Cisco optimization for x86 & fixed point

1. Only for fixed point on x86 platform (32bit and 64bit, uses SIMD
   intrinsics up to SSE4.2)
2. Use "configure --enable-fixed-point --enable-intrinsics" to enable
   optimization, default is disabled.
3. Official test cases are verified and passed.

Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>

4 files changed, 10 insertions, 7 deletions

diff --git a/silk/float/encode_frame_FLP.c b/silk/float/encode_frame_FLP.c
index d54e2686..2092a4d9 100644
--- a/silk/float/encode_frame_FLP.c
+++ b/silk/float/encode_frame_FLP.c
@@ -47,7 +47,7 @@ void silk_encode_do_VAD_FLP(
     /****************************/
     /* Voice Activity Detection */
     /****************************/
-    silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1 );
+    silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch );
 
     /**************************************************/
     /* Convert speech activity into VAD and DTX flags */
diff --git a/silk/float/find_pred_coefs_FLP.c b/silk/float/find_pred_coefs_FLP.c
index ea2c6c43..1d2dd72c 100644
--- a/silk/float/find_pred_coefs_FLP.c
+++ b/silk/float/find_pred_coefs_FLP.c
@@ -67,7 +67,8 @@ void silk_find_pred_coefs_FLP(
 
         /* Quantize LTP gain parameters */
         silk_quant_LTP_gains_FLP( psEncCtrl->LTPCoef, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex,
-            &psEnc->sCmn.sum_log_gain_Q7, WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr );
+            &psEnc->sCmn.sum_log_gain_Q7, WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr,
+            psEnc->sCmn.arch );
 
         /* Control LTP scaling */
         silk_LTP_scale_ctrl_FLP( psEnc, psEncCtrl, condCoding );
diff --git a/silk/float/main_FLP.h b/silk/float/main_FLP.h
index fb553b61..e5a75972 100644
--- a/silk/float/main_FLP.h
+++ b/silk/float/main_FLP.h
@@ -205,7 +205,8 @@ void silk_quant_LTP_gains_FLP(
     const silk_float                W[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I    Error weights                        */
     const opus_int                  mu_Q10,                             /* I    Mu value (R/D tradeoff)                     */
     const opus_int                  lowComplexity,                      /* I    Flag for low complexity                     */
-    const opus_int                  nb_subfr                            /* I    number of subframes                         */
+    const opus_int                  nb_subfr,                           /* I    number of subframes                         */
+    int                             arch                                /* I    Run-time architecture                       */
 );
 
 /* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */
diff --git a/silk/float/wrappers_FLP.c b/silk/float/wrappers_FLP.c
index 350599b2..6666b8ef 100644
--- a/silk/float/wrappers_FLP.c
+++ b/silk/float/wrappers_FLP.c
@@ -161,10 +161,10 @@ void silk_NSQ_wrapper_FLP(
     /* Call NSQ */
     if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) {
         silk_NSQ_del_dec( &psEnc->sCmn, psNSQ, psIndices, x_Q3, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14,
-            AR2_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14 );
+            AR2_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch );
     } else {
         silk_NSQ( &psEnc->sCmn, psNSQ, psIndices, x_Q3, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14,
-            AR2_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14 );
+            AR2_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch );
     }
 }
 
@@ -179,7 +179,8 @@ void silk_quant_LTP_gains_FLP(
     const silk_float                W[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I    Error weights                        */
     const opus_int                  mu_Q10,                             /* I    Mu value (R/D tradeoff)                     */
     const opus_int                  lowComplexity,                      /* I    Flag for low complexity                     */
-    const opus_int                  nb_subfr                            /* I    number of subframes                         */
+    const opus_int                  nb_subfr,                           /* I    number of subframes                         */
+    int                             arch                                /* I    Run-time architecture                       */
 )
 {
     opus_int   i;
@@ -193,7 +194,7 @@ void silk_quant_LTP_gains_FLP(
         W_Q18[ i ] = (opus_int32)silk_float2int( W[ i ] * 262144.0f );
     }
 
-    silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, sum_log_gain_Q7, W_Q18, mu_Q10, lowComplexity, nb_subfr );
+    silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, sum_log_gain_Q7, W_Q18, mu_Q10, lowComplexity, nb_subfr, arch );
 
     for( i = 0; i < nb_subfr * LTP_ORDER; i++ ) {
         B[ i ] = (silk_float)B_Q14[ i ] * ( 1.0f / 16384.0f );