[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 --- Comment #6 from Andrew Macleod --- oops, wasnt meant to actually attach that note to this PR.
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879
--- Comment #5 from Andrew Macleod ---
On 08/18/2017 06:13 PM, law at redhat dot com wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879
>
> Jeffrey A. Law changed:
>
> What|Removed |Added
>
> Status|NEW |RESOLVED
> Resolution|--- |FIXED
>
> --- Comment #4 from Jeffrey A. Law ---
> This was fixed by RIchi's change for pr71433.
>
Made a bunch of updates this week, and things are in decent shape I think.
so looking at that test case with the code base as of this afternoon, if
we expand irange to have 3 sub-ranges (which I think long term we will
may need)
void
foo (int a)
{
if (a != 5 && a != 10)
bar ();
if (a == 10)
baz ();
}
we get gimple
[0.00%] [count: INV]:
_1 = a_6(D) != 5;
_2 = a_6(D) != 10;
_3 = _1 & _2;
if (_3 != 0)
goto ; [INV] [count: INV]
else
goto ; [INV] [count: INV]
[0.00%] [count: INV]:
bar ();
[0.00%] [count: INV]:
if (a_6(D) == 10)
goto ; [INV] [count: INV]
else
goto ; [INV] [count: INV]
shows
BB 2: T: _1 [0x0001, 0x0001] precision = 1
BB 2: T: _2 [0x0001, 0x0001] precision = 1
BB 2: T: _3 [0x, 0x] precision = 1
BB 2: T: a_6(D) [0x8000, 0x4][0x6, 0x9][0xb,
0x7fff] precision = 32
BB 2: F: _1 [0x0, 0x0001] precision = 1
BB 2: F: _2 [0x0, 0x0001] precision = 1
BB 2: F: _3 [0x0, 0x0] precision = 1
BB 2: F: a_6(D) [0x5, 0x5][0xa, 0xa] precision = 32
BB 4: T: a_6(D) [0xa, 0xa] precision = 32
BB 4: F: a_6(D) [0x8000, 0x9][0xb, 0x7fff]
precision = 32
so the true side of bb2 shows the range of a_6 to be everything except 5
and 10 [0x8000, 0x4][0x6, 0x9][0xb, 0x7fff] precision = 32
and the true side of bb4, leading to the call to baz(), a_6 has to have
a value of 10.. if you intersect those 2 ranges on that path... voila!!
its a NULL range meaning that call can't happen on that path
2->3->4->5, letting you do your thing..
So that is promising...
Andrew
Over the weekend im going to consider options for handling an increase
of precision in ranges without going full on dynamic.
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 Jeffrey A. Law changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #4 from Jeffrey A. Law --- This was fixed by RIchi's change for pr71433.
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 Andrew Pinski changed: What|Removed |Added Keywords||missed-optimization Severity|normal |enhancement
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 --- Comment #3 from Jeffrey A. Law --- Which is the problem Andrew has started poking at :-) One of the paths being explored is creating those PHIs.
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 Richard Biener changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2016-05-02 Ever confirmed|0 |1 --- Comment #2 from Richard Biener --- The issue here is really that we don't insert PHIs for the SSA defs of inserted asserts so we have to play "tricks" like finding dominating asserts in the first place.
[Bug middle-end/70879] Missed jump threading opportunity with multiple != conditions
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70879 Jeffrey A. Law changed: What|Removed |Added CC||amacleod at redhat dot com --- Comment #1 from Jeffrey A. Law --- Andrew, thought this might be of interest... Look at bb3...
