndata(lastcolsum0_lo, thiscolsum0_hi);
shiftcolsum1_hi =
vis_faligndata(lastcolsum1_hi, lastcolsum1_lo);
shiftcolsum1_lo =
vis_faligndata(lastcolsum1_lo, thiscolsum1_hi);
acc0 = vis_fpadd16(acc0, deight);
acc1 = vis_fpadd16(acc1, deight);
acc2 = vis_fpadd16(acc2, dseven);
acc3 = vis_fpadd16(acc3, dseven);
acc4 = vis_fpadd16(acc4, deight);
acc5 = vis_fpadd16(acc5, deight);
acc6 = vis_fpadd16(acc6, dseven);
acc7 = vis_fpadd16(acc7, dseven);
ac0 = vis_fmul8x16(fthree1, shiftcolsum0_hi);
ac1 = vis_fmul8x16(fthree1, shiftcolsum0_lo);
ac2 = vis_fmul8x16(fone1, shiftcolsum0_hi);
ac3 = vis_fmul8x16(fone1, shiftcolsum0_lo);
ac4 = vis_fmul8x16(fthree1, shiftcolsum1_hi);
ac5 = vis_fmul8x16(fthree1, shiftcolsum1_lo);
ac6 = vis_fmul8x16(fone1, shiftcolsum1_hi);
ac7 = vis_fmul8x16(fone1, shiftcolsum1_lo);
acc0 = vis_fpadd16(acc0, ac0);
acc1 = vis_fpadd16(acc1, ac1);
acc2 = vis_fpadd16(acc2, ac2);
acc3 = vis_fpadd16(acc3, ac3);
acc4 = vis_fpadd16(acc4, ac4);
acc5 = vis_fpadd16(acc5, ac5);
acc6 = vis_fpadd16(acc6, ac6);
acc7 = vis_fpadd16(acc7, ac7);
data0 = vis_fpack16_pair(acc0, acc1);
data1 = vis_fpack16_pair(acc2, acc3);
data2 = vis_fpack16_pair(acc4, acc5);
data3 = vis_fpack16_pair(acc6, acc7);
dp0[0] = vis_fpmerge(vis_read_hi(data1), vis_read_hi(data0));
dp0[1] = vis_fpmerge(vis_read_lo(data1), vis_read_lo(data0));
dp1[0] = vis_fpmerge(vis_read_hi(data3), vis_read_hi(data2));
dp1[1] = vis_fpmerge(vis_read_lo(data3), vis_read_lo(data2));
dp0 += 2;
dp1 += 2;
lastcolsum0_hi = thiscolsum0_hi;
lastcolsum0_lo = thiscolsum0_lo;
lastcolsum1_hi = thiscolsum1_hi;
lastcolsum1_lo = thiscolsum1_lo;
}
if (i < n) {
acc0 = vis_fmul8x16(fone1, lastcolsum0_hi);
acc1 = vis_fmul8x16(fone1, lastcolsum0_lo);
acc2 = vis_fmul8x16(fthree1, lastcolsum0_hi);
acc3 = vis_fmul8x16(fthree1, lastcolsum0_lo);
acc4 = vis_fmul8x16(fone1, lastcolsum1_hi);
acc5 = vis_fmul8x16(fone1, lastcolsum1_lo);
acc6 = vis_fmul8x16(fthree1, lastcolsum1_hi);
acc7 = vis_fmul8x16(fthree1, lastcolsum1_lo);
shiftcolsum0_hi =
vis_faligndata(lastcolsum0_hi, lastcolsum0_lo);
shiftcolsum0_lo =
vis_faligndata(lastcolsum0_lo, lastcolsum0_lo);
shiftcolsum1_hi =
vis_faligndata(lastcolsum1_hi, lastcolsum1_lo);
shiftcolsum1_lo =
vis_faligndata(lastcolsum1_lo, lastcolsum1_lo);
acc0 = vis_fpadd16(acc0, deight);
acc1 = vis_fpadd16(acc1, deight);
acc2 = vis_fpadd16(acc2, dseven);
acc3 = vis_fpadd16(acc3, dseven);
acc4 = vis_fpadd16(acc4, deight);
acc5 = vis_fpadd16(acc5, deight);
acc6 = vis_fpadd16(acc6, dseven);
acc7 = vis_fpadd16(acc7, dseven);
ac0 = vis_fmul8x16(fthree1, shiftcolsum0_hi);
ac1 = vis_fmul8x16(fthree1, shiftcolsum0_lo);
ac2 = vis_fmul8x16(fone1, shiftcolsum0_hi);
ac3 = vis_fmul8x16(fone1, shiftcolsum0_lo);
ac4 = vis_fmul8x16(fthree1, shiftcolsum1_hi);
ac5 = vis_fmul8x16(fthree1, shiftcolsum1_lo);
ac6 = vis_fmul8x16(fone1, shiftcolsum1_hi);
ac7 = vis_fmul8x16(fone1, shiftcolsum1_lo);
acc0 = vis_fpadd16(acc0, ac0);
acc1 = vis_fpadd16(acc1, ac1);
acc2 = vis_fpadd16(acc2, ac2);
acc3 = vis_fpadd16(acc3, ac3);
acc4 = vis_fpadd16(acc4, ac4);
acc5 = vis_fpadd16(acc5, ac5);
acc6 = vis_fpadd16(acc6, ac6);
acc7 = vis_fpadd16(acc7, ac7);
data0 = vis_fpack16_pair(acc0, acc1);
data1 = vis_fpack16_pair(acc2, acc3);
data2 = vis_fpack16_pair(acc4, acc5);
data3 = vis_fpack16_pair(acc6, acc7);
acc0 = vis_fpmerge(vis_read_hi(data1), vis_read_hi(data0));
acc1 = vis_fpmerge(vis_read_hi(data3), vis_read_hi(data2));
emask = vis_edge8(dp0, dend0);
vis_pst_8(acc0, dp0, emask);
vis_pst_8(acc1, dp1, emask);
i += 4;
dp0++;
dp1++;
if (i < n) {
acc0 = vis_fpmerge(vis_read_lo(data1),
vis_read_lo(data0));
acc1 = vis_fpmerge(vis_read_lo(data3),
vis_read_lo(data2));
emask = vis_edge8(dp0, dend0);
vis_pst_8(acc0, dp0, emask);
vis_pst_8(acc1, dp1, emask);
}
}
vis_write_gsr(7);
dp0 = (mlib_d64 *)dst0;
dp1 = (mlib_d64 *)dst1;
ac0 = *dp0;
ac2 = *dp1;
#pragma pipeloop(0)
for (i = 0; i < 2 * n - 8; i += 8) {
ac1 = *dp0;
ac3 = *dp1;
*dp0 = vis_faligndata(ac0, ac1);
*dp1 = vis_faligndata(ac2, ac3);
dp0++;
dp1++;
ac0 = ac1;
ac2 = ac3;
}
if (i < 2 * n) {
ac1 = vis_ld_d64_nf(dp0);
ac3 = vis_ld_d64_nf(dp1);
emask = vis_edge8(dp0, dend0);
acc0 = vis_faligndata(ac0, ac1);
acc1 = vis_faligndata(ac2, ac3);
vis_pst_8(acc0, dp0, emask);
vis_pst_8(acc1, dp1, emask);
}
dst0[0] = (4 * (3 * src1[0] + src0[0]) + 8) >> 4;
dst1[0] = (4 * (3 * src1[0] + src2[0]) + 8) >> 4;
dst0[2 * n - 1] = (4 * (3 * src1[n - 1] + src0[n - 1]) + 7) >> 4;
dst1[2 * n - 1] = (4 * (3 * src1[n - 1] + src2[n - 1]) + 7) >> 4;
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorSubS_U8_U8_Mod(
mlib_u8 *z,
const mlib_u8 *x,
const mlib_u8 *c,
mlib_s32 n)
{
/* edge masks */
mlib_s32 emask;
/* offset of address alignment in destination */
mlib_s32 off;
mlib_s8 *pzend;
mlib_d64 *dpx, *dpz, *dpzend;
mlib_d64 dx, dx0, dx1, dr0, dr1, dr;
mlib_u8 uc = *((mlib_s8 *)c);
mlib_d64 uncontrol_mask = vis_to_double_dup(0xff00ff00);
/* prepare the scaling factors */
mlib_d64 dcl = vis_to_double_dup(uc | (uc << 16));
mlib_d64 dch = vis_to_double_dup((uc << 8) | (uc << 24));
mlib_s32 scal = uc << 24 | uc << 16 | uc << 8 | uc;
mlib_s32 sr1, sr2, sr3, sr3_;
mlib_s32 mask = 0x7f7f7f7f;
mlib_s32 x8, x12;
mlib_s32 nrest, i;
if (n <= 0)
return (MLIB_FAILURE);
pzend = (mlib_s8 *)z + n - 1;
dpzend = (mlib_d64 *)((mlib_addr)pzend & (~7));
/*
* check for 64-bit aligned special case
*/
if ((((mlib_addr)x | (mlib_addr)z) & 7) == 0) {
/*
* We can process source and destination vectors by 16 bytes.
*/
dpx = (mlib_d64 *)x;
dpz = (mlib_d64 *)z;
#pragma pipeloop(0)
for (i = 0; i < n >> 4; i++) {
mlib_u64 ld0;
dx = dpx[0];
SUBS_S8_MOD;
(*dpz++) = dr;
ld0 = *((mlib_u64 *)dpx + 1);
x8 = ld0 >> 32;
sr1 = x8 ^ ~scal;
sr2 = (scal | ~mask) - (x8 & mask);
sr3 = (sr1 & ~mask) ^ sr2;
x12 = ld0 & 0xFFFFFFFF;
sr1 = x12 ^ ~scal;
sr2 = (scal | ~mask) - (x12 & mask);
sr3_ = (sr1 & ~mask) ^ sr2;
(*dpz++) = vis_to_double(sr3, sr3_);
dpx += 2;
}
nrest = n & 0xf;
if (nrest >= 8) {
dx = (*dpx++);
SUBS_S8_MOD;
(*dpz++) = dr;
nrest -= 8;
}
if (nrest > 0) {
dx = (*dpx++);
SUBS_S8_MOD;
emask = vis_edge8(dpz, pzend);
vis_pst_8(dr, dpz, emask);
}
} else {
void mlib_v_ImageLookUpSI_U16_U8_2_D1(const mlib_u16 *src,
mlib_u8 *dst,
mlib_s32 xsize,
const mlib_u8 **table)
{
mlib_u16 *sp; /* pointer to source data */
mlib_s32 s0, s1, s2, s3, s4; /* source data */
mlib_u8 *dl; /* pointer to start of destination */
mlib_u8 *dend; /* pointer to end of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2; /* destination data */
mlib_d64 t3, t4, t5; /* destination data */
mlib_d64 t6, t7, acc; /* destination data */
mlib_s32 emask; /* edge mask */
mlib_s32 i, num; /* loop variable */
const mlib_u8 *tab0 = &table[0][0];
const mlib_u8 *tab1 = &table[1][0];
sp = (void *)src;
dl = dst;
dend = dl + 2 * xsize - 1;
vis_alignaddr((void *)0, 7);
s0 = *sp++;
*dl++ = tab0[s0];
dp = (mlib_d64 *) dl;
xsize--;
if (xsize >= 4) {
s1 = sp[0];
s2 = sp[1];
s3 = sp[2];
s4 = sp[3];
sp += 4;
#pragma pipeloop(0)
for (i = 0; i <= xsize - 8; i += 4, sp += 4) {
t7 = VIS_LD_U8_I(tab0, s4);
t6 = VIS_LD_U8_I(tab1, s3);
t5 = VIS_LD_U8_I(tab0, s3);
t4 = VIS_LD_U8_I(tab1, s2);
t3 = VIS_LD_U8_I(tab0, s2);
t2 = VIS_LD_U8_I(tab1, s1);
t1 = VIS_LD_U8_I(tab0, s1);
t0 = VIS_LD_U8_I(tab1, s0);
acc = vis_faligndata(t7, acc);
acc = vis_faligndata(t6, acc);
acc = vis_faligndata(t5, acc);
acc = vis_faligndata(t4, acc);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s4;
s1 = sp[0];
s2 = sp[1];
s3 = sp[2];
s4 = sp[3];
*dp++ = acc;
}
t7 = VIS_LD_U8_I(tab0, s4);
t6 = VIS_LD_U8_I(tab1, s3);
t5 = VIS_LD_U8_I(tab0, s3);
t4 = VIS_LD_U8_I(tab1, s2);
t3 = VIS_LD_U8_I(tab0, s2);
t2 = VIS_LD_U8_I(tab1, s1);
t1 = VIS_LD_U8_I(tab0, s1);
t0 = VIS_LD_U8_I(tab1, s0);
acc = vis_faligndata(t7, acc);
acc = vis_faligndata(t6, acc);
acc = vis_faligndata(t5, acc);
acc = vis_faligndata(t4, acc);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s4;
*dp++ = acc;
}
num = ((mlib_u8 *) dend - (mlib_u8 *) dp) >> 1;
sp += num;
num++;
#pragma pipeloop(0)
for (i = 0; i < num; i++) {
s1 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(tab1, s1);
acc = vis_faligndata(t0, acc);
t0 = VIS_LD_U8_I(tab0, s1);
acc = vis_faligndata(t0, acc);
}
t0 = VIS_LD_U8_I(tab1, s0);
acc = vis_faligndata(t0, acc);
emask = vis_edge8(dp, dend);
vis_pst_8(acc, dp, emask);
}
/* The case of even address of vector x */
static void
mlib_VectorDotProd_U8C_al_x(
mlib_d64 *z,
const void *x,
const void *y,
mlib_s32 n)
{
mlib_u8 *pxend, *px = (mlib_u8 *)x, *py = (mlib_u8 *)y;
mlib_d64 sum_r = 0.0, sum_i = 0.0;
mlib_d64 *dpx, *dpy, *dpxend;
mlib_d64 dx, dy, dy0, dy1;
mlib_d64 dx_r, dy_r, dy_i;
mlib_d64 d_iih, d_iil, d_irh, d_irl, d_rih, d_ril, d_rrh, d_rrl;
mlib_d64 d_ih, d_il, d_rh, d_rl;
mlib_d64 ds_r, ds_i, ds1_r, ds1_i;
mlib_d64 lb_mask = vis_to_double_dup(0x00FF00FF);
mlib_d64 edge[2], fzero = vis_fzero();
mlib_f32 fsum;
mlib_s32 d_left;
mlib_s32 emask, off;
edge[0] = edge[1] = 0;
dpx = (mlib_d64 *)((mlib_addr)px & (~7));
off = (mlib_addr)dpx - (mlib_addr)px;
dpy = vis_alignaddr((void *)py, off);
pxend = px + n + n - 1;
dpxend = (mlib_d64 *)((mlib_addr)pxend & (~7));
emask = vis_edge8(px, pxend);
vis_pst_8(dpx[0], edge, emask);
dx = edge[0];
dy = vis_ld_d64_nf(dpy);
if (((((mlib_addr)px) ^ ((mlib_addr)py)) & 7) == 0) {
vis_write_bmask(0x781A3C5E, 0);
while ((mlib_addr)dpx < (mlib_addr)dpxend) {
d_left = dpxend - dpx;
if (d_left > MAX_LOOP)
d_left = MAX_LOOP;
ds_i = ds_r = ds1_i = ds1_r = 0.0;
#pragma pipeloop(0)
for (; d_left > 0; d_left--) {
DPROD_U8C0;
SUM_U8C;
dx = dpx[1];
dy = dpy[1];
dpx++;
dpy++;
}
ds_i = vis_fpadd32(ds_i, ds1_i);
ds_r = vis_fpadd32(ds_r, ds1_r);
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
} else {
mlib_s32 mask = ((mlib_addr)(py + off)) & 7;
vis_write_bmask(0x11111111 * mask, 0x01234567);
dy1 = vis_ld_d64_nf(dpy+1);
dy = vis_bshuffle(dy, dy1);
SET_ALIGN_U8C;
while ((mlib_addr)dpx < (mlib_addr)dpxend) {
d_left = dpxend - dpx;
if (d_left > MAX_LOOP)
d_left = MAX_LOOP;
ds_i = ds_r = ds1_i = ds1_r = 0.0;
#pragma pipeloop(0)
for (; d_left > 0; d_left--) {
DPROD_U8C;
SUM_U8C;
dy0 = dy1;
dy1 = vis_ld_d64_nf(dpy+2);
dx = vis_ld_d64_nf(dpx+1);
dy = vis_bshuffle(dy0, dy1);
dpx++;
dpy++;
}
ds_i = vis_fpadd32(ds_i, ds1_i);
ds_r = vis_fpadd32(ds_r, ds1_r);
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
}
if ((mlib_addr)dpx <= (mlib_addr)pxend) {
emask = vis_edge8(dpx, pxend);
vis_pst_8(dx, edge + 1, emask);
dx = edge[1];
SET_ALIGN_U8C;
DPROD_U8C;
SUM_U8C_TAIL;
fsum = vis_read_hi(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_r);
sum_r += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_hi(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
fsum = vis_read_lo(ds_i);
sum_i += (mlib_d64)*((mlib_s32 *)&fsum);
}
z[0] = sum_r;
z[1] = sum_i;
#undef MAX_LOOP
}
void mlib_v_ImageLookUp_S16_U8_3_D1(const mlib_s16 *src,
mlib_u8 *dst,
mlib_s32 xsize,
const mlib_u8 *table0,
const mlib_u8 *table1,
const mlib_u8 *table2)
{
mlib_s16 *sp; /* pointer to source data */
mlib_s32 s0, s1, s2, s3; /* source data */
mlib_s32 s4, s5, s6, s7; /* source data */
mlib_u8 *dl; /* pointer to start of destination */
mlib_u8 *dend; /* pointer to end of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2; /* destination data */
mlib_d64 t3, t4, t5; /* destination data */
mlib_d64 t6, t7, acc; /* destination data */
mlib_s32 emask; /* edge mask */
mlib_s32 i, num; /* loop variable */
const mlib_u8 *table;
dl = dst;
sp = (void *)src;
dp = (mlib_d64 *) dl;
dend = dl + xsize - 1;
vis_alignaddr((void *)0, 7);
if (xsize >= 8) {
s0 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
s4 = sp[4];
s5 = sp[5];
s6 = sp[6];
s7 = sp[7];
sp += 8;
#pragma pipeloop(0)
for (i = 0; i <= xsize - 16; i += 8, sp += 8) {
t7 = VIS_LD_U8_I(table1, s7);
t6 = VIS_LD_U8_I(table0, s6);
t5 = VIS_LD_U8_I(table2, s5);
t4 = VIS_LD_U8_I(table1, s4);
t3 = VIS_LD_U8_I(table0, s3);
t2 = VIS_LD_U8_I(table2, s2);
t1 = VIS_LD_U8_I(table1, s1);
t0 = VIS_LD_U8_I(table0, s0);
acc = vis_faligndata(t7, acc);
acc = vis_faligndata(t6, acc);
acc = vis_faligndata(t5, acc);
acc = vis_faligndata(t4, acc);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
table = table0;
table0 = table2;
table2 = table1;
table1 = table;
s0 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
s4 = sp[4];
s5 = sp[5];
s6 = sp[6];
s7 = sp[7];
*dp++ = acc;
}
t7 = VIS_LD_U8_I(table1, s7);
t6 = VIS_LD_U8_I(table0, s6);
t5 = VIS_LD_U8_I(table2, s5);
t4 = VIS_LD_U8_I(table1, s4);
t3 = VIS_LD_U8_I(table0, s3);
t2 = VIS_LD_U8_I(table2, s2);
t1 = VIS_LD_U8_I(table1, s1);
t0 = VIS_LD_U8_I(table0, s0);
acc = vis_faligndata(t7, acc);
acc = vis_faligndata(t6, acc);
acc = vis_faligndata(t5, acc);
acc = vis_faligndata(t4, acc);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
table = table0;
table0 = table2;
table2 = table1;
table1 = table;
*dp++ = acc;
}
if ((mlib_addr) dp <= (mlib_addr) dend) {
num = (mlib_addr) dend - (mlib_addr) dp;
sp += num;
num++;
i = num - 3 * (num / 3);
if (i == 2) {
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table1, s0);
acc = vis_faligndata(t0, acc);
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc = vis_faligndata(t0, acc);
num -= 2;
}
else if (i == 1) {
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc = vis_faligndata(t0, acc);
num--;
}
#pragma pipeloop(0)
for (i = 0; i < num; i += 3) {
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table2, s0);
acc = vis_faligndata(t0, acc);
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table1, s0);
acc = vis_faligndata(t0, acc);
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(table0, s0);
acc = vis_faligndata(t0, acc);
}
emask = vis_edge8(dp, dend);
vis_pst_8(acc, dp, emask);
}
}
mlib_status
mlib_v_ImageAdd_U8(
mlib_image *dst,
const mlib_image *src1,
const mlib_image *src2)
{
mlib_s32 i, j, k;
mlib_s32 offdst, offsrc1, offsrc2, emask;
mlib_s32 amount;
mlib_d64 *dpp, *spp2, *spp1, *tmp_ptr;
mlib_d64 dd, dd0, dd1, sd10, sd11, sd20, sd21;
mlib_d64 sd1h, sd2h, sd1l, sd2l, rdh, rdl;
mlib_u8 *dend;
mlib_f32 nul = vis_to_float(0), fone = vis_to_float(0x100);
VALIDATE(mlib_u8);
/* initialize GSR scale factor */
vis_write_gsr(7 << 3);
sl1 = sp1;
sl2 = sp2;
dl = dp;
amount = width * channels;
offdst = ((mlib_addr)dp) & 7;
offsrc1 = ((mlib_addr)sp1) & 7;
offsrc2 = ((mlib_addr)sp2) & 7;
if ((offdst == offsrc1) && (offdst == offsrc2) &&
(((strided ^ stride1) & 7) == 0) &&
(((strided ^ stride2) & 7) == 0)) {
for (j = 0; j < height; j++) {
/* prepare the destination addresses */
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = (mlib_u8 *)dpp - dp;
/* prepare the source addresses */
spp1 = (mlib_d64 *)vis_alignaddr(sp1, 0);
spp2 = (mlib_d64 *)vis_alignaddr(sp2, 0);
dend = dp + amount - 1;
/* generate edge mask for the start point */
emask = vis_edge8(dp, dend);
if (emask != 0xff) {
sd10 = (*spp1++);
sd20 = (*spp2++);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp++, emask);
i += 8;
}
#pragma pipeloop(0)
for (; i <= amount - 8; i += 8) {
sd10 = (*spp1++);
sd20 = (*spp2++);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
(*dpp++) = dd;
}
if (i < amount) {
emask = vis_edge8(dpp, dend);
sd10 = (*spp1++);
sd20 = (*spp2++);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp, emask);
}
sp1 = sl1 += stride1;
sp2 = sl2 += stride2;
dp = dl += strided;
}
} else if ((offdst == offsrc1) && (((strided ^ stride1) & 7) == 0)) {
for (j = 0; j < height; j++) {
/* prepare the destination addresses */
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = (mlib_u8 *)dpp - dp;
/* prepare the source addresses */
spp1 = (mlib_d64 *)vis_alignaddr(sp1, 0);
spp2 = (mlib_d64 *)vis_alignaddr(sp2, i);
dend = dp + amount - 1;
/* generate edge mask for the start point */
emask = vis_edge8(dp, dend);
sd20 = vis_ld_d64_nf(spp2);
if (emask != 0xff) {
sd10 = (*spp1++);
sd21 = vis_ld_d64_nf(spp2 + 1);
sd20 = vis_faligndata(sd20, sd21);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp++, emask);
sd20 = sd21;
spp2++;
i += 8;
}
#pragma pipeloop(0)
for (; i <= amount - 8; i += 8) {
sd10 = (*spp1++);
sd21 = vis_ld_d64_nf(spp2 + 1);
sd20 = vis_faligndata(sd20, sd21);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
(*dpp++) = dd;
sd20 = sd21;
spp2++;
}
if (i < amount) {
emask = vis_edge8(dpp, dend);
sd10 = (*spp1++);
sd20 = vis_faligndata(sd20,
vis_ld_d64_nf(spp2 + 1));
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp, emask);
}
sp1 = sl1 += stride1;
sp2 = sl2 += stride2;
dp = dl += strided;
}
} else if ((offdst == offsrc2) && (((strided ^ stride2) & 7) == 0)) {
for (j = 0; j < height; j++) {
/* prepare the destination addresses */
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = (mlib_u8 *)dpp - dp;
/* prepare the source addresses */
spp2 = (mlib_d64 *)vis_alignaddr(sp2, 0);
spp1 = (mlib_d64 *)vis_alignaddr(sp1, i);
dend = dp + amount - 1;
/* generate edge mask for the start point */
emask = vis_edge8(dp, dend);
sd10 = vis_ld_d64_nf(spp1);
if (emask != 0xff) {
sd20 = (*spp2++);
sd11 = vis_ld_d64_nf(spp1 + 1);
sd10 = vis_faligndata(sd10, sd11);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp++, emask);
sd10 = sd11;
spp1++;
i += 8;
}
#pragma pipeloop(0)
for (; i <= amount - 8; i += 8) {
sd20 = (*spp2++);
sd11 = vis_ld_d64_nf(spp1 + 1);
sd10 = vis_faligndata(sd10, sd11);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
(*dpp++) = dd;
sd10 = sd11;
spp1++;
}
if (i < amount) {
emask = vis_edge8(dpp, dend);
sd20 = (*spp2++);
sd10 = vis_faligndata(sd10,
vis_ld_d64_nf(spp1 + 1));
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp, emask);
}
sp1 = sl1 += stride1;
sp2 = sl2 += stride2;
dp = dl += strided;
}
} else if ((offsrc1 == offsrc2) && (((stride1 ^ stride2) & 7) == 0)) {
for (j = 0; j < height; j++) {
/* prepare the source addresses */
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = (mlib_u8 *)dpp - dp;
/* prepare the destination addresses */
spp1 = (mlib_d64 *)vis_alignaddr(sp1, i);
spp2 = (mlib_d64 *)vis_alignaddr(sp2, i);
dend = dp + amount - 1;
/* generate edge mask for the start point */
emask = vis_edge8(dp, dend);
sd10 = vis_ld_d64_nf(spp1); spp1++;
sd20 = vis_ld_d64_nf(spp2); spp2++;
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd0);
if (emask != 0xff) {
sd10 = vis_ld_d64_nf(spp1); spp1++;
sd20 = vis_ld_d64_nf(spp2); spp2++;
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd1);
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = dd1;
i += 8;
}
#pragma pipeloop(0)
for (; i <= amount - 8; i += 8) {
sd10 = vis_ld_d64_nf(spp1); spp1++;
sd20 = vis_ld_d64_nf(spp2); spp2++;
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd1);
(*dpp++) = vis_faligndata(dd0, dd1);
dd0 = dd1;
}
if (i < amount) {
emask = vis_edge8(dpp, dend);
sd10 = vis_ld_d64_nf(spp1); spp1++;
sd20 = vis_ld_d64_nf(spp2); spp2++;
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd1);
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask);
}
sp1 = sl1 += stride1;
sp2 = sl2 += stride2;
dp = dl += strided;
}
} else {
/* common case */
for (j = 0; j < height; j++) {
/* prepare the destination addresses */
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = (mlib_u8 *)dpp - dp;
dend = dp + amount - 1;
/* generate edge mask for the start point */
emask = vis_edge8(dp, dend);
if (emask != 0xff) {
spp1 = (mlib_d64 *)vis_alignaddr(sp1, i);
sd10 = vis_faligndata(vis_ld_d64_nf(spp1),
vis_ld_d64_nf(spp1 + 1));
spp2 = (mlib_d64 *)vis_alignaddr(sp2, i);
sd20 = vis_faligndata(vis_ld_d64_nf(spp2),
vis_ld_d64_nf(spp2 + 1));
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
vis_pst_8(dd, dpp++, emask);
i += 8;
}
/* copy src1 to dst */
spp1 = (mlib_d64 *)vis_alignaddr(sp1, i);
sd11 = vis_ld_d64_nf(spp1);
tmp_ptr = dpp;
#pragma pipeloop(0)
for (k = i; k <= (amount - 8); k += 8) {
sd10 = sd11;
sd11 = vis_ld_d64_nf(spp1 + 1);
(*tmp_ptr++) = vis_faligndata(sd10, sd11);
spp1++;
}
sd11 = vis_faligndata(sd11, vis_ld_d64_nf(spp1 + 1));
spp2 = (mlib_d64 *)vis_alignaddr(sp2, i);
sd20 = vis_ld_d64_nf(spp2);
tmp_ptr = dpp;
#pragma pipeloop(0)
for (; i <= amount - 8; i += 8) {
sd10 = (*tmp_ptr++);
sd21 = vis_ld_d64_nf(spp2 + 1);
sd20 = vis_faligndata(sd20, sd21);
MLIB_V_ADDIMAGE_U8(sd10, sd20, dd);
(*dpp++) = dd;
sd20 = sd21;
spp2++;
}
if (i < amount) {
emask = vis_edge8(dpp, dend);
sd20 = vis_faligndata(sd20,
vis_ld_d64_nf(spp2 + 1));
MLIB_V_ADDIMAGE_U8(sd11, sd20, dd);
vis_pst_8(dd, dpp, emask);
}
sp1 = sl1 += stride1;
sp2 = sl2 += stride2;
dp = dl += strided;
}
}
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VideoDownSample422(
mlib_u8 *dst,
const mlib_u8 *src,
mlib_s32 n)
{
mlib_d64 *sp0 = (mlib_d64 *)src;
mlib_d64 *pd = (mlib_d64 *)dst;
mlib_d64 d0;
mlib_d64 tmp, data0, data1;
mlib_d64 acc0_hi, acc0_lo;
mlib_d64 round = vis_to_double_dup(0x1);
mlib_f32 fone = vis_to_float(0x1000000);
mlib_s32 i, edge;
if (n <= 0)
return (MLIB_FAILURE);
vis_write_gsr(6 << 3);
vis_write_bmask(0x02461357, 0);
#pragma pipeloop(0)
for (i = 0; i <= n - 16; i += 16) {
d0 = (*sp0++);
tmp = vis_bshuffle(d0, d0);
acc0_hi = vis_fmul8x16au(vis_read_hi(tmp), fone);
acc0_lo = vis_fmul8x16au(vis_read_lo(tmp), fone);
acc0_hi = vis_fpadd16(acc0_hi, acc0_lo);
data0 = vis_fpadd16(acc0_hi, round);
d0 = (*sp0++);
tmp = vis_bshuffle(d0, d0);
acc0_hi = vis_fmul8x16au(vis_read_hi(tmp), fone);
acc0_lo = vis_fmul8x16au(vis_read_lo(tmp), fone);
acc0_hi = vis_fpadd16(acc0_hi, acc0_lo);
data1 = vis_fpadd16(acc0_hi, round);
(*pd++) = vis_fpack16_pair(data0, data1);
}
if (i < n) {
d0 = (*sp0++);
tmp = vis_bshuffle(d0, d0);
acc0_hi = vis_fmul8x16au(vis_read_hi(tmp), fone);
acc0_lo = vis_fmul8x16au(vis_read_lo(tmp), fone);
acc0_hi = vis_fpadd16(acc0_hi, acc0_lo);
data0 = vis_fpadd16(acc0_hi, round);
d0 = vis_ld_d64_nf(sp0);
tmp = vis_bshuffle(d0, d0);
acc0_hi = vis_fmul8x16au(vis_read_hi(tmp), fone);
acc0_lo = vis_fmul8x16au(vis_read_lo(tmp), fone);
acc0_hi = vis_fpadd16(acc0_hi, acc0_lo);
data1 = vis_fpadd16(acc0_hi, round);
edge = vis_edge8(pd, (dst + (n / 2) - 1));
vis_pst_8(vis_fpack16_pair(data0, data1), pd, edge);
}
return (MLIB_SUCCESS);
}
void
mlib_v_ImageSqrShift_U8(
mlib_u8 *src,
mlib_s32 slb,
mlib_u8 *dst,
mlib_s32 dlb,
mlib_s32 xsize,
mlib_s32 ysize)
{
/* pointer to a line in source image */
mlib_u8 *sl;
/* 8-byte aligned pointer to source image */
mlib_d64 *sp;
/* pointer to a line in destination image */
mlib_u8 *dl;
/* pointer to end of a line in destination image */
mlib_u8 *dend;
/* 8-byte aligned pointer to destination image */
mlib_d64 *dp;
/* offset of address alignment in destination */
mlib_s32 off;
/* edge masks */
mlib_s32 emask;
/* source data */
mlib_d64 s0, s1;
/* source data */
mlib_d64 sd;
/* destination data */
mlib_d64 dd;
/* temporaries used in macro */
mlib_d64 sdh, sdl;
/* temporaries used in macro */
mlib_d64 rdh, rdl;
/* loop variable */
mlib_s32 i, j, n;
sl = src;
dl = dst;
/* row loop */
for (j = 0; j < ysize; j++) {
/* prepare the destination address */
dp = (mlib_d64 *)((mlib_addr)dl & (~7));
off = (mlib_addr)dp - (mlib_addr)dl;
dend = dl + xsize - 1;
/* prepare the source address */
sp = (mlib_d64 *)vis_alignaddr(sl, off);
/* generate edge mask for the start point */
emask = vis_edge8(dl, dend);
/* first 8 pixels */
s0 = vis_ld_d64_nf(sp); sp++;
s1 = vis_ld_d64_nf(sp); sp++;
sd = vis_faligndata(s0, s1);
MLIB_V_IMAGESQRSHIFT_U8(sd, dd);
vis_pst_8(dd, dp++, emask);
n = ((mlib_u8 *)(dend + 1) - (mlib_u8 *)dp) / 8;
/* 8-pixel column loop */
#pragma pipeloop(0)
for (i = 0; i < n; i++) {
s0 = s1;
s1 = vis_ld_d64_nf(sp); sp++;
sd = vis_faligndata(s0, s1);
MLIB_V_IMAGESQRSHIFT_U8(sd, dd);
(*dp++) = dd;
}
/* end point handling */
if ((mlib_addr)dp <= (mlib_addr)dend) {
emask = vis_edge8(dp, dend);
s0 = s1;
s1 = vis_ld_d64_nf(sp); sp++;
sd = vis_faligndata(s0, s1);
MLIB_V_IMAGESQRSHIFT_U8(sd, dd);
vis_pst_8(dd, dp++, emask);
}
sl += slb;
dl += dlb;
}
}
void
mlib_v_ImageColorRGB2Mono_U8_D1(
const mlib_u8 *src,
mlib_u8 *dst,
mlib_s32 dsize,
const mlib_d64 *weight)
{
mlib_u8 *dst_end;
mlib_d64 dd, d0, d1, d2, d3;
mlib_d64 rgdd0, bdd0, rgdd1, bdd1, ddt;
mlib_d64 *src_all, *dp;
mlib_f32 d32, e32, alpha, gamma, beta;
mlib_d64 sd0, sd1, sd2;
mlib_s32 i, emask;
mlib_s32 off;
mlib_s32 mask0 = 0x0369147a;
mlib_s32 mask1 = 0x258b258b;
mlib_s32 mask2 = 0x47ad58be;
mlib_s32 mask3 = 0x69cf69cf;
/* prepare the weight */
alpha = vis_to_float(weight[0] * 8192);
beta = vis_to_float(weight[1] * 8192);
gamma = vis_to_float(weight[2] * 8192);
vis_write_gsr(2 << 3);
dp = (mlib_d64 *)((mlib_addr)dst & (~7));
off = (mlib_addr)dp - (mlib_addr)dst;
dst_end = dst + (dsize - 1);
emask = vis_edge8(dst, dst_end);
src_all = vis_alignaddr((void *)src, (3 * off));
d0 = (*src_all++);
d1 = (*src_all++);
d2 = (*src_all++);
d3 = (*src_all++);
sd0 = vis_faligndata(d0, d1);
sd1 = vis_faligndata(d1, d2);
sd2 = vis_faligndata(d2, d3);
CHANNELSEPARATE_U8(sd0, sd1, sd2, rgdd0, bdd0, rgdd1, bdd1);
CHANNELWEIGHT_U8(rgdd0, bdd0, rgdd1, bdd1, dd);
vis_pst_8(dd, dp, emask);
dp++;
#pragma pipeloop(0)
for (i = 8 + off; i <= (dsize - 8); i += 8) {
d0 = d3;
d1 = (*src_all++);
d2 = (*src_all++);
d3 = (*src_all++);
sd0 = vis_faligndata(d0, d1);
sd1 = vis_faligndata(d1, d2);
sd2 = vis_faligndata(d2, d3);
CHANNELSEPARATE_U8(sd0, sd1, sd2, rgdd0, bdd0, rgdd1, bdd1);
CHANNELWEIGHT_U8(rgdd0, bdd0, rgdd1, bdd1, dd);
(*dp++) = dd;
}
if ((mlib_addr)dp <= (mlib_addr)dst_end) {
emask = vis_edge8(dp, dst_end);
d0 = d3;
d1 = (*src_all++);
d2 = (*src_all++);
d3 = (*src_all++);
sd0 = vis_faligndata(d0, d1);
sd1 = vis_faligndata(d1, d2);
sd2 = vis_faligndata(d2, d3);
CHANNELSEPARATE_U8(sd0, sd1, sd2, rgdd0, bdd0, rgdd1, bdd1);
CHANNELWEIGHT_U8(rgdd0, bdd0, rgdd1, bdd1, dd);
vis_pst_8(dd, dp, emask);
}
}
static mlib_status
mlib_v_VideoColorYUV2ABGR411_dst_nonalign(
mlib_u8 *abgr,
const mlib_u8 *y,
const mlib_u8 *u,
const mlib_u8 *v,
mlib_s32 width,
mlib_s32 height,
mlib_s32 abgr_stride,
mlib_s32 y_stride,
mlib_s32 uv_stride)
{
/* pointers to src address */
mlib_u8 *sp1, *sp2, *sp3, *sl1, *sl2, *sl3;
/* pointers to dst address */
mlib_u8 *dp, *dl, *dend;
/* all. pointer to y */
mlib_d64 *spy;
/* all. pointer to dst */
mlib_d64 *dpp;
/* u, v data */
mlib_f32 fu, fv;
/* y data */
mlib_d64 dy0, dy1, dy2;
mlib_d64 ddy1, ddy2, ddy3, ddy4;
mlib_d64 du0, du1;
mlib_d64 dv1, dv2;
mlib_d64 dr, dr1, dr2, dr3, dr4;
mlib_d64 dg, dg1, dg2, dg3, dg4;
mlib_d64 db, db1, db2, db3, db4;
mlib_d64 dd, dd0, dd1, dtmp;
/* used to load u, v into mlib_f32 */
mlib_f32 ffu[1], ffv[1];
/* used to load u, v into mlib_f32 */
mlib_u8 *ufu, *vfu;
/* 1.1644 * 4096 */
mlib_f32 f0 = vis_to_float(0x12a1);
/* 2.0184 * 8192 */
mlib_f32 f1 = vis_to_float(0x4097);
/* -0.3920 * 8192 */
mlib_f32 f4 = vis_to_float(0xf375);
/* -0.8132 * 8192 */
mlib_f32 f5 = vis_to_float(0xe5fa);
/* 1.5966 * 8192 */
mlib_f32 f8 = vis_to_float(0x3317);
/* -276.9856 * 32 */
mlib_d64 doff0 = vis_to_double_dup(0xdd60dd60);
/* 135.6352 * 32 */
mlib_d64 doff1 = vis_to_double_dup(0x10f410f4);
/* -222.9952 * 32 */
mlib_d64 doff2 = vis_to_double_dup(0xe420e420);
mlib_f32 fscale = vis_to_float(0x80808080);
/* loop variables */
mlib_s32 i, j;
/* alpha_ch. is not written */
mlib_s32 emask = 0x7777;
mlib_s32 emask1;
mlib_d64 *buf;
mlib_s32 inc;
ufu = (mlib_u8 *)ffu;
vfu = (mlib_u8 *)ffv;
/*
* initialize GSR scale factor
*/
vis_write_gsr(3 << 3);
buf = (mlib_d64 *)__mlib_malloc((width / 8 + 1) * sizeof (mlib_d64));
if (buf == NULL)
return (MLIB_FAILURE);
sp1 = sl1 = (mlib_u8 *)y;
sp2 = sl2 = (mlib_u8 *)u;
sp3 = sl3 = (mlib_u8 *)v;
dl = dp = (mlib_u8 *)abgr;
/*
* row loop
*/
for (j = 0; j < height; j++) {
spy = (mlib_d64 *)vis_alignaddr(sp1, 0);
dpp = buf;
dy0 = vis_ld_d64_nf(spy); spy++;
#pragma pipeloop(0)
for (i = 0; i < width; i += 8) {
dy1 = vis_ld_d64_nf(spy); spy++;
(*dpp++) = vis_faligndata(dy0, dy1);
dy0 = dy1;
}
spy = buf;
dend = dp + width * 4 - 1;
emask1 = vis_edge8(dp, dend);
dpp = (mlib_d64 *)vis_alignaddr(dp, 0);
i = dp - (mlib_u8 *)dpp;
emask >>= i;
vis_alignaddr((void *)(8 - i), 0);
inc = (emask1 != 0xff);
emask1 &= emask;
ufu[0] = vis_ld_u8_nf(sp2);
ufu[1] = vis_ld_u8_nf(sp2 + 1);
ufu[2] = vis_ld_u8_nf(sp2 + 2);
ufu[3] = vis_ld_u8_nf(sp2 + 3);
vfu[0] = vis_ld_u8_nf(sp3);
vfu[1] = vis_ld_u8_nf(sp3 + 1);
vfu[2] = vis_ld_u8_nf(sp3 + 2);
vfu[3] = vis_ld_u8_nf(sp3 + 3);
sp2 += 4;
sp3 += 4;
fu = ffu[0];
fv = ffv[0];
/*
* 16-pixel column loop
*/
#pragma pipeloop(0)
for (i = 0; i <= width - 16; i += 16) {
dy1 = (*spy++);
dy2 = (*spy++);
du0 = vis_fmul8x16al(fu, f1);
db = vis_fpadd16(du0, doff0);
du1 = vis_fmul8x16al(fu, f4);
dv1 = vis_fmul8x16al(fv, f5);
dtmp = vis_fpadd16(du1, dv1);
dg = vis_fpadd16(dtmp, doff1);
dv2 = vis_fmul8x16al(fv, f8);
dr = vis_fpadd16(dv2, doff2);
ddy1 = vis_fmul8x16al(vis_read_hi(dy1), f0);
ddy2 = vis_fmul8x16al(vis_read_lo(dy1), f0);
ufu[0] = vis_ld_u8_nf(sp2);
ddy3 = vis_fmul8x16al(vis_read_hi(dy2), f0);
ddy4 = vis_fmul8x16al(vis_read_lo(dy2), f0);
ufu[1] = vis_ld_u8_nf(sp2 + 1);
db1 = vis_fmul8x16au(fscale, vis_read_hi(db));
db1 = vis_fpadd16(ddy1, db1);
ufu[2] = vis_ld_u8_nf(sp2 + 2);
db2 = vis_fmul8x16al(fscale, vis_read_hi(db));
db2 = vis_fpadd16(ddy2, db2);
ufu[3] = vis_ld_u8_nf(sp2 + 3);
db3 = vis_fmul8x16au(fscale, vis_read_lo(db));
db3 = vis_fpadd16(ddy3, db3);
vfu[0] = vis_ld_u8_nf(sp3);
db4 = vis_fmul8x16al(fscale, vis_read_lo(db));
db4 = vis_fpadd16(ddy4, db4);
vfu[1] = vis_ld_u8_nf(sp3 + 1);
dg1 = vis_fmul8x16au(fscale, vis_read_hi(dg));
dg1 = vis_fpadd16(ddy1, dg1);
vfu[2] = vis_ld_u8_nf(sp3 + 2);
dg2 = vis_fmul8x16al(fscale, vis_read_hi(dg));
dg2 = vis_fpadd16(ddy2, dg2);
vfu[3] = vis_ld_u8_nf(sp3 + 3);
dg3 = vis_fmul8x16au(fscale, vis_read_lo(dg));
dg3 = vis_fpadd16(ddy3, dg3);
dg4 = vis_fmul8x16al(fscale, vis_read_lo(dg));
dg4 = vis_fpadd16(ddy4, dg4);
dr1 = vis_fmul8x16au(fscale, vis_read_hi(dr));
dr1 = vis_fpadd16(ddy1, dr1);
dr2 = vis_fmul8x16al(fscale, vis_read_hi(dr));
dr2 = vis_fpadd16(ddy2, dr2);
dr3 = vis_fmul8x16au(fscale, vis_read_lo(dr));
dr3 = vis_fpadd16(ddy3, dr3);
dr4 = vis_fmul8x16al(fscale, vis_read_lo(dr));
dr4 = vis_fpadd16(ddy4, dr4);
dr = vis_fpack16_pair(dr1, dr2);
dr1 = vis_fpack16_pair(dr3, dr4);
dg = vis_fpack16_pair(dg1, dg2);
dg1 = vis_fpack16_pair(dg3, dg4);
db = vis_fpack16_pair(db1, db2);
db1 = vis_fpack16_pair(db3, db4);
dg2 = vis_fpmerge(vis_read_hi(db), vis_read_hi(dg));
dg3 = vis_fpmerge(vis_read_hi(db), vis_read_hi(dr));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
dpp += inc;
inc = 1;
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
dg2 = vis_fpmerge(vis_read_lo(db), vis_read_lo(dg));
dg3 = vis_fpmerge(vis_read_lo(db), vis_read_lo(dr));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
dg2 = vis_fpmerge(vis_read_hi(db1), vis_read_hi(dg1));
dg3 = vis_fpmerge(vis_read_hi(db1), vis_read_hi(dr1));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
dg2 = vis_fpmerge(vis_read_lo(db1), vis_read_lo(dg1));
dg3 = vis_fpmerge(vis_read_lo(db1), vis_read_lo(dr1));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
fu = ffu[0];
fv = ffv[0];
sp2 += 4;
sp3 += 4;
emask1 = emask;
}
if (i <= width - 8) {
dy1 = (*spy++);
du0 = vis_fmul8x16al(fu, f1);
db = vis_fpadd16(du0, doff0);
du1 = vis_fmul8x16al(fu, f4);
dv1 = vis_fmul8x16al(fv, f5);
dtmp = vis_fpadd16(du1, dv1);
dg = vis_fpadd16(dtmp, doff1);
dv2 = vis_fmul8x16al(fv, f8);
dr = vis_fpadd16(dv2, doff2);
ddy1 = vis_fmul8x16al(vis_read_hi(dy1), f0);
ddy2 = vis_fmul8x16al(vis_read_lo(dy1), f0);
ufu[0] = ufu[2];
db1 = vis_fmul8x16au(fscale, vis_read_hi(db));
db1 = vis_fpadd16(ddy1, db1);
vfu[0] = vfu[2];
db2 = vis_fmul8x16al(fscale, vis_read_hi(db));
db2 = vis_fpadd16(ddy2, db2);
dg1 = vis_fmul8x16au(fscale, vis_read_hi(dg));
dg1 = vis_fpadd16(ddy1, dg1);
dg2 = vis_fmul8x16al(fscale, vis_read_hi(dg));
dg2 = vis_fpadd16(ddy2, dg2);
dr1 = vis_fmul8x16au(fscale, vis_read_hi(dr));
dr1 = vis_fpadd16(ddy1, dr1);
dr2 = vis_fmul8x16al(fscale, vis_read_hi(dr));
dr2 = vis_fpadd16(ddy2, dr2);
dr = vis_fpack16_pair(dr1, dr2);
dg = vis_fpack16_pair(dg1, dg2);
db = vis_fpack16_pair(db1, db2);
dg2 = vis_fpmerge(vis_read_hi(db), vis_read_hi(dg));
dg3 = vis_fpmerge(vis_read_hi(db), vis_read_hi(dr));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
dpp += inc;
inc = 1;
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
dg2 = vis_fpmerge(vis_read_lo(db), vis_read_lo(dg));
dg3 = vis_fpmerge(vis_read_lo(db), vis_read_lo(dr));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp++, emask);
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
fu = ffu[0];
fv = ffv[0];
i += 8;
emask1 = emask;
}
if (i < width) {
dy1 = vis_ld_d64_nf(spy);
du0 = vis_fmul8x16al(fu, f1);
db = vis_fpadd16(du0, doff0);
du1 = vis_fmul8x16al(fu, f4);
dv1 = vis_fmul8x16al(fv, f5);
dtmp = vis_fpadd16(du1, dv1);
dg = vis_fpadd16(dtmp, doff1);
dv2 = vis_fmul8x16al(fv, f8);
dr = vis_fpadd16(dv2, doff2);
ddy1 = vis_fmul8x16al(vis_read_hi(dy1), f0);
db1 = vis_fmul8x16au(fscale, vis_read_hi(db));
db1 = vis_fpadd16(ddy1, db1);
dg1 = vis_fmul8x16au(fscale, vis_read_hi(dg));
dg1 = vis_fpadd16(ddy1, dg1);
dr1 = vis_fmul8x16au(fscale, vis_read_hi(dr));
dr1 = vis_fpadd16(ddy1, dr1);
fu = vis_fpack16(db1);
dg2 = vis_fpmerge(fu, vis_fpack16(dg1));
dg3 = vis_fpmerge(fu, vis_fpack16(dr1));
dd1 = vis_fpmerge(vis_read_hi(dg2), vis_read_hi(dg3));
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
dpp += inc;
dd0 = vis_fpmerge(vis_read_lo(dg2), vis_read_lo(dg3));
dd = vis_faligndata(dd1, dd0);
vis_pst_8(dd, dpp++, emask);
}
emask1 = vis_edge8(dpp, dend);
emask1 &= emask;
dd = vis_faligndata(dd0, dd1);
vis_pst_8(dd, dpp, emask1);
sp1 = sl1 = sl1 + y_stride;
sp2 = sl2 = sl2 + uv_stride;
sp3 = sl3 = sl3 + uv_stride;
dl = dp = dl + abgr_stride;
emask = 0x7777;
}
__mlib_free(buf);
return (MLIB_SUCCESS);
}
