[FFmpeg-cvslog] avfilter/af_arnndn: use scalarproduct_float() in dct function

[Paul B Mahol]

ffmpeg | branch: master | Paul B Mahol <onemda at gmail.com> | Thu Aug 27 18:49:02 2020 +0200| [ace1cc616308ef8f4aedc7814778b08eca2fd92b] | committer: Paul B Mahol

avfilter/af_arnndn: use scalarproduct_float() in dct function

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=ace1cc616308ef8f4aedc7814778b08eca2fd92b
---

 libavfilter/af_arnndn.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/libavfilter/af_arnndn.c b/libavfilter/af_arnndn.c
index d2722fb202..123bc45157 100644
--- a/libavfilter/af_arnndn.c
+++ b/libavfilter/af_arnndn.c
@@ -141,7 +141,7 @@ typedef struct AudioRNNContext {
     DenoiseState *st;
 
     DECLARE_ALIGNED(32, float, window)[WINDOW_SIZE];
-    float dct_table[NB_BANDS*NB_BANDS];
+    DECLARE_ALIGNED(32, float, dct_table)[FFALIGN(NB_BANDS, 4)][FFALIGN(NB_BANDS, 4)];
 
     RNNModel *model;
 
@@ -991,11 +991,9 @@ static void pitch_search(const float *x_lp, float *y,
 static void dct(AudioRNNContext *s, float *out, const float *in)
 {
     for (int i = 0; i < NB_BANDS; i++) {
-        float sum = 0.f;
+        float sum;
 
-        for (int j = 0; j < NB_BANDS; j++) {
-            sum += in[j] * s->dct_table[j * NB_BANDS + i];
-        }
+        sum = s->fdsp->scalarproduct_float(in, s->dct_table[i], FFALIGN(NB_BANDS, 4));
         out[i] = sum * sqrtf(2.f / 22);
     }
 }
@@ -1006,7 +1004,7 @@ static int compute_frame_features(AudioRNNContext *s, DenoiseState *st, AVComple
     float E = 0;
     float *ceps_0, *ceps_1, *ceps_2;
     float spec_variability = 0;
-    float Ly[NB_BANDS];
+    LOCAL_ALIGNED_32(float, Ly, [NB_BANDS]);
     LOCAL_ALIGNED_32(float, p, [WINDOW_SIZE]);
     float pitch_buf[PITCH_BUF_SIZE>>1];
     int pitch_index;
@@ -1348,7 +1346,7 @@ static float rnnoise_channel(AudioRNNContext *s, DenoiseState *st, float *out, c
     AVComplexFloat P[WINDOW_SIZE];
     float x[FRAME_SIZE];
     float Ex[NB_BANDS], Ep[NB_BANDS];
-    float Exp[NB_BANDS];
+    LOCAL_ALIGNED_32(float, Exp, [NB_BANDS]);
     float features[NB_FEATURES];
     float g[NB_BANDS];
     float gf[FREQ_SIZE];
@@ -1476,9 +1474,9 @@ static av_cold int init(AVFilterContext *ctx)
 
     for (int i = 0; i < NB_BANDS; i++) {
         for (int j = 0; j < NB_BANDS; j++) {
-            s->dct_table[i*NB_BANDS + j] = cosf((i + .5f) * j * M_PI / NB_BANDS);
+            s->dct_table[j][i] = cosf((i + .5f) * j * M_PI / NB_BANDS);
             if (j == 0)
-                s->dct_table[i*NB_BANDS + j] *= sqrtf(.5);
+                s->dct_table[j][i] *= sqrtf(.5);
         }
     }