X87: [crankshaft] Fragmentation-free allocation folding.

  port 61f5fbbb (r36133)

  original commit message:
  The new allocation folding implementation avoids fragmentation between folded allocation. As a consequence, our heap will always be iterable i.e. we do not have to perform a garbage collection
  before iterating the heap.

BUG=

Review-Url: https://codereview.chromium.org/1969553003
Cr-Commit-Position: refs/heads/master@{#36158}

src/crankshaft/x87/lithium-codegen-x87.cc

@@ -5456,6 +5456,30 @@ void LCodeGen::DoAllocate(LAllocate* instr) {
   }
 }
 
+void LCodeGen::DoFastAllocate(LFastAllocate* instr) {
+  DCHECK(instr->hydrogen()->IsAllocationFolded());
+  Register result = ToRegister(instr->result());
+  Register temp = ToRegister(instr->temp());
+
+  AllocationFlags flags = NO_ALLOCATION_FLAGS;
+  if (instr->hydrogen()->MustAllocateDoubleAligned()) {
+    flags = static_cast<AllocationFlags>(flags | DOUBLE_ALIGNMENT);
+  }
+  if (instr->hydrogen()->IsOldSpaceAllocation()) {
+    DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
+    flags = static_cast<AllocationFlags>(flags | PRETENURE);
+  }
+  if (!instr->hydrogen()->IsAllocationFoldingDominator()) {
+    if (instr->size()->IsConstantOperand()) {
+      int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
+      CHECK(size <= Page::kMaxRegularHeapObjectSize);
+      __ FastAllocate(size, result, temp, flags);
+    } else {
+      Register size = ToRegister(instr->size());
+      __ FastAllocate(size, result, temp, flags);
+    }
+  }
+}
 
 void LCodeGen::DoDeferredAllocate(LAllocate* instr) {
   Register result = ToRegister(instr->result());
@@ -5513,31 +5537,6 @@ void LCodeGen::DoDeferredAllocate(LAllocate* instr) {
   }
 }
 
-void LCodeGen::DoFastAllocate(LFastAllocate* instr) {
-  DCHECK(instr->hydrogen()->IsAllocationFolded());
-  Register result = ToRegister(instr->result());
-  Register temp = ToRegister(instr->temp());
-
-  AllocationFlags flags = NO_ALLOCATION_FLAGS;
-  if (instr->hydrogen()->MustAllocateDoubleAligned()) {
-    flags = static_cast<AllocationFlags>(flags | DOUBLE_ALIGNMENT);
-  }
-  if (instr->hydrogen()->IsOldSpaceAllocation()) {
-    DCHECK(!instr->hydrogen()->IsNewSpaceAllocation());
-    flags = static_cast<AllocationFlags>(flags | PRETENURE);
-  }
-  if (!instr->hydrogen()->IsAllocationFoldingDominator()) {
-    if (instr->size()->IsConstantOperand()) {
-      int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
-      CHECK(size <= Page::kMaxRegularHeapObjectSize);
-      __ FastAllocate(size, result, temp, flags);
-    } else {
-      Register size = ToRegister(instr->size());
-      __ FastAllocate(size, result, temp, flags);
-    }
-  }
-}
-
 
 void LCodeGen::DoTypeof(LTypeof* instr) {
   DCHECK(ToRegister(instr->context()).is(esi));

src/x87/macro-assembler-x87.h

@@ -629,6 +629,9 @@ class MacroAssembler: public Assembler {
   void Allocate(Register object_size, Register result, Register result_end,
                 Register scratch, Label* gc_required, AllocationFlags flags);
 
+  // FastAllocate is right now only used for folded allocations. It just
+  // increments the top pointer without checking against limit. This can only
+  // be done if it was proved earlier that the allocation will succeed.
   void FastAllocate(int object_size, Register result, Register result_end,
                     AllocationFlags flags);
   void FastAllocate(Register object_size, Register result, Register result_end,