Another extract_subreg coalescing bug.

e.g.
vr1024<2> extract_subreg vr1025, 2
If vr1024 do not have the same register class as vr1025, it's not safe to coalesce this away. For example, vr1024 might be a GPR32 while vr1025 might be a GPR64.

llvm-svn: 50385

1 files changed, 3 insertions, 1 deletions

diff --git a/llvm/lib/CodeGen/SimpleRegisterCoalescing.cpp b/llvm/lib/CodeGen/SimpleRegisterCoalescing.cpp
index 3ea35c6eab2..6a110f0b87d 100644
--- a/llvm/lib/CodeGen/SimpleRegisterCoalescing.cpp
+++ b/llvm/lib/CodeGen/SimpleRegisterCoalescing.cpp
@@ -942,9 +942,11 @@ bool SimpleRegisterCoalescing::JoinCopy(CopyRec &TheCopy, bool &Again) {
       unsigned OldSubIdx = isExtSubReg ? CopyMI->getOperand(0).getSubReg()
         : CopyMI->getOperand(2).getSubReg();
       if (OldSubIdx) {
-        if (OldSubIdx == SubIdx)
+        if (OldSubIdx == SubIdx && !differingRegisterClasses(SrcReg, DstReg))
           // r1024<2> = EXTRACT_SUBREG r1025, 2. Then r1024 has already been
           // coalesced to a larger register so the subreg indices cancel out.
+          // Also check if the other larger register is of the same register
+          // class as the would be resulting register.
           SubIdx = 0;
         else {
           DOUT << "\t Sub-register indices mismatch.\n";