Make the ResampleEffect minimum tolerated error smaller; I could really see some...

diff --git a/resample_effect.cpp b/resample_effect.cpp
index e1e6199ec5b5b857add0df623d6fd05a25d4b1e0..5ab0468d41122c8d367e8d4907c5cdd2d5bc4a41 100644 (file)
--- a/resample_effect.cpp
+++ b/resample_effect.cpp
@@ -74,9 +74,12 @@ unsigned combine_samples(float *src, float *dst, unsigned num_src_samples, unsig
                float offset, total_weight, sum_sq_error;
                combine_two_samples(w1, w2, &offset, &total_weight, &sum_sq_error);
 
                float offset, total_weight, sum_sq_error;
                combine_two_samples(w1, w2, &offset, &total_weight, &sum_sq_error);
 

-               // If the interpolation error is larger than that of one level
-               // at 8-bit precision, don't combine.
-               if (sum_sq_error > 1.0f / (256.0f * 256.0f)) {
+               // If the interpolation error is larger than that of about sqrt(2) of
+               // a level at 8-bit precision, don't combine. (You'd think 1.0 was enough,
+               // but since the artifacts are not really random, they can get quite
+               // visible. On the other hand, going to 0.25f, I can see no change at
+               // all with 8-bit output, so it would not seem to be worth it.)
+               if (sum_sq_error > 0.5f / (256.0f * 256.0f)) {

                        continue;
                }
 
                        continue;
                }