fix / 2-layer inner_rx

  align the buffer to 256-bit

openair1/PHY/NR_TRANSPORT/nr_ulsch_demodulation.c

@@ -1347,8 +1347,12 @@ void inner_rx_qpsk_2layer (NR_DL_FRAME_PARMS *frame_parms,
                            int dmrs_symbol_flag,
                            int output_shift)
 {
-  int32_t rxFext[nb_rx_ant][length + 4] __attribute__((aligned(32)));
-  int32_t chFext[nb_layer*nb_rx_ant][length + 4] __attribute__((aligned(32)));
+  int add_shift = 0;
+  if (length % 8)
+    add_shift = 8 - length % 8;
+
+  int32_t rxFext[nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
+  int32_t chFext[nb_layer*nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
   for (int aarx = 0; aarx < nb_rx_ant; aarx++) 
   {
     for (int aatx = 0; aatx < nb_layer; aatx++) 
@@ -1365,8 +1369,8 @@ void inner_rx_qpsk_2layer (NR_DL_FRAME_PARMS *frame_parms,
                             frame_parms);
     }
   }
-  int32_t rho[nb_layer*nb_layer][length + 4] __attribute__((aligned(32)));
-  int32_t rxFext_comp[nb_layer][length + 4] __attribute__((aligned(32)));
+  int32_t rho[nb_layer*nb_layer][length + add_shift] __attribute__((aligned(32)));
+  int32_t rxFext_comp[nb_layer][length + add_shift] __attribute__((aligned(32)));
   for (int aarx = 0; aarx < nb_rx_ant; aarx++) 
   {
     for (int aatx = 0; aatx < nb_layer; aatx++) 
@@ -1374,7 +1378,7 @@ void inner_rx_qpsk_2layer (NR_DL_FRAME_PARMS *frame_parms,
       for (int atx = 0; atx < nb_layer; atx++) 
       {
 #ifdef USE_128BIT
-        simde__m128i mmtmpD0, mmtmpD1, mmtmpD2, mmtmpD3;
+        simde__m128i mmtmpD0, mmtmpD1, mmtmpD2, mmtmpD3, mmtmpD4;
         simde__m128i *rho128        = (simde__m128i *)rho[aatx*nb_layer+atx];
         simde__m128i *ul_ch128      = (simde__m128i *)chFext[aatx * nb_rx_ant + aarx];
         simde__m128i *ul_ch128_2    = (simde__m128i *)chFext[atx * nb_rx_ant + aarx];
@@ -1392,10 +1396,11 @@ void inner_rx_qpsk_2layer (NR_DL_FRAME_PARMS *frame_parms,
           mmtmpD1 = simde_mm_srai_epi32(mmtmpD1, output_shift);
           mmtmpD2 = simde_mm_unpacklo_epi32(mmtmpD0, mmtmpD1);
           mmtmpD3 = simde_mm_unpackhi_epi32(mmtmpD0, mmtmpD1);
+          mmtmpD4 = simde_mm_packs_epi32(mmtmpD2, mmtmpD3);
           if (aarx == 0)
-            rho128[i] = simde_mm_packs_epi32(mmtmpD2, mmtmpD3);
+            rho128[i] = mmtmpD4;
           else
-            rho128[i] = simde_mm_adds_epi16(rho128[i], simde_mm_packs_epi32(mmtmpD2, mmtmpD3));
+            rho128[i] = simde_mm_adds_epi16(rho128[i], mmtmpD4);
         }
 #else
         simde__m256i mmtmpD0, mmtmpD1, mmtmpD2, mmtmpD3, mmtmpD4;
@@ -1513,8 +1518,12 @@ void inner_rx_16qam_2layer (NR_DL_FRAME_PARMS *frame_parms,
                             int dmrs_symbol_flag,
                             int output_shift)
 {
-  int32_t rxFext[nb_rx_ant][length + 4] __attribute__((aligned(32)));
-  int32_t chFext[nb_layer*nb_rx_ant][length + 4] __attribute__((aligned(32)));
+  int add_shift = 0;
+  if (length % 8)
+    add_shift = 8 - length % 8;
+  
+  int32_t rxFext[nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
+  int32_t chFext[nb_layer*nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
   for (int aarx = 0; aarx < nb_rx_ant; aarx++) 
   {
     for (int aatx = 0; aatx < nb_layer; aatx++) 
@@ -1532,9 +1541,9 @@ void inner_rx_16qam_2layer (NR_DL_FRAME_PARMS *frame_parms,
     }
   }
   
-  int32_t rho[nb_layer*nb_layer][length + 4] __attribute__((aligned(32)));
-  int32_t rxFext_comp[nb_layer][length + 4] __attribute__((aligned(32)));
-  int32_t ul_ch_mag[nb_layer][length + 4] __attribute__((aligned(32)));
+  int32_t rho[nb_layer*nb_layer][length + add_shift] __attribute__((aligned(32)));
+  int32_t rxFext_comp[nb_layer][length + add_shift] __attribute__((aligned(32)));
+  int32_t ul_ch_mag[nb_layer][length + add_shift] __attribute__((aligned(32)));
   for (int aatx = 0; aatx < nb_layer; aatx++) 
   {
     for (int aarx = 0; aarx < nb_rx_ant; aarx++) 
@@ -1705,8 +1714,12 @@ void inner_rx_64qam_2layer (NR_DL_FRAME_PARMS *frame_parms,
                             int dmrs_symbol_flag,
                             int output_shift)
 {
-  int32_t rxFext[nb_rx_ant][length + 4] __attribute__((aligned(32)));
-  int32_t chFext[nb_layer*nb_rx_ant][length + 4] __attribute__((aligned(32)));
+  int add_shift = 0;
+  if (length % 8)
+    add_shift = 8 - length % 8;
+  
+  int32_t rxFext[nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
+  int32_t chFext[nb_layer*nb_rx_ant][length + add_shift] __attribute__((aligned(32)));
   for (int aarx = 0; aarx < nb_rx_ant; aarx++) 
   {
     for (int aatx = 0; aatx < nb_layer; aatx++) 
@@ -1723,9 +1736,9 @@ void inner_rx_64qam_2layer (NR_DL_FRAME_PARMS *frame_parms,
                             frame_parms);
     }
   }
-  int32_t rho[nb_layer*nb_layer][length + 4] __attribute__((aligned(32)));
-  int32_t rxFext_comp[nb_layer][length + 4] __attribute__((aligned(32)));
-  int32_t ul_ch_mag[nb_layer][length + 4] __attribute__((aligned(32)));
+  int32_t rho[nb_layer*nb_layer][length + add_shift] __attribute__((aligned(32)));
+  int32_t rxFext_comp[nb_layer][length + add_shift] __attribute__((aligned(32)));
+  int32_t ul_ch_mag[nb_layer][length + add_shift] __attribute__((aligned(32)));
   for (int aatx = 0; aatx < nb_layer; aatx++) 
   {
     for (int aarx = 0; aarx < nb_rx_ant; aarx++) 

openair1/PHY/NR_TRANSPORT/nr_ulsch_llr_computation.c

@@ -3746,6 +3746,20 @@ void nr_ulsch_shift_llr(int16_t **llr_layers, uint32_t nb_re, uint32_t rxdataF_e
   simde__m128i *llr_layers0 = (simde__m128i *)&llr_layers[0][rxdataF_ext_offset];
   simde__m128i *llr_layers1 = (simde__m128i *)&llr_layers[1][rxdataF_ext_offset];
 
+  uint8_t mem_offset = ((16 - ((long)llr_layers0)) & 0xF) >> 2;
+
+  if (mem_offset > 0) {
+    c16_t *llr_layers0_c16 = (c16_t *)&llr_layers[0][rxdataF_ext_offset];
+    c16_t *llr_layers1_c16 = (c16_t *)&llr_layers[1][rxdataF_ext_offset];
+    for (int i = 0; i < mem_offset; i++) 
+    {
+      llr_layers0_c16[i] = c16Shift(llr_layers0_c16[i], shift);
+      llr_layers1_c16[i] = c16Shift(llr_layers1_c16[i], shift);
+    }
+    llr_layers0 = (simde__m128i *)&llr_layers[0][rxdataF_ext_offset + (mem_offset << 1)];
+    llr_layers1 = (simde__m128i *)&llr_layers[1][rxdataF_ext_offset + (mem_offset << 1)];
+  }
+
   for (int i = 0; i < nb_re >> 2; i++) {
     llr_layers0[i] = simde_mm_srai_epi16(llr_layers0[i], shift);
     llr_layers1[i] = simde_mm_srai_epi16(llr_layers1[i], shift);