ARM NEON refactor VST2 w/ writeback instructions.

In addition to improving the representation, this adds support for assembly
parsing of these instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146588 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 130 insertions, 54 deletions

diff --git a/lib/Target/ARM/ARMExpandPseudoInsts.cpp b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
index 69011d73ca..099278286f 100644
--- a/lib/Target/ARM/ARMExpandPseudoInsts.cpp
+++ b/lib/Target/ARM/ARMExpandPseudoInsts.cpp
@@ -308,18 +308,24 @@ static const NEONLdStTableEntry NEONLdStTable[] = {
 { ARM::VST2LNq32Pseudo_UPD, ARM::VST2LNq32_UPD, false, true, true,  EvenDblSpc, 2, 2,true},
 
 { ARM::VST2d16Pseudo,       ARM::VST2d16,      false, false, false, SingleSpc,  2, 4 ,false},
-{ ARM::VST2d16Pseudo_UPD,   ARM::VST2d16_UPD, false, true, true,  SingleSpc,  2, 4 ,false},
+{ ARM::VST2d16PseudoWB_fixed,   ARM::VST2d16wb_fixed, false, true, false,  SingleSpc,  2, 4 ,false},
+{ ARM::VST2d16PseudoWB_register,   ARM::VST2d16wb_register, false, true, true,  SingleSpc,  2, 4 ,false},
 { ARM::VST2d32Pseudo,       ARM::VST2d32,      false, false, false, SingleSpc,  2, 2 ,false},
-{ ARM::VST2d32Pseudo_UPD,   ARM::VST2d32_UPD, false, true, true,  SingleSpc,  2, 2 ,false},
+{ ARM::VST2d32PseudoWB_fixed,   ARM::VST2d32wb_fixed, false, true, true,  SingleSpc,  2, 2 ,false},
+{ ARM::VST2d32PseudoWB_register,   ARM::VST2d32wb_register, false, true, true,  SingleSpc,  2, 2 ,false},
 { ARM::VST2d8Pseudo,        ARM::VST2d8,       false, false, false, SingleSpc,  2, 8 ,false},
-{ ARM::VST2d8Pseudo_UPD,    ARM::VST2d8_UPD, false, true, true,  SingleSpc,  2, 8 ,false},
+{ ARM::VST2d8PseudoWB_fixed,    ARM::VST2d8wb_fixed, false, true, false,  SingleSpc,  2, 8 ,false},
+{ ARM::VST2d8PseudoWB_register,    ARM::VST2d8wb_register, false, true, true,  SingleSpc,  2, 8 ,false},
 
 { ARM::VST2q16Pseudo,       ARM::VST2q16,      false, false, false, SingleSpc,  4, 4 ,false},
-{ ARM::VST2q16Pseudo_UPD,   ARM::VST2q16_UPD, false, true, true,  SingleSpc,  4, 4 ,false},
+{ ARM::VST2q16PseudoWB_fixed,   ARM::VST2q16wb_fixed, false, true, false,  SingleSpc,  4, 4 ,false},
+{ ARM::VST2q16PseudoWB_register,   ARM::VST2q16wb_register, false, true, true,  SingleSpc,  4, 4 ,false},
 { ARM::VST2q32Pseudo,       ARM::VST2q32,      false, false, false, SingleSpc,  4, 2 ,false},
-{ ARM::VST2q32Pseudo_UPD,   ARM::VST2q32_UPD, false, true, true,  SingleSpc,  4, 2 ,false},
+{ ARM::VST2q32PseudoWB_fixed,   ARM::VST2q32wb_fixed, false, true, false,  SingleSpc,  4, 2 ,false},
+{ ARM::VST2q32PseudoWB_register,   ARM::VST2q32wb_register, false, true, true,  SingleSpc,  4, 2 ,false},
 { ARM::VST2q8Pseudo,        ARM::VST2q8,       false, false, false, SingleSpc,  4, 8 ,false},
-{ ARM::VST2q8Pseudo_UPD,    ARM::VST2q8_UPD, false, true, true,  SingleSpc,  4, 8 ,false},
+{ ARM::VST2q8PseudoWB_fixed,    ARM::VST2q8wb_fixed, false, true, false,  SingleSpc,  4, 8 ,false},
+{ ARM::VST2q8PseudoWB_register,    ARM::VST2q8wb_register, false, true, true,  SingleSpc,  4, 8 ,false},
 
 { ARM::VST3LNd16Pseudo,     ARM::VST3LNd16,     false, false, false, SingleSpc, 3, 4 ,true},
 { ARM::VST3LNd16Pseudo_UPD, ARM::VST3LNd16_UPD, false, true, true,  SingleSpc, 3, 4 ,true},
@@ -1193,12 +1199,18 @@ bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
     case ARM::VST2q8Pseudo:
     case ARM::VST2q16Pseudo:
     case ARM::VST2q32Pseudo:
-    case ARM::VST2d8Pseudo_UPD:
-    case ARM::VST2d16Pseudo_UPD:
-    case ARM::VST2d32Pseudo_UPD:
-    case ARM::VST2q8Pseudo_UPD:
-    case ARM::VST2q16Pseudo_UPD:
-    case ARM::VST2q32Pseudo_UPD:
+    case ARM::VST2d8PseudoWB_fixed:
+    case ARM::VST2d16PseudoWB_fixed:
+    case ARM::VST2d32PseudoWB_fixed:
+    case ARM::VST2q8PseudoWB_fixed:
+    case ARM::VST2q16PseudoWB_fixed:
+    case ARM::VST2q32PseudoWB_fixed:
+    case ARM::VST2d8PseudoWB_register:
+    case ARM::VST2d16PseudoWB_register:
+    case ARM::VST2d32PseudoWB_register:
+    case ARM::VST2q8PseudoWB_register:
+    case ARM::VST2q16PseudoWB_register:
+    case ARM::VST2q32PseudoWB_register:
     case ARM::VST3d8Pseudo:
     case ARM::VST3d16Pseudo:
     case ARM::VST3d32Pseudo:
diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index ced13abca1..74731419fd 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -1589,6 +1589,12 @@ static unsigned getVLDSTRegisterUpdateOpcode(unsigned Opc) {
   case ARM::VLD2q16PseudoWB_fixed: return ARM::VLD2q16PseudoWB_register;
   case ARM::VLD2q32PseudoWB_fixed: return ARM::VLD2q32PseudoWB_register;
 
+  case ARM::VST2d8PseudoWB_fixed: return ARM::VST2d8PseudoWB_register;
+  case ARM::VST2d16PseudoWB_fixed: return ARM::VST2d16PseudoWB_register;
+  case ARM::VST2d32PseudoWB_fixed: return ARM::VST2d32PseudoWB_register;
+  case ARM::VST2q8PseudoWB_fixed: return ARM::VST2q8PseudoWB_register;
+  case ARM::VST2q16PseudoWB_fixed: return ARM::VST2q16PseudoWB_register;
+  case ARM::VST2q32PseudoWB_fixed: return ARM::VST2q32PseudoWB_register;
   }
   return Opc; // If not one we handle, return it unchanged.
 }
@@ -1806,9 +1812,9 @@ SDNode *ARMDAGToDAGISel::SelectVST(SDNode *N, bool isUpdating, unsigned NumVecs,
     Ops.push_back(Align);
     if (isUpdating) {
       SDValue Inc = N->getOperand(AddrOpIdx + 1);
-      // FIXME: VST1 fixed increment doesn't need Reg0. Remove the reg0
+      // FIXME: VST1/VST2 fixed increment doesn't need Reg0. Remove the reg0
       // case entirely when the rest are updated to that form, too.
-      if (NumVecs == 1 && !isa<ConstantSDNode>(Inc.getNode()))
+      if (NumVecs <= 2 && !isa<ConstantSDNode>(Inc.getNode()))
         Opc = getVLDSTRegisterUpdateOpcode(Opc);
       // We use a VST1 for v1i64 even if the pseudo says vld2/3/4, so
       // check for that explicitly too. Horribly hacky, but temporary.
@@ -2889,10 +2895,13 @@ SDNode *ARMDAGToDAGISel::Select(SDNode *N) {
   }
 
   case ARMISD::VST2_UPD: {
-    unsigned DOpcodes[] = { ARM::VST2d8Pseudo_UPD, ARM::VST2d16Pseudo_UPD,
-                            ARM::VST2d32Pseudo_UPD, ARM::VST1q64PseudoWB_fixed};
-    unsigned QOpcodes[] = { ARM::VST2q8Pseudo_UPD, ARM::VST2q16Pseudo_UPD,
-                            ARM::VST2q32Pseudo_UPD };
+    unsigned DOpcodes[] = { ARM::VST2d8PseudoWB_fixed,
+                            ARM::VST2d16PseudoWB_fixed,
+                            ARM::VST2d32PseudoWB_fixed,
+                            ARM::VST1q64PseudoWB_fixed};
+    unsigned QOpcodes[] = { ARM::VST2q8PseudoWB_fixed,
+                            ARM::VST2q16PseudoWB_fixed,
+                            ARM::VST2q32PseudoWB_fixed };
     return SelectVST(N, true, 2, DOpcodes, QOpcodes, 0);
   }
 
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 59911783f5..cd12ab9b40 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -1523,44 +1523,90 @@ def  VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
 def  VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
 
 // ...with address register writeback:
-class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
-  : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
-          (ins addrmode6:$Rn, am6offset:$Rm, VdTy:$Vd),
-          IIC_VST2u, "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
-  let Inst{5-4} = Rn{5-4};
-  let DecoderMethod = "DecodeVSTInstruction";
+//class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
+//  : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+//          (ins addrmode6:$Rn, am6offset:$Rm, VdTy:$Vd),
+//          IIC_VST2u, "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
+//  let Inst{5-4} = Rn{5-4};
+//  let DecoderMethod = "DecodeVSTInstruction";
+//}
+multiclass VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt,
+                   RegisterOperand VdTy> {
+  def _fixed : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+                     (ins addrmode6:$Rn, VdTy:$Vd), IIC_VLD1u,
+                     "vst2", Dt, "$Vd, $Rn!",
+                     "$Rn.addr = $wb", []> {
+    let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
+    let Inst{4} = Rn{4};
+    let DecoderMethod = "DecodeVSTInstruction";
+    let AsmMatchConverter = "cvtVSTwbFixed";
+  }
+  def _register : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+                        (ins addrmode6:$Rn, rGPR:$Rm, VdTy:$Vd), IIC_VLD1u,
+                        "vst2", Dt, "$Vd, $Rn, $Rm",
+                        "$Rn.addr = $wb", []> {
+    let Inst{4} = Rn{4};
+    let DecoderMethod = "DecodeVSTInstruction";
+    let AsmMatchConverter = "cvtVSTwbRegister";
+  }
 }
-class VST2QWB<bits<4> op7_4, string Dt>
-  : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
-          (ins addrmode6:$Rn, am6offset:$Rm, VecListFourD:$Vd), IIC_VST2x2u,
-          "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
-  let Inst{5-4} = Rn{5-4};
-  let DecoderMethod = "DecodeVSTInstruction";
+//class VST2QWB<bits<4> op7_4, string Dt>
+//  : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
+//          (ins addrmode6:$Rn, am6offset:$Rm, VecListFourD:$Vd), IIC_VST2x2u,
+//          "vst2", Dt, "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
+//  let Inst{5-4} = Rn{5-4};
+//  let DecoderMethod = "DecodeVSTInstruction";
+//}
+multiclass VST2QWB<bits<4> op7_4, string Dt> {
+  def _fixed : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
+                     (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1u,
+                     "vst2", Dt, "$Vd, $Rn!",
+                     "$Rn.addr = $wb", []> {
+    let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
+    let Inst{4} = Rn{4};
+    let DecoderMethod = "DecodeVSTInstruction";
+    let AsmMatchConverter = "cvtVSTwbFixed";
+  }
+  def _register : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
+                        (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
+                        IIC_VLD1u,
+                        "vst2", Dt, "$Vd, $Rn, $Rm",
+                        "$Rn.addr = $wb", []> {
+    let Inst{4} = Rn{4};
+    let DecoderMethod = "DecodeVSTInstruction";
+    let AsmMatchConverter = "cvtVSTwbRegister";
+  }
 }
 
-def VST2d8_UPD  : VST2DWB<0b1000, {0,0,?,?}, "8",  VecListTwoD>;
-def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
-def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
+defm VST2d8wb    : VST2DWB<0b1000, {0,0,?,?}, "8",  VecListTwoD>;
+defm VST2d16wb   : VST2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
+defm VST2d32wb   : VST2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
 
-def VST2q8_UPD  : VST2QWB<{0,0,?,?}, "8">;
-def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
-def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
+defm VST2q8wb    : VST2QWB<{0,0,?,?}, "8">;
+defm VST2q16wb   : VST2QWB<{0,1,?,?}, "16">;
+defm VST2q32wb   : VST2QWB<{1,0,?,?}, "32">;
 
-def VST2d8Pseudo_UPD  : VSTQWBPseudo<IIC_VST2u>;
-def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
-def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
+def VST2d8PseudoWB_fixed     : VSTQWBPseudo<IIC_VST2u>;
+def VST2d16PseudoWB_fixed    : VSTQWBPseudo<IIC_VST2u>;
+def VST2d32PseudoWB_fixed    : VSTQWBPseudo<IIC_VST2u>;
+def VST2d8PseudoWB_register  : VSTQWBPseudo<IIC_VST2u>;
+def VST2d16PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
+def VST2d32PseudoWB_register : VSTQWBPseudo<IIC_VST2u>;
 
-def VST2q8Pseudo_UPD  : VSTQQWBPseudo<IIC_VST2x2u>;
-def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
-def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q8PseudoWB_fixed     : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q16PseudoWB_fixed    : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q32PseudoWB_fixed    : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q8PseudoWB_register  : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q16PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
+def VST2q32PseudoWB_register : VSTQQWBPseudo<IIC_VST2x2u>;
 
 // ...with double-spaced registers
 def VST2b8      : VST2<0b1001, {0,0,?,?}, "8",  VecListTwoQ, IIC_VST2>;
 def VST2b16     : VST2<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VST2>;
 def VST2b32     : VST2<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VST2>;
-def VST2b8_UPD  : VST2DWB<0b1001, {0,0,?,?}, "8",  VecListTwoQ>;
-def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
-def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
+defm VST2b8wb   : VST2DWB<0b1001, {0,0,?,?}, "8",  VecListTwoQ>;
+defm VST2b16wb  : VST2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
+defm VST2b32wb  : VST2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
 
 //   VST3     : Vector Store (multiple 3-element structures)
 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
diff --git a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
index dee04ead53..45f285c395 100644
--- a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
+++ b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp
@@ -2221,15 +2221,24 @@ static DecodeStatus DecodeVSTInstruction(llvm::MCInst &Inst, unsigned Insn,
     case ARM::VST1d16Qwb_register:
     case ARM::VST1d32Qwb_register:
     case ARM::VST1d64Qwb_register:
-    case ARM::VST2d8_UPD:
-    case ARM::VST2d16_UPD:
-    case ARM::VST2d32_UPD:
-    case ARM::VST2q8_UPD:
-    case ARM::VST2q16_UPD:
-    case ARM::VST2q32_UPD:
-    case ARM::VST2b8_UPD:
-    case ARM::VST2b16_UPD:
-    case ARM::VST2b32_UPD:
+    case ARM::VST2d8wb_fixed:
+    case ARM::VST2d16wb_fixed:
+    case ARM::VST2d32wb_fixed:
+    case ARM::VST2d8wb_register:
+    case ARM::VST2d16wb_register:
+    case ARM::VST2d32wb_register:
+    case ARM::VST2q8wb_fixed:
+    case ARM::VST2q16wb_fixed:
+    case ARM::VST2q32wb_fixed:
+    case ARM::VST2q8wb_register:
+    case ARM::VST2q16wb_register:
+    case ARM::VST2q32wb_register:
+    case ARM::VST2b8wb_fixed:
+    case ARM::VST2b16wb_fixed:
+    case ARM::VST2b32wb_fixed:
+    case ARM::VST2b8wb_register:
+    case ARM::VST2b16wb_register:
+    case ARM::VST2b32wb_register:
     case ARM::VST3d8_UPD:
     case ARM::VST3d16_UPD:
     case ARM::VST3d32_UPD: