https://llvm.org/bugs/show_bug.cgi?id=31867
Bug ID: 31867
Summary: InstCombine uselessly checks for irreducible phi
cycles.
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Scalar Optimizations
Assignee: unassignedbugs at nondot.org
Reporter: dberlin at dberlin.org
CC: llvm-bugs at lists.llvm.org
Classification: Unclassified
InstCombinePhi has this code:
// We sometimes end up with phi cycles that non-obviously end up being the
// same value, for example:
// z = some value; x = phi (y, z); y = phi (x, z)
// where the phi nodes don't necessarily need to be in the same block. Do a
// quick check to see if the PHI node only contains a single non-phi value,
if
// so, scan to see if the phi cycle is actually equal to that value.
First, such code is impossible in minimal SSA. It should actually be impossible
with the SSAUpdater too, since it generates minimal SSA.
So i'm not sure what would generate it in the first place.
I can't find a non-handmade testcase either.
My suggestion would be to either make it a statistic, or remove it.
Second, this is the ultra slow way of doing it. If it does minimize all cases,
it does so non optimally (IE it is performed more than it needs to be). It's a
bit hard to analyze to determine if it actually minimizes all cases.
The algorithm to minimize these phis is precisely the phi minimization
algorithm added to the new ssaupdater, which is an SCC algorithm that will
minimize all phis in as short a time as possible.
(it depends on the cycle depth. In the case given as an example above, it will
be one pass).
You can run this exactly once on the CFG, at the beginning or end of
instcombine, and feel confident it is minimal ssa again.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170204/d8abcd07/attachment.html>