nearest neighbor scaling, I noticed that the bilinear version
actually used less CPU than the generic AGG code path with nn
scaling. So I wrote an optimized nn scaling routine for nn
based on the bilinear scaling code. So the indices into the
source bitmap are cached. I don't know if this is the optimal
nn scaling routine, but the CPU usage dropped significantly.
Only B_OP_COPY is optimized as of yet.
* Optimized the bilinear scaling. When more filtered pixels than
unfiltered pixels are anticipated, the loops are unrolled to
special case the very last row/column and bottom right pixel.
This eliminates the branches in the loops.
* Fixed a bug with partial scaled drawing of bitmaps when it
used the bilinear scaling, the bitmapShift was in the wrong
direction.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@27169 a95241bf-73f2-0310-859d-f6bbb57e9c96
Stephan Aßmus
c1111e046b