In my recent patch for tile clauses, I noticed some tile tests didn't appear to
fail, even though the loops they tiled were ill formed. I figured I'd find out
why in the fullness of time.
Didn't take long for time to be full. The omp_for parsing routines had no
knowledge of tile, so did not checking. This patch rectifies that.
I did notice a couple of nits in the c++ parser, also fixed here
1) we'd complain about a missing inner loop twice, which is a bit repetative.
Fixed by only emitting the missing for token error, if we didn't already emit an
error.
2) 'do ... while (1)' is an odd way to write a non-terminating loop. Changed to
'for (;;) ...' I guess the loop condition was originally something other than '1'.
applied to gomp4
nathan
