The Highest-Level Feature of C

[BartC]

Minor nit: unless the compiler can prove that the lower 10 bits of x is
zero, it'd have to do something like:

if ( (x & 0x3ff) != 0) {

More of a major nit if it doesn't work! Then the compiler would have to do
more work to decide whether the optimisation is worth doing.

 

[BartC]

Minor nit: unless the compiler can prove that the lower 10 bits of x is
zero, it'd have to do something like:

if ( (x & 0x3ff) != 0) {
/* Default case */
} else switch (x>>10) {
case 1:
case 2:
case 3:
}

The jump table method already requires the compiler to test the value to
be sure it is in range before using the jump table (to make sure it uses
an item IN the jump table), doing something like:

if((x & 0xF3FF) != 0) go to default
else go to table[x>>10] with entries in the table for 0, 1, 2, and 3

Wouldn't it still need a check for being in range after (or even before)
shifting?

would be nearly as efficient as the case of 1, 2, and 3, which would
need a comparison for x <= 3 (and > or >= 0).

(I've used a single unsigned compare here, when the jump-table includes an
entry for x=0.)