Avoid cexpf in srsran_vec_apply_cfo_simd

lib/src/phy/utils/vector_simd.c

@@ -1684,21 +1684,21 @@ void srsran_vec_apply_cfo_simd(const cf_t* x, float cfo, cf_t* z, int len)
 {
   const float TWOPI = 2.0f * (float)M_PI;
   int         i     = 0;
+  cf_t        osc   = cexpf(_Complex_I * TWOPI * cfo);
+  cf_t        phase = 1.0f;
 
 #if SRSRAN_SIMD_CF_SIZE
-  srsran_simd_aligned cf_t _osc[SRSRAN_SIMD_CF_SIZE];
+  // Load initial phases and oscillator
   srsran_simd_aligned cf_t _phase[SRSRAN_SIMD_CF_SIZE];
-
-  if (i < len - SRSRAN_SIMD_CF_SIZE + 1) {
-    for (int k = 0; k < SRSRAN_SIMD_CF_SIZE; k++) {
-      _osc[k]   = cexpf(_Complex_I * TWOPI * cfo * SRSRAN_SIMD_CF_SIZE);
-      _phase[k] = cexpf(_Complex_I * TWOPI * cfo * k);
-    }
+  _phase[0] = phase;
+  for (int k = 1; k < SRSRAN_SIMD_CF_SIZE; k++) {
+    _phase[k] = _phase[k - 1] * osc;
   }
-  simd_cf_t _simd_osc   = srsran_simd_cfi_load(_osc);
+  simd_cf_t _simd_osc   = srsran_simd_cf_set1(_phase[SRSRAN_SIMD_CF_SIZE - 1] * osc);
   simd_cf_t _simd_phase = srsran_simd_cfi_load(_phase);
 
   if (SRSRAN_IS_ALIGNED(x) && SRSRAN_IS_ALIGNED(z)) {
+    // For aligned memory
     for (; i < len - SRSRAN_SIMD_CF_SIZE + 1; i += SRSRAN_SIMD_CF_SIZE) {
       simd_cf_t a = srsran_simd_cfi_load(&x[i]);
 
@@ -1709,6 +1709,7 @@ void srsran_vec_apply_cfo_simd(const cf_t* x, float cfo, cf_t* z, int len)
       _simd_phase = srsran_simd_cf_prod(_simd_phase, _simd_osc);
     }
   } else {
+    // For unaligned memory
     for (; i < len - SRSRAN_SIMD_F_SIZE + 1; i += SRSRAN_SIMD_F_SIZE) {
       simd_cf_t a = srsran_simd_cfi_loadu(&x[i]);
 
@@ -1719,9 +1720,12 @@ void srsran_vec_apply_cfo_simd(const cf_t* x, float cfo, cf_t* z, int len)
       _simd_phase = srsran_simd_cf_prod(_simd_phase, _simd_osc);
     }
   }
+
+  // Stores the next phase
+  srsran_simd_cfi_store(_phase, _simd_phase);
+  phase = _phase[0];
 #endif
-  cf_t osc   = cexpf(_Complex_I * TWOPI * cfo);
-  cf_t phase = cexpf(_Complex_I * TWOPI * cfo * i);
+
   for (; i < len; i++) {
     z[i] = x[i] * phase;