Export of internal Abseil changes.

--
eb6cc81ef7e89e10fc9df47418af93e22fd116d2 by Abseil Team <absl-team@google.com>:

Workaround clang bug https://bugs.llvm.org/show_bug.cgi?id=38289

PiperOrigin-RevId: 206006290

--
509e9829295bfc429b82de42f2e073c756ea5709 by Jon Cohen <cohenjon@google.com>:

Remove make_unique ambiguity when using gcc 4.9 in C++14 mode.

gcc 4.9.4 has __cplusplus at 201300L instead of 201402L when in C++14 mode, I guess indicating incomplete support.  Anyways, it causes a problem with this check as in c++14 mode in old gcc we were defining absl::make_unique when std::make_unique was present

PiperOrigin-RevId: 205886589
GitOrigin-RevId: eb6cc81ef7e89e10fc9df47418af93e22fd116d2
Change-Id: I9acf3f3d0fd3b0b46ae099821f3bf21b72c28b2b

absl/memory/memory.h

@@ -83,7 +83,11 @@ struct MakeUniqueResult<T[N]> {
 
 }  // namespace memory_internal
 
-#if __cplusplus >= 201402L || defined(_MSC_VER)
+// gcc 4.8 has __cplusplus at 201301 but doesn't define make_unique.  Other
+// supported compilers either just define __cplusplus as 201103 but have
+// make_unique (msvc), or have make_unique whenever __cplusplus > 201103 (clang)
+#if (__cplusplus > 201103L || defined(_MSC_VER)) && \
+    !(defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ == 8)
 using std::make_unique;
 #else
 // -----------------------------------------------------------------------------

absl/strings/internal/str_format/float_conversion.cc

@@ -153,7 +153,14 @@ void PrintExponent(int exp, char e, Buffer *out) {
 
 template <typename Float, typename Int>
 constexpr bool CanFitMantissa() {
-  return std::numeric_limits<Float>::digits <= std::numeric_limits<Int>::digits;
+  return
+#if defined(__clang__) && !defined(__SSE3__)
+      // Workaround for clang bug: https://bugs.llvm.org/show_bug.cgi?id=38289
+      // Casting from long double to uint64_t is miscompiled and drops bits.
+      (!std::is_same<Float, long double>::value ||
+       !std::is_same<Int, uint64_t>::value) &&
+#endif
+      std::numeric_limits<Float>::digits <= std::numeric_limits<Int>::digits;
 }
 
 template <typename Float>