[llvm] r284152 - [RAGreedy] Empty live-ranges always succeed in last chance recoloring.
Quentin Colombet via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 13 12:27:48 PDT 2016
Author: qcolombet
Date: Thu Oct 13 14:27:48 2016
New Revision: 284152
URL: http://llvm.org/viewvc/llvm-project?rev=284152&view=rev
Log:
[RAGreedy] Empty live-ranges always succeed in last chance recoloring.
Relax the constraint for empty live-ranges while doing last chance
recoloring. Indeed, those live-ranges do not need an actual color to be
fond for the recoloring to work.
Empty live-range may happen as a result of splitting/spilling.
Unfortunately no test case for in-tree targets.
Modified:
llvm/trunk/lib/CodeGen/RegAllocGreedy.cpp
Modified: llvm/trunk/lib/CodeGen/RegAllocGreedy.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/RegAllocGreedy.cpp?rev=284152&r1=284151&r2=284152&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/RegAllocGreedy.cpp (original)
+++ llvm/trunk/lib/CodeGen/RegAllocGreedy.cpp Thu Oct 13 14:27:48 2016
@@ -2223,10 +2223,21 @@ bool RAGreedy::tryRecoloringCandidates(P
DEBUG(dbgs() << "Try to recolor: " << *LI << '\n');
unsigned PhysReg;
PhysReg = selectOrSplitImpl(*LI, NewVRegs, FixedRegisters, Depth + 1);
- if (PhysReg == ~0u || !PhysReg)
+ // When splitting happens, the live-range may actually be empty.
+ // In that case, this is okay to continue the recoloring even
+ // if we did not find an alternative color for it. Indeed,
+ // there will not be anything to color for LI in the end.
+ if (PhysReg == ~0u || (!PhysReg && !LI->empty()))
return false;
+
+ if (!PhysReg) {
+ assert(LI->empty() && "Only empty live-range do not require a register");
+ DEBUG(dbgs() << "Recoloring of " << *LI << " succeeded. Empty LI.\n");
+ continue;
+ }
DEBUG(dbgs() << "Recoloring of " << *LI
<< " succeeded with: " << PrintReg(PhysReg, TRI) << '\n');
+
Matrix->assign(*LI, PhysReg);
FixedRegisters.insert(LI->reg);
}
More information about the llvm-commits
mailing list