diff --git a/lib/include/srslte/phy/fec/viterbi.h b/lib/include/srslte/phy/fec/viterbi.h
index 043a6f9f9..d69750fb3 100644
--- a/lib/include/srslte/phy/fec/viterbi.h
+++ b/lib/include/srslte/phy/fec/viterbi.h
@@ -106,6 +106,12 @@ SRSLTE_API int srslte_viterbi_init_neon(srslte_viterbi_t *q,
                                    uint32_t max_frame_length, 
                                    bool tail_bitting);
 
+SRSLTE_API int srslte_viterbi_init_avx2(srslte_viterbi_t *q, 
+                                   srslte_viterbi_type_t type, 
+                                   int poly[3], 
+                                   uint32_t max_frame_length, 
+                                   bool tail_bitting);
+
 
 
 #endif
diff --git a/lib/include/srslte/phy/utils/vector_simd.h b/lib/include/srslte/phy/utils/vector_simd.h
index cd6eb4d28..3ecdf7b59 100644
--- a/lib/include/srslte/phy/utils/vector_simd.h
+++ b/lib/include/srslte/phy/utils/vector_simd.h
@@ -36,19 +36,45 @@ extern "C" {
 #include "srslte/config.h"
 
 SRSLTE_API int srslte_vec_dot_prod_sss_simd(short *x, short *y, uint32_t len); 
+
+SRSLTE_API int srslte_vec_dot_prod_sss_simd_avx(short *x, short *y, uint32_t len); 
+
+
   
 SRSLTE_API void srslte_vec_sum_sss_simd(short *x, short *y, short *z, uint32_t len);
 
+SRSLTE_API void srslte_vec_sum_sss_simd_avx(short *x, short *y, short *z, uint32_t len);
+
+
+
 SRSLTE_API void srslte_vec_sub_sss_simd(short *x, short *y, short *z, uint32_t len); 
 
+SRSLTE_API void srslte_vec_sub_sss_simd_avx(short *x, short *y, short *z, uint32_t len);
+
+
+
+
+
 SRSLTE_API void srslte_vec_prod_sss_simd(short *x, short *y, short *z, uint32_t len); 
 
+SRSLTE_API void srslte_vec_prod_sss_simd_avx(short *x, short *y, short *z, uint32_t len);
+
+
 SRSLTE_API void srslte_vec_sc_div2_sss_simd(short *x, int n_rightshift, short *z, uint32_t len); 
 
+SRSLTE_API  void srslte_vec_sc_div2_sss_simd_avx(short *x, int k, short *z, uint32_t len);
+
+
+
+
+
 SRSLTE_API void srslte_vec_lut_sss_simd(short *x, unsigned short *lut, short *y, uint32_t len); 
 
 SRSLTE_API void srslte_vec_convert_fi_simd(float *x, int16_t *z, float scale, uint32_t len); 
 
+
+
+SRSLTE_API void srslte_32fc_s32f_multiply_32fc_avx( cf_t *z,const cf_t *x,const float h,const uint32_t len);
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/src/phy/fec/viterbi.c b/lib/src/phy/fec/viterbi.c
index baebed0a1..444c29338 100644
--- a/lib/src/phy/fec/viterbi.c
+++ b/lib/src/phy/fec/viterbi.c
@@ -42,6 +42,14 @@
 
 #define DEFAULT_GAIN 100
 
+
+#define AVX_ON
+
+#ifdef LV_HAVE_AVX
+    #ifdef AVX_ON
+        #define USE_AVX
+    #endif
+#endif
 //#undef LV_HAVE_SSE
 
 int decode37(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length) {
@@ -120,6 +128,51 @@ void free37_sse(void *o) {
 #endif
 
 
+
+#ifdef LV_HAVE_AVX
+int decode37_avx2(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length) {
+  srslte_viterbi_t *q = o;
+
+  uint32_t best_state;
+
+  if (frame_length > q->framebits) {
+    fprintf(stderr, "Initialized decoder for max frame length %d bits\n",
+        q->framebits);
+    return -1;
+  }
+
+  /* Initialize Viterbi decoder */
+  init_viterbi37_avx2(q->ptr, q->tail_biting?-1:0);
+
+  /* Decode block */
+  if (q->tail_biting) {
+    for (int i=0;i<TB_ITER;i++) {
+      memcpy(&q->tmp[i*3*frame_length], symbols, 3*frame_length*sizeof(uint8_t));      
+    }
+    update_viterbi37_blk_avx2(q->ptr, q->tmp, TB_ITER*frame_length, &best_state);
+    chainback_viterbi37_avx2(q->ptr,  q->tmp, TB_ITER*frame_length, best_state);
+    memcpy(data, &q->tmp[((int) (TB_ITER/2))*frame_length], frame_length*sizeof(uint8_t));
+  } else {
+    update_viterbi37_blk_avx2(q->ptr, symbols, frame_length+q->K-1, NULL);
+    chainback_viterbi37_avx2(q->ptr, data, frame_length, 0);
+  }
+  
+  return q->framebits;
+}
+
+void free37_avx2(void *o) {
+  srslte_viterbi_t *q = o;
+  if (q->symbols_uc) {
+    free(q->symbols_uc);
+  }
+  if (q->tmp) {
+    free(q->tmp);
+  }
+  delete_viterbi37_avx2(q->ptr);
+}
+
+#endif
+
 #ifdef HAVE_NEON
 int decode37_neon(void *o, uint8_t *symbols, uint8_t *data, uint32_t frame_length) {
   srslte_viterbi_t *q = o;
@@ -286,6 +339,45 @@ int init37_neon(srslte_viterbi_t *q, int poly[3], uint32_t framebits, bool tail_
 }
 #endif
 
+
+#ifdef LV_HAVE_AVX
+int init37_avx2(srslte_viterbi_t *q, int poly[3], uint32_t framebits, bool tail_biting) {
+  q->K = 7;
+  q->R = 3;
+  q->framebits = framebits;
+  q->gain_quant_s = 4; 
+  q->gain_quant = DEFAULT_GAIN; 
+  q->tail_biting = tail_biting;
+  q->decode = decode37_avx2;
+  q->free = free37_avx2;
+  q->decode_f = NULL;
+  printf("USING AVX VITERBI\n");
+  q->symbols_uc = srslte_vec_malloc(3 * (q->framebits + q->K - 1) * sizeof(uint8_t));
+  if (!q->symbols_uc) {
+    perror("malloc");
+    return -1;
+  }
+  if (q->tail_biting) {
+    q->tmp = srslte_vec_malloc(TB_ITER*3*(q->framebits + q->K - 1) * sizeof(uint8_t));
+    if (!q->tmp) {
+      perror("malloc");
+      free37(q);
+      return -1;
+    }
+  } else {
+    q->tmp = NULL;
+  }
+  
+  if ((q->ptr = create_viterbi37_avx2(poly, TB_ITER*framebits)) == NULL) {
+    fprintf(stderr, "create_viterbi37 failed\n");
+    free37(q);
+    return -1;
+  } else {
+    return 0;
+  }     
+}
+#endif
+
 void srslte_viterbi_set_gain_quant(srslte_viterbi_t *q, float gain_quant) {
   q->gain_quant = gain_quant;
 }
@@ -299,7 +391,11 @@ int srslte_viterbi_init(srslte_viterbi_t *q, srslte_viterbi_type_t type, int pol
   switch (type) {
   case SRSLTE_VITERBI_37:
 #ifdef LV_HAVE_SSE
-    return init37_sse(q, poly, max_frame_length, tail_bitting);
+    #ifdef USE_AVX 
+        return init37_avx2(q, poly, max_frame_length, tail_bitting);
+    #else
+        return init37_sse(q, poly, max_frame_length, tail_bitting);
+    #endif
 #else
 	#ifdef HAVE_NEON
 	return init37_neon(q, poly, max_frame_length, tail_bitting);
@@ -320,6 +416,13 @@ int srslte_viterbi_init_sse(srslte_viterbi_t *q, srslte_viterbi_type_t type, int
 }
 #endif
 
+#ifdef LV_HAVE_AVX
+int srslte_viterbi_init_avx2(srslte_viterbi_t *q, srslte_viterbi_type_t type, int poly[3], uint32_t max_frame_length, bool tail_bitting) 
+{
+  return init37_avx2(q, poly, max_frame_length, tail_bitting);      
+}
+#endif
+
 void srslte_viterbi_free(srslte_viterbi_t *q) {
   if (q->free) {
     q->free(q);    
diff --git a/lib/src/phy/fec/viterbi37.h b/lib/src/phy/fec/viterbi37.h
index 2c7f8c57f..574f4fd87 100644
--- a/lib/src/phy/fec/viterbi37.h
+++ b/lib/src/phy/fec/viterbi37.h
@@ -88,3 +88,25 @@ int update_viterbi37_blk_neon(void *p,
                               uint32_t *best_state);
 
 
+void *create_viterbi37_avx2(int polys[3], 
+                            uint32_t len);
+
+int init_viterbi37_avx2(void *p, 
+                        int starting_state);
+
+
+void reset_blk_avx2(void *p, int nbits);
+
+int chainback_viterbi37_avx2(void *p, 
+                             uint8_t *data, 
+                             uint32_t nbits, 
+                             uint32_t endstate);
+
+void delete_viterbi37_avx2(void *p);
+
+int update_viterbi37_blk_avx2(void *p, 
+                              uint8_t *syms, 
+                              uint32_t nbits, 
+                              uint32_t *best_state);
+
+
diff --git a/lib/src/phy/utils/vector.c b/lib/src/phy/utils/vector.c
index a5921aeda..12809950b 100644
--- a/lib/src/phy/utils/vector.c
+++ b/lib/src/phy/utils/vector.c
@@ -109,7 +109,7 @@ void srslte_vec_sub_sss(short *x, short *y, short *z, uint32_t len) {
     z[i] = x[i]-y[i];
   }
 #else
-  srslte_vec_sub_sss_simd(x, y, z, len);
+  srslte_vec_sub_sss_simd_avx(x, y, z, len);
 #endif
 }
 
@@ -135,7 +135,7 @@ void srslte_vec_sum_sss(short *x, short *y, short *z, uint32_t len) {
     z[i] = x[i]+y[i];
   }
 #else
-  srslte_vec_sum_sss_simd(x, y, z, len);
+  srslte_vec_sum_sss_simd_avx(x, y, z, len);
 #endif
 }
 
@@ -204,7 +204,7 @@ void srslte_vec_sc_div2_sss(short *x, int n_rightshift, short *z, uint32_t len)
     z[i] = x[i]/pow2_div;
   }
 #else
-  srslte_vec_sc_div2_sss_simd(x, n_rightshift, z, len);
+  srslte_vec_sc_div2_sss_simd_avx(x, n_rightshift, z, len);
 #endif
 }
 
@@ -226,10 +226,7 @@ void srslte_vec_sc_prod_cfc(cf_t *x, float h, cf_t *z, uint32_t len) {
     z[i] = x[i]*h;
   }
 #else
-  cf_t hh;
-  __real__ hh = h;
-  __imag__ hh = 0;
-  volk_32fc_s32fc_multiply_32fc(z,x,hh,len);
+  srslte_32fc_s32f_multiply_32fc_avx(z,x, h, len);
 #endif
 }
 
@@ -514,7 +511,7 @@ void srslte_vec_prod_sss(short *x, short *y, short *z, uint32_t len) {
     z[i] = x[i]*y[i];
   }
 #else
-  srslte_vec_prod_sss_simd(x,y,z,len);
+  srslte_vec_prod_sss_simd_avx(x,y,z,len);
 #endif
 }
 
@@ -653,7 +650,7 @@ int32_t srslte_vec_dot_prod_sss(int16_t *x, int16_t *y, uint32_t len) {
   }
   return res;
 #else
-  return srslte_vec_dot_prod_sss_simd(x, y, len); 
+  return srslte_vec_dot_prod_sss_simd_avx(x, y, len); 
 #endif
 }
 
diff --git a/lib/src/phy/utils/vector_simd.c b/lib/src/phy/utils/vector_simd.c
index 817dfa2da..81694b110 100644
--- a/lib/src/phy/utils/vector_simd.c
+++ b/lib/src/phy/utils/vector_simd.c
@@ -40,6 +40,9 @@
 #include <smmintrin.h>
 #endif
 
+#ifdef LV_HAVE_AVX
+#include <immintrin.h>
+#endif
 
 
 int srslte_vec_dot_prod_sss_simd(short *x, short *y, uint32_t len)
@@ -83,6 +86,47 @@ int srslte_vec_dot_prod_sss_simd(short *x, short *y, uint32_t len)
   return result; 
 }
 
+
+int srslte_vec_dot_prod_sss_simd_avx(short *x, short *y, uint32_t len)
+{
+  int result = 0; 
+#ifdef LV_HAVE_AVX
+  unsigned int number = 0;
+  const unsigned int points = len / 16;
+
+  const __m256i* xPtr = (const __m256i*) x;
+  const __m256i* yPtr = (const __m256*) y;
+  
+  __m256i dotProdVal = _mm256_setzero_si256();
+
+  __m256i xVal, yVal, zVal;
+  for(;number < points; number++){
+
+    xVal = _mm256_load_si256(xPtr);
+    yVal = _mm256_loadu_si256(yPtr);
+    zVal = _mm256_mullo_epi16(xVal, yVal);
+    dotProdVal = _mm256_add_epi16(dotProdVal, zVal);
+    xPtr ++;
+    yPtr ++;
+  }
+  
+  short dotProdVector[16];
+  _mm256_store_si256((__m256i*) dotProdVector, dotProdVal);
+  for (int i=0;i<16;i++) {
+    result += dotProdVector[i]; 
+  }
+
+  number = points * 16;
+  for(;number < len; number++){
+    result += (x[number] * y[number]);
+  }
+  
+#endif
+  return result; 
+}
+
+
+
 void srslte_vec_sum_sss_simd(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
@@ -116,6 +160,39 @@ void srslte_vec_sum_sss_simd(short *x, short *y, short *z, uint32_t len)
 
 }
 
+void srslte_vec_sum_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
+{
+#ifdef LV_HAVE_SSE
+  unsigned int number = 0;
+  const unsigned int points = len / 16;
+
+  const __m256i* xPtr = (const __m256i*) x;
+  const __m256i* yPtr = (const __m256i*) y;
+  __m256i* zPtr = (__m256i*) z;
+
+  __m256i xVal, yVal, zVal;
+  for(;number < points; number++){
+
+    xVal = _mm256_load_si256(xPtr);
+    yVal = _mm256_loadu_si256(yPtr);
+
+    zVal = _mm256_add_epi16(xVal, yVal);
+    _mm256_store_si256(zPtr, zVal); 
+
+    xPtr ++;
+    yPtr ++;
+    zPtr ++;
+  }
+
+  number = points * 16;
+  for(;number < len; number++){
+    z[number] = x[number] + y[number];
+  }
+#endif
+
+}
+
+
 void srslte_vec_sub_sss_simd(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
@@ -148,6 +225,41 @@ void srslte_vec_sub_sss_simd(short *x, short *y, short *z, uint32_t len)
 #endif
 }
 
+void srslte_vec_sub_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
+{
+#ifdef LV_HAVE_AVX
+  unsigned int number = 0;
+  const unsigned int points = len / 16;
+
+  const __m256i* xPtr = (const __m256i*) x;
+  const __m256i* yPtr = (const __m256i*) y;
+  __m256i* zPtr = (__m256i*) z;
+
+  __m256i xVal, yVal, zVal;
+  for(;number < points; number++){
+
+    xVal = _mm256_load_si256(xPtr);
+    yVal = _mm256_loadu_si256(yPtr);
+
+    zVal = _mm256_sub_epi16(xVal, yVal);
+
+    _mm256_store_si256(zPtr, zVal); 
+
+    xPtr ++;
+    yPtr ++;
+    zPtr ++;
+  }
+
+  number = points * 16;
+  for(;number < len; number++){
+    z[number] = x[number] - y[number];
+  }
+  #endif
+}
+
+
+
+
 void srslte_vec_prod_sss_simd(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
@@ -180,6 +292,38 @@ void srslte_vec_prod_sss_simd(short *x, short *y, short *z, uint32_t len)
 #endif
 }
 
+void srslte_vec_prod_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
+{
+#ifdef LV_HAVE_SSE
+  unsigned int number = 0;
+  const unsigned int points = len / 16;
+
+  const __m256i* xPtr = (const __m256i*) x;
+  const __m256i* yPtr = (const __m256i*) y;
+  __m256i* zPtr = (__m256i*) z;
+
+  __m256i xVal, yVal, zVal;
+  for(;number < points; number++){
+
+    xVal = _mm256_load_si256(xPtr);
+    yVal = _mm256_loadu_si256(yPtr);
+
+    zVal = _mm256_mullo_epi16(xVal, yVal);
+
+    _mm256_store_si256(zPtr, zVal); 
+
+    xPtr ++;
+    yPtr ++;
+    zPtr ++;
+  }
+
+  number = points * 16;
+  for(;number < len; number++){
+    z[number] = x[number] * y[number];
+  }
+#endif
+}
+
 void srslte_vec_sc_div2_sss_simd(short *x, int k, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
@@ -210,6 +354,36 @@ void srslte_vec_sc_div2_sss_simd(short *x, int k, short *z, uint32_t len)
 #endif
 }
 
+void srslte_vec_sc_div2_sss_simd_avx(short *x, int k, short *z, uint32_t len)
+{
+#ifdef LV_HAVE_AVX
+  unsigned int number = 0;
+  const unsigned int points = len / 16;
+
+  const __m256i* xPtr = (const __m256i*) x;
+  __m256i* zPtr = (__m256i*) z;
+
+  __m256i xVal, zVal;
+  for(;number < points; number++){
+
+    xVal = _mm256_load_si256(xPtr);
+    
+    zVal = _mm256_srai_epi16(xVal, k);                 
+      
+    _mm256_store_si256(zPtr, zVal); 
+
+    xPtr ++;
+    zPtr ++;
+  }
+
+  number = points * 16;
+  short divn = (1<<k);
+  for(;number < len; number++){
+    z[number] = x[number] / divn;
+  }
+#endif
+}
+
 /* No improvement with AVX */
 void srslte_vec_lut_sss_simd(short *x, unsigned short *lut, short *y, uint32_t len)
 {
@@ -282,3 +456,33 @@ void srslte_vec_convert_fi_simd(float *x, int16_t *z, float scale, uint32_t len)
   }
 #endif
 }
+
+
+ void srslte_32fc_s32f_multiply_32fc_avx( cf_t *z,const cf_t *x,const float h,const uint32_t len)
+{
+#ifdef LV_HAVE_AVX
+   
+    unsigned int i = 0;
+    const unsigned int loops = len/4;
+    //__m256 outputVec;
+     cf_t *xPtr = x;
+     cf_t *zPtr = z;
+    
+    __m256 inputVec, outputVec;
+     const __m256 tapsVec  = _mm256_set1_ps(h);
+    for(;i < loops;i++)
+    {
+        inputVec  = _mm256_loadu_ps((float*)xPtr);
+       //__builtin_prefetch(xPtr+4);
+        outputVec = _mm256_mul_ps(inputVec,tapsVec);
+       _mm256_storeu_ps((float*)zPtr,outputVec);
+       xPtr += 4;
+       zPtr += 4;
+    }
+    
+    for(i = loops * 4;i < len;i++)
+    {
+        *zPtr++ = (*xPtr++) * h;
+    }
+#endif
+}
\ No newline at end of file
diff --git a/srslte/lib/fec/viterbi37_avx2.c b/srslte/lib/fec/viterbi37_avx2.c
new file mode 100644
index 000000000..bb8e90d10
--- /dev/null
+++ b/srslte/lib/fec/viterbi37_avx2.c
@@ -0,0 +1,339 @@
+/* Adapted Phil Karn's r=1/3 k=9 viterbi decoder to r=1/3 k=7
+ * 
+ * K=15 r=1/6 Viterbi decoder for x86 SSE2
+ * Copyright Mar 2004, Phil Karn, KA9Q
+ * May be used under the terms of the GNU Lesser General Public License (LGPL)
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <memory.h>
+#include <limits.h>
+#include "parity.h"
+
+//#define DEBUG
+
+#ifdef LV_HAVE_SSE
+
+#include <emmintrin.h>
+#include <tmmintrin.h>
+#include <immintrin.h>
+#include <emmintrin.h>
+#define _mm256_set_m128i(v0, v1)  _mm256_insertf128_si256(_mm256_castsi128_si256(v1), (v0), 1)
+
+#define _mm256_setr_m128i(v0, v1) _mm256_set_m128i((v1), (v0))
+
+typedef union { 
+  unsigned char c[64]; 
+  __m128i       v[4];   
+} metric_t;
+typedef union { 
+  unsigned int  w[2]; 
+  unsigned char  c[8]; 
+  unsigned short s[4]; 
+  __m64 v;  
+} decision_t;
+
+union branchtab27 { 
+  unsigned char c[32]; 
+  __m256i       v;  
+} Branchtab37_sse2[3];
+
+int firstGo;
+/* State info for instance of Viterbi decoder */
+struct v37 {
+  metric_t metrics1; /* path metric buffer 1 */
+  metric_t metrics2; /* path metric buffer 2 */
+  decision_t *dp; /* Pointer to current decision */
+  metric_t *old_metrics,*new_metrics; /* Pointers to path metrics, swapped on every bit */
+  decision_t *decisions; /* Beginning of decisions for block */
+  uint32_t len; 
+};
+
+void set_viterbi37_polynomial_avx2(int polys[3]) {
+  int state;
+
+  for(state=0;state < 32;state++){
+    Branchtab37_sse2[0].c[state] = (polys[0] < 0) ^ parity((2*state) & polys[0]) ? 255:0;
+    Branchtab37_sse2[1].c[state] = (polys[1] < 0) ^ parity((2*state) & polys[1]) ? 255:0;
+    Branchtab37_sse2[2].c[state] = (polys[2] < 0) ^ parity((2*state) & polys[2]) ? 255:0;
+  }
+}
+
+void clear_v37_avx2(struct v37 *vp) {
+  bzero(vp->decisions, sizeof(decision_t)*vp->len);
+  vp->dp = NULL; 
+  bzero(&vp->metrics1, sizeof(metric_t));
+  bzero(&vp->metrics2, sizeof(metric_t));
+  vp->old_metrics = NULL; 
+  vp->new_metrics = NULL; 
+}
+
+
+/* Initialize Viterbi decoder for start of new frame */
+int init_viterbi37_avx2(void *p, int starting_state) {
+  struct v37 *vp = p;
+  uint32_t i;
+  firstGo = 1;
+  for(i=0;i<64;i++)
+    vp->metrics1.c[i] = 63;
+
+  clear_v37_avx2(vp);
+
+  vp->old_metrics = &vp->metrics1;
+  vp->new_metrics = &vp->metrics2;
+  vp->dp = vp->decisions;
+  if (starting_state != -1) {
+    vp->old_metrics->c[starting_state & 63] = 0; /* Bias known start state */
+  }
+  return 0;
+}
+
+/* Create a new instance of a Viterbi decoder */
+void *create_viterbi37_avx2(int polys[3], uint32_t len) {
+  void *p;
+  struct v37 *vp;
+
+  set_viterbi37_polynomial_avx2(polys);
+  
+  /* Ordinary malloc() only returns 8-byte alignment, we need 16 */
+  if(posix_memalign(&p, sizeof(__m128i),sizeof(struct v37)))
+    return NULL;
+
+  vp = (struct v37 *)p;
+  if(posix_memalign(&p, sizeof(__m128i),(len+6)*sizeof(decision_t))) {
+    free(vp);
+    return NULL;
+  }
+  vp->decisions = (decision_t *)p;
+  vp->len = len+6;
+  return vp;
+}
+
+
+/* Viterbi chainback */
+int chainback_viterbi37_avx2(
+      void *p,
+      uint8_t *data, /* Decoded output data */
+      uint32_t nbits, /* Number of data bits */
+      uint32_t endstate) { /* Terminal encoder state */
+  struct v37 *vp = p;
+  
+  if (p == NULL)
+    return -1;
+
+  decision_t *d = (decision_t *)vp->decisions;
+
+  /* Make room beyond the end of the encoder register so we can
+   * accumulate a full byte of decoded data
+   */
+  endstate %= 64;
+  endstate <<= 2;
+  
+  /* The store into data[] only needs to be done every 8 bits.
+   * But this avoids a conditional branch, and the writes will
+   * combine in the cache anyway
+   */
+  d += 6; /* Look past tail */
+  while(nbits--) {
+    int k;
+
+    k = (d[nbits].c[(endstate>>2)/8] >> ((endstate>>2)%8)) & 1;
+    endstate = (endstate >> 1) | (k << 7);
+    data[nbits] = k;
+    //printf("nbits=%d, endstate=%3d, k=%d, w[0]=%d, w[1]=%d, c=%d\n", nbits, endstate, k, d[nbits].s[1]&1, d[nbits].s[2]&1, d[nbits].c[(endstate>>2)/8]&1);
+  }
+  return 0;
+}
+
+/* Delete instance of a Viterbi decoder */
+void delete_viterbi37_avx2(void *p){
+  struct v37 *vp = p;
+
+  if(vp != NULL){
+    free(vp->decisions);
+    free(vp);
+  }
+}
+void printer_256i(char *s, __m256i val) {
+  
+  printf("%s: ", s);
+  
+  uint8_t *x = (uint8_t*) &val; 
+  for (int i=0;i<32;i++) {
+    printf("%3d, ", x[i]);
+  }
+  printf("\n");
+}
+
+void printer_128i(char *s, __m128i val) {
+  
+  printf("%s: ", s);
+  
+  uint8_t *x = (uint8_t*) &val; 
+  for (int i=0;i<16;i++) {
+    printf("%3d, ", x[i]);
+  }
+  printf("\n");
+}
+
+void printer_m64(char *s, __m64 val) {
+  
+  printf("%s: ", s);
+  
+  uint8_t *x = (uint8_t*) &val; 
+  for (int i=0;i<8;i++) {
+    printf("%3d, ", x[i]);
+  }
+  printf("\n");
+}
+
+
+void update_viterbi37_blk_avx2(void *p,unsigned char *syms,int nbits, uint32_t *best_state) {
+  struct v37 *vp = p;
+  decision_t *d;
+
+  if(p == NULL)
+    return;
+  
+#ifdef DEBUG
+  printf("[");
+#endif
+  
+  d = (decision_t *) vp->dp;
+  
+  for (int s=0;s<nbits;s++) {
+    memset(d+s,0,sizeof(decision_t));
+  }
+  
+  while(nbits--) {
+    __m256i sym0v,sym1v,sym2v;
+    
+    void *tmp;
+
+    sym0v = _mm256_set1_epi8(syms[0]);
+    sym1v = _mm256_set1_epi8(syms[1]);
+    sym2v = _mm256_set1_epi8(syms[2]);
+
+    syms += 3;
+    
+     
+    __m256i decision0,decision1,survivor0,survivor1,metric,m_metric,m0,m1,m2,m3;
+
+    /* Form branch metrics */
+    m0 = _mm256_avg_epu8(_mm256_xor_si256(Branchtab37_sse2[0].v,sym0v),_mm256_xor_si256(Branchtab37_sse2[1].v,sym1v));
+    metric = _mm256_avg_epu8(_mm256_xor_si256(Branchtab37_sse2[2].v,sym2v),m0);
+    
+#ifdef DEBUG
+      print_128i("metric_initial", metric);
+#endif
+      /* There's no packed bytes right shift in SSE2, so we use the word version and mask
+       */
+    metric = _mm256_srli_epi16(metric,3);
+    metric = _mm256_and_si256(metric,_mm256_set1_epi8(31));
+    m_metric = _mm256_sub_epi8(_mm256_set1_epi8(31),metric);
+      
+#ifdef DEBUG
+      print_128i("metric        ", metric);
+      print_128i("m_metric      ", m_metric);
+#endif
+
+    __m256i temp = _mm256_set_m128i( vp->old_metrics->v[1], vp->old_metrics->v[0]);
+    m0 = _mm256_add_epi8(temp,metric); 
+    m2 = _mm256_add_epi8(temp,m_metric);
+     
+    temp = _mm256_set_m128i( vp->old_metrics->v[3], vp->old_metrics->v[2]);
+    m3 = _mm256_add_epi8(temp,metric);
+    m1 = _mm256_add_epi8(temp,m_metric);
+     
+     /* Compare and select, using modulo arithmetic */
+    decision0 = _mm256_cmpgt_epi8(_mm256_sub_epi8(m0,m1),_mm256_setzero_si256());
+    decision1 =_mm256_cmpgt_epi8(_mm256_sub_epi8(m2,m3),_mm256_setzero_si256());
+    survivor0 = _mm256_or_si256(_mm256_and_si256(decision0,m1),_mm256_andnot_si256(decision0,m0));
+    survivor1 = _mm256_or_si256(_mm256_and_si256(decision1,m3),_mm256_andnot_si256(decision1,m2));
+ 
+    unsigned int x  = _mm256_movemask_epi8(_mm256_unpackhi_epi8(decision0,decision1));
+    unsigned int y = _mm256_movemask_epi8(_mm256_unpacklo_epi8(decision0,decision1));
+  
+    d->s[0] = (short) y;
+    d->s[1] = (short) x;
+    d->s[2] = (short) (y >>16);
+    d->s[3] = (short)(x>> 16);
+
+      
+    __m256i unpack;
+    unpack =  _mm256_unpacklo_epi8(survivor0,survivor1);
+    vp->new_metrics->v[0] =_mm256_castsi256_si128(unpack);
+
+    vp->new_metrics->v[1] = _mm256_extractf128_si256(unpack,1);
+
+    unpack = _mm256_unpackhi_epi8(survivor0,survivor1);
+
+    vp->new_metrics->v[2] =_mm256_castsi256_si128(unpack);
+    vp->new_metrics->v[3] = _mm256_extractf128_si256(unpack,1);
+       
+    __m128i temp1 = vp->new_metrics->v[1];
+
+    vp->new_metrics->v[1] = vp->new_metrics->v[2];
+    vp->new_metrics->v[2] = temp1;
+
+    // See if we need to normalize
+    if (vp->new_metrics->c[0] > 100) {
+      int i;
+      uint8_t adjust; 
+      __m128i adjustv;
+      union { __m128i v; signed short w[8]; } t;
+      
+      adjustv = vp->new_metrics->v[0];
+      for(i=1;i<4;i++) {
+        adjustv = _mm_min_epu8(adjustv,vp->new_metrics->v[i]);
+      }
+      
+      adjustv = _mm_min_epu8(adjustv,_mm_srli_si128(adjustv,8));
+      adjustv = _mm_min_epu8(adjustv,_mm_srli_si128(adjustv,4));
+      adjustv = _mm_min_epu8(adjustv,_mm_srli_si128(adjustv,2));
+      
+      t.v = adjustv;
+      adjust = t.w[0];
+      adjustv = _mm_set1_epi8(adjust);
+      
+      /* We cannot use a saturated subtract, because we often have to adjust by more than SHRT_MAX
+       * This is okay since it can't overflow anyway
+       */
+      for(i=0;i<4;i++)
+        vp->new_metrics->v[i] = _mm_sub_epi8(vp->new_metrics->v[i],adjustv);
+     
+    }
+
+    firstGo = 0;
+    d++;
+    /* Swap pointers to old and new metrics */
+    tmp = vp->old_metrics;
+    vp->old_metrics = vp->new_metrics;
+    vp->new_metrics = tmp;
+  }
+  
+  if (best_state) {
+    uint32_t i, bst=0;
+    uint8_t minmetric=UINT8_MAX;
+    for (i=0;i<64;i++) {
+      if (vp->old_metrics->c[i] <= minmetric) {
+        bst = i;
+        minmetric = vp->old_metrics->c[i];
+      }
+    }
+    *best_state = bst;
+  }
+
+  #ifdef DEBUG
+  printf("];\n===========================================\n");
+#endif
+
+  vp->dp = d;
+}
+
+#endif
+
+
+