From 5eeee2c07cf0852aaf714c9ab6ca5cd1bfefb3af Mon Sep 17 00:00:00 2001
From: Sandesh Venkatesh <sandesh.venkatesh@ittiam.com>
Date: Tue, 1 Oct 2024 12:05:52 +0530
Subject: [PATCH] BASOP optimizations

[x] sum_32_fx 64-bit operator usage
[x] MAC basop usage in TNSAnalysisStereo
---
 lib_com/swb_bwe_com_fx.c           |  6 +++---
 lib_com/tools_fx.c                 | 20 ++++++++++----------
 lib_enc/cod_tcx.c                  |  4 ++--
 lib_enc/ivas_core_pre_proc_front.c |  2 +-
 4 files changed, 16 insertions(+), 16 deletions(-)

diff --git a/lib_com/swb_bwe_com_fx.c b/lib_com/swb_bwe_com_fx.c
index 29662510a..0025812db 100644
--- a/lib_com/swb_bwe_com_fx.c
+++ b/lib_com/swb_bwe_com_fx.c
@@ -543,7 +543,7 @@ void calc_normal_length_fx(
             pit++;
         }
 
-        L_tmp = L_mult0( peak, add( 15, THRES ) ); /*Q_syn */
+        L_tmp = L_mult0( peak, 15 + THRES ); /*Q_syn */
         IF( EQ_16( THRES, 8 ) )
         {
             L_tmp1 = L_shl( Mpy_32_16_1( L_mean, 32767 ), 3 );
@@ -877,7 +877,7 @@ void calc_norm_envelop_fx(
         }
         envelope[env_index] = L_tmp;
         move32();
-        env_index = add( env_index, 1 );
+        env_index++;
     }
 
     i = 0;
@@ -896,7 +896,7 @@ void calc_norm_envelop_fx(
         }
         envelope[env_index] = L_tmp;
         move32();
-        env_index = add( env_index, 1 );
+        env_index++;
         i = add( i, 1 );
     }
 
diff --git a/lib_com/tools_fx.c b/lib_com/tools_fx.c
index e65652e03..88f0a896f 100644
--- a/lib_com/tools_fx.c
+++ b/lib_com/tools_fx.c
@@ -423,22 +423,22 @@ Word32 sum_32_fx(
     const Word16 lvec, /* i  : length of input vector                */
     Word16 *e )
 {
-    Word16 i;
-    Word32 tmp;
+    Word16 i, shift;
+    Word64 tmp = 0;
+    move64();
+    Word32 ans;
 
-    tmp = 0;
-    Word16 exp = 0;
-    move32();
-    move16();
     FOR( i = 0; i < lvec; i++ )
     {
-        tmp = BASOP_Util_Add_Mant32Exp( tmp, exp, vec[i], *e, &exp );
+        tmp = W_add( tmp, vec[i] ); // e
     }
-
-    *e = exp;
+    shift = W_norm( tmp );
+    tmp = W_shl( tmp, shift ); // shift + (31 - e)
+    ans = W_extract_h( tmp );  // shift + (31 - e) - 32
+    *e = add( sub( *e, shift ), 32 );
     move16();
 
-    return tmp;
+    return ans;
 }
 
 Word32 sum2_fx_mod(                    /* o  : sum of all squared vector elements    Q(2x+1 -5)*/
diff --git a/lib_enc/cod_tcx.c b/lib_enc/cod_tcx.c
index 2acc3bab6..4508bfb64 100644
--- a/lib_enc/cod_tcx.c
+++ b/lib_enc/cod_tcx.c
@@ -278,8 +278,8 @@ void TNSAnalysisStereo_fx(
                          * both filters for the decision
                          */
 
-                        meanPredictionGain_fx = add( shr( pFilter[0]->predictionGain, 1 ), shr( pFilter[1]->predictionGain, 1 ) ); // Q7
-                        maxPredictionGain_fx = s_max( maxPredictionGain_fx, meanPredictionGain_fx );                               // Q7
+                        meanPredictionGain_fx = mac_r( L_mult( pFilter[0]->predictionGain, 16384 ), pFilter[1]->predictionGain, 16384 ); // Q7
+                        maxPredictionGain_fx = s_max( maxPredictionGain_fx, meanPredictionGain_fx );                                     // Q7
 
                         test();
                         test();
diff --git a/lib_enc/ivas_core_pre_proc_front.c b/lib_enc/ivas_core_pre_proc_front.c
index 72d95d33d..115f1aae3 100644
--- a/lib_enc/ivas_core_pre_proc_front.c
+++ b/lib_enc/ivas_core_pre_proc_front.c
@@ -2073,7 +2073,7 @@ ivas_error pre_proc_front_ivas_fx(
     Word16 stab_fac_fx;
     Word16 Q_new_loc;
     Word16 Q_r[2] = { 0 };
-    Word16 A_loc_fx[NB_SUBFR16k * ( M + 1 )];
+    Word16 A_loc_fx[NB_SUBFR16k * ( M + 1 )] = { 0 };
     Word16 epsP_l_loc[M + 1];
     Word16 epsP_h_loc[M + 1];
     Word32 ener_fx;
-- 
GitLab