On Wed, Jul 17, 2002 at 01:34:38PM -0400, Eric Thomas wrote: > > I went through this same mystery a few months ago. Apparently GCC pads > the generated assembly for better memory alignment, which makes certain > operations faster. And in some archs, legal. Non-align memory access is illegal in some archs, and non-existant in others (MIPS ignores the last 2/4(?) bits, anyway). In those cases the OS or the compiler has to access the two words were the data is stored. That's not a minor speed penalty... Anyway, whoever's playing with gcc generated assembly code should compile the program explicity without optimizations (-O0) or maybe with optimizations for size (-Os) (not sure about data size reduction). Regards, Luciano Rocha
This archive was generated by hypermail 2b30 : Wed Jul 17 2002 - 18:06:52 PDT