Question

I always subtract by 32768 when I have an unsigned short that I want to convert to a signed one.

Is that the fastest way to do it, or are there faster ways?

Answer 1

If the value is between from 0 to SHRT_MAX (inclusive), there’s nothing to worry about and the cast ((short)) is optional (unless your compiler is paranoid or configured to be paranoid).

If the unsigned short value can be greater than SHRT_MAX, about the only legal way to convert it to short is:

#include <limits.h>

short ushort2short(unsigned short s)
{
  if (s <= SHRT_MAX)
    return s; // or return (short)s;
  s -= SHRT_MAX + 1; // now s is 0 ... SHRT_MAX
  return (short)s - SHRT_MAX - 1;
}

This, of course, relies on signed shorts being 2’s complement with SHRT_MIN = –SHRT_MAX – 1.

A modern compiler will optimize away all the nonsense inside this function and just generate code to return s.

EDIT: Compiled the above with gcc 4.6.2 as gcc -Wall ush2sh.c -O2 -S -o ush2sh.s to this assembly code:

        .file   "ush2sh.c"
        .text
        .p2align 2,,3
        .globl  _ushort2short
        .def    _ushort2short;  .scl    2;      .type   32;     .endef
_ushort2short:
LFB0:
        .cfi_startproc
        movl    4(%esp), %eax
        ret
        .cfi_endproc
LFE0: