On 13 February 2012 17:22, Daniel Murphy <[email protected]> wrote: > "Iain Buclaw" <[email protected]> wrote in message > news:[email protected]... >> It's the fact that it does *nothing* to change code generation, as D >> code compiles down to the same as C equivalent. > > Isn't that what a C backend is supposed to do? >
My point is that D is low level / slick enough, that it generates near-identical code to C/C++ in comparison benchmarks - recent example I can recall, someone did a speed test on vectors in GDC vs G++, and found both to be pretty much on par with each other. Then again, I wouldn't expect less when both use the same backend. >> Code represented like that is exactly what gdc emits to it's backend >> for your example, albeit, a in a little bit more simplified language >> (and a little bit more optimized once it goes through all codegen >> passes =) > > That's a good point. Maybe this effort would be better spent trying to > build avr-gdc instead of the compiler centipede. > > I think it starts with a runtime library that is written for the given architecture in mind. The compiler is already there in my opinion, and I have seen little reason for it not to be. -- Iain Buclaw *(p < e ? p++ : p) = (c & 0x0f) + '0';
