Refactor PUSCH RE mapping

RE mapping time is reduced by 10x. Functions are made generic to be
later reused for PDSCH RE mapping at gNB.

openair1/PHY/NR_REFSIG/ptrs_nr.c

@@ -85,6 +85,13 @@ void set_ptrs_symb_idx(uint16_t *ptrs_symbols,
   }
 }
 
+unsigned int get_first_ptrs_re(const rnti_t rnti, const uint8_t K_ptrs, const uint16_t nRB, const uint8_t k_RE_ref)
+{
+  const uint16_t nRB_Kptrs = nRB % K_ptrs;
+  const uint16_t k_RB_ref = nRB_Kptrs ? (rnti % nRB_Kptrs) : (rnti % K_ptrs);
+  return (k_RE_ref + k_RB_ref * NR_NB_SC_PER_RB);
+}
+
 /*******************************************************************
 *
 * NAME :         is_ptrs_subcarrier

openair1/PHY/NR_REFSIG/ptrs_nr.h

@@ -53,6 +53,8 @@ void set_ptrs_symb_idx(uint16_t *ptrs_symbols,
                        uint8_t L_ptrs,
                        uint16_t dmrs_symb_pos);
 
+unsigned int get_first_ptrs_re(const rnti_t rnti, const uint8_t K_ptrs, const uint16_t nRB, const uint8_t k_RE_ref);
+
 uint8_t is_ptrs_subcarrier(uint16_t k,
                            uint16_t n_rnti,
                            uint8_t K_ptrs,

openair1/PHY/TOOLS/cmult_sv.c

@@ -24,6 +24,18 @@
 #include <simde/simde-common.h>
 #include <simde/x86/sse.h>
 
+void mult_real_vector_single_vector(const c16_t *x, const int16_t *alpha, c16_t *y, const unsigned int N)
+{
+  const simd_q15_t *alpha_128 = (const simd_q15_t *)alpha;
+  const simd_q15_t *x_128 = (const simd_q15_t *)x;
+  simd_q15_t *y_128 = (simd_q15_t *)y;
+  const unsigned int num_adds = (N + 3) / 4; // ceil(N/4)
+
+  for (uint_fast32_t n = 0; n < num_adds; n++) {
+    y_128[n] = mulhi_int16(x_128[n], *alpha_128);
+  }
+}
+
 void multadd_complex_vector_real_scalar(int16_t *x,
                                         int16_t alpha,
                                         int16_t *y,
@@ -35,19 +47,20 @@ void multadd_complex_vector_real_scalar(int16_t *x,
   int n;
 
   alpha_128 = set1_int16(alpha);
+  const uint32_t num_adds = (N + 3) / 4; // ceil(N/4)
 
-  if (zero_flag == 1)
-    for (n=0; n<N>>2; n++) {
-     // print_shorts("x_128[n]=", &x_128[n]);
-     // print_shorts("alpha_128", &alpha_128);
+  if (zero_flag == 1) {
+    for (n = 0; n < num_adds; n++) {
+      // print_shorts("x_128[n]=", &x_128[n]);
+      // print_shorts("alpha_128", &alpha_128);
       y_128[n] = mulhi_int16(x_128[n],alpha_128);
      // print_shorts("y_128[n]=", &y_128[n]);
     }
-
-  else
-    for (n=0; n<N>>2; n++) {
+  } else {
+    for (n = 0; n < num_adds; n++) {
       y_128[n] = adds_int16(y_128[n],mulhi_int16(x_128[n],alpha_128));
     }
+  }
 
   simde_mm_empty();
   simde_m_empty();

openair1/PHY/TOOLS/tools_defs.h

@@ -52,6 +52,9 @@
 extern "C" {
 #endif
 
+#define ALIGNARRAYSIZE(a, b) (((a + b - 1) / b) * b)
+#define ALNARS_16_4(a) ALIGNARRAYSIZE(a, 4)
+
   typedef struct complexd {
     double r;
     double i;
@@ -821,6 +824,8 @@ double interp(double x, double *xs, double *ys, int count);
 void simde_mm128_separate_real_imag_parts(simde__m128i *out_re, simde__m128i *out_im, simde__m128i in0, simde__m128i in1);
 void simde_mm256_separate_real_imag_parts(simde__m256i *out_re, simde__m256i *out_im, simde__m256i in0, simde__m256i in1);
 
+void mult_real_vector_single_vector(const c16_t *x, const int16_t *alpha, c16_t *y, const unsigned int N);
+
 #ifdef __cplusplus
 }
 #endif

openair1/SIMULATION/NR_PHY/ulsim.c

@@ -1421,17 +1421,6 @@ int main(int argc, char *argv[])
           errors_decoding++;
         }
       }
-      if (n_trials == 1) {
-        for (int r = 0; r < UE->ul_harq_processes[harq_pid].C; r++)
-          for (int i = 0; i < UE->ul_harq_processes[harq_pid].K >> 3; i++) {
-            if ((UE->ul_harq_processes[harq_pid].c[r][i] ^ ulsch_gNB->harq_process->c[r][i]) != 0)
-              printf("************");
-            /*printf("r %d: in[%d] %x, out[%d] %x (%x)\n",r,
-              i,UE->ul_harq_processes[harq_pid].c[r][i],
-              i,ulsch_gNB->harq_process->c[r][i],
-              UE->ul_harq_processes[harq_pid].c[r][i]^ulsch_gNB->harq_process->c[r][i]);*/
-          }
-      }
       if (errors_decoding > 0 && error_flag == 0) {
         n_false_positive++;
         if (n_trials==1)