swr: 5.1->stereo double, float and s16

2.5 times faster than the generic code

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

2 files changed, 22 insertions, 0 deletions

diff --git a/libswresample/rematrix.c b/libswresample/rematrix.c
index 1b8e65a1f4..96246275b2 100644
--- a/libswresample/rematrix.c
+++ b/libswresample/rematrix.c
@@ -307,6 +307,7 @@ int swri_rematrix_init(SwrContext *s){
         *((int*)s->native_one) = 32768;
         s->mix_1_1_f = copy_s16;
         s->mix_2_1_f = sum2_s16;
+        s->mix_any_f = get_mix_any_func_s16(s);
     }else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
         s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(float));
         s->native_one    = av_mallocz(sizeof(float));
@@ -316,6 +317,7 @@ int swri_rematrix_init(SwrContext *s){
         *((float*)s->native_one) = 1.0;
         s->mix_1_1_f = copy_float;
         s->mix_2_1_f = sum2_float;
+        s->mix_any_f = get_mix_any_func_float(s);
     }else if(s->midbuf.fmt == AV_SAMPLE_FMT_DBLP){
         s->native_matrix = av_mallocz(nb_in * nb_out * sizeof(double));
         s->native_one    = av_mallocz(sizeof(double));
@@ -325,6 +327,7 @@ int swri_rematrix_init(SwrContext *s){
         *((double*)s->native_one) = 1.0;
         s->mix_1_1_f = copy_double;
         s->mix_2_1_f = sum2_double;
+        s->mix_any_f = get_mix_any_func_double(s);
     }else
         av_assert0(0);
     //FIXME quantize for integeres
diff --git a/libswresample/rematrix_template.c b/libswresample/rematrix_template.c
index 0b5123c1cb..690ea7c6c7 100644
--- a/libswresample/rematrix_template.c
+++ b/libswresample/rematrix_template.c
@@ -35,3 +35,22 @@ static void RENAME(copy)(SAMPLE *out, const SAMPLE *in, COEFF *coeffp, int index
         out[i] = R(coeff*in[i]);
 }
 
+static void RENAME(mix6to2)(SAMPLE **out, const SAMPLE **in, COEFF *coeffp, int len){
+    int i;
+
+    for(i=0; i<len; i++) {
+        INTER t = in[2][i]*coeffp[0*6+2] + in[3][i]*coeffp[0*6+3];
+        out[0][i] = R(t + in[0][i]*coeffp[0*6+0] + in[4][i]*coeffp[0*6+4]);
+        out[1][i] = R(t + in[1][i]*coeffp[1*6+1] + in[5][i]*coeffp[1*6+5]);
+    }
+}
+
+static mix_any_func_type *RENAME(get_mix_any_func)(SwrContext *s){
+    if(   s->out_ch_layout == AV_CH_LAYOUT_STEREO && (s->in_ch_layout == AV_CH_LAYOUT_5POINT1 || s->in_ch_layout == AV_CH_LAYOUT_5POINT1_BACK)
+       && s->matrix[0][2] == s->matrix[1][2] && s->matrix[0][3] == s->matrix[1][3]
+       && !s->matrix[0][1] && !s->matrix[0][5] && !s->matrix[1][0] && !s->matrix[1][4]
+    )
+        return RENAME(mix6to2);
+
+    return NULL;
+}