Hmm, this is likely to be slower, isn't it? Though I've not tried a benchmark ahem. C always uses double for all intermediate values, so:

    float a, b = ... something;
    float c = a * b;

Compiles to something like:

    float c = double_to_float(double_multiply(float_to_double(a), float_to_double(b)));

If you just use double everywhere you'll save a lot of unnecessary conversions.

float can useful for specifying storage (eg. float a[10] is nice and small), but you should generally use double everywhere else.

Actually, I say that, but now I can't find a reference to float -> double promotion in the standard. I'll keep looking.

Yes, this might be a C99 thing. I think the problem here is that t is integer and p could be a float, which isn't automatically promoted. I've updated the PR to explicitly cast the multiplication up to double precision instead.

I tried with compiler explorer and this:

float fred(float a, float b, float c) { return a * b + c; }

Generates single-precision float mul and add instructions on all compilers I tested, so I'm wrong heh. Perhaps I was thinking of the promotion rules around untyped arguments.