[svn-commits] alecdavis: branch 1.8 r372212 - /branches/1.8/main/dsp.c

[SVN commits to the Digium repositories]

Author: alecdavis
Date: Wed Sep  5 01:45:43 2012
New Revision: 372212

URL: http://svnview.digium.com/svn/asterisk?view=rev&rev=372212
Log:
dsp.c: optimize goerztzel sample loops, in dtmf_detect, mf_detect and tone_detect

use a temporary short int when repeatedly used to call goertzel_sample.

alecdavis (license 585)
Reported by: alecdavis
Tested by: alecdavis
 
Review: https://reviewboard.asterisk.org/r/2093/


Modified:
    branches/1.8/main/dsp.c

Modified: branches/1.8/main/dsp.c
URL: http://svnview.digium.com/svn/asterisk/branches/1.8/main/dsp.c?view=diff&rev=372212&r1=372211&r2=372212
==============================================================================
--- branches/1.8/main/dsp.c (original)
+++ branches/1.8/main/dsp.c Wed Sep  5 01:45:43 2012
@@ -533,6 +533,7 @@
 	int limit;
 	int res = 0;
 	int16_t *ptr;
+	short samp;
 	int start, end;
 	fragment_t mute = {0, 0};
 
@@ -550,10 +551,11 @@
 		end = start + limit;
 
 		for (i = limit, ptr = amp ; i > 0; i--, ptr++) {
+			samp = *ptr;
 			/* signed 32 bit int should be enough to suqare any possible signed 16 bit value */
-			s->energy += (int32_t) *ptr * (int32_t) *ptr;
-
-			goertzel_sample(&s->tone, *ptr);
+			s->energy += (int32_t) samp * (int32_t) samp;
+
+			goertzel_sample(&s->tone, samp);
 		}
 
 		s->samples_pending -= limit;
@@ -646,10 +648,10 @@
 {
 	float row_energy[4];
 	float col_energy[4];
-	float famp;
 	int i;
 	int j;
 	int sample;
+	short samp;
 	int best_row;
 	int best_col;
 	int hit;
@@ -672,18 +674,18 @@
 		/* The following unrolled loop takes only 35% (rough estimate) of the 
 		   time of a rolled loop on the machine on which it was developed */
 		for (j = sample; j < limit; j++) {
-			famp = amp[j];
-			s->td.dtmf.energy += famp*famp;
+			samp = amp[j];
+			s->td.dtmf.energy += (int32_t) samp * (int32_t) samp;
 			/* With GCC 2.95, the following unrolled code seems to take about 35%
 			   (rough estimate) as long as a neat little 0-3 loop */
-			goertzel_sample(s->td.dtmf.row_out, amp[j]);
-			goertzel_sample(s->td.dtmf.col_out, amp[j]);
-			goertzel_sample(s->td.dtmf.row_out + 1, amp[j]);
-			goertzel_sample(s->td.dtmf.col_out + 1, amp[j]);
-			goertzel_sample(s->td.dtmf.row_out + 2, amp[j]);
-			goertzel_sample(s->td.dtmf.col_out + 2, amp[j]);
-			goertzel_sample(s->td.dtmf.row_out + 3, amp[j]);
-			goertzel_sample(s->td.dtmf.col_out + 3, amp[j]);
+			goertzel_sample(s->td.dtmf.row_out, samp);
+			goertzel_sample(s->td.dtmf.col_out, samp);
+			goertzel_sample(s->td.dtmf.row_out + 1, samp);
+			goertzel_sample(s->td.dtmf.col_out + 1, samp);
+			goertzel_sample(s->td.dtmf.row_out + 2, samp);
+			goertzel_sample(s->td.dtmf.col_out + 2, samp);
+			goertzel_sample(s->td.dtmf.row_out + 3, samp);
+			goertzel_sample(s->td.dtmf.col_out + 3, samp);
 		}
 		s->td.dtmf.current_sample += (limit - sample);
 		if (s->td.dtmf.current_sample < DTMF_GSIZE) {
@@ -801,6 +803,7 @@
 	int i;
 	int j;
 	int sample;
+	short samp;
 	int hit;
 	int limit;
 	fragment_t mute = {0, 0};
@@ -824,12 +827,13 @@
 		for (j = sample;  j < limit;  j++) {
 			/* With GCC 2.95, the following unrolled code seems to take about 35%
 			   (rough estimate) as long as a neat little 0-3 loop */
-			goertzel_sample(s->td.mf.tone_out, amp[j]);
-			goertzel_sample(s->td.mf.tone_out + 1, amp[j]);
-			goertzel_sample(s->td.mf.tone_out + 2, amp[j]);
-			goertzel_sample(s->td.mf.tone_out + 3, amp[j]);
-			goertzel_sample(s->td.mf.tone_out + 4, amp[j]);
-			goertzel_sample(s->td.mf.tone_out + 5, amp[j]);
+			samp = amp[j];
+			goertzel_sample(s->td.mf.tone_out, samp);
+			goertzel_sample(s->td.mf.tone_out + 1, samp);
+			goertzel_sample(s->td.mf.tone_out + 2, samp);
+			goertzel_sample(s->td.mf.tone_out + 3, samp);
+			goertzel_sample(s->td.mf.tone_out + 4, samp);
+			goertzel_sample(s->td.mf.tone_out + 5, samp);
 		}
 		s->td.mf.current_sample += (limit - sample);
 		if (s->td.mf.current_sample < MF_GSIZE) {