Make ptx barrier work irrespective of the cl_mem_fence_flags

This generates a "bar.sync 0” instruction, which not only causes the
threads to wait, but does acts as a memory fence, as required by
OpenCL. The fence does not differentiate between local and global
memory. Unfortunately, there is no similar instruction which does
not include a memory fence. Hence, we cannot optimize the case
where neither CLK_LOCAL_MEM_FENCE nor CLK_GLOBAL_MEM_FENCE is
passed.

llvm-svn: 315228