Record a correct deoptimization environment for slow-path code.

runtime/vm/flow_graph_compiler.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_XXX.
 
 #include "vm/flow_graph_compiler.h"
 
 #include "vm/cha.h"
 #include "vm/dart_entry.h"
@@ skipping 379 matching lines @@
   ASSERT(is_optimizing());
   CompilerDeoptInfo* info =
       new CompilerDeoptInfo(deopt_id,
                             kDeoptAtCall,
                             pending_deoptimization_env_);
   info->set_pc_offset(assembler()->CodeSize());
   deopt_infos_.Add(info);
 }
 
 
+// This function must be in sync with FlowGraphCompiler::SaveLiveRegisters
+// and FlowGraphCompiler::SlowPathEnvironmentFor.
 void FlowGraphCompiler::RecordSafepoint(LocationSummary* locs) {
   if (is_optimizing()) {
     BitmapBuilder* bitmap = locs->stack_bitmap();
     ASSERT(bitmap != NULL);
     ASSERT(bitmap->Length() <= StackSize());
     // Pad the bitmap out to describe all the spill slots.
     bitmap->SetLength(StackSize());
 
     // Mark the bits in the stack map in the same order we push registers in
     // slow path code (see FlowGraphCompiler::SaveLiveRegisters).
     //
     // Slow path code can have registers at the safepoint.
     if (!locs->always_calls()) {
       RegisterSet* regs = locs->live_registers();
-      if (regs->fpu_regs_count() > 0) {
+      if (regs->FpuRegisterCount() > 0) {
         // Denote FPU registers with 0 bits in the stackmap.  Based on the
         // assumption that there are normally few live FPU registers, this
         // encoding is simpler and roughly as compact as storing a separate
         // count of FPU registers.
         //
         // FPU registers have the highest register number at the highest
         // address (i.e., first in the stackmap).
         const intptr_t kFpuRegisterSpillFactor =
             kFpuRegisterSize / kWordSize;
         for (intptr_t i = kNumberOfFpuRegisters - 1; i >= 0; --i) {
@@ skipping 16 matching lines @@
     }
 
     intptr_t register_bit_count = bitmap->Length() - StackSize();
     stackmap_table_builder_->AddEntry(assembler()->CodeSize(),
                                       bitmap,
                                       register_bit_count);
   }
 }
 
 
+// This function must be in sync with FlowGraphCompiler::RecordSafepoint and
+// FlowGraphCompiler::SaveLiveRegisters.
+Environment* FlowGraphCompiler::SlowPathEnvironmentFor(
+    Instruction* instruction) {
+  if (instruction->env() == NULL) return NULL;
+
+  Environment* env = instruction->env()->DeepCopy();
+  // 1. Iterate the registers in the order they will be spilled to compute
+  //    the slots they will be spilled to.
+  intptr_t next_slot = StackSize();
+  RegisterSet* regs = instruction->locs()->live_registers();
+  intptr_t fpu_reg_slots[kNumberOfFpuRegisters];
+  intptr_t cpu_reg_slots[kNumberOfCpuRegisters];
+  const intptr_t kFpuRegisterSpillFactor = kFpuRegisterSize / kWordSize;
+  // FPU registers are spilled first from highest to lowest register number.
+  for (intptr_t i = kNumberOfFpuRegisters - 1; i >= 0; --i) {
+    FpuRegister reg = static_cast<FpuRegister>(i);
+    if (regs->ContainsFpuRegister(reg)) {
+      // We use the lowest address (thus highest index) to identify a
+      // multi-word spill slot.
+      next_slot += kFpuRegisterSpillFactor;
+      fpu_reg_slots[i] = (next_slot - 1);
+    } else {
+      fpu_reg_slots[i] = -1;
+    }
+  }
+  // General purpose registers are spilled from lowest to highest register
+  // number.
+  for (intptr_t i = 0; i < kNumberOfCpuRegisters; ++i) {
+    Register reg = static_cast<Register>(i);
+    if (regs->ContainsRegister(reg)) {
+      cpu_reg_slots[i] = next_slot++;
+    } else {
+      cpu_reg_slots[i] = -1;
+    }
+  }
+
+  // 2. Iterate the environment and replace register locations with the
+  //    corresponding spill slot locations.
+  for (Environment::DeepIterator it(env); !it.Done(); it.Advance()) {
+    Location loc = it.CurrentLocation();
+    if (loc.IsRegister()) {
+      intptr_t index = cpu_reg_slots[loc.reg()];
+      ASSERT(index >= 0);

[Florian Schneider, 2013/10/17 14:38:52]

Shouldn't this be even stronger?

ASSERT(index >= 0 && index >= StackSize());

+      it.SetCurrentLocation(Location::StackSlot(index));
+    } else if (loc.IsFpuRegister()) {
+      assembler()->Comment("You betcha!");

[Kevin Millikin (Google), 2013/10/14 16:12:52]

Ignore this.

[srdjan, 2013/12/03 22:42:10]

But it was submitted. Can we replace it with something more meaningful :-)?

+      intptr_t index = fpu_reg_slots[loc.fpu_reg()];
+      ASSERT(index >= 0);

[Florian Schneider, 2013/10/17 14:38:52]

ASSERT(index >= 0 && index >= StackSize());

+      Value* value = it.CurrentValue();
+      switch (value->definition()->representation()) {
+        case kUnboxedDouble:
+        case kUnboxedMint:
+          it.SetCurrentLocation(Location::DoubleStackSlot(index));

[Kevin Millikin (Google), 2013/10/14 16:12:52]

As discussed offline, I'm not sure that this index is correct.  We may need to
subtract two here.

I'll verify with a test case.

+          break;
+        case kUnboxedFloat32x4:
+        case kUnboxedUint32x4:
+          it.SetCurrentLocation(Location::QuadStackSlot(index));

[Florian Schneider, 2013/10/17 14:38:52]

Try to also have a test case for the quad-word case here.

+          break;
+        default:
+          UNREACHABLE();
+      }
+    }
+  }
+
+  return env;
+}
+
+
 Label* FlowGraphCompiler::AddDeoptStub(intptr_t deopt_id,
                                        DeoptReasonId reason) {
   ASSERT(is_optimizing_);
   CompilerDeoptInfoWithStub* stub =
       new CompilerDeoptInfoWithStub(deopt_id,
                                     reason,
                                     pending_deoptimization_env_);
   deopt_infos_.Add(stub);
   return stub->entry_label();
 }
@@ skipping 673 matching lines @@
 
   for (int i = 0; i < len; i++) {
     sorted->Add(CidTarget(ic_data.GetReceiverClassIdAt(i),
                           &Function::ZoneHandle(ic_data.GetTargetAt(i)),
                           ic_data.GetCountAt(i)));
   }
   sorted->Sort(HighestCountFirst);
 }
 
 }  // namespace dart