[Bug]: arm_neon.h inclusion causing issues about abseil-cpp

[aweits]

As a clarification, the primary issue here is that when nvcc calls the c++ preprocessor for .cu files, arm_neon.h gets included and that brings along definitions of types that only the respective c++ compiler will understand. nvcc then calls cudafe++ on the resultant file and errors out. This is confusing and the output might lead folks to misunderstand where the error is originating from. Running nvcc with the --dryrun or --verbose flag will show all the individual steps being run in the background, and incrementally executing these steps allows one to see what's going on.

[aweits]

The above is probably overly broad (!defined(__NVCC__)), and should be scoped to this (!(defined(__NVCC__) && defined(__CUDACC__))) per https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html and https://llvm.org/docs/CompileCudaWithLLVM.html

diff --git a/absl/base/config.h b/absl/base/config.h
index 5fa9f0ef..503a77af 100644
--- a/absl/base/config.h
+++ b/absl/base/config.h
@@ -962,7 +962,7 @@ static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != 'h' ||
 // https://llvm.org/docs/CompileCudaWithLLVM.html#detecting-clang-vs-nvcc-from-code
 #ifdef ABSL_INTERNAL_HAVE_ARM_NEON
 #error ABSL_INTERNAL_HAVE_ARM_NEON cannot be directly set
-#elif defined(__ARM_NEON) && !defined(__CUDA_ARCH__)
+#elif defined(__ARM_NEON) && !(defined(__NVCC__) && defined(__CUDACC__))
 #define ABSL_INTERNAL_HAVE_ARM_NEON 1
 #endif