Diederik van Lierop wrote:
... For now I've fixed my patch (see the attachement) by limiting it to the non-MMX c++ code. This fixes the bitmap pixel alignment issue for some users, but not for everyone. There must be someone reading this who can port this to the assembly code too. It really is only a few lines of code! Until then this bug cannot be completely solved, .... unless we start using Cairo (?)
I would recommend timing to see whether the MMX code is (still) really much faster than the normal C++ code. If it's not significantly faster than the old 32bit C++ code I would go with the higher quality code.
As for porting the code to assembly, if it is indeed making use of MMX I have doubts about the feasibility/usefulness of porting the code, as the assembly would likely not be as efficient for 64bit integers as it would be for 32bit integers (assuming the right instructions do exist). But of course this would have to be tested.