Fix overflow when using 32-bit ImDrawIdx

[bear24rw]

For backends that don't support VtxOffset (the imgui emscripten example) we are forced to use a 32bit index buffer instead. This patch fixes an issue from #41 where the calculation of the maximum index would overflow. The benchmark plot now renders correctly in the emscripten example when compiling with -DImDrawIdx="unsigned int".

[epezent]

Thanks for the fix! Can you make max_idx either static or a compile time constant so it's not reevaluated each time.

[bear24rw]

Good idea, I just force pushed that change.

[epezent]

@bear24rw , I don't know if you noticed, but I ended up changing your PR slightly. I replaced your ImPow calculation with some template code:

template <typename T>
struct MaxIdx { static const unsigned int Value; };
template <> const unsigned int MaxIdx<unsigned short>::Value = 65535;
template <> const unsigned int MaxIdx<unsigned int>::Value   = 4294967295;

...
unsigned int cnt = ImMin(prims, (MaxIdx<ImDrawIdx>::Value - DrawList._VtxCurrentIdx) / Renderer::VtxConsumed);
...

Does this still work as expected for you?

[bear24rw]

Does this still work as expected for you?

Yes!