AL causes CYC,SAACYC to run until maxoptcyc on prd.sac AWLF unit test
|
|
Bugzilla Link |
711 |
Created on |
May 14, 2010 20:25 |
Version |
svn |
OS |
Linux |
Architecture |
PC |
Attachments |
prd.sac |
Extended Description
Created an attachment (id=715)
source code to reproduce fault
It looks to me that AL has gone for a burton. sac2c prd.sac (and everything else...) runs for maxoptcyc in both CYC and SAACYC.
I took a look at the prdreverseAKD.sac unit test, and observed these
differences around cycle 23.
After esdcse (right before AL):
_al_2410 = _add_SxS_( _pinl_672__ea_584__icc_450, _ivexp_1865_aeb);
_al_2411 = _add_SxS_( _ivexp_1868_is, _al_2410);
After AL:
_al_2410 = _add_SxS_( _pinl_672__ea_584__icc_450, _ivexp_1865_aeb);
_al_2440 = _add_SxS_( _pinl_672__ea_584__icc_450, _ivexp_1865_aeb);
_al_2441 = _add_SxS_( _ivexp_1868_is, _al_2440);
_al_2411 = _al_2441;
Doesn't hardly seem worth it, eh? _al2410 is now dead code.
This is the behavior on 16836:MODIFIED, on rattler (not checked in).
It is NOT, however, the sole cause of the opt loop, because running with
-noal still runs for maxoptcyc cycles.
Build #16794:MODIFIED, on obelix, exhibits the same behavior.