diff options
author | Hal Finkel <hfinkel@anl.gov> | 2013-04-05 05:01:13 +0000 |
---|---|---|
committer | Hal Finkel <hfinkel@anl.gov> | 2013-04-05 05:01:13 +0000 |
commit | afe2f43e4ea0a514413bc742fc03ef5c675ae86a (patch) | |
tree | 07c9b2a02903c4c0ebe76969ede44868ef9ebec4 /include/llvm/CodeGen | |
parent | fc61e635fd09e0cb852313f5533fb7fe694158fb (diff) | |
download | llvm-afe2f43e4ea0a514413bc742fc03ef5c675ae86a.tar.gz llvm-afe2f43e4ea0a514413bc742fc03ef5c675ae86a.tar.bz2 llvm-afe2f43e4ea0a514413bc742fc03ef5c675ae86a.tar.xz |
Fix bug in PEI's virtual-register scavenging
This change fixes a bug that I introduced in r178058. After a register is
scavenged using one of the available spills slots the instruction defining the
virtual register needs to be moved to after the spill code. The scavenger has
already processed the defining instruction so that registers killed by that
instruction are available for definition in that same instruction. Unfortunately,
after this, the scavenger needs to iterate through the spill code and then
visit, again, the instruction that defines the now-scavenged register. In order
to avoid confusion, the register scavenger needs the ability to 'back up'
through the spill code so that it can again process the instructions in the
appropriate order. Prior to this fix, once the scavenger reached the
just-moved instruction, it would assert if it killed any registers because,
having already processed the instruction, it believed they were undefined.
Unfortunately, I don't yet have a small test case. Thanks to Pranav Bhandarkar
for diagnosing the problem and testing this fix.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178845 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'include/llvm/CodeGen')
-rw-r--r-- | include/llvm/CodeGen/RegisterScavenging.h | 17 |
1 files changed, 17 insertions, 0 deletions
diff --git a/include/llvm/CodeGen/RegisterScavenging.h b/include/llvm/CodeGen/RegisterScavenging.h index 49d16892f8..95bf29167c 100644 --- a/include/llvm/CodeGen/RegisterScavenging.h +++ b/include/llvm/CodeGen/RegisterScavenging.h @@ -93,6 +93,15 @@ public: while (MBBI != I) forward(); } + /// Invert the behavior of forward() on the current instruction (undo the + /// changes to the available registers made by forward()). + void unprocess(); + + /// Unprocess instructions until you reach the provided iterator. + void unprocess(MachineBasicBlock::iterator I) { + while (MBBI != I) unprocess(); + } + /// skipTo - Move the internal MBB iterator but do not update register states. void skipTo(MachineBasicBlock::iterator I) { if (I == MachineBasicBlock::iterator(NULL)) @@ -100,6 +109,10 @@ public: MBBI = I; } + MachineBasicBlock::iterator getCurrentPosition() const { + return MBBI; + } + /// getRegsUsed - return all registers currently in use in used. void getRegsUsed(BitVector &used, bool includeReserved); @@ -171,6 +184,10 @@ private: RegsAvailable |= Regs; } + /// Processes the current instruction and fill the KillRegs and DefRegs bit + /// vectors. + void determineKillsAndDefs(); + /// Add Reg and all its sub-registers to BV. void addRegWithSubRegs(BitVector &BV, unsigned Reg); |