Reland of Reland of "ARM64: Add NEON support"

This reverts commit c5aad5f2
The CL was reverted due to missing Chromium dependencies.

This commit removes the simulator trace-based tests, and the associated header file dependencies, previously pulled in by DEPS. The NEON support now has only hand-written tests, in test-assembler-arm64.cc. The remaining tests can be added in a later patch.

BUG=chromium:718439

Original issue's description:
> Reland "ARM64: Add NEON support"
>
> This reverts commit cc047635.
> The CL was reverted due to a missing DEPS mirror.
>
> Original issue's description:
> > ARM64: Add NEON support
> >
> > Add assembler, disassembler and simulator support for NEON in the ARM64 backend.
> >
> > BUG=
> >
> > Review-Url: https://codereview.chromium.org/2622643005
> > Cr-Commit-Position: refs/heads/master@{#44306}
>
> BUG=
>
> Review-Url: https://codereview.chromium.org/2812573003
> Cr-Commit-Position: refs/heads/master@{#44652}

Review-Url: https://codereview.chromium.org/2896303003
Cr-Commit-Position: refs/heads/master@{#45633}

BUILD.gn

@@ -2205,6 +2205,7 @@ v8_source_set("v8_base") {
       "src/arm64/macro-assembler-arm64.h",
       "src/arm64/simulator-arm64.cc",
       "src/arm64/simulator-arm64.h",
+      "src/arm64/simulator-logic-arm64.cc",
       "src/arm64/utils-arm64.cc",
       "src/arm64/utils-arm64.h",
       "src/compiler/arm64/code-generator-arm64.cc",

src/arm64/assembler-arm64-inl.h

@@ -57,6 +57,15 @@ inline int CPURegister::SizeInBytes() const {
   return reg_size / 8;
 }
 
+inline bool CPURegister::Is8Bits() const {
+  DCHECK(IsValid());
+  return reg_size == 8;
+}
+
+inline bool CPURegister::Is16Bits() const {
+  DCHECK(IsValid());
+  return reg_size == 16;
+}
 
 inline bool CPURegister::Is32Bits() const {
   DCHECK(IsValid());
@@ -69,9 +78,13 @@ inline bool CPURegister::Is64Bits() const {
   return reg_size == 64;
 }
 
+inline bool CPURegister::Is128Bits() const {
+  DCHECK(IsValid());
+  return reg_size == 128;
+}
 
 inline bool CPURegister::IsValid() const {
-  if (IsValidRegister() || IsValidFPRegister()) {
+  if (IsValidRegister() || IsValidVRegister()) {
     DCHECK(!IsNone());
     return true;
   } else {
@@ -87,14 +100,14 @@ inline bool CPURegister::IsValidRegister() const {
          ((reg_code < kNumberOfRegisters) || (reg_code == kSPRegInternalCode));
 }
 
-
-inline bool CPURegister::IsValidFPRegister() const {
-  return IsFPRegister() &&
-         ((reg_size == kSRegSizeInBits) || (reg_size == kDRegSizeInBits)) &&
-         (reg_code < kNumberOfFPRegisters);
+inline bool CPURegister::IsValidVRegister() const {
+  return IsVRegister() &&
+         ((reg_size == kBRegSizeInBits) || (reg_size == kHRegSizeInBits) ||
+          (reg_size == kSRegSizeInBits) || (reg_size == kDRegSizeInBits) ||
+          (reg_size == kQRegSizeInBits)) &&
+         (reg_code < kNumberOfVRegisters);
 }
 
-
 inline bool CPURegister::IsNone() const {
   // kNoRegister types should always have size 0 and code 0.
   DCHECK((reg_type != kNoRegister) || (reg_code == 0));
@@ -120,11 +133,7 @@ inline bool CPURegister::IsRegister() const {
   return reg_type == kRegister;
 }
 
-
-inline bool CPURegister::IsFPRegister() const {
-  return reg_type == kFPRegister;
-}
-
+inline bool CPURegister::IsVRegister() const { return reg_type == kVRegister; }
 
 inline bool CPURegister::IsSameSizeAndType(const CPURegister& other) const {
   return (reg_size == other.reg_size) && (reg_type == other.reg_type);
@@ -200,7 +209,7 @@ inline Register Register::XRegFromCode(unsigned code) {
   if (code == kSPRegInternalCode) {
     return csp;
   } else {
-    DCHECK(code < kNumberOfRegisters);
+    DCHECK_LT(code, static_cast<unsigned>(kNumberOfRegisters));
     return Register::Create(code, kXRegSizeInBits);
   }
 }
@@ -210,23 +219,40 @@ inline Register Register::WRegFromCode(unsigned code) {
   if (code == kSPRegInternalCode) {
     return wcsp;
   } else {
-    DCHECK(code < kNumberOfRegisters);
+    DCHECK_LT(code, static_cast<unsigned>(kNumberOfRegisters));
     return Register::Create(code, kWRegSizeInBits);
   }
 }
 
+inline VRegister VRegister::BRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kBRegSizeInBits);
+}
+
+inline VRegister VRegister::HRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kHRegSizeInBits);
+}
 
-inline FPRegister FPRegister::SRegFromCode(unsigned code) {
-  DCHECK(code < kNumberOfFPRegisters);
-  return FPRegister::Create(code, kSRegSizeInBits);
+inline VRegister VRegister::SRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kSRegSizeInBits);
 }
 
+inline VRegister VRegister::DRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kDRegSizeInBits);
+}
 
-inline FPRegister FPRegister::DRegFromCode(unsigned code) {
-  DCHECK(code < kNumberOfFPRegisters);
-  return FPRegister::Create(code, kDRegSizeInBits);
+inline VRegister VRegister::QRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kQRegSizeInBits);
 }
 
+inline VRegister VRegister::VRegFromCode(unsigned code) {
+  DCHECK_LT(code, static_cast<unsigned>(kNumberOfVRegisters));
+  return VRegister::Create(code, kVRegSizeInBits);
+}
 
 inline Register CPURegister::W() const {
   DCHECK(IsValidRegister());
@@ -239,16 +265,34 @@ inline Register CPURegister::X() const {
   return Register::XRegFromCode(reg_code);
 }
 
+inline VRegister CPURegister::V() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::VRegFromCode(reg_code);
+}
+
+inline VRegister CPURegister::B() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::BRegFromCode(reg_code);
+}
+
+inline VRegister CPURegister::H() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::HRegFromCode(reg_code);
+}
 
-inline FPRegister CPURegister::S() const {
-  DCHECK(IsValidFPRegister());
-  return FPRegister::SRegFromCode(reg_code);
+inline VRegister CPURegister::S() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::SRegFromCode(reg_code);
 }
 
+inline VRegister CPURegister::D() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::DRegFromCode(reg_code);
+}
 
-inline FPRegister CPURegister::D() const {
-  DCHECK(IsValidFPRegister());
-  return FPRegister::DRegFromCode(reg_code);
+inline VRegister CPURegister::Q() const {
+  DCHECK(IsValidVRegister());
+  return VRegister::QRegFromCode(reg_code);
 }
 
 
@@ -491,7 +535,7 @@ MemOperand::MemOperand(Register base, const Operand& offset, AddrMode addrmode)
 
     regoffset_ = NoReg;
   } else if (offset.IsShiftedRegister()) {
-    DCHECK(addrmode == Offset);
+    DCHECK((addrmode == Offset) || (addrmode == PostIndex));
 
     regoffset_ = offset.reg();
     shift_ = offset.shift();
@@ -877,21 +921,20 @@ LoadStoreOp Assembler::LoadOpFor(const CPURegister& rt) {
   if (rt.IsRegister()) {
     return rt.Is64Bits() ? LDR_x : LDR_w;
   } else {
-    DCHECK(rt.IsFPRegister());
-    return rt.Is64Bits() ? LDR_d : LDR_s;
-  }
-}
-
-
-LoadStorePairOp Assembler::LoadPairOpFor(const CPURegister& rt,
-                                         const CPURegister& rt2) {
-  DCHECK(AreSameSizeAndType(rt, rt2));
-  USE(rt2);
-  if (rt.IsRegister()) {
-    return rt.Is64Bits() ? LDP_x : LDP_w;
-  } else {
-    DCHECK(rt.IsFPRegister());
-    return rt.Is64Bits() ? LDP_d : LDP_s;
+    DCHECK(rt.IsVRegister());
+    switch (rt.SizeInBits()) {
+      case kBRegSizeInBits:
+        return LDR_b;
+      case kHRegSizeInBits:
+        return LDR_h;
+      case kSRegSizeInBits:
+        return LDR_s;
+      case kDRegSizeInBits:
+        return LDR_d;
+      default:
+        DCHECK(rt.IsQ());
+        return LDR_q;
+    }
   }
 }
 
@@ -901,11 +944,29 @@ LoadStoreOp Assembler::StoreOpFor(const CPURegister& rt) {
   if (rt.IsRegister()) {
     return rt.Is64Bits() ? STR_x : STR_w;
   } else {
-    DCHECK(rt.IsFPRegister());
-    return rt.Is64Bits() ? STR_d : STR_s;
+    DCHECK(rt.IsVRegister());
+    switch (rt.SizeInBits()) {
+      case kBRegSizeInBits:
+        return STR_b;
+      case kHRegSizeInBits:
+        return STR_h;
+      case kSRegSizeInBits:
+        return STR_s;
+      case kDRegSizeInBits:
+        return STR_d;
+      default:
+        DCHECK(rt.IsQ());
+        return STR_q;
+    }
   }
 }
 
+LoadStorePairOp Assembler::LoadPairOpFor(const CPURegister& rt,
+                                         const CPURegister& rt2) {
+  DCHECK_EQ(STP_w | LoadStorePairLBit, LDP_w);
+  return static_cast<LoadStorePairOp>(StorePairOpFor(rt, rt2) |
+                                      LoadStorePairLBit);
+}
 
 LoadStorePairOp Assembler::StorePairOpFor(const CPURegister& rt,
                                           const CPURegister& rt2) {
@@ -914,8 +975,16 @@ LoadStorePairOp Assembler::StorePairOpFor(const CPURegister& rt,
   if (rt.IsRegister()) {
     return rt.Is64Bits() ? STP_x : STP_w;
   } else {
-    DCHECK(rt.IsFPRegister());
-    return rt.Is64Bits() ? STP_d : STP_s;
+    DCHECK(rt.IsVRegister());
+    switch (rt.SizeInBits()) {
+      case kSRegSizeInBits:
+        return STP_s;
+      case kDRegSizeInBits:
+        return STP_d;
+      default:
+        DCHECK(rt.IsQ());
+        return STP_q;
+    }
   }
 }
 
@@ -924,7 +993,7 @@ LoadLiteralOp Assembler::LoadLiteralOpFor(const CPURegister& rt) {
   if (rt.IsRegister()) {
     return rt.Is64Bits() ? LDR_x_lit : LDR_w_lit;
   } else {
-    DCHECK(rt.IsFPRegister());
+    DCHECK(rt.IsVRegister());
     return rt.Is64Bits() ? LDR_d_lit : LDR_s_lit;
   }
 }
@@ -1107,9 +1176,8 @@ Instr Assembler::ImmLS(int imm9) {
   return truncate_to_int9(imm9) << ImmLS_offset;
 }
 
-
-Instr Assembler::ImmLSPair(int imm7, LSDataSize size) {
-  DCHECK(((imm7 >> size) << size) == imm7);
+Instr Assembler::ImmLSPair(int imm7, unsigned size) {
+  DCHECK_EQ((imm7 >> size) << size, imm7);
   int scaled_imm7 = imm7 >> size;
   DCHECK(is_int7(scaled_imm7));
   return truncate_to_int7(scaled_imm7) << ImmLSPair_offset;
@@ -1151,10 +1219,17 @@ Instr Assembler::ImmBarrierType(int imm2) {
   return imm2 << ImmBarrierType_offset;
 }
 
-
-LSDataSize Assembler::CalcLSDataSize(LoadStoreOp op) {
-  DCHECK((SizeLS_offset + SizeLS_width) == (kInstructionSize * 8));
-  return static_cast<LSDataSize>(op >> SizeLS_offset);
+unsigned Assembler::CalcLSDataSize(LoadStoreOp op) {
+  DCHECK((LSSize_offset + LSSize_width) == (kInstructionSize * 8));
+  unsigned size = static_cast<Instr>(op >> LSSize_offset);
+  if ((op & LSVector_mask) != 0) {
+    // Vector register memory operations encode the access size in the "size"
+    // and "opc" fields.
+    if ((size == 0) && ((op & LSOpc_mask) >> LSOpc_offset) >= 2) {
+      size = kQRegSizeLog2;
+    }
+  }
+  return size;
 }
 
 
@@ -1169,11 +1244,7 @@ Instr Assembler::ShiftMoveWide(int shift) {
   return shift << ShiftMoveWide_offset;
 }
 
-
-Instr Assembler::FPType(FPRegister fd) {
-  return fd.Is64Bits() ? FP64 : FP32;
-}
-
+Instr Assembler::FPType(VRegister fd) { return fd.Is64Bits() ? FP64 : FP32; }
 
 Instr Assembler::FPScale(unsigned scale) {
   DCHECK(is_uint6(scale));

src/arm64/code-stubs-arm64.cc

@@ -147,8 +147,8 @@ void DoubleToIStub::Generate(MacroAssembler* masm) {
 // See call site for description.
 static void EmitIdenticalObjectComparison(MacroAssembler* masm, Register left,
                                           Register right, Register scratch,
-                                          FPRegister double_scratch,
-                                          Label* slow, Condition cond) {
+                                          VRegister double_scratch, Label* slow,
+                                          Condition cond) {
   DCHECK(!AreAliased(left, right, scratch));
   Label not_identical, return_equal, heap_number;
   Register result = x0;
@@ -292,12 +292,9 @@ static void EmitStrictTwoHeapObjectCompare(MacroAssembler* masm,
 
 
 // See call site for description.
-static void EmitSmiNonsmiComparison(MacroAssembler* masm,
-                                    Register left,
-                                    Register right,
-                                    FPRegister left_d,
-                                    FPRegister right_d,
-                                    Label* slow,
+static void EmitSmiNonsmiComparison(MacroAssembler* masm, Register left,
+                                    Register right, VRegister left_d,
+                                    VRegister right_d, Label* slow,
                                     bool strict) {
   DCHECK(!AreAliased(left_d, right_d));
   DCHECK((left.is(x0) && right.is(x1)) ||
@@ -476,8 +473,8 @@ void CompareICStub::GenerateGeneric(MacroAssembler* masm) {
   // In case 3, we have found out that we were dealing with a number-number
   // comparison. The double values of the numbers have been loaded, right into
   // rhs_d, left into lhs_d.
-  FPRegister rhs_d = d0;
-  FPRegister lhs_d = d1;
+  VRegister rhs_d = d0;
+  VRegister lhs_d = d1;
   EmitSmiNonsmiComparison(masm, lhs, rhs, lhs_d, rhs_d, &slow, strict());
 
   __ Bind(&both_loaded_as_doubles);
@@ -613,7 +610,7 @@ void CompareICStub::GenerateGeneric(MacroAssembler* masm) {
 
 void StoreBufferOverflowStub::Generate(MacroAssembler* masm) {
   CPURegList saved_regs = kCallerSaved;
-  CPURegList saved_fp_regs = kCallerSavedFP;
+  CPURegList saved_fp_regs = kCallerSavedV;
 
   // We don't allow a GC during a store buffer overflow so there is no need to
   // store the registers in any particular way, but we do have to store and
@@ -686,12 +683,12 @@ void MathPowStub::Generate(MacroAssembler* masm) {
   Register exponent_integer = MathPowIntegerDescriptor::exponent();
   DCHECK(exponent_integer.is(x12));
   Register saved_lr = x19;
-  FPRegister result_double = d0;
-  FPRegister base_double = d0;
-  FPRegister exponent_double = d1;
-  FPRegister base_double_copy = d2;
-  FPRegister scratch1_double = d6;
-  FPRegister scratch0_double = d7;
+  VRegister result_double = d0;
+  VRegister base_double = d0;
+  VRegister exponent_double = d1;
+  VRegister base_double_copy = d2;
+  VRegister scratch1_double = d6;
+  VRegister scratch0_double = d7;
 
   // A fast-path for integer exponents.
   Label exponent_is_smi, exponent_is_integer;
@@ -1582,8 +1579,8 @@ void CompareICStub::GenerateNumbers(MacroAssembler* masm) {
   Register result = x0;
   Register rhs = x0;
   Register lhs = x1;
-  FPRegister rhs_d = d0;
-  FPRegister lhs_d = d1;
+  VRegister rhs_d = d0;
+  VRegister lhs_d = d1;
 
   if (left() == CompareICState::SMI) {
     __ JumpIfNotSmi(lhs, &miss);
@@ -2042,7 +2039,7 @@ RecordWriteStub::RegisterAllocation::RegisterAllocation(Register object,
       address_(address),
       scratch0_(scratch),
       saved_regs_(kCallerSaved),
-      saved_fp_regs_(kCallerSavedFP) {
+      saved_fp_regs_(kCallerSavedV) {
   DCHECK(!AreAliased(scratch, object, address));
 
   // The SaveCallerSaveRegisters method needs to save caller-saved

src/arm64/decoder-arm64-inl.h

@@ -213,6 +213,11 @@ void Decoder<V>::DecodeLoadStore(Instruction* instr) {
          (instr->Bits(27, 24) == 0xC) ||
          (instr->Bits(27, 24) == 0xD) );
 
+  if ((instr->Bit(28) == 0) && (instr->Bit(29) == 0) && (instr->Bit(26) == 1)) {
+    DecodeNEONLoadStore(instr);
+    return;
+  }
+
   if (instr->Bit(24) == 0) {
     if (instr->Bit(28) == 0) {
       if (instr->Bit(29) == 0) {
@@ -226,8 +231,6 @@ void Decoder<V>::DecodeLoadStore(Instruction* instr) {
           } else {
             V::VisitLoadStoreAcquireRelease(instr);
           }
-        } else {
-          DecodeAdvSIMDLoadStore(instr);
         }
       } else {
         if ((instr->Bits(31, 30) == 0x3) ||
@@ -513,16 +516,14 @@ void Decoder<V>::DecodeFP(Instruction* instr) {
          (instr->Bits(27, 24) == 0xF) );
 
   if (instr->Bit(28) == 0) {
-    DecodeAdvSIMDDataProcessing(instr);
+    DecodeNEONVectorDataProcessing(instr);
   } else {
-    if (instr->Bit(29) == 1) {
+    if (instr->Bits(31, 30) == 0x3) {
       V::VisitUnallocated(instr);
+    } else if (instr->Bits(31, 30) == 0x1) {
+      DecodeNEONScalarDataProcessing(instr);
     } else {
-      if (instr->Bits(31, 30) == 0x3) {
-        V::VisitUnallocated(instr);
-      } else if (instr->Bits(31, 30) == 0x1) {
-        DecodeAdvSIMDDataProcessing(instr);
-      } else {
+      if (instr->Bit(29) == 0) {
         if (instr->Bit(24) == 0) {
           if (instr->Bit(21) == 0) {
             if ((instr->Bit(23) == 1) ||
@@ -629,25 +630,190 @@ void Decoder<V>::DecodeFP(Instruction* instr) {
             V::VisitFPDataProcessing3Source(instr);
           }
         }
+      } else {
+        V::VisitUnallocated(instr);
       }
     }
   }
 }
 
-
-template<typename V>
-void Decoder<V>::DecodeAdvSIMDLoadStore(Instruction* instr) {
-  // TODO(all): Implement Advanced SIMD load/store instruction decode.
+template <typename V>
+void Decoder<V>::DecodeNEONLoadStore(Instruction* instr) {
   DCHECK(instr->Bits(29, 25) == 0x6);
-  V::VisitUnimplemented(instr);
+  if (instr->Bit(31) == 0) {
+    if ((instr->Bit(24) == 0) && (instr->Bit(21) == 1)) {
+      V::VisitUnallocated(instr);
+      return;
+    }
+
+    if (instr->Bit(23) == 0) {
+      if (instr->Bits(20, 16) == 0) {
+        if (instr->Bit(24) == 0) {
+          V::VisitNEONLoadStoreMultiStruct(instr);
+        } else {
+          V::VisitNEONLoadStoreSingleStruct(instr);
+        }
+      } else {
+        V::VisitUnallocated(instr);
+      }
+    } else {
+      if (instr->Bit(24) == 0) {
+        V::VisitNEONLoadStoreMultiStructPostIndex(instr);
+      } else {
+        V::VisitNEONLoadStoreSingleStructPostIndex(instr);
+      }
+    }
+  } else {
+    V::VisitUnallocated(instr);
+  }
 }
 
+template <typename V>
+void Decoder<V>::DecodeNEONVectorDataProcessing(Instruction* instr) {
+  DCHECK(instr->Bits(28, 25) == 0x7);
+  if (instr->Bit(31) == 0) {
+    if (instr->Bit(24) == 0) {
+      if (instr->Bit(21) == 0) {
+        if (instr->Bit(15) == 0) {
+          if (instr->Bit(10) == 0) {
+            if (instr->Bit(29) == 0) {
+              if (instr->Bit(11) == 0) {
+                V::VisitNEONTable(instr);
+              } else {
+                V::VisitNEONPerm(instr);
+              }
+            } else {
+              V::VisitNEONExtract(instr);
+            }
+          } else {
+            if (instr->Bits(23, 22) == 0) {
+              V::VisitNEONCopy(instr);
+            } else {
+              V::VisitUnallocated(instr);
+            }
+          }
+        } else {
+          V::VisitUnallocated(instr);
+        }
+      } else {
+        if (instr->Bit(10) == 0) {
+          if (instr->Bit(11) == 0) {
+            V::VisitNEON3Different(instr);
+          } else {
+            if (instr->Bits(18, 17) == 0) {
+              if (instr->Bit(20) == 0) {
+                if (instr->Bit(19) == 0) {
+                  V::VisitNEON2RegMisc(instr);
+                } else {
+                  if (instr->Bits(30, 29) == 0x2) {
+                    V::VisitUnallocated(instr);
+                  } else {
+                    V::VisitUnallocated(instr);
+                  }
+                }
+              } else {
+                if (instr->Bit(19) == 0) {
+                  V::VisitNEONAcrossLanes(instr);
+                } else {
+                  V::VisitUnallocated(instr);
+                }
+              }
+            } else {
+              V::VisitUnallocated(instr);
+            }
+          }
+        } else {
+          V::VisitNEON3Same(instr);
+        }
+      }
+    } else {
+      if (instr->Bit(10) == 0) {
+        V::VisitNEONByIndexedElement(instr);
+      } else {
+        if (instr->Bit(23) == 0) {
+          if (instr->Bits(22, 19) == 0) {
+            V::VisitNEONModifiedImmediate(instr);
+          } else {
+            V::VisitNEONShiftImmediate(instr);
+          }
+        } else {
+          V::VisitUnallocated(instr);
+        }
+      }
+    }
+  } else {
+    V::VisitUnallocated(instr);
+  }
+}
 
-template<typename V>
-void Decoder<V>::DecodeAdvSIMDDataProcessing(Instruction* instr) {
-  // TODO(all): Implement Advanced SIMD data processing instruction decode.
-  DCHECK(instr->Bits(27, 25) == 0x7);
-  V::VisitUnimplemented(instr);
+template <typename V>
+void Decoder<V>::DecodeNEONScalarDataProcessing(Instruction* instr) {
+  DCHECK(instr->Bits(28, 25) == 0xF);
+  if (instr->Bit(24) == 0) {
+    if (instr->Bit(21) == 0) {
+      if (instr->Bit(15) == 0) {
+        if (instr->Bit(10) == 0) {
+          if (instr->Bit(29) == 0) {
+            if (instr->Bit(11) == 0) {
+              V::VisitUnallocated(instr);
+            } else {
+              V::VisitUnallocated(instr);
+            }
+          } else {
+            V::VisitUnallocated(instr);
+          }
+        } else {
+          if (instr->Bits(23, 22) == 0) {
+            V::VisitNEONScalarCopy(instr);
+          } else {
+            V::VisitUnallocated(instr);
+          }
+        }
+      } else {
+        V::VisitUnallocated(instr);
+      }
+    } else {
+      if (instr->Bit(10) == 0) {
+        if (instr->Bit(11) == 0) {
+          V::VisitNEONScalar3Diff(instr);
+        } else {
+          if (instr->Bits(18, 17) == 0) {
+            if (instr->Bit(20) == 0) {
+              if (instr->Bit(19) == 0) {
+                V::VisitNEONScalar2RegMisc(instr);
+              } else {
+                if (instr->Bit(29) == 0) {
+                  V::VisitUnallocated(instr);
+                } else {
+                  V::VisitUnallocated(instr);
+                }
+              }
+            } else {
+              if (instr->Bit(19) == 0) {
+                V::VisitNEONScalarPairwise(instr);
+              } else {
+                V::VisitUnallocated(instr);
+              }
+            }
+          } else {
+            V::VisitUnallocated(instr);
+          }
+        }
+      } else {
+        V::VisitNEONScalar3Same(instr);
+      }
+    }
+  } else {
+    if (instr->Bit(10) == 0) {
+      V::VisitNEONScalarByIndexedElement(instr);
+    } else {
+      if (instr->Bit(23) == 0) {
+        V::VisitNEONScalarShiftImmediate(instr);
+      } else {
+        V::VisitUnallocated(instr);
+      }
+    }
+  }
 }
 
 

src/arm64/decoder-arm64.h

@@ -16,50 +16,72 @@ namespace internal {
 
 // List macro containing all visitors needed by the decoder class.
 
-#define VISITOR_LIST(V)            \
-  V(PCRelAddressing)               \
-  V(AddSubImmediate)               \
-  V(LogicalImmediate)              \
-  V(MoveWideImmediate)             \
-  V(Bitfield)                      \
-  V(Extract)                       \
-  V(UnconditionalBranch)           \
-  V(UnconditionalBranchToRegister) \
-  V(CompareBranch)                 \
-  V(TestBranch)                    \
-  V(ConditionalBranch)             \
-  V(System)                        \
-  V(Exception)                     \
-  V(LoadStorePairPostIndex)        \
-  V(LoadStorePairOffset)           \
-  V(LoadStorePairPreIndex)         \
-  V(LoadLiteral)                   \
-  V(LoadStoreUnscaledOffset)       \
-  V(LoadStorePostIndex)            \
-  V(LoadStorePreIndex)             \
-  V(LoadStoreRegisterOffset)       \
-  V(LoadStoreUnsignedOffset)       \
-  V(LoadStoreAcquireRelease)       \
-  V(LogicalShifted)                \
-  V(AddSubShifted)                 \
-  V(AddSubExtended)                \
-  V(AddSubWithCarry)               \
-  V(ConditionalCompareRegister)    \
-  V(ConditionalCompareImmediate)   \
-  V(ConditionalSelect)             \
-  V(DataProcessing1Source)         \
-  V(DataProcessing2Source)         \
-  V(DataProcessing3Source)         \
-  V(FPCompare)                     \
-  V(FPConditionalCompare)          \
-  V(FPConditionalSelect)           \
-  V(FPImmediate)                   \
-  V(FPDataProcessing1Source)       \
-  V(FPDataProcessing2Source)       \
-  V(FPDataProcessing3Source)       \
-  V(FPIntegerConvert)              \
-  V(FPFixedPointConvert)           \
-  V(Unallocated)                   \
+#define VISITOR_LIST(V)                 \
+  V(PCRelAddressing)                    \
+  V(AddSubImmediate)                    \
+  V(LogicalImmediate)                   \
+  V(MoveWideImmediate)                  \
+  V(Bitfield)                           \
+  V(Extract)                            \
+  V(UnconditionalBranch)                \
+  V(UnconditionalBranchToRegister)      \
+  V(CompareBranch)                      \
+  V(TestBranch)                         \
+  V(ConditionalBranch)                  \
+  V(System)                             \
+  V(Exception)                          \
+  V(LoadStorePairPostIndex)             \
+  V(LoadStorePairOffset)                \
+  V(LoadStorePairPreIndex)              \
+  V(LoadLiteral)                        \
+  V(LoadStoreUnscaledOffset)            \
+  V(LoadStorePostIndex)                 \
+  V(LoadStorePreIndex)                  \
+  V(LoadStoreRegisterOffset)            \
+  V(LoadStoreUnsignedOffset)            \
+  V(LoadStoreAcquireRelease)            \
+  V(LogicalShifted)                     \
+  V(AddSubShifted)                      \
+  V(AddSubExtended)                     \
+  V(AddSubWithCarry)                    \
+  V(ConditionalCompareRegister)         \
+  V(ConditionalCompareImmediate)        \
+  V(ConditionalSelect)                  \
+  V(DataProcessing1Source)              \
+  V(DataProcessing2Source)              \
+  V(DataProcessing3Source)              \
+  V(FPCompare)                          \
+  V(FPConditionalCompare)               \
+  V(FPConditionalSelect)                \
+  V(FPImmediate)                        \
+  V(FPDataProcessing1Source)            \
+  V(FPDataProcessing2Source)            \
+  V(FPDataProcessing3Source)            \
+  V(FPIntegerConvert)                   \
+  V(FPFixedPointConvert)                \
+  V(NEON2RegMisc)                       \
+  V(NEON3Different)                     \
+  V(NEON3Same)                          \
+  V(NEONAcrossLanes)                    \
+  V(NEONByIndexedElement)               \
+  V(NEONCopy)                           \
+  V(NEONExtract)                        \
+  V(NEONLoadStoreMultiStruct)           \
+  V(NEONLoadStoreMultiStructPostIndex)  \
+  V(NEONLoadStoreSingleStruct)          \
+  V(NEONLoadStoreSingleStructPostIndex) \
+  V(NEONModifiedImmediate)              \
+  V(NEONScalar2RegMisc)                 \
+  V(NEONScalar3Diff)                    \
+  V(NEONScalar3Same)                    \
+  V(NEONScalarByIndexedElement)         \
+  V(NEONScalarCopy)                     \
+  V(NEONScalarPairwise)                 \
+  V(NEONScalarShiftImmediate)           \
+  V(NEONShiftImmediate)                 \
+  V(NEONTable)                          \
+  V(NEONPerm)                           \
+  V(Unallocated)                        \
   V(Unimplemented)
 
 // The Visitor interface. Disassembler and simulator (and other tools)
@@ -109,6 +131,8 @@ class DispatchingDecoderVisitor : public DecoderVisitor {
   // stored by the decoder.
   void RemoveVisitor(DecoderVisitor* visitor);
 
+  void VisitNEONShiftImmediate(const Instruction* instr);
+
   #define DECLARE(A) void Visit##A(Instruction* instr);
   VISITOR_LIST(DECLARE)
   #undef DECLARE
@@ -173,12 +197,17 @@ class Decoder : public V {
   // Decode the Advanced SIMD (NEON) load/store part of the instruction tree,
   // and call the corresponding visitors.
   // On entry, instruction bits 29:25 = 0x6.
-  void DecodeAdvSIMDLoadStore(Instruction* instr);
+  void DecodeNEONLoadStore(Instruction* instr);
 
   // Decode the Advanced SIMD (NEON) data processing part of the instruction
   // tree, and call the corresponding visitors.
   // On entry, instruction bits 27:25 = 0x7.
-  void DecodeAdvSIMDDataProcessing(Instruction* instr);
+  void DecodeNEONVectorDataProcessing(Instruction* instr);
+
+  // Decode the Advanced SIMD (NEON) scalar data processing part of the
+  // instruction tree, and call the corresponding visitors.
+  // On entry, instruction bits 28:25 = 0xF.
+  void DecodeNEONScalarDataProcessing(Instruction* instr);
 };
 
 

src/arm64/deoptimizer-arm64.cc

@@ -118,13 +118,13 @@ void Deoptimizer::TableEntryGenerator::Generate() {
 
   // Save all allocatable double registers.
   CPURegList saved_double_registers(
-      CPURegister::kFPRegister, kDRegSizeInBits,
+      CPURegister::kVRegister, kDRegSizeInBits,
       RegisterConfiguration::Crankshaft()->allocatable_double_codes_mask());
   __ PushCPURegList(saved_double_registers);
 
   // Save all allocatable float registers.
   CPURegList saved_float_registers(
-      CPURegister::kFPRegister, kSRegSizeInBits,
+      CPURegister::kVRegister, kSRegSizeInBits,
       RegisterConfiguration::Crankshaft()->allocatable_float_codes_mask());
   __ PushCPURegList(saved_float_registers);
 

src/arm64/disasm-arm64.h

@@ -5,6 +5,7 @@
 #ifndef V8_ARM64_DISASM_ARM64_H
 #define V8_ARM64_DISASM_ARM64_H
 
+#include "src/arm64/assembler-arm64.h"
 #include "src/arm64/decoder-arm64.h"
 #include "src/arm64/instructions-arm64.h"
 #include "src/globals.h"
@@ -29,6 +30,13 @@ class DisassemblingDecoder : public DecoderVisitor {
  protected:
   virtual void ProcessOutput(Instruction* instr);
 
+  // Default output functions.  The functions below implement a default way of
+  // printing elements in the disassembly. A sub-class can override these to
+  // customize the disassembly output.
+
+  // Prints the name of a register.
+  virtual void AppendRegisterNameToOutput(const CPURegister& reg);
+
   void Format(Instruction* instr, const char* mnemonic, const char* format);
   void Substitute(Instruction* instr, const char* string);
   int SubstituteField(Instruction* instr, const char* format);

src/compiler/arm64/instruction-selector-arm64.cc

@@ -103,13 +103,13 @@ class Arm64OperandGenerator final : public OperandGenerator {
       case kArithmeticImm:
         return Assembler::IsImmAddSub(value);
       case kLoadStoreImm8:
-        return IsLoadStoreImmediate(value, LSByte);
+        return IsLoadStoreImmediate(value, 0);
       case kLoadStoreImm16:
-        return IsLoadStoreImmediate(value, LSHalfword);
+        return IsLoadStoreImmediate(value, 1);
       case kLoadStoreImm32:
-        return IsLoadStoreImmediate(value, LSWord);
+        return IsLoadStoreImmediate(value, 2);
       case kLoadStoreImm64:
-        return IsLoadStoreImmediate(value, LSDoubleWord);
+        return IsLoadStoreImmediate(value, 3);
       case kNoImmediate:
         return false;
       case kShift32Imm:  // Fall through.
@@ -130,7 +130,7 @@ class Arm64OperandGenerator final : public OperandGenerator {
   }
 
  private:
-  bool IsLoadStoreImmediate(int64_t value, LSDataSize size) {
+  bool IsLoadStoreImmediate(int64_t value, unsigned size) {
     return Assembler::IsImmLSScaled(value, size) ||
            Assembler::IsImmLSUnscaled(value);
   }

src/crankshaft/arm64/delayed-masm-arm64-inl.h

@@ -44,14 +44,12 @@ void DelayedMasm::Mov(const Register& rd,
   __ Mov(rd, operand, discard_mode);
 }
 
-
-void DelayedMasm::Fmov(FPRegister fd, FPRegister fn) {
+void DelayedMasm::Fmov(VRegister fd, VRegister fn) {
   EmitPending();
   __ Fmov(fd, fn);
 }
 
-
-void DelayedMasm::Fmov(FPRegister fd, double imm) {
+void DelayedMasm::Fmov(VRegister fd, double imm) {
   EmitPending();
   __ Fmov(fd, imm);
 }

src/crankshaft/arm64/delayed-masm-arm64.h

@@ -61,8 +61,8 @@ class DelayedMasm BASE_EMBEDDED {
   inline void Mov(const Register& rd,
                   const Operand& operand,
                   DiscardMoveMode discard_mode = kDontDiscardForSameWReg);
-  inline void Fmov(FPRegister fd, FPRegister fn);
-  inline void Fmov(FPRegister fd, double imm);
+  inline void Fmov(VRegister fd, VRegister fn);
+  inline void Fmov(VRegister fd, double imm);
   inline void LoadObject(Register result, Handle<Object> object);
   // Instructions which try to merge which the pending instructions.
   void StackSlotMove(LOperand* src, LOperand* dst);

src/crankshaft/arm64/lithium-codegen-arm64.h

@@ -159,10 +159,9 @@ class LCodeGen: public LCodeGenBase {
                          const Register& value,
                          uint64_t mask);
 
-  template<class InstrType>
-  void EmitBranchIfNonZeroNumber(InstrType instr,
-                                 const FPRegister& value,
-                                 const FPRegister& scratch);
+  template <class InstrType>
+  void EmitBranchIfNonZeroNumber(InstrType instr, const VRegister& value,
+                                 const VRegister& scratch);
 
   template<class InstrType>
   void EmitBranchIfHeapNumber(InstrType instr,

src/crankshaft/arm64/lithium-gap-resolver-arm64.h

@@ -68,7 +68,7 @@ class LGapResolver BASE_EMBEDDED {
   // These two methods switch from one mode to the other.
   void AcquireSavedValueRegister() { masm_.AcquireScratchRegister(); }
   void ReleaseSavedValueRegister() { masm_.ReleaseScratchRegister(); }
-  const FPRegister& SavedFPValueRegister() {
+  const VRegister& SavedFPValueRegister() {
     // We use the Crankshaft floating-point scratch register to break a cycle
     // involving double values as the MacroAssembler will not need it for the
     // operations performed by the gap resolver.

src/v8.gyp

@@ -1595,6 +1595,7 @@
             'arm64/macro-assembler-arm64-inl.h',
             'arm64/simulator-arm64.cc',
             'arm64/simulator-arm64.h',
+            'arm64/simulator-logic-arm64.cc',
             'arm64/utils-arm64.cc',
             'arm64/utils-arm64.h',
             'arm64/eh-frame-arm64.cc',