void scale_frame_down2x2_simd(yuv_frame_t* sin, yuv_frame_t* sout)
{
int wo=sout->width;
int ho=sout->height;
int so=sout->stride_y;
int si=sin->stride_y;
int i, j;
v128 ones = v128_dup_8(1);
v128 z = v128_dup_8(0);
for (i=0; i<ho; ++i) {
for (j=0; j<=wo-8; j+=8) {
v128 a = v128_load_aligned(&sin->y[(2*i+0)*si+2*j]);
v128 b = v128_load_aligned(&sin->y[(2*i+1)*si+2*j]);
v128 c = v128_avg_u8(a,b);
v128 d = v128_shr_s16(v128_madd_us8(c,ones),1);
v64_store_aligned(&sout->y[i*so+j], v128_low_v64(v128_pack_s16_u8(z,d)));
}
for (; j<wo; ++j) {
sout->y[i*so+j]=( ((sin->y[(2*i+0)*si+(2*j+0)] + sin->y[(2*i+1)*si+(2*j+0)]+1)>>1)+
+ ((sin->y[(2*i+0)*si+(2*j+1)] + sin->y[(2*i+1)*si+(2*j+1)]+1)>>1) )>>1;
}
}
#if USE_CHROMA
int soc=sout->stride_c;
int sic=sin->stride_c;
ho /= 2;
wo /= 2;
for (int i=0; i<ho; ++i) {
for (j=0; j<=wo-8; j+=8) {
v128 a = v128_load_aligned(&sin->u[(2*i+0)*sic+2*j]);
v128 b = v128_load_aligned(&sin->u[(2*i+1)*sic+2*j]);
v128 c = v128_avg_u8(a,b);
v128 d = v128_shr_s16(v128_madd_us8(c,ones),1);
v64_store_aligned(&sout->u[i*soc+j], v128_low_v64(v128_pack_s16_u8(z,d)));
}
for (; j<wo; ++j) {
sout->u[i*soc+j]=( ((sin->u[(2*i+0)*sic+(2*j+0)] + sin->u[(2*i+1)*sic+(2*j+0)]+1)>>1)+
+ ((sin->u[(2*i+0)*sic+(2*j+1)] + sin->u[(2*i+1)*sic+(2*j+1)]+1)>>1) )>>1;
}
for (j=0; j<=wo-8; j+=8) {
v128 a = v128_load_aligned(&sin->v[(2*i+0)*sic+2*j]);
v128 b = v128_load_aligned(&sin->v[(2*i+1)*sic+2*j]);
v128 c = v128_avg_u8(a,b);
v128 d = v128_shr_s16(v128_madd_us8(c,ones),1);
v64_store_aligned(&sout->v[i*soc+j], v128_low_v64(v128_pack_s16_u8(z,d)));
}
for (; j<wo; ++j) {
sout->v[i*soc+j]=( ((sin->v[(2*i+0)*sic+(2*j+0)] + sin->v[(2*i+1)*sic+(2*j+0)]+1)>>1)+
+ ((sin->v[(2*i+0)*sic+(2*j+1)] + sin->v[(2*i+1)*sic+(2*j+1)]+1)>>1) )>>1;
}
}
#endif
pad_yuv_frame(sout);
}
int ssd_calc_simd(uint8_t *a, uint8_t *b, int astride, int bstride, int size)
{
int i, j;
if (size == 8) {
ssd64_internal s = v64_ssd_u8_init();
s = v64_ssd_u8(s, v64_load_aligned(a + 0*astride), v64_load_aligned(b + 0*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 1*astride), v64_load_aligned(b + 1*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 2*astride), v64_load_aligned(b + 2*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 3*astride), v64_load_aligned(b + 3*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 4*astride), v64_load_aligned(b + 4*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 5*astride), v64_load_aligned(b + 5*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 6*astride), v64_load_aligned(b + 6*bstride));
s = v64_ssd_u8(s, v64_load_aligned(a + 7*astride), v64_load_aligned(b + 7*bstride));
return v64_ssd_u8_sum(s);
} else {
ssd128_internal s = v128_ssd_u8_init();
for (i = 0; i < size; i++)
for (j = 0; j < size; j += 16)
s = v128_ssd_u8(s, v128_load_aligned(a + i*astride + j), v128_load_aligned(b + i*bstride + j));
return v128_ssd_u8_sum(s);
}
}
int sad_calc_simd(uint8_t *a, uint8_t *b, int astride, int bstride, int width, int height)
{
int i, j;
if (width == 8) {
sad64_internal s = v64_sad_u8_init();
for (i = 0; i < height; i += 4) {
s = v64_sad_u8(s, v64_load_aligned(a + 0*astride), v64_load_aligned(b + 0*bstride));
s = v64_sad_u8(s, v64_load_aligned(a + 1*astride), v64_load_aligned(b + 1*bstride));
s = v64_sad_u8(s, v64_load_aligned(a + 2*astride), v64_load_aligned(b + 2*bstride));
s = v64_sad_u8(s, v64_load_aligned(a + 3*astride), v64_load_aligned(b + 3*bstride));
a += 4*astride;
b += 4*bstride;
}
return v64_sad_u8_sum(s);
} else {
sad128_internal s = v128_sad_u8_init();
for (i = 0; i < height; i++)
for (j = 0; j < width; j += 16)
s = v128_sad_u8(s, v128_load_aligned(a + i*astride + j), v128_load_aligned(b + i*bstride + j));
return v128_sad_u8_sum(s);
}
}
