Improve saturated_cast performance

base/numerics/safe_conversions.h

Index: base/numerics/safe_conversions.h
diff --git a/base/numerics/safe_conversions.h b/base/numerics/safe_conversions.h
index 66b278ff3d3e41b3ce33f1d56bea183f12dc7b61..99cbef4ca814c669514c0a206e6162458ed8c8e5 100644
--- a/base/numerics/safe_conversions.h
+++ b/base/numerics/safe_conversions.h
@@ -110,35 +110,35 @@ namespace internal {
 // that the specified numeric conversion will saturate by default rather than
 // overflow or underflow, and NaN assignment to an integral will return 0.
 // All boundary condition behaviors can be overriden with a custom handler.
+template <template <typename>
+          class SaturationHandler = SaturatedCastDefaultHandler,
+          typename Dst,
+          typename Src>
+constexpr Dst saturated_cast_impl(const Src value,
+                                  const RangeConstraint constraint) {
+  return constraint.IsValid()
+             ? static_cast<Dst>(value)
+             : (constraint.IsOverflow()
+                    ? SaturationHandler<Dst>::HandleOverflow()
+                    // Skip this check for integral Src, which cannot be NaN.
+                    : (std::is_integral<Src>::value || constraint.IsUnderflow()
+                           ? SaturationHandler<Dst>::HandleUnderflow()
+                           : SaturationHandler<Dst>::HandleNaN()));
+}
+
+// saturated_cast<> is analogous to static_cast<> for numeric types, except
+// that the specified numeric conversion will saturate by default rather than
+// overflow or underflow, and NaN assignment to an integral will return 0.
+// All boundary condition behaviors can be overriden with a custom handler.
 template <typename Dst,
           template <typename>
           class SaturationHandler = SaturatedCastDefaultHandler,
           typename Src>
 constexpr Dst saturated_cast(Src value) {
-  static_assert(
-      SaturationHandler<Dst>::lowest() < SaturationHandler<Dst>::max(), "");
-  // While this looks like a lot of code, it's all constexpr and all but
-  // one variable are compile-time constants (enforced by a static_assert).
-  // So, it should evaluate to the minimum number of comparisons required
-  // for the range check, which is 0-3, depending on the exact source and
-  // destination types, and whatever custom range is specified.
   using SrcType = typename UnderlyingType<Src>::type;
-  return IsGreaterOrEqual<SrcType, Dst>::Test(
-             value, NarrowingRange<Dst, SrcType, SaturationHandler>::lowest())
-             ? (IsLessOrEqual<SrcType, Dst>::Test(
-                    value,
-                    NarrowingRange<Dst, SrcType, SaturationHandler>::max())
-                    ? static_cast<Dst>(value)
-                    : SaturationHandler<Dst>::HandleOverflow())
-             // This last branch is a little confusing. It's specifically to
-             // catch NaN when converting from float to integral.
-             : (std::is_integral<SrcType>::value ||
-                        std::is_floating_point<Dst>::value ||
-                        IsLessOrEqual<SrcType, Dst>::Test(
-                            value, NarrowingRange<Dst, SrcType,
-                                                  SaturationHandler>::max())
-                    ? SaturationHandler<Dst>::HandleUnderflow()
-                    : SaturationHandler<Dst>::HandleNaN());
+  return saturated_cast_impl<SaturationHandler, Dst>(
+      static_cast<SrcType>(value),
+      DstRangeRelationToSrcRange<Dst, SaturationHandler, SrcType>(value));
 }
 
 // strict_cast<> is analogous to static_cast<> for numeric types, except that