X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FARM%2FARMInstrFormats.td;h=a547fd0444f0e3ba6f4ded95f8ceb75c88b310df;hb=61512ba251097888963a8f07a35605564bcfc537;hp=891c2976f5f1a58bf507beb8f5ae916009c5c5b3;hpb=d2fbdb7f5c85d2191514953bdba0fae7b788e623;p=oota-llvm.git

diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td
index 891c2976f5f..a547fd0444f 100644
--- a/lib/Target/ARM/ARMInstrFormats.td
+++ b/lib/Target/ARM/ARMInstrFormats.td
@@ -1,4 +1,4 @@
-//===- ARMInstrFormats.td - ARM Instruction Formats --*- tablegen -*---------=//
+//===- ARMInstrFormats.td - ARM Instruction Formats ----------*- tablegen -*-=//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -71,7 +71,7 @@ def NVTBLFrm      : Format<41>;
 
 // Misc flags.
 
-// the instruction has a Rn register operand.
+// The instruction has an Rn register operand.
 // UnaryDP - Indicates this is a unary data processing instruction, i.e.
 // it doesn't have a Rn operand.
 class UnaryDP    { bit isUnaryDataProc = 1; }
@@ -84,6 +84,7 @@ class Xform16Bit { bit canXformTo16Bit = 1; }
 // ARM Instruction flags.  These need to match ARMBaseInstrInfo.h.
 //
 
+// FIXME: Once the JIT is MC-ized, these can go away.
 // Addressing mode.
 class AddrMode<bits<5> val> {
   bits<5> Value = val;
@@ -126,16 +127,16 @@ def IndexModePost : IndexMode<2>;
 def IndexModeUpd  : IndexMode<3>;
 
 // Instruction execution domain.
-class Domain<bits<2> val> {
-  bits<2> Value = val;
+class Domain<bits<3> val> {
+  bits<3> Value = val;
 }
 def GenericDomain : Domain<0>;
 def VFPDomain     : Domain<1>; // Instructions in VFP domain only
 def NeonDomain    : Domain<2>; // Instructions in Neon domain only
 def VFPNeonDomain : Domain<3>; // Instructions in both VFP & Neon domains
+def VFPNeonA8Domain : Domain<5>; // Instructions in VFP & Neon under A8
 
 //===----------------------------------------------------------------------===//
-
 // ARM special operands.
 //
 
@@ -144,6 +145,39 @@ def CondCodeOperand : AsmOperandClass {
   let SuperClasses = [];
 }
 
+def CCOutOperand : AsmOperandClass {
+  let Name = "CCOut";
+  let SuperClasses = [];
+}
+
+def MemBarrierOptOperand : AsmOperandClass {
+  let Name = "MemBarrierOpt";
+  let SuperClasses = [];
+  let ParserMethod = "tryParseMemBarrierOptOperand";
+}
+
+def ProcIFlagsOperand : AsmOperandClass {
+  let Name = "ProcIFlags";
+  let SuperClasses = [];
+  let ParserMethod = "tryParseProcIFlagsOperand";
+}
+
+def MSRMaskOperand : AsmOperandClass {
+  let Name = "MSRMask";
+  let SuperClasses = [];
+  let ParserMethod = "tryParseMSRMaskOperand";
+}
+
+// ARM imod and iflag operands, used only by the CPS instruction.
+def imod_op : Operand<i32> {
+  let PrintMethod = "printCPSIMod";
+}
+
+def iflags_op : Operand<i32> {
+  let PrintMethod = "printCPSIFlag";
+  let ParserMatchClass = ProcIFlagsOperand;
+}
+
 // ARM Predicate operand. Default to 14 = always (AL). Second part is CC
 // register whose default is 0 (no register).
 def pred : PredicateOperand<OtherVT, (ops i32imm, CCR),
@@ -154,14 +188,16 @@ def pred : PredicateOperand<OtherVT, (ops i32imm, CCR),
 
 // Conditional code result for instructions whose 's' bit is set, e.g. subs.
 def cc_out : OptionalDefOperand<OtherVT, (ops CCR), (ops (i32 zero_reg))> {
-  string EncoderMethod = "getCCOutOpValue";
+  let EncoderMethod = "getCCOutOpValue";
   let PrintMethod = "printSBitModifierOperand";
+  let ParserMatchClass = CCOutOperand;
 }
 
 // Same as cc_out except it defaults to setting CPSR.
 def s_cc_out : OptionalDefOperand<OtherVT, (ops CCR), (ops (i32 CPSR))> {
-  string EncoderMethod = "getCCOutOpValue";
+  let EncoderMethod = "getCCOutOpValue";
   let PrintMethod = "printSBitModifierOperand";
+  let ParserMatchClass = CCOutOperand;
 }
 
 // ARM special operands for disassembly only.
@@ -170,22 +206,33 @@ def setend_op : Operand<i32> {
   let PrintMethod = "printSetendOperand";
 }
 
-def cps_opt : Operand<i32> {
-  let PrintMethod = "printCPSOptionOperand";
-}
-
 def msr_mask : Operand<i32> {
   let PrintMethod = "printMSRMaskOperand";
+  let ParserMatchClass = MSRMaskOperand;
 }
 
-// A8.6.117, A8.6.118.  Different instructions are generated for #0 and #-0.
-// The neg_zero operand translates -0 to -1, -1 to -2, ..., etc.
-def neg_zero : Operand<i32> {
-  let PrintMethod = "printNegZeroOperand";
+// Shift Right Immediate - A shift right immediate is encoded differently from
+// other shift immediates. The imm6 field is encoded like so:
+//
+//    Offset    Encoding
+//     8        imm6<5:3> = '001', 8 - <imm> is encoded in imm6<2:0>
+//     16       imm6<5:4> = '01', 16 - <imm> is encoded in imm6<3:0>
+//     32       imm6<5> = '1', 32 - <imm> is encoded in imm6<4:0>
+//     64       64 - <imm> is encoded in imm6<5:0>
+def shr_imm8  : Operand<i32> {
+  let EncoderMethod = "getShiftRight8Imm";
+}
+def shr_imm16 : Operand<i32> {
+  let EncoderMethod = "getShiftRight16Imm";
+}
+def shr_imm32 : Operand<i32> {
+  let EncoderMethod = "getShiftRight32Imm";
+}
+def shr_imm64 : Operand<i32> {
+  let EncoderMethod = "getShiftRight64Imm";
 }
 
 //===----------------------------------------------------------------------===//
-
 // ARM Instruction templates.
 //
 
@@ -204,6 +251,9 @@ class InstTemplate<AddrMode am, SizeFlagVal sz, IndexMode im,
   bit isUnaryDataProc = 0;
   bit canXformTo16Bit = 0;
 
+  // If this is a pseudo instruction, mark it isCodeGenOnly.
+  let isCodeGenOnly = !eq(!cast<string>(f), "Pseudo");
+
   // The layout of TSFlags should be kept in sync with ARMBaseInstrInfo.h.
   let TSFlags{4-0}   = AM.Value;
   let TSFlags{7-5}   = SZ.Value;
@@ -211,7 +261,7 @@ class InstTemplate<AddrMode am, SizeFlagVal sz, IndexMode im,
   let TSFlags{15-10} = Form;
   let TSFlags{16}    = isUnaryDataProc;
   let TSFlags{17}    = canXformTo16Bit;
-  let TSFlags{19-18} = D.Value;
+  let TSFlags{20-18} = D.Value;
 
   let Constraints = cstr;
   let Itinerary = itin;
@@ -231,16 +281,41 @@ class InstThumb<AddrMode am, SizeFlagVal sz, IndexMode im,
                 Format f, Domain d, string cstr, InstrItinClass itin>
   : InstTemplate<am, sz, im, f, d, cstr, itin>;
 
-class PseudoInst<dag oops, dag iops, InstrItinClass itin,
-                 string asm, list<dag> pattern>
+class PseudoInst<dag oops, dag iops, InstrItinClass itin, list<dag> pattern>
+  // FIXME: This really should derive from InstTemplate instead, as pseudos
+  //        don't need encoding information. TableGen doesn't like that
+  //        currently. Need to figure out why and fix it.
   : InstARM<AddrModeNone, SizeSpecial, IndexModeNone, Pseudo, GenericDomain,
             "", itin> {
   let OutOperandList = oops;
   let InOperandList = iops;
-  let AsmString = asm;
   let Pattern = pattern;
+  let isCodeGenOnly = 1;
 }
 
+// PseudoInst that's ARM-mode only.
+class ARMPseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin,
+                    list<dag> pattern>
+  : PseudoInst<oops, iops, itin, pattern> {
+  let SZ = sz;
+  list<Predicate> Predicates = [IsARM];
+}
+
+// PseudoInst that's Thumb-mode only.
+class tPseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin,
+                    list<dag> pattern>
+  : PseudoInst<oops, iops, itin, pattern> {
+  let SZ = sz;
+  list<Predicate> Predicates = [IsThumb];
+}
+
+// PseudoInst that's Thumb2-mode only.
+class t2PseudoInst<dag oops, dag iops, SizeFlagVal sz, InstrItinClass itin,
+                    list<dag> pattern>
+  : PseudoInst<oops, iops, itin, pattern> {
+  let SZ = sz;
+  list<Predicate> Predicates = [IsThumb2];
+}
 // Almost all ARM instructions are predicable.
 class I<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
         IndexMode im, Format f, InstrItinClass itin,
@@ -332,10 +407,6 @@ class ABXI<bits<4> opcod, dag oops, dag iops, InstrItinClass itin,
        asm, "", pattern> {
   let Inst{27-24} = opcod;
 }
-class ABXIx2<dag oops, dag iops, InstrItinClass itin,
-             string asm, list<dag> pattern>
-  : XI<oops, iops, AddrModeNone, Size8Bytes, IndexModeNone, Pseudo, itin,
-       asm, "", pattern>;
 
 // BR_JT instructions
 class JTI<dag oops, dag iops, InstrItinClass itin,
@@ -348,19 +419,42 @@ class AIldrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin,
               string opc, string asm, list<dag> pattern>
   : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin,
       opc, asm, "", pattern> {
+  bits<4> Rt;
+  bits<4> Rn;
   let Inst{27-23} = 0b00011;
   let Inst{22-21} = opcod;
   let Inst{20}    = 1;
+  let Inst{19-16} = Rn;
+  let Inst{15-12} = Rt;
   let Inst{11-0}  = 0b111110011111;
 }
 class AIstrex<bits<2> opcod, dag oops, dag iops, InstrItinClass itin,
               string opc, string asm, list<dag> pattern>
   : I<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, LdStExFrm, itin,
       opc, asm, "", pattern> {
+  bits<4> Rd;
+  bits<4> Rt;
+  bits<4> addr;
   let Inst{27-23} = 0b00011;
   let Inst{22-21} = opcod;
   let Inst{20}    = 0;
+  let Inst{19-16} = addr;
+  let Inst{15-12} = Rd;
   let Inst{11-4}  = 0b11111001;
+  let Inst{3-0}   = Rt;
+}
+class AIswp<bit b, dag oops, dag iops, string opc, list<dag> pattern>
+  : AI<oops, iops, MiscFrm, NoItinerary, opc, "\t$Rt, $Rt2, [$Rn]", pattern> {
+  bits<4> Rt;
+  bits<4> Rt2;
+  bits<4> Rn;
+  let Inst{27-23} = 0b00010;
+  let Inst{22} = b;
+  let Inst{21-20} = 0b00;
+  let Inst{19-16} = Rn;
+  let Inst{15-12} = Rt;
+  let Inst{11-4} = 0b00001001;
+  let Inst{3-0} = Rt2;
 }
 
 // addrmode1 instructions
@@ -385,419 +479,171 @@ class AXI1<bits<4> opcod, dag oops, dag iops, Format f, InstrItinClass itin,
   let Inst{24-21} = opcod;
   let Inst{27-26} = 0b00;
 }
-class AI1x2<dag oops, dag iops, Format f, InstrItinClass itin,
-            string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode1, Size8Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern>;
-
-
-// addrmode2 loads and stores
-class AI2<dag oops, dag iops, Format f, InstrItinClass itin,
-          string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{27-26} = 0b01;
-}
 
 // loads
 
-// LDR/LDRB
-class AIldr1<bits<3> op, bit opc22, dag oops, dag iops, AddrMode am, Format f,
-             InstrItinClass itin, string opc, string asm, list<dag> pattern>
-  : I<oops, iops, am, Size4Bytes, IndexModeNone, f, itin, opc, asm,
-      "", pattern> {
-  let Inst{27-25} = op;
-  let Inst{24} = 1;  // 24 == P
-  // 23 == U
-  let Inst{22} = opc22;
-  let Inst{21} = 0;  // 21 == W
-  let Inst{20} = 1;
-}
-// LDRH/LDRSB/LDRSH/LDRD
-class AIldr2<bits<4> op, bit opc22, bit opc20, dag oops, dag iops, AddrMode am,
+// LDR/LDRB/STR/STRB/...
+class AI2ldst<bits<3> op, bit isLd, bit isByte, dag oops, dag iops, AddrMode am,
              Format f, InstrItinClass itin, string opc, string asm,
              list<dag> pattern>
   : I<oops, iops, am, Size4Bytes, IndexModeNone, f, itin, opc, asm,
       "", pattern> {
-  let Inst{27-25} = 0b000;
+  let Inst{27-25} = op;
   let Inst{24} = 1;  // 24 == P
   // 23 == U
-  let Inst{22} = opc22;
+  let Inst{22} = isByte;
   let Inst{21} = 0;  // 21 == W
-  let Inst{20} = opc20;
-
-  let Inst{7-4} = op;
-}
-
-
-
-
-class AI2ldw<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AXI2ldw<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AI2ldb<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AXI2ldb<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-
-// stores
-class AI2stw<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AXI2stw<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AI2stb<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AXI2stb<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode2, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-
-// Pre-indexed loads
-class AI2ldwpr<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AI2ldbpr<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-
-// Pre-indexed stores
-class AI2stwpr<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
+  let Inst{20} = isLd;
 }
-class AI2stbpr<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePre, f, itin,
+// Indexed load/stores
+class AI2ldstidx<bit isLd, bit isByte, bit isPre, dag oops, dag iops,
+                IndexMode im, Format f, InstrItinClass itin, string opc,
+                string asm, string cstr, list<dag> pattern>
+  : I<oops, iops, AddrMode2, Size4Bytes, im, f, itin,
       opc, asm, cstr, pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-26} = 0b01;
-}
-
-// Post-indexed loads
-class AI2ldwpo<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AI2ldbpo<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-26} = 0b01;
-}
-
-// Post-indexed stores
-class AI2stwpo<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 0; // B bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-26} = 0b01;
-}
-class AI2stbpo<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode2, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{22}    = 1; // B bit
-  let Inst{24}    = 0; // P bit
+  bits<4> Rt;
   let Inst{27-26} = 0b01;
+  let Inst{24}    = isPre; // P bit
+  let Inst{22}    = isByte; // B bit
+  let Inst{21}    = isPre; // W bit
+  let Inst{20}    = isLd; // L bit
+  let Inst{15-12} = Rt;
+}
+class AI2stridx<bit isByte, bit isPre, dag oops, dag iops,
+                IndexMode im, Format f, InstrItinClass itin, string opc,
+                string asm, string cstr, list<dag> pattern>
+  : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr,
+               pattern> {
+  // AM2 store w/ two operands: (GPR, am2offset)
+  // {13}     1 == Rm, 0 == imm12
+  // {12}     isAdd
+  // {11-0}   imm12/Rm
+  bits<14> offset;
+  bits<4> Rn;
+  let Inst{25} = offset{13};
+  let Inst{23} = offset{12};
+  let Inst{19-16} = Rn;
+  let Inst{11-0} = offset{11-0};
+}
+// FIXME: Merge with the above class when addrmode2 gets used for STR, STRB
+// but for now use this class for STRT and STRBT.
+class AI2stridxT<bit isByte, bit isPre, dag oops, dag iops,
+                IndexMode im, Format f, InstrItinClass itin, string opc,
+                string asm, string cstr, list<dag> pattern>
+  : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr,
+               pattern> {
+  // AM2 store w/ two operands: (GPR, am2offset)
+  // {17-14}  Rn
+  // {13}     1 == Rm, 0 == imm12
+  // {12}     isAdd
+  // {11-0}   imm12/Rm
+  bits<18> addr;
+  let Inst{25} = addr{13};
+  let Inst{23} = addr{12};
+  let Inst{19-16} = addr{17-14};
+  let Inst{11-0} = addr{11-0};
 }
 
 // addrmode3 instructions
-class AI3<dag oops, dag iops, Format f, InstrItinClass itin,
-          string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern>;
-class AXI3<dag oops, dag iops, Format f, InstrItinClass itin,
-           string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern>;
-
-// loads
-class AI3ldh<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
+class AI3ld<bits<4> op, bit op20, dag oops, dag iops, Format f,
+            InstrItinClass itin, string opc, string asm, list<dag> pattern>
   : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
       opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
+  bits<14> addr;
+  bits<4> Rt;
   let Inst{27-25} = 0b000;
-}
-class AXI3ldh<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
-}
-class AI3ldsh<dag oops, dag iops, Format f, InstrItinClass itin,
-              string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
+  let Inst{24}    = 1;            // P bit
+  let Inst{23}    = addr{8};      // U bit
+  let Inst{22}    = addr{13};     // 1 == imm8, 0 == Rm
+  let Inst{21}    = 0;            // W bit
+  let Inst{20}    = op20;         // L bit
+  let Inst{19-16} = addr{12-9};   // Rn
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{11-8}  = addr{7-4};    // imm7_4/zero
+  let Inst{7-4}   = op;
+  let Inst{3-0}   = addr{3-0};    // imm3_0/Rm
+}
+
+class AI3ldstidx<bits<4> op, bit op20, bit isLd, bit isPre, dag oops, dag iops,
+                IndexMode im, Format f, InstrItinClass itin, string opc,
+                string asm, string cstr, list<dag> pattern>
+  : I<oops, iops, AddrMode3, Size4Bytes, im, f, itin,
+      opc, asm, cstr, pattern> {
+  bits<4> Rt;
   let Inst{27-25} = 0b000;
-}
-class AXI3ldsh<dag oops, dag iops, Format f, InstrItinClass itin,
-               string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
-}
-class AI3ldsb<dag oops, dag iops, Format f, InstrItinClass itin,
-              string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
+  let Inst{24}    = isPre;        // P bit
+  let Inst{21}    = isPre;        // W bit
+  let Inst{20}    = op20;         // L bit
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{7-4}   = op;
+}
+
+// FIXME: Merge with the above class when addrmode2 gets used for LDR, LDRB
+// but for now use this class for LDRSBT, LDRHT, LDSHT.
+class AI3ldstidxT<bits<4> op, bit op20, bit isLd, bit isPre, dag oops, dag iops,
+                  IndexMode im, Format f, InstrItinClass itin, string opc,
+                  string asm, string cstr, list<dag> pattern>
+  : I<oops, iops, AddrMode3, Size4Bytes, im, f, itin,
+      opc, asm, cstr, pattern> {
+  // {13}     1 == imm8, 0 == Rm
+  // {12-9}   Rn
+  // {8}      isAdd
+  // {7-4}    imm7_4/zero
+  // {3-0}    imm3_0/Rm
+  bits<14> addr;
+  bits<4> Rt;
   let Inst{27-25} = 0b000;
-}
-class AXI3ldsb<dag oops, dag iops, Format f, InstrItinClass itin,
-               string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
-}
-class AI3ldd<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
+  let Inst{24}    = isPre;        // P bit
+  let Inst{23}    = addr{8};      // U bit
+  let Inst{22}    = addr{13};     // 1 == imm8, 0 == Rm
+  let Inst{20}    = op20;         // L bit
+  let Inst{19-16} = addr{12-9};   // Rn
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{11-8}  = addr{7-4};    // imm7_4/zero
+  let Inst{7-4}   = op;
+  let Inst{3-0}   = addr{3-0};    // imm3_0/Rm
+  let AsmMatchConverter = "CvtLdWriteBackRegAddrMode3";
+}
+
+class AI3stridx<bits<4> op, bit isByte, bit isPre, dag oops, dag iops,
+                IndexMode im, Format f, InstrItinClass itin, string opc,
+                string asm, string cstr, list<dag> pattern>
+  : AI2ldstidx<0, isByte, isPre, oops, iops, im, f, itin, opc, asm, cstr,
+               pattern> {
+  // AM3 store w/ two operands: (GPR, am3offset)
+  bits<14> offset;
+  bits<4> Rt;
+  bits<4> Rn;
   let Inst{27-25} = 0b000;
+  let Inst{23}    = offset{8};
+  let Inst{22}    = offset{9};
+  let Inst{19-16} = Rn;
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{11-8}  = offset{7-4};  // imm7_4/zero
+  let Inst{7-4}   = op;
+  let Inst{3-0}   = offset{3-0};  // imm3_0/Rm
 }
 
 // stores
-class AI3sth<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-      opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AXI3sth<dag oops, dag iops, Format f, InstrItinClass itin,
-              string asm, list<dag> pattern>
-  : XI<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
-       asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
-}
-class AI3std<dag oops, dag iops, Format f, InstrItinClass itin,
+class AI3str<bits<4> op, dag oops, dag iops, Format f, InstrItinClass itin,
              string opc, string asm, list<dag> pattern>
   : I<oops, iops, AddrMode3, Size4Bytes, IndexModeNone, f, itin,
       opc, asm, "", pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 1; // P bit
+  bits<14> addr;
+  bits<4> Rt;
   let Inst{27-25} = 0b000;
+  let Inst{24}    = 1;            // P bit
+  let Inst{23}    = addr{8};      // U bit
+  let Inst{22}    = addr{13};     // 1 == imm8, 0 == Rm
+  let Inst{21}    = 0;            // W bit
+  let Inst{20}    = 0;            // L bit
+  let Inst{19-16} = addr{12-9};   // Rn
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{11-8}  = addr{7-4};    // imm7_4/zero
+  let Inst{7-4}   = op;
+  let Inst{3-0}   = addr{3-0};    // imm3_0/Rm
 }
 
-// Pre-indexed loads
-class AI3ldhpr<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3ldshpr<dag oops, dag iops, Format f, InstrItinClass itin,
-                string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3ldsbpr<dag oops, dag iops, Format f, InstrItinClass itin,
-                string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3lddpr<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePre, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 1; // W bit
-  let Inst{24}    = 1; // P bit
-  let Inst{27-25} = 0b000;
-}
-
-
 // Pre-indexed stores
 class AI3sthpr<dag oops, dag iops, Format f, InstrItinClass itin,
                string opc, string asm, string cstr, list<dag> pattern>
@@ -826,71 +672,30 @@ class AI3stdpr<dag oops, dag iops, Format f, InstrItinClass itin,
   let Inst{27-25} = 0b000;
 }
 
-// Post-indexed loads
-class AI3ldhpo<dag oops, dag iops, Format f, InstrItinClass itin,
-               string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 0; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3ldshpo<dag oops, dag iops, Format f, InstrItinClass itin,
-                string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 1; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3ldsbpo<dag oops, dag iops, Format f, InstrItinClass itin,
-                string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr,pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 1; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-25} = 0b000;
-}
-class AI3lddpo<dag oops, dag iops, Format f, InstrItinClass itin,
-             string opc, string asm, string cstr, list<dag> pattern>
-  : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin,
-      opc, asm, cstr, pattern> {
-  let Inst{4}     = 1;
-  let Inst{5}     = 0; // H bit
-  let Inst{6}     = 1; // S bit
-  let Inst{7}     = 1;
-  let Inst{20}    = 0; // L bit
-  let Inst{21}    = 0; // W bit
-  let Inst{24}    = 0; // P bit
-  let Inst{27-25} = 0b000;
-}
-
 // Post-indexed stores
 class AI3sthpo<dag oops, dag iops, Format f, InstrItinClass itin,
                string opc, string asm, string cstr, list<dag> pattern>
   : I<oops, iops, AddrMode3, Size4Bytes, IndexModePost, f, itin,
       opc, asm, cstr,pattern> {
+  // {13}     1 == imm8, 0 == Rm
+  // {12-9}   Rn
+  // {8}      isAdd
+  // {7-4}    imm7_4/zero
+  // {3-0}    imm3_0/Rm
+  bits<14> addr;
+  bits<4> Rt;
+  let Inst{3-0}   = addr{3-0};    // imm3_0/Rm
   let Inst{4}     = 1;
   let Inst{5}     = 1; // H bit
   let Inst{6}     = 0; // S bit
   let Inst{7}     = 1;
+  let Inst{11-8}  = addr{7-4};    // imm7_4/zero
+  let Inst{15-12} = Rt;           // Rt
+  let Inst{19-16} = addr{12-9};   // Rn
   let Inst{20}    = 0; // L bit
   let Inst{21}    = 0; // W bit
+  let Inst{22}    = addr{13};     // 1 == imm8, 0 == Rm
+  let Inst{23}    = addr{8};      // U bit
   let Inst{24}    = 0; // P bit
   let Inst{27-25} = 0b000;
 }
@@ -909,21 +714,17 @@ class AI3stdpo<dag oops, dag iops, Format f, InstrItinClass itin,
 }
 
 // addrmode4 instructions
-class AXI4ld<dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin,
-             string asm, string cstr, list<dag> pattern>
-  : XI<oops, iops, AddrMode4, Size4Bytes, im, f, itin,
-       asm, cstr, pattern> {
-  let Inst{20}    = 1; // L bit
-  let Inst{22}    = 0; // S bit
+class AXI4<dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin,
+           string asm, string cstr, list<dag> pattern>
+  : XI<oops, iops, AddrMode4, Size4Bytes, im, f, itin, asm, cstr, pattern> {
+  bits<4>  p;
+  bits<16> regs;
+  bits<4>  Rn;
+  let Inst{31-28} = p;
   let Inst{27-25} = 0b100;
-}
-class AXI4st<dag oops, dag iops, IndexMode im, Format f, InstrItinClass itin,
-             string asm, string cstr, list<dag> pattern>
-  : XI<oops, iops, AddrMode4, Size4Bytes, im, f, itin,
-       asm, cstr, pattern> {
-  let Inst{20}    = 0; // L bit
   let Inst{22}    = 0; // S bit
-  let Inst{27-25} = 0b100;
+  let Inst{19-16} = Rn;
+  let Inst{15-0}  = regs;
 }
 
 // Unsigned multiply, multiply-accumulate instructions.
@@ -1059,6 +860,9 @@ class APKHI<bits<8> opcod, bit tb, dag oops, dag iops, InstrItinClass itin,
 class ARMPat<dag pattern, dag result> : Pat<pattern, result> {
   list<Predicate> Predicates = [IsARM];
 }
+class ARMV5TPat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [IsARM, HasV5T];
+}
 class ARMV5TEPat<dag pattern, dag result> : Pat<pattern, result> {
   list<Predicate> Predicates = [IsARM, HasV5TE];
 }
@@ -1067,12 +871,9 @@ class ARMV6Pat<dag pattern, dag result> : Pat<pattern, result> {
 }
 
 //===----------------------------------------------------------------------===//
-//
 // Thumb Instruction Format Definitions.
 //
 
-// TI - Thumb instruction.
-
 class ThumbI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
              InstrItinClass itin, string asm, string cstr, list<dag> pattern>
   : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> {
@@ -1083,6 +884,7 @@ class ThumbI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   list<Predicate> Predicates = [IsThumb];
 }
 
+// TI - Thumb instruction.
 class TI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern>
   : ThumbI<oops, iops, AddrModeNone, Size2Bytes, itin, asm, "", pattern>;
 
@@ -1103,6 +905,13 @@ class TIx2<bits<5> opcod1, bits<2> opcod2, bit opcod3,
   let Inst{12}    = opcod3;
 }
 
+// Move to/from coprocessor instructions
+class T1Cop<dag oops, dag iops, string asm, list<dag> pattern>
+  : ThumbI<oops, iops, AddrModeNone, Size4Bytes, NoItinerary, asm, "", pattern>,
+    Encoding, Requires<[IsThumb, HasV6]> {
+  let Inst{31-28} = 0b1110;
+}
+
 // BR_JT instructions
 class TJTI<dag oops, dag iops, InstrItinClass itin, string asm,
            list<dag> pattern>
@@ -1116,7 +925,7 @@ class Thumb1I<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   let InOperandList = iops;
   let AsmString = asm;
   let Pattern = pattern;
-  list<Predicate> Predicates = [IsThumb1Only];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only];
 }
 
 class T1I<dag oops, dag iops, InstrItinClass itin,
@@ -1125,9 +934,6 @@ class T1I<dag oops, dag iops, InstrItinClass itin,
 class T1Ix2<dag oops, dag iops, InstrItinClass itin,
             string asm, list<dag> pattern>
   : Thumb1I<oops, iops, AddrModeNone, Size4Bytes, itin, asm, "", pattern>;
-class T1JTI<dag oops, dag iops, InstrItinClass itin,
-            string asm, list<dag> pattern>
-  : Thumb1I<oops, iops, AddrModeNone, SizeSpecial, itin, asm, "", pattern>;
 
 // Two-address instructions
 class T1It<dag oops, dag iops, InstrItinClass itin,
@@ -1144,7 +950,7 @@ class Thumb1sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   let InOperandList = !con(iops, (ins pred:$p));
   let AsmString = !strconcat(opc, "${s}${p}", asm);
   let Pattern = pattern;
-  list<Predicate> Predicates = [IsThumb1Only];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only];
 }
 
 class T1sI<dag oops, dag iops, InstrItinClass itin,
@@ -1155,7 +961,7 @@ class T1sI<dag oops, dag iops, InstrItinClass itin,
 class T1sIt<dag oops, dag iops, InstrItinClass itin,
             string opc, string asm, list<dag> pattern>
   : Thumb1sI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm,
-             "$lhs = $dst", pattern>;
+             "$Rn = $Rdn", pattern>;
 
 // Thumb1 instruction that can be predicated.
 class Thumb1pI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
@@ -1166,7 +972,7 @@ class Thumb1pI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   let InOperandList = !con(iops, (ins pred:$p));
   let AsmString = !strconcat(opc, "${p}", asm);
   let Pattern = pattern;
-  list<Predicate> Predicates = [IsThumb1Only];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only];
 }
 
 class T1pI<dag oops, dag iops, InstrItinClass itin,
@@ -1177,17 +983,8 @@ class T1pI<dag oops, dag iops, InstrItinClass itin,
 class T1pIt<dag oops, dag iops, InstrItinClass itin,
             string opc, string asm, list<dag> pattern>
   : Thumb1pI<oops, iops, AddrModeNone, Size2Bytes, itin, opc, asm,
-             "$lhs = $dst", pattern>;
+             "$Rn = $Rdn", pattern>;
 
-class T1pI1<dag oops, dag iops, InstrItinClass itin,
-            string opc, string asm, list<dag> pattern>
-  : Thumb1pI<oops, iops, AddrModeT1_1, Size2Bytes, itin, opc, asm, "", pattern>;
-class T1pI2<dag oops, dag iops, InstrItinClass itin,
-            string opc, string asm, list<dag> pattern>
-  : Thumb1pI<oops, iops, AddrModeT1_2, Size2Bytes, itin, opc, asm, "", pattern>;
-class T1pI4<dag oops, dag iops, InstrItinClass itin,
-            string opc, string asm, list<dag> pattern>
-  : Thumb1pI<oops, iops, AddrModeT1_4, Size2Bytes, itin, opc, asm, "", pattern>;
 class T1pIs<dag oops, dag iops,
             InstrItinClass itin, string opc, string asm, list<dag> pattern>
   : Thumb1pI<oops, iops, AddrModeT1_s, Size2Bytes, itin, opc, asm, "", pattern>;
@@ -1216,7 +1013,7 @@ class T1DataProcessing<bits<4> opcode> : Encoding16 {
 // A6.2.3 Special data instructions and branch and exchange encoding.
 class T1Special<bits<4> opcode> : Encoding16 {
   let Inst{15-10} = 0b010001;
-  let Inst{9-6} = opcode;
+  let Inst{9-6}   = opcode;
 }
 
 // A6.2.4 Load/store single data item encoding.
@@ -1224,12 +1021,37 @@ class T1LoadStore<bits<4> opA, bits<3> opB> : Encoding16 {
   let Inst{15-12} = opA;
   let Inst{11-9}  = opB;
 }
-class T1LdSt<bits<3> opB>     : T1LoadStore<0b0101, opB>;
-class T1LdSt4Imm<bits<3> opB> : T1LoadStore<0b0110, opB>; // Immediate, 4 bytes
-class T1LdSt1Imm<bits<3> opB> : T1LoadStore<0b0111, opB>; // Immediate, 1 byte
-class T1LdSt2Imm<bits<3> opB> : T1LoadStore<0b1000, opB>; // Immediate, 2 bytes
 class T1LdStSP<bits<3> opB>   : T1LoadStore<0b1001, opB>; // SP relative
 
+// Helper classes to encode Thumb1 loads and stores. For immediates, the
+// following bits are used for "opA" (see A6.2.4):
+//
+//   0b0110 => Immediate, 4 bytes
+//   0b1000 => Immediate, 2 bytes
+//   0b0111 => Immediate, 1 byte
+class T1pILdStEncode<bits<3> opcode, dag oops, dag iops, AddrMode am,
+                     InstrItinClass itin, string opc, string asm,
+                     list<dag> pattern>
+  : Thumb1pI<oops, iops, am, Size2Bytes, itin, opc, asm, "", pattern>,
+    T1LoadStore<0b0101, opcode> {
+  bits<3> Rt;
+  bits<8> addr;
+  let Inst{8-6} = addr{5-3};    // Rm
+  let Inst{5-3} = addr{2-0};    // Rn
+  let Inst{2-0} = Rt;
+}
+class T1pILdStEncodeImm<bits<4> opA, bit opB, dag oops, dag iops, AddrMode am,
+                        InstrItinClass itin, string opc, string asm,
+                        list<dag> pattern>
+  : Thumb1pI<oops, iops, am, Size2Bytes, itin, opc, asm, "", pattern>,
+    T1LoadStore<opA, {opB,?,?}> {
+  bits<3> Rt;
+  bits<8> addr;
+  let Inst{10-6} = addr{7-3};   // imm5
+  let Inst{5-3}  = addr{2-0};   // Rn
+  let Inst{2-0}  = Rt;
+}
+
 // A6.2.5 Miscellaneous 16-bit instructions encoding.
 class T1Misc<bits<7> opcode> : Encoding16 {
   let Inst{15-12} = 0b1011;
@@ -1258,6 +1080,9 @@ class Thumb2sI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
                InstrItinClass itin,
                string opc, string asm, string cstr, list<dag> pattern>
   : InstARM<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> {
+  bits<1> s; // condition-code set flag ('1' if the insn should set the flags)
+  let Inst{20} = s;
+
   let OutOperandList = oops;
   let InOperandList = !con(iops, (ins pred:$p, cc_out:$s));
   let AsmString = !strconcat(opc, "${s}${p}", asm);
@@ -1285,7 +1110,7 @@ class ThumbXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   let InOperandList = iops;
   let AsmString = asm;
   let Pattern = pattern;
-  list<Predicate> Predicates = [IsThumb1Only];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only];
 }
 
 class T2I<dag oops, dag iops, InstrItinClass itin,
@@ -1303,17 +1128,23 @@ class T2Iso<dag oops, dag iops, InstrItinClass itin,
 class T2Ipc<dag oops, dag iops, InstrItinClass itin,
             string opc, string asm, list<dag> pattern>
   : Thumb2I<oops, iops, AddrModeT2_pc, Size4Bytes, itin, opc, asm, "", pattern>;
-class T2Ii8s4<bit P, bit W, bit load, dag oops, dag iops, InstrItinClass itin,
+class T2Ii8s4<bit P, bit W, bit isLoad, dag oops, dag iops, InstrItinClass itin,
               string opc, string asm, list<dag> pattern>
   : Thumb2I<oops, iops, AddrModeT2_i8s4, Size4Bytes, itin, opc, asm, "",
             pattern> {
-  let Inst{31-27} = 0b11101;
-  let Inst{26-25} = 0b00;
+  bits<4> Rt;
+  bits<4> Rt2;
+  bits<13> addr;
+  let Inst{31-25} = 0b1110100;
   let Inst{24}    = P;
-  let Inst{23}    = ?; // The U bit.
+  let Inst{23}    = addr{8};
   let Inst{22}    = 1;
   let Inst{21}    = W;
-  let Inst{20}    = load;
+  let Inst{20}    = isLoad;
+  let Inst{19-16} = addr{12-9};
+  let Inst{15-12} = Rt{3-0};
+  let Inst{11-8}  = Rt2{3-0};
+  let Inst{7-0}   = addr{7-0};
 }
 
 class T2sI<dag oops, dag iops, InstrItinClass itin,
@@ -1327,9 +1158,11 @@ class T2JTI<dag oops, dag iops, InstrItinClass itin,
             string asm, list<dag> pattern>
   : Thumb2XI<oops, iops, AddrModeNone, SizeSpecial, itin, asm, "", pattern>;
 
-class T2Ix2<dag oops, dag iops, InstrItinClass itin,
-            string opc, string asm, list<dag> pattern>
-  : Thumb2I<oops, iops, AddrModeNone, Size8Bytes, itin, opc, asm, "", pattern>;
+// Move to/from coprocessor instructions
+class T2Cop<dag oops, dag iops, string asm, list<dag> pattern>
+  : T2XI<oops, iops, NoItinerary, asm, pattern>, Requires<[IsThumb2, HasV6]> {
+  let Inst{31-28} = 0b1111;
+}
 
 // Two-address instructions
 class T2XIt<dag oops, dag iops, InstrItinClass itin,
@@ -1357,29 +1190,30 @@ class T2Iidxldst<bit signed, bits<2> opcod, bit load, bit pre,
   // (P, W) = (1, 1) Pre-indexed or (0, 1) Post-indexed
   let Inst{10}    = pre; // The P bit.
   let Inst{8}     = 1; // The W bit.
-}
 
-// Helper class for disassembly only
-// A6.3.16 & A6.3.17
-// T2Imac - Thumb2 multiply [accumulate, and absolute difference] instructions.
-class T2I_mac<bit long, bits<3> op22_20, bits<4> op7_4, dag oops, dag iops,
-             InstrItinClass itin, string opc, string asm, list<dag> pattern>
-  : T2I<oops, iops, itin, opc, asm, pattern> {
-  let Inst{31-27} = 0b11111;
-  let Inst{26-24} = 0b011;
-  let Inst{23}    = long;
-  let Inst{22-20} = op22_20;
-  let Inst{7-4}   = op7_4;
+  bits<9> addr;
+  let Inst{7-0} = addr{7-0};
+  let Inst{9}   = addr{8}; // Sign bit
+
+  bits<4> Rt;
+  bits<4> Rn;
+  let Inst{15-12} = Rt{3-0};
+  let Inst{19-16} = Rn{3-0};
 }
 
 // Tv5Pat - Same as Pat<>, but requires V5T Thumb mode.
 class Tv5Pat<dag pattern, dag result> : Pat<pattern, result> {
-  list<Predicate> Predicates = [IsThumb1Only, HasV5T];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only, HasV5T];
 }
 
 // T1Pat - Same as Pat<>, but requires that the compiler be in Thumb1 mode.
 class T1Pat<dag pattern, dag result> : Pat<pattern, result> {
-  list<Predicate> Predicates = [IsThumb1Only];
+  list<Predicate> Predicates = [IsThumb, IsThumb1Only];
+}
+
+// T2v6Pat - Same as Pat<>, but requires V6T2 Thumb2 mode.
+class T2v6Pat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [IsThumb2, HasV6T2];
 }
 
 // T2Pat - Same as Pat<>, but requires that the compiler be in Thumb2 mode.
@@ -1404,6 +1238,7 @@ class VFPI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
   let InOperandList = !con(iops, (ins pred:$p));
   let AsmString = !strconcat(opc, "${p}", asm);
   let Pattern = pattern;
+  let PostEncoderMethod = "VFPThumb2PostEncoder";
   list<Predicate> Predicates = [HasVFP2];
 }
 
@@ -1412,17 +1247,22 @@ class VFPXI<dag oops, dag iops, AddrMode am, SizeFlagVal sz,
             IndexMode im, Format f, InstrItinClass itin,
             string asm, string cstr, list<dag> pattern>
   : InstARM<am, sz, im, f, VFPDomain, cstr, itin> {
+  bits<4> p;
+  let Inst{31-28} = p;
   let OutOperandList = oops;
   let InOperandList = iops;
   let AsmString = asm;
   let Pattern = pattern;
+  let PostEncoderMethod = "VFPThumb2PostEncoder";
   list<Predicate> Predicates = [HasVFP2];
 }
 
 class VFPAI<dag oops, dag iops, Format f, InstrItinClass itin,
             string opc, string asm, list<dag> pattern>
   : VFPI<oops, iops, AddrModeNone, Size4Bytes, IndexModeNone, f, itin,
-         opc, asm, "", pattern>;
+         opc, asm, "", pattern> {
+  let PostEncoderMethod = "VFPThumb2PostEncoder";
+}
 
 // ARM VFP addrmode5 loads and stores
 class ADI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops,
@@ -1430,13 +1270,24 @@ class ADI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops,
            string opc, string asm, list<dag> pattern>
   : VFPI<oops, iops, AddrMode5, Size4Bytes, IndexModeNone,
          VFPLdStFrm, itin, opc, asm, "", pattern> {
+  // Instruction operands.
+  bits<5>  Dd;
+  bits<13> addr;
+
+  // Encode instruction operands.
+  let Inst{23}    = addr{8};      // U (add = (U == '1'))
+  let Inst{22}    = Dd{4};
+  let Inst{19-16} = addr{12-9};   // Rn
+  let Inst{15-12} = Dd{3-0};
+  let Inst{7-0}   = addr{7-0};    // imm8
+
   // TODO: Mark the instructions with the appropriate subtarget info.
   let Inst{27-24} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{11-9}  = 0b101;
   let Inst{8}     = 1;          // Double precision
 
-  // 64-bit loads & stores operate on both NEON and VFP pipelines.
+  // Loads & stores operate on both NEON and VFP pipelines.
   let D = VFPNeonDomain;
 }
 
@@ -1445,11 +1296,25 @@ class ASI5<bits<4> opcod1, bits<2> opcod2, dag oops, dag iops,
            string opc, string asm, list<dag> pattern>
   : VFPI<oops, iops, AddrMode5, Size4Bytes, IndexModeNone,
          VFPLdStFrm, itin, opc, asm, "", pattern> {
+  // Instruction operands.
+  bits<5>  Sd;
+  bits<13> addr;
+
+  // Encode instruction operands.
+  let Inst{23}    = addr{8};      // U (add = (U == '1'))
+  let Inst{22}    = Sd{0};
+  let Inst{19-16} = addr{12-9};   // Rn
+  let Inst{15-12} = Sd{4-1};
+  let Inst{7-0}   = addr{7-0};    // imm8
+
   // TODO: Mark the instructions with the appropriate subtarget info.
   let Inst{27-24} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{11-9}  = 0b101;
   let Inst{8}     = 0;          // Single precision
+
+  // Loads & stores operate on both NEON and VFP pipelines.
+  let D = VFPNeonDomain;
 }
 
 // VFP Load / store multiple pseudo instructions.
@@ -1468,19 +1333,36 @@ class AXDI4<dag oops, dag iops, IndexMode im, InstrItinClass itin,
             string asm, string cstr, list<dag> pattern>
   : VFPXI<oops, iops, AddrMode4, Size4Bytes, im,
           VFPLdStMulFrm, itin, asm, cstr, pattern> {
+  // Instruction operands.
+  bits<4>  Rn;
+  bits<13> regs;
+
+  // Encode instruction operands.
+  let Inst{19-16} = Rn;
+  let Inst{22}    = regs{12};
+  let Inst{15-12} = regs{11-8};
+  let Inst{7-0}   = regs{7-0};
+
   // TODO: Mark the instructions with the appropriate subtarget info.
   let Inst{27-25} = 0b110;
   let Inst{11-9}  = 0b101;
   let Inst{8}     = 1;          // Double precision
-
-  // 64-bit loads & stores operate on both NEON and VFP pipelines.
-  let D = VFPNeonDomain;
 }
 
 class AXSI4<dag oops, dag iops, IndexMode im, InstrItinClass itin,
             string asm, string cstr, list<dag> pattern>
   : VFPXI<oops, iops, AddrMode4, Size4Bytes, im,
           VFPLdStMulFrm, itin, asm, cstr, pattern> {
+  // Instruction operands.
+  bits<4> Rn;
+  bits<13> regs;
+
+  // Encode instruction operands.
+  let Inst{19-16} = Rn;
+  let Inst{22}    = regs{8};
+  let Inst{15-12} = regs{12-9};
+  let Inst{7-0}   = regs{7-0};
+
   // TODO: Mark the instructions with the appropriate subtarget info.
   let Inst{27-25} = 0b110;
   let Inst{11-9}  = 0b101;
@@ -1492,6 +1374,16 @@ class ADuI<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3, bits<2> opcod4,
            bit opcod5, dag oops, dag iops, InstrItinClass itin, string opc,
            string asm, list<dag> pattern>
   : VFPAI<oops, iops, VFPUnaryFrm, itin, opc, asm, pattern> {
+  // Instruction operands.
+  bits<5> Dd;
+  bits<5> Dm;
+
+  // Encode instruction operands.
+  let Inst{3-0}   = Dm{3-0};
+  let Inst{5}     = Dm{4};
+  let Inst{15-12} = Dd{3-0};
+  let Inst{22}    = Dd{4};
+
   let Inst{27-23} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{19-16} = opcod3;
@@ -1506,26 +1398,25 @@ class ADbI<bits<5> opcod1, bits<2> opcod2, bit op6, bit op4, dag oops,
            dag iops, InstrItinClass itin, string opc, string asm,
            list<dag> pattern>
   : VFPAI<oops, iops, VFPBinaryFrm, itin, opc, asm, pattern> {
-  let Inst{27-23} = opcod1;
-  let Inst{21-20} = opcod2;
-  let Inst{11-9}  = 0b101;
-  let Inst{8}     = 1;          // Double precision
-  let Inst{6}     = op6;
-  let Inst{4}     = op4;
-}
+  // Instruction operands.
+  bits<5> Dd;
+  bits<5> Dn;
+  bits<5> Dm;
+
+  // Encode instruction operands.
+  let Inst{3-0}   = Dm{3-0};
+  let Inst{5}     = Dm{4};
+  let Inst{19-16} = Dn{3-0};
+  let Inst{7}     = Dn{4};
+  let Inst{15-12} = Dd{3-0};
+  let Inst{22}    = Dd{4};
 
-// Double precision, binary, VML[AS] (for additional predicate)
-class ADbI_vmlX<bits<5> opcod1, bits<2> opcod2, bit op6, bit op4, dag oops,
-           dag iops, InstrItinClass itin, string opc, string asm,
-           list<dag> pattern>
-  : VFPAI<oops, iops, VFPBinaryFrm, itin, opc, asm, pattern> {
   let Inst{27-23} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{11-9}  = 0b101;
   let Inst{8}     = 1;          // Double precision
   let Inst{6}     = op6;
   let Inst{4}     = op4;
-  list<Predicate> Predicates = [HasVFP2, UseVMLx];
 }
 
 // Single precision, unary
@@ -1533,6 +1424,16 @@ class ASuI<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3, bits<2> opcod4,
            bit opcod5, dag oops, dag iops, InstrItinClass itin, string opc,
            string asm, list<dag> pattern>
   : VFPAI<oops, iops, VFPUnaryFrm, itin, opc, asm, pattern> {
+  // Instruction operands.
+  bits<5> Sd;
+  bits<5> Sm;
+
+  // Encode instruction operands.
+  let Inst{3-0}   = Sm{4-1};
+  let Inst{5}     = Sm{0};
+  let Inst{15-12} = Sd{4-1};
+  let Inst{22}    = Sd{0};
+
   let Inst{27-23} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{19-16} = opcod3;
@@ -1542,8 +1443,8 @@ class ASuI<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3, bits<2> opcod4,
   let Inst{4}     = opcod5;
 }
 
-// Single precision unary, if no NEON
-// Same as ASuI except not available if NEON is enabled
+// Single precision unary, if no NEON. Same as ASuI except not available if
+// NEON is enabled.
 class ASuIn<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3, bits<2> opcod4,
             bit opcod5, dag oops, dag iops, InstrItinClass itin, string opc,
             string asm, list<dag> pattern>
@@ -1556,6 +1457,19 @@ class ASuIn<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3, bits<2> opcod4,
 class ASbI<bits<5> opcod1, bits<2> opcod2, bit op6, bit op4, dag oops, dag iops,
            InstrItinClass itin, string opc, string asm, list<dag> pattern>
   : VFPAI<oops, iops, VFPBinaryFrm, itin, opc, asm, pattern> {
+  // Instruction operands.
+  bits<5> Sd;
+  bits<5> Sn;
+  bits<5> Sm;
+
+  // Encode instruction operands.
+  let Inst{3-0}   = Sm{4-1};
+  let Inst{5}     = Sm{0};
+  let Inst{19-16} = Sn{4-1};
+  let Inst{7}     = Sn{0};
+  let Inst{15-12} = Sd{4-1};
+  let Inst{22}    = Sd{0};
+
   let Inst{27-23} = opcod1;
   let Inst{21-20} = opcod2;
   let Inst{11-9}  = 0b101;
@@ -1564,13 +1478,26 @@ class ASbI<bits<5> opcod1, bits<2> opcod2, bit op6, bit op4, dag oops, dag iops,
   let Inst{4}     = op4;
 }
 
-// Single precision binary, if no NEON
-// Same as ASbI except not available if NEON is enabled
+// Single precision binary, if no NEON. Same as ASbI except not available if
+// NEON is enabled.
 class ASbIn<bits<5> opcod1, bits<2> opcod2, bit op6, bit op4, dag oops,
             dag iops, InstrItinClass itin, string opc, string asm,
             list<dag> pattern>
   : ASbI<opcod1, opcod2, op6, op4, oops, iops, itin, opc, asm, pattern> {
   list<Predicate> Predicates = [HasVFP2,DontUseNEONForFP];
+
+  // Instruction operands.
+  bits<5> Sd;
+  bits<5> Sn;
+  bits<5> Sm;
+
+  // Encode instruction operands.
+  let Inst{3-0}   = Sm{4-1};
+  let Inst{5}     = Sm{0};
+  let Inst{19-16} = Sn{4-1};
+  let Inst{7}     = Sn{0};
+  let Inst{15-12} = Sd{4-1};
+  let Inst{22}    = Sd{0};
 }
 
 // VFP conversion instructions
@@ -1668,6 +1595,25 @@ class NLdSt<bit op23, bits<2> op21_20, bits<4> op11_8, bits<4> op7_4,
   let Inst{21-20} = op21_20;
   let Inst{11-8}  = op11_8;
   let Inst{7-4}   = op7_4;
+
+  let PostEncoderMethod = "NEONThumb2LoadStorePostEncoder";
+
+  bits<5> Vd;
+  bits<6> Rn;
+  bits<4> Rm;
+
+  let Inst{22}    = Vd{4};
+  let Inst{15-12} = Vd{3-0};
+  let Inst{19-16} = Rn{3-0};
+  let Inst{3-0}   = Rm{3-0};
+}
+
+class NLdStLn<bit op23, bits<2> op21_20, bits<4> op11_8, bits<4> op7_4,
+            dag oops, dag iops, InstrItinClass itin,
+            string opc, string dt, string asm, string cstr, list<dag> pattern>
+  : NLdSt<op23, op21_20, op11_8, op7_4, oops, iops, itin, opc,
+          dt, asm, cstr, pattern> {
+  bits<3> lane;
 }
 
 class PseudoNLdSt<dag oops, dag iops, InstrItinClass itin, string cstr>
@@ -1693,6 +1639,7 @@ class NDataI<dag oops, dag iops, Format f, InstrItinClass itin,
   : NeonI<oops, iops, AddrModeNone, IndexModeNone, f, itin, opc, dt, asm, cstr,
           pattern> {
   let Inst{31-25} = 0b1111001;
+  let PostEncoderMethod = "NEONThumb2DataIPostEncoder";
 }
 
 class NDataXI<dag oops, dag iops, Format f, InstrItinClass itin,
@@ -1700,6 +1647,7 @@ class NDataXI<dag oops, dag iops, Format f, InstrItinClass itin,
   : NeonXI<oops, iops, AddrModeNone, IndexModeNone, f, itin, opc, asm,
            cstr, pattern> {
   let Inst{31-25} = 0b1111001;
+  let PostEncoderMethod = "NEONThumb2DataIPostEncoder";
 }
 
 // NEON "one register and a modified immediate" format.
@@ -1716,11 +1664,11 @@ class N1ModImm<bit op23, bits<3> op21_19, bits<4> op11_8, bit op7, bit op6,
   let Inst{6}     = op6;
   let Inst{5}     = op5;
   let Inst{4}     = op4;
-  
+
   // Instruction operands.
   bits<5> Vd;
   bits<13> SIMM;
-  
+
   let Inst{15-12} = Vd{3-0};
   let Inst{22}    = Vd{4};
   let Inst{24}    = SIMM{7};
@@ -1741,7 +1689,7 @@ class N2V<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18, bits<2> op17_16,
   let Inst{11-7}  = op11_7;
   let Inst{6}     = op6;
   let Inst{4}     = op4;
-  
+
   // Instruction operands.
   bits<5> Vd;
   bits<5> Vm;
@@ -1765,7 +1713,7 @@ class N2VX<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18, bits<2> op17_16,
   let Inst{11-7}  = op11_7;
   let Inst{6}     = op6;
   let Inst{4}     = op4;
-  
+
   // Instruction operands.
   bits<5> Vd;
   bits<5> Vm;
@@ -1787,7 +1735,7 @@ class N2VImm<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
   let Inst{7}    = op7;
   let Inst{6}    = op6;
   let Inst{4}    = op4;
-  
+
   // Instruction operands.
   bits<5> Vd;
   bits<5> Vm;
@@ -1801,7 +1749,8 @@ class N2VImm<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
 }
 
 // NEON 3 vector register format.
-class N3V<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
+
+class N3VCommon<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
           dag oops, dag iops, Format f, InstrItinClass itin,
           string opc, string dt, string asm, string cstr, list<dag> pattern>
   : NDataI<oops, iops, f, itin, opc, dt, asm, cstr, pattern> {
@@ -1811,7 +1760,14 @@ class N3V<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
   let Inst{11-8}  = op11_8;
   let Inst{6}     = op6;
   let Inst{4}     = op4;
-  
+}
+
+class N3V<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
+          dag oops, dag iops, Format f, InstrItinClass itin,
+          string opc, string dt, string asm, string cstr, list<dag> pattern>
+  : N3VCommon<op24, op23, op21_20, op11_8, op6, op4,
+              oops, iops, f, itin, opc, dt, asm, cstr, pattern> {
+
   // Instruction operands.
   bits<5> Vd;
   bits<5> Vn;
@@ -1825,6 +1781,47 @@ class N3V<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
   let Inst{5}     = Vm{4};
 }
 
+class N3VLane32<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
+          dag oops, dag iops, Format f, InstrItinClass itin,
+          string opc, string dt, string asm, string cstr, list<dag> pattern>
+  : N3VCommon<op24, op23, op21_20, op11_8, op6, op4,
+              oops, iops, f, itin, opc, dt, asm, cstr, pattern> {
+
+  // Instruction operands.
+  bits<5> Vd;
+  bits<5> Vn;
+  bits<5> Vm;
+  bit lane;
+
+  let Inst{15-12} = Vd{3-0};
+  let Inst{22}    = Vd{4};
+  let Inst{19-16} = Vn{3-0};
+  let Inst{7}     = Vn{4};
+  let Inst{3-0}   = Vm{3-0};
+  let Inst{5}     = lane;
+}
+
+class N3VLane16<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6, bit op4,
+          dag oops, dag iops, Format f, InstrItinClass itin,
+          string opc, string dt, string asm, string cstr, list<dag> pattern>
+  : N3VCommon<op24, op23, op21_20, op11_8, op6, op4,
+              oops, iops, f, itin, opc, dt, asm, cstr, pattern> {
+
+  // Instruction operands.
+  bits<5> Vd;
+  bits<5> Vn;
+  bits<5> Vm;
+  bits<2> lane;
+
+  let Inst{15-12} = Vd{3-0};
+  let Inst{22}    = Vd{4};
+  let Inst{19-16} = Vn{3-0};
+  let Inst{7}     = Vn{4};
+  let Inst{2-0}   = Vm{2-0};
+  let Inst{5}     = lane{1};
+  let Inst{3}     = lane{0};
+}
+
 // Same as N3V except it doesn't have a data type suffix.
 class N3VX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6,
            bit op4,
@@ -1837,7 +1834,7 @@ class N3VX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op6,
   let Inst{11-8}  = op11_8;
   let Inst{6}     = op6;
   let Inst{4}     = op4;
-  
+
   // Instruction operands.
   bits<5> Vd;
   bits<5> Vn;
@@ -1861,18 +1858,22 @@ class NVLaneOp<bits<8> opcod1, bits<4> opcod2, bits<2> opcod3,
   let Inst{11-8}  = opcod2;
   let Inst{6-5}   = opcod3;
   let Inst{4}     = 1;
+  // A8.6.303, A8.6.328, A8.6.329
+  let Inst{3-0}   = 0b0000;
 
   let OutOperandList = oops;
   let InOperandList = !con(iops, (ins pred:$p));
   let AsmString = !strconcat(opc, "${p}", ".", dt, "\t", asm);
   let Pattern = pattern;
   list<Predicate> Predicates = [HasNEON];
-  
+
+  let PostEncoderMethod = "NEONThumb2DupPostEncoder";
+
   bits<5> V;
   bits<4> R;
   bits<4> p;
   bits<4> lane;
-  
+
   let Inst{31-28} = p{3-0};
   let Inst{7}     = V{4};
   let Inst{19-16} = V{3-0};
@@ -1905,11 +1906,11 @@ class NVDupLane<bits<4> op19_16, bit op6, dag oops, dag iops,
   let Inst{11-7}  = 0b11000;
   let Inst{6}     = op6;
   let Inst{4}     = 0;
-  
+
   bits<5> Vd;
   bits<5> Vm;
   bits<4> lane;
-  
+
   let Inst{22}     = Vd{4};
   let Inst{15-12} = Vd{3-0};
   let Inst{5}     = Vm{4};