Adds single-precision and SIMD load/store to arm64.

runtime/vm/flow_graph_compiler_arm64.cc

 // Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_ARM64.
 #if defined(TARGET_ARCH_ARM64)
 
 #include "vm/flow_graph_compiler.h"
 
 #include "vm/ast_printer.h"
@@ skipping 237 matching lines @@
   const intptr_t from_index = num_type_args - num_type_params;
   const TypeArguments& type_arguments =
       TypeArguments::ZoneHandle(type.arguments());
   const bool is_raw_type = type_arguments.IsNull() ||
       type_arguments.IsRaw(from_index, num_type_params);
   // Signature class is an instantiated parameterized type.
   if (!type_class.IsSignatureClass()) {
     if (is_raw_type) {
       const Register kClassIdReg = R2;
       // dynamic type argument, check only classes.
-      __ LoadClassId(kClassIdReg, kInstanceReg);
+      __ LoadClassId(kClassIdReg, kInstanceReg, PP);
       __ CompareImmediate(kClassIdReg, type_class.id(), PP);
       __ b(is_instance_lbl, EQ);
       // List is a very common case.
       if (IsListClass(type_class)) {
         GenerateListTypeCheck(kClassIdReg, is_instance_lbl);
       }
       return GenerateSubtype1TestCacheLookup(
           token_pos, type_class, is_instance_lbl, is_not_instance_lbl);
     }
     // If one type argument only, check if type argument is Object or dynamic.
@@ skipping 60 matching lines @@
   if (smi_class.IsSubtypeOf(TypeArguments::Handle(),
                             type_class,
                             TypeArguments::Handle(),
                             NULL)) {
     __ b(is_instance_lbl, EQ);
   } else {
     __ b(is_not_instance_lbl, EQ);
   }
   // Compare if the classes are equal.
   const Register kClassIdReg = R2;
-  __ LoadClassId(kClassIdReg, kInstanceReg);
+  __ LoadClassId(kClassIdReg, kInstanceReg, PP);
   __ CompareImmediate(kClassIdReg, type_class.id(), PP);
   __ b(is_instance_lbl, EQ);
   // See ClassFinalizer::ResolveSuperTypeAndInterfaces for list of restricted
   // interfaces.
   // Bool interface can be implemented only by core class Bool.
   if (type.IsBoolType()) {
     __ CompareImmediate(kClassIdReg, kBoolCid, PP);
     __ b(is_instance_lbl, EQ);
     __ b(is_not_instance_lbl);
     return false;
   }
   if (type.IsFunctionType()) {
     // Check if instance is a closure.
-    __ LoadClassById(R3, kClassIdReg);
-    __ LoadFieldFromOffset(R3, R3, Class::signature_function_offset());
+    __ LoadClassById(R3, kClassIdReg, PP);
+    __ LoadFieldFromOffset(R3, R3, Class::signature_function_offset(), PP);
     __ CompareObject(R3, Object::null_object(), PP);
     __ b(is_instance_lbl, NE);
   }
   // Custom checking for numbers (Smi, Mint, Bigint and Double).
   // Note that instance is not Smi (checked above).
   if (type.IsSubtypeOf(Type::Handle(Type::Number()), NULL)) {
     GenerateNumberTypeCheck(
         kClassIdReg, type, is_instance_lbl, is_not_instance_lbl);
     return false;
   }
@@ skipping 13 matching lines @@
 // TODO(srdjan): Implement a quicker subtype check, as type test
 // arrays can grow too high, but they may be useful when optimizing
 // code (type-feedback).
 RawSubtypeTestCache* FlowGraphCompiler::GenerateSubtype1TestCacheLookup(
     intptr_t token_pos,
     const Class& type_class,
     Label* is_instance_lbl,
     Label* is_not_instance_lbl) {
   __ Comment("Subtype1TestCacheLookup");
   const Register kInstanceReg = R0;
-  __ LoadClass(R1, kInstanceReg);
+  __ LoadClass(R1, kInstanceReg, PP);
   // R1: instance class.
   // Check immediate superclass equality.
-  __ LoadFieldFromOffset(R2, R1, Class::super_type_offset());
-  __ LoadFieldFromOffset(R2, R2, Type::type_class_offset());
+  __ LoadFieldFromOffset(R2, R1, Class::super_type_offset(), PP);
+  __ LoadFieldFromOffset(R2, R2, Type::type_class_offset(), PP);
   __ CompareObject(R2, type_class, PP);
   __ b(is_instance_lbl, EQ);
 
   const Register kTypeArgumentsReg = kNoRegister;
   const Register kTempReg = kNoRegister;
   return GenerateCallSubtypeTestStub(kTestTypeOneArg,
                                      kInstanceReg,
                                      kTypeArgumentsReg,
                                      kTempReg,
                                      is_instance_lbl,
@@ skipping 13 matching lines @@
   // Skip check if destination is a dynamic type.
   if (type.IsTypeParameter()) {
     const TypeParameter& type_param = TypeParameter::Cast(type);
     // Load instantiator (or null) and instantiator type arguments on stack.
     __ ldr(R1, Address(SP));  // Get instantiator type arguments.
     // R1: instantiator type arguments.
     // Check if type arguments are null, i.e. equivalent to vector of dynamic.
     __ CompareObject(R1, Object::null_object(), PP);
     __ b(is_instance_lbl, EQ);
     __ LoadFieldFromOffset(
-        R2, R1, TypeArguments::type_at_offset(type_param.index()));
+        R2, R1, TypeArguments::type_at_offset(type_param.index()), PP);
     // R2: concrete type of type.
     // Check if type argument is dynamic.
     __ CompareObject(R2, Type::ZoneHandle(Type::DynamicType()), PP);
     __ b(is_instance_lbl, EQ);
     __ CompareObject(R2, Type::ZoneHandle(Type::ObjectType()), PP);
     __ b(is_instance_lbl, EQ);
 
     // For Smi check quickly against int and num interfaces.
     Label not_smi;
     __ tsti(R0, kSmiTagMask);  // Value is Smi?
@@ skipping 294 matching lines @@
       num_fixed_params + num_opt_pos_params + num_opt_named_params;
   ASSERT(function.NumParameters() == num_params);
   ASSERT(parsed_function().first_parameter_index() == kFirstLocalSlotFromFp);
 
   // Check that min_num_pos_args <= num_pos_args <= max_num_pos_args,
   // where num_pos_args is the number of positional arguments passed in.
   const int min_num_pos_args = num_fixed_params;
   const int max_num_pos_args = num_fixed_params + num_opt_pos_params;
 
   __ LoadFieldFromOffset(
-      R8, R4, ArgumentsDescriptor::positional_count_offset());
+      R8, R4, ArgumentsDescriptor::positional_count_offset(), PP);
   // Check that min_num_pos_args <= num_pos_args.
   Label wrong_num_arguments;
   __ CompareImmediate(R8, Smi::RawValue(min_num_pos_args), PP);
   __ b(&wrong_num_arguments, LT);
   // Check that num_pos_args <= max_num_pos_args.
   __ CompareImmediate(R8, Smi::RawValue(max_num_pos_args), PP);
   __ b(&wrong_num_arguments, GT);
 
   // Copy positional arguments.
   // Argument i passed at fp[kParamEndSlotFromFp + num_args - i] is copied
   // to fp[kFirstLocalSlotFromFp - i].
 
-  __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset());
+  __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset(), PP);
   // Since R7 and R8 are Smi, use LSL 2 instead of LSL 3.
   // Let R7 point to the last passed positional argument, i.e. to
   // fp[kParamEndSlotFromFp + num_args - (num_pos_args - 1)].
   __ sub(R7, R7, Operand(R8));
   __ add(R7, FP, Operand(R7, LSL, 2));
   __ add(R7, R7, Operand((kParamEndSlotFromFp + 1) * kWordSize));
 
   // Let R6 point to the last copied positional argument, i.e. to
   // fp[kFirstLocalSlotFromFp - (num_pos_args - 1)].
   __ AddImmediate(R6, FP, (kFirstLocalSlotFromFp + 1) * kWordSize, PP);
@@ skipping 33 matching lines @@
         const intptr_t result = opt_param_name.CompareTo(param_i->name());
         ASSERT(result != 0);
         if (result > 0) break;
         opt_param[i + 1] = opt_param[i];
         opt_param_position[i + 1] = opt_param_position[i];
       }
       opt_param[i + 1] = parameter;
       opt_param_position[i + 1] = pos;
     }
     // Generate code handling each optional parameter in alphabetical order.
-    __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset());
+    __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset(), PP);
     __ LoadFieldFromOffset(
-        R8, R4, ArgumentsDescriptor::positional_count_offset());
+        R8, R4, ArgumentsDescriptor::positional_count_offset(), PP);
     __ SmiUntag(R8);
     // Let R7 point to the first passed argument, i.e. to
     // fp[kParamEndSlotFromFp + num_args - 0]; num_args (R7) is Smi.
     __ add(R7, FP, Operand(R7, LSL, 2));
     __ AddImmediate(R7, R7, kParamEndSlotFromFp * kWordSize, PP);
     // Let R6 point to the entry of the first named argument.
     __ add(R6, R4, Operand(
         ArgumentsDescriptor::first_named_entry_offset() - kHeapObjectTag));
     for (int i = 0; i < num_opt_named_params; i++) {
       Label load_default_value, assign_optional_parameter;
       const int param_pos = opt_param_position[i];
       // Check if this named parameter was passed in.
       // Load R5 with the name of the argument.
-      __ LoadFromOffset(R5, R6, ArgumentsDescriptor::name_offset());
+      __ LoadFromOffset(R5, R6, ArgumentsDescriptor::name_offset(), PP);
       ASSERT(opt_param[i]->name().IsSymbol());
       __ CompareObject(R5, opt_param[i]->name(), PP);
       __ b(&load_default_value, NE);
       // Load R5 with passed-in argument at provided arg_pos, i.e. at
       // fp[kParamEndSlotFromFp + num_args - arg_pos].
-      __ LoadFromOffset(R5, R6, ArgumentsDescriptor::position_offset());
+      __ LoadFromOffset(R5, R6, ArgumentsDescriptor::position_offset(), PP);
       // R5 is arg_pos as Smi.
       // Point to next named entry.
       __ add(R6, R6, Operand(ArgumentsDescriptor::named_entry_size()));
       // Negate and untag R5 so we can use in scaled address mode.
       __ subs(R5, ZR, Operand(R5, ASR, 1));
       Address argument_addr(R7, R5, UXTX, Address::Scaled);  // R5 is untagged.
       __ ldr(R5, argument_addr);
       __ b(&assign_optional_parameter);
       __ Bind(&load_default_value);
       // Load R5 with default argument.
       const Object& value = Object::ZoneHandle(
           parsed_function().default_parameter_values().At(
               param_pos - num_fixed_params));
       __ LoadObject(R5, value, PP);
       __ Bind(&assign_optional_parameter);
       // Assign R5 to fp[kFirstLocalSlotFromFp - param_pos].
       // We do not use the final allocation index of the variable here, i.e.
       // scope->VariableAt(i)->index(), because captured variables still need
       // to be copied to the context that is not yet allocated.
       const intptr_t computed_param_pos = kFirstLocalSlotFromFp - param_pos;
-      __ StoreToOffset(R5, FP, computed_param_pos * kWordSize);
+      __ StoreToOffset(R5, FP, computed_param_pos * kWordSize, PP);
     }
     delete[] opt_param;
     delete[] opt_param_position;
     if (check_correct_named_args) {
       // Check that R6 now points to the null terminator in the arguments
       // descriptor.
       __ ldr(R5, Address(R6));
       __ CompareObject(R5, Object::null_object(), PP);
       __ b(&all_arguments_processed, EQ);
     }
   } else {
     ASSERT(num_opt_pos_params > 0);
     __ LoadFieldFromOffset(
-        R8, R4, ArgumentsDescriptor::positional_count_offset());
+        R8, R4, ArgumentsDescriptor::positional_count_offset(), PP);
     __ SmiUntag(R8);
     for (int i = 0; i < num_opt_pos_params; i++) {
       Label next_parameter;
       // Handle this optional positional parameter only if k or fewer positional
       // arguments have been passed, where k is param_pos, the position of this
       // optional parameter in the formal parameter list.
       const int param_pos = num_fixed_params + i;
       __ CompareImmediate(R8, param_pos, PP);
       __ b(&next_parameter, GT);
       // Load R5 with default argument.
       const Object& value = Object::ZoneHandle(
           parsed_function().default_parameter_values().At(i));
       __ LoadObject(R5, value, PP);
       // Assign R5 to fp[kFirstLocalSlotFromFp - param_pos].
       // We do not use the final allocation index of the variable here, i.e.
       // scope->VariableAt(i)->index(), because captured variables still need
       // to be copied to the context that is not yet allocated.
       const intptr_t computed_param_pos = kFirstLocalSlotFromFp - param_pos;
-      __ StoreToOffset(R5, FP, computed_param_pos * kWordSize);
+      __ StoreToOffset(R5, FP, computed_param_pos * kWordSize, PP);
       __ Bind(&next_parameter);
     }
     if (check_correct_named_args) {
-      __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset());
+      __ LoadFieldFromOffset(R7, R4, ArgumentsDescriptor::count_offset(), PP);
       __ SmiUntag(R7);
       // Check that R8 equals R7, i.e. no named arguments passed.
       __ CompareRegisters(R8, R7);
       __ b(&all_arguments_processed, EQ);
     }
   }
 
   __ Bind(&wrong_num_arguments);
   if (function.IsClosureFunction()) {
     // Invoke noSuchMethod function passing "call" as the original name.
@@ skipping 11 matching lines @@
   }
 
   __ Bind(&all_arguments_processed);
   // Nullify originally passed arguments only after they have been copied and
   // checked, otherwise noSuchMethod would not see their original values.
   // This step can be skipped in case we decide that formal parameters are
   // implicitly final, since garbage collecting the unmodified value is not
   // an issue anymore.
 
   // R4 : arguments descriptor array.
-  __ LoadFieldFromOffset(R8, R4, ArgumentsDescriptor::count_offset());
+  __ LoadFieldFromOffset(R8, R4, ArgumentsDescriptor::count_offset(), PP);
   __ SmiUntag(R8);
   __ add(R7, FP, Operand((kParamEndSlotFromFp + 1) * kWordSize));
   const Address original_argument_addr(R7, R8, UXTX, Address::Scaled);
   __ LoadObject(TMP, Object::null_object(), PP);
   Label null_args_loop, null_args_loop_condition;
   __ b(&null_args_loop_condition);
   __ Bind(&null_args_loop);
   __ str(TMP, original_argument_addr);
   __ Bind(&null_args_loop_condition);
   __ subs(R8, R8, Operand(1));
   __ b(&null_args_loop, PL);
 }
 
 
 void FlowGraphCompiler::GenerateInlinedGetter(intptr_t offset) {
   // LR: return address.
   // SP: receiver.
   // Sequence node has one return node, its input is load field node.
   __ Comment("Inlined Getter");
-  __ LoadFromOffset(R0, SP, 0 * kWordSize);
-  __ LoadFromOffset(R0, R0, offset - kHeapObjectTag);
+  __ LoadFromOffset(R0, SP, 0 * kWordSize, PP);
+  __ LoadFromOffset(R0, R0, offset - kHeapObjectTag, PP);
   __ ret();
 }
 
 
 void FlowGraphCompiler::GenerateInlinedSetter(intptr_t offset) {
   // LR: return address.
   // SP+1: receiver.
   // SP+0: value.
   // Sequence node has one store node and one return NULL node.
   __ Comment("Inlined Setter");
-  __ LoadFromOffset(R0, SP, 1 * kWordSize);  // Receiver.
-  __ LoadFromOffset(R1, SP, 0 * kWordSize);  // Value.
-  __ StoreIntoObject(R0, FieldAddress(R0, offset), R1);
+  __ LoadFromOffset(R0, SP, 1 * kWordSize, PP);  // Receiver.
+  __ LoadFromOffset(R1, SP, 0 * kWordSize, PP);  // Value.
+  __ StoreIntoObjectOffset(R0, offset, R1, PP);
   __ LoadObject(R0, Object::null_object(), PP);
   __ ret();
 }
 
 
 void FlowGraphCompiler::EmitFrameEntry() {
   const Function& function = parsed_function().function();
   Register new_pp = kNoPP;
   if (CanOptimizeFunction() &&
       function.IsOptimizable() &&
       (!is_optimizing() || may_reoptimize())) {
     const Register function_reg = R6;
     new_pp = R13;
 
     // Set up pool pointer in new_pp.
     __ LoadPoolPointer(new_pp);
 
     // Load function object using the callee's pool pointer.
     __ LoadObject(function_reg, function, new_pp);
 
     // Patch point is after the eventually inlined function object.
     AddCurrentDescriptor(PcDescriptors::kEntryPatch,
                          Isolate::kNoDeoptId,
                          0);  // No token position.
     intptr_t threshold = FLAG_optimization_counter_threshold;
-    __ LoadFieldFromOffset(R7, function_reg, Function::usage_counter_offset());
+    __ LoadFieldFromOffset(
+        R7, function_reg, Function::usage_counter_offset(), new_pp);
     if (is_optimizing()) {
       // Reoptimization of an optimized function is triggered by counting in
       // IC stubs, but not at the entry of the function.
       threshold = FLAG_reoptimization_counter_threshold;
     } else {
       __ add(R7, R7, Operand(1));
-      __ StoreFieldToOffset(R7, function_reg, Function::usage_counter_offset());
+      __ StoreFieldToOffset(
+          R7, function_reg, Function::usage_counter_offset(), new_pp);
     }
     __ CompareImmediate(R7, threshold, new_pp);
     ASSERT(function_reg == R6);
     Label dont_optimize;
     __ b(&dont_optimize, LT);
     __ Branch(&StubCode::OptimizeFunctionLabel(), new_pp);
     __ Bind(&dont_optimize);
   } else if (!flow_graph().IsCompiledForOsr()) {
     // We have to load the PP here too because a load of an external label
     // may be patched at the AddCurrentDescriptor below.
@@ skipping 49 matching lines @@
     ASSERT(!parsed_function().function().HasOptionalParameters());
     const bool check_arguments = !flow_graph().IsCompiledForOsr();
 #else
     const bool check_arguments =
         function.IsClosureFunction() && !flow_graph().IsCompiledForOsr();
 #endif
     if (check_arguments) {
       __ Comment("Check argument count");
       // Check that exactly num_fixed arguments are passed in.
       Label correct_num_arguments, wrong_num_arguments;
-      __ LoadFieldFromOffset(R0, R4, ArgumentsDescriptor::count_offset());
+      __ LoadFieldFromOffset(R0, R4, ArgumentsDescriptor::count_offset(), PP);
       __ CompareImmediate(R0, Smi::RawValue(num_fixed_params), PP);
       __ b(&wrong_num_arguments, NE);
       __ LoadFieldFromOffset(R1, R4,
-            ArgumentsDescriptor::positional_count_offset());
+            ArgumentsDescriptor::positional_count_offset(), PP);
       __ CompareRegisters(R0, R1);
       __ b(&correct_num_arguments, EQ);
       __ Bind(&wrong_num_arguments);
       if (function.IsClosureFunction()) {
         // Invoke noSuchMethod function passing the original function name.
         // For closure functions, use "call" as the original name.
         const String& name =
             String::Handle(function.IsClosureFunction()
                              ? Symbols::Call().raw()
                              : function.name());
@@ skipping 16 matching lines @@
   }
 
   // In unoptimized code, initialize (non-argument) stack allocated slots to
   // null.
   if (!is_optimizing() && (num_locals > 0)) {
     __ Comment("Initialize spill slots");
     const intptr_t slot_base = parsed_function().first_stack_local_index();
     __ LoadObject(R0, Object::null_object(), PP);
     for (intptr_t i = 0; i < num_locals; ++i) {
       // Subtract index i (locals lie at lower addresses than FP).
-      __ StoreToOffset(R0, FP, (slot_base - i) * kWordSize);
+      __ StoreToOffset(R0, FP, (slot_base - i) * kWordSize, PP);
     }
   }
 
   VisitBlocks();
 
   __ hlt(0);
   GenerateDeferredCode();
   // Emit function patching code. This will be swapped with the first 3
   // instructions at entry point.
   AddCurrentDescriptor(PcDescriptors::kPatchCode,
@@ skipping 67 matching lines @@
 void FlowGraphCompiler::EmitEdgeCounter() {
   // We do not check for overflow when incrementing the edge counter.  The
   // function should normally be optimized long before the counter can
   // overflow; and though we do not reset the counters when we optimize or
   // deoptimize, there is a bound on the number of
   // optimization/deoptimization cycles we will attempt.
   const Array& counter = Array::ZoneHandle(Array::New(1, Heap::kOld));
   counter.SetAt(0, Smi::Handle(Smi::New(0)));
   __ Comment("Edge counter");
   __ LoadObject(R0, counter, PP);
-  __ LoadFieldFromOffset(TMP, R0, Array::element_offset(0));
+  __ LoadFieldFromOffset(TMP, R0, Array::element_offset(0), PP);
   __ add(TMP, TMP, Operand(Smi::RawValue(1)));
-  __ StoreFieldToOffset(TMP, R0, Array::element_offset(0));
+  __ StoreFieldToOffset(TMP, R0, Array::element_offset(0), PP);
 }
 
 
 void FlowGraphCompiler::EmitOptimizedInstanceCall(
     ExternalLabel* target_label,
     const ICData& ic_data,
     intptr_t argument_count,
     intptr_t deopt_id,
     intptr_t token_pos,
     LocationSummary* locs) {
@@ skipping 40 matching lines @@
     intptr_t token_pos,
     LocationSummary* locs) {
   MegamorphicCacheTable* table = Isolate::Current()->megamorphic_cache_table();
   const String& name = String::Handle(ic_data.target_name());
   const Array& arguments_descriptor =
       Array::ZoneHandle(ic_data.arguments_descriptor());
   ASSERT(!arguments_descriptor.IsNull() && (arguments_descriptor.Length() > 0));
   const MegamorphicCache& cache =
       MegamorphicCache::ZoneHandle(table->Lookup(name, arguments_descriptor));
   Label not_smi, load_cache;
-  __ LoadFromOffset(R0, SP, (argument_count - 1) * kWordSize);
+  __ LoadFromOffset(R0, SP, (argument_count - 1) * kWordSize, PP);
   __ tsti(R0, kSmiTagMask);
   __ b(&not_smi, NE);
   __ LoadImmediate(R0, Smi::RawValue(kSmiCid), PP);
   __ b(&load_cache);
 
   __ Bind(&not_smi);
-  __ LoadClassId(R0, R0);
+  __ LoadClassId(R0, R0, PP);
   __ SmiTag(R0);
 
   // R0: class ID of the receiver (smi).
   __ Bind(&load_cache);
   __ LoadObject(R1, cache, PP);
-  __ LoadFieldFromOffset(R2, R1, MegamorphicCache::buckets_offset());
-  __ LoadFieldFromOffset(R1, R1, MegamorphicCache::mask_offset());
+  __ LoadFieldFromOffset(R2, R1, MegamorphicCache::buckets_offset(), PP);
+  __ LoadFieldFromOffset(R1, R1, MegamorphicCache::mask_offset(), PP);
   // R2: cache buckets array.
   // R1: mask.
   __ mov(R3, R0);
 
   Label loop, update, call_target_function;
   __ b(&loop);
 
   __ Bind(&update);
   __ add(R3, R3, Operand(Smi::RawValue(1)));
   __ Bind(&loop);
   __ and_(R3, R3, Operand(R1));
   const intptr_t base = Array::data_offset();
   // R3 is smi tagged, but table entries are 8 bytes, so LSL 2.
   __ add(TMP, R2, Operand(R3, LSL, 2));
-  __ LoadFieldFromOffset(R4, TMP, base);
+  __ LoadFieldFromOffset(R4, TMP, base, PP);
 
   ASSERT(kIllegalCid == 0);
   __ tst(R4, Operand(R4));
   __ b(&call_target_function, EQ);
   __ CompareRegisters(R4, R0);
   __ b(&update, NE);
 
   __ Bind(&call_target_function);
   // Call the target found in the cache.  For a class id match, this is a
   // proper target for the given name and arguments descriptor.  If the
   // illegal class id was found, the target is a cache miss handler that can
   // be invoked as a normal Dart function.
   __ add(TMP, R2, Operand(R3, LSL, 2));
-  __ LoadFieldFromOffset(R0, TMP, base + kWordSize);
-  __ LoadFieldFromOffset(R1, R0, Function::code_offset());
-  __ LoadFieldFromOffset(R1, R1, Code::instructions_offset());
+  __ LoadFieldFromOffset(R0, TMP, base + kWordSize, PP);
+  __ LoadFieldFromOffset(R1, R0, Function::code_offset(), PP);
+  __ LoadFieldFromOffset(R1, R1, Code::instructions_offset(), PP);
   __ LoadObject(R5, ic_data, PP);
   __ LoadObject(R4, arguments_descriptor, PP);
   __ AddImmediate(R1, R1, Instructions::HeaderSize() - kHeapObjectTag, PP);
   __ blr(R1);
   AddCurrentDescriptor(PcDescriptors::kOther, Isolate::kNoDeoptId, token_pos);
   RecordSafepoint(locs);
   AddDeoptIndexAtCall(Isolate::ToDeoptAfter(deopt_id), token_pos);
   __ Drop(argument_count);
 }
 
@@ skipping 263 matching lines @@
   MoveOperands* move = moves_[index];
   const Location source = move->src();
   const Location destination = move->dest();
 
   if (source.IsRegister()) {
     if (destination.IsRegister()) {
       __ mov(destination.reg(), source.reg());
     } else {
       ASSERT(destination.IsStackSlot());
       const intptr_t dest_offset = destination.ToStackSlotOffset();
-      __ StoreToOffset(source.reg(), FP, dest_offset);
+      __ StoreToOffset(source.reg(), FP, dest_offset, PP);
     }
   } else if (source.IsStackSlot()) {
     if (destination.IsRegister()) {
       const intptr_t source_offset = source.ToStackSlotOffset();
-      __ LoadFromOffset(destination.reg(), FP, source_offset);
+      __ LoadFromOffset(destination.reg(), FP, source_offset, PP);
     } else {
       ASSERT(destination.IsStackSlot());
       const intptr_t source_offset = source.ToStackSlotOffset();
       const intptr_t dest_offset = destination.ToStackSlotOffset();
-      __ LoadFromOffset(TMP, FP, source_offset);
-      __ StoreToOffset(TMP, FP, dest_offset);
+      __ LoadFromOffset(TMP, FP, source_offset, PP);
+      __ StoreToOffset(TMP, FP, dest_offset, PP);
     }
   } else if (source.IsFpuRegister()) {
     if (destination.IsFpuRegister()) {
       __ fmovdd(destination.fpu_reg(), source.fpu_reg());
     } else {
       if (destination.IsDoubleStackSlot()) {
         const intptr_t dest_offset = destination.ToStackSlotOffset();
         VRegister src = source.fpu_reg();
-        __ StoreDToOffset(src, FP, dest_offset);
+        __ StoreDToOffset(src, FP, dest_offset, PP);
       } else {
         ASSERT(destination.IsQuadStackSlot());
         UNIMPLEMENTED();
       }
     }
   } else if (source.IsDoubleStackSlot()) {
     if (destination.IsFpuRegister()) {
       const intptr_t dest_offset = source.ToStackSlotOffset();
       const VRegister dst = destination.fpu_reg();
-      __ LoadDFromOffset(dst, FP, dest_offset);
+      __ LoadDFromOffset(dst, FP, dest_offset, PP);
     } else {
       ASSERT(destination.IsDoubleStackSlot());
       const intptr_t source_offset = source.ToStackSlotOffset();
       const intptr_t dest_offset = destination.ToStackSlotOffset();
-      __ LoadDFromOffset(VTMP, FP, source_offset);
-      __ StoreDToOffset(VTMP, FP, dest_offset);
+      __ LoadDFromOffset(VTMP, FP, source_offset, PP);
+      __ StoreDToOffset(VTMP, FP, dest_offset, PP);
     }
   } else if (source.IsQuadStackSlot()) {
     UNIMPLEMENTED();
   } else {
     ASSERT(source.IsConstant());
     const Object& constant = source.constant();
     if (destination.IsRegister()) {
       __ LoadObject(destination.reg(), constant, PP);
     } else if (destination.IsFpuRegister()) {
       const VRegister dst = destination.fpu_reg();
       __ LoadObject(TMP, constant, PP);
-      __ LoadDFieldFromOffset(dst, TMP, Double::value_offset());
+      __ LoadDFieldFromOffset(dst, TMP, Double::value_offset(), PP);
     } else if (destination.IsDoubleStackSlot()) {
       const intptr_t dest_offset = destination.ToStackSlotOffset();
       __ LoadObject(TMP, constant, PP);
-      __ LoadDFieldFromOffset(VTMP, TMP, Double::value_offset());
-      __ StoreDToOffset(VTMP, FP, dest_offset);
+      __ LoadDFieldFromOffset(VTMP, TMP, Double::value_offset(), PP);
+      __ StoreDToOffset(VTMP, FP, dest_offset, PP);
     } else {
       ASSERT(destination.IsStackSlot());
       const intptr_t dest_offset = destination.ToStackSlotOffset();
       __ LoadObject(TMP, constant, PP);
-      __ StoreToOffset(TMP, FP, dest_offset);
+      __ StoreToOffset(TMP, FP, dest_offset, PP);
     }
   }
 
   move->Eliminate();
 }
 
 
 void ParallelMoveResolver::EmitSwap(int index) {
   MoveOperands* move = moves_[index];
   const Location source = move->src();
@@ skipping 24 matching lines @@
            source.IsQuadStackSlot());
     bool double_width = destination.IsDoubleStackSlot() ||
                         source.IsDoubleStackSlot();
     VRegister reg = source.IsFpuRegister() ? source.fpu_reg()
                                           : destination.fpu_reg();
     const intptr_t slot_offset = source.IsFpuRegister()
         ? destination.ToStackSlotOffset()
         : source.ToStackSlotOffset();
 
     if (double_width) {
-      __ LoadDFromOffset(VTMP, FP, slot_offset);
-      __ StoreDToOffset(reg, FP, slot_offset);
+      __ LoadDFromOffset(VTMP, FP, slot_offset, PP);
+      __ StoreDToOffset(reg, FP, slot_offset, PP);
       __ fmovdd(reg, VTMP);
     } else {
       UNIMPLEMENTED();
     }
   } else if (source.IsDoubleStackSlot() && destination.IsDoubleStackSlot()) {
     const intptr_t source_offset = source.ToStackSlotOffset();
     const intptr_t dest_offset = destination.ToStackSlotOffset();
 
     ScratchFpuRegisterScope ensure_scratch(this, VTMP);
     VRegister scratch = ensure_scratch.reg();
-    __ LoadDFromOffset(VTMP, FP, source_offset);
-    __ LoadDFromOffset(scratch, FP, dest_offset);
-    __ StoreDToOffset(VTMP, FP, dest_offset);
-    __ StoreDToOffset(scratch, FP, source_offset);
+    __ LoadDFromOffset(VTMP, FP, source_offset, PP);
+    __ LoadDFromOffset(scratch, FP, dest_offset, PP);
+    __ StoreDToOffset(VTMP, FP, dest_offset, PP);
+    __ StoreDToOffset(scratch, FP, source_offset, PP);
   } else if (source.IsQuadStackSlot() && destination.IsQuadStackSlot()) {
     UNIMPLEMENTED();
   } else {
     UNREACHABLE();
   }
 
   // The swap of source and destination has executed a move from source to
   // destination.
   move->Eliminate();
 
@@ skipping 65 matching lines @@
 void ParallelMoveResolver::RestoreFpuScratch(FpuRegister reg) {
   UNIMPLEMENTED();
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM64