WLIR fails to remove WL from LoopFun, testforAKD.sac is 7X slower. WLBSC?
|
|
Bugzilla Link |
659 |
Created on |
Jan 11, 2010 20:15 |
Resolution |
INVALID |
Resolved on |
Jan 11, 2010 20:19 |
Version |
svn |
OS |
Linux |
Architecture |
PC |
Attachments |
testforAKD.sac |
Extended Description
Created an attachment (id=649)
Source code to reproduce fault
sac/apex/testforAKD/testforAKD.sac is running about 7X slower, in part because of this WL within a LoopFun:
_pinl_1545__wlbsc_1045_sc_e = _idx_shape_sel_( 0, r_4);
_pinl_1546__wlbsc_1046_sc_bound = [ _pinl_1545__wlbsc_1045_sc_e ];
_wlbsc_2847_sc_bound = [ 0 ];
_pinl_1539_z = with {
(_wlbsc_2847_sc_bound <= _pinl_1535_iv=[_pinl_1540__eat_184] < _pinl_1546__wlbsc_1046_sc_bound genwidth [ _pinl_1545__wlbsc_1045_sc_e ])
{
_ivexi_2739 = _attachextrema_( _pinl_1535_iv, _wlur_2142, _pinl_1546__wlbsc_1046_sc_bound);
_dup_2747__pinl_1561__flat_95 = _sel_VxA_( _ivexi_2739, r_4);
_dup_2748__pinl_1550__flat_213 = _add_SxS_( _dup_2747__pinl_1561__flat_95, _pinl_1504__flat_95);
} : _dup_2748__pinl_1550__flat_213 ;
} :
genarray( _pinl_1546__wlbsc_1046_sc_bound, _pinl_1499__flat_94);
r_4 is a parameter to the LoopFun, so this WL should have been moved outside
the LoopFun.
Marked as "major" because of the magnitude of the slowdown.