https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94021
--- Comment #9 from Andrew Macleod <amacleod at redhat dot com> ---
(In reply to Jakub Jelinek from comment #6)
> CCing Andrew and Aldy to see what the ranger does or can do, talking about
> I mean, if we have:
> h_1 = x_2 / 3600;
> if (x_2 <= -3599 && x_2 <= 89999)
> use (h_1);
> figure out that h_1 is set to x_2 / 3600 and even when that
> SSA_NAME_DEF_STMT is not in a guarded block, its use is in one and so from
> the [-3599, 89999] range of x_2 at the point of use derive that h_1 there is
> [0, 24]?
> Surely if it is like:
> if (x_2 <= -3599 && x_2 <= 89999)
> {
> h_1 = x_2 / 3600;
> use (h_1);
> }
> I'd expect it to handle it that way.
Certainly we get the latter case. The earlier case is currently...
inconsistent. Something I hope to address in the next release.
if we have precisely:
h_1 = x_2 / 3600;
if (x_2 <= -3599 && x_2 <= 89999)
use (h_1);
and if that is calculated in such a way that all of the conditions are
evaluated in a single basic block, then the GORI engine well mark h1 and x_2
both as exports and the evaluator will calculate the desired value for h_1.
Once we start to pull them further apart, the current implementation loses the
ability to recalculate h_1 when we get new ranges for x_2.
I have plans to segregate the def chains from the export lists in blocks, and
allow for greater ability to recalculate things like this.
When I look at #c4 in EVRP, I see:
=========== BB 4 ============
<bb 4> :
# x_10 = PHI <x_13(D)(2), x_14(3)>
h_15 = x_10 / 3600;
_1 = x_10 % 3600;
m_16 = _1 / 60;
h.0_2 = (unsigned int) h_15;
_3 = h.0_2 > 23;
_5 = _3;
if (_5 != 0)
goto <bb 6>; [INV]
else
goto <bb 5>; [INV]
_1 : int [0, 3599]
h.0_2 : unsigned int [0, 596523]
x_10 : int [0, +INF]
h_15 : int [0, 596523]
m_16 : int [0, 59]
4->6 (T) h.0_2 : unsigned int [0, 596523]
4->6 (T) _5 : _Bool [1, 1]
4->6 (T) x_10 : int [0, +INF]
4->6 (T) h_15 : int [0, 596523]
4->5 (F) h.0_2 : unsigned int [0, 596523]
4->5 (F) _5 : _Bool [0, 0]
4->5 (F) x_10 : int [0, +INF]
4->5 (F) h_15 : int [0, 596523]
and then later on:
=========== BB 8 ============
x_10 int [0, +INF]
<bb 8> :
if (x_10 <= 89999)
goto <bb 9>; [INV]
else
goto <bb 10>; [INV]
8->9 (T) x_10 : int [0, 89999]
8->10 (F) x_10 : int [90000, +INF]
=========== BB 9 ============
<bb 9> :
__builtin_snprintf (&a, 8, "%s%02i%02i", "+", h_15, m_16);
The defchains already indicate that h_15 is dependant on the value of x_10, and
I am hoping to enable recalculation of h_15 when a dependant range has
changed.. and not just when they are exported from the same block.
so in this case, when we ask for the range of h_15 in BB_9, we should be able
to see that x_10 has a range of int [0, 89999] and trigger a recalculation of
h_15 using "current" values. and come up with h_15 = [0,24]
The pieces are all there, but they need to be assembled in a non time consuming
way :-)
It is on the radar for next release.
