X86: change MOV64ri64i32 into MOV32ri64

The MOV64ri64i32 instruction required hacky MCInst lowering because it
was allocated as setting a GR64, but the eventual instruction ("movl")
only set a GR32. This converts it into a so-called "MOV32ri64" which
still accepts a (appropriate) 64-bit immediate but defines a GR32.
This is then converted to the full GR64 by a SUBREG_TO_REG operation,
thus keeping everyone happy.

This fixes a typo in the opcode field of the original patch, which
should make the legact JIT work again (& adds test for that problem).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@183068 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 72 insertions, 42 deletions

diff --git a/lib/Target/X86/X86CodeEmitter.cpp b/lib/Target/X86/X86CodeEmitter.cpp
index 8fea6ed..c3734a2 100644
--- a/lib/Target/X86/X86CodeEmitter.cpp
+++ b/lib/Target/X86/X86CodeEmitter.cpp
@@ -1270,7 +1270,7 @@ void Emitter<CodeEmitter>::emitInstruction(MachineInstr &MI,
 
     unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
       : (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
-    if (Opcode == X86::MOV64ri64i32)
+    if (Opcode == X86::MOV32ri64)
       rt = X86::reloc_absolute_word;  // FIXME: add X86II flag?
     // This should not occur on Darwin for relocatable objects.
     if (Opcode == X86::MOV64ri)
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 0b6940e..02f5fe4 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -200,6 +200,7 @@ namespace {
     bool SelectAddr(SDNode *Parent, SDValue N, SDValue &Base,
                     SDValue &Scale, SDValue &Index, SDValue &Disp,
                     SDValue &Segment);
+    bool SelectMOV64Imm32(SDValue N, SDValue &Imm);
     bool SelectLEAAddr(SDValue N, SDValue &Base,
                        SDValue &Scale, SDValue &Index, SDValue &Disp,
                        SDValue &Segment);
@@ -1380,6 +1381,33 @@ bool X86DAGToDAGISel::SelectScalarSSELoad(SDNode *Root,
 }
 
 
+bool X86DAGToDAGISel::SelectMOV64Imm32(SDValue N, SDValue &Imm) {
+  if (const ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N)) {
+    uint64_t ImmVal = CN->getZExtValue();
+    if ((uint32_t)ImmVal != (uint64_t)ImmVal)
+      return false;
+
+    Imm = CurDAG->getTargetConstant(ImmVal, MVT::i64);
+    return true;
+  }
+
+  // In static codegen with small code model, we can get the address of a label
+  // into a register with 'movl'. TableGen has already made sure we're looking
+  // at a label of some kind.
+  assert(N->getOpcode() == X86ISD::Wrapper && "Unexpected node type for MOV32ri64");
+  N = N.getOperand(0);
+
+  if (N->getOpcode() != ISD::TargetConstantPool &&
+      N->getOpcode() != ISD::TargetJumpTable &&
+      N->getOpcode() != ISD::TargetGlobalAddress &&
+      N->getOpcode() != ISD::TargetExternalSymbol &&
+      N->getOpcode() != ISD::TargetBlockAddress)
+    return false;
+
+  Imm = N;
+  return TM.getCodeModel() == CodeModel::Small;
+}
+
 /// SelectLEAAddr - it calls SelectAddr and determines if the maximal addressing
 /// mode it matches can be cost effectively emitted as an LEA instruction.
 bool X86DAGToDAGISel::SelectLEAAddr(SDValue N,
diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td
index 213993a..8a7ee7d 100644
--- a/lib/Target/X86/X86InstrCompiler.td
+++ b/lib/Target/X86/X86InstrCompiler.td
@@ -236,10 +236,18 @@ def : Pat<(i64 0), (SUBREG_TO_REG (i64 0), (MOV32r0), sub_32bit)> {
 // use MOV32ri with a SUBREG_TO_REG to represent the zero-extension, however
 // that would make it more difficult to rematerialize.
 let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1,
-    isCodeGenOnly = 1 in
-def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src),
-                        "", [(set GR64:$dst, i64immZExt32:$src)],
-                        IIC_ALU_NONMEM>, Sched<[WriteALU]>;
+    isCodeGenOnly = 1, neverHasSideEffects = 1 in
+def MOV32ri64 : Ii32<0xb8, AddRegFrm, (outs GR32:$dst), (ins i64i32imm:$src),
+                     "", [], IIC_ALU_NONMEM>, Sched<[WriteALU]>;
+
+// This 64-bit pseudo-move can be used for both a 64-bit constant that is
+// actually the zero-extension of a 32-bit constant, and for labels in the
+// x86-64 small code model.
+def mov64imm32 : ComplexPattern<i64, 1, "SelectMOV64Imm32", [imm, X86Wrapper]>;
+
+let AddedComplexity = 1 in
+def : Pat<(i64 mov64imm32:$src),
+          (SUBREG_TO_REG (i64 0), (MOV32ri64 mov64imm32:$src), sub_32bit)>;
 
 // Use sbb to materialize carry bit.
 let Uses = [EFLAGS], Defs = [EFLAGS], isPseudo = 1, SchedRW = [WriteALU] in {
@@ -924,20 +932,6 @@ def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
 def : Pat<(i64 (X86Wrapper tblockaddress:$dst)),
           (MOV64ri tblockaddress:$dst)>, Requires<[FarData]>;
 
-// In static codegen with small code model, we can get the address of a label
-// into a register with 'movl'.  FIXME: This is a hack, the 'imm' predicate of
-// the MOV64ri64i32 should accept these.
-def : Pat<(i64 (X86Wrapper tconstpool  :$dst)),
-          (MOV64ri64i32 tconstpool  :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tjumptable  :$dst)),
-          (MOV64ri64i32 tjumptable  :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
-          (MOV64ri64i32 tglobaladdr :$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
-          (MOV64ri64i32 texternalsym:$dst)>, Requires<[SmallCode]>;
-def : Pat<(i64 (X86Wrapper tblockaddress:$dst)),
-          (MOV64ri64i32 tblockaddress:$dst)>, Requires<[SmallCode]>;
-
 // In kernel code model, we can get the address of a label
 // into a register with 'movq'.  FIXME: This is a hack, the 'imm' predicate of
 // the MOV64ri32 should accept these.
diff --git a/lib/Target/X86/X86MCInstLower.cpp b/lib/Target/X86/X86MCInstLower.cpp
index b423e1e..4b6503b 100644
--- a/lib/Target/X86/X86MCInstLower.cpp
+++ b/lib/Target/X86/X86MCInstLower.cpp
@@ -226,13 +226,6 @@ MCOperand X86MCInstLower::LowerSymbolOperand(const MachineOperand &MO,
 
 
 
-static void lower_subreg32(MCInst *MI, unsigned OpNo) {
-  // Convert registers in the addr mode according to subreg32.
-  unsigned Reg = MI->getOperand(OpNo).getReg();
-  if (Reg != 0)
-    MI->getOperand(OpNo).setReg(getX86SubSuperRegister(Reg, MVT::i32));
-}
-
 static void lower_lea64_32mem(MCInst *MI, unsigned OpNo) {
   // Convert registers in the addr mode according to subreg64.
   for (unsigned i = 0; i != 4; ++i) {
@@ -246,11 +239,6 @@ static void lower_lea64_32mem(MCInst *MI, unsigned OpNo) {
   }
 }
 
-/// LowerSubReg32_Op0 - Things like MOVZX16rr8 -> MOVZX32rr8.
-static void LowerSubReg32_Op0(MCInst &OutMI, unsigned NewOpc) {
-  OutMI.setOpcode(NewOpc);
-  lower_subreg32(&OutMI, 0);
-}
 /// LowerUnaryToTwoAddr - R = setb   -> R = sbb R, R
 static void LowerUnaryToTwoAddr(MCInst &OutMI, unsigned NewOpc) {
   OutMI.setOpcode(NewOpc);
@@ -388,9 +376,12 @@ ReSimplify:
     assert(OutMI.getOperand(1+X86::AddrSegmentReg).getReg() == 0 &&
            "LEA has segment specified!");
     break;
-  case X86::MOV64ri64i32: LowerSubReg32_Op0(OutMI, X86::MOV32ri); break;
   case X86::MOV32r0:      LowerUnaryToTwoAddr(OutMI, X86::XOR32rr); break;
 
+  case X86::MOV32ri64:
+    OutMI.setOpcode(X86::MOV32ri);
+    break;
+
   // Commute operands to get a smaller encoding by using VEX.R instead of VEX.B
   // if one of the registers is extended, but other isn't.
   case X86::VMOVAPDrr:
diff --git a/test/CodeGen/X86/abi-isel.ll b/test/CodeGen/X86/abi-isel.ll
index 658ccaa..955fc62 100644
--- a/test/CodeGen/X86/abi-isel.ll
+++ b/test/CodeGen/X86/abi-isel.ll
@@ -6629,8 +6629,8 @@ entry:
 ; LINUX-32-PIC-NEXT: 	ret
 
 ; LINUX-64-PIC: bam02:
-; LINUX-64-PIC: 	movq	ptr@GOTPCREL(%rip), [[RCX:%r.x]]
-; LINUX-64-PIC-NEXT: 	movl	$262144, %eax
+; LINUX-64-PIC: 	movl	$262144, %eax
+; LINUX-64-PIC-NEXT: 	movq	ptr@GOTPCREL(%rip), [[RCX:%r.x]]
 ; LINUX-64-PIC-NEXT: 	addq	([[RCX]]), %rax
 ; LINUX-64-PIC-NEXT: 	ret
 
@@ -6655,20 +6655,20 @@ entry:
 ; DARWIN-32-PIC-NEXT: 	ret
 
 ; DARWIN-64-STATIC: _bam02:
-; DARWIN-64-STATIC: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
-; DARWIN-64-STATIC-NEXT: 	movl	$262144, %eax
+; DARWIN-64-STATIC: 	movl	$262144, %eax
+; DARWIN-64-STATIC-NEXT: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
 ; DARWIN-64-STATIC-NEXT: 	addq	([[RCX]]), %rax
 ; DARWIN-64-STATIC-NEXT: 	ret
 
 ; DARWIN-64-DYNAMIC: _bam02:
-; DARWIN-64-DYNAMIC: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
-; DARWIN-64-DYNAMIC-NEXT: 	movl	$262144, %eax
+; DARWIN-64-DYNAMIC: 	movl	$262144, %eax
+; DARWIN-64-DYNAMIC-NEXT: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
 ; DARWIN-64-DYNAMIC-NEXT: 	addq	([[RCX]]), %rax
 ; DARWIN-64-DYNAMIC-NEXT: 	ret
 
 ; DARWIN-64-PIC: _bam02:
-; DARWIN-64-PIC: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
-; DARWIN-64-PIC-NEXT: 	movl	$262144, %eax
+; DARWIN-64-PIC: 	movl	$262144, %eax
+; DARWIN-64-PIC-NEXT: 	movq	_ptr@GOTPCREL(%rip), [[RCX:%r.x]]
 ; DARWIN-64-PIC-NEXT: 	addq	([[RCX]]), %rax
 ; DARWIN-64-PIC-NEXT: 	ret
 }
@@ -6791,8 +6791,8 @@ entry:
 ; LINUX-32-PIC-NEXT: 	ret
 
 ; LINUX-64-PIC: bam05:
-; LINUX-64-PIC: 	movq	dptr@GOTPCREL(%rip), [[RCX:%r.x]]
-; LINUX-64-PIC-NEXT: 	movl	$262144, %eax
+; LINUX-64-PIC: 	movl	$262144, %eax
+; LINUX-64-PIC-NEXT: 	movq	dptr@GOTPCREL(%rip), [[RCX:%r.x]]
 ; LINUX-64-PIC-NEXT: 	addq	([[RCX]]), %rax
 ; LINUX-64-PIC-NEXT: 	ret
 
diff --git a/test/ExecutionEngine/mov64zext32.ll b/test/ExecutionEngine/mov64zext32.ll
new file mode 100644
index 0000000..00241c8
--- /dev/null
+++ b/test/ExecutionEngine/mov64zext32.ll
@@ -0,0 +1,17 @@
+; RUN: %lli %s > /dev/null
+
+define i64 @foo() {
+  ret i64 42
+}
+
+define i32 @main() {
+  %val = call i64 @foo()
+  %is42 = icmp eq i64 %val, 42
+  br i1 %is42, label %good, label %bad
+
+good:
+  ret i32 0
+
+bad:
+  ret i32 1
+} 
\ No newline at end of file