Index: lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
===================================================================
--- lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -347,6 +347,11 @@
   bool ParseSectionDirectiveHSAText();
 
 public:
+public:
+  enum AMDGPUMatchResultTy {
+    Match_PreferE32 = FIRST_TARGET_MATCH_RESULT_TY
+  };
+
   AMDGPUAsmParser(MCSubtargetInfo &STI, MCAsmParser &_Parser,
                const MCInstrInfo &MII,
                const MCTargetOptions &Options)
@@ -556,6 +561,11 @@
       (getForcedEncodingSize() == 64 && !(TSFlags & SIInstrFlags::VOP3)))
     return Match_InvalidOperand;
 
+  if ((TSFlags & SIInstrFlags::VOP3) &&
+      (TSFlags & SIInstrFlags::VOPAsmPrefer32Bit) &&
+      getForcedEncodingSize() != 64)
+    return Match_PreferE32;
+
   return Match_Success;
 }
 
@@ -614,6 +624,9 @@
       }
       return Error(ErrorLoc, "invalid operand for instruction");
     }
+    case Match_PreferE32:
+      return Error(IDLoc, "internal error: instruction without _e64 suffix "
+                          "should be encoded as e32");
   }
   llvm_unreachable("Implement any new match types added!");
 }
@@ -1701,8 +1714,12 @@
 }
 
 void AMDGPUAsmParser::cvtVOP3(MCInst &Inst, const OperandVector &Operands) {
-  ((AMDGPUOperand &)*Operands[1]).addRegOperands(Inst, 1);
-  unsigned i = 2;
+
+  unsigned i = 1;
+  const MCInstrDesc &Desc = MII.get(Inst.getOpcode());
+  if (Desc.getNumDefs() > 0) {
+    ((AMDGPUOperand &)*Operands[i++]).addRegOperands(Inst, 1);
+  }
 
   std::map<enum AMDGPUOperand::ImmTy, unsigned> OptionalIdx;
 
Index: lib/Target/AMDGPU/SIDefines.h
===================================================================
--- lib/Target/AMDGPU/SIDefines.h
+++ lib/Target/AMDGPU/SIDefines.h
@@ -37,7 +37,8 @@
   MIMG = 1 << 18,
   FLAT = 1 << 19,
   WQM = 1 << 20,
-  VGPRSpill = 1 << 21
+  VGPRSpill = 1 << 21,
+  VOPAsmPrefer32Bit = 1 << 22
 };
 }
 
Index: lib/Target/AMDGPU/SIInstrFormats.td
===================================================================
--- lib/Target/AMDGPU/SIInstrFormats.td
+++ lib/Target/AMDGPU/SIInstrFormats.td
@@ -40,6 +40,7 @@
   field bits<1> FLAT = 0;
   field bits<1> WQM = 0;
   field bits<1> VGPRSpill = 0;
+  field bits<1> VOPAsmPrefer32Bit = 0;
 
   // These need to be kept in sync with the enum in SIInstrFlags.
   let TSFlags{0} = VM_CNT;
@@ -68,6 +69,7 @@
   let TSFlags{19} = FLAT;
   let TSFlags{20} = WQM;
   let TSFlags{21} = VGPRSpill;
+  let TSFlags{22} = VOPAsmPrefer32Bit;
 
   let SchedRW = [Write32Bit];
 }
Index: lib/Target/AMDGPU/SIInstrInfo.td
===================================================================
--- lib/Target/AMDGPU/SIInstrInfo.td
+++ lib/Target/AMDGPU/SIInstrInfo.td
@@ -989,11 +989,12 @@
   let ParserMatchClass = InputModsMatchClass;
 }
 
-class getNumSrcArgs<ValueType Src1, ValueType Src2> {
+class getNumSrcArgs<ValueType Src0, ValueType Src1, ValueType Src2> {
   int ret =
-    !if (!eq(Src1.Value, untyped.Value),      1,   // VOP1
+    !if (!eq(Src0.Value, untyped.Value),      0,
+      !if (!eq(Src1.Value, untyped.Value),    1,   // VOP1
          !if (!eq(Src2.Value, untyped.Value), 2,   // VOP2
-                                              3)); // VOP3
+                                              3))); // VOP3
 }
 
 // Returns the register class to use for the destination of VOP[123C]
@@ -1085,17 +1086,20 @@
 // Returns the assembly string for the inputs and outputs of a VOP[12C]
 // instruction.  This does not add the _e32 suffix, so it can be reused
 // by getAsm64.
-class getAsm32 <int NumSrcArgs> {
+class getAsm32 <bit HasDst, int NumSrcArgs> {
+  string dst = "$dst";
+  string src0 = ", $src0";
   string src1 = ", $src1";
   string src2 = ", $src2";
-  string ret = "$dst, $src0"#
-               !if(!eq(NumSrcArgs, 1), "", src1)#
-               !if(!eq(NumSrcArgs, 3), src2, "");
+  string ret = !if(HasDst, dst, "") #
+               !if(!eq(NumSrcArgs, 1), src0, "") #
+               !if(!eq(NumSrcArgs, 2), src0#src1, "") #
+               !if(!eq(NumSrcArgs, 3), src0#src1#src2, "");
 }
 
 // Returns the assembly string for the inputs and outputs of a VOP3
 // instruction.
-class getAsm64 <int NumSrcArgs, bit HasModifiers> {
+class getAsm64 <bit HasDst, int NumSrcArgs, bit HasModifiers> {
   string src0 = !if(!eq(NumSrcArgs, 1), "$src0_modifiers", "$src0_modifiers,");
   string src1 = !if(!eq(NumSrcArgs, 1), "",
                    !if(!eq(NumSrcArgs, 2), " $src1_modifiers",
@@ -1103,7 +1107,7 @@
   string src2 = !if(!eq(NumSrcArgs, 3), " $src2_modifiers", "");
   string ret =
   !if(!eq(HasModifiers, 0),
-      getAsm32<NumSrcArgs>.ret,
+      getAsm32<HasDst, NumSrcArgs>.ret,
       "$dst, "#src0#src1#src2#"$clamp"#"$omod");
 }
 
@@ -1122,11 +1126,12 @@
   field RegisterOperand Src1RC64 = getVOP3SrcForVT<Src1VT>.ret;
   field RegisterOperand Src2RC64 = getVOP3SrcForVT<Src2VT>.ret;
 
-  field bit HasDst32 = !if(!eq(DstVT, untyped), 0, 1);
-  field int NumSrcArgs = getNumSrcArgs<Src1VT, Src2VT>.ret;
+  field bit HasDst = !if(!eq(DstVT.Value, untyped.Value), 0, 1);
+  field bit HasDst32 = HasDst;
+  field int NumSrcArgs = getNumSrcArgs<Src0VT, Src1VT, Src2VT>.ret;
   field bit HasModifiers = hasModifiers<Src0VT>.ret;
 
-  field dag Outs = (outs DstRC:$dst);
+  field dag Outs = !if(HasDst,(outs DstRC:$dst),(outs));
 
   // VOP3b instructions are a special case with a second explicit
   // output. This is manually overridden for them.
@@ -1137,8 +1142,8 @@
   field dag Ins64 = getIns64<Src0RC64, Src1RC64, Src2RC64, NumSrcArgs,
                              HasModifiers>.ret;
 
-  field string Asm32 = getAsm32<NumSrcArgs>.ret;
-  field string Asm64 = getAsm64<NumSrcArgs, HasModifiers>.ret;
+  field string Asm32 = getAsm32<HasDst, NumSrcArgs>.ret;
+  field string Asm64 = getAsm64<HasDst, NumSrcArgs, HasModifiers>.ret;
 }
 
 // FIXME: I think these F16/I16 profiles will need to use f16/i16 types in order
@@ -1151,6 +1156,8 @@
 def VOP_F16_F16_I16 : VOPProfile <[f16, f16, i32, untyped]>;
 def VOP_I16_I16_I16 : VOPProfile <[i32, i32, i32, untyped]>;
 
+def VOP_NONE : VOPProfile <[untyped, untyped, untyped, untyped]>;
+
 def VOP_F32_F32 : VOPProfile <[f32, f32, untyped, untyped]>;
 def VOP_F32_F64 : VOPProfile <[f32, f64, untyped, untyped]>;
 def VOP_F32_I32 : VOPProfile <[f32, i32, untyped, untyped]>;
@@ -1246,8 +1253,8 @@
   let Ins32 = (ins Src0RC32:$src0, Src1RC32:$src1, VGPR_32:$src2);
   let Ins64 = getIns64<Src0RC64, Src1RC64, RegisterOperand<VGPR_32>, 3,
                              HasModifiers>.ret;
-  let Asm32 = getAsm32<2>.ret;
-  let Asm64 = getAsm64<2, HasModifiers>.ret;
+  let Asm32 = getAsm32<1, 2>.ret;
+  let Asm64 = getAsm64<1, 2, HasModifiers>.ret;
 }
 def VOP_F64_F64_F64_F64 : VOPProfile <[f64, f64, f64, f64]>;
 def VOP_I32_I32_I32_I32 : VOPProfile <[i32, i32, i32, i32]>;
@@ -1256,8 +1263,15 @@
 class SIInstAlias <string asm, Instruction inst, VOPProfile p> :
     InstAlias <asm, (inst)>, PredicateControl {
 
-  field bit isCompare;
   field bit isCommutable;
+  field bit isCompare;
+  field bit isMoveImm;
+  field bit isReMaterializable;
+  field bit isAsCheapAsAMove;
+  list <SchedReadWrite> SchedRW;
+  field bits<8> vdst;
+  field bits<9> src0;
+  field bit VOPAsmPrefer32Bit;
 
   let ResultInst =
     !if (p.HasDst32,
@@ -1424,6 +1438,9 @@
   MnemonicAlias<opName#"_e64", opName> {
   let isPseudo = 1;
   let isCodeGenOnly = 1;
+
+  field bit vdst;
+  field bit src0;
 }
 
 class VOP3_Real_si <bits<9> op, dag outs, dag ins, string asm, string opName> :
Index: lib/Target/AMDGPU/SIInstructions.td
===================================================================
--- lib/Target/AMDGPU/SIInstructions.td
+++ lib/Target/AMDGPU/SIInstructions.td
@@ -1156,8 +1156,8 @@
 // VOP1 Instructions
 //===----------------------------------------------------------------------===//
 
-let vdst = 0, src0 = 0 in {
-defm V_NOP : VOP1_m <vop1<0x0>, (outs), (ins), "v_nop", [], "v_nop">;
+let vdst = 0, src0 = 0, VOPAsmPrefer32Bit = 1 in {
+defm V_NOP : VOP1Inst <vop1<0x0>, "v_nop", VOP_NONE>;
 }
 
 let isMoveImm = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in {
@@ -1332,10 +1332,8 @@
 defm V_FREXP_MANT_F32 : VOP1Inst <vop1<0x40, 0x34>, "v_frexp_mant_f32",
   VOP_F32_F32
 >;
-let vdst = 0, src0 = 0 in {
-defm V_CLREXCP : VOP1_m <vop1<0x41,0x35>, (outs), (ins), "v_clrexcp", [],
-  "v_clrexcp"
->;
+let vdst = 0, src0 = 0, VOPAsmPrefer32Bit = 1 in {
+defm V_CLREXCP : VOP1Inst <vop1<0x41,0x35>, "v_clrexcp", VOP_NONE>;
 }
 defm V_MOVRELD_B32 : VOP1Inst <vop1<0x42, 0x36>, "v_movreld_b32", VOP_I32_I32>;
 defm V_MOVRELS_B32 : VOP1Inst <vop1<0x43, 0x37>, "v_movrels_b32", VOP_I32_I32>;
Index: test/MC/AMDGPU/vop1.s
===================================================================
--- test/MC/AMDGPU/vop1.s
+++ test/MC/AMDGPU/vop1.s
@@ -8,6 +8,25 @@
 // RUN: not llvm-mc -arch=amdgcn -mcpu=bonaire -show-encoding %s 2>&1 | FileCheck %s --check-prefix=NOSICI
 // RUN: not llvm-mc -arch=amdgcn -mcpu=tonga -show-encoding %s 2>&1 | FileCheck %s -check-prefix=NOVI
 
+// Force 32-bit encoding
+
+// GCN: v_mov_b32_e32 v1, v2 ; encoding: [0x02,0x03,0x02,0x7e]
+v_mov_b32_e32 v1, v2
+
+// Force 32-bit encoding for special instructions
+// FIXME: We should be printing _e32 suffixes for these:
+
+// GCN: v_nop ; encoding: [0x00,0x00,0x00,0x7e]
+v_nop_e32
+
+// SICI: v_clrexcp ; encoding: [0x00,0x82,0x00,0x7e]
+// VI:   v_clrexcp ; encoding: [0x00,0x6a,0x00,0x7e]
+v_clrexcp_e32
+
+//===----------------------------------------------------------------------===//
+// Instructions
+//===----------------------------------------------------------------------===//
+
 
 // GCN: v_nop ; encoding: [0x00,0x00,0x00,0x7e]
 v_nop
Index: test/MC/AMDGPU/vop3-vop1-nosrc.s
===================================================================
--- /dev/null
+++ test/MC/AMDGPU/vop3-vop1-nosrc.s
@@ -0,0 +1,14 @@
+// RUN: llvm-mc -arch=amdgcn -mcpu=SI -show-encoding %s | FileCheck %s --check-prefix=SICI
+// RUN: llvm-mc -arch=amdgcn -mcpu=tonga -show-encoding %s | FileCheck %s --check-prefix=VI
+// XFAIL: *
+
+// FIXME: We should be printing _e64 suffixes for these. 
+// FIXME: When this is fixed delete this file and fix test case in vop3.s
+
+v_nop_e64
+// SICI: v_nop_e64 ; encoding: [0x00,0x00,0x00,0xd3,0x00,0x00,0x00,0x00]
+// VI:   v_nop_e64 ; encoding: [0x00,0x00,0x40,0xd1,0x00,0x00,0x00,0x00]
+
+v_clrexcp_e64
+// SICI: v_clrexcp_e64 ; encoding: [0x00,0x00,0x82,0xd3,0x00,0x00,0x00,0x00]
+// VI:   v_clrexcp_e64 ; encoding: [0x00,0x00,0x75,0xd1,0x00,0x00,0x00,0x00]
Index: test/MC/AMDGPU/vop3.s
===================================================================
--- test/MC/AMDGPU/vop3.s
+++ test/MC/AMDGPU/vop3.s
@@ -118,6 +118,23 @@
 // VOP1 Instructions
 //===----------------------------------------------------------------------===//
 
+// Test forced e64 encoding with e32 operands
+
+v_mov_b32_e64 v1, v2
+// SICI: v_mov_b32_e64 v1, v2 ; encoding: [0x01,0x00,0x02,0xd3,0x02,0x01,0x00,0x00]
+// VI:   v_mov_b32_e64 v1, v2 ; encoding: [0x01,0x00,0x41,0xd1,0x02,0x01,0x00,0x00]
+
+// Force e64 encoding for special instructions.
+// FIXME, we should be printing the _e64 suffix for v_nop and v_clrexcp.
+
+v_nop_e64
+// SICI: v_nop ; encoding: [0x00,0x00,0x00,0xd3,0x00,0x00,0x00,0x00]
+// VI:   v_nop ; encoding: [0x00,0x00,0x40,0xd1,0x00,0x00,0x00,0x00]
+
+v_clrexcp_e64
+// SICI: v_clrexcp ; encoding: [0x00,0x00,0x82,0xd3,0x00,0x00,0x00,0x00]
+// VI:   v_clrexcp ; encoding: [0x00,0x00,0x75,0xd1,0x00,0x00,0x00,0x00]
+
 //
 // Modifier tests:
 //