fix after rebase with develop

individually

openair1/PHY/CODING/nrLDPC_defs.h

@@ -45,6 +45,8 @@ typedef struct {
   time_stats_t *tprep;
   time_stats_t *tparity;
   time_stats_t *toutput;
+  time_stats_t *dlsch_rate_matching_stats;
+  time_stats_t *dlsch_interleaving_stats;
   int Kr;
   uint32_t Kb;
   uint32_t Zc;

openair1/PHY/CODING/nr_rate_matching.c

@@ -41,43 +41,131 @@ void nr_interleaving_ldpc(uint32_t E, uint8_t Qm, uint8_t *e,uint8_t *f)
   memset(f,0,E*sizeof(uint8_t));
   uint8_t *e0,*e1,*e2,*e3,*e4,*e5,*e6,*e7;
   uint8_t *fp;
-#if 0 //def __WASAVX2__
-  simde__m256i tmp0,tmp1,tmp2,tmp0b,tmp1b,tmp3,tmp4,tmp5;
-  simde__m256i *e0_256,*e1_256,*e2_256,*e3_256,*e4_256,*e5_256,*e6_256,*e7_256;
+  int j,k;
+#if 1 //def __WASAVX2__
+#ifdef __AVX512F__
+  __m512i tmp0,tmp1,tmp2,tmp3,tmp4,tmp5;
+  __m512i *e0_512,*e1_512,*e2_512,*e3_512,*e4_512,*e5_512,*e6_512,*e7_512;
 
-  simde__m256i *f_256=(simde__m256i *)f;
+  __m512i *f_512=(__m512i *)f;
+#else
+  simde__m128i tmp0,tmp1,tmp2,tmp3,tmp4,tmp5,tmp6,tmp7;
+  simde__m128i *e0_128,*e1_128,*e2_128,*e3_128,*e4_128,*e5_128,*e6_128,*e7_128;
 
-  uint8_t *fp2;
+  simde__m128i *f_128=(simde__m128i *)f;
+#endif
   switch(Qm) {
   case 2:
     e0=e;
     e1=e0+EQm;
-    e0_256=(simde__m256i *)e0;
-    e1_256=(simde__m256i *)e1;
-    for (int k=0,j=0;j<EQm>>5;j++,k+=2) {
-      f_256[k]   = simde_mm256_unpacklo_epi8(e0_256[j],e1_256[j]);
-      f_256[k+1] = simde_mm256_unpackhi_epi8(e0_256[j],e1_256[j]); 
+#ifdef __AVX512F__
+    e0_512=(__m512i *)e0;
+    e1_512=(__m512i *)e1;
+    __m512i p8a = _mm512_set_epi8(95,31,94,30,93,29,92,28,91,27,90,26,89,25,88,24,87,23,86,22,85,21,84,20,83,19,82,18,81,17,80,16,79,15,78,14,77,13,76,12,75,11,74,10,73,9,72,8,71,7,70,6,69,5,68,4,67,3,66,2,65,1,64,0);
+    __m512i p8b = _mm512_set_epi8(127,63,126,62,125,61,124,60,123,59,122,58,121,57,120,56,119,55,118,54,117,53,116,52,115,51,114,50,113,49,112,48,111,47,110,46,109,45,108,44,107,43,106,42,105,41,104,40,103,39,102,38,101,37,100,36,99,35,98,34,97,33,96,32);
+
+    __m512i e0j,e1j;
+    for (k=0,j=0;j<EQm>>6;j++,k+=2) {
+      e0j    = _mm512_loadu_si512(e0_512+j);	    
+      e1j    = _mm512_loadu_si512(e1_512+j);	    
+      _mm512_storeu_si512(f_512+k,_mm512_permutex2var_epi8(e0j,p8a,e1j)); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
+      _mm512_storeu_si512(f_512+k+1,_mm512_permutex2var_epi8(e0j,p8b,e1j)); 
     }
+    if ((j<<6) != EQm) {
+      int k2=k<<6;
+      fp = &f[k2];
+      for (int j2=(j<<6) ; j2< EQm ; j2++) {
+        *fp++ = e0[j2];
+        *fp++ = e1[j2];
+      }
+    }
+#else
+    e0_128=(simde__m128i *)e0;
+    e1_128=(simde__m128i *)e1;
+    for (k=0,j=0;j<EQm>>4;j++,k+=2) {
+      f_128[k]   = simde_mm_unpacklo_epi8(e0_128[j],e1_128[j]);
+      f_128[k+1] = simde_mm_unpackhi_epi8(e0_128[j],e1_128[j]); 
+    }
+    if ((j<<4) != EQm) {
+      int k2=k<<4;
+      fp = &f[k2];
+      for (int j2=(j<<4) ; j2< EQm ; j2++) {
+        *fp++ = e0[j2];
+        *fp++ = e1[j2];
+      }
+    }
+#endif
     break;
   case 4:
     e0=e;
     e1=e0+EQm;
     e2=e1+EQm;
     e3=e2+EQm;
-    e0_256=(simde__m256i *)e0;
-    e1_256=(simde__m256i *)e1;
-    e2_256=(simde__m256i *)e2;
-    e3_256=(simde__m256i *)e3;
-    for (int k=0,j=0;j<EQm>>5;j++,k+=4) {
-      tmp0   = simde_mm256_unpacklo_epi8(e0_256[j],e1_256[j]); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
-      tmp1   = simde_mm256_unpacklo_epi8(e2_256[j],e3_256[j]); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
-      f_256[k]   = simde_mm256_unpacklo_epi8(tmp0,tmp1);   // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
-      f_256[k+1] = simde_mm256_unpackhi_epi8(tmp0,tmp1);   // e0(i+8) e1(i+8) e2(i+8) e3(i+8) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
-      tmp0   = simde_mm256_unpackhi_epi8(e0_256[j],e1_256[j]); // e0(i+16) e1(i+16) e0(i+17) e1(i+17) .... e0(i+31) e1(i+31)
-      tmp1   = simde_mm256_unpackhi_epi8(e2_256[j],e3_256[j]); // e2(i+16) e3(i+16) e2(i+17) e3(i+17) .... e2(i+31) e3(i+31)
-      f_256[k+2] = simde_mm256_unpacklo_epi8(tmp0,tmp1);
-      f_256[k+3] = simde_mm256_unpackhi_epi8(tmp0,tmp1); 
+#ifdef __AVX512F__
+    e0_512=(__m512i *)e0;
+    e1_512=(__m512i *)e1;
+    e2_512=(__m512i *)e2;
+    e3_512=(__m512i *)e3;
+    p8a = _mm512_set_epi8(95,31,94,30,93,29,92,28,91,27,90,26,89,25,88,24,87,23,86,22,85,21,84,20,83,19,82,18,81,17,80,16,79,15,78,14,77,13,76,12,75,11,74,10,73,9,72,8,71,7,70,6,69,5,68,4,67,3,66,2,65,1,64,0);
+    p8b = _mm512_set_epi8(127,63,126,62,125,61,124,60,123,59,122,58,121,57,120,56,119,55,118,54,117,53,116,52,115,51,114,50,113,49,112,48,111,47,110,46,109,45,108,44,107,43,106,42,105,41,104,40,103,39,102,38,101,37,100,36,99,35,98,34,97,33,96,32);
+
+    __m512i p16a = _mm512_set_epi16(47,15,46,14,45,13,44,12,43,11,42,10,41,9,40,8,39,7,38,6,37,5,36,4,35,3,34,2,33,1,32,0);
+    __m512i p16b = _mm512_set_epi16(63,31,62,30,61,29,60,28,59,27,58,26,57,25,56,24,55,23,54,22,53,21,52,20,51,19,50,18,49,17,48,16);
+
+    __m512i p32a = _mm512_set_epi32(23,7,22,6,21,5,20,4,19,3,18,2,17,1,16,0);
+    __m512i p32b = _mm512_set_epi32(31,15,30,14,29,13,28,12,27,11,26,10,25,9,24,8);
+    __m512i e2j,e3j;
+    for (k=0,j=0;j<EQm>>6;j++,k+=4) {
+      e0j    = _mm512_loadu_si512(e0_512+j);	    
+      e1j    = _mm512_loadu_si512(e1_512+j);	    
+      e2j    = _mm512_loadu_si512(e2_512+j);	    
+      e3j    = _mm512_loadu_si512(e3_512+j);	    
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8a,e1j); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8a,e3j); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
+      _mm512_storeu_si512(f_512+k,_mm512_permutex2var_epi16(tmp0,p16a,tmp1));  // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
+      _mm512_storeu_si512(f_512+k+1,_mm512_permutex2var_epi16(tmp0,p16b,tmp1)); 
+
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8b,e1j); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8b,e3j); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
+      _mm512_storeu_si512(f_512+k+2,_mm512_permutex2var_epi16(tmp0,p16a,tmp1));  // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
+      _mm512_storeu_si512(f_512+k+3,_mm512_permutex2var_epi16(tmp0,p16b,tmp1)); 
     }
+    if ((j<<6) != EQm) {
+      int k2=k<<6;
+      fp = &f[k2];
+      for (int j2=(j<<6) ; j2< EQm ; j2++) {
+        *fp++ = e0[j2];
+        *fp++ = e1[j2];
+        *fp++ = e2[j2];
+        *fp++ = e3[j2];
+      }
+    }
+#else
+    e0_128=(simde__m128i *)e0;
+    e1_128=(simde__m128i *)e1;
+    e2_128=(simde__m128i *)e2;
+    e3_128=(simde__m128i *)e3;
+    for (k=0,j=0;j<EQm>>4;j++,k+=4) {
+      tmp0   = simde_mm_unpacklo_epi8(e0_128[j],e1_128[j]); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+7) e1(i+7)
+      tmp1   = simde_mm_unpacklo_epi8(e2_128[j],e3_128[j]); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+7) e3(i+7)
+      f_128[k]   = simde_mm_unpacklo_epi16(tmp0,tmp1);   // e0(i) e1(i) e2(i) e3(i) ... e0(i+3) e1(i+3) e2(i+3) e3(i+3)
+      f_128[k+1] = simde_mm_unpackhi_epi16(tmp0,tmp1);   // e0(i+4) e1(i+4) e2(i+4) e3(i+4) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
+      tmp0   = simde_mm_unpackhi_epi8(e0_128[j],e1_128[j]); // e0(i+8) e1(i+8) e0(i+9) e1(i+9) .... e0(i+15) e1(i+15)
+      tmp1   = simde_mm_unpackhi_epi8(e2_128[j],e3_128[j]); // e2(i+8) e3(i+9) e2(i+10) e3(i+10) .... e2(i+31) e3(i+31)
+      f_128[k+2] = simde_mm_unpacklo_epi16(tmp0,tmp1);
+      f_128[k+3] = simde_mm_unpackhi_epi16(tmp0,tmp1); 
+    }
+    if ((j<<4) != EQm) {
+      int k2=k<<4;
+      fp = &f[k2];
+      for (int j2=(j<<4) ; j2< EQm ; j2++) {
+        *fp++ = e0[j2];
+        *fp++ = e1[j2];
+        *fp++ = e2[j2];
+        *fp++ = e3[j2];
+      }
+    }
+#endif
     break;
   case 6:
     e0=e;
@@ -86,95 +174,77 @@ void nr_interleaving_ldpc(uint32_t E, uint8_t Qm, uint8_t *e,uint8_t *f)
     e3=e2+EQm;
     e4=e3+EQm;
     e5=e4+EQm;
-    e0_256=(simde__m256i *)e0;
-    e1_256=(simde__m256i *)e1;
-    e2_256=(simde__m256i *)e2;
-    e3_256=(simde__m256i *)e3;
-    e4_256=(simde__m256i *)e4;
-    e5_256=(simde__m256i *)e5;
+#ifdef __AVX512F__
+    e0_512=(__m512i *)e0;
+    e1_512=(__m512i *)e1;
+    e2_512=(__m512i *)e2;
+    e3_512=(__m512i *)e3;
+    e4_512=(__m512i *)e4;
+    e5_512=(__m512i *)e5;
+    __m512i e4j,e5j;
+    p8a = _mm512_set_epi8(95,31,94,30,93,29,92,28,91,27,90,26,89,25,88,24,87,23,86,22,85,21,84,20,83,19,82,18,81,17,80,16,79,15,78,14,77,13,76,12,75,11,74,10,73,9,72,8,71,7,70,6,69,5,68,4,67,3,66,2,65,1,64,0);
+    p8b = _mm512_set_epi8(127,63,126,62,125,61,124,60,123,59,122,58,121,57,120,56,119,55,118,54,117,53,116,52,115,51,114,50,113,49,112,48,111,47,110,46,109,45,108,44,107,43,106,42,105,41,104,40,103,39,102,38,101,37,100,36,99,35,98,34,97,33,96,32);
 
-    for (int j=0,k=0;j<EQm>>5;j++,k+=192) {
-      fp  = f+k;
-      fp2 = fp+96;
+    p16a = _mm512_set_epi16(47,15,46,14,45,13,44,12,43,11,42,10,41,9,40,8,39,7,38,6,37,5,36,4,35,3,34,2,33,1,32,0);
+    p16b = _mm512_set_epi16(63,31,62,30,61,29,60,28,59,27,58,26,57,25,56,24,55,23,54,22,53,21,52,20,51,19,50,18,49,17,48,16);
 
-      tmp0   = simde_mm256_unpacklo_epi8(e0_256[j],e1_256[j]); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
-      tmp1   = simde_mm256_unpacklo_epi8(e2_256[j],e3_256[j]); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
-      tmp0b  = simde_mm256_unpacklo_epi16(tmp0,tmp1); // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
-      tmp1b  = simde_mm256_unpackhi_epi16(tmp0,tmp1); // e0(i+8) e1(i+8) e2(i+8) e3(i+8) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
-      tmp0   = simde_mm256_unpacklo_epi8(e4_256[j],e5_256[j]); // e4(i) e5(i) e4(i+1) e5(i+1) .... e4(i+15) e5(i+15)
-      *((uint32_t*)fp)      = simde_mm256_extract_epi32(tmp0b,0);
-      *((uint16_t*)(fp+4))  = simde_mm256_extract_epi16(tmp0,0);
-      *((uint32_t*)(fp+6))  = simde_mm256_extract_epi32(tmp0b,1);
-      *((uint16_t*)(fp+10)) = simde_mm256_extract_epi16(tmp0,1);
-      *((uint32_t*)(fp+12)) = simde_mm256_extract_epi32(tmp0b,2);
-      *((uint16_t*)(fp+16)) = simde_mm256_extract_epi16(tmp0,2);
-      *((uint32_t*)(fp+18)) = simde_mm256_extract_epi32(tmp0b,3);
-      *((uint16_t*)(fp+22)) = simde_mm256_extract_epi16(tmp0,3);
-      *((uint32_t*)(fp+24)) = simde_mm256_extract_epi32(tmp0b,4);
-      *((uint16_t*)(fp+26)) = simde_mm256_extract_epi16(tmp0,4);
-      *((uint32_t*)(fp+30)) = simde_mm256_extract_epi32(tmp0b,5);
-      *((uint16_t*)(fp+34)) = simde_mm256_extract_epi16(tmp0,5);
-      *((uint32_t*)(fp+36)) = simde_mm256_extract_epi32(tmp0,6);
-      *((uint16_t*)(fp+40)) = simde_mm256_extract_epi16(tmp0,6);
-      *((uint32_t*)(fp+42)) = simde_mm256_extract_epi32(tmp0b,7);
-      *((uint16_t*)(fp+46)) = simde_mm256_extract_epi16(tmp0,7);
+    __m512i p16c  = _mm512_set_epi16(21,20,41,19,18,40,17,16,39,15,14,38,13,12,37,11,10,36,9,8,35,7,6,34,5,4,33,3,2,32,1,0);
+    __m512i p16d  = _mm512_set_epi16(10,52,9,8,51,7,6,50,5,4,49,3,2,48,1,0,47,31,30,46,29,28,45,27,26,44,25,24,43,23,22,42);
+    __m512i p16d2 = _mm512_set_epi16(32+10,30,32+9,32+8,27,32+7,32+6,24,32+5,32+4,21,32+3,32+2,18,32+1,32+0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0);
+    __m512i p16e  = _mm512_set_epi16(63,31,30,62,29,28,61,27,26,60,25,24,59,23,22,58,21,20,57,19,18,56,17,16,55,15,14,54,13,12,53,11);
 
-      *((uint32_t*)(fp+48)) = simde_mm256_extract_epi32(tmp1b,0);
-      *((uint16_t*)(fp+52)) = simde_mm256_extract_epi16(tmp0,8);
-      *((uint32_t*)(fp+56)) = simde_mm256_extract_epi32(tmp1b,1);
-      *((uint16_t*)(fp+60)) = simde_mm256_extract_epi16(tmp0,9);
-      *((uint32_t*)(fp+62)) = simde_mm256_extract_epi32(tmp1b,2);
-      *((uint16_t*)(fp+66)) = simde_mm256_extract_epi16(tmp0,10);
-      *((uint32_t*)(fp+68)) = simde_mm256_extract_epi32(tmp1b,3);
-      *((uint16_t*)(fp+72)) = simde_mm256_extract_epi16(tmp0,11);
-      *((uint32_t*)(fp+74)) = simde_mm256_extract_epi32(tmp1b,4);
-      *((uint16_t*)(fp+76)) = simde_mm256_extract_epi16(tmp0,12);
-      *((uint32_t*)(fp+80)) = simde_mm256_extract_epi32(tmp1b,5);
-      *((uint16_t*)(fp+82)) = simde_mm256_extract_epi16(tmp0,13);
-      *((uint32_t*)(fp+86)) = simde_mm256_extract_epi32(tmp1b,6);
-      *((uint16_t*)(fp+90)) = simde_mm256_extract_epi16(tmp0,14);
-      *((uint32_t*)(fp+92)) = simde_mm256_extract_epi32(tmp1b,7);
-      *((uint16_t*)(fp+94)) = simde_mm256_extract_epi16(tmp0,15);
+    for (j=0,k=0;j<EQm>>6;j++,k+=6) {
+      e0j    = _mm512_loadu_si512(e0_512+j);	    
+      e1j    = _mm512_loadu_si512(e1_512+j);	    
+      e2j    = _mm512_loadu_si512(e2_512+j);	    
+      e3j    = _mm512_loadu_si512(e3_512+j);	    
+      e4j    = _mm512_loadu_si512(e4_512+j);	    
+      e5j    = _mm512_loadu_si512(e5_512+j);	    
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8a,e1j); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+31) e1(i+31)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8a,e3j); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+31) e3(i+31)
+      tmp2   = _mm512_permutex2var_epi8(e4j,p8a,e5j); // e4(i) e5(i) e4(i+1) e5(i+1) .... e4(i+31) e5(i+31)
+      tmp3   = _mm512_permutex2var_epi16(tmp0,p16a,tmp1);  // e0(i) e1(i) e2(i) e3(i) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16b,tmp1);  // e0(i+16) e1(i+16) e2(i+16) e3(i+16) ... e0(i+31) e1(i+31) e2(i+31) e3(i+31)
+      _mm512_storeu_si512(f_512+k,_mm512_permutex2var_epi16(tmp3,p16c,tmp2));  // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) ... e0(i+9) e1(i+9) e2(i+9) e3(i+9) e4(i+9) e5(i+9) e0(i+10) e1(i+10)
+      tmp5   = _mm512_permutex2var_epi16(tmp3,p16d,tmp2);  // e2(i+10) e3(i+10) e4(i+10) e5(i+10) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) x x x x e4(i+16) e5(i+16) x x x x .... e4(i+20) e5(i+20) x x
+      _mm512_storeu_si512(f_512+k+1,_mm512_permutex2var_epi16(tmp5,p16d2,tmp4)); // e2(i+10) e3(i+10) e4(i+10) e5(i+10) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) e0(i+16) e1(i+16) e2(i+16) e3(i+16) e4(i+16) e5(i+16)  e0(i+20) e1(i+20) e2(i+20) e3(i+20) e4(i+20) e5(i+20) e0(i+21) e1(i+21)  
+      _mm512_storeu_si512(f_512+k+2,_mm512_permutex2var_epi16(tmp4,p16e,tmp2)); // e2(i+21) e3(i+21) e4(i+21) e5(i+21) .... e0(i+31) e1(i+31) e2(i+31) e3(i+31) e4(i+31) e5(i+31) 
 
-      tmp0   = simde_mm256_unpackhi_epi8(e0_256[j],e1_256[j]); // e0(i+16) e1(i+16) e0(i+17) e1(i+17) .... e0(i+31) e1(i+31)
-      tmp1   = simde_mm256_unpackhi_epi8(e2_256[j],e3_256[j]); // e2(i+16) e3(i+16) e2(i+17) e3(i+17) .... e2(i+31) e3(i+31)
-      tmp0b  = simde_mm256_unpacklo_epi16(tmp0,tmp1); // e0(i+16) e1(i+16) e2(i+16) e3(i+16) ... e0(i+23) e1(i+23) e2(i+23) e3(i+23)
-      tmp1b  = simde_mm256_unpackhi_epi16(tmp0,tmp1); // e0(i+24) e1(i+24) e2(i+24) e3(i+24) ... e0(i+31) e1(i+31) e2(i+31) e3(i+31)
-      tmp0   = simde_mm256_unpackhi_epi8(e4_256[j],e5_256[j]); // e4(i+16) e5(i+16) e4(i+17) e5(i+17) .... e4(i+31) e5(i+31)
-      *((uint32_t*)fp2)      = simde_mm256_extract_epi32(tmp0b,0);
-      *((uint16_t*)(fp2+4))  = simde_mm256_extract_epi16(tmp0,0);
-      *((uint32_t*)(fp2+6))  = simde_mm256_extract_epi32(tmp0b,1);
-      *((uint16_t*)(fp2+10)) = simde_mm256_extract_epi16(tmp0,1);
-      *((uint32_t*)(fp2+12)) = simde_mm256_extract_epi32(tmp0b,2);
-      *((uint16_t*)(fp2+16)) = simde_mm256_extract_epi16(tmp0,2);
-      *((uint32_t*)(fp2+18)) = simde_mm256_extract_epi32(tmp0b,3);
-      *((uint16_t*)(fp2+22)) = simde_mm256_extract_epi16(tmp0,3);
-      *((uint32_t*)(fp2+24)) = simde_mm256_extract_epi32(tmp0b,4);
-      *((uint16_t*)(fp2+26)) = simde_mm256_extract_epi16(tmp0,4);
-      *((uint32_t*)(fp2+30)) = simde_mm256_extract_epi32(tmp0b,5);
-      *((uint16_t*)(fp2+34)) = simde_mm256_extract_epi16(tmp0,5);
-      *((uint32_t*)(fp2+36)) = simde_mm256_extract_epi32(tmp0,6);
-      *((uint16_t*)(fp2+40)) = simde_mm256_extract_epi16(tmp0,6);
-      *((uint32_t*)(fp2+42)) = simde_mm256_extract_epi32(tmp0b,7);
-      *((uint16_t*)(fp2+46)) = simde_mm256_extract_epi16(tmp0,7);
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8b,e1j); // e0(i+32) e1(i+32) e0(i+32) e1(i+32) .... e0(i+63) e1(i+63)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8b,e3j); // e2(i+32) e3(i+32) e2(i+32) e3(i+32) .... e2(i+63) e3(i+63)
+      tmp2   = _mm512_permutex2var_epi8(e4j,p8b,e5j); // e4(i+32) e5(i+32) e4(i+32) e5(i+32) .... e4(i+63) e5(i+63)
+      tmp3   = _mm512_permutex2var_epi16(tmp0,p16a,tmp1);  // e0(i) e1(i) e2(i) e3(i) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16b,tmp1);  // e0(i+16) e1(i+16) e2(i+16) e3(i+16) ... e0(i+31) e1(i+31) e2(i+31) e3(i+31)
 
-      *((uint32_t*)(fp2+48)) = simde_mm256_extract_epi32(tmp1b,0);
-      *((uint16_t*)(fp2+52)) = simde_mm256_extract_epi16(tmp0,8);
-      *((uint32_t*)(fp2+56)) = simde_mm256_extract_epi32(tmp1b,1);
-      *((uint16_t*)(fp2+60)) = simde_mm256_extract_epi16(tmp0,9);
-      *((uint32_t*)(fp2+62)) = simde_mm256_extract_epi32(tmp1b,2);
-      *((uint16_t*)(fp2+66)) = simde_mm256_extract_epi16(tmp0,10);
-      *((uint32_t*)(fp2+68)) = simde_mm256_extract_epi32(tmp1b,3);
-      *((uint16_t*)(fp2+72)) = simde_mm256_extract_epi16(tmp0,11);
-      *((uint32_t*)(fp2+74)) = simde_mm256_extract_epi32(tmp1b,4);
-      *((uint16_t*)(fp2+76)) = simde_mm256_extract_epi16(tmp0,12);
-      *((uint32_t*)(fp2+80)) = simde_mm256_extract_epi32(tmp1b,5);
-      *((uint16_t*)(fp2+82)) = simde_mm256_extract_epi16(tmp0,13);
-      *((uint32_t*)(fp2+86)) = simde_mm256_extract_epi32(tmp1b,6);
-      *((uint16_t*)(fp2+90)) = simde_mm256_extract_epi16(tmp0,14);
-      *((uint32_t*)(fp2+92)) = simde_mm256_extract_epi32(tmp1b,7);
-      *((uint16_t*)(fp2+94)) = simde_mm256_extract_epi16(tmp0,15);
+      _mm512_storeu_si512(f_512+k+3,_mm512_permutex2var_epi16(tmp3,p16c,tmp2));  // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) ... e0(i+9) e1(i+9) e2(i+9) e3(i+9) e4(i+9) e5(i+9) e0(i+10) e1(i+10)
+      tmp5   = _mm512_permutex2var_epi16(tmp3,p16d,tmp2);  // e2(i+10) e3(i+10) e4(i+10) e5(i+10) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) x x x x e4(i+16) e5(i+16) x x x x .... e4(i+20) e5(i+20) x x
+      _mm512_storeu_si512(f_512+k+4,_mm512_permutex2var_epi16(tmp5,p16d2,tmp4)); // e2(i+10) e3(i+10) e4(i+10) e5(i+10) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) e0(i+16) e1(i+16) e2(i+16) e3(i+16) e4(i+16) e5(i+16)  e0(i+20) e1(i+20) e2(i+20) e3(i+20) e4(i+20) e5(i+20) e0(i+21) e1(i+21)  
+      _mm512_storeu_si512(f_512+k+5,_mm512_permutex2var_epi16(tmp4,p16e,tmp2)); // e2(i+21) e3(i+21) e4(i+21) e5(i+21) .... e0(i+31) e1(i+31) e2(i+31) e3(i+31) e4(i+31) e5(i+31) 
     }
+    if ((j<<6) != EQm) {
+      int k2=k<<6;
+      fp = &f[k2];
+      for (int j2=(j<<6) ; j2< EQm ; j2++) {
+        *fp++ = e0[j2];
+        *fp++ = e1[j2];
+        *fp++ = e2[j2];
+        *fp++ = e3[j2];
+        *fp++ = e4[j2];
+        *fp++ = e5[j2];
+      }
+    }
+#else
+    fp = f;
+    for (int j = 0; j< EQm; j++){
+      *fp++ = e0[j];
+      *fp++ = e1[j];
+      *fp++ = e2[j];
+      *fp++ = e3[j];
+      *fp++ = e4[j];
+      *fp++ = e5[j];
+    }
+
+#endif
     break;
   case 8:
     e0=e;
@@ -186,43 +256,135 @@ void nr_interleaving_ldpc(uint32_t E, uint8_t Qm, uint8_t *e,uint8_t *f)
     e6=e5+EQm;
     e7=e6+EQm;
 
-    e0_256=(simde__m256i *)e0;
-    e1_256=(simde__m256i *)e1;
-    e2_256=(simde__m256i *)e2;
-    e3_256=(simde__m256i *)e3;
-    e4_256=(simde__m256i *)e4;
-    e5_256=(simde__m256i *)e5;
-    e6_256=(simde__m256i *)e6;
-    e7_256=(simde__m256i *)e7;
-    for (int k=0,j=0;j<EQm>>5;j++,k+=8) {
-      tmp0   = simde_mm256_unpacklo_epi8(e0_256[j],e1_256[j]); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
-      tmp1   = simde_mm256_unpacklo_epi8(e2_256[j],e3_256[j]); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
-      tmp2   = simde_mm256_unpacklo_epi8(e4_256[j],e5_256[j]); // e4(i) e5(i) e4(i+1) e5(i+1) .... e4(i+15) e5(i+15)
-      tmp3   = simde_mm256_unpacklo_epi8(e6_256[j],e7_256[j]); // e6(i) e7(i) e6(i+1) e7(i+1) .... e6(i+15) e7(i+15)
-      tmp4   = simde_mm256_unpacklo_epi16(tmp0,tmp1);  // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
-      tmp5   = simde_mm256_unpacklo_epi16(tmp2,tmp3);  // e4(i) e5(i) e6(i) e7(i) ... e4(i+7) e5(i+7) e6(i+7) e7(i+7)
-      f_256[k]   = simde_mm256_unpacklo_epi16(tmp4,tmp5);  // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) e6(i) e7(i)... e0(i+3) e1(i+3) e2(i+3) e3(i+3) e4(i+3) e5(i+3) e6(i+3) e7(i+3))
-      f_256[k+1] = simde_mm256_unpackhi_epi16(tmp4,tmp5);  // e0(i+4) e1(i+4) e2(i+4) e3(i+4) e4(i+4) e5(i+4) e6(i+4) e7(i+4)... e0(i+7) e1(i+7) e2(i+7) e3(i+7) e4(i+7) e5(i+7) e6(i+7) e7(i+7))
+#ifdef __AVX512F__
+    e0_512=(__m512i *)e0;
+    e1_512=(__m512i *)e1;
+    e2_512=(__m512i *)e2;
+    e3_512=(__m512i *)e3;
+    e4_512=(__m512i *)e4;
+    e5_512=(__m512i *)e5;
+    e6_512=(__m512i *)e6;
+    e7_512=(__m512i *)e7;
+    __m512i e6j,e7j;
+    p8a = _mm512_set_epi8(95,31,94,30,93,29,92,28,91,27,90,26,89,25,88,24,87,23,86,22,85,21,84,20,83,19,82,18,81,17,80,16,79,15,78,14,77,13,76,12,75,11,74,10,73,9,72,8,71,7,70,6,69,5,68,4,67,3,66,2,65,1,64,0);
+    p8b = _mm512_set_epi8(127,63,126,62,125,61,124,60,123,59,122,58,121,57,120,56,119,55,118,54,117,53,116,52,115,51,114,50,113,49,112,48,111,47,110,46,109,45,108,44,107,43,106,42,105,41,104,40,103,39,102,38,101,37,100,36,99,35,98,34,97,33,96,32);
 
-      tmp4   = simde_mm256_unpackhi_epi16(tmp0,tmp1);  // e0(i+8) e1(i+8) e2(i+8) e3(i+8) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
-      tmp5   = simde_mm256_unpackhi_epi16(tmp2,tmp3);  // e4(i+8) e5(i+8) e6(i+8) e7(i+8) ... e4(i+15) e5(i+15) e6(i+15) e7(i+15)
-      f_256[k+2]   = simde_mm256_unpacklo_epi16(tmp4,tmp5);  // e0(i+8) e1(i+8) e2(i+8) e3(i+8) e4(i+8) e5(i+8) e6(i+8) e7(i+8)... e0(i+11) e1(i+11) e2(i+11) e3(i+11) e4(i+11) e5(i+11) e6(i+11) e7(i+11))
-      f_256[k+3] = simde_mm256_unpackhi_epi16(tmp4,tmp5);  // e0(i+12) e1(i+12) e2(i+12) e3(i+12) e4(i+12) e5(i+12) e6(i+12) e7(i+12)... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) e6(i+15) e7(i+15))
+    p16a = _mm512_set_epi16(47,15,46,14,45,13,44,12,43,11,42,10,41,9,40,8,39,7,38,6,37,5,36,4,35,3,34,2,33,1,32,0);
+    p16b = _mm512_set_epi16(63,31,62,30,61,29,60,28,59,27,58,26,57,25,56,24,55,23,54,22,53,21,52,20,51,19,50,18,49,17,48,16);
 
-      tmp0   = simde_mm256_unpackhi_epi8(e0_256[j],e1_256[j]); // e0(i+16) e1(i+16) e0(i+17) e1(i+17) .... e0(i+31) e1(i+31)
-      tmp1   = simde_mm256_unpackhi_epi8(e2_256[j],e3_256[j]); // e2(i+16) e3(i+16) e2(i+17) e3(i+17) .... e2(i+31) e3(i+31)
-      tmp2   = simde_mm256_unpackhi_epi8(e4_256[j],e5_256[j]); // e4(i+16) e5(i+16) e4(i+17) e5(i+17) .... e4(i+31) e5(i+31)
-      tmp3   = simde_mm256_unpackhi_epi8(e6_256[j],e7_256[j]); // e6(i+16) e7(i+16) e6(i+17) e7(i+17) .... e6(i+31) e7(i+31)
-      tmp4   = simde_mm256_unpacklo_epi16(tmp0,tmp1);  // e0(i+!6) e1(i+16) e2(i+16) e3(i+16) ... e0(i+23) e1(i+23) e2(i+23) e3(i+23)
-      tmp5   = simde_mm256_unpacklo_epi16(tmp2,tmp3);  // e4(i+16) e5(i+16) e6(i+16) e7(i+16) ... e4(i+23) e5(i+23) e6(i+23) e7(i+23)
-      f_256[k+4] = simde_mm256_unpacklo_epi16(tmp4,tmp5);  // e0(i+16) e1(i+16) e2(i+16) e3(i+16) e4(i+16) e5(i+16) e6(i+16) e7(i+16)... e0(i+19) e1(i+19) e2(i+19) e3(i+19) e4(i+19) e5(i+19) e6(i+19) e7(i+19))
-      f_256[k+5] = simde_mm256_unpackhi_epi16(tmp4,tmp5);  // e0(i+20) e1(i+20) e2(i+20) e3(i+20) e4(i+20) e5(i+20) e6(i+20) e7(i+20)... e0(i+23) e1(i+23) e2(i+23) e3(i+23) e4(i+23) e5(i+23) e6(i+23) e7(i+23))
+    p32a = _mm512_set_epi32(23,7,22,6,21,5,20,4,19,3,18,2,17,1,16,0);
+    p32b = _mm512_set_epi32(31,15,30,14,29,13,28,12,27,11,26,10,25,9,24,8);
+    for (k=0,j=0;j<EQm>>6;j++,k+=8) {
+      e0j    = _mm512_loadu_si512(e0_512+j);	    
+      e1j    = _mm512_loadu_si512(e1_512+j);	    
+      e2j    = _mm512_loadu_si512(e2_512+j);	    
+      e3j    = _mm512_loadu_si512(e3_512+j);	    
+      e4j    = _mm512_loadu_si512(e4_512+j);	    
+      e5j    = _mm512_loadu_si512(e5_512+j);	    
+      e6j    = _mm512_loadu_si512(e6_512+j);	    
+      e7j    = _mm512_loadu_si512(e7_512+j);	    
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8a,e1j); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+15) e1(i+15)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8a,e3j); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+15) e3(i+15)
+      tmp2   = _mm512_permutex2var_epi8(e4j,p8a,e5j); // e4(i) e5(i) e4(i+1) e5(i+1) .... e4(i+15) e5(i+15)
+      tmp3   = _mm512_permutex2var_epi8(e6j,p8a,e7j); // e6(i) e7(i) e6(i+1) e7(i+1) .... e6(i+15) e7(i+15)
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16a,tmp1);  // e0(i) e1(i) e2(i) e3(i) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)
+      tmp5   = _mm512_permutex2var_epi16(tmp2,p16a,tmp3);  // e4(i) e5(i) e6(i) e7(i) ... e4(i+7) e5(i+7) e6(i+7) e7(i+7)
+      _mm512_storeu_si512(f_512+k,_mm512_permutex2var_epi32(tmp4,p32a,tmp5));  // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) e6(i) e7(i)... e0(i+3) e1(i+3) e2(i+3) e3(i+3) e4(i+3) e5(i+3) e6(i+3) e7(i+3))
+      _mm512_storeu_si512(f_512+k+1,_mm512_permutex2var_epi32(tmp4,p32b,tmp5));  // e0(i+4) e1(i+4) e2(i+4) e3(i+4) e4(i+4) e5(i+4) e6(i+4) e7(i+4)... e0(i+7) e1(i+7) e2(i+7) e3(i+7) e4(i+7) e5(i+7) e6(i+7) e7(i+7))
 
-      tmp4   = simde_mm256_unpackhi_epi16(tmp0,tmp1);  // e0(i+24) e1(i+24) e2(i+24) e3(i+24) ... e0(i+31) e1(i+31) e2(i+31) e3(i+31)
-      tmp5   = simde_mm256_unpackhi_epi16(tmp2,tmp3);  // e4(i+24) e5(i+24) e6(i+24) e7(i+24) ... e4(i+31) e5(i+31) e6(i+31) e7(i+31)
-      f_256[k+6] = simde_mm256_unpacklo_epi16(tmp4,tmp5);  // e0(i+24) e1(i+24) e2(i+24) e3(i+24) e4(i+24) e5(i+24) e6(i+24) e7(i+24)... e0(i+27) e1(i+27) e2(i+27) e3(i+27) e4(i+27) e5(i+27) e6(i+27) e7(i+27))
-      f_256[k+7] = simde_mm256_unpackhi_epi16(tmp4,tmp5);  // e0(i+28) e1(i+28) e2(i+28) e3(i+28) e4(i+28) e5(i+28) e6(i+28) e7(i+28)... e0(i+31) e1(i+31) e2(i+31) e3(i+31) e4(i+31) e5(i+31) e6(i+31) e7(i+31))
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16b,tmp1);  // e0(i+8) e1(i+8) e2(i+8) e3(i+8) ... e0(i+15) e1(i+15) e2(i+15) e3(i+15)
+      tmp5   = _mm512_permutex2var_epi16(tmp2,p16b,tmp3);  // e4(i+8) e5(i+8) e6(i+8) e7(i+8) ... e4(i+15) e5(i+15) e6(i+15) e7(i+15)
+      _mm512_storeu_si512(f_512+k+2,_mm512_permutex2var_epi32(tmp4,p32a,tmp5));  // e0(i+8) e1(i+8) e2(i+8) e3(i+8) e4(i+8) e5(i+8) e6(i+8) e7(i+8)... e0(i+11) e1(i+11) e2(i+11) e3(i+11) e4(i+11) e5(i+11) e6(i+11) e7(i+11))
+      _mm512_storeu_si512(f_512+k+3,_mm512_permutex2var_epi32(tmp4,p32b,tmp5));  // e0(i+12) e1(i+12) e2(i+12) e3(i+12) e4(i+12) e5(i+12) e6(i+12) e7(i+12)... e0(i+15) e1(i+15) e2(i+15) e3(i+15) e4(i+15) e5(i+15) e6(i+15) e7(i+15))
+
+      tmp0   = _mm512_permutex2var_epi8(e0j,p8b,e1j); // e0(i+16) e1(i+16) e0(i+17) e1(i+17) .... e0(i+31) e1(i+31)
+      tmp1   = _mm512_permutex2var_epi8(e2j,p8b,e3j); // e2(i+16) e3(i+16) e2(i+17) e3(i+17) .... e2(i+31) e3(i+31)
+      tmp2   = _mm512_permutex2var_epi8(e4j,p8b,e5j); // e4(i+16) e5(i+16) e4(i+17) e5(i+17) .... e4(i+31) e5(i+31)
+      tmp3   = _mm512_permutex2var_epi8(e6j,p8b,e7j); // e6(i+16) e7(i+16) e6(i+17) e7(i+17) .... e6(i+31) e7(i+31)
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16a,tmp1);  // e0(i+!6) e1(i+16) e2(i+16) e3(i+16) ... e0(i+23) e1(i+23) e2(i+23) e3(i+23)
+      tmp5   = _mm512_permutex2var_epi16(tmp2,p16a,tmp3);  // e4(i+16) e5(i+16) e6(i+16) e7(i+16) ... e4(i+23) e5(i+23) e6(i+23) e7(i+23)
+      _mm512_storeu_si512(f_512+k+4,_mm512_permutex2var_epi32(tmp4,p32a,tmp5));  // e0(i+16) e1(i+16) e2(i+16) e3(i+16) e4(i+16) e5(i+16) e6(i+16) e7(i+16)... e0(i+19) e1(i+19) e2(i+19) e3(i+19) e4(i+19) e5(i+19) e6(i+19) e7(i+19))
+      _mm512_storeu_si512(f_512+k+5,_mm512_permutex2var_epi32(tmp4,p32b,tmp5));  // e0(i+20) e1(i+20) e2(i+20) e3(i+20) e4(i+20) e5(i+20) e6(i+20) e7(i+20)... e0(i+23) e1(i+23) e2(i+23) e3(i+23) e4(i+23) e5(i+23) e6(i+23) e7(i+23))
+
+      tmp4   = _mm512_permutex2var_epi16(tmp0,p16b,tmp1);  // e0(i+24) e1(i+24) e2(i+24) e3(i+24) ... e0(i+31) e1(i+31) e2(i+31) e3(i+31)
+      tmp5   = _mm512_permutex2var_epi16(tmp2,p16b,tmp3);  // e4(i+24) e5(i+24) e6(i+24) e7(i+24) ... e4(i+31) e5(i+31) e6(i+31) e7(i+31)
+      _mm512_storeu_si512(f_512+k+6,_mm512_permutex2var_epi32(tmp4,p32a,tmp5));  // e0(i+24) e1(i+24) e2(i+24) e3(i+24) e4(i+24) e5(i+24) e6(i+24) e7(i+24)... e0(i+27) e1(i+27) e2(i+27) e3(i+27) e4(i+27) e5(i+27) e6(i+27) e7(i+27))
+      _mm512_storeu_si512(f_512+k+7,_mm512_permutex2var_epi32(tmp4,p32b,tmp5));  // e0(i+28) e1(i+28) e2(i+28) e3(i+28) e4(i+28) e5(i+28) e6(i+28) e7(i+28)... e0(i+31) e1(i+31) e2(i+31) e3(i+31) e4(i+31) e5(i+31) e6(i+31) e7(i+31))
     }
+    if ((j<<6) != EQm) {
+      int k2=k<<6;
+      fp=&f[k2];
+      for (int j2=(j<<6) ; j2< EQm ; j2++) {
+         *fp++ = e0[j2];
+         *fp++ = e1[j2];
+         *fp++ = e2[j2];
+         *fp++ = e3[j2];
+         *fp++ = e4[j2];
+         *fp++ = e5[j2];
+         *fp++ = e6[j2];
+         *fp++ = e7[j2];
+      }
+    }
+#else
+    e0_128=(simde__m128i *)e0;
+    e1_128=(simde__m128i *)e1;
+    e2_128=(simde__m128i *)e2;
+    e3_128=(simde__m128i *)e3;
+    e4_128=(simde__m128i *)e4;
+    e5_128=(simde__m128i *)e5;
+    e6_128=(simde__m128i *)e6;
+    e7_128=(simde__m128i *)e7;
+    simde__m128i e0j,e1j,e2j,e3j,e4j,e5j,e6j,e7j;
+    
+    for (k=0,j=0;j<EQm>>4;j++,k+=8) {
+      e0j    = simde_mm_loadu_si128(e0_128+j);	    
+      e1j    = simde_mm_loadu_si128(e1_128+j);	    
+      e2j    = simde_mm_loadu_si128(e2_128+j);	    
+      e3j    = simde_mm_loadu_si128(e3_128+j);	    
+      e4j    = simde_mm_loadu_si128(e4_128+j);	    
+      e5j    = simde_mm_loadu_si128(e5_128+j);	    
+      e6j    = simde_mm_loadu_si128(e6_128+j);	    
+      e7j    = simde_mm_loadu_si128(e7_128+j);	    
+      tmp0   = simde_mm_unpacklo_epi8(e0j,e1j); // e0(i) e1(i) e0(i+1) e1(i+1) .... e0(i+7) e1(i+7)
+      tmp1   = simde_mm_unpacklo_epi8(e2j,e3j); // e2(i) e3(i) e2(i+1) e3(i+1) .... e2(i+7) e3(i+7)
+      tmp2   = simde_mm_unpacklo_epi16(tmp0,tmp1);// e0(i) e1(i) e2(i) e3(i) e0(i+1) e1(i+1) e2(i+1) e3(i+1) ... e0(i+3) e1(i+3) e2(i+3) e3(i+3)	
+      tmp3   = simde_mm_unpackhi_epi16(tmp0,tmp1);// e0(i+4) e1(i+4) e2(i+4) e3(i+4) e0(i+5) e1(i+5) e2(i+5) e3(i+5) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)	
+      tmp4   = simde_mm_unpacklo_epi8(e4j,e5j); // e4(i) e5(i) e4(i+1) e5(i+1) .... e4(i+7) e5(i+7)
+      tmp5   = simde_mm_unpacklo_epi8(e6j,e7j); // e6(i) e7(i) e6(i+1) e7(i+1) .... e6(i+7) e7(i+7)
+      tmp6   = simde_mm_unpacklo_epi16(tmp4,tmp5);// e4(i) e5(i) e6(i) e7(i) e4(i+1) e5(i+1) e6(i+1) e7(i+1) ... e4(i+3) e5(i+3) e6(i+) e7(i+3)	
+      tmp7   = simde_mm_unpackhi_epi16(tmp4,tmp5);// e4(i+4) e5(i+4) e6(i+4) e7(i+4) e4(i+5) e5(i+5) e6(i+5) e7(i+5) ... e4(i+7) e5(i+7) e6(i+7) e7(i+7)	
+      simde_mm_storeu_si128(f_128+k,simde_mm_unpacklo_epi32(tmp2,tmp6));   // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) e6(i) e7(i) e0(i+1) ... e7(i+1)
+      simde_mm_storeu_si128(f_128+k+1,simde_mm_unpackhi_epi32(tmp2,tmp6));   // e0(i+2) e1(i+2) e2(i+2) e3(i+2) e4(i+2) e5(i+2) e6(i+2) e7(i+2) e0(i+3) e1(i+3) ... e7(i+3)
+      simde_mm_storeu_si128(f_128+k+2,simde_mm_unpacklo_epi32(tmp3,tmp7));   // e0(i+4) e1(i+4) e2(i+4) e3(i+4) e4(i+4) e5(i+4) e6(i+4) e7+4(i) e0(i+5) ... e7(i+5)
+      simde_mm_storeu_si128(f_128+k+3,simde_mm_unpackhi_epi32(tmp3,tmp7));   // e0(i+6) e1(i+6) e2(i+6) e3(i+6) e4(i+6) e5(i+6) e6(i+6) e7(i+6) e0(i+7) e0(i+7) ... e7(i+7)
+
+
+      tmp0   = simde_mm_unpackhi_epi8(e0j,e1j); // e0(i+8) e1(i+8) e0(i+9) e1(i+9) .... e0(i+15) e1(i+15)
+      tmp1   = simde_mm_unpackhi_epi8(e2j,e3j); // e2(i+8) e3(i+8) e2(i+9) e3(i+9) .... e2(i+15) e3(i+15)
+      tmp2   = simde_mm_unpacklo_epi16(tmp0,tmp1);// e0(i) e1(i) e2(i) e3(i) e0(i+1) e1(i+1) e2(i+1) e3(i+1) ... e0(i+3) e1(i+3) e2(i+3) e3(i+3)	
+      tmp3   = simde_mm_unpackhi_epi16(tmp0,tmp1);// e0(i+4) e1(i+4) e2(i+4) e3(i+4) e0(i+5) e1(i+5) e2(i+5) e3(i+5) ... e0(i+7) e1(i+7) e2(i+7) e3(i+7)	
+      tmp4   = simde_mm_unpackhi_epi8(e4j,e5j); // e4(i+8) e5(i+8) e4(i+9) e5(i+9) .... e4(i+15) e515i+7)
+      tmp5   = simde_mm_unpackhi_epi8(e6j,e7j); // e6(i+8) e7(i+8) e6(i+9) e7(i+9) .... e6(i+15) e7(i+15)
+      tmp6   = simde_mm_unpacklo_epi16(tmp4,tmp5);// e4(i) e5(i) e6(i) e7(i) e4(i+1) e5(i+1) e6(i+1) e7(i+1) ... e4(i+3) e5(i+3) e6(i+) e7(i+3)	
+      tmp7   = simde_mm_unpackhi_epi16(tmp4,tmp5);// e4(i+4) e5(i+4) e6(i+4) e7(i+4) e4(i+5) e5(i+5) e6(i+5) e7(i+5) ... e4(i+7) e5(i+7) e6(i+7) e7(i+7)	
+      simde_mm_storeu_si128(f_128+k+4,simde_mm_unpacklo_epi32(tmp2,tmp6));   // e0(i) e1(i) e2(i) e3(i) e4(i) e5(i) e6(i) e7(i) e0(i+1) ... e7(i+1)
+      simde_mm_storeu_si128(f_128+k+5,simde_mm_unpackhi_epi32(tmp2,tmp6));   // e0(i+2) e1(i+2) e2(i+2) e3(i+2) e4(i+2) e5(i+2) e6(i+2) e7(i+2) e0(i+3) e1(i+3) ... e7(i+3)
+      simde_mm_storeu_si128(f_128+k+6,simde_mm_unpacklo_epi32(tmp3,tmp7));   // e0(i+4) e1(i+4) e2(i+4) e3(i+4) e4(i+4) e5(i+4) e6(i+4) e7+4(i) e0(i+5) ... e7(i+5)
+      simde_mm_storeu_si128(f_128+k+7,simde_mm_unpackhi_epi32(tmp3,tmp7));   // e0(i+6) e1(i+6) e2(i+6) e3(i+6) e4(i+6) e5(i+6) e6(i+6) e7(i+6) e0(i+7) e0(i+7) ... e7(i+7)
+    }
+    if ((j<<4) != EQm) {
+      int k2=k<<4;
+      fp=&f[k2];
+      for (int j2=(j<<4) ; j2< EQm ; j2++) {
+         *fp++ = e0[j2];
+         *fp++ = e1[j2];
+         *fp++ = e2[j2];
+         *fp++ = e3[j2];
+         *fp++ = e4[j2];
+         *fp++ = e5[j2];
+      }
+    }
+#endif
     break;
   default: AssertFatal(1==0,"Should be here!\n");
   }

openair1/PHY/CODING/nr_segmentation.c

@@ -151,13 +151,8 @@ else
     for (r=0; r<*C; r++) {
 
       k = 0;
-
-      while (k<((Kprime - L)>>3)) {
-        output_buffers[r][k] = input_buffer[s];
-		//printf("encoding segment %d : byte %d (%d) => %d\n",r,k,(Kprime-L)>>3,input_buffer[s]);
-        k++;
-        s++;
-      }
+      memcpy(output_buffers[r],input_buffer+s,(Kprime-L)>>3);
+      s+=(Kprime-L)>>3;
 
       if (*C > 1) { // add CRC
         crc = crc24b(output_buffers[r],Kprime-L)>>8;

openair1/PHY/MODULATION/nr_modulation.c

@@ -24,8 +24,10 @@
 #include "executables/softmodem-common.h"
 
 // #define DEBUG_DLSCH_PRECODING_PRINT_WITH_TRIVIAL // TODO: For debug, to be removed if want to merge to develop
-
-//Table 6.3.1.5-1 Precoding Matrix W 1 layer 2 antenna ports 'n' = -1 and 'o' = -j
+//#define DEBUG_LAYER_MAPPING
+#define USE_NEON
+//#define USE_GATHER
+// Table 6.3.1.5-1 Precoding Matrix W 1 layer 2 antenna ports 'n' = -1 and 'o' = -j
 const char nr_W_1l_2p[6][2][1] = {
     {{'1'}, {'0'}}, // pmi 0
     {{'0'}, {'1'}},
@@ -35,7 +37,7 @@ const char nr_W_1l_2p[6][2][1] = {
     {{'1'}, {'o'}} // pmi 5
 };
 
-//Table 6.3.1.5-3 Precoding Matrix W 1 layer 4 antenna ports 'n' = -1 and 'o' = -j
+// Table 6.3.1.5-3 Precoding Matrix W 1 layer 4 antenna ports 'n' = -1 and 'o' = -j
 const char nr_W_1l_4p[28][4][1] = {
     {{'1'}, {'0'}, {'0'}, {'0'}}, // pmi 0
     {{'0'}, {'1'}, {'0'}, {'0'}},
@@ -68,14 +70,14 @@ const char nr_W_1l_4p[28][4][1] = {
     {{'1'}, {'o'}, {'o'}, {'n'}} // pmi 27
 };
 
-//Table 6.3.1.5-4 Precoding Matrix W 2 antenna ports layers 2  'n' = -1 and 'o' = -j
+// Table 6.3.1.5-4 Precoding Matrix W 2 antenna ports layers 2  'n' = -1 and 'o' = -j
 const char nr_W_2l_2p[3][2][2] = {
     {{'1', '0'}, {'0', '1'}}, // pmi 0
     {{'1', '1'}, {'1', 'n'}},
     {{'1', '1'}, {'j', 'o'}} // pmi 2
 };
 
-//Table 6.3.1.5-5 Precoding Matrix W 2 layers 4 antenna ports 'n' = -1 and 'o' = -j
+// Table 6.3.1.5-5 Precoding Matrix W 2 layers 4 antenna ports 'n' = -1 and 'o' = -j
 const char nr_W_2l_4p[22][4][2] = {
     {{'1', '0'}, {'0', '1'}, {'0', '0'}, {'0', '0'}}, // pmi 0
     {{'1', '0'}, {'0', '0'}, {'0', '1'}, {'0', '0'}}, {{'1', '0'}, {'0', '0'}, {'0', '0'}, {'0', '1'}},
@@ -94,7 +96,7 @@ const char nr_W_2l_4p[22][4][2] = {
     {{'1', '1'}, {'o', 'o'}, {'1', 'n'}, {'o', 'j'}}, {{'1', '1'}, {'o', 'o'}, {'j', 'o'}, {'1', 'n'}} // pmi 21
 };
 
-//Table 6.3.1.5-6 Precoding Matrix W 3 layers 4 antenna ports 'n' = -1 and 'o' = -j
+// Table 6.3.1.5-6 Precoding Matrix W 3 layers 4 antenna ports 'n' = -1 and 'o' = -j
 const char nr_W_3l_4p[7][4][3] = {{{'1', '0', '0'}, {'0', '1', '0'}, {'0', '0', '1'}, {'0', '0', '0'}}, // pmi 0
                                   {{'1', '0', '0'}, {'0', '1', '0'}, {'1', '0', '0'}, {'0', '0', '1'}},
                                   {{'1', '0', '0'}, {'0', '1', '0'}, {'n', '0', '0'}, {'0', '0', '1'}},
@@ -103,7 +105,7 @@ const char nr_W_3l_4p[7][4][3] = {{{'1', '0', '0'}, {'0', '1', '0'}, {'0', '0',
                                   {{'1', '1', '1'}, {'n', '1', 'n'}, {'1', '1', 'n'}, {'n', '1', '1'}},
                                   {{'1', '1', '1'}, {'n', '1', 'n'}, {'j', 'j', 'o'}, {'o', 'j', 'j'}}};
 
-//Table 6.3.1.5-7 Precoding Matrix W 4 layers 4 antenna ports 'n' = -1 and 'o' = -j
+// Table 6.3.1.5-7 Precoding Matrix W 4 layers 4 antenna ports 'n' = -1 and 'o' = -j
 const char nr_W_4l_4p[5][4][4] = {
     {{'1', '0', '0', '0'}, {'0', '1', '0', '0'}, {'0', '0', '1', '0'}, {'0', '0', '0', '1'}}, // pmi 0
     {{'1', '1', '0', '0'}, {'0', '0', '1', '1'}, {'1', 'n', '0', '0'}, {'0', '0', '1', 'n'}},
@@ -112,137 +114,133 @@ const char nr_W_4l_4p[5][4][4] = {
     {{'1', '1', '1', '1'}, {'1', 'n', '1', 'n'}, {'j', 'j', 'o', 'o'}, {'j', 'o', 'o', 'j'}} // pmi 4
 };
 
-void nr_modulation(const uint32_t *in,
-                   uint32_t length,
-                   uint16_t mod_order,
-                   int16_t *out)
+void nr_modulation(const uint32_t *in, uint32_t length, uint16_t mod_order, int16_t *out)
 {
-  uint16_t mask = ((1<<mod_order)-1);
-  int32_t* nr_mod_table32;
-  int32_t* out32 = (int32_t*) out;
+  uint16_t mask = ((1 << mod_order) - 1);
+  int32_t *nr_mod_table32;
+  int32_t *out32 = (int32_t *)out;
   const uint8_t *in_bytes = (const uint8_t *)in;
   const uint64_t *in64 = (const uint64_t *)in;
-  int64_t* out64 = (int64_t*) out;
-  uint32_t i=0;
+  int64_t *out64 = (int64_t *)out;
+  uint32_t i = 0;
 
 #if defined(__SSE2__)
   simde__m128i *nr_mod_table128;
   simde__m128i *out128;
 #endif
 
-  LOG_D(PHY,"nr_modulation: length %d, mod_order %d\n",length,mod_order);
+  LOG_D(PHY, "nr_modulation: length %d, mod_order %d\n", length, mod_order);
 
   switch (mod_order) {
-
 #if defined(__SSE2__)
-  case 2:
-    nr_mod_table128 = (simde__m128i *)nr_qpsk_byte_mod_table;
-    out128 = (simde__m128i *)out;
-    for (i=0; i<length/8; i++)
-      out128[i] = nr_mod_table128[in_bytes[i]];
-    // the bits that are left out
-    i = i*8/2;
-    nr_mod_table32 = (int32_t*) nr_qpsk_mod_table;
-    while (i<length/2) {
-      const int idx = ((in_bytes[(i * 2) / 8] >> ((i * 2) & 0x7)) & mask);
-      out32[i] = nr_mod_table32[idx];
-      i++;
-    }
-    return;
+    case 2:
+      nr_mod_table128 = (simde__m128i *)nr_qpsk_byte_mod_table;
+      out128 = (simde__m128i *)out;
+      for (i = 0; i < length / 8; i++)
+        out128[i] = nr_mod_table128[in_bytes[i]];
+      // the bits that are left out
+      i = i * 8 / 2;
+      nr_mod_table32 = (int32_t *)nr_qpsk_mod_table;
+      while (i < length / 2) {
+        const int idx = ((in_bytes[(i * 2) / 8] >> ((i * 2) & 0x7)) & mask);
+        out32[i] = nr_mod_table32[idx];
+        i++;
+      }
+      return;
 #else
-  case 2:
-    nr_mod_table32 = (int32_t*) nr_qpsk_mod_table;
-    for (i=0; i<length/mod_order; i++) {
-      const int idx = ((in[i * 2 / 32] >> ((i * 2) & 0x1f)) & mask);
-      out32[i] = nr_mod_table32[idx];
-    }
-    return;
+    case 2:
+      nr_mod_table32 = (int32_t *)nr_qpsk_mod_table;
+      for (i = 0; i < length / mod_order; i++) {
+        const int idx = ((in[i * 2 / 32] >> ((i * 2) & 0x1f)) & mask);
+        out32[i] = nr_mod_table32[idx];
+      }
+      return;
 #endif
 
-  case 4:
-    out64 = (int64_t*) out;
-    for (i=0; i<length/8; i++)
-      out64[i] = nr_16qam_byte_mod_table[in_bytes[i]];
-    // the bits that are left out
-    i = i*8/4;
-    while (i<length/4) {
-      const int idx = ((in_bytes[(i * 4) / 8] >> ((i * 4) & 0x7)) & mask);
-      out32[i] = nr_16qam_mod_table[idx];
-      i++;
-    }
-    return;
-
-  case 6:
-    if (length > (3*64))
-      for (i = 0; i < length - 3 * 64; i += 3 * 64) {
-        uint64_t x = *in64++;
-        uint64_t x1 = x & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x >> 12) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x >> 24) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x >> 36) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x >> 48) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        uint64_t x2 = (x >> 60);
-        x = *in64++;
-        x2 |= x<<4;
-        x1 = x2 & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 12) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 24) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 36) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 48) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x2 = ((x>>56)&0xf0) | (x2>>60);
-        x = *in64++;
-        x2 |= x<<8;
-        x1 = x2 & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 12) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 24) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 36) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x1 = (x2 >> 48) & 0xfff;
-        *out64++ = nr_64qam_mod_table[x1];
-        x2 = ((x>>52)&0xff0) | (x2>>60);
-        *out64++ = nr_64qam_mod_table[x2];
+    case 4:
+      out64 = (int64_t *)out;
+      for (i = 0; i < length / 8; i++)
+        out64[i] = nr_16qam_byte_mod_table[in_bytes[i]];
+      // the bits that are left out
+      i = i * 8 / 4;
+      while (i < length / 4) {
+        const int idx = ((in_bytes[(i * 4) / 8] >> ((i * 4) & 0x7)) & mask);
+        out32[i] = nr_16qam_mod_table[idx];
+        i++;
       }
-    
-    while (i + 24 <= length) {
-      uint32_t xx = 0;
-      memcpy(&xx, in_bytes + i / 8, 3);
-      uint64_t x1 = xx & 0xfff;
-      *out64++ = nr_64qam_mod_table[x1];
-      x1 = (xx >> 12) & 0xfff;
-      *out64++ = nr_64qam_mod_table[x1];
-      i += 24;
-    }
-    if (i != length) {
-      uint32_t xx = 0;
-      memcpy(&xx, in_bytes + i / 8, 2);
-      uint64_t x1 = xx & 0xfff;
-      *out64++ = nr_64qam_mod_table[x1];
-    }
-    return;
+      return;
 
-  case 8:
-    nr_mod_table32 = (int32_t*) nr_256qam_mod_table;
-    for (i=0; i<length/8; i++)
-      out32[i] = nr_mod_table32[in_bytes[i]];
-    return;
+    case 6:
+      if (length > (3 * 64))
+        for (i = 0; i < length - 3 * 64; i += 3 * 64) {
+          uint64_t x = *in64++;
+          uint64_t x1 = x & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x >> 12) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x >> 24) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x >> 36) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x >> 48) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          uint64_t x2 = (x >> 60);
+          x = *in64++;
+          x2 |= x << 4;
+          x1 = x2 & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 12) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 24) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 36) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 48) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x2 = ((x >> 56) & 0xf0) | (x2 >> 60);
+          x = *in64++;
+          x2 |= x << 8;
+          x1 = x2 & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 12) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 24) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 36) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x1 = (x2 >> 48) & 0xfff;
+          *out64++ = nr_64qam_mod_table[x1];
+          x2 = ((x >> 52) & 0xff0) | (x2 >> 60);
+          *out64++ = nr_64qam_mod_table[x2];
+        }
 
-  default:
-    break;
+      while (i + 24 <= length) {
+        uint32_t xx = 0;
+        memcpy(&xx, in_bytes + i / 8, 3);
+        uint64_t x1 = xx & 0xfff;
+        *out64++ = nr_64qam_mod_table[x1];
+        x1 = (xx >> 12) & 0xfff;
+        *out64++ = nr_64qam_mod_table[x1];
+        i += 24;
+      }
+      if (i != length) {
+        uint32_t xx = 0;
+        memcpy(&xx, in_bytes + i / 8, 2);
+        uint64_t x1 = xx & 0xfff;
+        *out64++ = nr_64qam_mod_table[x1];
+      }
+      return;
+
+    case 8:
+      nr_mod_table32 = (int32_t *)nr_256qam_mod_table;
+      for (i = 0; i < length / 8; i++)
+        out32[i] = nr_mod_table32[in_bytes[i]];
+      return;
+
+    default:
+      break;
   }
-  AssertFatal(false,"Invalid or unsupported modulation order %d\n",mod_order);
+  AssertFatal(false, "Invalid or unsupported modulation order %d\n", mod_order);
 }
 
 void nr_layer_mapping(int nbCodes,
@@ -251,36 +249,296 @@ void nr_layer_mapping(int nbCodes,
                       uint8_t n_layers,
                       int layerSz,
                       uint32_t n_symbs,
-                      c16_t tx_layer[layerSz],
-                      int layer)
+                      c16_t tx_layers[][layerSz])
 {
-  LOG_D(PHY,"Doing layer mapping for %d layers, %d symbols\n",n_layers,n_symbs);
+  LOG_D(PHY, "Doing layer mapping for %d layers, %d symbols\n", n_layers, n_symbs);
 
   switch (n_layers) {
-
     case 1:
-      memcpy(tx_layer, mod_symbs[0], n_symbs * sizeof(**mod_symbs));
+      memcpy(tx_layers[0], mod_symbs[0], n_symbs * sizeof(**mod_symbs));
       break;
 
-    case 2:
-    case 3:
-    case 4:
-    for (int i = 0; i < n_symbs / n_layers; i++) {
-      const c16_t *base = mod_symbs[0] + n_layers * i;
-      tx_layer[i] = base[layer];
-    }
-      break;
+    case 2: {
+#if defined(__AVX512BW__)
+      int i=0;
+      __m512i perm2a = _mm512_set_epi32(30, 28, 26, 24, 22, 20, 18, 16, 14, 12, 10, 8, 6, 4, 2, 0);
+      __m512i perm2b = _mm512_set_epi32(31, 29, 27, 25, 23, 21, 19, 17, 15, 13, 11, 9, 7, 5, 3, 1);
+#ifndef USE_GATHER
+      __m512i a, b;
+      int j;
+      for (i = 0, j = 0; i < n_symbs >> 4; i += 2, j++) {
+        a = ((__m512i *)mod_symbs[0])[i];
+        b = ((__m512i *)mod_symbs[0])[i + 1];
+        ((__m512i *)tx_layers[0])[j] = _mm512_permutex2var_epi32(a, perm2a, b);
+        ((__m512i *)tx_layers[1])[j] = _mm512_permutex2var_epi32(a, perm2b, b);
+      }
+      if (i << 4 != n_symbs) {
+        for (int i2 = ((n_symbs >> 4) << 4); i2 < n_symbs; i2 += 2) {
+          tx_layers[0][i2 >> 1] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 1] = mod_symbs[0][i2 + 1];
+        }
+      }
+#else
+  for (i=0; i <n_symbs; i+=32) {
+    *(__m512i *)(tx_layers[0]+i/2)=_mm512_i32gather_epi32(perm2a, mod_symbs[0]+i, 4);
+    *(__m512i *)(tx_layers[1]+i/2)=_mm512_i32gather_epi32(perm2b, mod_symbs[0]+i, 4);
+  }
+  for (; i < n_symbs; i += 2) {
+    tx_layers[0][i >> 1] = mod_symbs[0][i];
+    tx_layers[1][i >> 1] = mod_symbs[0][i + 1];
+  }
+#endif
+#elif defined(__aarch64__)
+      int i;
+#ifdef USE_NEON
+      // SIMDe doesn't handle this properly, gcc up to 14.2 neither
+      uint8_t const perm0[16] = {0, 1, 2, 3, 8, 9, 10, 11, 4, 5, 6, 7, 12, 13, 14, 15};
+      uint8x16_t perm = vld1q_u8(perm0);
+      uint8x16_t d;
+      for (i = 0; i < n_symbs >> 2; i++) {
+        d = vqtbl1q_u8(((uint8x16_t *)mod_symbs[0])[i], perm);
+        ((int64_t *)tx_layers[0])[i] = vgetq_lane_u64((uint64x2_t)d, 0);
+        ((int64_t *)tx_layers[1])[i] = vgetq_lane_u64((uint64x2_t)d, 1);
+      }
+      if (i << 2 != n_symbs) {
+        for (int i2 = ((n_symbs >> 3) << 3); i2 < n_symbs; i2 += 2) {
+          tx_layers[0][i2 >> 1] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 1] = mod_symbs[0][i2 + 1];
+        }
+      }
+#else
+      for (i = 0; i < n_symbs; i += 2) {
+        tx_layers[0][i >> 1] = mod_symbs[0][i];
+        tx_layers[1][i >> 1] = mod_symbs[0][i + 1];
+      }
+#endif
+#else
+      int i;
+      simde__m256i perm2 = simde_mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
+      simde__m256i d;
+      for (i = 0; i < n_symbs >> 3; i++) {
+        d = simde_mm256_permutevar8x32_epi32(((simde__m256i *)mod_symbs[0])[i], perm2);
+        ((simde__m128i *)tx_layers[0])[i] = simde_mm256_extractf128_si256(d, 0);
+        ((simde__m128i *)tx_layers[1])[i] = simde_mm256_extractf128_si256(d, 1);
+      }
+      if (i << 3 != n_symbs) {
+        for (int i2 = ((n_symbs >> 3) << 3); i2 < n_symbs; i2 += 2) {
+          tx_layers[0][i2 >> 1] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 1] = mod_symbs[0][i2 + 1];
+        }
+      }
+#endif
+    } break;
+    case 3: {
+#if defined(__AVX512BW__)
+      __m512i i0, i1, i2, d0;
+      __m512i perm3_0 =
+          _mm512_set_epi32(13 + 16, 10 + 16, 7 + 16, 4 + 16, 1 + 16, 14 + 16, 11 + 16, 8 + 16, 5 + 16, 2 + 16, 15, 12, 9, 6, 3, 0);
+      __m512i perm3_0b = _mm512_set_epi32(13 + 16, 10 + 16, 7 + 16, 4 + 16, 1 + 16, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+      __m512i perm3_1 = _mm512_set_epi32(14 + 16,
+                                         11 + 16,
+                                         8 + 16,
+                                         5 + 16,
+                                         2 + 16,
+                                         15 + 16,
+                                         12 + 16,
+                                         9 + 16,
+                                         6 + 16,
+                                         3 + 16,
+                                         0 + 16,
+                                         13,
+                                         10,
+                                         7,
+                                         4,
+                                         1);
+      __m512i perm3_1b = _mm512_set_epi32(14 + 16, 11 + 16, 8 + 16, 5 + 16, 2 + 16, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+      __m512i perm3_2 = _mm512_set_epi32(15 + 16,
+                                         12 + 16,
+                                         9 + 16,
+                                         6 + 16,
+                                         3 + 16,
+                                         0 + 16,
+                                         13 + 16,
+                                         10 + 16,
+                                         7 + 16,
+                                         4 + 16,
+                                         1 + 16,
+                                         14,
+                                         11,
+                                         8,
+                                         5,
+                                         2);
+      __m512i perm3_2b = _mm512_set_epi32(15 + 16, 12 + 16, 9 + 16, 6 + 16, 3 + 16, 0 + 16, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+      int i,n;
+      for (i = 0, n = 0; i < n_symbs >> 4; i += 3, n++) {
+        i0 = ((__m512i *)mod_symbs[0])[i];
+        i1 = ((__m512i *)mod_symbs[0])[i + 1];
+        i2 = ((__m512i *)mod_symbs[0])[i + 2];
+        d0 = _mm512_permutex2var_epi32(i0, perm3_0, i1);
+        ((__m512i *)tx_layers[0])[n] = _mm512_permutex2var_epi32(d0, perm3_0b, i2); // 11000000
+        d0 = _mm512_permutex2var_epi32(i0, perm3_1, i1);
+        ((__m512i *)tx_layers[1])[n] = _mm512_permutex2var_epi32(d0, perm3_1b, i2); // 11000000
+        d0 = _mm512_permutex2var_epi32(i0, perm3_2, i1);
+        ((__m512i *)tx_layers[2])[n] = _mm512_permutex2var_epi32(d0, perm3_2b, i2); // 11000000
+      }
+#elif defined(__aarch64__)
+      for (int i = 0; i < n_symbs; i += 3) {
+        tx_layers[0][i / 3] = mod_symbs[0][i];
+        tx_layers[1][i / 3] = mod_symbs[0][i + 1];
+        tx_layers[2][i / 3] = mod_symbs[0][i + 2];
+      }
+
+#else
+      simde__m256i i0, i1, i2, d0, d1, d2, d3;
+      simde__m256i perm3_0 = simde_mm256_set_epi32(5, 2, 7, 4, 1, 6, 3, 0);
+      simde__m256i perm3_1 = simde_mm256_set_epi32(6, 3, 0, 5, 2, 7, 4, 1);
+      simde__m256i perm3_2 = simde_mm256_set_epi32(7, 4, 1, 6, 3, 0, 5, 2);
+      int i,n;
+      for (i = 0, n = 0; i < n_symbs >> 3; i += 3, n++) {
+        i0 = ((simde__m256i *)mod_symbs[0])[i];
+        i1 = ((simde__m256i *)mod_symbs[0])[i + 1];
+        i2 = ((simde__m256i *)mod_symbs[0])[i + 2];
+        d0 = simde_mm256_permutevar8x32_epi32(i0, perm3_0);
+        d1 = simde_mm256_permutevar8x32_epi32(i1, perm3_0);
+        d2 = simde_mm256_permutevar8x32_epi32(i2, perm3_0);
+        d3 = simde_mm256_blend_epi32(d0, d1, 0x38); // 00111000
+        ((simde__m256i *)tx_layers[0])[n] = simde_mm256_blend_epi32(d3, d2, 0xc0); // 11000000
+        d0 = simde_mm256_permutevar8x32_epi32(i0, perm3_1);
+        d1 = simde_mm256_permutevar8x32_epi32(i1, perm3_1);
+        d2 = simde_mm256_permutevar8x32_epi32(i2, perm3_1);
+        d3 = simde_mm256_blend_epi32(d0, d1, 0x18); // 00011000
+        ((simde__m256i *)tx_layers[1])[n] = simde_mm256_blend_epi32(d3, d2, 0xe0); // 11100000
+        d0 = simde_mm256_permutevar8x32_epi32(i0, perm3_2);
+        d1 = simde_mm256_permutevar8x32_epi32(i1, perm3_2);
+        d2 = simde_mm256_permutevar8x32_epi32(i2, perm3_2);
+        d3 = simde_mm256_blend_epi32(d0, d1, 0x1c); // 00011100
+        ((simde__m256i *)tx_layers[2])[n] = simde_mm256_blend_epi32(d3, d2, 0xe0); // 11100000
+#ifdef DEBUG_LAYER_MAPPING
+        printf("\nsymb %d/%d\n", i << 3, n_symbs);
+        printf(" layer 0:\t");
+        for (int j = 0; j < 8 * 6; j += 6) {
+          printf("%d %d ", ((int16_t *)&mod_symbs[0][i << 3])[j], ((int16_t *)&mod_symbs[0][i << 3])[j + 1]);
+        }
+        printf("\n layer 1:\t");
+        for (int j = 2; j < 8 * 6; j += 6) {
+          printf("%d %d ", ((int16_t *)&mod_symbs[0][i << 3])[j], ((int16_t *)&mod_symbs[0][i << 3])[j + 1]);
+        }
+        printf("\n layer 2:\t");
+        for (int j = 4; j < 8 * 6; j += 6) {
+          printf("%d %d ", ((int16_t *)&mod_symbs[0][i << 3])[j], ((int16_t *)&mod_symbs[0][i << 3])[j + 1]);
+        }
+        printf("\n Mapping layer 0:\t");
+        for (int j = 0; j < 16; j++) {
+          printf("%d ", ((int16_t *)&tx_layers[0][n << 3])[j]);
+        }
+        printf("\n Mapping layer 1:\t");
+        for (int j = 0; j < 16; j++) {
+          printf("%d ", ((int16_t *)&tx_layers[1][n << 3])[j]);
+        }
+        printf("\n Mapping layer 2:\t");
+        for (int j = 0; j < 16; j++) {
+          printf("%d ", ((int16_t *)&tx_layers[2][n << 3])[j]);
+        }
+#endif
+      }
+      if (i << 3 != n_symbs) {
+        for (int i2 = ((n_symbs >> 3) << 3); i2 < n_symbs; i2 += 3) {
+          tx_layers[0][i2 / 3] = mod_symbs[0][i2];
+          tx_layers[1][i2 / 3] = mod_symbs[0][i2 + 1];
+          tx_layers[2][i2 / 3] = mod_symbs[0][i2 + 2];
+        }
+      }
+#endif
+    } break;
+
+    case 4: {
+#if defined(__AVX512BW__)
+      __m512i perm4 = _mm512_set_epi32(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0);
+      __m512i e;
+      int i;
+      for (i = 0; i < n_symbs >> 4; i++) {
+        e = _mm512_permutexvar_epi32(perm4, ((__m512i *)mod_symbs[0])[i]);
+        ((__m128i *)tx_layers[0])[i] = _mm512_extracti64x2_epi64(e, 0);
+        ((__m128i *)tx_layers[1])[i] = _mm512_extracti64x2_epi64(e, 1);
+        ((__m128i *)tx_layers[2])[i] = _mm512_extracti64x2_epi64(e, 2);
+        ((__m128i *)tx_layers[3])[i] = _mm512_extracti64x2_epi64(e, 3);
+      }
+      if (i << 4 != n_symbs) {
+        for (int i2 = ((n_symbs >> 4) << 4); i2 < n_symbs; i2 += 4) {
+          tx_layers[0][i2 >> 2] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 2] = mod_symbs[0][i2 + 1];
+          tx_layers[2][i2 >> 2] = mod_symbs[0][i2 + 2];
+          tx_layers[3][i2 >> 2] = mod_symbs[0][i2 + 3];
+        }
+      }
+#elif defined(__aarch64__)
+      int i;
+#ifdef USE_NEON
+      // SIMDe doesn't handle this properly, gcc up to 14.2 neither
+      uint32x4_t d4;
+      for (i = 0; i < n_symbs >> 2; i++) {
+        d4 = ((uint32x4_t *)mod_symbs[0])[i];
+        ((uint32_t *)tx_layers[0])[i] = vgetq_lane_u32(d4, 0);
+        ((uint32_t *)tx_layers[1])[i] = vgetq_lane_u32(d4, 1);
+        ((uint32_t *)tx_layers[2])[i] = vgetq_lane_u32(d4, 0);
+        ((uint32_t *)tx_layers[3])[i] = vgetq_lane_u32(d4, 1);
+      }
+      if (i << 2 != n_symbs) {
+        for (int i2 = ((n_symbs >> 2) << 2); i2 < n_symbs; i2 += 4) {
+          tx_layers[0][i2 >> 2] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 2] = mod_symbs[0][i2 + 1];
+          tx_layers[0][i2 >> 2] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 2] = mod_symbs[0][i2 + 1];
+        }
+      }
+#else
+      for (int i = 0; i < n_symbs; i += 4) {
+        tx_layers[0][i >> 2] = mod_symbs[0][i];
+        tx_layers[1][i >> 2] = mod_symbs[0][i + 1];
+        tx_layers[2][i >> 2] = mod_symbs[0][i + 2];
+        tx_layers[3][i >> 2] = mod_symbs[0][i + 3];
+      }
+#endif
+#else
+      simde__m256i perm4 = simde_mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
+      simde__m256i e;
+      int i;
+      for (i = 0; i < n_symbs >> 3; i++) {
+        e = simde_mm256_permutevar8x32_epi32(((simde__m256i *)mod_symbs[0])[i], perm4);
+        ((uint64_t *)tx_layers[0])[i] = simde_mm256_extract_epi64(e, 0);
+        ((uint64_t *)tx_layers[1])[i] = simde_mm256_extract_epi64(e, 1);
+        ((uint64_t *)tx_layers[2])[i] = simde_mm256_extract_epi64(e, 2);
+        ((uint64_t *)tx_layers[3])[i] = simde_mm256_extract_epi64(e, 3);
+      }
+      if (i << 3 != n_symbs) {
+        for (int i2 = ((n_symbs >> 3) << 3); i2 < n_symbs; i2 += 3) {
+          tx_layers[0][i2 >> 1] = mod_symbs[0][i2];
+          tx_layers[1][i2 >> 1] = mod_symbs[0][i2 + 1];
+          tx_layers[2][i2 >> 1] = mod_symbs[0][i2 + 2];
+          tx_layers[3][i2 >> 1] = mod_symbs[0][i2 + 2];
+        }
+      }
+#endif
+    } break;
 
     case 5:
-      if (layer < 2)
-        for (int i = 0; i < n_symbs; i += 2) {
+    case 6:
+    case 7:
+    case 8:
+      /*
+      // Layer 0,1
+      for (int i = 0; i < n_symbs; i += 2) {
           const int txIdx = i / 2;
-          tx_layer[txIdx] = mod_symbs[0][i + layer];
-        }
+          tx_layer[0][txIdx] = mod_symbs[0][i];
+          tx_layer[1][txIdx] = mod_symbs[0][i + 1];
+      }
+      // layers 2,3,4
       else
         for (int i = 0; i < n_symbs; i += 3) {
           const int txIdx = i / 3;
-          tx_layer[txIdx] = mod_symbs[1][i + layer];
+          tx_layer[2][txIdx] = mod_symbs[1][i + 2];
+          tx_layer[3][txIdx] = mod_symbs[1][i + 3];
+          tx_layer[4][txIdx] = mod_symbs[1][i + 4];
         }
       break;
 
@@ -288,7 +546,12 @@ void nr_layer_mapping(int nbCodes,
       for (int q=0; q<2; q++)
         for (int i = 0; i < n_symbs; i += 3) {
           const int txIdx = i / 3;
-          tx_layer[txIdx] = mod_symbs[q][i + layer];
+          tx_layer[0][txIdx] = mod_symbs[q][i + layer];
+          tx_layer[1][txIdx] = mod_symbs[q][i + layer];
+          tx_layer[2][txIdx] = mod_symbs[q][i + layer];
+          tx_layer[3][txIdx] = mod_symbs[q][i + layer];
+          tx_layer[4][txIdx] = mod_symbs[q][i + layer];
+          tx_layer[5][txIdx] = mod_symbs[q][i + layer];
         }
       break;
 
@@ -312,7 +575,7 @@ void nr_layer_mapping(int nbCodes,
         tx_layer[txIdx] = mod_symbs[q][i + layer];
       }
       break;
-
+*/
     default:
       AssertFatal(0, "Invalid number of layers %d\n", n_layers);
   }
@@ -337,7 +600,7 @@ void nr_dft(c16_t *z, c16_t *d, uint32_t Msc_PUSCH)
   simde__m128i norm128;
 
   if ((Msc_PUSCH % 1536) > 0) {
-    for (i = 0, ip = 0; i < Msc_PUSCH; i++, ip+=4) {
+    for (i = 0, ip = 0; i < Msc_PUSCH; i++, ip += 4) {
       dft_in0[ip] = d[i];
     }
   }
@@ -357,19 +620,18 @@ void nr_dft(c16_t *z, c16_t *d, uint32_t Msc_PUSCH)
   }
 
   if ((Msc_PUSCH % 1536) > 0) {
-    for (i = 0, ip = 0; i < Msc_PUSCH; i++, ip+=4)
+    for (i = 0, ip = 0; i < Msc_PUSCH; i++, ip += 4)
       z[i] = dft_out0[ip];
   }
-
 }
 
 void perform_symbol_rotation(NR_DL_FRAME_PARMS *fp, double f0, c16_t *symbol_rotation)
 {
-  const int nsymb = fp->symbols_per_slot * fp->slots_per_frame/10;
-  const double Tc=(1/480e3/4096);
-  const double Nu=2048*64*(1/(float)(1<<fp->numerology_index));
-  const double Ncp0=16*64 + (144*64*(1/(float)(1<<fp->numerology_index)));
-  const double Ncp1=(144*64*(1/(float)(1<<fp->numerology_index)));
+  const int nsymb = fp->symbols_per_slot * fp->slots_per_frame / 10;
+  const double Tc = (1 / 480e3 / 4096);
+  const double Nu = 2048 * 64 * (1 / (float)(1 << fp->numerology_index));
+  const double Ncp0 = 16 * 64 + (144 * 64 * (1 / (float)(1 << fp->numerology_index)));
+  const double Ncp1 = (144 * 64 * (1 / (float)(1 << fp->numerology_index)));
 
   LOG_D(PHY, "Doing symbol rotation calculation for TX/RX, f0 %f Hz, Nsymb %d\n", f0, nsymb);
 
@@ -430,7 +692,9 @@ void init_timeshift_rotation(NR_DL_FRAME_PARMS *fp)
     fp->timeshift_symbol_rotation[i].i = (int16_t)round(exp_im * 32767);
 
     if (i < 10)
-      LOG_D(PHY,"Timeshift symbol rotation %d => (%d,%d) %f\n",i,
+      LOG_D(PHY,
+            "Timeshift symbol rotation %d => (%d,%d) %f\n",
+            i,
             fp->timeshift_symbol_rotation[i].r,
             fp->timeshift_symbol_rotation[i].i,
             poff);
@@ -441,13 +705,13 @@ c16_t nr_layer_precoder(int sz, c16_t datatx_F_precoding[][sz], const char *prec
 {
   c16_t precodatatx_F = {0};
 
-  for (int al = 0; al<n_layers; al++) {
+  for (int al = 0; al < n_layers; al++) {
     c16_t antenna = datatx_F_precoding[al][re_offset];
     switch (prec_matrix[al]) {
-      case '0': //multiply by zero
+      case '0': // multiply by zero
         break;
 
-      case '1': //multiply by 1
+      case '1': // multiply by 1
         precodatatx_F = c16add(precodatatx_F, antenna);
         break;
 
@@ -492,15 +756,15 @@ c16_t nr_layer_precoder_cm(int n_layers,
 }
 
 void nr_layer_precoder_simd(const int n_layers,
-                           const int n_symbols,
-                           const int symSz,
-                           const c16_t txdataF_res_mapped[n_layers][n_symbols][symSz],
-                           const int ant,
-                           const nfapi_nr_pm_pdu_t *pmi_pdu,
-                           const int symbol,
-                           const int sc_offset,
-                           const int re_cnt,
-                           c16_t *txdataF_precoded)
+                            const int n_symbols,
+                            const int symSz,
+                            const c16_t txdataF_res_mapped[n_layers][n_symbols][symSz],
+                            const int ant,
+                            const nfapi_nr_pm_pdu_t *pmi_pdu,
+                            const int symbol,
+                            const int sc_offset,
+                            const int re_cnt,
+                            c16_t *txdataF_precoded)
 {
   uint32_t sc = sc_offset;
   c16_t prec_weight = {0};
@@ -510,22 +774,25 @@ void nr_layer_precoder_simd(const int n_layers,
   // 256 SIMD: Do 8 RE in one iteration, 3 iterations for 2 RB
 #ifdef __AVX2__
   const uint32_t re_cnt_align8 = re_cnt & ~7;
-  for(; sc < sc_offset + (re_cnt_align8); sc += sizeof(simde__m256i) / sizeof(prec_weight)) {
+  for (; sc < sc_offset + (re_cnt_align8); sc += sizeof(simde__m256i) / sizeof(prec_weight)) {
     // Matrix multiplication for 4 elements of the result (sizeof(simde__m256i) / sizeof(*prec_matrix) = 8)
     simde__m256i y = simde_mm256_set1_epi16(0); // Y = W[0]*X[0] + W[1]*X[1] + ... + W[nrOfLayers-1]*X[nrOfLayers-1]
-    for(int nl = 0; nl < n_layers; nl++) {
+    for (int nl = 0; nl < n_layers; nl++) {
       prec_weight.r = pmi_pdu->weights[nl][ant].precoder_weight_Re;
       prec_weight.i = pmi_pdu->weights[nl][ant].precoder_weight_Im;
 
       const simde__m256i x = simde_mm256_loadu_epi32(&txdataF_res_mapped[nl][symbol][sc]);
 
       // Rearrange precoding matrix weight to match complex multiplication and broadcast it to match SIMD size
-      const simde__m256i w_c   = simde_mm256_set1_epi32(c16toI32(c16conj(prec_weight)));   // broadcast conjugate of w
-      const simde__m256i w_s   = simde_mm256_set1_epi32(c16toI32(c16swap(prec_weight)));   // broadcast swapped real and img of w
+      const simde__m256i w_c = simde_mm256_set1_epi32(c16toI32(c16conj(prec_weight))); // broadcast conjugate of w
+      const simde__m256i w_s = simde_mm256_set1_epi32(c16toI32(c16swap(prec_weight))); // broadcast swapped real and img of w
 
       // Multiplication and shift
-      const simde__m256i reals = simde_mm256_srai_epi32(simde_mm256_madd_epi16(x, w_c), 15); // (int32_t) .r = (x.r * w.r - x.i * w.i) >> 15
-      const simde__m256i imags = simde_mm256_slli_epi32(simde_mm256_madd_epi16(x, w_s), 1);  // (int32_t) .i = (x.r * w.i + x.i * w.r) << 1, since higher 16 bit of each 32 bit is taken by blend_epi16
+      const simde__m256i reals =
+          simde_mm256_srai_epi32(simde_mm256_madd_epi16(x, w_c), 15); // (int32_t) .r = (x.r * w.r - x.i * w.i) >> 15
+      const simde__m256i imags = simde_mm256_slli_epi32(
+          simde_mm256_madd_epi16(x, w_s),
+          1); // (int32_t) .i = (x.r * w.i + x.i * w.r) << 1, since higher 16 bit of each 32 bit is taken by blend_epi16
 
       // Re-arrange to match c16_t format
       const simde__m256i produ = simde_mm256_blend_epi16(reals, imags, 0xAA);
@@ -540,43 +807,40 @@ void nr_layer_precoder_simd(const int n_layers,
 
   // 128 SIMD: Do 4 RE in one iteration, 3 iterations for 1 RB
   const uint32_t re_cnt_align4 = re_cnt & ~3;
-  for(; sc < sc_offset+re_cnt_align4; sc += sizeof(simde__m128i) / sizeof(prec_weight)) {
-    #ifdef DEBUG_DLSCH_PRECODING_PRINT_WITH_TRIVIAL // Get result with trivial solution, TODO: To be removed
-      c16_t y_triv[4];
-      for(int i = 0; i < 4; i++)
-        y_triv[i] = nr_layer_precoder_cm(n_layers,
-                                         NR_SYMBOLS_PER_SLOT,
-                                         symSz,
-                                         txdataF_res_mapped,
-                                         ant,
-                                         pmi_pdu,
-                                         symbol,
-                                         sc + i);
-      memcpy(&txdataF_precoded[sc], y_triv, sizeof(y_triv));
-    #endif
+  for (; sc < sc_offset + re_cnt_align4; sc += sizeof(simde__m128i) / sizeof(prec_weight)) {
+#ifdef DEBUG_DLSCH_PRECODING_PRINT_WITH_TRIVIAL // Get result with trivial solution, TODO: To be removed
+    c16_t y_triv[4];
+    for (int i = 0; i < 4; i++)
+      y_triv[i] = nr_layer_precoder_cm(n_layers, NR_SYMBOLS_PER_SLOT, symSz, txdataF_res_mapped, ant, pmi_pdu, symbol, sc + i);
+    memcpy(&txdataF_precoded[sc], y_triv, sizeof(y_triv));
+#endif
 
     // Matrix multiplication for 4 elements of the result (sizeof(simde__m128i) / sizeof(c16_t) = 4)
     simde__m128i y = simde_mm_set1_epi16(0); // Y = W[0]*X[0] + W[1]*X[1] + ... + W[nrOfLayers-1]*X[nrOfLayers-1]
-    for(int nl = 0; nl < n_layers; nl++) {
+    for (int nl = 0; nl < n_layers; nl++) {
       prec_weight.r = pmi_pdu->weights[nl][ant].precoder_weight_Re;
       prec_weight.i = pmi_pdu->weights[nl][ant].precoder_weight_Im;
 
       const simde__m128i x = simde_mm_loadu_epi32(&txdataF_res_mapped[nl][symbol][sc]);
 
       // Rearrange precoding matrix weight to match complex multiplication and broadcast it to match SIMD size
-      const simde__m128i w_c   = simde_mm_set1_epi32(c16toI32(c16conj(prec_weight)));   // broadcast conjugate of w
-      const simde__m128i w_s   = simde_mm_set1_epi32(c16toI32(c16swap(prec_weight)));   // broadcast swapped real and img of w
+      const simde__m128i w_c = simde_mm_set1_epi32(c16toI32(c16conj(prec_weight))); // broadcast conjugate of w
+      const simde__m128i w_s = simde_mm_set1_epi32(c16toI32(c16swap(prec_weight))); // broadcast swapped real and img of w
 
       // Multiplication and shift
-      const simde__m128i reals = simde_mm_srai_epi32(simde_mm_madd_epi16(x, w_c), 15); // (int32_t) .r = (x.r * w.r - x.i * w.i) >> 15
-      const simde__m128i imags = simde_mm_slli_epi32(simde_mm_madd_epi16(x, w_s), 1);  // (int32_t) .i = (x.r * w.i + x.i * w.r) << 1, since higher 16 bit of each 32 bit is taken by blend_epi16
+      const simde__m128i reals =
+          simde_mm_srai_epi32(simde_mm_madd_epi16(x, w_c), 15); // (int32_t) .r = (x.r * w.r - x.i * w.i) >> 15
+      const simde__m128i imags = simde_mm_slli_epi32(
+          simde_mm_madd_epi16(x, w_s),
+          1); // (int32_t) .i = (x.r * w.i + x.i * w.r) << 1, since higher 16 bit of each 32 bit is taken by blend_epi16
 
       /* Re-arrange to match c16_t format
-        bit index: 0            | 16              | 32           | 48              | 64           | 80              | 96           | 112
-        reals =   {R0.r[15..30] | R0.r[31] (0)*15 | R1.r[15..30] | R1.r[31] (0)*15 | R2.r[15..30] | R2.r[31] (0)*15 | R3.r[15..30] | R3.r[31] (0)*15}
-        imags =   {0 R0.i[0..14]| R0.i[15..30]    | 0 R1.i[0..14]| R1.i[15..30]    | 0 R2.i[0..14]| R2.i[15..30]    | 0 R3.i[0..14]| R3.i[15..30]   }
-        16b from  {reals        | imags           | reals        | imags           | reals        | imags           | reals        | imags          }
-        produ =   {R0.r[15..30] | R0.i[15..30]    | R1.r[15..30] | R1.i[15..30]    | R2.r[15..30] | R2.i[15..30]    | R3.r[15..30] | R3.i[15..30]   }
+        bit index: 0            | 16              | 32           | 48              | 64           | 80              | 96           |
+        112 reals =   {R0.r[15..30] | R0.r[31] (0)*15 | R1.r[15..30] | R1.r[31] (0)*15 | R2.r[15..30] | R2.r[31] (0)*15 |
+        R3.r[15..30] | R3.r[31] (0)*15} imags =   {0 R0.i[0..14]| R0.i[15..30]    | 0 R1.i[0..14]| R1.i[15..30]    | 0 R2.i[0..14]|
+        R2.i[15..30]    | 0 R3.i[0..14]| R3.i[15..30]   } 16b from  {reals        | imags           | reals        | imags | reals
+        | imags           | reals        | imags          } produ =   {R0.r[15..30] | R0.i[15..30]    | R1.r[15..30] | R1.i[15..30]
+        | R2.r[15..30] | R2.i[15..30]    | R3.r[15..30] | R3.i[15..30]   }
       */
       const simde__m128i produ = simde_mm_blend_epi16(reals, imags, 0xAA);
 
@@ -586,30 +850,30 @@ void nr_layer_precoder_simd(const int n_layers,
     // Store the result to txdataF
     simde_mm_storeu_si128(&txdataF_precoded[sc], y);
 
-    #ifdef DEBUG_DLSCH_PRECODING_PRINT_WITH_TRIVIAL // Print simd and trivial result, TODO: To be removed
-      c16_t *y_simd = (c16_t*) &y;
-      printf("debug_to_be_removed re_cnt=%d, sc=%u, y_simd=(%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d)\n",
-             re_cnt,
-             sc,
-             y_simd[0].r,
-             y_simd[0].i,
-             y_simd[1].r,
-             y_simd[1].i,
-             y_simd[2].r,
-             y_simd[2].i,
-             y_simd[3].r,
-             y_simd[3].i);
-      printf("debug_to_be_removed re_cnt=%d, sc=%u, y_triv=(%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d)\n",
-             re_cnt,
-             sc,
-             y_triv[0].r,
-             y_triv[0].i,
-             y_triv[1].r,
-             y_triv[1].i,
-             y_triv[2].r,
-             y_triv[2].i,
-             y_triv[3].r,
-             y_triv[3].i);
-    #endif
+#ifdef DEBUG_DLSCH_PRECODING_PRINT_WITH_TRIVIAL // Print simd and trivial result, TODO: To be removed
+    c16_t *y_simd = (c16_t *)&y;
+    printf("debug_to_be_removed re_cnt=%d, sc=%u, y_simd=(%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d)\n",
+           re_cnt,
+           sc,
+           y_simd[0].r,
+           y_simd[0].i,
+           y_simd[1].r,
+           y_simd[1].i,
+           y_simd[2].r,
+           y_simd[2].i,
+           y_simd[3].r,
+           y_simd[3].i);
+    printf("debug_to_be_removed re_cnt=%d, sc=%u, y_triv=(%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d), (%+4d,%+4d)\n",
+           re_cnt,
+           sc,
+           y_triv[0].r,
+           y_triv[0].i,
+           y_triv[1].r,
+           y_triv[1].i,
+           y_triv[2].r,
+           y_triv[2].i,
+           y_triv[3].r,
+           y_triv[3].i);
+#endif
   }
 }

openair1/PHY/MODULATION/nr_modulation.h

@@ -60,8 +60,7 @@ void nr_layer_mapping(int nbCodes,
                       uint8_t n_layers,
                       int layerSz,
                       uint32_t n_symbs,
-                      c16_t tx_layers[layerSz],
-                      int l);
+                      c16_t tx_layers[][layerSz]);
 
 /*! \brief Perform NR layer mapping. TS 38.211 V15.4.0 subclause 7.3.1.3
   @param[in] ulsch_ue, double Pointer to NR_UE_ULSCH_t struct

openair1/PHY/NR_TRANSPORT/nr_dci.c

@@ -248,9 +248,11 @@ void nr_generate_dci_top(processingData_L1tx_t *msgTx, int slot, int txdataF_off
 {
   PHY_VARS_gNB *gNB = msgTx->gNB;
   NR_DL_FRAME_PARMS *frame_parms = &gNB->frame_parms;
+  start_meas(&gNB->dci_generation_stats);
   for (int i = 0; i < msgTx->num_ul_pdcch; i++)
     nr_generate_dci(msgTx->gNB, &msgTx->ul_pdcch_pdu[i].pdcch_pdu.pdcch_pdu_rel15, txdataF_offset, frame_parms, slot);
   for (int i = 0; i < msgTx->num_dl_pdcch; i++)
     nr_generate_dci(msgTx->gNB, &msgTx->pdcch_pdu[i].pdcch_pdu_rel15, txdataF_offset, frame_parms, slot);
+  stop_meas(&gNB->dci_generation_stats);
 }
 

openair1/PHY/NR_TRANSPORT/nr_dlsch_coding.c

@@ -176,6 +176,7 @@ static void ldpc8blocks(void *p)
 
     uint8_t e[E];
     bzero (e, E);
+    start_meas(impp->dlsch_rate_matching_stats);
     nr_rate_matching_ldpc(Tbslbrm,
                           impp->BG,
                           impp->Zc,
@@ -186,6 +187,7 @@ static void ldpc8blocks(void *p)
                           impp->K - impp->F - 2 * impp->Zc,
                           rel15->rvIndex[0],
                           E);
+    stop_meas(impp->dlsch_rate_matching_stats);
     if (impp->K - impp->F - 2 * impp->Zc > E) {
       LOG_E(PHY,
             "dlsch coding A %d  Kr %d G %d (nb_rb %d, nb_symb_sch %d, nb_re_dmrs %d, length_dmrs %d, mod_order %d)\n",
@@ -217,10 +219,12 @@ static void ldpc8blocks(void *p)
       printf("output ratematching e[%d]= %d r_offset %u\n", i,e[i], r_offset);
 
 #endif
+    start_meas(impp->dlsch_interleaving_stats);
     nr_interleaving_ldpc(E,
                          mod_order,
                          e,
                          impp->output+r_offset);
+    stop_meas(impp->dlsch_interleaving_stats);
 #ifdef DEBUG_DLSCH_CODING
 
     for (int i =0; i<16; i++)
@@ -377,7 +381,8 @@ int nr_dlsch_encoding(PHY_VARS_gNB *gNB,
     encoder_implemparams_t arr[n_seg];
     task_ans_t ans[n_seg];
     memset(ans, 0, n_seg * sizeof(task_ans_t));
-
+    impp.dlsch_rate_matching_stats = dlsch_rate_matching_stats;
+    impp.dlsch_interleaving_stats = dlsch_interleaving_stats;
     for (int j = 0; j < n_seg; j++) {
       encoder_implemparams_t *perJobImpp = &arr[j];
       *perJobImpp = impp;

openair1/PHY/defs_gNB.h

@@ -535,6 +535,8 @@ typedef struct PHY_VARS_gNB_s {
   time_stats_t dlsch_interleaving_stats;
   time_stats_t dlsch_segmentation_stats;
 
+  time_stats_t dci_generation_stats;
+  time_stats_t phase_comp_stats;
   time_stats_t rx_pusch_stats;
   time_stats_t rx_pusch_init_stats;
   time_stats_t rx_pusch_symbol_processing_stats;

openair1/SCHED_NR/phy_procedures_nr_gNB.c

@@ -286,6 +286,7 @@ void phy_procedures_gNB_TX(processingData_L1tx_t *msgTx,
 
   //apply the OFDM symbol rotation here
   if (gNB->phase_comp) {
+    start_meas(&gNB->phase_comp_stats);
     for(int i = 0; i < gNB->common_vars.num_beams_period; ++i) {
       for (int aa = 0; aa < cfg->carrier_config.num_tx_ant.value; aa++) {
         apply_nr_rotation_TX(fp,
@@ -300,6 +301,7 @@ void phy_procedures_gNB_TX(processingData_L1tx_t *msgTx,
           T_INT(aa), T_BUFFER(&gNB->common_vars.txdataF[aa][txdataF_offset], fp->samples_per_slot_wCP*sizeof(int32_t)));
       }
     }
+    stop_meas(&gNB->phase_comp_stats);
   }
 
   VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_PHY_PROCEDURES_gNB_TX + gNB->CC_id, 0);

openair1/SIMULATION/NR_PHY/dlsim.c

@@ -969,10 +969,12 @@ int main(int argc, char **argv)
     reset_meas(&gNB->dlsch_segmentation_stats);
     reset_meas(&gNB->dlsch_modulation_stats);
     reset_meas(&gNB->dlsch_encoding_stats);
+    reset_meas(&gNB->dci_generation_stats);
     reset_meas(&gNB->tinput);
     reset_meas(&gNB->tprep);
     reset_meas(&gNB->tparity);
     reset_meas(&gNB->toutput);
+    reset_meas(&gNB->phase_comp_stats);
 
     uint32_t errors_scrambling[16] = {0};
     int n_errors[16] = {0};
@@ -1270,6 +1272,7 @@ int main(int argc, char **argv)
              UE->dl_harq_processes[0][slot].C,
              msgDataTx->dlsch[0][0].harq_process.pdsch_pdu.pdsch_pdu_rel15.TBSize[0] << 3);
       printDistribution(&gNB->phy_proc_tx,table_tx,"PHY proc tx");
+      printStatIndent2(&gNB->dci_generation_stats,"DCI encoding time");
       printStatIndent2(&gNB->dlsch_encoding_stats,"DLSCH encoding time");
       printStatIndent3(&gNB->dlsch_segmentation_stats,"DLSCH segmentation time");
       printStatIndent3(&gNB->tinput,"DLSCH LDPC input processing time");
@@ -1282,6 +1285,7 @@ int main(int argc, char **argv)
       printStatIndent2(&gNB->dlsch_scrambling_stats, "DLSCH scrambling time");
       printStatIndent2(&gNB->dlsch_resource_mapping_stats, "DLSCH Resource Mapping time");
       printStatIndent2(&gNB->dlsch_precoding_stats,"DLSCH Layer Precoding time");
+      if (gNB->phase_comp) printStatIndent2(&gNB->phase_comp_stats,"Phase Compensation");
 
       printf("\nUE function statistics (per %d us slot)\n", 1000 >> *scc->ssbSubcarrierSpacing);
       for (int i = RX_PDSCH_STATS; i <= DLSCH_PROCEDURES_STATS; i++) {

openair2/LAYER2/nr_rlc/nr_rlc_entity.c

@@ -100,6 +100,8 @@ nr_rlc_entity_t *new_nr_rlc_entity_am(
     exit(1);
   }
 
+  tx_maxsize *= 100;
+  rx_maxsize *= 100;
   ret->tx_maxsize = tx_maxsize;
   ret->rx_maxsize = rx_maxsize;
 

radio/fhi_72/oran-config.c

@@ -452,7 +452,7 @@ static uint64_t get_u64_mask(const paramdef_t *pd)
   for (int i = 0; i < pd->numelt; ++i) {
     int num = pd->iptr[i];
     AssertFatal(num >= 0 && num < 64, "cannot put element of %d in 64-bit mask\n", num);
-    mask |= 1 << num;
+    mask |= 1LL << num;
   }
   return mask;
 }