Kyrill Tkachov wrote:
> On 25/04/16 20:21, Wilco Dijkstra wrote:
> > The GCC switch expansion is awful, so
> > even with a good indirect predictor it is better to use conditional
> > branches.
>
> In what way is it awful? If there's something we can do better at
> can you file a bug report with a testcase so that we can work on
> improving it rather than tweaking a heuristic in the backend.
In every way :-(
Try this simple example and see whether you agree this is terrible,
especially on targets that don't use PC-relative tables...
int i;
int func(int a)
{
switch(a)
{
case 0: i = 20; break;
case 1: i = 50; break;
case 2: i = 29; break;
case 3: i = 20; break;
case 4: i = 50; break;
case 5: i = 29; break;
case 6: i = 20; break;
case 7: i = 50; break;
case 8: i = 29; break;
case 9: i = 79; break;
case 110: i = 27; break;
default: i = 77; break;
}
return i;
}
See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70861.
It's also disappointing this doesn't get turned into:
return i = (unsigned)a <= 110 ? table[a] : 77;
Wilco
