From d48dcc25b4dd027b60729a4765b0493d4faa1872 Mon Sep 17 00:00:00 2001
From: yagoda <tallonj@tcd.ie>
Date: Wed, 31 May 2017 22:19:26 +0100
Subject: [PATCH] updating avx vectors and viterbi

---
 srslte/include/srslte/utils/vector_simd.h | 26 ++++-----
 srslte/lib/fec/test/viterbi_test.c        |  2 +-
 srslte/lib/fec/viterbi.c                  |  9 +--
 srslte/lib/utils/vector.c                 | 68 +++++++++++++++--------
 srslte/lib/utils/vector_simd.c            | 35 +++++++-----
 5 files changed, 81 insertions(+), 59 deletions(-)

diff --git a/srslte/include/srslte/utils/vector_simd.h b/srslte/include/srslte/utils/vector_simd.h
index 3ecdf7b59..5cea166b3 100644
--- a/srslte/include/srslte/utils/vector_simd.h
+++ b/srslte/include/srslte/utils/vector_simd.h
@@ -35,46 +35,46 @@ extern "C" {
 #include <stdint.h>
 #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_sse(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 int srslte_vec_dot_prod_sss_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_sse(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_sum_sss_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_sse(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_sub_sss_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_sse(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_prod_sss_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_sse(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_sc_div2_sss_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_lut_sss_sse(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_vec_convert_fi_sse(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);
+SRSLTE_API void srslte_vec_mult_scalar_cf_f_avx( cf_t *z,const cf_t *x,const float h,const uint32_t len);
 #ifdef __cplusplus
 }
 #endif
diff --git a/srslte/lib/fec/test/viterbi_test.c b/srslte/lib/fec/test/viterbi_test.c
index 3d3e7f64a..f619b50b9 100644
--- a/srslte/lib/fec/test/viterbi_test.c
+++ b/srslte/lib/fec/test/viterbi_test.c
@@ -213,7 +213,7 @@ int main(int argc, char **argv) {
       gettimeofday(&t[1], NULL);
       int M = 1; 
       
-      srslte_vec_fprint_b(stdout, data_tx, frame_length);
+      //srslte_vec_fprint_b(stdout, data_tx, frame_length);
       
       for (int i=0;i<M;i++) {
         srslte_viterbi_decode_uc(&dec, llr_c, data_rx, frame_length);      
diff --git a/srslte/lib/fec/viterbi.c b/srslte/lib/fec/viterbi.c
index 5f1592ebb..cc3cdd6e1 100644
--- a/srslte/lib/fec/viterbi.c
+++ b/srslte/lib/fec/viterbi.c
@@ -43,13 +43,6 @@
 #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) {
@@ -391,7 +384,7 @@ 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
-    #ifdef USE_AVX 
+    #ifdef LV_HAVE_AVX 
         return init37_avx2(q, poly, max_frame_length, tail_bitting);
     #else
         return init37_sse(q, poly, max_frame_length, tail_bitting);
diff --git a/srslte/lib/utils/vector.c b/srslte/lib/utils/vector.c
index ce0ce63ef..0aa1c67ea 100644
--- a/srslte/lib/utils/vector.c
+++ b/srslte/lib/utils/vector.c
@@ -103,13 +103,17 @@ void srslte_vec_sub_fff(float *x, float *y, float *z, uint32_t len) {
 }
 
 void srslte_vec_sub_sss(short *x, short *y, short *z, uint32_t len) {
-#ifndef LV_HAVE_SSE
-  int i;
+#ifdef LV_HAVE_AVX
+  srslte_vec_sub_sss_avx(x, y, z, len);
+#else
+#ifdef LV_HAVE_SSE
+  srslte_vec_sub_sss_sse(x, y, z, len);
+#else
+    int i;
   for (i=0;i<len;i++) {
     z[i] = x[i]-y[i];
   }
-#else
-  srslte_vec_sub_sss_simd_avx(x, y, z, len);
+#endif
 #endif
 }
 
@@ -129,13 +133,17 @@ void srslte_vec_sum_fff(float *x, float *y, float *z, uint32_t len) {
 }
 
 void srslte_vec_sum_sss(short *x, short *y, short *z, uint32_t len) {
-#ifndef LV_HAVE_SSE
+#ifdef LV_HAVE_AVX
+  srslte_vec_sum_sss_avx(x, y, z, len);
+#else
+#ifdef LV_HAVE_SSE
+  srslte_vec_sum_sss_sse(x, y, z, len);
+#else
   int i;
   for (i=0;i<len;i++) {
     z[i] = x[i]+y[i];
   }
-#else
-  srslte_vec_sum_sss_simd_avx(x, y, z, len);
+#endif
 #endif
 }
 
@@ -197,14 +205,18 @@ void srslte_vec_sc_prod_sfs(short *x, float h, short *z, uint32_t len) {
 }
 
 void srslte_vec_sc_div2_sss(short *x, int n_rightshift, short *z, uint32_t len) {
-#ifndef LV_HAVE_SSE
+#ifdef LV_HAVE_AVX
+  srslte_vec_sc_div2_sss_avx(x, n_rightshift, z, len);
+#else
+#ifdef LV_HAVE_SSE
+  srslte_vec_sc_div2_sss_sse(x, n_rightshift, z, len);
+#else
   int i;
   int pow2_div = 1<<n_rightshift;
   for (i=0;i<len;i++) {
     z[i] = x[i]/pow2_div;
   }
-#else
-  srslte_vec_sc_div2_sss_simd_avx(x, n_rightshift, z, len);
+#endif
 #endif
 }
 
@@ -220,13 +232,14 @@ void srslte_vec_norm_cfc(cf_t *x, float amplitude, cf_t *y, uint32_t len) {
 }
 
 void srslte_vec_sc_prod_cfc(cf_t *x, float h, cf_t *z, uint32_t len) {
-#ifndef HAVE_VOLK_MULT_FUNCTION
+#ifdef LV_HAVE_AVX
+  srslte_vec_mult_scalar_cf_f_avx(z,x, h, len);  
+#else
   int i;
   for (i=0;i<len;i++) {
     z[i] = x[i]*h;
   }
-#else
-  srslte_32fc_s32f_multiply_32fc_avx(z,x, h, len);
+
 #endif
 }
 
@@ -271,7 +284,7 @@ void srslte_vec_convert_fi(float *x, int16_t *z, float scale, uint32_t len) {
     z[i] = (int16_t) (x[i]*scale);
   }
 #else 
-  srslte_vec_convert_fi_simd(x, z, scale, len);
+  srslte_vec_convert_fi_sse(x, z, scale, len);
 #endif
 }
 
@@ -284,14 +297,13 @@ void srslte_vec_lut_fuf(float *x, uint32_t *lut, float *y, uint32_t len) {
 void srslte_vec_lut_sss(short *x, unsigned short *lut, short *y, uint32_t len) {
 #ifdef DEBUG_MODE
 #warning FIXME: Disabling SSE/AVX in srslte_vec_lut_sss
-  srslte_vec_lut_sss_simd(x, lut, y, len);
 #else
-#ifndef LV_HAVE_SSE
+#ifdef LV_HAVE_SSE
   for (int i=0;i<len;i++) {
     y[lut[i]] = x[i];
   }
 #else
-  srslte_vec_lut_sss_simd(x, lut, y, len);
+  srslte_vec_lut_sss_sse(x, lut, y, len);
 #endif
 #endif
 }
@@ -505,13 +517,19 @@ void srslte_vec_prod_fff(float *x, float *y, float *z, uint32_t len) {
 }
 
 void srslte_vec_prod_sss(short *x, short *y, short *z, uint32_t len) {
-#ifndef LV_HAVE_SSE
+
+
+#ifdef LV_HAVE_AVX
+  srslte_vec_prod_sss_avx(x,y,z,len);
+#else
+#ifdef LV_HAVE_SSE
+  srslte_vec_prod_sss_sse(x,y,z,len);
+#else
   int i;
   for (i=0;i<len;i++) {
     z[i] = x[i]*y[i];
   }
-#else
-  srslte_vec_prod_sss_simd_avx(x,y,z,len);
+#endif
 #endif
 }
 
@@ -642,15 +660,19 @@ float srslte_vec_dot_prod_fff(float *x, float *y, uint32_t len) {
 }
 
 int32_t srslte_vec_dot_prod_sss(int16_t *x, int16_t *y, uint32_t len) {
-#ifndef LV_HAVE_SSE
+#ifdef LV_HAVE_AVX
+  return srslte_vec_dot_prod_sss_avx(x, y, len);
+#else
+#ifdef LV_HAVE_SSE
+  return srslte_vec_dot_prod_sss_sse(x, y, len);
+#else
   uint32_t i;
   int32_t res = 0;
   for (i=0;i<len;i++) {
     res += x[i]*y[i];
   }
   return res;
-#else
-  return srslte_vec_dot_prod_sss_simd_avx(x, y, len); 
+#endif
 #endif
 }
 
diff --git a/srslte/lib/utils/vector_simd.c b/srslte/lib/utils/vector_simd.c
index a6b57be3a..68a34bd67 100644
--- a/srslte/lib/utils/vector_simd.c
+++ b/srslte/lib/utils/vector_simd.c
@@ -45,7 +45,7 @@
 #endif
 
 
-int srslte_vec_dot_prod_sss_simd(short *x, short *y, uint32_t len)
+int srslte_vec_dot_prod_sss_sse(short *x, short *y, uint32_t len)
 {
   int result = 0; 
 #ifdef LV_HAVE_SSE
@@ -87,7 +87,7 @@ int srslte_vec_dot_prod_sss_simd(short *x, short *y, uint32_t len)
 }
 
 
-int srslte_vec_dot_prod_sss_simd_avx(short *x, short *y, uint32_t len)
+int srslte_vec_dot_prod_sss_avx(short *x, short *y, uint32_t len)
 {
   int result = 0; 
 #ifdef LV_HAVE_AVX
@@ -127,7 +127,7 @@ int srslte_vec_dot_prod_sss_simd_avx(short *x, short *y, uint32_t len)
 
 
 
-void srslte_vec_sum_sss_simd(short *x, short *y, short *z, uint32_t len)
+void srslte_vec_sum_sss_sse(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -160,7 +160,7 @@ 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)
+void srslte_vec_sum_sss_avx(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -193,7 +193,7 @@ void srslte_vec_sum_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
 }
 
 
-void srslte_vec_sub_sss_simd(short *x, short *y, short *z, uint32_t len)
+void srslte_vec_sub_sss_sse(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -225,7 +225,7 @@ 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)
+void srslte_vec_sub_sss_avx(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_AVX
   unsigned int number = 0;
@@ -260,7 +260,7 @@ void srslte_vec_sub_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
 
 
 
-void srslte_vec_prod_sss_simd(short *x, short *y, short *z, uint32_t len)
+void srslte_vec_prod_sss_sse(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -292,7 +292,7 @@ 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)
+void srslte_vec_prod_sss_avx(short *x, short *y, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -324,7 +324,12 @@ void srslte_vec_prod_sss_simd_avx(short *x, short *y, short *z, uint32_t len)
 #endif
 }
 
-void srslte_vec_sc_div2_sss_simd(short *x, int k, short *z, uint32_t len)
+
+
+
+
+
+void srslte_vec_sc_div2_sss_sse(short *x, int k, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -354,7 +359,7 @@ 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)
+void srslte_vec_sc_div2_sss_avx(short *x, int k, short *z, uint32_t len)
 {
 #ifdef LV_HAVE_AVX
   unsigned int number = 0;
@@ -384,8 +389,10 @@ void srslte_vec_sc_div2_sss_simd_avx(short *x, int k, short *z, uint32_t len)
 #endif
 }
 
+
+
 /* No improvement with AVX */
-void srslte_vec_lut_sss_simd(short *x, unsigned short *lut, short *y, uint32_t len)
+void srslte_vec_lut_sss_sse(short *x, unsigned short *lut, short *y, uint32_t len)
 {
 #ifndef DEBUG_MODE
 #ifdef LV_HAVE_SSE
@@ -419,7 +426,7 @@ void srslte_vec_lut_sss_simd(short *x, unsigned short *lut, short *y, uint32_t l
 }
 
 /* Modified from volk_32f_s32f_convert_16i_a_simd2. Removed clipping */
-void srslte_vec_convert_fi_simd(float *x, int16_t *z, float scale, uint32_t len)
+void srslte_vec_convert_fi_sse(float *x, int16_t *z, float scale, uint32_t len)
 {
 #ifdef LV_HAVE_SSE
   unsigned int number = 0;
@@ -457,8 +464,8 @@ 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)
+//srslte_32fc_s32f_multiply_32fc_avx
+ void srslte_vec_mult_scalar_cf_f_avx( cf_t *z,const cf_t *x,const float h,const uint32_t len)
 {
 #ifdef LV_HAVE_AVX