-
Logged In: YES
user_id=256904
Manually triage found that the bug locates in the file
SymMatrix.cc, PU: HepMatrix operator*(const HepSymMatrix
&m1,const HepSymMatrix &m2)
Note the operator * is overloaded several times.
This is a Pro64 bug. Pro64 .13 has the same failures.
And it failed even at -O0. So triage is difficult.
2003-02-28 07:59:25 UTC in Open Research Compiler - Aurora
-
Description:
There's a compilation failure when building CLHEP
library (refer to bug #693490). If we substitute a good
Transform3D.o for the erroneous one, the library can be
made up (So only one compilation failure). But the test
of TestInversion_EXE failed (Result comparison failed.
The diff is given below. W/ GCC built CLHEP lib, the
result is correct. So the bug lies in the...
2003-02-27 06:47:37 UTC in Open Research Compiler - Aurora
-
This is an SWP bug. The following scheule is finalized
by SWP and the dependence between mov_t_pr and
cmp is ignored by SWP.
Attached is the bzip2.s generated by compiling bzip2.c
in Spec2000C/bzip2 with '-O3 -S' option. (I used
ORC1.0. Similar problems have been reported with
ORC1.1.)
>>Here is one of the problematic code segments:
>>...
2002-12-09 02:58:42 UTC in Open Research Compiler - Aurora
-
gap failed at peak mode after William's checkin
in Jun 1.
2002-07-12 10:36:04 UTC in Open Research Compiler - Aurora
-
Logged In: YES
user_id=256904
The probability of an edge on regional CFG may need to be
computed and maintained based on regional CFG, instead of
copied from global CFG.
Liu Yang, would you please take a look at this problem.
2002-05-22 02:55:10 UTC in Open Research Compiler - Aurora
-
There're such code like:
IPFEC_Enable_Region_Formation =
IPFEC_Enable_Region_Formation && CG_Enable_Ipfec_Phases
&& CG_opt_level > 1;
in Config_Ipfec_Flags() in cg.cxx.
when compile a PU, say PU 1, if skip option is set,
then CG_opt_level will be set to 1, and
IPFEC_Enable_Region_Formation will be set to FALSE.
That flag will keep to be FALSE for all PUs followed...
2002-03-26 09:41:00 UTC in Open Research Compiler - Aurora
-
Logged In: YES
user_id=256904
Sorry coming back to this problem so late.
The problem here is not how to handle phisical predicate
registers. For example:
(tn3) cmp.xx.xx tn1, tn2 = a, b
(tn1) br.cond
We have to query PRDB whether tn1 and tn2 are complementary
w/ respect to their home BB. If the answer is true, we can
negate the branch. Otherwise we have the choice to insert an...
2002-03-26 06:30:53 UTC in Open Research Compiler - Aurora
-
Logged In: YES
user_id=256904
Sorry, this's Gan Ge's item.
2002-03-21 03:21:27 UTC in Open Research Compiler - Aurora
-
Logged In: YES
user_id=256904
Further investigation show that the real reason is not
as specified in above. Currently, the implementation of
cflow optimizations takes such an approach:
- firstly take a snapshot of the flow graph.
keep the info in the BBINFO structures.
- do various optimizations on the CFG, including
reordering.
- modify branches according to the changes of the...
2002-03-19 07:04:50 UTC in Open Research Compiler - Aurora
-
An assertion failed when try to use the tracing
facility of CFLOW when doing freq reorder.
To reproduce, use:
orcc -show -keep -S -O3 -fomit-frame-pointer -fb_opt
fb.mid -fb_type=2 -fb_phase=4 deflate.c
-Wb,-OPT:skip_b=2:skip_a=2 -Wb,-tt47:0x40
Error info:
### Assertion failure at line 233 of
../../common/com/opcode.h:
### Compiler Error in file deflate.c during CFLOW
(second...
2002-03-07 01:38:07 UTC in Open Research Compiler - Aurora
