Add DoubleToSmi optimistically, preventing boxing/unboxing of doubles and propagate smi-nessof the …

runtime/vm/intermediate_language_ia32.cc

Index: runtime/vm/intermediate_language_ia32.cc
===================================================================
--- runtime/vm/intermediate_language_ia32.cc	(revision 16185)
+++ runtime/vm/intermediate_language_ia32.cc	(working copy)
@@ -2315,6 +2315,31 @@
 }
 
 
+LocationSummary* DoubleToSmiInstr::MakeLocationSummary() const {
+  const intptr_t kNumInputs = 1;
+  const intptr_t kNumTemps = 0;
+  LocationSummary* result = new LocationSummary(
+      kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  result->set_in(0, Location::RequiresXmmRegister());
+  result->set_out(Location:: Location::RequiresRegister());
+  return result;
+}
+
+
+void DoubleToSmiInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  Label* deopt = compiler->AddDeoptStub(deopt_id(), kDeoptDoubleToSmi);
+  Register result = locs()->out().reg();
+  XmmRegister value = locs()->in(0).xmm_reg();
+  __ cvttsd2si(result, value);

[Vyacheslav Egorov (Google), 2012/12/17 12:53:44]

This code is almost exactly repeating code in the Intrinsifier::Double_toInt. 

I hope we will find a way to remove this duplication soon.

[srdjan, 2012/12/18 00:12:47]

Code duplication of short patterns is OK. I think it is more readable and
actually easier to maintain, than calling a 'macro' assembler function that does
something.

[Vyacheslav Egorov (Google), 2012/12/18 11:45:10]

Well, the discrepancy in x64 demonstrates that it is not easier to maintain at
all. However at the moment I see no easy alternative. 

+  // Overflow is signalled with minint.
+  Label do_call, done;
+  // Check for overflow and that it fits into Smi.
+  __ cmpl(result, Immediate(0xC0000000));
+  __ j(NEGATIVE, deopt);
+  __ SmiTag(result);
+}
+
+
 LocationSummary* PolymorphicInstanceCallInstr::MakeLocationSummary() const {
   return MakeCallSummary();
 }