[FFmpeg-cvslog] swr: work with 4 noise shaping coeffs at a time

[Michael Niedermayer]

ffmpeg | branch: master | Michael Niedermayer <michaelni at gmx.at> | Sat Jan 12 19:11:08 2013 +0100| [c8737d348be526d6403856f424a73ad1c81b157f] | committer: Michael Niedermayer

swr: work with 4 noise shaping coeffs at a time

63->38 kcycles

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=c8737d348be526d6403856f424a73ad1c81b157f
---

 libswresample/dither_template.c |   11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/libswresample/dither_template.c b/libswresample/dither_template.c
index 8cf0c57..ecbe55c 100644
--- a/libswresample/dither_template.c
+++ b/libswresample/dither_template.c
@@ -29,6 +29,9 @@ void RENAME(swri_noise_shaping)(SwrContext *s, AudioData *dsts, const AudioData
     float S   = s->dither.ns_scale;
     float S_1 = s->dither.ns_scale_1;
 
+    av_assert2((taps&3) != 2);
+    av_assert2((taps&3) != 3 || s->dither.ns_coeffs[taps] == 0);
+
     for (ch=0; ch<srcs->ch_count; ch++) {
         const float *noise = ((const float *)noises->ch[ch]) + s->dither.noise_pos;
         const DELEM *src = (const DELEM*)srcs->ch[ch];
@@ -38,7 +41,13 @@ void RENAME(swri_noise_shaping)(SwrContext *s, AudioData *dsts, const AudioData
         pos  = s->dither.ns_pos;
         for (i=0; i<count; i++) {
             double d1, d = src[i]*S_1;
-            for(j=0; j<taps; j++)
+            for(j=0; j<taps-2; j+=4) {
+                d -= ns_coeffs[j    ] * ns_errors[pos + j    ]
+                    +ns_coeffs[j + 1] * ns_errors[pos + j + 1]
+                    +ns_coeffs[j + 2] * ns_errors[pos + j + 2]
+                    +ns_coeffs[j + 3] * ns_errors[pos + j + 3];
+            }
+            if(j < taps)
                 d -= ns_coeffs[j] * ns_errors[pos + j];
             pos = pos ? pos - 1 : taps - 1;
             d1 = rint(d + noise[i]);