[CUDA] Make unattributed constexpr functions implicitly host+device.

With this patch, by a constexpr function is implicitly host+device
unless:

 a) it's a variadic function (variadic functions are not allowed on the
    device side), or
 b) it's preceeded by a __device__ overload in a system header.

The restriction on overloading __host__ __device__ functions on the
basis of their CUDA attributes remains in place, but we use (b) to allow
us to define __device__ overloads for constexpr functions in cmath,
which would otherwise be __host__ __device__ and thus not overloadable.

You can disable this behavior with -fno-cuda-host-device-constexpr.

Reviewers: tra, rnk, rsmith

Subscribers: cfe-commits

Differential Revision: http://reviews.llvm.org/D18380

llvm-svn: 264964

1 files changed, 3 insertions, 0 deletions

diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index cc657ed5a3f..5bb036ab26e 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -1560,6 +1560,9 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   if (Args.hasArg(OPT_fcuda_allow_variadic_functions))
     Opts.CUDAAllowVariadicFunctions = 1;
 
+  if (Args.hasArg(OPT_fno_cuda_host_device_constexpr))
+    Opts.CUDAHostDeviceConstexpr = 0;
+
   if (Opts.ObjC1) {
     if (Arg *arg = Args.getLastArg(OPT_fobjc_runtime_EQ)) {
       StringRef value = arg->getValue();