[turbofan] IA: Float64ToUint32 supports mem operand

src/compiler/ia32/code-generator-ia32.cc

Index: src/compiler/ia32/code-generator-ia32.cc
diff --git a/src/compiler/ia32/code-generator-ia32.cc b/src/compiler/ia32/code-generator-ia32.cc
index 3a3fd795d8d38cd164fc68e7f6129265548d05f5..a0d30e6ba8038b79527f1da11748388c65896682 100644
--- a/src/compiler/ia32/code-generator-ia32.cc
+++ b/src/compiler/ia32/code-generator-ia32.cc
@@ -302,11 +302,17 @@ void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
       break;
     case kSSEFloat64ToUint32: {
       XMMRegister scratch = xmm0;
+      Label done;
       __ Move(scratch, -2147483648.0);
-      // TODO(turbofan): IA32 SSE subsd() should take an operand.
-      __ addsd(scratch, i.InputDoubleRegister(0));
+      __ addsd(scratch, i.InputOperand(0));
       __ cvttsd2si(i.OutputRegister(), scratch);
       __ add(i.OutputRegister(), Immediate(0x80000000));
+      __ j(not_carry, &done, Label::kNear);

[titzer, 2014/09/18 14:03:06]

What does this extra condition here check?

Note that this the implementation of ChangeFloat64ToUint32, which explicitly
assumes that all of its inputs lie in the range [0...2^32-1] and are integers.

[Weiliang, 2014/09/18 14:11:01]

cvttsd2si truncates the result by rounding toward zero. So when changing the
range from [0...2^32-1] to [-2^31, 2^31-1], we need to compensate the result.

+      __ cvtsi2sd(scratch, i.OutputRegister());
+      __ ucomisd(scratch, i.InputOperand(0));
+      __ j(equal, &done, Label::kNear);
+      __ sub(i.OutputRegister(), Immediate(1));
+      __ bind(&done);
       break;
     }
     case kSSEInt32ToFloat64: