[Bug rtl-optimization/116002] GCC Compiler time-hog with large basic block in Function

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116002

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #5)
> With -O2 we have PTA showing up
> 
>  tree PTA                           :  12.21 ( 77%)
> 
> PTA solving should be pretty quick, but it seems we take a lot of iterations
> for unknown reasons:
> 
>   32.27%         20663  cc1      cc1               [.] topo_visit
>   16.29%         10432  cc1      cc1               [.] solve_constraints
>   14.69%          9412  cc1      cc1               [.] find
>   14.10%          9021  cc1      cc1               [.] bitmap_clear_bit
>    1.19%           764  cc1      cc1               [.] bitmap_set_bit

the issue is that we have only constraints like

_7 = &NONLOCAL
e_30 = _7 + UNKNOWN
e_30 = f_27 + UNKNOWN
_8 = f_27 + UNKNOWN
_8 = f_27 + UNKNOWN
_9 = _8 + UNKNOWN

those form a graph with N nodes and no copy edges as those constraints
are all complex.  Each solving iteration a few nodes get a new solution
but those dependent have already been evaluated so we bottleneck on
computing the optimal order to visit.

I'm testing a patch for this.