Commit 70b838ff authored by Elena Lukashova's avatar Elena Lukashova

Fixing a bug in dlsch_channel_level:

instead of _mm_srai_epi16
_mm_srai_epi32 should be used as _mm_madd_epi16
packs multiply-adds 16 bits and packs into 4
blocks of 32 bits.
parent 759c1342
...@@ -3738,8 +3738,8 @@ void dlsch_channel_level(int **dl_ch_estimates_ext, ...@@ -3738,8 +3738,8 @@ void dlsch_channel_level(int **dl_ch_estimates_ext,
for (rb=0;rb<nb_rb;rb++) { for (rb=0;rb<nb_rb;rb++) {
//printf("rb %d : ",rb); //printf("rb %d : ",rb);
avg128D = _mm_add_epi32(avg128D,_mm_srai_epi16(_mm_madd_epi16(dl_ch128[0],dl_ch128[0]),x)); avg128D = _mm_add_epi32(avg128D,_mm_srai_epi32(_mm_madd_epi16(dl_ch128[0],dl_ch128[0]),x));
avg128D = _mm_add_epi32(avg128D,_mm_srai_epi16(_mm_madd_epi16(dl_ch128[1],dl_ch128[1]),x)); avg128D = _mm_add_epi32(avg128D,_mm_srai_epi32(_mm_madd_epi16(dl_ch128[1],dl_ch128[1]),x));
//avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[0],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[0], coeff128),15))); //avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[0],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[0], coeff128),15)));
//avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[1],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[1], coeff128),15))); //avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[1],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[1], coeff128),15)));
...@@ -3748,7 +3748,7 @@ void dlsch_channel_level(int **dl_ch_estimates_ext, ...@@ -3748,7 +3748,7 @@ void dlsch_channel_level(int **dl_ch_estimates_ext,
dl_ch128+=2; dl_ch128+=2;
} }
else { else {
avg128D = _mm_add_epi32(avg128D,_mm_srai_epi16(_mm_madd_epi16(dl_ch128[2],dl_ch128[2]),x)); avg128D = _mm_add_epi32(avg128D,_mm_srai_epi32(_mm_madd_epi16(dl_ch128[2],dl_ch128[2]),x));
//avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[2],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[2], coeff128),15))); //avg128D = _mm_add_epi32(avg128D,_mm_madd_epi16(dl_ch128[2],_mm_srai_epi16(_mm_mulhi_epi16(dl_ch128[2], coeff128),15)));
dl_ch128+=3; dl_ch128+=3;
} }
...@@ -3863,8 +3863,8 @@ void dlsch_channel_level_core(int **dl_ch_estimates_ext, ...@@ -3863,8 +3863,8 @@ void dlsch_channel_level_core(int **dl_ch_estimates_ext,
length2 = length>>3; length2 = length>>3;
for (ii=0;ii<length2;ii++) { for (ii=0;ii<length2;ii++) {
avg128D = _mm_add_epi32(avg128D,_mm_srai_epi16(_mm_madd_epi16(dl_ch128[0],dl_ch128[0]),x)); avg128D = _mm_add_epi32(avg128D,_mm_srai_epi32(_mm_madd_epi16(dl_ch128[0],dl_ch128[0]),x));
avg128D = _mm_add_epi32(avg128D,_mm_srai_epi16(_mm_madd_epi16(dl_ch128[1],dl_ch128[1]),x)); avg128D = _mm_add_epi32(avg128D,_mm_srai_epi32(_mm_madd_epi16(dl_ch128[1],dl_ch128[1]),x));
dl_ch128+=2; dl_ch128+=2;
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment