Mark rematerialized super/sub registers as dead.

When we're rematerializing into a not-quite-right register we already add the
real definition as an imp-def, but we should also be marking the "official"
register as dead, since nothing else is going to use it as a result of this
remat.

Not doing this can affect pressure tracking.

rdar://problem/14158833

llvm-svn: 184002

llvm/lib/CodeGen/RegisterCoalescer.cpp

@@ -843,6 +843,7 @@ bool RegisterCoalescer::reMaterializeTrivialDef(CoalescerPair &CP,
     // been asked for. If so it must implicitly define the whole thing.
     assert(TargetRegisterInfo::isPhysicalRegister(DstReg) &&
            "Only expect virtual or physical registers in remat");
+    NewMI->getOperand(0).setIsDead(true);
     NewMI->addOperand(MachineOperand::CreateReg(CopyDstReg,
                                                 true  /*IsDef*/,
                                                 true  /*IsImp*/,

llvm/test/CodeGen/X86/remat-phys-dead.ll

@@ -0,0 +1,23 @@
+; REQUIRES: asserts
+; RUN: llc -mtriple=x86_64-apple-darwin -debug -o /dev/null < %s 2>&1 | FileCheck %s
+
+; We need to make sure that rematerialization into a physical register marks the
+; super- or sub-register as dead after this rematerialization since only the
+; original register is actually used later. Largely irrelevant for a trivial
+; example like this, since EAX is never used again, but easy to test.
+
+define i8 @test_remat() {
+  ret i8 0
+; CHECK: REGISTER COALESCING
+; CHECK: Remat: %EAX<def,dead> = MOV32r0 %EFLAGS<imp-def,dead>, %AL<imp-def>
+}
+
+; On the other hand, if it's already the correct width, we really shouldn't be
+; marking the definition register as dead.
+
+define i32 @test_remat32() {
+  ret i32 0
+; CHECK: REGISTER COALESCING
+; CHECK: Remat: %EAX<def> = MOV32r0 %EFLAGS<imp-def,dead>
+}
+