Implement Math.f16round

[littledivy]

Math.f16round can be implemented with the WebAssembly half-precision proposal which adds f16 type and operations.

This hasn't landed in V8 yet. I haven't tested this patch but it should work with this CL

diff --git a/compiler/builtins.js b/compiler/builtins.js
index 2645859..4272046 100644
--- a/compiler/builtins.js
+++ b/compiler/builtins.js
@@ -362,6 +362,21 @@ export const BuiltinFuncs = function() {
     ]
   };

+  this.__Math_f16round = {
+    floatOnly: true,
+    params: [ valtypeBinary ],
+    locals: [],
+    returns: [ valtypeBinary ],
+    returnType: TYPES.number,
+    wasm: [
+      [ Opcodes.local_get, 0 ],
+      // f32x4.demote_f64x2
+      [ Opcodes.f32_demote_f64 ],
+      // f16x8.demote_f32x4_zero
+      [ Opcodes.f16_demote_f32_s ],
+    ]
+  };
+
   // todo: this does not overflow correctly
   this.__Math_imul = {
     floatOnly: true,
@@ -1113,4 +1128,4 @@ export const BuiltinFuncs = function() {
   };

   PrecompiledBuiltins.BuiltinFuncs.call(this);
-};
\ No newline at end of file
+};
diff --git a/compiler/wasmSpec.js b/compiler/wasmSpec.js
index cd1d598..e712375 100644
--- a/compiler/wasmSpec.js
+++ b/compiler/wasmSpec.js
@@ -197,6 +197,8 @@ export const Opcodes = {
   i32_trunc_sat_f64_s: [ 0xfc, 0x02 ],
   i32_trunc_sat_f64_u: [ 0xfc, 0x03 ],

+  f16_demote_f32_s: [0xfd, 0x14d],
+
   memory_init: [ 0xfc, 0x08 ],
   data_drop: [ 0xfc, 0x09 ],
   memory_copy: [ 0xfc, 0x0a ],
@@ -235,4 +237,4 @@ export const ValtypeSize = {
   // special
   i8: 1,
   i16: 2
-};
\ No newline at end of file
+};