[wasm] Detect unrepresentability in the float32-to-int32 conversion correctly on arm.

src/compiler/arm64/code-generator-arm64.cc

Index: src/compiler/arm64/code-generator-arm64.cc
diff --git a/src/compiler/arm64/code-generator-arm64.cc b/src/compiler/arm64/code-generator-arm64.cc
index 280a53712b70f88afb9e7cb0ef202771c42f8631..ef0344e16269c0f40561f88eb26047debe5b63fa 100644
--- a/src/compiler/arm64/code-generator-arm64.cc
+++ b/src/compiler/arm64/code-generator-arm64.cc
@@ -1323,12 +1323,21 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     case kArm64Float32ToInt32:
       __ Fcvtzs(i.OutputRegister32(), i.InputFloat32Register(0));
+      // Avoid INT32_MAX as an overflow indicator and use INT32_MIN instead,
+      // because INT32_MIN allows easier out-of-bounds detection.
+      __ Cmn(i.OutputRegister32(), 1);
+      __ Csinc(i.OutputRegister32(), i.OutputRegister32(), i.OutputRegister32(),
+               vc);
       break;
     case kArm64Float64ToInt32:
       __ Fcvtzs(i.OutputRegister32(), i.InputDoubleRegister(0));
       break;
     case kArm64Float32ToUint32:
       __ Fcvtzu(i.OutputRegister32(), i.InputFloat32Register(0));
+      // Avoid UINT32_MAX as an overflow indicator and use 0 instead,
+      // because 0 allows easier out-of-bounds detection.
+      __ Cmn(i.OutputRegister32(), 1);
+      __ Adc(i.OutputRegister(), i.OutputRegister(), Operand(0));

[Rodolph Perfetta (ARM), 2016/06/29 15:42:18]

i.OutputRegister32(), I made the mistake in my previous suggestion.

[ahaas, 2016/06/29 15:49:53]

Done. Thanks for looking at it so carefully again.

       break;
     case kArm64Float64ToUint32:
       __ Fcvtzu(i.OutputRegister32(), i.InputDoubleRegister(0));