Fix a bug in the local spiller, where we could take code like this:
store r12 -> [ss#2]
R3 = load [ss#1]
use R3
R3 = load [ss#2]
R4 = load [ss#1]
and turn it into this code:
store R12 -> [ss#2]
R3 = load [ss#1]
use R3
R3 = R12
R4 = R3 <- oops!
The problem was that promoting R3 = load[ss#2] to a copy missed the fact that
the instruction invalidated R3 at that point.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@23638 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/VirtRegMap.cpp b/lib/CodeGen/VirtRegMap.cpp
index 633cf2b..f7306c5 100644
--- a/lib/CodeGen/VirtRegMap.cpp
+++ b/lib/CodeGen/VirtRegMap.cpp
@@ -490,11 +490,11 @@
if (DestReg != It->second) {
MRI->copyRegToReg(MBB, &MI, DestReg, It->second,
MF.getSSARegMap()->getRegClass(VirtReg));
- // Revisit the copy if the destination is a vreg.
- if (MRegisterInfo::isVirtualRegister(DestReg)) {
- NextMII = &MI;
- --NextMII; // backtrack to the copy.
- }
+ // Revisit the copy so we make sure to notice the effects of the
+ // operation on the destreg (either needing to RA it if it's
+ // virtual or needing to clobber any values if it's physical).
+ NextMII = &MI;
+ --NextMII; // backtrack to the copy.
}
MBB.erase(&MI);
goto ProcessNextInst;