Enable stub generation using Hydrogen/Lithium (again)

src/arm/deoptimizer-arm.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 204 matching lines @@
       if (value == 1) return i;
     }
   }
   UNREACHABLE();
   return -1;
 }
 
 
 void Deoptimizer::DoComputeOsrOutputFrame() {
   DeoptimizationInputData* data = DeoptimizationInputData::cast(
-      optimized_code_->deoptimization_data());
+      compiled_code_->deoptimization_data());
   unsigned ast_id = data->OsrAstId()->value();
 
   int bailout_id = LookupBailoutId(data, BailoutId(ast_id));
   unsigned translation_index = data->TranslationIndex(bailout_id)->value();
   ByteArray* translations = data->TranslationByteArray();
 
   TranslationIterator iterator(translations, translation_index);
   Translation::Opcode opcode =
       static_cast<Translation::Opcode>(iterator.Next());
   ASSERT(Translation::BEGIN == opcode);
@@ skipping 13 matching lines @@
   USE(closure_id);
   ASSERT_EQ(Translation::kSelfLiteralId, closure_id);
   unsigned height = iterator.Next();
   unsigned height_in_bytes = height * kPointerSize;
   USE(height_in_bytes);
 
   unsigned fixed_size = ComputeFixedSize(function_);
   unsigned input_frame_size = input_->GetFrameSize();
   ASSERT(fixed_size + height_in_bytes == input_frame_size);
 
-  unsigned stack_slot_size = optimized_code_->stack_slots() * kPointerSize;
+  unsigned stack_slot_size = compiled_code_->stack_slots() * kPointerSize;
   unsigned outgoing_height = data->ArgumentsStackHeight(bailout_id)->value();
   unsigned outgoing_size = outgoing_height * kPointerSize;
   unsigned output_frame_size = fixed_size + stack_slot_size + outgoing_size;
   ASSERT(outgoing_size == 0);  // OSR does not happen in the middle of a call.
 
   if (FLAG_trace_osr) {
     PrintF("[on-stack replacement: begin 0x%08" V8PRIxPTR " ",
            reinterpret_cast<intptr_t>(function_));
     function_->PrintName();
     PrintF(" => node=%u, frame=%d->%d]\n",
@@ skipping 71 matching lines @@
     delete output_[0];
     output_[0] = input_;
     output_[0]->SetPc(reinterpret_cast<uint32_t>(from_));
   } else {
     // Set up the frame pointer and the context pointer.
     output_[0]->SetRegister(fp.code(), input_->GetRegister(fp.code()));
     output_[0]->SetRegister(cp.code(), input_->GetRegister(cp.code()));
 
     unsigned pc_offset = data->OsrPcOffset()->value();
     uint32_t pc = reinterpret_cast<uint32_t>(
-        optimized_code_->entry() + pc_offset);
+        compiled_code_->entry() + pc_offset);
     output_[0]->SetPc(pc);
   }
   Code* continuation = isolate_->builtins()->builtin(Builtins::kNotifyOSR);
   output_[0]->SetContinuation(
       reinterpret_cast<uint32_t>(continuation->entry()));
 
   if (FLAG_trace_osr) {
     PrintF("[on-stack replacement translation %s: 0x%08" V8PRIxPTR " ",
            ok ? "finished" : "aborted",
            reinterpret_cast<intptr_t>(function_));
@@ skipping 92 matching lines @@
   Builtins* builtins = isolate_->builtins();
   Code* adaptor_trampoline =
       builtins->builtin(Builtins::kArgumentsAdaptorTrampoline);
   uint32_t pc = reinterpret_cast<uint32_t>(
       adaptor_trampoline->instruction_start() +
       isolate_->heap()->arguments_adaptor_deopt_pc_offset()->value());
   output_frame->SetPc(pc);
 }
 
 
+void Deoptimizer::DoCompiledStubFrame(TranslationIterator* iterator,
+                                      int frame_index) {
+  //
+  //               FROM                                  TO             <-fp
+  //    |          ....           |          |          ....           |
+  //    +-------------------------+          +-------------------------+
+  //    | JSFunction continuation |          | JSFunction continuation |
+  //    +-------------------------+          +-------------------------+<-sp
+  // |  |   saved frame (fp)      |
+  // |  +=========================+<-fp
+  // |  |   JSFunction context    |
+  // v  +-------------------------+
+  //    |   COMPILED_STUB marker  |          fp = saved frame
+  //    +-------------------------+          f8 = JSFunction context
+  //    |                         |
+  //    | ...                     |
+  //    |                         |
+  //    +-------------------------+<-sp
+  //
+  //
+  int output_frame_size = 1 * kPointerSize;
+  FrameDescription* output_frame =
+      new(output_frame_size) FrameDescription(output_frame_size, 0);
+  Code* notify_miss =
+      isolate_->builtins()->builtin(Builtins::kNotifyICMiss);
+  output_frame->SetState(Smi::FromInt(FullCodeGenerator::NO_REGISTERS));
+  output_frame->SetContinuation(
+      reinterpret_cast<intptr_t>(notify_miss->entry()));
+
+  ASSERT(compiled_code_->kind() == Code::COMPILED_STUB);
+  int major_key = compiled_code_->major_key();
+  CodeStubInterfaceDescriptor* descriptor =
+      isolate_->code_stub_interface_descriptors()[major_key];
+  Handle<Code> miss_ic(descriptor->deoptimization_handler);
+  output_frame->SetPc(reinterpret_cast<intptr_t>(miss_ic->instruction_start()));
+  unsigned input_frame_size = input_->GetFrameSize();
+  intptr_t value = input_->GetFrameSlot(input_frame_size - kPointerSize);
+  output_frame->SetFrameSlot(0, value);
+  value = input_->GetFrameSlot(input_frame_size - 2 * kPointerSize);
+  output_frame->SetRegister(fp.code(), value);
+  output_frame->SetFp(value);
+  value = input_->GetFrameSlot(input_frame_size - 3 * kPointerSize);
+  output_frame->SetRegister(cp.code(), value);
+
+  Translation::Opcode opcode =
+      static_cast<Translation::Opcode>(iterator->Next());
+  ASSERT(opcode == Translation::REGISTER);
+  USE(opcode);
+  int input_reg = iterator->Next();
+  intptr_t input_value = input_->GetRegister(input_reg);
+  output_frame->SetRegister(r1.code(), input_value);
+
+  int32_t next = iterator->Next();
+  opcode = static_cast<Translation::Opcode>(next);
+  ASSERT(opcode == Translation::REGISTER);
+  input_reg = iterator->Next();
+  input_value = input_->GetRegister(input_reg);
+  output_frame->SetRegister(r0.code(), input_value);
+
+  ASSERT(frame_index == 0);
+  output_[frame_index] = output_frame;
+}
+
+
 void Deoptimizer::DoComputeConstructStubFrame(TranslationIterator* iterator,
                                               int frame_index) {
   Builtins* builtins = isolate_->builtins();
   Code* construct_stub = builtins->builtin(Builtins::kJSConstructStubGeneric);
   JSFunction* function = JSFunction::cast(ComputeLiteral(iterator->Next()));
   unsigned height = iterator->Next();
   unsigned height_in_bytes = height * kPointerSize;
   if (FLAG_trace_deopt) {
     PrintF("  translating construct stub => height=%d\n", height_in_bytes);
   }
@@ skipping 407 matching lines @@
 void Deoptimizer::FillInputFrame(Address tos, JavaScriptFrame* frame) {
   // Set the register values. The values are not important as there are no
   // callee saved registers in JavaScript frames, so all registers are
   // spilled. Registers fp and sp are set to the correct values though.
 
   for (int i = 0; i < Register::kNumRegisters; i++) {
     input_->SetRegister(i, i * 4);
   }
   input_->SetRegister(sp.code(), reinterpret_cast<intptr_t>(frame->sp()));
   input_->SetRegister(fp.code(), reinterpret_cast<intptr_t>(frame->fp()));
-  for (int i = 0; i < DoubleRegister::kNumAllocatableRegisters; i++) {
+  for (int i = 0; i < DoubleRegister::NumAllocatableRegisters(); i++) {
     input_->SetDoubleRegister(i, 0.0);
   }
 
   // Fill the frame content from the actual data on the frame.
   for (unsigned i = 0; i < input_->GetFrameSize(); i += kPointerSize) {
     input_->SetFrameSlot(i, Memory::uint32_at(tos + i));
   }
 }
 
 
 #define __ masm()->
 
 // This code tries to be close to ia32 code so that any changes can be
 // easily ported.
 void Deoptimizer::EntryGenerator::Generate() {
   GeneratePrologue();
 
   Isolate* isolate = masm()->isolate();
 
-  CpuFeatures::Scope scope(VFP3);
   // Save all general purpose registers before messing with them.
   const int kNumberOfRegisters = Register::kNumRegisters;
 
   // Everything but pc, lr and ip which will be saved but not restored.
   RegList restored_regs = kJSCallerSaved | kCalleeSaved | ip.bit();
 
   const int kDoubleRegsSize =
-      kDoubleSize * DwVfpRegister::kNumAllocatableRegisters;
+      kDoubleSize * DwVfpRegister::kMaxNumAllocatableRegisters;
 
-  // Save all VFP registers before messing with them.
-  DwVfpRegister first = DwVfpRegister::FromAllocationIndex(0);
-  DwVfpRegister last =
-      DwVfpRegister::FromAllocationIndex(
-          DwVfpRegister::kNumAllocatableRegisters - 1);
-  ASSERT(last.code() > first.code());
-  ASSERT((last.code() - first.code()) ==
-      (DwVfpRegister::kNumAllocatableRegisters - 1));
+  if (CpuFeatures::IsSupported(VFP2)) {
+    CpuFeatures::Scope scope(VFP2);
+    // Save all VFP registers before messing with them.
+    DwVfpRegister first = DwVfpRegister::FromAllocationIndex(0);
+    DwVfpRegister last =
+        DwVfpRegister::FromAllocationIndex(
+            DwVfpRegister::kMaxNumAllocatableRegisters - 1);
+    ASSERT(last.code() > first.code());
+    ASSERT((last.code() - first.code()) ==
+           (DwVfpRegister::kMaxNumAllocatableRegisters - 1));
 #ifdef DEBUG
-  for (int i = 0; i <= (DwVfpRegister::kNumAllocatableRegisters - 1); i++) {
-    ASSERT((DwVfpRegister::FromAllocationIndex(i).code() <= last.code()) &&
-           (DwVfpRegister::FromAllocationIndex(i).code() >= first.code()));
+    int max = DwVfpRegister::kMaxNumAllocatableRegisters - 1;
+    for (int i = 0; i <= max; i++) {
+      ASSERT((DwVfpRegister::FromAllocationIndex(i).code() <= last.code()) &&
+             (DwVfpRegister::FromAllocationIndex(i).code() >= first.code()));
+    }
+#endif
+    __ vstm(db_w, sp, first, last);
+  } else {
+    __ sub(sp, sp, Operand(kDoubleRegsSize));
   }
-#endif
-  __ vstm(db_w, sp, first, last);
 
   // Push all 16 registers (needed to populate FrameDescription::registers_).
   // TODO(1588) Note that using pc with stm is deprecated, so we should perhaps
   // handle this a bit differently.
   __ stm(db_w, sp, restored_regs  | sp.bit() | lr.bit() | pc.bit());
 
   const int kSavedRegistersAreaSize =
       (kNumberOfRegisters * kPointerSize) + kDoubleRegsSize;
 
   // Get the bailout id from the stack.
@@ skipping 38 matching lines @@
   __ ldr(r1, MemOperand(r0, Deoptimizer::input_offset()));
 
   // Copy core registers into FrameDescription::registers_[kNumRegisters].
   ASSERT(Register::kNumRegisters == kNumberOfRegisters);
   for (int i = 0; i < kNumberOfRegisters; i++) {
     int offset = (i * kPointerSize) + FrameDescription::registers_offset();
     __ ldr(r2, MemOperand(sp, i * kPointerSize));
     __ str(r2, MemOperand(r1, offset));
   }
 
-  // Copy VFP registers to
-  // double_registers_[DoubleRegister::kNumAllocatableRegisters]
-  int double_regs_offset = FrameDescription::double_registers_offset();
-  for (int i = 0; i < DwVfpRegister::kNumAllocatableRegisters; ++i) {
-    int dst_offset = i * kDoubleSize + double_regs_offset;
-    int src_offset = i * kDoubleSize + kNumberOfRegisters * kPointerSize;
-    __ vldr(d0, sp, src_offset);
-    __ vstr(d0, r1, dst_offset);
+  if (CpuFeatures::IsSupported(VFP2)) {
+    CpuFeatures::Scope scope(VFP2);
+    // Copy VFP registers to
+    // double_registers_[DoubleRegister::kMaxNumAllocatableRegisters]
+    int double_regs_offset = FrameDescription::double_registers_offset();
+    for (int i = 0; i < DwVfpRegister::NumAllocatableRegisters(); ++i) {
+      int dst_offset = i * kDoubleSize + double_regs_offset;
+      int src_offset = i * kDoubleSize + kNumberOfRegisters * kPointerSize;
+      __ vldr(d0, sp, src_offset);
+      __ vstr(d0, r1, dst_offset);
+    }
   }
 
   // Remove the bailout id, eventually return address, and the saved registers
   // from the stack.
   if (type() == EAGER || type() == OSR) {
     __ add(sp, sp, Operand(kSavedRegistersAreaSize + (1 * kPointerSize)));
   } else {
     __ add(sp, sp, Operand(kSavedRegistersAreaSize + (2 * kPointerSize)));
   }
 
   // Compute a pointer to the unwinding limit in register r2; that is
   // the first stack slot not part of the input frame.
   __ ldr(r2, MemOperand(r1, FrameDescription::frame_size_offset()));
   __ add(r2, r2, sp);
 
   // Unwind the stack down to - but not including - the unwinding
   // limit and copy the contents of the activation frame to the input
   // frame description.
   __ add(r3,  r1, Operand(FrameDescription::frame_content_offset()));
   Label pop_loop;
+  Label pop_loop_header;
+  __ b(&pop_loop_header);
   __ bind(&pop_loop);
   __ pop(r4);
   __ str(r4, MemOperand(r3, 0));
   __ add(r3, r3, Operand(sizeof(uint32_t)));
+  __ bind(&pop_loop_header);
   __ cmp(r2, sp);
   __ b(ne, &pop_loop);
 
   // Compute the output frame in the deoptimizer.
   __ push(r0);  // Preserve deoptimizer object across call.
   // r0: deoptimizer object; r1: scratch.
   __ PrepareCallCFunction(1, r1);
   // Call Deoptimizer::ComputeOutputFrames().
   {
     AllowExternalCallThatCantCauseGC scope(masm());
     __ CallCFunction(
         ExternalReference::compute_output_frames_function(isolate), 1);
   }
   __ pop(r0);  // Restore deoptimizer object (class Deoptimizer).
 
   // Replace the current (input) frame with the output frames.
-  Label outer_push_loop, inner_push_loop;
+  Label outer_push_loop, inner_push_loop,
+      outer_loop_header, inner_loop_header;
   // Outer loop state: r0 = current "FrameDescription** output_",
   // r1 = one past the last FrameDescription**.
   __ ldr(r1, MemOperand(r0, Deoptimizer::output_count_offset()));
   __ ldr(r0, MemOperand(r0, Deoptimizer::output_offset()));  // r0 is output_.
   __ add(r1, r0, Operand(r1, LSL, 2));
+  __ jmp(&outer_loop_header);
   __ bind(&outer_push_loop);
   // Inner loop state: r2 = current FrameDescription*, r3 = loop index.
   __ ldr(r2, MemOperand(r0, 0));  // output_[ix]
   __ ldr(r3, MemOperand(r2, FrameDescription::frame_size_offset()));
+  __ jmp(&inner_loop_header);
   __ bind(&inner_push_loop);
   __ sub(r3, r3, Operand(sizeof(uint32_t)));
   __ add(r6, r2, Operand(r3));
   __ ldr(r7, MemOperand(r6, FrameDescription::frame_content_offset()));
   __ push(r7);
+  __ bind(&inner_loop_header);
   __ cmp(r3, Operand(0));
   __ b(ne, &inner_push_loop);  // test for gt?
   __ add(r0, r0, Operand(kPointerSize));
+  __ bind(&outer_loop_header);
   __ cmp(r0, r1);
   __ b(lt, &outer_push_loop);
 
   // Push state, pc, and continuation from the last output frame.
   if (type() != OSR) {
     __ ldr(r6, MemOperand(r2, FrameDescription::state_offset()));
     __ push(r6);
   }
 
   __ ldr(r6, MemOperand(r2, FrameDescription::pc_offset()));
@@ skipping 40 matching lines @@
     __ push(ip);
     __ b(&done);
     ASSERT(masm()->pc_offset() - start == table_entry_size_);
   }
   __ bind(&done);
 }
 
 #undef __
 
 } }  // namespace v8::internal