12->16bit rescaling (transmission) for USRP device

targets/ARCH/USRP/USERSPACE/LIB/usrp_lib.cpp

@@ -148,6 +148,35 @@ static int trx_usrp_write(openair0_device *device, openair0_timestamp timestamp,
   int ret=0;
   usrp_state_t *s = (usrp_state_t*)device->priv;
   
+  int nsamps2;  // aligned to upper 32 or 16 byte boundary
+#if defined(__x86_64) || defined(__i386__)
+#ifdef __AVX2__
+  nsamps2 = (nsamps+7)>>3;
+  __m256i buff_tx[2][nsamps2];
+#else
+  nsamps2 = (nsamps+3)>>2;
+  __m128i buff_tx[2][nsamps2];
+#endif
+#elif defined(__arm__)
+  nsamps2 = (nsamps+3)>>2;
+  int16x8_t buff_tx[2][nsamps2];
+#endif
+  
+  // bring RX data into 12 LSBs for softmodem RX
+  for (int i=0; i<cc; i++) {
+    for (int j=0; j<nsamps2; j++) {
+#if defined(__x86_64__) || defined(__i386__)
+#ifdef __AVX2__
+      buff_tx[i][j] = _mm256_slli_epi16(((__m256i*)buff[i])[j],4);
+#else
+      buff_tx[i][j] = _mm_slli_epi16(((__m128i*)buff128[i])[j],4);
+#endif
+#elif defined(__arm__)
+      buff_tx[i][j] = vshlq_n_s16(((int16x8_t*)buff128[i])[j],4);
+#endif
+    }
+  }
+
   s->tx_md.time_spec = uhd::time_spec_t::from_ticks(timestamp, s->sample_rate);
   s->tx_md.has_time_spec = flags;
   
@@ -174,10 +203,10 @@ static int trx_usrp_write(openair0_device *device, openair0_timestamp timestamp,
   if (cc>1) {
     std::vector<void *> buff_ptrs;
     for (int i=0; i<cc; i++)
-            buff_ptrs.push_back(buff[i]);
+      buff_ptrs.push_back(buff_tx[i]);
     ret = (int)s->tx_stream->send(buff_ptrs, nsamps, s->tx_md,1e-3);
   } else
-        ret = (int)s->tx_stream->send(buff[0], nsamps, s->tx_md,1e-3);
+    ret = (int)s->tx_stream->send(buff_tx[0], nsamps, s->tx_md,1e-3);