[OPENMP][NVPTX]Fix dynamic scheduling.

Summary:
Previous implementation may cause the runtime crash when the number of
teams is > 1024. Patch fixes this problem + reduces number of the atomic
operations by 32 times.

Reviewers: grokos, gtbercea, kkwli0

Subscribers: guansong, jfb, openmp-commits, caomhin

Differential Revision: https://reviews.llvm.org/D56332

git-svn-id: https://llvm.org/svn/llvm-project/openmp/trunk@350524 91177308-0d34-0410-b5e6-96231b3b80d8