Clang: fix -Wimplicit-const-int-float-conversion

Some of these integers exceed the precision of float. In
that case the number is rounded. The rounding shouldn't
matter too much anyway, so just shut the compiler up.