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_VideoColorSplit3_S16(
mlib_s16 *color1,
mlib_s16 *color2,
mlib_s16 *color3,
const mlib_s16 *colors,
mlib_s32 n)
{
mlib_d64 *sp = (mlib_d64 *)colors;
mlib_d64 *dp0 = (mlib_d64 *)color1;
mlib_d64 *dp1 = (mlib_d64 *)color2;
mlib_d64 *dp2 = (mlib_d64 *)color3;
mlib_d64 sd0, sd1, sd2, dd0, dd1, dd2, dd3;
mlib_s32 i;
vis_write_gsr(4);
vis_write_bmask(0x02CE13DF, 0);
#pragma pipeloop(0)
#pragma unroll(4)
for (i = 0; i <= (n - 4); i += 4) {
sd0 = sp[0];
sd1 = sp[1];
sd2 = sp[2];
dd1 = vis_fpmerge(vis_read_lo(sd0), vis_read_hi(sd2));
dd0 = vis_fpmerge(vis_read_hi(sd0), vis_read_lo(sd1));
(*dp0++) = vis_bshuffle(dd0, dd1);
dd2 = vis_fpmerge(vis_read_hi(sd1), vis_read_lo(sd2));
dd3 = vis_faligndata(dd0, dd2);
(*dp1++) = vis_bshuffle(dd3, dd3);
(*dp2++) = vis_bshuffle(dd1, dd2);
sp += 3;
}
/*
* last 4 pixels
*/
if (i < n) {
mlib_s32 emask = 0xF0 >> (n & 3);
mlib_d64 st0, st1, st2;
sd0 = sp[0];
sd1 = vis_ld_d64_nf(sp + 1);
sd2 = vis_ld_d64_nf(sp + 2);
dd1 = vis_fpmerge(vis_read_lo(sd0), vis_read_hi(sd2));
dd0 = vis_fpmerge(vis_read_hi(sd0), vis_read_lo(sd1));
st0 = vis_bshuffle(dd0, dd1);
dd2 = vis_fpmerge(vis_read_hi(sd1), vis_read_lo(sd2));
dd3 = vis_faligndata(dd0, dd2);
st1 = vis_bshuffle(dd3, dd3);
st2 = vis_bshuffle(dd1, dd2);
vis_pst_16(st0, dp0, emask);
vis_pst_16(st1, dp1, emask);
vis_pst_16(st2, dp2, emask);
}
void mlib_v_ImageLookUpSI_S32_S16_4_DstOff2_D1(mlib_s32 *src,
mlib_s16 *dst,
mlib_s32 xsize,
mlib_s16 **table)
{
mlib_s32 *sp; /* pointer to source data */
mlib_s32 s0, s1; /* source data */
mlib_s16 *dl; /* pointer to start of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2, t3; /* destination data */
mlib_d64 acc; /* destination data */
mlib_s32 i; /* loop variable */
mlib_s16 *tab0 = &table[0][(mlib_u32)2147483648];
mlib_s16 *tab1 = &table[1][(mlib_u32)2147483648];
mlib_s16 *tab2 = &table[2][(mlib_u32)2147483648];
mlib_s16 *tab3 = &table[3][(mlib_u32)2147483648];
sp = src;
dl = dst;
dp = (mlib_d64 *) dl;
vis_alignaddr((void *) 0, 6);
s0 = *sp++;
if (xsize >= 1) {
s1 = *sp++;
#pragma pipeloop(0)
for(i = 0; i <= xsize - 2; i++) {
t3 = vis_ld_u16_i(tab1, ((mlib_addr)2*s1));
t2 = vis_ld_u16_i(tab0, ((mlib_addr)2*s1));
t1 = vis_ld_u16_i(tab3, ((mlib_addr)2*s0));
t0 = vis_ld_u16_i(tab2, ((mlib_addr)2*s0));
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s1;
s1 = *sp++;
*dp++ = acc;
}
t3 = vis_ld_u16_i(tab1, ((mlib_addr)2*s1));
t2 = vis_ld_u16_i(tab0, ((mlib_addr)2*s1));
t1 = vis_ld_u16_i(tab3, ((mlib_addr)2*s0));
t0 = vis_ld_u16_i(tab2, ((mlib_addr)2*s0));
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s1;
*dp++ = acc;
}
dl = (mlib_s16*)dp;
dl[0] = tab2[s0];
dl[1] = tab3[s0];
}
mlib_status
__mlib_VideoColorSplit2_S16(
mlib_s16 *color1,
mlib_s16 *color2,
const mlib_s16 *colors,
mlib_s32 n)
{
mlib_d64 *sp = (mlib_d64 *)colors;
mlib_d64 *dp0 = (mlib_d64 *)color1;
mlib_d64 *dp1 = (mlib_d64 *)color2;
mlib_d64 sd0, sd1, dd0, dd1, dd2, dd3;
mlib_s32 i;
vis_write_gsr64(((mlib_u64)0x014589cd << 32) | 2);
/*
* 8-pixels loop
*/
sd0 = sp[0];
sd1 = vis_ld_d64_nf(sp + 1);
dd0 = vis_faligndata(sd0, sd1);
dd1 = vis_faligndata(sd1, sd0);
dd2 = vis_bshuffle(sd0, sd1);
dd3 = vis_bshuffle(dd0, dd1);
sd0 = vis_ld_d64_nf(sp + 2);
sd1 = vis_ld_d64_nf(sp + 3);
dd0 = vis_faligndata(sd0, sd1);
dd1 = vis_faligndata(sd1, sd0);
#pragma pipeloop(0)
for (i = 0; i < (n / 4); i++) {
(*dp0++) = dd2;
(*dp1++) = dd3;
dd2 = vis_bshuffle(sd0, sd1);
dd3 = vis_bshuffle(dd0, dd1);
sd0 = vis_ld_d64_nf(sp + 4);
sd1 = vis_ld_d64_nf(sp + 5);
dd0 = vis_faligndata(sd0, sd1);
dd1 = vis_faligndata(sd1, sd0);
sp += 2;
}
/*
* last 8 pixels
*/
if (n & 3) {
mlib_s32 emask = 0xF0 >> (n & 3);
sd0 = sp[0];
sd1 = vis_ld_d64_nf(sp + 1);
dd0 = vis_faligndata(sd0, sd1);
dd1 = vis_faligndata(sd1, sd0);
dd2 = vis_bshuffle(sd0, sd1);
dd3 = vis_bshuffle(dd0, dd1);
vis_pst_16(dd2, (mlib_f32 *)dp0, emask);
vis_pst_16(dd3, (mlib_f32 *)dp1, emask);
}
void
mlib_v_ImageColorRGB2Mono_U8_A8D2X8(
const mlib_u8 *src,
mlib_s32 slb,
mlib_u8 *dst,
mlib_s32 dlb,
mlib_s32 xsize,
mlib_s32 ysize,
const mlib_d64 *weight)
{
mlib_d64 *sp, *dp;
mlib_d64 *sl, *dl;
mlib_d64 sd0, sd1, sd2, sd3;
mlib_d64 dd;
mlib_d64 rgdd0, bdd0, rgdd1, bdd1, ddt;
mlib_f32 d32, e32, alpha, gamma, beta;
mlib_s32 i, j;
mlib_s32 mask0 = 0x0369147a;
mlib_s32 mask1 = 0x258b258b;
sp = sl = (mlib_d64 *)src;
dp = dl = (mlib_d64 *)dst;
/* 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) + 4);
for (j = 0; j < ysize; j++) {
#pragma pipeloop(0)
for (i = 0; i < xsize / 8; i++) {
sd0 = (*sp++);
sd1 = (*sp++);
sd2 = (*sp++);
sd3 = vis_faligndata(sd2, sd2);
sd2 = vis_faligndata(sd1, sd2);
CHANNELSEPARATE_U8_AL(sd0, sd1, sd2, sd3, rgdd0, bdd0,
rgdd1, bdd1);
CHANNELWEIGHT_U8(rgdd0, bdd0, rgdd1, bdd1, dd);
(*dp++) = dd;
}
sp = sl = (mlib_d64 *)((mlib_u8 *)sl + slb);
dp = dl = (mlib_d64 *)((mlib_u8 *)dl + dlb);
}
}
void mlib_v_ImageLookUp_U8_U16_124_SrcOff0_D1(const mlib_u8 *src,
mlib_u16 *dst,
mlib_s32 xsize,
const mlib_u16 *table0,
const mlib_u16 *table1,
const mlib_u16 *table2,
const mlib_u16 *table3)
{
mlib_u32 *sa; /* aligned pointer to source data */
mlib_u8 *sp; /* pointer to source data */
mlib_u32 s0; /* source data */
mlib_u16 *dl; /* pointer to start of destination */
mlib_u16 *dend; /* pointer to end of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2; /* destination data */
mlib_d64 t3, acc0; /* destination data */
mlib_s32 emask; /* edge mask */
mlib_s32 i, num; /* loop variable */
sa = (mlib_u32*)src;
dl = dst;
dp = (mlib_d64 *) dl;
dend = dl + xsize - 1;
vis_alignaddr((void *) 0, 6);
i = 0;
if (xsize >= 4) {
s0 = *sa++;
#pragma pipeloop(0)
for(i = 0; i <= xsize - 8; i+=4) {
t3 = VIS_LD_U16_I(table3, (s0 << 1) & 0x1FE);
t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = *sa++;
*dp++ = acc0;
}
t3 = VIS_LD_U16_I(table3, (s0 << 1) & 0x1FE);
t2 = VIS_LD_U16_I(table2, (s0 >> 7) & 0x1FE);
t1 = VIS_LD_U16_I(table1, (s0 >> 15) & 0x1FE);
t0 = VIS_LD_U16_I(table0, (s0 >> 23) & 0x1FE);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
*dp++ = acc0;
}
void mlib_v_ImageLookUpSI_S16_U16_4_DstOff0_D1(const mlib_s16 *src,
mlib_u16 *dst,
mlib_s32 xsize,
const mlib_u16 **table)
{
mlib_s16 *sp; /* pointer to source data */
mlib_s32 s0; /* source data */
mlib_u16 *dl; /* pointer to start of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2, t3; /* destination data */
mlib_d64 acc; /* destination data */
mlib_s32 i; /* loop variable */
const mlib_u16 *tab0 = &table[0][32768];
const mlib_u16 *tab1 = &table[1][32768];
const mlib_u16 *tab2 = &table[2][32768];
const mlib_u16 *tab3 = &table[3][32768];
sp = (void *)src;
dl = dst;
dp = (mlib_d64 *) dl;
vis_alignaddr((void *) 0, 6);
if (xsize >= 1) {
s0 = (*sp++) << 1;
#pragma pipeloop(0)
for(i = 0; i <= xsize - 2; i++) {
t3 = VIS_LD_U16_I(tab3, s0);
t2 = VIS_LD_U16_I(tab2, s0);
t1 = VIS_LD_U16_I(tab1, s0);
t0 = VIS_LD_U16_I(tab0, s0);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = (*sp++) << 1;
*dp++ = acc;
}
t3 = VIS_LD_U16_I(tab3, s0);
t2 = VIS_LD_U16_I(tab2, s0);
t1 = VIS_LD_U16_I(tab1, s0);
t0 = VIS_LD_U16_I(tab0, s0);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
*dp++ = acc;
}
}
static void FUNC(
m3) (
FUNC_M_ARG)
{
mlib_s32 i;
mlib_d64 k0 = pkern[0];
mlib_d64 k1 = pkern[1];
mlib_d64 k2 = pkern[2];
mlib_d64 a0, a1, a2, aa, sum;
mlib_d64 *perror = vis_alignaddr(perror1, 0);
a0 = (*perror++);
for (i = 0; i < sw; i++) {
aa = (*perror++);
a1 = vis_faligndata(a0, aa);
a2 = vis_faligndata(a1, vis_faligndata(aa, aa));
sum = vis_fpadd16(buffd[i], FMUL_16x16(k0, a0));
sum = vis_fpadd16(sum, FMUL_16x16(k1, a1));
buffd[i] = vis_fpadd16(sum, FMUL_16x16(k2, a2));
a0 = aa;
}
}
static void FUNC(
m1) (
FUNC_M_ARG)
{
mlib_s32 i;
mlib_d64 k0 = pkern[0];
mlib_d64 a0, e0, e1;
mlib_d64 *perror = vis_alignaddr(perror1, 0);
e0 = (*perror++);
for (i = 0; i < (sw + 3) / 4; i++) {
e1 = (*perror++);
a0 = vis_faligndata(e0, e1);
buffd[i] = vis_fpadd16(buffd[i], FMUL_16x16(k0, a0));
e0 = e1;
}
}
void ADD_SUFF(IntRgbxToIntArgbConvert)(BLIT_PARAMS)
{
mlib_s32 dstScan = pDstInfo->scanStride;
mlib_s32 srcScan = pSrcInfo->scanStride;
mlib_d64 dd, mask;
mlib_s32 i, i0, j;
if (dstScan == 4*width && srcScan == 4*width) {
width *= height;
height = 1;
}
mask = vis_to_double_dup(0xFF000000);
vis_alignaddr(NULL, 7);
for (j = 0; j < height; j++) {
mlib_u32 *src = srcBase;
mlib_u32 *dst = dstBase;
i = i0 = 0;
if ((mlib_s32)dst & 7) {
dst[i] = 0xff000000 | (src[i] >> 8);
i0 = 1;
}
#pragma pipeloop(0)
for (i = i0; i <= (mlib_s32)width - 2; i += 2) {
dd = vis_freg_pair(((mlib_f32*)src)[i], ((mlib_f32*)src)[i + 1]);
dd = vis_faligndata(dd, dd);
*(mlib_d64*)(dst + i) = vis_for(dd, mask);
}
if (i < width) {
dst[i] = 0xff000000 | (src[i] >> 8);
}
void
mlib_v_ImageLookUp_U8_U8_124_SrcOff0_D1(
const mlib_u8 *src,
mlib_u8 *dst,
mlib_s32 xsize,
const mlib_u8 *table0,
const mlib_u8 *table1,
const mlib_u8 *table2,
const mlib_u8 *table3)
{
/* aligned pointer to source data */
mlib_u32 *sa;
/* pointer to source data */
mlib_u8 *sp;
/* source data */
mlib_u32 s0, s1;
/* 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, acc1;
/* edge mask */
mlib_s32 emask;
/* loop variable */
mlib_s32 i, num;
sa = (mlib_u32 *)src;
dl = dst;
dp = (mlib_d64 *)dl;
dend = dl + xsize - 1;
vis_alignaddr((void *)0, 7);
if (xsize >= 8) {
s0 = sa[0];
s1 = sa[1];
sa += 2;
vis_write_bmask(0x012389ab, 0);
#pragma pipeloop(0)
for (i = 0; i <= xsize - 16; i += 8, sa += 2) {
t7 = VIS_LD_U8_I(table3, s1 & 0xFF);
t6 = VIS_LD_U8_I(table2, (s1 >> 8) & 0xFF);
t5 = VIS_LD_U8_I(table1, (s1 >> 16) & 0xFF);
t4 = VIS_LD_U8_I(table0, s1 >> 24);
t3 = VIS_LD_U8_I(table3, s0 & 0xFF);
t2 = VIS_LD_U8_I(table2, (s0 >> 8) & 0xFF);
t1 = VIS_LD_U8_I(table1, (s0 >> 16) & 0xFF);
t0 = VIS_LD_U8_I(table0, s0 >> 24);
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 = sa[0];
s1 = sa[1];
(*dp++) = vis_bshuffle(acc0, acc1);
}
t7 = VIS_LD_U8_I(table3, s1 & 0xFF);
t6 = VIS_LD_U8_I(table2, (s1 >> 8) & 0xFF);
t5 = VIS_LD_U8_I(table1, (s1 >> 16) & 0xFF);
t4 = VIS_LD_U8_I(table0, s1 >> 24);
t3 = VIS_LD_U8_I(table3, s0 & 0xFF);
t2 = VIS_LD_U8_I(table2, (s0 >> 8) & 0xFF);
t1 = VIS_LD_U8_I(table1, (s0 >> 16) & 0xFF);
t0 = VIS_LD_U8_I(table0, s0 >> 24);
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);
}
mlib_status
__mlib_VectorConvert_S32_S16_Mod(
mlib_s32 *z,
const mlib_s16 *x,
mlib_s32 n)
{
mlib_s32 i;
const mlib_s16 *src = x;
mlib_s32 *dst = z;
mlib_d64 *ddsrc, *ddst;
mlib_s32 len_64, even_length, rest_64, length = n;
mlib_d64 dd1, dd2, dd3, dd4;
mlib_f32 two_16_ones = vis_to_float(0x10001);
if (length < 16) {
EXPAND(mlib_s16, mlib_s32);
}
while ((mlib_addr)dst & 7) {
(*dst++) = (*src++);
length--;
}
ddsrc = (mlib_d64 *)vis_alignaddr((void *)src, 0);
ddst = (mlib_d64 *)dst;
rest_64 = length & 3;
len_64 = length >> 2;
even_length = len_64 << 2;
dd2 = ddsrc[0];
if (!((mlib_addr)(src) & 7)) {
/*
* Source vector is 64-bit aligned. We can process it without
* vis_faligndata.
* Peeling of 1 iteration.
*/
if (i = (len_64 & 1)) {
dd3 = (*ddsrc++);
/*
* Now obtaining of the 4*32 - signed objects
*/
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd3));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd3));
}
#pragma pipeloop(1)
#pragma unroll(1)
for (; i < len_64; i += 2) {
dd3 = (*ddsrc++);
dd4 = (*ddsrc++);
/*
* Now obtaining of the 4*32 - signed objects
*/
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd3));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd3));
/*
* Now obtaining of the 4*32 - signed objects
*/
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd4));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd4));
}
} else {
/*
* Source vector is not 64-bit aligned. Use vis_faligndata.
* Peeling of 1 iteration.
*/
i = 1;
if ((len_64 & 1)) {
i++;
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + 1);
dd3 = vis_faligndata(dd1, dd2);
/*
* Now obtaining of the 4*32 - signed objects
*/
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd3));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd3));
}
/*
* Now loop with step == 2.
*/
#pragma pipeloop(1)
#pragma unroll(1)
for (; i <= len_64; i += 2) {
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + i);
dd3 = vis_faligndata(dd1, dd2);
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + i + 1);
dd4 = vis_faligndata(dd1, dd2);
/*
* Now obtaining of the 4*32 - signed objects
*/
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd3));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd3));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_hi(dd4));
(*ddst++) =
vis_fmuld8ulx16(two_16_ones, vis_read_lo(dd4));
}
}
for (i = 0; i < rest_64; i++)
dst[even_length + i] = src[even_length + i];
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorConvert_S32_S8_Mod(
mlib_s32 *z,
const mlib_s8 *x,
mlib_s32 n)
{
mlib_s8 *psrc = (mlib_s8 *)x;
mlib_s32 *pdst = (mlib_s32 *)z;
mlib_f32 fone = vis_to_float(0x10001);
mlib_d64 *dpsrc, dsrc0, dsrc1, dsrc, dst0, dst1, dst2, dst3, done =
vis_to_double_dup(0x1000100);
mlib_s32 i = 0;
if (n <= 0)
return (MLIB_FAILURE);
if ((mlib_addr)pdst & 7) {
(*pdst++) = (*psrc++);
i = 1;
}
dpsrc = (mlib_d64 *)vis_alignaddr(psrc, 0);
dsrc = vis_ld_d64_nf(dpsrc);
vis_write_bmask(0x00012223, 0);
if ((mlib_addr)psrc & 7) {
dsrc1 = vis_ld_d64_nf(dpsrc + 1);
dsrc = vis_faligndata(dsrc, dsrc1);
#pragma pipeloop(1)
#pragma unroll(1)
for (; i <= (n - 8); i += 8) {
dst1 = vis_fpmerge(vis_read_hi(dsrc),
vis_read_hi(dsrc));
dst1 = vis_fmul8sux16(dst1, done);
dst0 = vis_bshuffle(dst1, dst1);
dst1 = vis_fmuld8ulx16(fone, vis_read_lo(dst1));
dst3 = vis_fpmerge(vis_read_lo(dsrc),
vis_read_lo(dsrc));
dst3 = vis_fmul8sux16(dst3, done);
dst2 = vis_fmuld8ulx16(fone, vis_read_hi(dst3));
dst3 = vis_fmuld8ulx16(fone, vis_read_lo(dst3));
dsrc0 = dsrc1;
dsrc1 = vis_ld_d64_nf(dpsrc + 2);
dsrc = vis_faligndata(dsrc0, dsrc1);
((mlib_d64 *)pdst)[0] = dst0;
((mlib_d64 *)pdst)[1] = dst1;
((mlib_d64 *)pdst)[2] = dst2;
((mlib_d64 *)pdst)[3] = dst3;
pdst += 8;
psrc += 8;
dpsrc++;
}
} else {
#pragma pipeloop(1)
#pragma unroll(1)
for (; i <= (n - 8); i += 8) {
dst1 = vis_fpmerge(vis_read_hi(dsrc),
vis_read_hi(dsrc));
dst1 = vis_fmul8sux16(dst1, done);
dst0 = vis_bshuffle(dst1, dst1);
dst1 = vis_fmuld8ulx16(fone, vis_read_lo(dst1));
dst3 = vis_fpmerge(vis_read_lo(dsrc),
vis_read_lo(dsrc));
dst3 = vis_fmul8sux16(dst3, done);
dst2 = vis_bshuffle(dst3, dst3);
dst3 = vis_fmuld8ulx16(fone, vis_read_lo(dst3));
dsrc = vis_ld_d64_nf(dpsrc + 1);
((mlib_d64 *)pdst)[0] = dst0;
((mlib_d64 *)pdst)[1] = dst1;
((mlib_d64 *)pdst)[2] = dst2;
((mlib_d64 *)pdst)[3] = dst3;
pdst += 8;
psrc += 8;
dpsrc++;
}
}
for (; i < n; i++)
(*pdst++) = (*psrc++);
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorConvert_S16_S8_Mod(
mlib_s16 *z,
const mlib_s8 *x,
mlib_s32 n)
{
mlib_s32 i;
const mlib_s8 *src = x;
mlib_s16 *dst = z;
mlib_d64 *ddsrc, *ddst;
mlib_d64 four_16_ones = vis_to_double_dup(0x01000100);
mlib_f32 fzero = vis_fzeros();
mlib_s32 len_64, even_length, rest_64, length = n, off;
mlib_d64 dd0, dd1, dd2, dd4, dd5, dd6, dd7;
if (length < 16) {
EXPAND(mlib_s8, mlib_s16);
}
while ((mlib_addr)dst & 7) {
(*dst++) = (*src++);
length--;
}
ddsrc = (mlib_d64 *)vis_alignaddr((void *)src, 0);
ddst = (mlib_d64 *)dst;
rest_64 = length & 7;
len_64 = length >> 3;
even_length = len_64 << 3;
dd2 = ddsrc[0];
off = (mlib_addr)src & 7;
if (!off) {
/*
* Both vectors are 64-bit aligned.
*/
/*
* Peeling of 1 iteration.
*/
if (i = (len_64 & 1)) {
dd1 = (*ddsrc++);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_hi(dd1),
fzero), four_16_ones);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_lo(dd1),
fzero), four_16_ones);
}
#pragma pipeloop(0)
#pragma unroll(4)
for (; i < len_64; i += 2) {
dd1 = (*ddsrc++);
dd2 = (*ddsrc++);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_hi(dd1),
fzero), four_16_ones);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_lo(dd1),
fzero), four_16_ones);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_hi(dd2),
fzero), four_16_ones);
(*ddst++) =
vis_fmul8sux16(vis_fpmerge(vis_read_lo(dd2),
fzero), four_16_ones);
}
} else {
/*
* Source vector is not 64-bit aligned.
* Peeling of 1 iteration. Then loop with step==2.
*/
vis_alignaddr((void *)0, 1);
vis_write_bmask(0x11111111 * off, 0x04152637);
i = 1;
if (len_64 & 1) {
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + 1); i++;
dd4 = vis_bshuffle(dd1, dd2);
dd5 = vis_faligndata(dd4, dd4);
(*ddst++) = vis_fmul8sux16(dd4, four_16_ones);
(*ddst++) = vis_fmul8sux16(dd5, four_16_ones);
}
#pragma pipeloop(0)
#pragma unroll(4)
for (; i <= len_64; i += 2) {
dd0 = dd2;
dd1 = vis_ld_d64_nf(ddsrc + i);
dd2 = vis_ld_d64_nf(ddsrc + i + 1);
dd4 = vis_bshuffle(dd0, dd1);
dd6 = vis_bshuffle(dd1, dd2);
dd5 = vis_faligndata(dd4, dd4);
dd7 = vis_faligndata(dd6, dd6);
(*ddst++) = vis_fmul8sux16(dd4, four_16_ones);
(*ddst++) = vis_fmul8sux16(dd5, four_16_ones);
(*ddst++) = vis_fmul8sux16(dd6, four_16_ones);
(*ddst++) = vis_fmul8sux16(dd7, four_16_ones);
}
}
for (i = 0; i < rest_64; i++)
dst[even_length + i] = src[even_length + i];
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorConvert_S32_U8_Mod(
mlib_s32 *z,
const mlib_u8 *x,
mlib_s32 n)
{
mlib_u8 *psrc = (mlib_u8 *)x;
mlib_s32 *pdst = (mlib_s32 *)z;
mlib_f32 fzero = vis_fzero(), fone1 = vis_to_float(0x100), fone2 =
vis_to_float(0x10001);
mlib_d64 *dpsrc, dsrc0, dsrc1, dsrc, dst0, dst1, dst2, dst3;
mlib_s32 i = 0, off;
if (n <= 0)
return (MLIB_FAILURE);
if ((mlib_addr)pdst & 7) {
(*pdst++) = (*psrc++);
i = 1;
}
dpsrc = (mlib_d64 *)vis_alignaddr(psrc, 0);
dsrc = dpsrc[0];
off = (mlib_addr)psrc & 7;
if (off) {
dsrc1 = dsrc;
vis_alignaddr((void *)0, 7);
vis_write_bmask(0x11111111 * off, 0x40516273);
#pragma pipeloop(0)
#pragma unroll(2)
for (; i <= (n - 8); i += 8) {
dsrc0 = dsrc1;
dsrc1 = vis_ld_d64_nf(dpsrc + 1);
dsrc = vis_bshuffle(dsrc0, dsrc1);
dst0 = vis_fmuld8ulx16(vis_read_hi(dsrc), fone2);
dst1 = vis_fmuld8ulx16(vis_read_lo(dsrc), fone2);
dsrc = vis_faligndata(dsrc, dsrc);
dst2 = vis_fmuld8ulx16(vis_read_hi(dsrc), fone2);
dst3 = vis_fmuld8ulx16(vis_read_lo(dsrc), fone2);
((mlib_d64 *)pdst)[0] = dst0;
((mlib_d64 *)pdst)[1] = dst1;
((mlib_d64 *)pdst)[2] = dst2;
((mlib_d64 *)pdst)[3] = dst3;
pdst += 8;
psrc += 8;
dpsrc++;
}
} else {
#pragma pipeloop(1)
#pragma unroll(1)
for (; i <= (n - 8); i += 8) {
dst1 = vis_fmul8x16al(vis_read_hi(dsrc), fone1);
dst0 = vis_fpmerge(fzero, vis_read_hi(dst1));
dst1 = vis_fpmerge(fzero, vis_read_lo(dst1));
dst3 = vis_fpmerge(vis_read_lo(dsrc),
vis_read_lo(dsrc));
dst2 = vis_fmuld8ulx16(vis_read_hi(dst3), fone2);
dst3 = vis_fmuld8ulx16(vis_read_lo(dst3), fone2);
dsrc = vis_ld_d64_nf(dpsrc + 1);
((mlib_d64 *)pdst)[0] = dst0;
((mlib_d64 *)pdst)[1] = dst1;
((mlib_d64 *)pdst)[2] = dst2;
((mlib_d64 *)pdst)[3] = dst3;
pdst += 8;
psrc += 8;
dpsrc++;
}
}
for (; i < n; i++)
(*pdst++) = (*psrc++);
return (MLIB_SUCCESS);
}
void mlib_v_ImageLookUpSI_S16_U16_3_D1(const mlib_s16 *src,
mlib_u16 *dst,
mlib_s32 xsize,
const mlib_u16 **table)
{
mlib_s16 *sp; /* pointer to source data */
mlib_u16 *dl; /* pointer to start of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2, t3; /* destination data */
mlib_d64 acc0, acc1, acc2; /* destination data */
mlib_s32 i; /* loop variable */
const mlib_u16 *tab0 = &table[0][32768];
const mlib_u16 *tab1 = &table[1][32768];
const mlib_u16 *tab2 = &table[2][32768];
mlib_s32 s00, s01, s02, s03;
sp = (void *)src;
dl = dst;
dp = (mlib_d64 *) dl;
vis_alignaddr((void *) 0, 6);
i = 0;
if (xsize >= 4) {
s00 = (sp[0] << 1);
s01 = (sp[1] << 1);
s02 = (sp[2] << 1);
s03 = (sp[3] << 1);
sp += 4;
#pragma pipeloop(0)
for(i = 0; i <= xsize - 8; i+=4, sp+=4) {
t3 = VIS_LD_U16_I(tab0, s01);
t2 = VIS_LD_U16_I(tab2, s00);
t1 = VIS_LD_U16_I(tab1, s00);
t0 = VIS_LD_U16_I(tab0, s00);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
t3 = VIS_LD_U16_I(tab1, s02);
t2 = VIS_LD_U16_I(tab0, s02);
t1 = VIS_LD_U16_I(tab2, s01);
t0 = VIS_LD_U16_I(tab1, s01);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc1 = vis_faligndata(t1, acc1);
acc1 = vis_faligndata(t0, acc1);
t3 = VIS_LD_U16_I(tab2, s03);
t2 = VIS_LD_U16_I(tab1, s03);
t1 = VIS_LD_U16_I(tab0, s03);
t0 = VIS_LD_U16_I(tab2, s02);
acc2 = vis_faligndata(t3, acc2);
acc2 = vis_faligndata(t2, acc2);
acc2 = vis_faligndata(t1, acc2);
acc2 = vis_faligndata(t0, acc2);
s00 = (sp[0] << 1);
s01 = (sp[1] << 1);
s02 = (sp[2] << 1);
s03 = (sp[3] << 1);
*dp++ = acc0;
*dp++ = acc1;
*dp++ = acc2;
}
t3 = VIS_LD_U16_I(tab0, s01);
t2 = VIS_LD_U16_I(tab2, s00);
t1 = VIS_LD_U16_I(tab1, s00);
t0 = VIS_LD_U16_I(tab0, s00);
acc0 = vis_faligndata(t3, acc0);
acc0 = vis_faligndata(t2, acc0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
t3 = VIS_LD_U16_I(tab1, s02);
t2 = VIS_LD_U16_I(tab0, s02);
t1 = VIS_LD_U16_I(tab2, s01);
t0 = VIS_LD_U16_I(tab1, s01);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc1 = vis_faligndata(t1, acc1);
acc1 = vis_faligndata(t0, acc1);
t3 = VIS_LD_U16_I(tab2, s03);
t2 = VIS_LD_U16_I(tab1, s03);
t1 = VIS_LD_U16_I(tab0, s03);
t0 = VIS_LD_U16_I(tab2, s02);
acc2 = vis_faligndata(t3, acc2);
acc2 = vis_faligndata(t2, acc2);
acc2 = vis_faligndata(t1, acc2);
acc2 = vis_faligndata(t0, acc2);
*dp++ = acc0;
*dp++ = acc1;
*dp++ = acc2;
i += 4;
}
dl = (mlib_u16*)dp;
#pragma pipeloop(0)
for (; i < xsize; i++) {
s00 = sp[0];
dl[0] = tab0[s00];
dl[1] = tab1[s00];
dl[2] = tab2[s00];
dl += 3; sp ++;
}
}
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);
}
void mlib_v_ImageLookUpSI_U16_U8_2_DstA8D1(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; /* source data */
mlib_u16 *dl; /* pointer to start of destination */
mlib_u16 *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 = (mlib_u16 *) dst;
dp = (mlib_d64 *) dl;
dend = dl + xsize - 1;
vis_alignaddr((void *)0, 7);
if (xsize >= 4) {
s0 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
sp += 4;
#pragma pipeloop(0)
for (i = 0; i <= xsize - 8; i += 4, sp += 4) {
t7 = VIS_LD_U8_I(tab1, s3);
t6 = VIS_LD_U8_I(tab0, s3);
t5 = VIS_LD_U8_I(tab1, s2);
t4 = VIS_LD_U8_I(tab0, s2);
t3 = VIS_LD_U8_I(tab1, s1);
t2 = VIS_LD_U8_I(tab0, s1);
t1 = VIS_LD_U8_I(tab1, s0);
t0 = VIS_LD_U8_I(tab0, 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 = sp[0];
s1 = sp[1];
s2 = sp[2];
s3 = sp[3];
*dp++ = acc;
}
t7 = VIS_LD_U8_I(tab1, s3);
t6 = VIS_LD_U8_I(tab0, s3);
t5 = VIS_LD_U8_I(tab1, s2);
t4 = VIS_LD_U8_I(tab0, s2);
t3 = VIS_LD_U8_I(tab1, s1);
t2 = VIS_LD_U8_I(tab0, s1);
t1 = VIS_LD_U8_I(tab1, s0);
t0 = VIS_LD_U8_I(tab0, 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);
*dp++ = acc;
}
if ((mlib_addr) dp <= (mlib_addr) dend) {
num = (mlib_u16 *) dend - (mlib_u16 *) dp;
sp += num;
num++;
#pragma pipeloop(0)
for (i = 0; i < num; i++) {
s0 = (mlib_s32) * sp;
sp--;
t0 = VIS_LD_U8_I(tab1, s0);
acc = vis_faligndata(t0, acc);
t0 = VIS_LD_U8_I(tab0, s0);
acc = vis_faligndata(t0, acc);
}
emask = vis_edge16(dp, dend);
vis_pst_16(acc, dp, emask);
}
}
void mlib_v_ImageLookUpSI_S16_U16_2_DstA8D1(const mlib_s16 *src,
mlib_u16 *dst,
mlib_s32 xsize,
const mlib_u16 **table)
{
mlib_s16 *sp; /* pointer to source data */
mlib_s32 s0, s1; /* source data */
mlib_u16 *dl; /* pointer to start of destination */
mlib_d64 *dp; /* aligned pointer to destination */
mlib_d64 t0, t1, t2; /* destination data */
mlib_d64 t3, acc; /* destination data */
mlib_s32 i; /* loop variable */
const mlib_u16 *tab0 = &table[0][32768];
const mlib_u16 *tab1 = &table[1][32768];
sp = (void *)src;
dl = dst;
dp = (mlib_d64 *) dl;
vis_alignaddr((void *) 0, 6);
if (xsize >= 2) {
s0 = (sp[0] << 1);
s1 = (sp[1] << 1);
sp += 2;
#pragma pipeloop(0)
for(i = 0; i <= xsize - 4; i+=2, sp+=2) {
t3 = VIS_LD_U16_I(tab1, s1);
t2 = VIS_LD_U16_I(tab0, s1);
t1 = VIS_LD_U16_I(tab1, s0);
t0 = VIS_LD_U16_I(tab0, s0);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = (sp[0] << 1);
s1 = (sp[1] << 1);
*dp++ = acc;
}
t3 = VIS_LD_U16_I(tab1, s1);
t2 = VIS_LD_U16_I(tab0, s1);
t1 = VIS_LD_U16_I(tab1, s0);
t0 = VIS_LD_U16_I(tab0, s0);
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
*dp++ = acc;
}
if ((xsize & 1) != 0) {
s0 = (sp[0] << 1);
t1 = VIS_LD_U16_I(tab1, s0);
t0 = VIS_LD_U16_I(tab0, s0);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
*(mlib_f32*)dp = vis_read_hi(acc);
}
}
void
mlib_v_ImageLookUpSI_S32_S16_2_D1(
const mlib_s32 *src,
mlib_s16 *dst,
mlib_s32 xsize,
const mlib_s16 **table)
{
/* pointer to source data */
mlib_s32 *sp;
/* source data */
mlib_s32 s0, s1, s2;
/* pointer to start of destination */
mlib_s16 *dl;
/* aligned pointer to destination */
mlib_d64 *dp;
/* destination data */
mlib_d64 t0, t1, t2;
/* destination data */
mlib_d64 t3, acc;
/* loop variable */
mlib_s32 i;
const mlib_s16 *tab0 = &table[0][(mlib_u32)2147483648u];
const mlib_s16 *tab1 = &table[1][(mlib_u32)2147483648u];
sp = (void *)src;
dl = dst;
vis_alignaddr((void *)0, 6);
s0 = (*sp++);
(*dl++) = tab0[s0];
dp = (mlib_d64 *)dl;
xsize--;
if (xsize >= 2) {
s1 = sp[0];
s2 = sp[1];
sp += 2;
#pragma pipeloop(0)
for (i = 0; i <= xsize - 4; i += 2, sp += 2) {
t3 = VIS_LD_U16_I(tab0, ((mlib_addr)2 * s2));
t2 = VIS_LD_U16_I(tab1, ((mlib_addr)2 * s1));
t1 = VIS_LD_U16_I(tab0, ((mlib_addr)2 * s1));
t0 = VIS_LD_U16_I(tab1, ((mlib_addr)2 * s0));
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s2;
s1 = sp[0];
s2 = sp[1];
(*dp++) = acc;
}
t3 = VIS_LD_U16_I(tab0, ((mlib_addr)2 * s2));
t2 = VIS_LD_U16_I(tab1, ((mlib_addr)2 * s1));
t1 = VIS_LD_U16_I(tab0, ((mlib_addr)2 * s1));
t0 = VIS_LD_U16_I(tab1, ((mlib_addr)2 * s0));
acc = vis_faligndata(t3, acc);
acc = vis_faligndata(t2, acc);
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
s0 = s2;
(*dp++) = acc;
}
dl = (mlib_s16 *)dp;
if ((xsize & 1) != 0) {
s1 = sp[0];
t1 = VIS_LD_U16_I(tab0, ((mlib_addr)2 * s1));
t0 = VIS_LD_U16_I(tab1, ((mlib_addr)2 * s0));
acc = vis_faligndata(t1, acc);
acc = vis_faligndata(t0, acc);
*(mlib_f32 *)dp = vis_read_hi(acc);
s0 = s1;
dl += 2;
}
*dl = tab1[s0];
}
void mlib_v_ImageLookUpSI_U16_U8_4_DstOff3_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; /* source data */
mlib_u8 *dl; /* pointer to start 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 i; /* loop variable */
const mlib_u8 *tab0 = &table[0][0];
const mlib_u8 *tab1 = &table[1][0];
const mlib_u8 *tab2 = &table[2][0];
const mlib_u8 *tab3 = &table[3][0];
sp = (void *)src;
dl = dst;
dp = (mlib_d64 *) dl;
vis_alignaddr((void *)0, 7);
s0 = *sp++;
if (xsize >= 2) {
s1 = sp[0];
s2 = sp[1];
sp += 2;
#pragma pipeloop(0)
for (i = 0; i <= xsize - 4; i += 2, sp += 2) {
t7 = VIS_LD_U8_I(tab2, s2);
t6 = VIS_LD_U8_I(tab1, s2);
t5 = VIS_LD_U8_I(tab0, s2);
t4 = VIS_LD_U8_I(tab3, s1);
t3 = VIS_LD_U8_I(tab2, s1);
t2 = VIS_LD_U8_I(tab1, s1);
t1 = VIS_LD_U8_I(tab0, s1);
t0 = VIS_LD_U8_I(tab3, 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 = s2;
s1 = sp[0];
s2 = sp[1];
*dp++ = acc;
}
t7 = VIS_LD_U8_I(tab2, s2);
t6 = VIS_LD_U8_I(tab1, s2);
t5 = VIS_LD_U8_I(tab0, s2);
t4 = VIS_LD_U8_I(tab3, s1);
t3 = VIS_LD_U8_I(tab2, s1);
t2 = VIS_LD_U8_I(tab1, s1);
t1 = VIS_LD_U8_I(tab0, s1);
t0 = VIS_LD_U8_I(tab3, 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 = s2;
*dp++ = acc;
}
dl = (mlib_u8 *) dp;
if ((xsize & 1) != 0) {
s1 = sp[0];
t7 = VIS_LD_U8_I(tab2, s1);
t6 = VIS_LD_U8_I(tab1, s1);
t5 = VIS_LD_U8_I(tab0, s1);
t4 = VIS_LD_U8_I(tab3, s0);
acc = vis_faligndata(t7, acc);
acc = vis_faligndata(t6, acc);
acc = vis_faligndata(t5, acc);
acc = vis_faligndata(t4, acc);
*(mlib_f32 *) dl = vis_read_hi(acc);
dl += 4;
s0 = s1;
}
dl[0] = tab3[s0];
}
mlib_status
__mlib_VectorConjRev_S16C_S16C_Sat(
mlib_s16 *zz,
const mlib_s16 *xx,
mlib_s32 n)
{
mlib_s16 *x = (mlib_s16 *)xx, *z = (mlib_s16 *)zz;
mlib_s16 *src = (mlib_s16 *)x, *dst = (mlib_s16 *)&z[2 * n];
mlib_d64 *dsrc, *ddst;
mlib_d64 d1, d2, d3, d4, dl, dh, d_rest;
mlib_d64 dlog0 = vis_to_double_dup(0x0000ffff), dtwo =
vis_to_double(0, 2);
mlib_f32 f_two = vis_to_float(0x20002);
mlib_s16 c;
mlib_s32 i, rest_64, len_64, even_length, odd = 0, length =
(mlib_s32)n * 2;
mlib_s32 re_part;
CHECK(x, z);
if ((n < 16)) {
CONJREVC(mlib_s16,
MLIB_S16_MAX,
MLIB_S16_MIN);
}
while (((mlib_addr)dst) & 7) {
if ((c = src[1]) == MLIB_S16_MIN)
*--dst = MLIB_S16_MAX;
else
*--dst = -c;
length -= 2;
src += 2;
if (((mlib_addr)dst) & 7) {
*--dst = src[-2];
} else {
re_part = src[-2];
odd = 1;
break;
}
}
vis_write_gsr(15 << 3);
ddst = (mlib_d64 *)dst;
rest_64 = length & 3;
len_64 = length >> 2;
even_length = len_64 << 2;
if (!odd) {
/*
* Aligning loop finished with imaginary part. The following processing
* starts with real part.
*/
if (!((mlib_addr)src & 7)) {
/*
* Src address is 8-byte aligned.
*/
dsrc = (mlib_d64 *)src;
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d3 = (*dsrc++);
CONJ16;
*--ddst = d4;
}
} else {
dsrc = (mlib_d64 *)vis_alignaddr(src, 0);
d2 = (*dsrc++);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = (*dsrc++);
d3 = vis_faligndata(d1, d2);
CONJ16;
*--ddst = d4;
}
}
} else {
/*
* Aligning loop finished with real part. Th following processing
* starts with imaginary part.
*/
if (!((mlib_addr)src & 7)) {
/*
* Src address is 8-byte aligned.
*/
dsrc = (mlib_d64 *)vis_alignaddr(src, 2);
d_rest = vis_to_double((re_part << 16), 0);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d3 = (*dsrc++);
CONJ16;
*--ddst = vis_faligndata(d4, d_rest);
d_rest = d4;
}
ddst--;
d_rest = vis_faligndata(d_rest, d_rest);
vis_pst_16(d_rest, ddst, 0x1);
} else {
dsrc = (mlib_d64 *)vis_alignaddr(src, 0);
d2 = (*dsrc++);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = (*dsrc++);
d3 = vis_faligndata(d1, d2);
CONJ16;
*--ddst = d4;
}
vis_write_gsr(2);
d2 = *ddst;
d3 = vis_faligndata(d1, d2);
vis_pst_16(d3, (ddst - 1), 0x1);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = *(ddst + 1);
(*ddst++) = vis_faligndata(d1, d2);
}
dst[-1] = re_part;
}
dst--;
}
if (!rest_64)
return (MLIB_SUCCESS);
for (i = 0; i < rest_64; i += 2) {
dst[-even_length - 2 - i] = src[even_length + i];
if ((c = src[even_length + i + 1]) == MLIB_S16_MIN)
dst[-even_length - 2 - i + 1] = MLIB_S16_MAX;
else
dst[-even_length - 2 - i + 1] = -c;
}
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VectorConjRev_S8C_S8C_Sat(
mlib_s8 *zz,
const mlib_s8 *xx,
mlib_s32 n)
{
const mlib_s8 *x = xx;
mlib_s8 *z = zz;
mlib_s8 *src = (mlib_s8 *)x, *dst = z + 2 * (n);
mlib_d64 *dsrc, *ddst;
mlib_d64 d1, d2, d3, d4, dl, dh, d_rest;
mlib_d64 dcntr0 = vis_to_double_dup(0x00800080);
mlib_d64 dxor0 = vis_to_double_dup(0x007f007f);
mlib_d64 done = vis_to_double_dup(1);
mlib_s8 c;
mlib_s32 i, rest_64, len_64, even_length, odd = 0, length =
(mlib_s32)n * 2;
mlib_s32 re_part;
mlib_f32 f_null = vis_to_float(0);
CHECK(x, z);
if (n < 8) {
CONJREVC(mlib_s8,
MLIB_S8_MAX,
MLIB_S8_MIN);
}
while (((mlib_addr)dst) & 7) {
if ((c = src[1]) == MLIB_S8_MIN)
*--dst = MLIB_S8_MAX;
else
*--dst = -c;
length -= 2;
src += 2;
if (((mlib_addr)dst) & 7) {
*--dst = src[-2];
} else {
re_part = src[-2];
odd = 1;
break;
}
}
vis_write_gsr(7 << 3);
ddst = (mlib_d64 *)dst;
rest_64 = length & 7;
len_64 = length >> 3;
even_length = len_64 << 3;
if (!odd) {
/*
* Aligning loop finished with imaginary part. The following processing
* starts with real part.
*/
if (!((mlib_addr)src & 7)) {
/*
* Src address is 8-byte aligned.
*/
dsrc = (mlib_d64 *)src;
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d3 = (*dsrc++);
CONJ8;
*--ddst = d4;
}
} else {
dsrc = (mlib_d64 *)vis_alignaddr(src, 0);
d2 = (*dsrc++);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = (*dsrc++);
d3 = vis_faligndata(d1, d2);
CONJ8;
*--ddst = d4;
}
}
} else {
/*
* Aligning loop finished with real part. Th following processing
* starts with imaginary part.
*/
if (!((mlib_addr)src & 7)) {
/*
* Src address is 8-byte aligned.
*/
dsrc = (mlib_d64 *)vis_alignaddr(src, 1);
d_rest = vis_to_double((re_part << 24), 0);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d3 = (*dsrc++);
CONJ8;
*--ddst = vis_faligndata(d4, d_rest);
d_rest = d4;
}
ddst--;
d_rest = vis_faligndata(d_rest, d_rest);
vis_pst_8(d_rest, ddst, 0x1);
} else {
dsrc = (mlib_d64 *)vis_alignaddr(src, 0);
d2 = (*dsrc++);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = (*dsrc++);
d3 = vis_faligndata(d1, d2);
CONJ8;
*--ddst = d4;
}
vis_write_gsr(1);
d2 = *ddst;
d3 = vis_faligndata(d1, d2);
vis_pst_8(d3, (ddst - 1), 0x1);
#pragma pipeloop(0)
for (i = 0; i < len_64; i++) {
d1 = d2;
d2 = *(ddst + 1);
(*ddst++) = vis_faligndata(d1, d2);
}
dst[-1] = re_part;
}
dst--;
}
if (!rest_64)
return (MLIB_SUCCESS);
for (i = 0; i < rest_64; i += 2) {
dst[-even_length - 2 - i] = src[even_length + i];
if ((c = src[even_length + i + 1]) == MLIB_S8_MIN)
dst[-even_length - 2 - i + 1] = MLIB_S8_MAX;
else
dst[-even_length - 2 - i + 1] = -c;
}
return (MLIB_SUCCESS);
}
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_VideoColorYUV2RGB444_nonalign(
mlib_u8 *rgb,
const mlib_u8 *y,
const mlib_u8 *u,
const mlib_u8 *v,
mlib_s32 width,
mlib_s32 height,
mlib_s32 rgb_stride,
mlib_s32 yuv_stride)
{
/* all. pointer to y, u, v */
mlib_d64 *spy, *dfu, *dfv;
/* y data */
mlib_d64 dy0, dy1, dy3;
mlib_d64 du, dv, du0, du1, dv0, dv1;
/* (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_dup(0x1be01be0);
mlib_d64 k_135_6352 = vis_to_double_dup(0x10f410f4);
mlib_d64 k_276_9856 = vis_to_double_dup(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 red, green, blue, *ddp, dd0, dd1, dd2;
/* loop variable */
mlib_s32 i, j;
mlib_d64 *buf, BUFF[16 * 1024];
mlib_u8 *tmp, *dp;
if (width * 3 > 16 * 1024) {
tmp = __mlib_malloc(width * 3 * sizeof (mlib_u8) + 7);
if (tmp == NULL)
return (MLIB_FAILURE);
buf = (mlib_d64 *)((mlib_addr)(tmp + 7) & ~7);
} else {
buf = (mlib_d64 *)BUFF;
}
dp = (mlib_u8 *)buf;
ddp = (mlib_d64 *)dp;
for (j = 0; j < height; j++) {
dfu = (mlib_d64 *)vis_alignaddr((void *)u, 0);
du0 = (*dfu++);
du1 = vis_ld_d64_nf(dfu); dfu++;
du = vis_faligndata(du0, du1);
du0 = du1;
dfv = (mlib_d64 *)vis_alignaddr((void *)v, 0);
dv0 = (*dfv++);
dv1 = vis_ld_d64_nf(dfv); dfv++;
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);
spy = (mlib_d64 *)vis_alignaddr((void *)y, 0);
dy0 = (*spy++);
dy3 = vis_ld_d64_nf(spy); spy++;
dy1 = vis_faligndata(dy0, dy3);
dy0 = dy3;
/* 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);
vis_alignaddr((void *)u, 0);
du1 = vis_ld_d64_nf(dfu); dfu++;
du = vis_faligndata(du0, du1);
du0 = du1;
/* Y*1.1644 */
y_11644_hi = vis_fmul8x16au(vis_read_hi(dy1), k12);
b_hi = vis_fpsub16(u_20184_hi, k_276_9856);
vis_alignaddr((void *)v, 0);
dv1 = vis_ld_d64_nf(dfv); dfv++;
dv = vis_faligndata(dv0, dv1);
dv0 = dv1;
/* Y*1.1644 */
y_11644_lo = vis_fmul8x16au(vis_read_lo(dy1), k12);
b_lo = vis_fpsub16(u_20184_lo, k_276_9856);
/* U*(-0.3920); */
u_3920_hi = vis_fmul8x16au(vis_read_hi(du), k34);
r_hi = vis_fpsub16(v_15966_hi, k_222_9952);
/* V*(-0.8132); */
v_8132_hi = vis_fmul8x16al(vis_read_hi(dv), k34);
r_lo = vis_fpsub16(v_15966_lo, k_222_9952);
u_3920_lo = vis_fmul8x16au(vis_read_lo(du), k34);
g_hi = vis_fpadd16(g_hi, y_11644_hi);
v_8132_lo = vis_fmul8x16al(vis_read_lo(dv), k34);
g_lo = vis_fpadd16(g_lo, y_11644_lo);
green = vis_fpack16_pair(g_hi, g_lo);
b_hi = vis_fpadd16(b_hi, y_11644_hi);
b_lo = vis_fpadd16(b_lo, y_11644_lo);
blue = vis_fpack16_pair(b_hi, b_lo);
r_hi = vis_fpadd16(r_hi, y_11644_hi);
r_lo = vis_fpadd16(r_lo, y_11644_lo);
red = vis_fpack16_pair(r_hi, r_lo);
vis_alignaddr((void *)y, 0);
dy3 = vis_ld_d64_nf(spy); spy++;
dy1 = vis_faligndata(dy0, dy3);
dy0 = dy3;
#pragma pipeloop(0)
for (i = 0; i <= width - 8; i += 8) {
vis_write_bmask(0x0801902A, 0);
dd0 = vis_bshuffle(red, green);
vis_write_bmask(0x03B04C05, 0);
dd1 = vis_bshuffle(red, green);
vis_write_bmask(0xD06E07F0, 0);
dd2 = vis_bshuffle(red, green);
vis_write_bmask(0x01834967, 0);
ddp[0] = vis_bshuffle(dd0, blue);
vis_write_bmask(0xA12B45C7, 0);
ddp[1] = vis_bshuffle(dd1, blue);
vis_write_bmask(0x0D23E56F, 0);
ddp[2] = vis_bshuffle(dd2, blue);
/* 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);
vis_alignaddr((void *)u, 0);
du1 = vis_ld_d64_nf(dfu); dfu++;
du = vis_faligndata(du0, du1);
du0 = du1;
/* Y*1.1644 */
y_11644_hi = vis_fmul8x16au(vis_read_hi(dy1), k12);
b_hi = vis_fpsub16(u_20184_hi, k_276_9856);
vis_alignaddr((void *)v, 0);
dv1 = vis_ld_d64_nf(dfv); dfv++;
dv = vis_faligndata(dv0, dv1);
dv0 = dv1;
/* Y*1.1644 */
y_11644_lo = vis_fmul8x16au(vis_read_lo(dy1), k12);
b_lo = vis_fpsub16(u_20184_lo, k_276_9856);
/* U*(-0.3920); */
u_3920_hi = vis_fmul8x16au(vis_read_hi(du), k34);
r_hi = vis_fpsub16(v_15966_hi, k_222_9952);
/* V*(-0.8132); */
v_8132_hi = vis_fmul8x16al(vis_read_hi(dv), k34);
r_lo = vis_fpsub16(v_15966_lo, k_222_9952);
u_3920_lo = vis_fmul8x16au(vis_read_lo(du), k34);
g_hi = vis_fpadd16(g_hi, y_11644_hi);
v_8132_lo = vis_fmul8x16al(vis_read_lo(dv), k34);
g_lo = vis_fpadd16(g_lo, y_11644_lo);
green = vis_fpack16_pair(g_hi, g_lo);
b_hi = vis_fpadd16(b_hi, y_11644_hi);
b_lo = vis_fpadd16(b_lo, y_11644_lo);
blue = vis_fpack16_pair(b_hi, b_lo);
r_hi = vis_fpadd16(r_hi, y_11644_hi);
r_lo = vis_fpadd16(r_lo, y_11644_lo);
red = vis_fpack16_pair(r_hi, r_lo);
vis_alignaddr((void *)y, 0);
dy3 = vis_ld_d64_nf(spy); spy++;
dy1 = vis_faligndata(dy0, dy3);
dy0 = dy3;
ddp += 3;
}
dp = (mlib_u8 *)ddp;
vis_alignaddr((void *)(width - i), 0);
blue = vis_faligndata(blue, blue);
green = vis_faligndata(green, green);
red = vis_faligndata(red, red);
dp += ((width - i - 1) * 3);
vis_alignaddr((void *)spy, 7);
for (; i < width; i++) {
STORE_PIXEL(0, 1, 2);
dp -= 3;
}
__mlib_VectorCopy_U8(rgb, (mlib_u8 *)buf, width * 3);
rgb += rgb_stride;
dp = (mlib_u8 *)buf;
ddp = (mlib_d64 *)dp;
y += yuv_stride;
u += yuv_stride;
v += yuv_stride;
}
if (width * 3 > 16 * 1024)
__mlib_free(tmp);
return (MLIB_SUCCESS);
}
void
mlib_v_ImageLookUpSI_U16_S16_2_D1(
const mlib_u16 *src,
mlib_s16 *dst,
mlib_s32 xsize,
const mlib_s16 **table)
{
/* pointer to source data */
mlib_u16 *sp;
/* source data */
mlib_s32 s0, s1, s2;
/* pointer to start of destination */
mlib_s16 *dl;
/* aligned pointer to destination */
mlib_d64 *dp;
/* destination data */
mlib_d64 t0, t1, t2;
/* destination data */
mlib_d64 t3, acc0;
/* loop variable */
mlib_s32 i;
const mlib_s16 *tab0 = &table[0][0];
const mlib_s16 *tab1 = &table[1][0];
/* destination data */
mlib_d64 acc1;
sp = (void *)src;
dl = dst;
vis_alignaddr((void *)0, 6);
s0 = (*sp++);
(*dl++) = tab0[s0];
dp = (mlib_d64 *)dl;
xsize--;
s0 <<= 1;
if (xsize >= 2) {
s1 = (sp[0] << 1);
s2 = (sp[1] << 1);
sp += 2;
vis_write_bmask(0x012389ab, 0);
#pragma pipeloop(0)
for (i = 0; i <= xsize - 4; i += 2, sp += 2) {
t3 = VIS_LD_U16_I(tab0, s2);
t2 = VIS_LD_U16_I(tab1, s1);
t1 = VIS_LD_U16_I(tab0, s1);
t0 = VIS_LD_U16_I(tab1, s0);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = s2;
s1 = (sp[0] << 1);
s2 = (sp[1] << 1);
(*dp++) = vis_bshuffle(acc0, acc1);
}
t3 = VIS_LD_U16_I(tab0, s2);
t2 = VIS_LD_U16_I(tab1, s1);
t1 = VIS_LD_U16_I(tab0, s1);
t0 = VIS_LD_U16_I(tab1, s0);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = s2;
(*dp++) = vis_bshuffle(acc0, acc1);
}
dl = (mlib_s16 *)dp;
if ((xsize & 1) != 0) {
s1 = (sp[0] << 1);
t1 = VIS_LD_U16_I(tab0, s1);
t0 = VIS_LD_U16_I(tab1, s0);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
*(mlib_f32 *)dp = vis_read_hi(acc0);
s0 = s1;
dl += 2;
}
s0 >>= 1;
*dl = tab1[s0];
}
mlib_status
__mlib_VectorConvert_S16_U8_Mod(
mlib_s16 *z,
const mlib_u8 *x,
mlib_s32 n)
{
mlib_s32 i;
const mlib_u8 *src = x;
mlib_s16 *dst = z;
mlib_d64 *ddsrc, *ddst;
mlib_s32 len_64, even_length, rest_64, length = n;
mlib_f32 fzero = vis_fzeros();
mlib_d64 dd1, dd2, dd3, dd4;
mlib_f32 fm = vis_to_float(0x100);
if (length < 16) {
EXPAND(mlib_u8, mlib_s16);
}
while ((mlib_addr)dst & 7) {
(*dst++) = (*src++);
length--;
}
ddsrc = (mlib_d64 *)vis_alignaddr((void *)src, 0);
ddst = (mlib_d64 *)dst;
rest_64 = length & 7;
len_64 = length >> 3;
even_length = len_64 << 3;
dd2 = ddsrc[0];
if (!((mlib_addr)src & 7)) {
/*
* Both vectors are 64-bit aligned. We can process without
* vis_faligndata
* Peeling the 1 iteration. Then loop with step==2.
*/
if (i = (len_64 & 1)) {
dd1 = (*ddsrc++);
(*ddst++) = vis_fpmerge(fzero, vis_read_hi(dd1));
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd1));
}
#pragma pipeloop(1)
#pragma unroll(1)
for (; i < len_64; i += 2) {
dd1 = (*ddsrc++);
dd2 = (*ddsrc++);
(*ddst++) = vis_fmul8x16al(vis_read_hi(dd1), fm);
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd1));
(*ddst++) = vis_fmul8x16al(vis_read_hi(dd2), fm);
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd2));
}
} else {
/*
* Source vector is not 64-bit aligned. Use vis_faligndata.
* Peeling the 1 iteration. Then loop with step==2.
*/
i = 1;
if (len_64 & 1) {
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + 1); i++;
dd3 = vis_faligndata(dd1, dd2);
(*ddst++) = vis_fpmerge(fzero, vis_read_hi(dd3));
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd3));
}
#pragma pipeloop(0)
#pragma unroll(2)
for (; i <= len_64; i += 2) {
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + i);
dd3 = vis_faligndata(dd1, dd2);
dd1 = dd2;
dd2 = vis_ld_d64_nf(ddsrc + i + 1);
dd4 = vis_faligndata(dd1, dd2);
(*ddst++) = vis_fmul8x16al(vis_read_hi(dd3), fm);
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd3));
(*ddst++) = vis_fmul8x16al(vis_read_hi(dd4), fm);
(*ddst++) = vis_fpmerge(fzero, vis_read_lo(dd4));
}
}
for (i = 0; i < rest_64; i++)
dst[even_length + i] = src[even_length + i];
return (MLIB_SUCCESS);
}
mlib_status
__mlib_VideoInterpX_U8_U8(
mlib_u8 *curr_block,
const mlib_u8 *ref_block,
mlib_s32 width,
mlib_s32 height,
mlib_s32 frame_stride,
mlib_s32 field_stride)
{
mlib_d64 s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, *sd, *dd;
mlib_d64 dzero = vis_fzero();
const mlib_f32 fm1 = vis_to_float(0x100);
mlib_f32 fzero = vis_read_hi(dzero);
mlib_d64 rounder = vis_fone();
mlib_s32 y;
vis_write_gsr((6 << 3) + ((mlib_u32)ref_block & 7));
dd = (mlib_d64 *)curr_block;
sd = (mlib_d64 *)((mlib_addr)ref_block & ~7);
if (width == 8) {
y = height >> 2;
if (((mlib_s32)(ref_block + 1) & 7)) {
do {
s0 = sd[0];
s1 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d0 = vis_faligndata(s0, s1);
s2 = sd[0];
s3 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d1 = vis_faligndata(s2, s3);
s4 = sd[0];
s5 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d2 = vis_faligndata(s4, s5);
s6 = sd[0];
s7 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d3 = vis_faligndata(s6, s7);
vis_alignaddr((void *)(ref_block + 1), 0);
s0 = vis_faligndata(s0, s1);
s1 = vis_faligndata(s2, s3);
s2 = vis_faligndata(s4, s5);
s3 = vis_faligndata(s6, s7);
MLIB_V_VIDEOINTERP(d0, d0, s0);
MLIB_V_VIDEOINTERP(d1, d1, s1);
MLIB_V_VIDEOINTERP(d2, d2, s2);
MLIB_V_VIDEOINTERP4(d3, d3, s3);
*dd = d0;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d1;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d2;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d3;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
vis_alignaddr((void *)ref_block, 0);
} while (--y);
} else {
do {
s0 = sd[0];
s1 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d0 = vis_faligndata(s0, s1);
s2 = sd[0];
s3 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d1 = vis_faligndata(s2, s3);
s4 = sd[0];
s5 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d2 = vis_faligndata(s4, s5);
s6 = sd[0];
s7 = sd[1];
sd = (mlib_d64 *)((mlib_u8 *)sd + field_stride);
d3 = vis_faligndata(s6, s7);
MLIB_V_VIDEOINTERP4(d0, d0, s1);
MLIB_V_VIDEOINTERP4(d1, d1, s3);
MLIB_V_VIDEOINTERP4(d2, d2, s5);
MLIB_V_VIDEOINTERP4(d3, d3, s7);
*dd = d0;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d1;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d2;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
*dd = d3;
dd = (mlib_d64 *)((mlib_u8 *)dd + field_stride);
} while (--y);
}
} else {
void
mlib_v_ImageLookUpSI_U16_S16_4_DstOff3_D1(
const mlib_u16 *src,
mlib_s16 *dst,
mlib_s32 xsize,
const mlib_s16 **table)
{
/* pointer to source data */
mlib_u16 *sp;
/* source data */
mlib_s32 s0, s1;
/* pointer to start of destination */
mlib_s16 *dl;
/* aligned pointer to destination */
mlib_d64 *dp;
/* destination data */
mlib_d64 t0, t1, t2, t3;
/* destination data */
mlib_d64 acc0;
/* loop variable */
mlib_s32 i;
const mlib_s16 *tab0 = &table[0][0];
const mlib_s16 *tab1 = &table[1][0];
const mlib_s16 *tab2 = &table[2][0];
const mlib_s16 *tab3 = &table[3][0];
/* destination data */
mlib_d64 acc1;
sp = (void *)src;
dl = dst;
dp = (mlib_d64 *)dl;
vis_alignaddr((void *)0, 6);
s0 = ((*sp++)) << 1;
if (xsize >= 1) {
s1 = ((*sp++)) << 1;
vis_write_bmask(0x012389ab, 0);
#pragma pipeloop(0)
for (i = 0; i <= xsize - 2; i++) {
t3 = VIS_LD_U16_I(tab2, s1);
t2 = VIS_LD_U16_I(tab1, s1);
t1 = VIS_LD_U16_I(tab0, s1);
t0 = VIS_LD_U16_I(tab3, s0);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = s1;
s1 = ((*sp++)) << 1;
(*dp++) = vis_bshuffle(acc0, acc1);
}
t3 = VIS_LD_U16_I(tab2, s1);
t2 = VIS_LD_U16_I(tab1, s1);
t1 = VIS_LD_U16_I(tab0, s1);
t0 = VIS_LD_U16_I(tab3, s0);
acc1 = vis_faligndata(t3, acc1);
acc1 = vis_faligndata(t2, acc1);
acc0 = vis_faligndata(t1, acc0);
acc0 = vis_faligndata(t0, acc0);
s0 = s1;
(*dp++) = vis_bshuffle(acc0, acc1);
}
dl = (mlib_s16 *)dp;
s0 >>= 1;
dl[0] = tab3[s0];
}
