Change deoptimize entrypoint to save everything.

And implement FPU register retrieval from stack on x86.

On Nexus 9, AOSP ToT, the boot.oat size reduction is
  prebuilt multi-part boot image:
    - 32-bit boot.oat: -20KiB (-0.03%)
    - 64-bit boot.oat: -45KiB (-0.06%)
  on-device built single boot image:
    - 32-bit boot.oat: -24KiB (-0.04%)
    - 64-bit boot.oat: -36KiB (-0.05%)

Test: Run ART test suite on host and Nexus 9.
Bug: 30212852
Change-Id: I5d98e2a24363136d73dfec6100ab02f8eb101911
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 3923f52..ceceedd 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -494,7 +494,6 @@
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARM64* arm64_codegen = down_cast<CodeGeneratorARM64*>(codegen);
     __ Bind(GetEntryLabel());
-    SaveLiveRegisters(codegen, instruction_->GetLocations());
     arm64_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
     CheckEntrypointTypes<kQuickDeoptimize, void, void>();
   }
@@ -3060,6 +3059,7 @@
 void LocationsBuilderARM64::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
+  locations->SetCustomSlowPathCallerSaves(RegisterSet());  // No caller-save registers.
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }