new sincos256_ps, log256_ps and exp_256 function using AVX instructions

openair1/SIMULATION/TOOLS/defs.h

@@ -313,6 +313,17 @@ void multipath_channel_freq_SSE_float(channel_desc_t *desc,
 		       uint8_t UE_id,
 		       uint8_t CC_id,
 		       uint8_t th_id);
+void multipath_channel_freq_AVX_float(channel_desc_t *desc,
+                       float *tx_sig_re[2],
+                       float *tx_sig_im[2],
+                       float *rx_sig_re[2],
+                       float *rx_sig_im[2],
+                       uint32_t length,
+                       uint8_t keep_channel,
+		       uint8_t eNB_id,
+		       uint8_t UE_id,
+		       uint8_t CC_id,
+		       uint8_t th_id);
 /**\fn void multipath_channel_prach(channel_desc_t *desc,
            double tx_sig_re[2],
            double tx_sig_im[2],
@@ -474,7 +485,9 @@ void table_nor(unsigned long seed);
 double nfix(void);
 __m128 nfix_SSE(__m128i iz);
 __m128 log_ps(__m128 x);
+__m256 log256_ps(__m256 x);
 __m128 exp_ps(__m128 x);
+__m256 exp256_ps(__m256 x);
 double uniformrandom(void);
 void uniformrandomSSE(__m128d *d1,__m128d *d2);
 double ziggurat(double mean, double variance);
@@ -482,14 +495,17 @@ __m128 ziggurat_SSE_float(void);
 void boxmuller_SSE_float(__m128 *data1, __m128 *data2);
 int freq_channel(channel_desc_t *desc,uint16_t nb_rb, int16_t n_samples);
 int freq_channel_SSE_float(channel_desc_t *desc,uint16_t nb_rb, int16_t n_samples);
+int freq_channel_AVX_float(channel_desc_t *desc,uint16_t nb_rb, int16_t n_samples);
 int freq_channel_prach(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples,int16_t prach_fmt,int16_t n_ra_prb);
 int freq_channel_prach_SSE_float(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples,int16_t prach_fmt,int16_t n_ra_prb);
 int init_freq_channel(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples);
 int init_freq_channel_SSE_float(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples);
+int init_freq_channel_AVX_float(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples);
 int init_freq_channel_prach(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples,int16_t prach_fmt,int16_t n_ra_prb);
 int init_freq_channel_prach_SSE_float(channel_desc_t *desc,uint16_t nb_rb,int16_t n_samples,int16_t prach_fmt,int16_t n_ra_prb);
 
 void sincos_ps(__m128 x, __m128 *s, __m128 *c);
+void sincos256_ps(__m256 x, __m256 *s, __m256 *c);
 
 uint8_t multipath_channel_nosigconv(channel_desc_t *desc);
 void multipath_tv_channel(channel_desc_t *desc,

openair1/SIMULATION/TOOLS/multipath_channel.c

@@ -615,6 +615,127 @@ void multipath_channel_freq_SSE_float(channel_desc_t *desc,
 	} // f,f2,f3
   //}//k
 }
+void multipath_channel_freq_AVX_float(channel_desc_t *desc,
+                       float *tx_sig_re[2],
+                       float *tx_sig_im[2],
+                       float *rx_sig_re[2],
+                       float *rx_sig_im[2],
+                       uint32_t length,
+                       uint8_t keep_channel,
+		       uint8_t eNB_id,
+		       uint8_t UE_id,
+		       uint8_t CC_id,
+		       uint8_t th_id)
+{
+  int ii,j,f;
+  __m256 rx_tmp256_re_f,rx_tmp256_im_f,rx_tmp256_re,rx_tmp256_im, rx_tmp256_1,rx_tmp256_2,rx_tmp256_3,rx_tmp256_4,tx256_re,tx256_im,chF256_x,chF256_y,pathloss256;
+  //struct complex rx_tmp;
+
+  float path_loss = pow(10,desc->path_loss_dB/20);
+  pathloss256 = _mm256_set1_ps(path_loss);
+
+  int nb_rb, n_samples, ofdm_symbol_size, symbols_per_tti;
+  nb_rb=PHY_vars_UE_g[UE_id][CC_id]->frame_parms.N_RB_DL;
+  n_samples=PHY_vars_UE_g[UE_id][CC_id]->frame_parms.N_RB_DL*12+1;
+  ofdm_symbol_size=length/PHY_vars_UE_g[UE_id][CC_id]->frame_parms.symbols_per_tti;
+  symbols_per_tti=length/PHY_vars_UE_g[UE_id][CC_id]->frame_parms.ofdm_symbol_size;
+
+
+#ifdef DEBUG_CH
+  printf("[CHANNEL_FREQ] keep = %d : path_loss = %g (%f), nb_rx %d, nb_tx %d, dd %d, len %d \n",keep_channel,path_loss,desc->path_loss_dB,desc->nb_rx,desc->nb_tx,dd,desc->channel_length);
+#endif		
+
+  if (keep_channel) {
+  	// do nothing - keep channel
+  } else {
+        random_channel_freq(desc,0);
+  	freq_channel_SSE_float(desc,nb_rb,n_samples);//Find desc->chF
+  }
+	for (j=0;j<(symbols_per_tti>>2);j++){
+		for (ii=0; ii<desc->nb_rx; ii++) {
+			_mm256_storeu_ps(&rx_sig_re[ii][4*j*ofdm_symbol_size],_mm256_setzero_ps());
+			_mm256_storeu_ps(&rx_sig_im[ii][4*j*ofdm_symbol_size],_mm256_setzero_ps());
+		}
+	}
+	for (f=0;f<((ofdm_symbol_size*symbols_per_tti)>>3); f++) {//f2 = 0-1024*14-1 ---- for 10 Mhz BW
+		//printf("f is %d\n",f);
+		for (ii=0; ii<desc->nb_rx; ii++) {
+			//rx_tmp.x = 0;
+			//rx_tmp.y = 0;
+			rx_tmp256_re_f = _mm256_setzero_ps();
+      			rx_tmp256_im_f = _mm256_setzero_ps();
+			if (f%(ofdm_symbol_size>>2)<(n_samples>>2))//1-300
+			{
+				for (j=0; j<desc->nb_tx; j++) {	
+					//first n_samples>>1 samples of each frequency ofdm symbol out of ofdm_symbol_size
+					//RX_RE(k) += TX_RE(k).chF(k).x	- TX_IM(k).chF(k).y	
+					//RX_IM(k) += TX_IM(k).chF(k).x + TX_RE(k).chF(k).y
+					tx256_re = _mm256_loadu_ps(&tx_sig_re[j][(4*f+1)]);
+            				tx256_im = _mm256_loadu_ps(&tx_sig_im[j][(4*f+1)]);
+          				chF256_x = _mm256_set1_ps(desc->chFf[ii+(j*desc->nb_rx)].x[(4*(f%(ofdm_symbol_size>>2)))+(n_samples>>2)]);
+          				chF256_y = _mm256_set1_ps(desc->chFf[ii+(j*desc->nb_rx)].y[(4*(f%(ofdm_symbol_size>>2)))+(n_samples>>2)]);
+					//rx_tmp.x += (tx_sig_re[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f+(n_samples>>1)-1].x)//tx128_re*ch128_x
+					//	     -(tx_sig_im[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f+(n_samples>>1)-1].y);//-tx128_im*ch128_y
+					//rx_tmp.y += (tx_sig_im[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f+(n_samples>>1)-1].x)//tx128_im*ch128_x
+					//	     +(tx_sig_re[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f+(n_samples>>1)-1].y);//+tx128_re*ch128_y
+					rx_tmp256_1    = _mm256_mul_ps(tx256_re,chF256_x);
+					rx_tmp256_2    = _mm256_mul_ps(tx256_im,chF256_y);
+					rx_tmp256_3    = _mm256_mul_ps(tx256_im,chF256_x);
+					rx_tmp256_4    = _mm256_mul_ps(tx256_re,chF256_y);
+					rx_tmp256_re   = _mm256_sub_ps(rx_tmp256_1,rx_tmp256_2);
+					rx_tmp256_im   = _mm256_add_ps(rx_tmp256_3,rx_tmp256_4);
+					rx_tmp256_re_f = _mm256_add_ps(rx_tmp256_re_f,rx_tmp256_re);
+					rx_tmp256_im_f = _mm256_add_ps(rx_tmp256_im_f,rx_tmp256_im);
+				}  // j    
+				//rx_sig_re[ii][f+k*ofdm_symbol_size] =  rx_tmp.x*path_loss;
+				//rx_sig_im[ii][f+k*ofdm_symbol_size] =  rx_tmp.y*path_loss;
+				rx_tmp256_re_f = _mm256_mul_ps(rx_tmp256_re_f,pathloss256);
+			        rx_tmp256_im_f = _mm256_mul_ps(rx_tmp256_im_f,pathloss256);
+			        _mm256_storeu_ps(&rx_sig_re[ii][(4*f+1)],rx_tmp256_re_f);
+			        _mm256_storeu_ps(&rx_sig_im[ii][(4*f+1)],rx_tmp256_im_f);
+			}
+			else if (f%(ofdm_symbol_size>>2)>(n_samples>>2) && f%(ofdm_symbol_size>>2)<(ofdm_symbol_size>>2)-(n_samples>>2))
+			{
+				//rx_sig_re[ii][f+k*ofdm_symbol_size] =  0;
+				//rx_sig_im[ii][f+k*ofdm_symbol_size] =  0;
+			        //_mm_storeu_pd(&rx_sig_re[ii][2*f],_mm_setzero_pd());
+			        //_mm_storeu_pd(&rx_sig_im[ii][2*f],_mm_setzero_pd());
+				break;
+			}
+			else
+			{
+				for (j=0; j<desc->nb_tx; j++) {	
+					//last n_samples>>1 samples of each frequency ofdm symbol out of ofdm_symbol_size
+					//RX_RE(k) += TX_RE(k).chF(k).x - TX_IM(k).chF(k).y
+					//RX_IM(k) += TX_IM(k).chF(k).x + TX_RE(k).chF(k).y
+					tx256_re = _mm256_loadu_ps(&tx_sig_re[j][4*f]);
+            				tx256_im = _mm256_loadu_ps(&tx_sig_im[j][4*f]);
+          				chF256_x = _mm256_set1_ps(desc->chFf[ii+(j*desc->nb_rx)].x[4*(f%(ofdm_symbol_size>>2)-((ofdm_symbol_size>>2)-(n_samples>>2)))]);
+          				chF256_y = _mm256_set1_ps(desc->chFf[ii+(j*desc->nb_rx)].y[4*(f%(ofdm_symbol_size>>2)-((ofdm_symbol_size>>2)-(n_samples>>2)))]);
+					//rx_tmp.x += (tx_sig_re[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f2].x)
+					//	     -(tx_sig_im[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f2].y);
+					//rx_tmp.y += (tx_sig_im[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f2].x)
+					//	     +(tx_sig_re[j][f+k*ofdm_symbol_size] * desc->chF[ii+(j*desc->nb_rx)][f2].y);
+					rx_tmp256_1    = _mm256_mul_ps(tx256_re,chF256_x);
+					rx_tmp256_2    = _mm256_mul_ps(tx256_im,chF256_y);
+					rx_tmp256_3    = _mm256_mul_ps(tx256_im,chF256_x);
+					rx_tmp256_4    = _mm256_mul_ps(tx256_re,chF256_y);
+					rx_tmp256_re   = _mm256_sub_ps(rx_tmp256_1,rx_tmp256_2);
+					rx_tmp256_im   = _mm256_add_ps(rx_tmp256_3,rx_tmp256_4);
+					rx_tmp256_re_f = _mm256_add_ps(rx_tmp256_re_f,rx_tmp256_re);
+					rx_tmp256_im_f = _mm256_add_ps(rx_tmp256_im_f,rx_tmp256_im);
+				}  // j    
+				//rx_sig_re[ii][f+k*ofdm_symbol_size] = rx_tmp.x*path_loss;
+				//rx_sig_im[ii][f+k*ofdm_symbol_size] = rx_tmp.y*path_loss;
+				rx_tmp256_re_f = _mm256_mul_ps(rx_tmp256_re_f,pathloss256);
+			        rx_tmp256_im_f = _mm256_mul_ps(rx_tmp256_im_f,pathloss256);
+			        _mm256_storeu_ps(&rx_sig_re[ii][4*f],rx_tmp256_re_f);
+			        _mm256_storeu_ps(&rx_sig_im[ii][4*f],rx_tmp256_im_f);
+			}
+		} // ii
+	} // f,f2,f3
+  //}//k
+}
 #ifdef CHANNEL_SSE
 void multipath_channel_prach(channel_desc_t *desc,
                        double *tx_sig_re[2],

openair1/SIMULATION/TOOLS/rangen_double.c

@@ -33,7 +33,7 @@
 #ifdef USER_MODE
 #include <stdio.h>
 #include <stdlib.h>
-#include <stdin.h>
+#include <stdint.h>
 #include <math.h>
 #include <time.h>
 #endif