Merge remote-tracking branch 'origin/perf_improv_polar_encoder' into integration_2024_w14

common/utils/nr/nr_common.c

@@ -32,6 +32,7 @@
 
 #include <stdint.h>
 #include "assertions.h"
+#include "common/utils/assertions.h"
 #include "nr_common.h"
 #include <complex.h>
 
@@ -52,6 +53,15 @@ static const uint8_t bit_reverse_table_256[] = {
     0x3B, 0xBB, 0x7B, 0xFB, 0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7, 0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7,
     0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF, 0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF};
 
+void reverse_bits_u8(uint8_t const* in, size_t sz, uint8_t* out)
+{
+  DevAssert(in != NULL);
+  DevAssert(out != NULL);
+
+  for(size_t i = 0; i < sz; ++i)
+    out[i] = bit_reverse_table_256[in[i]];
+}
+
 // Reverse bits implementation based on http://graphics.stanford.edu/~seander/bithacks.html
 uint64_t reverse_bits(uint64_t in, int n_bits)
 {

common/utils/nr/nr_common.h

@@ -34,6 +34,7 @@
 #define __COMMON_UTILS_NR_NR_COMMON__H__
 
 #include <stdint.h>
+#include <stdlib.h>
 #include "assertions.h"
 #include "PHY/defs_common.h"
 
@@ -175,8 +176,11 @@ static inline int get_num_dmrs(uint16_t dmrs_mask ) {
 }
 
 uint64_t reverse_bits(uint64_t in, int n_bits);
+void reverse_bits_u8(uint8_t const* in, size_t sz, uint8_t* out);
+
 uint64_t from_nrarfcn(int nr_bandP, uint8_t scs_index, uint32_t dl_nrarfcn);
 uint32_t to_nrarfcn(int nr_bandP, uint64_t dl_CarrierFreq, uint8_t scs_index, uint32_t bw);
+
 int get_first_ul_slot(int nrofDownlinkSlots, int nrofDownlinkSymbols, int nrofUplinkSymbols);
 int cce_to_reg_interleaving(const int R, int k, int n_shift, const int C, int L, const int N_regs);
 int get_SLIV(uint8_t S, uint8_t L);

openair1/PHY/CODING/nrPolar_tools/nr_polar_defs.h

@@ -282,7 +282,7 @@ void nr_polar_generate_u(uint64_t *u,
                          uint16_t N,
                          uint8_t n_pc);
 
-void nr_polar_uxG(uint64_t *D, const uint64_t *u, const fourDimArray_t *G_N_tab, uint16_t N);
+void nr_polar_uxG(uint8_t const* u, size_t N, uint8_t* D);
 
 void nr_polar_info_extraction_from_u(uint64_t *Cprime,
                                      const uint8_t *u,

openair1/PHY/CODING/nrPolar_tools/nr_polar_encoder.c

@@ -485,12 +485,13 @@ void build_polar_tables(t_nrPolar_params *polarParams) {
     polarParams->rm_tab[tcnt] = polarParams->rate_matching_pattern[outpos] >> shift;
 }
 
+
 void polar_encoder_fast(uint64_t *A,
                         void *out,
-                        int32_t crcmask,
-                        uint8_t ones_flag,
-                        int8_t messageType,
-                        uint16_t messageLength,
+                        int32_t crcmask, 
+                        uint8_t ones_flag,  
+                        int8_t messageType, 
+                        uint16_t messageLength, 
                         uint8_t aggregation_level) {
                         
   t_nrPolar_params *polarParams=nr_polar_params(messageType, messageLength, aggregation_level, false);
@@ -687,8 +688,8 @@ void polar_encoder_fast(uint64_t *A,
   printf("\n");
 #endif
 
-  uint64_t D[8] = {0};
-  nr_polar_uxG(D, u, polarParams->G_N_tab, polarParams->N);
+  uint64_t D[8];
+  nr_polar_uxG((uint8_t *)u, polarParams->N, (uint8_t *)D);
 
 #ifdef POLAR_CODING_DEBUG
   printf("d: ");
@@ -701,6 +702,7 @@ void polar_encoder_fast(uint64_t *A,
     printf("%lu", (D[n1] >> n2) & 1);
   }
   printf("\n");
+  fflush(stdout);
 #endif
 
   memset((void*)out,0,polarParams->encoderLength>>3);

openair1/PHY/CODING/nrPolar_tools/nr_polar_procedures.c

@@ -30,8 +30,12 @@
  * \warning
  */
 
+#include "common/utils/nr/nr_common.h"
 #include "PHY/CODING/nrPolar_tools/nr_polar_defs.h"
 
+#include <assert.h>
+#include <stdlib.h>
+#include <stdint.h>
 
 // TS 38.212 - Section 5.3.1.2 Polar encoding
 void nr_polar_generate_u(uint64_t *u,
@@ -119,23 +123,71 @@ void nr_polar_info_extraction_from_u(uint64_t *Cprime,
   }
 }
 
-void nr_polar_uxG(uint64_t *D, const uint64_t *u, const fourDimArray_t *G_N_tab, uint16_t N)
+
+static
+void encode_packed_byte(uint8_t* in_out)
+{
+  //    *in_out ^= 0xAA & (*in_out << 1);
+  //    *in_out ^= 0xCC & (*in_out << 2);
+  //    *in_out ^= *in_out  in_out << 4;
+  //
+  *in_out ^= 0x55 & (*in_out >> 1);
+  *in_out ^= 0x33 & (*in_out >> 2);
+  *in_out ^= *in_out >> 4;
+}
+
+static
+void polar_encode_bits(uint8_t* in_out, size_t N)
+{
+  size_t const num_bytes_per_block = N >> 3;
+  for(size_t i = 0; i < num_bytes_per_block; ++i){
+    encode_packed_byte(&in_out[i]);
+  }
+}
+
+static 
+uint32_t log2_floor(uint32_t x) 
 {
-  const int N64 = N / 64;
-  cast2Darray(g_n, uint64_t, G_N_tab);
-  for (int n = 0; n < N; n++) {
-    const uint64_t *Gn = g_n[N - 1 - n];
-
-    int n_ones = 0;
-    for (int a = 0; a < N64; a++)
-      n_ones += count_bits_set(u[a] & Gn[a]);
-
-    int n1 = n / 64;
-    int n2 = n - (n1 * 64);
-    D[n1] |= ((uint64_t)n_ones & 1) << n2;
+  if(x == 0)
+    return 0; 
+  uint32_t clz = __builtin_clz(x);
+  return 31U - clz;
+}
+
+static
+void polar_encode_bytes(uint8_t* in_out, size_t N)
+{
+  size_t brnch_sz = 1;
+  size_t n_brnch = N >> 4;
+  size_t const blck_pwr = log2_floor(N); 
+  for (size_t stage = 3; stage < blck_pwr; ++stage) {
+    for (size_t brnch = 0; brnch < n_brnch; ++brnch) {
+        for(size_t byte = 0; byte < brnch_sz; ++byte){
+          size_t const dst = 2*brnch_sz*brnch + byte;
+          in_out[dst] ^= in_out[dst + brnch_sz];
+        }
+    }
+    n_brnch >>= 1;
+    brnch_sz <<= 1;
   }
 }
 
+void nr_polar_uxG(uint8_t const *u, size_t N, uint8_t *D2)
+{
+  assert(N > 7); 
+  uint8_t tmp[N/8];
+  for(int i = 0; i < N/8; ++i)
+    tmp[i] = u[N/8 - 1 - i];
+
+  reverse_bits_u8(tmp, N/8, D2);
+
+  // Do the encoding/xor for the bottom 3 levels of the tree.
+  // Thus, data remaining to encode, is in 2^3 type.
+  polar_encode_bits(D2, N);
+  // Xor the remaining tree levels. Use bytes for it
+  polar_encode_bytes(D2, N);
+}
+
 void nr_polar_bit_insertion(uint8_t *input,
 			    uint8_t *output,
 			    uint16_t N,