Add float-to-half (binary16) conversion functions.

Based on code by Fabian Giesen at
https://fgiesen.wordpress.com/2012/03/28/half-to-float-done-quic/.

These will be needed for creating binary16 textures from floating point data.

BUG=skia:3103

Review URL: https://codereview.chromium.org/760753003
diff --git a/src/core/SkHalf.cpp b/src/core/SkHalf.cpp
new file mode 100644
index 0000000..0db979b
--- /dev/null
+++ b/src/core/SkHalf.cpp
@@ -0,0 +1,97 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "SkHalf.h"
+#include "SkFloatBits.h"
+
+uint16_t halfMantissa(SkHalf h) {
+    return h & 0x03ff;
+}
+
+uint16_t halfExponent(SkHalf h) {
+    return (h >> 10) & 0x001f;
+}
+
+uint16_t halfSign(SkHalf h) {
+    return h >> 15;
+}
+
+union FloatUIntUnion {
+    uint32_t fUInt;    // this must come first for the initializations below to work
+    float    fFloat;
+};
+
+// based on Fabien Giesen's float_to_half_fast3()
+// see https://gist.github.com/rygorous/2156668
+SkHalf SkFloatToHalf(float f) {
+    static const uint32_t f32infty = { 255 << 23 };
+    static const uint32_t f16infty = { 31 << 23 };
+    static const FloatUIntUnion magic = { 15 << 23 };
+    static const uint32_t sign_mask = 0x80000000u;
+    static const uint32_t round_mask = ~0xfffu;
+    SkHalf o = 0;
+    
+    FloatUIntUnion floatUnion;
+    floatUnion.fFloat = f;
+    
+    uint32_t sign = floatUnion.fUInt & sign_mask;
+    floatUnion.fUInt ^= sign;
+    
+    // NOTE all the integer compares in this function can be safely
+    // compiled into signed compares since all operands are below
+    // 0x80000000. Important if you want fast straight SSE2 code
+    // (since there's no unsigned PCMPGTD).
+    
+    // Inf or NaN (all exponent bits set)
+    if (floatUnion.fUInt >= f32infty)
+        // NaN->qNaN and Inf->Inf
+        o = (floatUnion.fUInt > f32infty) ? 0x7e00 : 0x7c00;
+    // (De)normalized number or zero
+    else {
+        floatUnion.fUInt &= round_mask;
+        floatUnion.fFloat *= magic.fFloat;
+        floatUnion.fUInt -= round_mask;
+        // Clamp to signed infinity if overflowed
+        if (floatUnion.fUInt > f16infty) {
+            floatUnion.fUInt = f16infty;
+        }
+        
+        o = floatUnion.fUInt >> 13; // Take the bits!
+    }
+    
+    o |= sign >> 16;
+    return o;
+}
+
+// based on Fabien Giesen's half_to_float_fast2()
+// see https://fgiesen.wordpress.com/2012/03/28/half-to-float-done-quic/
+float SkHalfToFloat(SkHalf h) {
+    static const FloatUIntUnion magic = { 126 << 23 };
+    FloatUIntUnion o;
+    
+    if (halfExponent(h) == 0)
+    {
+        // Zero / Denormal
+        o.fUInt = magic.fUInt + halfMantissa(h);
+        o.fFloat -= magic.fFloat;
+    }
+    else
+    {
+        // Set mantissa
+        o.fUInt = halfMantissa(h) << 13;
+        // Set exponent
+        if (halfExponent(h) == 0x1f)
+            // Inf/NaN
+            o.fUInt |= (255 << 23);
+        else
+            o.fUInt |= ((127 - 15 + halfExponent(h)) << 23);
+    }
+    
+    // Set sign
+    o.fUInt |= (halfSign(h) << 31);
+    return o.fFloat;
+}