[arm64] Use SP-offset rather than FP-offset.

src/compiler/arm64/code-generator-arm64.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 
 #include "src/arm64/frames-arm64.h"
 #include "src/arm64/macro-assembler-arm64.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
@@ skipping 189 matching lines @@
     }
     UNREACHABLE();
     return Operand(-1);
   }
 
   MemOperand ToMemOperand(InstructionOperand* op, MacroAssembler* masm) const {
     DCHECK(op != NULL);
     DCHECK(op->IsStackSlot() || op->IsDoubleStackSlot());
     FrameOffset offset =
         linkage()->GetFrameOffset(AllocatedOperand::cast(op)->index(), frame());
+    if (offset.from_frame_pointer()) {
+      int from_sp =
+          offset.offset() + (frame()->GetSpToFpSlotCount() * kPointerSize);
+      // Convert FP-offsets to SP-offsets if it results in better code.
+      if (Assembler::IsImmLSUnscaled(from_sp) ||
+          Assembler::IsImmLSScaled(from_sp, LSDoubleWord)) {
+        offset = FrameOffset::FromStackPointer(from_sp);
+      }
+    }
     return MemOperand(offset.from_stack_pointer() ? masm->StackPointer() : fp,
                       offset.offset());
   }
 };
 
 
 namespace {
 
 class OutOfLineLoadNaN32 final : public OutOfLineCode {
  public:
@@ skipping 204 matching lines @@
     case kArchCallCodeObject: {
       EnsureSpaceForLazyDeopt();
       if (instr->InputAt(0)->IsImmediate()) {
         __ Call(Handle<Code>::cast(i.InputHeapObject(0)),
                 RelocInfo::CODE_TARGET);
       } else {
         Register target = i.InputRegister(0);
         __ Add(target, target, Code::kHeaderSize - kHeapObjectTag);
         __ Call(target);
       }
+      frame()->ClearOutgoingParameterSlots();
       RecordCallPosition(instr);
       break;
     }
     case kArchTailCallCodeObject: {
       AssembleDeconstructActivationRecord();
       if (instr->InputAt(0)->IsImmediate()) {
         __ Jump(Handle<Code>::cast(i.InputHeapObject(0)),
                 RelocInfo::CODE_TARGET);
       } else {
         Register target = i.InputRegister(0);
         __ Add(target, target, Code::kHeaderSize - kHeapObjectTag);
         __ Jump(target);
       }
+      frame()->ClearOutgoingParameterSlots();
       break;
     }
     case kArchCallJSFunction: {
       EnsureSpaceForLazyDeopt();
       Register func = i.InputRegister(0);
       if (FLAG_debug_code) {
         // Check the function's context matches the context argument.
         UseScratchRegisterScope scope(masm());
         Register temp = scope.AcquireX();
         __ Ldr(temp, FieldMemOperand(func, JSFunction::kContextOffset));
         __ cmp(cp, temp);
         __ Assert(eq, kWrongFunctionContext);
       }
       __ Ldr(x10, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
       __ Call(x10);
+      frame()->ClearOutgoingParameterSlots();
       RecordCallPosition(instr);
       break;
     }
     case kArchTailCallJSFunction: {
       Register func = i.InputRegister(0);
       if (FLAG_debug_code) {
         // Check the function's context matches the context argument.
         UseScratchRegisterScope scope(masm());
         Register temp = scope.AcquireX();
         __ Ldr(temp, FieldMemOperand(func, JSFunction::kContextOffset));
         __ cmp(cp, temp);
         __ Assert(eq, kWrongFunctionContext);
       }
       AssembleDeconstructActivationRecord();
       __ Ldr(x10, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
       __ Jump(x10);
+      frame()->ClearOutgoingParameterSlots();
       break;
     }
     case kArchLazyBailout: {
       EnsureSpaceForLazyDeopt();
       RecordCallPosition(instr);
       break;
     }
     case kArchPrepareCallCFunction:
       // We don't need kArchPrepareCallCFunction on arm64 as the instruction
       // selector already perform a Claim to reserve space on the stack and
       // guarantee correct alignment of stack pointer.
       UNREACHABLE();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());
       if (instr->InputAt(0)->IsImmediate()) {
         ExternalReference ref = i.InputExternalReference(0);
         __ CallCFunction(ref, num_parameters, 0);
       } else {
         Register func = i.InputRegister(0);
         __ CallCFunction(func, num_parameters, 0);
       }
+      // CallCFunction only supports register arguments so we never need to call
+      // frame()->ClearOutgoingParameterSlots() here.
+      DCHECK(frame()->GetOutgoingParameterSlotCount() == 0);
       break;
     }
     case kArchJmp:
       AssembleArchJump(i.InputRpo(0));
       break;
     case kArchTableSwitch:
       AssembleArchTableSwitch(instr);
       break;
     case kArchLookupSwitch:
       AssembleArchLookupSwitch(instr);
@@ skipping 243 matching lines @@
       __ Bfi(i.OutputRegister(), i.InputRegister(1), i.InputInt6(2),
              i.InputInt6(3));
       break;
     case kArm64TestAndBranch32:
     case kArm64TestAndBranch:
       // Pseudo instructions turned into tbz/tbnz in AssembleArchBranch.
       break;
     case kArm64CompareAndBranch32:
       // Pseudo instruction turned into cbz/cbnz in AssembleArchBranch.
       break;
-    case kArm64Claim: {
+    case kArm64ClaimForCallArguments: {
       __ Claim(i.InputInt32(0));
+      frame()->AllocateOutgoingParameterSlots(i.InputInt32(0));
       break;
     }
     case kArm64Poke: {
       Operand operand(i.InputInt32(1) * kPointerSize);
       __ Poke(i.InputRegister(0), operand);
       break;
     }
     case kArm64PokePair: {
       int slot = i.InputInt32(2) - 1;
       __ PokePair(i.InputRegister(1), i.InputRegister(0), slot * kPointerSize);
@@ skipping 387 matching lines @@
 
 
 void CodeGenerator::AssembleDeoptimizerCall(
     int deoptimization_id, Deoptimizer::BailoutType bailout_type) {
   Address deopt_entry = Deoptimizer::GetDeoptimizationEntry(
       isolate(), deoptimization_id, bailout_type);
   __ Call(deopt_entry, RelocInfo::RUNTIME_ENTRY);
 }
 
 
-// TODO(dcarney): increase stack slots in frame once before first use.
-static int AlignedStackSlots(int stack_slots) {
-  if (stack_slots & 1) stack_slots++;
-  return stack_slots;
-}
-
-
 void CodeGenerator::AssemblePrologue() {
   CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
   if (descriptor->kind() == CallDescriptor::kCallAddress) {
     __ SetStackPointer(csp);
     __ Push(lr, fp);
     __ Mov(fp, csp);
   } else if (descriptor->IsJSFunctionCall()) {
     CompilationInfo* info = this->info();
     __ SetStackPointer(jssp);
     __ Prologue(info->IsCodePreAgingActive());
@@ skipping 13 matching lines @@
     // frame is still on the stack. Optimized code uses OSR values directly from
     // the unoptimized frame. Thus, all that needs to be done is to allocate the
     // remaining stack slots.
     if (FLAG_code_comments) __ RecordComment("-- OSR entrypoint --");
     osr_pc_offset_ = __ pc_offset();
     // TODO(titzer): cannot address target function == local #-1
     __ ldr(x1, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
     stack_shrink_slots -= OsrHelper(info()).UnoptimizedFrameSlots();
   }
 
-  if (stack_shrink_slots > 0) {
-    Register sp = __ StackPointer();
-    if (!sp.Is(csp)) {
-      __ Sub(sp, sp, stack_shrink_slots * kPointerSize);
-    }
-    __ Sub(csp, csp, AlignedStackSlots(stack_shrink_slots) * kPointerSize);
+  if (csp.Is(masm()->StackPointer())) {
+    // The system stack pointer requires 16-byte alignment at function call
+    // boundaries.
+    stack_shrink_slots += frame()->AlignSavedCalleeRegisterSlots();
   }
+  __ Claim(stack_shrink_slots);
 
   // Save FP registers.
   CPURegList saves_fp = CPURegList(CPURegister::kFPRegister, kDRegSizeInBits,
                                    descriptor->CalleeSavedFPRegisters());
   int saved_count = saves_fp.Count();
   if (saved_count != 0) {
     DCHECK(saves_fp.list() == CPURegList::GetCalleeSavedFP().list());
     __ PushCPURegList(saves_fp);
     frame()->AllocateSavedCalleeRegisterSlots(saved_count *
                                               (kDoubleSize / kPointerSize));
@@ skipping 231 matching lines @@
       padding_size -= kInstructionSize;
     }
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8