Change static calls in unoptimized code to always call via a stub. Using ICData, the call count of …

runtime/vm/stub_code_mips.cc

Index: runtime/vm/stub_code_mips.cc
===================================================================
--- runtime/vm/stub_code_mips.cc	(revision 24288)
+++ runtime/vm/stub_code_mips.cc	(working copy)
@@ -1798,13 +1798,83 @@
 }
 
 
+// Intermediary stub between a static call and its target. ICData contains
+// the target function and the call count.
+// S5: ICData
+void StubCode::GenerateUnoptimizedStaticCallStub(Assembler* assembler) {
+  GenerateUsageCounterIncrement(assembler, T0);
+  __ TraceSimMsg("UnoptimizedStaticCallStub");
+#if defined(DEBUG)
+  { Label ok;
+    // Check that the IC data array has NumberOfArgumentsChecked() == 0.
+    // 'num_args_tested' is stored as an untagged int.
+    __ lw(T0, FieldAddress(S5, ICData::num_args_tested_offset()));
+    __ BranchEqual(T0, 0, &ok);

[zra, 2013/06/21 19:03:35]

__ beq(T0, ZR, &ok)

[srdjan, 2013/06/21 20:30:18]

Done.

+    __ Stop("Incorrect IC data for unoptimized static call");
+    __ Bind(&ok);
+  }
+#endif  // DEBUG
+
+  // S5: IC data object (preserved).
+  __ lw(T0, FieldAddress(S5, ICData::ic_data_offset()));
+  // T0: ic_data_array with entries: target functions and count.
+  __ AddImmediate(T0, Array::data_offset() - kHeapObjectTag);
+  // T0: points directly to the first ic data array element.
+  const intptr_t target_offset = ICData::TargetIndexFor(0) * kWordSize;
+  const intptr_t count_offset = ICData::CountIndexFor(0) * kWordSize;
+
+  // Increment count for this call.
+  Label increment_done;
+  __ lw(T4, Address(T0, count_offset));
+  __ AddImmediateDetectOverflow(T4, T4, Smi::RawValue(1), T5, T6);
+  __ bgez(T5, &increment_done);  // No overflow.
+  __ delay_slot()->sw(T4, Address(T0, count_offset));
+
+  __ LoadImmediate(T1, Smi::RawValue(Smi::kMaxValue));
+  __ sw(T1, Address(T0, count_offset));
+
+  __ Bind(&increment_done);
+
+  Label target_is_compiled;
+  // Get function and call it, if possible.
+  __ lw(T3, Address(T0, target_offset));
+  __ lw(T4, FieldAddress(T3, Function::code_offset()));
+  __ LoadImmediate(TMP, reinterpret_cast<intptr_t>(Object::null()));
+  __ bne(T4, TMP, &target_is_compiled);
+
+  __ EnterStubFrame();
+  // Preserve target function and IC data object.
+  int num_slots = 3;  // Two preserved registers, one argument (function).

[zra, 2013/06/21 19:03:35]

When the number of slots is fixed, I've been foregoing a variable for the number
of slots like this.

[srdjan, 2013/06/21 20:30:18]

Done.

+  __ addiu(SP, SP, Immediate(-num_slots * kWordSize));
+  __ sw(S5, Address(SP, (num_slots - 1) * kWordSize));  // Preserve IC data.
+  __ sw(T3, Address(SP, (num_slots - 2) * kWordSize));  // Preserve function.
+  __ sw(T3, Address(SP, (num_slots - 3) * kWordSize));  // Function argument.
+  __ CallRuntime(kCompileFunctionRuntimeEntry);
+  __ lw(T3, Address(SP, (num_slots - 2) * kWordSize));  // Restore function.
+  __ lw(S5, Address(SP, (num_slots - 1) * kWordSize));  // Restore IC data.
+  __ addiu(SP, SP, Immediate(num_slots * kWordSize));
+  // T3: target function.
+  __ lw(T4, FieldAddress(T3, Function::code_offset()));
+  __ LeaveStubFrame();
+
+  __ Bind(&target_is_compiled);
+  // T4: target code.
+  __ lw(T3, FieldAddress(T4, Code::instructions_offset()));
+  __ AddImmediate(T3, Instructions::HeaderSize() - kHeapObjectTag);
+  __ jr(T3);
+  // Load arguments descriptor into S4.
+  __ delay_slot()->
+      lw(S4,  FieldAddress(S5, ICData::arguments_descriptor_offset()));
+}
+
+
 void StubCode::GenerateBreakpointRuntimeStub(Assembler* assembler) {
   __ Unimplemented("BreakpointRuntime stub");
 }
 
 
 //  RA: return address (Dart code).
-//  S4: Arguments descriptor array.
+//  S5: IC data (unoptimized static call).
 void StubCode::GenerateBreakpointStaticStub(Assembler* assembler) {
   __ TraceSimMsg("BreakpointStaticStub");
   // Create a stub frame as we are pushing some objects on the stack before
@@ -1812,13 +1882,13 @@
   __ EnterStubFrame();
   // Preserve arguments descriptor and make room for result.
   __ addiu(SP, SP, Immediate(-2 * kWordSize));
-  __ sw(S4, Address(SP, 1 * kWordSize));
+  __ sw(S5, Address(SP, 1 * kWordSize));
   __ LoadImmediate(TMP, reinterpret_cast<intptr_t>(Object::null()));
   __ sw(TMP, Address(SP, 0 * kWordSize));
   __ CallRuntime(kBreakpointStaticHandlerRuntimeEntry);
   // Pop code object result and restore arguments descriptor.
   __ lw(T0, Address(SP, 0 * kWordSize));
-  __ lw(S4, Address(SP, 1 * kWordSize));
+  __ lw(S5, Address(SP, 1 * kWordSize));
   __ addiu(SP, SP, Immediate(2 * kWordSize));
   __ LeaveStubFrame();
 
@@ -1826,6 +1896,8 @@
   // ensures that the call target is compiled.
   __ lw(T0, FieldAddress(T0, Code::instructions_offset()));
   __ AddImmediate(T0, Instructions::HeaderSize() - kHeapObjectTag);
+  // Load arguments descriptor into S4.
+  __ lw(S4, FieldAddress(S5, ICData::arguments_descriptor_offset()));
   __ jr(T0);
 }