[CUDA] Use activemask.b32 instruction to implement __activemask w/ CUDA-9.2+

[tra]

Summary

vote.ballot instruction is gone in recent CUDA versions and
vote.sync.ballot can not be used because it needs a thread mask parameter.
Fortunately PTX 6.2 (introduced with CUDA-9.2) provides activemask.b32
instruction for this.

clang/lib/Headers/__clang_cuda_intrinsics.h

 inline __device__ int __uni_sync(unsigned int mask, int pred) {
   return __nvvm_vote_uni_sync(mask, pred);
 }
 
 inline __device__ unsigned int __ballot_sync(unsigned int mask, int pred) {
   return __nvvm_vote_ballot_sync(mask, pred);
 }
 
-inline __device__ unsigned int __activemask() { return __nvvm_vote_ballot(1); }
+inline __device__ unsigned int __activemask() {
+#if CUDA_VERSION < 9020
+  return __nvvm_vote_ballot(1);
+#else
+  unsigned int mask;
+  asm volatile("activemask.b32 %0;" : "=r"(mask));
+  return mask;
+#endif
+}
 
 inline __device__ unsigned int __fns(unsigned mask, unsigned base, int offset) {
   return __nvvm_fns(mask, base, offset);
 }
 
 #endif // !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 300
 
 // Define __match* builtins CUDA-9 headers expect to see.