optimize quantizaton (about 3x faster)

further opt is easily possible but could lead to overflows depening upon coefficient range, so this wont be done yet as it would make the code somewhat less flexible


git-svn-id: file:///var/local/repositories/ffmpeg/trunk@3354 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b

libavcodec/snow.c

@@ -1801,29 +1801,33 @@ static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int b
     const int h= b->height;
     const int qlog= clip(s->qlog + b->qlog, 0, 128);
     const int qmul= qexp[qlog&7]<<(qlog>>3);
-    int x,y;
+    int x,y, thres1, thres2;
+    START_TIMER
 
     assert(QROOT==8);
 
     bias= bias ? 0 : (3*qmul)>>3;
+    thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
+    thres2= 2*thres1;
     
     if(!bias){
         for(y=0; y<h; y++){
             for(x=0; x<w; x++){
                 int i= src[x + y*stride];
-                //FIXME use threshold
-                //FIXME optimize
-                //FIXME bias
+                
+                if((unsigned)(i+thres1) > thres2){
                     if(i>=0){
                         i<<= QEXPSHIFT;
-                    i/= qmul;
+                        i/= qmul; //FIXME optimize
                         src[x + y*stride]=  i;
                     }else{
                         i= -i;
                         i<<= QEXPSHIFT;
-                    i/= qmul;
+                        i/= qmul; //FIXME optimize
                         src[x + y*stride]= -i;
                     }
+                }else
+                    src[x + y*stride]= 0;
             }
         }
     }else{
@@ -1831,21 +1835,24 @@ static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int b
             for(x=0; x<w; x++){
                 int i= src[x + y*stride]; 
                 
-                //FIXME use threshold
-                //FIXME optimize
-                //FIXME bias
+                if((unsigned)(i+thres1) > thres2){
                     if(i>=0){
                         i<<= QEXPSHIFT;
-                    i= (i + bias) / qmul;
+                        i= (i + bias) / qmul; //FIXME optimize
                         src[x + y*stride]=  i;
                     }else{
                         i= -i;
                         i<<= QEXPSHIFT;
-                    i= (i + bias) / qmul;
+                        i= (i + bias) / qmul; //FIXME optimize
                         src[x + y*stride]= -i;
                     }
+                }else
+                    src[x + y*stride]= 0;
+            }
         }
     }
+    if(level+1 == s->spatial_decomposition_count){
+//        STOP_TIMER("quantize")
     }
 }