Enable stub generation using Hydrogen/Lithium (again)

src/ia32/deoptimizer-ia32.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 289 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();
   // TODO(kasperl): This should not be the bailout_id_. It should be
   // the ast id. Confusing.
   ASSERT(bailout_id_ == ast_id);
 
   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);
@@ skipping 16 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, ebp:esp=0x%08x:0x%08x]\n",
@@ skipping 90 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(ebp.code(), frame_pointer);
     output_[0]->SetRegister(esi.code(), input_->GetRegister(esi.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 =
       function_->GetIsolate()->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",
@@ skipping 93 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             <-ebp
+  //    |          ....           |          |          ....           |
+  //    +-------------------------+          +-------------------------+
+  //    | JSFunction continuation |          | JSFunction continuation |
+  //    +-------------------------+          +-------------------------+<-esp
+  // |  |   saved frame (ebp)     |
+  // |  +=========================+<-ebp
+  // |  |   JSFunction context    |
+  // v  +-------------------------+
+  //    |   COMPILED_STUB marker  |          ebp = saved frame
+  //    +-------------------------+          esi = JSFunction context
+  //    |                         |
+  //    | ...                     |
+  //    |                         |
+  //    +-------------------------+<-esp
+  //
+  //
+  int output_frame_size = 1 * kPointerSize;
+  FrameDescription* output_frame =
+      new(output_frame_size) FrameDescription(output_frame_size, 0);
+  Code* continuation =

[Jakob Kummerow, 2012/11/28 16:28:22]

Please choose another name either for this or for the "JSFunction continuation"
in the diagram above. (Same on the other platforms.)

[danno, 2012/11/30 16:23:24]

Done.

+      isolate_->builtins()->builtin(Builtins::kNotifyICMiss);
+  output_frame->SetState(Smi::FromInt(FullCodeGenerator::NO_REGISTERS));
+  output_frame->SetContinuation(
+      reinterpret_cast<uint32_t>(continuation->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(ebp.code(), value);
+  output_frame->SetFp(value);
+  value = input_->GetFrameSlot(input_frame_size - 3 * kPointerSize);
+  output_frame->SetRegister(esi.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(edx.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(ecx.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 408 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 ebp and esp are set to the correct values though.
 
   for (int i = 0; i < Register::kNumRegisters; i++) {
     input_->SetRegister(i, i * 4);
   }
   input_->SetRegister(esp.code(), reinterpret_cast<intptr_t>(frame->sp()));
   input_->SetRegister(ebp.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()->
 
 void Deoptimizer::EntryGenerator::Generate() {
   GeneratePrologue();
-  CpuFeatures::Scope scope(SSE2);
 
   Isolate* isolate = masm()->isolate();
 
   // Save all general purpose registers before messing with them.
   const int kNumberOfRegisters = Register::kNumRegisters;
 
   const int kDoubleRegsSize = kDoubleSize *
                               XMMRegister::kNumAllocatableRegisters;
   __ sub(esp, Immediate(kDoubleRegsSize));
-  for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
-    XMMRegister xmm_reg = XMMRegister::FromAllocationIndex(i);
-    int offset = i * kDoubleSize;
-    __ movdbl(Operand(esp, offset), xmm_reg);
+  if (CpuFeatures::IsSupported(SSE2)) {
+    CpuFeatures::Scope scope(SSE2);
+    for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
+      XMMRegister xmm_reg = XMMRegister::FromAllocationIndex(i);
+      int offset = i * kDoubleSize;
+      __ movdbl(Operand(esp, offset), xmm_reg);
+    }
   }
 
   __ pushad();
 
   const int kSavedRegistersAreaSize = kNumberOfRegisters * kPointerSize +
                                       kDoubleRegsSize;
 
   // Get the bailout id from the stack.
   __ mov(ebx, Operand(esp, kSavedRegistersAreaSize));
 
@@ skipping 27 matching lines @@
   // Preserve deoptimizer object in register eax and get the input
   // frame descriptor pointer.
   __ mov(ebx, Operand(eax, Deoptimizer::input_offset()));
 
   // Fill in the input registers.
   for (int i = kNumberOfRegisters - 1; i >= 0; i--) {
     int offset = (i * kPointerSize) + FrameDescription::registers_offset();
     __ pop(Operand(ebx, offset));
   }
 
-  // Fill in the double input registers.
   int double_regs_offset = FrameDescription::double_registers_offset();
-  for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
-    int dst_offset = i * kDoubleSize + double_regs_offset;
-    int src_offset = i * kDoubleSize;
-    __ movdbl(xmm0, Operand(esp, src_offset));
-    __ movdbl(Operand(ebx, dst_offset), xmm0);
+  if (CpuFeatures::IsSupported(SSE2)) {
+    CpuFeatures::Scope scope(SSE2);
+    // Fill in the double input registers.
+    for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
+      int dst_offset = i * kDoubleSize + double_regs_offset;
+      int src_offset = i * kDoubleSize;
+      __ movdbl(xmm0, Operand(esp, src_offset));
+      __ movdbl(Operand(ebx, dst_offset), xmm0);
+    }
   }
+  __ fninit();
 
   // Remove the bailout id and the double registers from the stack.
   if (type() == EAGER) {
     __ add(esp, Immediate(kDoubleRegsSize + kPointerSize));
   } else {
     __ add(esp, Immediate(kDoubleRegsSize + 2 * kPointerSize));
   }
 
   // Compute a pointer to the unwinding limit in register ecx; that is
   // the first stack slot not part of the input frame.
   __ mov(ecx, Operand(ebx, FrameDescription::frame_size_offset()));
   __ add(ecx, esp);
 
   // Unwind the stack down to - but not including - the unwinding
   // limit and copy the contents of the activation frame to the input
   // frame description.
   __ lea(edx, Operand(ebx, FrameDescription::frame_content_offset()));
+  Label pop_loop_header;
+  __ jmp(&pop_loop_header);
   Label pop_loop;
   __ bind(&pop_loop);
   __ pop(Operand(edx, 0));
   __ add(edx, Immediate(sizeof(uint32_t)));
+  __ bind(&pop_loop_header);
   __ cmp(ecx, esp);
   __ j(not_equal, &pop_loop);
 
   // Compute the output frame in the deoptimizer.
   __ push(eax);
   __ PrepareCallCFunction(1, ebx);
   __ mov(Operand(esp, 0 * kPointerSize), eax);
   {
     AllowExternalCallThatCantCauseGC scope(masm());
     __ CallCFunction(
@@ skipping 17 matching lines @@
     // If frame needs dynamic alignment push padding.
     Label no_padding;
     __ cmp(Operand(eax, Deoptimizer::has_alignment_padding_offset()),
            Immediate(0));
     __ j(equal, &no_padding);
     __ push(Immediate(kAlignmentZapValue));
     __ bind(&no_padding);
   }
 
   // Replace the current 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: eax = current FrameDescription**, edx = one past the
   // last FrameDescription**.
   __ mov(edx, Operand(eax, Deoptimizer::output_count_offset()));
   __ mov(eax, Operand(eax, Deoptimizer::output_offset()));
   __ lea(edx, Operand(eax, edx, times_4, 0));
+  __ jmp(&outer_loop_header);
   __ bind(&outer_push_loop);
   // Inner loop state: ebx = current FrameDescription*, ecx = loop index.
   __ mov(ebx, Operand(eax, 0));
   __ mov(ecx, Operand(ebx, FrameDescription::frame_size_offset()));
+  __ jmp(&inner_loop_header);
   __ bind(&inner_push_loop);
   __ sub(ecx, Immediate(sizeof(uint32_t)));
   __ push(Operand(ebx, ecx, times_1, FrameDescription::frame_content_offset()));
+  __ bind(&inner_loop_header);
   __ test(ecx, ecx);
   __ j(not_zero, &inner_push_loop);
   __ add(eax, Immediate(kPointerSize));
+  __ bind(&outer_loop_header);
   __ cmp(eax, edx);
   __ j(below, &outer_push_loop);
 
   // In case of OSR, we have to restore the XMM registers.
   if (type() == OSR) {
-    for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
-      XMMRegister xmm_reg = XMMRegister::FromAllocationIndex(i);
-      int src_offset = i * kDoubleSize + double_regs_offset;
-      __ movdbl(xmm_reg, Operand(ebx, src_offset));
+    if (CpuFeatures::IsSupported(SSE2)) {
+      CpuFeatures::Scope scope(SSE2);
+      for (int i = 0; i < XMMRegister::kNumAllocatableRegisters; ++i) {
+        XMMRegister xmm_reg = XMMRegister::FromAllocationIndex(i);
+        int src_offset = i * kDoubleSize + double_regs_offset;
+        __ movdbl(xmm_reg, Operand(ebx, src_offset));
+      }
     }
   }
 
   // Push state, pc, and continuation from the last output frame.
   if (type() != OSR) {
     __ push(Operand(ebx, FrameDescription::state_offset()));
   }
   __ push(Operand(ebx, FrameDescription::pc_offset()));
   __ push(Operand(ebx, FrameDescription::continuation_offset()));
 
@@ skipping 24 matching lines @@
   }
   __ bind(&done);
 }
 
 #undef __
 
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_IA32