SpritesMind.Net

Stef wrote:From my last checks out was getting worse, but maybe absolute latter version are better, I've to admit I didn't checked from sometime now.
Of course for better test you need to benchmark 100% C code, I say that because sgdk uses assembly sometime.

Yeah, we'd want to call a C function that did a certain amount of work, with the only ints being a simple vint that incremented a ticks var. Something like that. How about those old BYTE benchmarks?

Yeah that could be a good solution.
It's better to have different kind of benchmark (memory copy / memory compare, basic operation, line drawing...).
Maybe i could integrate a benchmark sample in SGDK (based on what we can find around) so who want to test the compiler and flags optimization impact could use it =)

If you make some kind of benchmarks, I'll convert it to work on my toolchain and try it under 4.8.3 at all the different O levels.

And yes, we want a variety of things that may show where the compiler's strengths lay.