Index: clang/lib/Headers/__clang_cuda_cmath.h
===================================================================
--- clang/lib/Headers/__clang_cuda_cmath.h
+++ clang/lib/Headers/__clang_cuda_cmath.h
@@ -131,15 +131,6 @@
 __DEVICE__ float log(float __x) { return ::logf(__x); }
 __DEVICE__ float log10(float __x) { return ::log10f(__x); }
 __DEVICE__ float modf(float __x, float *__iptr) { return ::modff(__x, __iptr); }
-__DEVICE__ float nexttoward(float __from, double __to) {
-  return __builtin_nexttowardf(__from, __to);
-}
-__DEVICE__ double nexttoward(double __from, double __to) {
-  return __builtin_nexttoward(__from, __to);
-}
-__DEVICE__ float nexttowardf(float __from, double __to) {
-  return __builtin_nexttowardf(__from, __to);
-}
 __DEVICE__ float pow(float __base, float __exp) {
   return ::powf(__base, __exp);
 }
@@ -157,6 +148,10 @@
 __DEVICE__ float tan(float __x) { return ::tanf(__x); }
 __DEVICE__ float tanh(float __x) { return ::tanhf(__x); }
 
+// Notably missing above are nexttoward and nextafter.  We omit them because
+// libdevice doesn't provide an implementation, and we don't want to be in the
+// business of implementing tricky libm functions in this header.
+
 // Now we've defined everything we promised we'd define in
 // __clang_cuda_math_forward_declares.h.  We need to do two additional things to
 // fix up our math functions.
@@ -252,7 +247,6 @@
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_1(long, lrint)
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_1(long, lround)
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_1(double, nearbyint);
-__CUDA_CLANG_FN_INTEGER_OVERLOAD_2(double, nextafter);
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_2(double, pow);
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_2(double, remainder);
 __CUDA_CLANG_FN_INTEGER_OVERLOAD_1(double, rint);
@@ -295,13 +289,6 @@
   return std::ldexp((double)__x, __exp);
 }
 
-template <typename __T>
-__DEVICE__ typename __clang_cuda_enable_if<std::numeric_limits<__T>::is_integer,
-                                           double>::type
-nexttoward(__T __from, double __to) {
-  return std::nexttoward((double)__from, __to);
-}
-
 template <typename __T1, typename __T2>
 __DEVICE__ typename __clang_cuda_enable_if<
     std::numeric_limits<__T1>::is_specialized &&
@@ -387,8 +374,6 @@
 using ::lrint;
 using ::lround;
 using ::nearbyint;
-using ::nextafter;
-using ::nexttoward;
 using ::pow;
 using ::remainder;
 using ::remquo;
@@ -455,9 +440,6 @@
 using ::lroundf;
 using ::modff;
 using ::nearbyintf;
-using ::nextafterf;
-using ::nexttowardf;
-using ::nexttowardf;
 using ::powf;
 using ::remainderf;
 using ::remquof;
Index: clang/lib/Headers/__clang_cuda_math_forward_declares.h
===================================================================
--- clang/lib/Headers/__clang_cuda_math_forward_declares.h
+++ clang/lib/Headers/__clang_cuda_math_forward_declares.h
@@ -147,11 +147,6 @@
 __DEVICE__ float nanf(const char *);
 __DEVICE__ double nearbyint(double);
 __DEVICE__ float nearbyint(float);
-__DEVICE__ double nextafter(double, double);
-__DEVICE__ float nextafter(float, float);
-__DEVICE__ double nexttoward(double, double);
-__DEVICE__ float nexttoward(float, double);
-__DEVICE__ float nexttowardf(float, double);
 __DEVICE__ double pow(double, double);
 __DEVICE__ double pow(double, int);
 __DEVICE__ float pow(float, float);
@@ -185,6 +180,10 @@
 __DEVICE__ double trunc(double);
 __DEVICE__ float trunc(float);
 
+// Notably missing above are nexttoward and nextafter, which we don't define on
+// the device side, because libdevice doesn't give us an implementation, and we
+// don't want to be in the business of writing one ourselves.
+
 // We need to define these overloads in exactly the namespace our standard
 // library uses (including the right inline namespace), otherwise they won't be
 // picked up by other functions in the standard library (e.g. functions in
@@ -254,8 +253,6 @@
 using ::nan;
 using ::nanf;
 using ::nearbyint;
-using ::nextafter;
-using ::nexttoward;
 using ::pow;
 using ::remainder;
 using ::remquo;