Enable stub generation using Hydrogen/Lithium (again)

src/ia32/deoptimizer-ia32.cc

Index: src/ia32/deoptimizer-ia32.cc
diff --git a/src/ia32/deoptimizer-ia32.cc b/src/ia32/deoptimizer-ia32.cc
index 99ad5225bc0bada7edb634e15935f9647fd92e7f..32f9a31abd3713925d6c54bcfe1b1dfe8a15a732 100644
--- a/src/ia32/deoptimizer-ia32.cc
+++ b/src/ia32/deoptimizer-ia32.cc
@@ -569,6 +569,53 @@ void Deoptimizer::DoComputeArgumentsAdaptorFrame(TranslationIterator* iterator,
 }
 
 
+void Deoptimizer::DoCompiledStubPseudoFrame(TranslationIterator* iterator,
+                                            int frame_index) {
+  //  Builtins* builtins = isolate_->builtins();
+  //  Code::Kind stub_kind = static_cast<Code::Kind>(iterator->Next());
+  FrameDescription* output_frame = new(0) FrameDescription(0, 0);
+  Code* continuation =
+      isolate_->builtins()->builtin(Builtins::kNotifyICMiss);
+  output_frame->SetState(Smi::FromInt(FullCodeGenerator::NO_REGISTERS));
+  output_frame->SetContinuation(
+      reinterpret_cast<uint32_t>(continuation->entry()));
+  Handle<Code> miss_ic = isolate_->builtins()->KeyedLoadIC_Miss();
+  output_frame->SetPc(
+      reinterpret_cast<intptr_t>(miss_ic->instruction_start()));
+
+  Code::Kind stub_kind = static_cast<Code::Kind>(iterator->Next());
+  ASSERT(stub_kind == Code::KEYED_LOAD_IC);
+  USE(stub_kind);
+
+  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);
+
+  Address frame_ptr =
+      reinterpret_cast<Address>(input_->GetRegister(ebp.code()));
+  Address context_ptr_raw = frame_ptr + StandardFrameConstants::kContextOffset;
+  Context** context_ptr =
+      reinterpret_cast<Context**>(context_ptr_raw);
+  output_frame->SetRegister(esi.code(),
+                            reinterpret_cast<intptr_t>(*context_ptr));
+  output_frame->SetRegister(ebp.code(), reinterpret_cast<intptr_t>(frame_ptr));
+
+  ASSERT(frame_index == 0);
+  output_[frame_index] = output_frame;
+}
+
+
 void Deoptimizer::DoComputeConstructStubFrame(TranslationIterator* iterator,
                                               int frame_index) {
   Builtins* builtins = isolate_->builtins();
@@ -997,7 +1044,7 @@ void Deoptimizer::FillInputFrame(Address tos, JavaScriptFrame* frame) {
   }
   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);
   }
 
@@ -1012,7 +1059,6 @@ void Deoptimizer::FillInputFrame(Address tos, JavaScriptFrame* frame) {
 
 void Deoptimizer::EntryGenerator::Generate() {
   GeneratePrologue();
-  CpuFeatures::Scope scope(SSE2);
 
   Isolate* isolate = masm()->isolate();
 
@@ -1022,10 +1068,13 @@ void Deoptimizer::EntryGenerator::Generate() {
   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();
@@ -1073,14 +1122,18 @@ void Deoptimizer::EntryGenerator::Generate() {
     __ 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) {
@@ -1098,10 +1151,13 @@ void Deoptimizer::EntryGenerator::Generate() {
   // 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);
 
@@ -1139,31 +1195,39 @@ void Deoptimizer::EntryGenerator::Generate() {
   }
 
   // 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));
+      }
     }
   }