Help in understanding ccp propagator

6 msg	current gcc trunk testsuite failure on cygwin: ...
1 msg	question about switch table
1 msg	Libjava is broken again

Help in understanding ccp propagator
\ Revital1 Eres (3 Jun 2007)
. \ Daniel Berlin (3 Jun 2007)
. . \ Revital1 Eres (4 Jun 2007)
. . . \ Daniel Berlin (4 Jun 2007)
. . . . \ Revital1 Eres (5 Jun 2007)
. . . . . \ Daniel Berlin (5 Jun 2007)
. . . . . . \ Revital1 Eres (12 Jun 2007)
. . . . . . . \ Daniel Berlin (13 Jun 2007)
. . . . . . . . \ Revital1 Eres (17 Jun 2007)
. . . . . . . . . \ Richard Guenther (18 Jun 2007)

2 msg	Redefinition of symbol ???
10 msg	When was decimal floating point added to gcc?
1 msg	gcc-4.3-20070601 is now available
2 msg	What is purpose of numbered variables??
7 msg	Predictive commoning miscompiles 482.sphinx3 in...
2 msg	DOC PATCH: attribute(alloc_size)
3 msg	signal unwind and fp state
1 msg	synchronous cancel and 'terminate called withou...
4 msg	__builtin_apply_args - GCC 4.1.1
6 msg	Optimizations for itanium
29 msg	Git repository with full GCC history
1 msg	Linux Foundation Travel Fund and GCC Summit
1 msg	cxx-reflection branch ?
6 msg	Access to symbol table??
1 msg	Debug info generation (target: Fr30) - GCC 4.1.1
17 msg	Very Fast: Directly Coded Lexical Analyzer

Subject:	Help in understanding ccp propagator
Group:	Gcc
From:	Revital1 Eres
Date:	3 Jun 2007

Hello,

I will greatly appreciate any suggestions regarding the following
problem I have with the ccp propagator. I am testing the new store
ccp patch which propagates constants by walking the virtual use-def
chain (http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00055.html) and I
encountered the following problem while testing tree_join.cc file which
is under libstdc++-v3 testsuite:

The final propagator replaces the right operand (D.65705_98) in the
following if condition with a constant zero which causes the program to be
aborted as i0_62 is not always zero:

# i0_62 = PHI <i0_5(29), 0(3)>
<L139>:;
D.61410.first = i0_62;
...
D.65705_98 = D.61410.first;
- if (D.65704_96 >= 0)
+ if (D.65704_96 >= D.65705_98)
goto <bb 11>;
else
goto <bb 14>;

Tracing the execution of the propagator it seems that the if statement
is first been propogated with zero after simulating the execution; but not
updated after the lattice value of the variables it depends on changed.

Here is the scenario of the propagator:

1) D.61410.first is been updated to zero:

19921 Visiting PHI node: i0_62 = PHI <i0_5(29), 0(3)>
19922 Argument #0 (29 -> 4 not executable)
19923
19924 Argument #1 (3 -> 4 executable)
19925 0 Value: CONSTANT 0
19926
19927 PHI node value: CONSTANT 0
19928
19929 Lattice value changed to CONSTANT 0. Adding SSA edges to worklist.
19930
19931 Visiting statement:
19932 D.61410.first = i0_62;
19933
19934 Lattice value changed to CONSTANT 0. Adding SSA edges to worklist.
19935

2) D.65705_98 is been upadted to zero:

20235 Visiting statement:
20236 D.65705_98 = D.61410.first;
20237
20238 Lattice value changed to CONSTANT 0. Adding SSA edges to worklist.
20239
20240 Visiting statement:
20241 if (D.65704_96 >= D.65705_98)
....

3) D.61410.first latice value changed to VARYING.

20360 Simulating statement (from ssa_edges): D.61410.first = i0_62;
20361
20362 Visiting statement:
20363 D.61410.first = i0_62;
20364
20365 Lattice value changed to VARYING. Adding SSA edges to worklist.

4) if D.65705_98 is been replaced to zero:

25422 Substituing values and folding statements
25423
25424 Folded statement Folded statement: if (D.65704_96 >= D.65705_98)
25425 into: if (D.65704_96 >= 0)

I am not sure why after the lattice value of D.61410.first has been
changed to VARYING D.65705_98 and the if statement is not been updated
as well.

Thanks,
Revital