[nvptx] Don't emit barriers for empty loops -- fix

[nvptx] Don't emit barriers for empty loops -- fix

When compiling an empty loop:
...
  long long v1;
  #pragma acc parallel num_gangs (640) num_workers(1) vector_length (128)
  #pragma acc loop
    for (v1 = 0; v1 < 20; v1 += 2)
        ;
...
the compiler emits two subsequent bar.syncs.  This triggers some bug on my
quadro m1200 (I'm assuming in the ptxas/JIT compiler) that hangs the testcase.

This patch works around the bug by doing an optimization: we detect that this is
an empty loop (a forked immediately followed by a joining), and don't emit the
barriers.

The patch does not include the test-case yet, since vector_length (128) is not
yet supported at this point.

2019-01-07  Tom de Vries  <tdevries@suse.de>

	PR target/85381
	* config/nvptx/nvptx.c (nvptx_process_pars): Don't emit barriers for
	empty loops.

From-SVN: r267630