Dick Hollenbeck wrote: >> That could be worked around by using C99 initializers. I am not fond of >> replacing a small amount of static data with a bigger chunk of code. >> > > Are you thinking it is bigger because it looks bigger in your editor, > or bigger because it generates more code? Switch statements can be > handled quite efficiently by the compiler if the case statements are > of contiguous values, and I believe that is the case we are talking > about. Sure? The best I have seen are jump tables, which means 1 pointer + 1 load instruction + 1 branch/return per case, compared with a single indexed memory load - this looks quite bloaty to me, not only in terms of code size, but also execution speed (pipeline flush). A short test using gcc 4.3 (i386) and 4.2.3 (arm) seems to confirm this suspection.
> C99: are you saying use #defines as the array indices in the > initializers? > That could be an acceptable alternative. That was the general idea. If it is not sufficient, I would next suggest trying compile-time asserts (which add zero overhead/code) to make sure breakage does not go undetected. cu Michael _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
