change qscale -> lambda for the motion estimation

finetune bit/distortion weighting factor used in motion estimation, the old coeffs where finetuned relative to incorrect mv_penalty tables which where then fixed later but the coeffs where not
this _may_ fix the long standing blocking artifacts, but may also introduce mudding artefacts theoretically, so please tell us if u stumble across any so we can either fix them or export this variable so the user can change it

Originally committed as revision 3189 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 6 insertions, 7 deletions

diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c
index d2ab2b2e65..b92b6dfad8 100644
--- a/libavcodec/motion_est.c
+++ b/libavcodec/motion_est.c
@@ -223,18 +223,17 @@ static inline int get_penalty_factor(MpegEncContext *s, int type){
     switch(type&0xFF){
     default:
     case FF_CMP_SAD:
-        return s->qscale*2;
+        return s->lambda>>FF_LAMBDA_SHIFT;
     case FF_CMP_DCT:
-        return s->qscale*3;
+        return (3*s->lambda)>>(FF_LAMBDA_SHIFT+1);
     case FF_CMP_SATD:
-        return s->qscale*6;
+        return (2*s->lambda)>>FF_LAMBDA_SHIFT;
+    case FF_CMP_RD:
+    case FF_CMP_PSNR:
     case FF_CMP_SSE:
-        return s->qscale*s->qscale*2;
+        return s->lambda2>>FF_LAMBDA_SHIFT;
     case FF_CMP_BIT:
         return 1;
-    case FF_CMP_RD:
-    case FF_CMP_PSNR:
-        return (s->qscale*s->qscale*185 + 64)>>7;
     }
 }