using thread pool to parallize the precoding and ifft for each phy antenna in dlsim

cmake_targets/CMakeLists.txt

@@ -1025,6 +1025,7 @@ set(PHY_SRC
   ${OPENAIR1_DIR}/PHY/TOOLS/sqrt.c
   ${OPENAIR1_DIR}/PHY/TOOLS/time_meas.c
   ${OPENAIR1_DIR}/PHY/TOOLS/lut.c
+  ${OPENAIR1_DIR}/PHY/TOOLS/thread_pool.c
   )
 if (${SMBV})
   set(PHY_SRC "${PHY_SRC} ${OPENAIR1_DIR}/PHY/TOOLS/smbv.c")
@@ -1840,7 +1841,7 @@ target_link_libraries (oaisim_nos1 ${T_LIB})
 # Unitary tests for each piece of L1: example, mbmssim is MBMS L1 simulator
 #####################################
 
-foreach(myExe dlsim dlsim_tm7 ulsim pbchsim scansim mbmssim pdcchsim pucchsim prachsim syncsim)
+foreach(myExe dlsim ulsim pbchsim scansim mbmssim pdcchsim pucchsim prachsim syncsim)
   add_executable(${myExe}
     ${OPENAIR_BIN_DIR}/messages_xml.h
     ${OPENAIR1_DIR}/SIMULATION/LTE_PHY/${myExe}.c
@@ -1921,7 +1922,6 @@ if (${T_TRACER})
   add_dependencies(oaisim generate_T)
   add_dependencies(oaisim_nos1 generate_T)
   add_dependencies(dlsim generate_T)
-  add_dependencies(dlsim_tm7 generate_T)
   add_dependencies(ulsim generate_T)
   add_dependencies(pbchsim generate_T)
   add_dependencies(scansim generate_T)

openair1/PHY/INIT/lte_init.c

@@ -22,6 +22,8 @@
 #include "defs.h"
 #include "SCHED/defs.h"
 #include "PHY/extern.h"
+#include "PHY/TOOLS/thread_pool.h"
+#include "PHY/MODULATION/defs.h"
 #include "SIMULATION/TOOLS/defs.h"
 #include "RadioResourceConfigCommonSIB.h"
 #include "RadioResourceConfigDedicated.h"
@@ -1308,8 +1310,10 @@ int phy_init_lte_eNB(PHY_VARS_eNB *eNB,
       eNB->sinr_dB = (double*) malloc16_clear( fp->N_RB_DL*12*sizeof(double) );
     }
   } //eNB_id
-  
-  
+
+  /* Create thread pool */
+  eNB->pool = new_thread_pool(do_OFDM_mod_thread, eNB);
+  sleep(1);
   
   if (abstraction_flag==0) {
     if (eNB->node_function != NGFI_RRU_IF4p5) {

openair1/PHY/MODULATION/beamforming.c

@@ -64,16 +64,16 @@ int beam_precoding(int32_t **txdataF,
   memset(txdataF_BF[aa],0,sizeof(int32_t)*(frame_parms->ofdm_symbol_size));
 
   for (p=0; p<14; p++) {
-    //if (p==0 || p==1 || p==5 || p>7)
-    //  mult_cpx_conj_vector((int16_t*)txdataF[p], (int16_t*)beam_weights[p][aa], (int16_t*)txdataF_BF[aa], frame_parms->ofdm_symbol_size, 15);
+    //if (p==0 || p==1 || p==4 || p==5 || p>=7)
+      //multadd_cpx_vector((int16_t*)txdataF[p],(int16_t*)beam_weights[p][aa], (int16_t*)txdataF_BF[aa], 0, frame_parms->ofdm_symbol_size, 15);
     for (re=0;re<frame_parms->ofdm_symbol_size;re++) {
-      if ((p==0 || p==1 || p==5 || p>=7) && txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re]!=0) {
+      if ((p==0 || p==1 || p==4 || p==5 || p>=7) && txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re]!=0) {
         ((int16_t*)&txdataF_BF[aa][re])[0] += (int16_t)((((int16_t*)&txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re])[0]*((int16_t*)&beam_weights[p][aa][re])[0])>>15);
         ((int16_t*)&txdataF_BF[aa][re])[0] -= (int16_t)((((int16_t*)&txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re])[1]*((int16_t*)&beam_weights[p][aa][re])[1])>>15);
         ((int16_t*)&txdataF_BF[aa][re])[1] += (int16_t)((((int16_t*)&txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re])[0]*((int16_t*)&beam_weights[p][aa][re])[1])>>15);
         ((int16_t*)&txdataF_BF[aa][re])[1] += (int16_t)((((int16_t*)&txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re])[1]*((int16_t*)&beam_weights[p][aa][re])[0])>>15);
 
-	  /*
+/*
           printf("beamforming.c:txdataF[%d][%d]=%d+j%d, beam_weights[%d][%d][%d]=%d+j%d,txdata_BF[%d][%d]=%d+j%d\n",
                  p,slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re,
                  ((int16_t*)&txdataF[p][slot_offset_F+symbol*frame_parms->ofdm_symbol_size+re])[0],
@@ -83,7 +83,7 @@ int beam_precoding(int32_t **txdataF,
                  aa,re,
                  ((int16_t*)&txdataF_BF[aa][re])[0],
                  ((int16_t*)&txdataF_BF[aa][re])[1]); 
-	  */
+*/
       } 
     }
   }

openair1/PHY/MODULATION/defs.h

@@ -46,6 +46,14 @@ void PHY_ofdm_mod(int *input,
                   Extension_t etype
                  );
 
+/**
+\brief This function is a thread performing beamforming precoding and
+ifft for each physical antenna
+
+@param eNB The eNB context
+*/
+void *do_OFDM_mod_thread(PHY_VARS_eNB *eNB);
+
 #ifdef OPENAIR_LTE
 
 /*!
@@ -82,7 +90,7 @@ void normal_prefix_mod(int32_t *txdataF,int32_t *txdata,uint8_t nsymb,LTE_DL_FRA
 
 void do_OFDM_mod(int32_t **txdataF, int32_t **txdata, uint32_t frame,uint16_t next_slot, LTE_DL_FRAME_PARMS *frame_parms);
 
-void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t next_slot, LTE_DL_FRAME_PARMS *frame_parms);
+void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t next_slot, LTE_DL_FRAME_PARMS *frame_parms, int antenna);
 
 void remove_7_5_kHz(PHY_VARS_eNB *phy_vars_eNB,uint8_t subframe);
 

openair1/PHY/MODULATION/ofdm_mod.c

@@ -29,9 +29,11 @@ This section deals with basic functions for OFDM Modulation.
 
 */
 
+#include <sys/syscall.h> 
 #include "PHY/defs.h"
 #include "UTIL/LOG/log.h"
 #include "UTIL/LOG/vcd_signal_dumper.h"
+#include "PHY/TOOLS/thread_pool.h"
 
 //static short temp2[2048*4] __attribute__((aligned(16)));
 
@@ -285,7 +287,7 @@ void do_OFDM_mod(int32_t **txdataF, int32_t **txdata, uint32_t frame,uint16_t ne
 }
 
 // OFDM modulation for each symbol
-void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t next_slot, LTE_DL_FRAME_PARMS *frame_parms)
+void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t next_slot, LTE_DL_FRAME_PARMS *frame_parms, int antenna)
 {
 
   int aa, l, slot_offset;
@@ -293,19 +295,80 @@ void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t ne
   int32_t **txdataF_BF = eNB_common_vars->txdataF_BF[eNB_id];
   int32_t **txdata = eNB_common_vars->txdata[eNB_id];
 
+  aa = antenna;
   slot_offset = (next_slot)*(frame_parms->samples_per_tti>>1);
 
   //printf("Thread %d starting ... aa %d (%llu)\n",omp_get_thread_num(),aa,rdtsc());
+
+
   for (l=0; l<frame_parms->symbols_per_tti>>1; l++) {
+
+    //printf("do_OFDM_mod_l, slot=%d, l=%d, NUMBER_OF_OFDM_CARRIERS=%d,OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES=%d\n",next_slot, l,NUMBER_OF_OFDM_CARRIERS,OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES);
+    VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,1);
+    beam_precoding(txdataF,txdataF_BF,frame_parms,eNB_common_vars->beam_weights[eNB_id],next_slot,l,aa);
+    VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,0);
+
+    //PMCH case not implemented...
+
+    if (frame_parms->Ncp == 1)
+      PHY_ofdm_mod(txdataF_BF[aa],         // input
+                   &txdata[aa][slot_offset+l*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES],            // output
+                   frame_parms->ofdm_symbol_size,
+                   1,                                   // number of symbols
+                   frame_parms->nb_prefix_samples,      // number of prefix samples
+                   CYCLIC_PREFIX);
+    else {
+      if (l==0) {
+        PHY_ofdm_mod(txdataF_BF[aa],        // input
+                     &txdata[aa][slot_offset],           // output
+                     frame_parms->ofdm_symbol_size,
+                     1,                                  // number of symbols
+                     frame_parms->nb_prefix_samples0,    // number of prefix samples
+                     CYCLIC_PREFIX);
+      } else {
+        PHY_ofdm_mod(txdataF_BF[aa],        // input
+                     &txdata[aa][slot_offset+OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES0+(l-1)*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES],           // output
+                     frame_parms->ofdm_symbol_size,
+                     1,                                  // number of symbols
+                     frame_parms->nb_prefix_samples,     // number of prefix samples
+                     CYCLIC_PREFIX);
+
+        /* printf("txdata[%d][%d]=%d\n",aa,slot_offset+OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES0+(l-1)*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES,txdata[aa][slot_offset+OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES0+(l-1)*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES]);
+ * */
+      }
+    }
+  }
+
+}
+
+// OFDM modulation thread
+void *do_OFDM_mod_thread(PHY_VARS_eNB *eNB)
+{
+  thread_pool *pool = eNB->pool;
+  int eNB_id = 0;
+  LTE_DL_FRAME_PARMS *frame_parms = &eNB->frame_parms;
+
+  int32_t **txdataF = eNB->common_vars.txdataF[eNB_id];
+  int32_t **txdataF_BF = eNB->common_vars.txdataF_BF[eNB_id];
+  int32_t **txdata = eNB->common_vars.txdata[eNB_id];
+  int aa, l, slot_offset;
+
+
+  //printf("Thread %d starting ... aa %d (%llu)\n",omp_get_thread_num(),aa,rdtsc());
+
+  while(1) {
+
+    aa = thread_pool_wait(pool);
+    slot_offset = (pool->next_slot)*(frame_parms->samples_per_tti>>1);
   
-    for (aa=0; aa<frame_parms->nb_antennas_tx; aa++) {
+    printf("OFDM mod thread %ld starts processing antenna %d slot %d\n", syscall(SYS_gettid), aa, pool->next_slot);
+    for (l=0; l<frame_parms->symbols_per_tti>>1; l++) {
 
-      //printf("do_OFDM_mod_l, slot=%d, l=%d, NUMBER_OF_OFDM_CARRIERS=%d,OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES=%d\n",next_slot, l,NUMBER_OF_OFDM_CARRIERS,OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES);
-      VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,1);
-      beam_precoding(txdataF,txdataF_BF,frame_parms,eNB_common_vars->beam_weights[eNB_id],next_slot,l,aa);
-      VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,0);
+      //VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,1);
+      beam_precoding(txdataF,txdataF_BF,frame_parms,eNB->common_vars.beam_weights[eNB_id],pool->next_slot,l,aa);
+      //VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_BEAM_PRECODING,0);
 
-      //PMCH case not implemented... 
+      //PMCH case not implemented...
 
       if (frame_parms->Ncp == 1)
         PHY_ofdm_mod(txdataF_BF[aa],         // input
@@ -331,11 +394,13 @@ void do_OFDM_mod_symbol(LTE_eNB_COMMON *eNB_common_vars, int eNB_id, uint16_t ne
                        frame_parms->nb_prefix_samples,     // number of prefix samples
                        CYCLIC_PREFIX);
 
-          /* printf("txdata[%d][%d]=%d\n",aa,slot_offset+OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES0+(l-1)*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES,txdata[aa][slot_offset+OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES0+(l-1)*OFDM_SYMBOL_SIZE_COMPLEX_SAMPLES]);
- * */
         }
       }
     }
-  }
 
+    printf("OFDM mod thread %ld finished processing antenna %d slot %d\n", syscall(SYS_gettid), aa, pool->next_slot);
+
+    /* signal end of computation to thread pool */
+    thread_pool_done(pool);
+  }
 }

openair1/PHY/defs.h

@@ -122,6 +122,7 @@ static inline void* malloc16_clear( size_t size )
 #include "PHY/TOOLS/time_meas.h"
 #include "PHY/CODING/defs.h"
 #include "PHY/TOOLS/defs.h"
+#include "PHY/TOOLS/thread_pool.h"
 #include "platform_types.h"
 
 #ifdef OPENAIR_LTE
@@ -411,6 +412,7 @@ typedef struct PHY_VARS_eNB_s {
   eNB_func_t           node_function;
   eNB_timing_t         node_timing;
   int                  single_thread_flag;
+  thread_pool          *pool;
   openair0_rf_map      rf_map;
   int                  abstraction_flag;
   void                 (*do_prach)(struct PHY_VARS_eNB_s *eNB);

openair1/SIMULATION/LTE_PHY/dlsim.c

@@ -53,6 +53,7 @@
 
 
 #include "PHY/TOOLS/lte_phy_scope.h"
+#include "PHY/TOOLS/thread_pool.h"
 
 PHY_VARS_eNB *eNB;
 PHY_VARS_UE *UE;
@@ -1331,6 +1332,8 @@ int main(int argc, char **argv)
   //int len;
   uint8_t num_rounds = 4;//,fix_rounds=0;
 
+  int thread_pool_active = 0;
+
   int u;
   int n=0;
   int abstx=0;
@@ -1419,7 +1422,7 @@ int main(int argc, char **argv)
   //  num_layers = 1;
   perfect_ce = 0;
 
-  while ((c = getopt (argc, argv, "ahdpZDe:Em:n:o:s:f:t:c:g:r:F:x:q:y:z:AM:N:I:i:O:R:S:C:T:b:u:v:w:B:Pl:WXY")) != -1) {
+  while ((c = getopt (argc, argv, "ahdpZDe:Em:n:o:s:f:t:c:g:r:F:x:q:y:z:AM:N:I:i:O:R:S:C:T:b:u:v:w:B:Pl:WXYU")) != -1) {
     switch (c) {
     case 'a':
       awgn_flag = 1;
@@ -1486,6 +1489,7 @@ int main(int argc, char **argv)
     case 'W':
       two_thread_flag = 1;
       break;
+
     case 'l':
       offset_mumimo_llr_drange_fix=atoi(optarg);
       break;
@@ -1627,6 +1631,10 @@ int main(int argc, char **argv)
 
       break;
 
+    case 'U':
+      thread_pool_active = 1;
+      break;
+
     case 'v':
       i_mod = atoi(optarg);
 
@@ -1745,6 +1753,7 @@ int main(int argc, char **argv)
       printf("-O Set the percenatge of effective rate to testbench the modem performance (typically 30 and 70, range 1-100) \n");
       printf("-I Input filename for TrCH data (binary)\n");
       printf("-u Enables the Interference Aware Receiver for TM5 (default is normal receiver)\n");
+      printf("-U Enable thread pool for beam precoding and OFDM modulation\n");
       exit(1);
       break;
     }
@@ -2406,15 +2415,37 @@ int main(int argc, char **argv)
 			  &eNB->frame_parms);
 	    */
 
-            do_OFDM_mod_symbol(&eNB->common_vars,
-                               eNB_id,
-                               (subframe*2),
-                               &eNB->frame_parms);
+            if (thread_pool_active == 0) {
+              for (aa=0; aa<eNB->frame_parms.nb_antennas_tx; aa++) {
+                do_OFDM_mod_symbol(&eNB->common_vars,
+                                   eNB_id,
+                                   (subframe*2),
+                                   &eNB->frame_parms,
+                                   aa);
+
+                do_OFDM_mod_symbol(&eNB->common_vars,
+                                   eNB_id,
+                                   (subframe*2)+1,
+                                   &eNB->frame_parms,
+                                   aa);
+              }
+            } else {
+              eNB->pool->next_slot = 2*subframe;
+              //printf("dlsim:slot=%d\n", eNB->pool->next_slot);
+
+              while (eNB->pool->next_slot < subframe*2+2) {
+                /* start all threads */
+                thread_pool_start(eNB->pool);
 
-            do_OFDM_mod_symbol(&eNB->common_vars,
-                               eNB_id,
-                               (subframe*2)+1,
-                               &eNB->frame_parms);
+                ///* wait all threads finishes */
+                thread_pool_join(eNB->pool);
+                //usleep(1000*1000);
+
+                /* set next_slot */
+                eNB->pool->next_slot++;
+                //pool->next_slot %= 20;
+              }
+            }
 
 
 	    stop_meas(&eNB->ofdm_mod_stats);
@@ -2453,8 +2484,9 @@ int main(int argc, char **argv)
 	        write_output("txsigF0.m","txsF0", &eNB->common_vars.txdataF[eNB_id][0][subframe*nsymb*eNB->frame_parms.ofdm_symbol_size],nsymb*eNB->frame_parms.ofdm_symbol_size,1,1);
               } else if (transmission_mode == 7) {
                 write_output("txsigF0.m","txsF0", &eNB->common_vars.txdataF[eNB_id][5][subframe*nsymb*eNB->frame_parms.ofdm_symbol_size],nsymb*eNB->frame_parms.ofdm_symbol_size,1,1);
-                write_output("txsigF0_BF.m","txsF0_BF", &eNB->common_vars.txdataF_BF[eNB_id][0][0],eNB->frame_parms.ofdm_symbol_size,1,1);
               }
+
+              write_output("txsigF0_BF.m","txsF0_BF", &eNB->common_vars.txdataF_BF[eNB_id][0][0],eNB->frame_parms.ofdm_symbol_size,1,1);
             }
 	  }
 

targets/RT/USER/lte-enb.c

@@ -295,17 +295,21 @@ void do_OFDM_mod_rt(int subframe,PHY_VARS_eNB *phy_vars_eNB)
 
     VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_OFDM_MODULATION,1);
 
-    do_OFDM_mod_symbol(&phy_vars_eNB->common_vars,
-                  0,
-                  subframe<<1,
-                  &phy_vars_eNB->frame_parms);
- 
-    // if S-subframe generate first slot only 
-    if (subframe_select(&phy_vars_eNB->frame_parms,subframe) == SF_DL) {
+    for (aa=0; aa<phy_vars_eNB->frame_parms.nb_antennas_tx; aa++) {
       do_OFDM_mod_symbol(&phy_vars_eNB->common_vars,
-                    0,
-                    1+(subframe<<1),
-                    &phy_vars_eNB->frame_parms);
+                         0,
+                         subframe<<1,
+                         &phy_vars_eNB->frame_parms,
+                         aa);
+
+      // if S-subframe generate first slot only
+      if (subframe_select(&phy_vars_eNB->frame_parms,subframe) == SF_DL) {
+        do_OFDM_mod_symbol(&phy_vars_eNB->common_vars,
+                           0,
+                           1+(subframe<<1),
+                           &phy_vars_eNB->frame_parms,
+                           aa);
+      }
     }
 
     VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_ENB_OFDM_MODULATION,0);