void luma_sample_interp_8x8_3_3_SSE2(unsigned char* sortie, unsigned char* image,const short PicWidthInPix,const short OutStride){
unsigned char* tmpimg = image + 1;
unsigned char* tmpimg2 = image + PicWidthInPix;
__m64 *ptr_out = (__m64 *)sortie;
int i;
for(i = 0; i < 8 * PicWidthInPix; i += PicWidthInPix, ptr_out += (OutStride >> 3))
{
__m128i verthalf = _mm_movpi64_epi64(interpolvline_2(tmpimg + i,PicWidthInPix));
__m128i horizhalf = _mm_movpi64_epi64(interpolhline_2(tmpimg2 + i));
*ptr_out = _mm_movepi64_pi64(_mm_avg_epu8(verthalf,horizhalf));
}
_mm_empty();
}
void luma_sample_interp_8x8_3_2_SSE2(unsigned char* sortie, unsigned char* image,const short PicWidthInPix,const short OutStride){
__m64 *ptr_out = (__m64 *)sortie;
unsigned char* tmpimg = image + 1;
__m128i temp[13];
int i;
for(i = 0; i < 13;i++) {
_mm_store_si128(temp + i, interpolhline128(image + PicWidthInPix * (i - 2)));
}
for(i = 0; i < 8;i++, ptr_out += (OutStride >> 3))
{
__m128i verthalf = _mm_movpi64_epi64(interpolvline_2(tmpimg + i*PicWidthInPix,PicWidthInPix));
__m128i centerpix = _mm_movpi64_epi64(interpolvline128_2(temp+i));
*ptr_out = _mm_movepi64_pi64(_mm_avg_epu8(verthalf,centerpix));
}
_mm_empty();
}
__m64 interpolvline_1( unsigned char* image, int PicWidthInPix){
__m128i xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7;
__m64 ret;
xmm7 = _mm_setzero_si128();
xmm0 = _mm_movpi64_epi64(*((__m64*)(image - 2*PicWidthInPix)));
xmm0 = _mm_unpacklo_epi8(xmm0,xmm7);
xmm1 = _mm_movpi64_epi64(*((__m64*)(image - 1*PicWidthInPix)));
xmm1 = _mm_unpacklo_epi8(xmm1,xmm7);
xmm2 = _mm_movpi64_epi64(*((__m64*)(image - 0*PicWidthInPix)));
xmm2 = _mm_unpacklo_epi8(xmm2,xmm7);
xmm3 = _mm_movpi64_epi64(*((__m64*)(image + 1*PicWidthInPix)));
xmm3 = _mm_unpacklo_epi8(xmm3,xmm7);
xmm4 = _mm_movpi64_epi64(*((__m64*)(image + 2*PicWidthInPix)));
xmm4 = _mm_unpacklo_epi8(xmm4,xmm7);
xmm5 = _mm_movpi64_epi64(*((__m64*)(image + 3*PicWidthInPix)));
xmm5 = _mm_unpacklo_epi8(xmm5,xmm7);
// filter on 8 values
xmm6 = _mm_add_epi16(xmm2,xmm3);
xmm6 = _mm_slli_epi16(xmm6,2);
xmm6 = _mm_sub_epi16(xmm6,xmm1);
xmm6 = _mm_sub_epi16(xmm6,xmm4);
xmm1 = _mm_set_epi32(0x00050005,0x00050005,0x00050005,0x00050005);
xmm6 = _mm_mullo_epi16(xmm6,xmm1);
xmm6 = _mm_add_epi16(xmm6,xmm0);
xmm6 = _mm_add_epi16(xmm6,xmm5);
xmm6 = _mm_add_epi16(xmm6,_mm_set_epi32(0x00100010,0x00100010,0x00100010,0x00100010));
xmm6 = _mm_max_epi16(xmm6, xmm7); // preventing negative values
xmm6 = _mm_srli_epi16(xmm6,5);
xmm2 = _mm_packus_epi16(xmm2,xmm7);
xmm3 = _mm_packus_epi16(xmm3,xmm7);
xmm6 = _mm_packus_epi16(xmm6,xmm7);
xmm5 = _mm_unpacklo_epi8(xmm2,xmm6);
xmm4 = _mm_unpacklo_epi8(xmm6,xmm3);
xmm6 = _mm_avg_epu8(xmm4,xmm5);
xmm6 = _mm_slli_epi16(xmm6,8);
xmm6 = _mm_srli_epi16(xmm6,8);
xmm6 = _mm_packus_epi16(xmm6,xmm7);
ret = _mm_movepi64_pi64(xmm6);
_mm_empty();
return(ret);
}
void lines_scale2(const unsigned char *src, unsigned y, unsigned char *dst1, unsigned char *dst2, unsigned nPix)
{
const unsigned char
*u = src + ((y-1) & 7)*sc2lines_width,
*m = src + ((y+0) & 7)*sc2lines_width,
*l = src + ((y+1) & 7)*sc2lines_width;
for (unsigned i = 0; i < nPix; i += 8) {
__m64 uu = *(__m64*)(u+i);
__m64 ll = *(__m64*)(l+i);
__m64 cmp = _mm_cmpeq_pi8(uu,ll);
if (_mm_movemask_pi8(cmp) != 0xFF) {
__m128i mm = _mm_loadu_si128((__m128i*)(m+i-4));
__m128i uu = _mm_loadu_si128((__m128i*)(u+i-4));
__m128i ll = _mm_loadu_si128((__m128i*)(l+i-4));
__m128i md = _mm_slli_si128(mm,1);
__m128i mf = _mm_srli_si128(mm,1);
__m128i maskall = _mm_or_si128(_mm_cmpeq_epi8(md,mf), _mm_cmpeq_epi8(uu,ll));
__m128i e0, e1, v1, v2, v3;
e0 = _mm_cmpeq_epi8(md,uu);
e0 = _mm_andnot_si128(maskall, e0);
e0 = _mm_srli_si128(e0,4);
e0 = _mm_unpacklo_epi8(e0, _mm_setzero_si128());
e1 = _mm_cmpeq_epi8(mf,uu);
e1 = _mm_andnot_si128(maskall, e1);
e1 = _mm_srli_si128(e1,4);
e1 = _mm_unpacklo_epi8(_mm_setzero_si128(), e1);
e0 = _mm_or_si128(e0, e1);
v1 = _mm_srli_si128(mm,4);
v1 = _mm_unpacklo_epi8(v1,v1);
v2 = _mm_srli_si128(uu,4);
v2 = _mm_unpacklo_epi8(v2,v2);
_mm_store_si128((__m128i*)(dst1 + 2*i), _mm_or_si128( _mm_and_si128(e0,v2), _mm_andnot_si128(e0,v1) ) );
e0 = _mm_cmpeq_epi8(md,ll);
e0 = _mm_andnot_si128(maskall, e0);
e0 = _mm_srli_si128(e0,4);
e0 = _mm_unpacklo_epi8(e0, _mm_setzero_si128());
e1 = _mm_cmpeq_epi8(mf,ll);
e1 = _mm_andnot_si128(maskall, e1);
e1 = _mm_srli_si128(e1,4);
e1 = _mm_unpacklo_epi8(_mm_setzero_si128(), e1);
e0 = _mm_or_si128(e0, e1);
v3 = _mm_srli_si128(ll,4);
v3 = _mm_unpacklo_epi8(v3,v3);
_mm_store_si128((__m128i*)(dst2 + 2*i), _mm_or_si128( _mm_and_si128(e0,v3), _mm_andnot_si128(e0,v1) ) );
} else {
__m64 v0 = *(__m64*)(m+i);
__m128i v1 = _mm_movpi64_epi64(v0);
v1 = _mm_unpacklo_epi8(v1,v1);
_mm_store_si128((__m128i*)(dst1 + 2*i), v1);
_mm_store_si128((__m128i*)(dst2 + 2*i), v1);
}
}
}
