Normalize the resample weight after bilinear combining.
authorSteinar H. Gunderson <sgunderson@bigfoot.com>
Sat, 25 Jan 2014 18:41:40 +0000 (19:41 +0100)
committerSteinar H. Gunderson <sgunderson@bigfoot.com>
Sat, 25 Jan 2014 20:14:07 +0000 (21:14 +0100)
commit0d58598f7c593c34de04ec1a0e4849f5cfb44792
treed9e7d5ee83025e4672d17a980c6df7c9521e140c
parent88ee2edc41d536c7057424df185fc6b49f993896
Normalize the resample weight after bilinear combining.

We introduce a small bit of error in the combining (due to having to
compensate for lack of subpixel sampling precision), so normalize
after it rather than before it. Also, do a second normalization pass,
which seemingly helps sometimes (probably due to inaccuracies in the
float sum).

This seems to kill about half the precision loss on Intel, at least.
resample_effect.cpp