Hello, I thought you might be interested in following. There was this "feature" that gcc did not generate autoincrementing addresses for byte accesses. I looked at it and got it fixed imho. Now it seems to produce a bit more optimized code. Code looks good for my eyes, but please review the patch.
Also included in the patch is a fix for a case where negation of a 32-bit operand generates some unnecessary movs. Now the code generated looks again fine. But please review... The patch was for gcc-core 3.3.6 I'm afraid that I may have screwed up the diff, but the changes were quite short so it should be easy to apply that patch even by hand. -- Henri Moilanen ps. I'm sorry if this is duplicate. But it looks like gmail and sourceforge.net have had some incompatibilities...
gcc.patch
Description: Binary data
