Hello,
Converting an unsigned integer to a signed integer is easy in C. The
implicit conversion works fine. However, converting from a signed
integer to an unsigned integer invokes implementation defined
behaviour (or can raise an implementation defined signal) if "the
value cannot be represented in it". So I wrote some simple code to
convert a uint_fast32_t to an int_fast32_t.
if (positive > (int_fast64_t) INT32_MAX) {
return -((UINT32_MAX - (int_fast64_t) positive) + 1u);
}
return positive;
Unfortunately, GCC does not optimize this away the comparison and
calculations on X86_64. This is not a bug report. I have not checked
to see if the latest version of GCC catches this. Besides, Clang does
not catch this either and so maybe this optimization would be
difficult to implement. What I want to know is, how can I make this
code friendly to GCC's optimizer?
Maybe if I need the best performance I would need to rely upon
implementation defined behaviour or use a builtin intrinsic. But
before I look into those options I thought I'd check with the mailing
list.
