https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117875
--- Comment #10 from rguenther at suse dot de <rguenther at suse dot de> --- On Thu, 5 Dec 2024, hubicka at ucw dot cz wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117875 > > --- Comment #9 from Jan Hubicka <hubicka at ucw dot cz> --- > > > But maybe I'm missing something? > > > > I guess the issue is that with > > > > # k_24 = PHI <1(13), k_29(16)> > > > > to easily see this we'd have to compute the range of > > (unsigned int) M_9(D) - 1 and the range of (unsigned int) M_9(D) - > > (unsigned) > > k_29 and then see those are the same singleton. I don't think we can > > arrive here when using the range of k_24 itself, so maybe I'm asking too > > much of VRP here. > > Yes, I don't see how ranger can figure out easily the singleton range, > since final value of the iteration is variable. > Last year I looked into it and I kind of remember implementing special > matching code, but can not find the patch anymore. The code checking > loop1->any_estimate can be generalized to check for estimate of one of > the loop and adjust the other accordingly. > > It is scev_cprop pass that does replace the use of IV value use out of > loops by its bound. As in: > #include <stdio.h> > int test(int n) > { > int i; > if (n < 0) > return -1; > for (i = 0; i < n; i++) > printf ("%i\n",i); > return i; > } > > Maybe we want to be able to run it on a specific loop and run it on > loop1 between split finishes updating IL and before profile update? That can be done already, I refactored the code to be run on a loop a while back. Just call final_value_replacement_loop (loop), special casing it to a single PHI might be possible as well, of course.
