TLD: missing TLD macros when accessing upcrt_forall_control

[nenadv]

After Steven's commits for TLD fixes of #100, ... I get only one failure on intrepid test suite:

[intrepid/test10]   0sec  20140725_110545  FAILED (CRASH=SIGABRT/NEW)           
commandline: [ /eng/upc/dev/nenad/clang-upc-pthreads/bld/upcr-packed-dbg/dbg_cupc2c/upcrun -q -n 4  -p 4 ./test10 ]
PassExpr: passed                                                                
FailExpr: rror                                                                  
--- App stdout ---                                                              
--- App stderr ---                                                              
test10: error at element [0,0]. Does not have affinity with thread 3, inside nested upc_forall.
*** Caught a fatal signal: SIGABRT(6) on node 0/1                               
test10: error at element [0,0]. Does not have affinity with thread 2, inside nested upc_forall.
test10: error at element [0,1]. Does not have affinity with thread 0, inside nested upc_forall.
NOTICE: Before reporting bugs, run with GASNET_BACKTRACE=1 in the environment to generate a backtrace.
*** Caught a fatal signal: SIGABRT(6) on node 0/1 

I am running the full test suite now.

[PHHargrove]

The failure within a nested forall could be related to making the state which records the nesting into TLD (or failing to do so). Without looking at code I am guessing the accesses to the control variable in the transformed forall are not using the TLD access macros.

FYI: I am also retesting this pthreads support today.

[PHHargrove]

Without looking at code I am guessing the accesses to the control variable in the transformed forall are not using the TLD access macros.

That guess has been confirmed:

$ grep upcrt_forall_control test10.trans.c
int32_t UPCR_TLD_DEFINE_TENTATIVE(upcrt_forall_control, 4, 4);
            if (upcrt_forall_control) {
                upcrt_forall_control = 1;
                upcrt_forall_control = 0;

[PHHargrove]

There appear to be many tests in the full suite that fail due to the lack of TLD macros when accessing upcrt_forall_control. That omission leads both to nested forall's which don't know they are nested and non-nested foralls which erroneously execute as if they are nested.

Use of upc_forall is sufficiently pervasive in the test suite that I cannot (yet) definitively determine if any of the numerous runtime failures seen in a full harness run are not due to this particular bug.

[PHHargrove]

I have a possible fix in testing now.

[PHHargrove]

With the fix mentioned in my previous comment, the number of "NEW" runtime failures drops from 78 to just 20.

Fixed in e5008aa