in simd/powerpc/jdsample-altivec.c [129:314]
void jsimd_h2v2_fancy_upsample_altivec(int max_v_samp_factor,
JDIMENSION downsampled_width,
JSAMPARRAY input_data,
JSAMPARRAY *output_data_ptr)
{
JSAMPARRAY output_data = *output_data_ptr;
JSAMPROW inptr_1, inptr0, inptr1, outptr0, outptr1;
int inrow, outrow, incol;
__vector unsigned char this_1, this0, this1, out;
__vector short this_1l, this_1h, this0l, this0h, this1l, this1h,
lastcolsum_1h, lastcolsum1h,
p_lastcolsum_1l, p_lastcolsum_1h, p_lastcolsum1l, p_lastcolsum1h,
thiscolsum_1l, thiscolsum_1h, thiscolsum1l, thiscolsum1h,
nextcolsum_1l = { 0 }, nextcolsum_1h = { 0 },
nextcolsum1l = { 0 }, nextcolsum1h = { 0 },
p_nextcolsum_1l, p_nextcolsum_1h, p_nextcolsum1l, p_nextcolsum1h,
tmpl, tmph, outle, outhe, outlo, outho;
/* Constants */
__vector unsigned char pb_zero = { __16X(0) },
last_index_col0 =
{ 0, 1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 },
last_index =
{ 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 },
next_index =
{ 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17 },
next_index_lastcol =
{ 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 14, 15 },
#ifdef __BIG_ENDIAN__
merge_pack_index =
{ 1, 17, 3, 19, 5, 21, 7, 23, 9, 25, 11, 27, 13, 29, 15, 31 };
#else
merge_pack_index =
{ 0, 16, 2, 18, 4, 20, 6, 22, 8, 24, 10, 26, 12, 28, 14, 30 };
#endif
__vector short pw_zero = { __8X(0) }, pw_three = { __8X(3) },
pw_seven = { __8X(7) }, pw_eight = { __8X(8) };
__vector unsigned short pw_four = { __8X(4) };
for (inrow = 0, outrow = 0; outrow < max_v_samp_factor; inrow++) {
inptr_1 = input_data[inrow - 1];
inptr0 = input_data[inrow];
inptr1 = input_data[inrow + 1];
outptr0 = output_data[outrow++];
outptr1 = output_data[outrow++];
if (downsampled_width & 15) {
inptr_1[downsampled_width] = inptr_1[downsampled_width - 1];
inptr0[downsampled_width] = inptr0[downsampled_width - 1];
inptr1[downsampled_width] = inptr1[downsampled_width - 1];
}
this0 = vec_ld(0, inptr0);
this0l = (__vector short)VEC_UNPACKHU(this0);
this0h = (__vector short)VEC_UNPACKLU(this0);
this0l = vec_mladd(this0l, pw_three, pw_zero);
this0h = vec_mladd(this0h, pw_three, pw_zero);
this_1 = vec_ld(0, inptr_1);
this_1l = (__vector short)VEC_UNPACKHU(this_1);
this_1h = (__vector short)VEC_UNPACKLU(this_1);
thiscolsum_1l = vec_add(this0l, this_1l);
thiscolsum_1h = vec_add(this0h, this_1h);
lastcolsum_1h = thiscolsum_1h;
p_lastcolsum_1l = vec_perm(thiscolsum_1l, thiscolsum_1l, last_index_col0);
p_lastcolsum_1h = vec_perm(thiscolsum_1l, thiscolsum_1h, last_index);
this1 = vec_ld(0, inptr1);
this1l = (__vector short)VEC_UNPACKHU(this1);
this1h = (__vector short)VEC_UNPACKLU(this1);
thiscolsum1l = vec_add(this0l, this1l);
thiscolsum1h = vec_add(this0h, this1h);
lastcolsum1h = thiscolsum1h;
p_lastcolsum1l = vec_perm(thiscolsum1l, thiscolsum1l, last_index_col0);
p_lastcolsum1h = vec_perm(thiscolsum1l, thiscolsum1h, last_index);
for (incol = downsampled_width; incol > 0;
incol -= 16, inptr_1 += 16, inptr0 += 16, inptr1 += 16,
outptr0 += 32, outptr1 += 32) {
if (downsampled_width - incol > 0) {
p_lastcolsum_1l = vec_perm(lastcolsum_1h, thiscolsum_1l, last_index);
p_lastcolsum_1h = vec_perm(thiscolsum_1l, thiscolsum_1h, last_index);
p_lastcolsum1l = vec_perm(lastcolsum1h, thiscolsum1l, last_index);
p_lastcolsum1h = vec_perm(thiscolsum1l, thiscolsum1h, last_index);
lastcolsum_1h = thiscolsum_1h; lastcolsum1h = thiscolsum1h;
}
if (incol <= 16) {
p_nextcolsum_1l = vec_perm(thiscolsum_1l, thiscolsum_1h, next_index);
p_nextcolsum_1h = vec_perm(thiscolsum_1h, thiscolsum_1h,
next_index_lastcol);
p_nextcolsum1l = vec_perm(thiscolsum1l, thiscolsum1h, next_index);
p_nextcolsum1h = vec_perm(thiscolsum1h, thiscolsum1h,
next_index_lastcol);
} else {
this0 = vec_ld(16, inptr0);
this0l = (__vector short)VEC_UNPACKHU(this0);
this0h = (__vector short)VEC_UNPACKLU(this0);
this0l = vec_mladd(this0l, pw_three, pw_zero);
this0h = vec_mladd(this0h, pw_three, pw_zero);
this_1 = vec_ld(16, inptr_1);
this_1l = (__vector short)VEC_UNPACKHU(this_1);
this_1h = (__vector short)VEC_UNPACKLU(this_1);
nextcolsum_1l = vec_add(this0l, this_1l);
nextcolsum_1h = vec_add(this0h, this_1h);
p_nextcolsum_1l = vec_perm(thiscolsum_1l, thiscolsum_1h, next_index);
p_nextcolsum_1h = vec_perm(thiscolsum_1h, nextcolsum_1l, next_index);
this1 = vec_ld(16, inptr1);
this1l = (__vector short)VEC_UNPACKHU(this1);
this1h = (__vector short)VEC_UNPACKLU(this1);
nextcolsum1l = vec_add(this0l, this1l);
nextcolsum1h = vec_add(this0h, this1h);
p_nextcolsum1l = vec_perm(thiscolsum1l, thiscolsum1h, next_index);
p_nextcolsum1h = vec_perm(thiscolsum1h, nextcolsum1l, next_index);
}
/* Process the upper row */
tmpl = vec_mladd(thiscolsum_1l, pw_three, pw_zero);
outle = vec_add(tmpl, p_lastcolsum_1l);
outle = vec_add(outle, pw_eight);
outle = vec_sr(outle, pw_four);
outlo = vec_add(tmpl, p_nextcolsum_1l);
outlo = vec_add(outlo, pw_seven);
outlo = vec_sr(outlo, pw_four);
out = vec_perm((__vector unsigned char)outle,
(__vector unsigned char)outlo, merge_pack_index);
vec_st(out, 0, outptr0);
if (incol > 8) {
tmph = vec_mladd(thiscolsum_1h, pw_three, pw_zero);
outhe = vec_add(tmph, p_lastcolsum_1h);
outhe = vec_add(outhe, pw_eight);
outhe = vec_sr(outhe, pw_four);
outho = vec_add(tmph, p_nextcolsum_1h);
outho = vec_add(outho, pw_seven);
outho = vec_sr(outho, pw_four);
out = vec_perm((__vector unsigned char)outhe,
(__vector unsigned char)outho, merge_pack_index);
vec_st(out, 16, outptr0);
}
/* Process the lower row */
tmpl = vec_mladd(thiscolsum1l, pw_three, pw_zero);
outle = vec_add(tmpl, p_lastcolsum1l);
outle = vec_add(outle, pw_eight);
outle = vec_sr(outle, pw_four);
outlo = vec_add(tmpl, p_nextcolsum1l);
outlo = vec_add(outlo, pw_seven);
outlo = vec_sr(outlo, pw_four);
out = vec_perm((__vector unsigned char)outle,
(__vector unsigned char)outlo, merge_pack_index);
vec_st(out, 0, outptr1);
if (incol > 8) {
tmph = vec_mladd(thiscolsum1h, pw_three, pw_zero);
outhe = vec_add(tmph, p_lastcolsum1h);
outhe = vec_add(outhe, pw_eight);
outhe = vec_sr(outhe, pw_four);
outho = vec_add(tmph, p_nextcolsum1h);
outho = vec_add(outho, pw_seven);
outho = vec_sr(outho, pw_four);
out = vec_perm((__vector unsigned char)outhe,
(__vector unsigned char)outho, merge_pack_index);
vec_st(out, 16, outptr1);
}
thiscolsum_1l = nextcolsum_1l; thiscolsum_1h = nextcolsum_1h;
thiscolsum1l = nextcolsum1l; thiscolsum1h = nextcolsum1h;
}
}
}