Speed up silk_warped_autocorrelation_FLP()

Reducing the dependency chain between tmp1 and tmp2 at the
cost of an extra multiply.

1 files changed, 4 insertions, 2 deletions

diff --git a/silk/float/warped_autocorrelation_FLP.c b/silk/float/warped_autocorrelation_FLP.c
index 09186e73..116dab92 100644
--- a/silk/float/warped_autocorrelation_FLP.c
+++ b/silk/float/warped_autocorrelation_FLP.c
@@ -54,11 +54,13 @@ void silk_warped_autocorrelation_FLP(
         /* Loop over allpass sections */
         for( i = 0; i < order; i += 2 ) {
             /* Output of allpass section */
-            tmp2 = state[ i ] + warping * ( state[ i + 1 ] - tmp1 );
+            /* We voluntarily use two multiples instead of factoring the expression to
+               reduce the length of the dependency chain (tmp1->tmp2->tmp1... ). */
+            tmp2 = state[ i ] + warping * state[ i + 1 ] - warping * tmp1;
             state[ i ] = tmp1;
             C[ i ] += state[ 0 ] * tmp1;
             /* Output of allpass section */
-            tmp1 = state[ i + 1 ] + warping * ( state[ i + 2 ] - tmp2 );
+            tmp1 = state[ i + 1 ] + warping * state[ i + 2 ] - warping * tmp2;
             state[ i + 1 ] = tmp2;
             C[ i + 1 ] += state[ 0 ] * tmp2;
         }