Revert "ART: Reference.getReferent intrinsic for arm and arm64"

Reverting because GenerateCalleeMethodStaticOrDirectCall()
prevents replacing kDexCacheViaMethod with kRuntimeCall
where we would not retrieve the target method at all and
leave the runtime to retrieve and call it just like for
unresolved methods.

The intrinsic should be re-implemented by loading the
flags through HLoadClass.

Note that the intrinsic was unimplemented for CC and a bit
broken for non-CC, using LDR instead of LDRB for loading
the flags.

Test: Rely on TreeHugger.
Bug: 32535355
Bug: 30627598

This reverts commit d8c052ac0aa3382c4807add33afa32580ffeecbb.

Change-Id: I81fd14dac60c94ac543e336f4f3c888259fc8bd7

9 files changed, 8 insertions, 195 deletions

diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index c66bd77d6b..f231712243 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -8946,7 +8946,8 @@ Register CodeGeneratorARM::GetInvokeStaticOrDirectExtraParameter(HInvokeStaticOr
   // save one load. However, since this is just an intrinsic slow path we prefer this
   // simple and more robust approach rather that trying to determine if that's the case.
   SlowPathCode* slow_path = GetCurrentSlowPath();
-  if (slow_path != nullptr && slow_path->IsCoreRegisterSaved(location.AsRegister<Register>())) {
+  DCHECK(slow_path != nullptr);  // For intrinsified invokes the call is emitted on the slow path.
+  if (slow_path->IsCoreRegisterSaved(location.AsRegister<Register>())) {
     int stack_offset = slow_path->GetStackOffsetOfCoreRegister(location.AsRegister<Register>());
     __ LoadFromOffset(kLoadWord, temp, SP, stack_offset);
     return temp;
@@ -8954,8 +8955,7 @@ Register CodeGeneratorARM::GetInvokeStaticOrDirectExtraParameter(HInvokeStaticOr
   return location.AsRegister<Register>();
 }
 
-Location CodeGeneratorARM::GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
-                                                                  Location temp) {
+void CodeGeneratorARM::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
     case HInvokeStaticOrDirect::MethodLoadKind::kStringInit: {
@@ -9016,11 +9016,6 @@ Location CodeGeneratorARM::GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticO
       break;
     }
   }
-  return callee_method;
-}
-
-void CodeGeneratorARM::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
-  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
diff --git a/compiler/optimizing/code_generator_arm.h b/compiler/optimizing/code_generator_arm.h
index 2409a4d38d..5f37d3bff1 100644
--- a/compiler/optimizing/code_generator_arm.h
+++ b/compiler/optimizing/code_generator_arm.h
@@ -455,7 +455,6 @@ class CodeGeneratorARM : public CodeGenerator {
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
-  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 7d9c61b76c..c11beaca43 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -4497,8 +4497,7 @@ HInvokeStaticOrDirect::DispatchInfo CodeGeneratorARM64::GetSupportedInvokeStatic
   return desired_dispatch_info;
 }
 
-Location CodeGeneratorARM64::GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
-                                                                    Location temp) {
+void CodeGeneratorARM64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
   // Make sure that ArtMethod* is passed in kArtMethodRegister as per the calling convention.
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
@@ -4563,12 +4562,6 @@ Location CodeGeneratorARM64::GenerateCalleeMethodStaticOrDirectCall(HInvokeStati
       break;
     }
   }
-  return callee_method;
-}
-
-void CodeGeneratorARM64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
-  // All registers are assumed to be correctly set up.
-  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
diff --git a/compiler/optimizing/code_generator_arm64.h b/compiler/optimizing/code_generator_arm64.h
index 7a4b3d4805..747fc9f0b1 100644
--- a/compiler/optimizing/code_generator_arm64.h
+++ b/compiler/optimizing/code_generator_arm64.h
@@ -540,7 +540,6 @@ class CodeGeneratorARM64 : public CodeGenerator {
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
-  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 1f8e1efd5e..2d6f4f21cc 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -9119,7 +9119,7 @@ vixl32::Register CodeGeneratorARMVIXL::GetInvokeStaticOrDirectExtraParameter(
   return RegisterFrom(location);
 }
 
-Location CodeGeneratorARMVIXL::GenerateCalleeMethodStaticOrDirectCall(
+void CodeGeneratorARMVIXL::GenerateStaticOrDirectCall(
     HInvokeStaticOrDirect* invoke, Location temp) {
   Location callee_method = temp;  // For all kinds except kRecursive, callee will be in temp.
   switch (invoke->GetMethodLoadKind()) {
@@ -9177,12 +9177,6 @@ Location CodeGeneratorARMVIXL::GenerateCalleeMethodStaticOrDirectCall(
       break;
     }
   }
-  return callee_method;
-}
-
-void CodeGeneratorARMVIXL::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
-                                                      Location temp) {
-  Location callee_method = GenerateCalleeMethodStaticOrDirectCall(invoke, temp);
 
   switch (invoke->GetCodePtrLocation()) {
     case HInvokeStaticOrDirect::CodePtrLocation::kCallSelf:
diff --git a/compiler/optimizing/code_generator_arm_vixl.h b/compiler/optimizing/code_generator_arm_vixl.h
index ef809510ad..f6e4de33a8 100644
--- a/compiler/optimizing/code_generator_arm_vixl.h
+++ b/compiler/optimizing/code_generator_arm_vixl.h
@@ -538,7 +538,6 @@ class CodeGeneratorARMVIXL : public CodeGenerator {
       const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
       HInvokeStaticOrDirect* invoke) OVERRIDE;
 
-  Location GenerateCalleeMethodStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
   void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
   void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
 
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index 9803c9a0e9..8fc6f45729 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -2624,58 +2624,6 @@ void IntrinsicCodeGeneratorARM::VisitDoubleIsInfinite(HInvoke* invoke) {
   codegen_->GenerateConditionWithZero(kCondEQ, out, out);
 }
 
-void IntrinsicLocationsBuilderARM::VisitReferenceGetReferent(HInvoke* invoke) {
-  if (kEmitCompilerReadBarrier) {
-    // Do not intrinsify this call with the read barrier configuration.
-    return;
-  }
-  LocationSummary* locations = new (arena_) LocationSummary(invoke,
-                                                            LocationSummary::kCallOnSlowPath,
-                                                            kIntrinsified);
-  locations->SetInAt(0, Location::RequiresRegister());
-  locations->SetOut(Location::SameAsFirstInput());
-  locations->AddTemp(Location::RequiresRegister());
-}
-
-void IntrinsicCodeGeneratorARM::VisitReferenceGetReferent(HInvoke* invoke) {
-  DCHECK(!kEmitCompilerReadBarrier);
-  ArmAssembler* const assembler = GetAssembler();
-  LocationSummary* locations = invoke->GetLocations();
-
-  Register obj = locations->InAt(0).AsRegister<Register>();
-  Register out = locations->Out().AsRegister<Register>();
-
-  SlowPathCode* slow_path = new (GetAllocator()) IntrinsicSlowPathARM(invoke);
-  codegen_->AddSlowPath(slow_path);
-
-  // Load ArtMethod first.
-  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
-  DCHECK(invoke_direct != nullptr);
-  Register temp = codegen_->GenerateCalleeMethodStaticOrDirectCall(
-      invoke_direct, locations->GetTemp(0)).AsRegister<Register>();
-
-  // Now get declaring class.
-  __ ldr(temp, Address(temp, ArtMethod::DeclaringClassOffset().Int32Value()));
-
-  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
-  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
-  DCHECK_NE(slow_path_flag_offset, 0u);
-  DCHECK_NE(disable_flag_offset, 0u);
-  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
-
-  // Check static flags that prevent using intrinsic.
-  __ ldr(IP, Address(temp, disable_flag_offset));
-  __ ldr(temp, Address(temp, slow_path_flag_offset));
-  __ orr(IP, IP, ShifterOperand(temp));
-  __ CompareAndBranchIfNonZero(IP, slow_path->GetEntryLabel());
-
-  // Fast path.
-  __ ldr(out, Address(obj, mirror::Reference::ReferentOffset().Int32Value()));
-  codegen_->MaybeRecordImplicitNullCheck(invoke);
-  __ MaybeUnpoisonHeapReference(out);
-  __ Bind(slow_path->GetExitLabel());
-}
-
 void IntrinsicLocationsBuilderARM::VisitIntegerValueOf(HInvoke* invoke) {
   InvokeRuntimeCallingConvention calling_convention;
   IntrinsicVisitor::ComputeIntegerValueOfLocations(
@@ -2782,6 +2730,7 @@ UNIMPLEMENTED_INTRINSIC(ARM, MathRoundDouble)   // Could be done by changing rou
 UNIMPLEMENTED_INTRINSIC(ARM, MathRoundFloat)    // Could be done by changing rounding mode, maybe?
 UNIMPLEMENTED_INTRINSIC(ARM, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARM, SystemArrayCopyChar)
+UNIMPLEMENTED_INTRINSIC(ARM, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(ARM, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index b511c5a18d..93fba08028 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -2897,69 +2897,6 @@ void IntrinsicCodeGeneratorARM64::VisitDoubleIsInfinite(HInvoke* invoke) {
   GenIsInfinite(invoke->GetLocations(), /* is64bit */ true, GetVIXLAssembler());
 }
 
-void IntrinsicLocationsBuilderARM64::VisitReferenceGetReferent(HInvoke* invoke) {
-  if (kEmitCompilerReadBarrier) {
-    // Do not intrinsify this call with the read barrier configuration.
-    return;
-  }
-  LocationSummary* locations = new (arena_) LocationSummary(invoke,
-                                                            LocationSummary::kCallOnSlowPath,
-                                                            kIntrinsified);
-  locations->SetInAt(0, Location::RequiresRegister());
-  locations->SetOut(Location::SameAsFirstInput());
-  locations->AddTemp(Location::RequiresRegister());
-}
-
-void IntrinsicCodeGeneratorARM64::VisitReferenceGetReferent(HInvoke* invoke) {
-  DCHECK(!kEmitCompilerReadBarrier);
-  MacroAssembler* masm = GetVIXLAssembler();
-  LocationSummary* locations = invoke->GetLocations();
-
-  Register obj = InputRegisterAt(invoke, 0);
-  Register out = OutputRegister(invoke);
-
-  SlowPathCodeARM64* slow_path = new (GetAllocator()) IntrinsicSlowPathARM64(invoke);
-  codegen_->AddSlowPath(slow_path);
-
-  // Load ArtMethod first.
-  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
-  DCHECK(invoke_direct != nullptr);
-  Register temp0 = XRegisterFrom(codegen_->GenerateCalleeMethodStaticOrDirectCall(
-                                 invoke_direct, locations->GetTemp(0)));
-
-  // Now get declaring class.
-  __ Ldr(temp0.W(), MemOperand(temp0, ArtMethod::DeclaringClassOffset().Int32Value()));
-
-  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
-  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
-  DCHECK_NE(slow_path_flag_offset, 0u);
-  DCHECK_NE(disable_flag_offset, 0u);
-  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
-
-  // Check static flags that prevent using intrinsic.
-  if (slow_path_flag_offset == disable_flag_offset + 1) {
-    // Load two adjacent flags in one 64-bit load.
-    __ Ldr(temp0, MemOperand(temp0, disable_flag_offset));
-  } else {
-    UseScratchRegisterScope temps(masm);
-    Register temp1 = temps.AcquireW();
-    __ Ldr(temp1.W(), MemOperand(temp0, disable_flag_offset));
-    __ Ldr(temp0.W(), MemOperand(temp0, slow_path_flag_offset));
-    __ Orr(temp0, temp1, temp0);
-  }
-  __ Cbnz(temp0, slow_path->GetEntryLabel());
-
-  {
-    // Ensure that between load and MaybeRecordImplicitNullCheck there are no pools emitted.
-    vixl::EmissionCheckScope guard(codegen_->GetVIXLAssembler(), kMaxMacroInstructionSizeInBytes);
-    // Fast path.
-    __ Ldr(out, HeapOperand(obj, mirror::Reference::ReferentOffset().Int32Value()));
-    codegen_->MaybeRecordImplicitNullCheck(invoke);
-  }
-  codegen_->GetAssembler()->MaybeUnpoisonHeapReference(out);
-  __ Bind(slow_path->GetExitLabel());
-}
-
 void IntrinsicLocationsBuilderARM64::VisitIntegerValueOf(HInvoke* invoke) {
   InvokeRuntimeCallingConvention calling_convention;
   IntrinsicVisitor::ComputeIntegerValueOfLocations(
@@ -3055,6 +2992,7 @@ void IntrinsicCodeGeneratorARM64::VisitThreadInterrupted(HInvoke* invoke) {
   __ Bind(&done);
 }
 
+UNIMPLEMENTED_INTRINSIC(ARM64, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARM64, IntegerLowestOneBit)
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 1a33b0ee01..3c0f2f78e8 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -3000,60 +3000,6 @@ void IntrinsicCodeGeneratorARMVIXL::VisitDoubleIsInfinite(HInvoke* invoke) {
   codegen_->GenerateConditionWithZero(kCondEQ, out, out);
 }
 
-void IntrinsicLocationsBuilderARMVIXL::VisitReferenceGetReferent(HInvoke* invoke) {
-  if (kEmitCompilerReadBarrier) {
-    // Do not intrinsify this call with the read barrier configuration.
-    return;
-  }
-  LocationSummary* locations = new (arena_) LocationSummary(invoke,
-                                                            LocationSummary::kCallOnSlowPath,
-                                                            kIntrinsified);
-  locations->SetInAt(0, Location::RequiresRegister());
-  locations->SetOut(Location::SameAsFirstInput());
-  locations->AddTemp(Location::RequiresRegister());
-}
-
-void IntrinsicCodeGeneratorARMVIXL::VisitReferenceGetReferent(HInvoke* invoke) {
-  DCHECK(!kEmitCompilerReadBarrier);
-  ArmVIXLAssembler* assembler = GetAssembler();
-  LocationSummary* locations = invoke->GetLocations();
-
-  vixl32::Register obj = InputRegisterAt(invoke, 0);
-  vixl32::Register out = OutputRegister(invoke);
-
-  SlowPathCodeARMVIXL* slow_path = new (GetAllocator()) IntrinsicSlowPathARMVIXL(invoke);
-  codegen_->AddSlowPath(slow_path);
-
-  // Load ArtMethod first.
-  HInvokeStaticOrDirect* invoke_direct = invoke->AsInvokeStaticOrDirect();
-  DCHECK(invoke_direct != nullptr);
-  vixl32::Register temp0 = RegisterFrom(codegen_->GenerateCalleeMethodStaticOrDirectCall(
-      invoke_direct, locations->GetTemp(0)));
-
-  // Now get declaring class.
-  __ Ldr(temp0, MemOperand(temp0, ArtMethod::DeclaringClassOffset().Int32Value()));
-
-  uint32_t slow_path_flag_offset = codegen_->GetReferenceSlowFlagOffset();
-  uint32_t disable_flag_offset = codegen_->GetReferenceDisableFlagOffset();
-  DCHECK_NE(slow_path_flag_offset, 0u);
-  DCHECK_NE(disable_flag_offset, 0u);
-  DCHECK_NE(slow_path_flag_offset, disable_flag_offset);
-
-  // Check static flags that prevent using intrinsic.
-  UseScratchRegisterScope temps(assembler->GetVIXLAssembler());
-  vixl32::Register temp1 = temps.Acquire();
-  __ Ldr(temp1, MemOperand(temp0, disable_flag_offset));
-  __ Ldr(temp0, MemOperand(temp0, slow_path_flag_offset));
-  __ Orr(temp0, temp1, temp0);
-  __ CompareAndBranchIfNonZero(temp0, slow_path->GetEntryLabel());
-
-  // Fast path.
-  __ Ldr(out, MemOperand(obj, mirror::Reference::ReferentOffset().Int32Value()));
-  codegen_->MaybeRecordImplicitNullCheck(invoke);
-  assembler->MaybeUnpoisonHeapReference(out);
-  __ Bind(slow_path->GetExitLabel());
-}
-
 void IntrinsicLocationsBuilderARMVIXL::VisitMathCeil(HInvoke* invoke) {
   if (features_.HasARMv8AInstructions()) {
     CreateFPToFPLocations(arena_, invoke);
@@ -3178,6 +3124,7 @@ void IntrinsicCodeGeneratorARMVIXL::VisitThreadInterrupted(HInvoke* invoke) {
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathRoundDouble)   // Could be done by changing rounding mode, maybe?
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeCASLong)     // High register pressure.
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyChar)
+UNIMPLEMENTED_INTRINSIC(ARMVIXL, ReferenceGetReferent)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, IntegerHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, LongHighestOneBit)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, IntegerLowestOneBit)