https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117875
--- Comment #11 from Andrew Macleod <amacleod at redhat dot com> ---
(In reply to Richard Biener from comment #8)
> (In reply to Richard Biener from comment #7)
> > I'm talking about the number of iterations of the second loop (after loop
> > splitting), the niter expression is (unsigned int) M_9(D) - (unsigned int)
> > k_24.
> > We know k_29 == M_9(D) when exiting the first loop, so on that entry the
> > expression computes zero. For the case the first loop is short-cut we know
> > M_9(D) == 1 and thus the difference is zero as well.
> >
> > So I expect the range of (unsigned int) M_9(D) - (unsigned int) k_2 to be
> > [0,0].
> >
> > 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.
Is there any place smart enough to be able to recognize that because the first
loop is a bump by one and with the core:
# k_15 = PHI <k_12(8), 1(14)>
<...>
k_12 = k_15 + 1;
if (k_12 < M_9(D))
the exit condition could be rewritten to
if (k_12 == M_9(D)) ?
THe loop is only entered by the branch if M_9(D) > 1 feeding the 1(14)
initializer, so the relation k_15 < M_9 is always true. (We could know that
with a little work)
That might enable a couple of things. (Ranger will also have a more precise
relation on that exit edge, which may also help inn other unexpected ways.)
Im coincidentally looking at PHI argument relations for 116753 (we currently do
nothing other than basic equality if all the arguments are in the same
equivalence set).
k_12 = k_15 + 1;
if (k_12 < M_9(D))
goto <bb 8>; [89.00%] // latch
else
goto <bb 16>; [11.00%]
<bb 16> [local count: 67276368]:
# k_29 = PHI <k_12(3)>
if (M_9(D) >= k_29)
goto <bb 15>; [99.95%]
else
goto <bb 6>; [0.05%]
X
<bb 15> [local count: 105119324]:
# k_24 = PHI <1(13), k_29(16)>
the second branch gets eliminated anyway, but feeding into the k_24 PHI, we'll
be able to recognize k_29 == M_9(D) with the condition rewritten. It also
not particularly difficult to determine that the constant 1 is actually a
rewritten M_9(D) if we tried hard enough.
ranger knows that M_9(D) is [1, 1] on that edge. with that info, it would not
be a stretch to be able to put M_9(D) and k_24 in the same equivalency set...
That could give you the 0 result you are looking for. If the 2 operands are in
an equivalency set, operator_minus will DTRT:
if (rel == VREL_EQ)
rel_range = int_range<2> (type, wi::zero (prec), wi::zero (prec));
