[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Comment #33 from pinskia at gcc dot gnu dot org 2005-10-04 12:54 --- Fixed. -- pinskia at gcc dot gnu dot org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Comment #32 from cvs-commit at gcc dot gnu dot org 2005-10-04 05:57 --- Subject: Bug 23049 CVSROOT:/cvs/gcc Module name:gcc Changes by: [EMAIL PROTECTED] 2005-10-04 05:57:38 Modified files: gcc: ChangeLog tree-ssa-dom.c gcc/doc: tree-ssa.texi gcc/testsuite : ChangeLog gcc/testsuite/gcc.dg: ucnid-4.c Added files: gcc/testsuite/gcc.dg: pr23049.c Log message: 2005-10-05 Steven Bosscher <[EMAIL PROTECTED]> gcc/ PR tree-optimization/23049 * tree-ssa-dom.c (thread_across_edge): Make sure that the condition of a COND_EXPR is folded before calling fold on the whole rhs of a conditional assignment. * doc/tree-ssa.texi: Update the GIMPLE grammar for a valid rhs to document that a COND_EXPR may appear there. testsuite/ * gcc.dg/pr23049.c: New test. * gcc.dg/ucnid-4.c: Fix test. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.10073&r2=2.10074 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-dom.c.diff?cvsroot=gcc&r1=2.128&r2=2.129 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/doc/tree-ssa.texi.diff?cvsroot=gcc&r1=1.31&r2=1.32 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.6130&r2=1.6131 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/pr23049.c.diff?cvsroot=gcc&r1=NONE&r2=1.1 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/ucnid-4.c.diff?cvsroot=gcc&r1=1.3&r2=1.4 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-26 13:20 --- This is also testsuite/gcc.c-torture/execute/builtin-bitops-1.c failure with "-ftree-vectorize -maltivec" on powerpc-darwin, see http://gcc.gnu.org/ml/gcc-testresults/2005-09/msg01199.html http://gcc.gnu.org/ml/gcc-testresults/2005-09/msg01201.html -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From law at redhat dot com 2005-09-20 15:11 --- Subject: Re: [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x On Sun, 2005-09-18 at 15:59 +, steven at gcc dot gnu dot org wrote: > --- Additional Comments From steven at gcc dot gnu dot org 2005-09-18 > 15:59 --- > Actually fold_stmt has nothing to do with it. It is fold (cached_lhs) in > trhead_across_edge that is the problem here. Jeff, what do you think about > this patch? Presumably COND_EXPR on the RHS of a statement is the only special case that we need to worry about. I can't immediately think of others. Seems pretty reasonable, in the case were the COND_EXPR_COND does not fold, you could safely set pre_fold_lhs to TREE_OPERAND (stmt, 1) which would allow us to catch cases where the THEN/ELSE clauses of the COND_EXPR are equivalent due to substitutions. But I doubt it's terribly important. You might use pre_fold_expr rather than pre_fold_lhs which is a little confusing since we're looking at the RHS of a statement :-) jeff -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From steven at gcc dot gnu dot org 2005-09-18
15:59 ---
Actually fold_stmt has nothing to do with it. It is fold (cached_lhs) in
trhead_across_edge that is the problem here. Jeff, what do you think about
this patch?
Index: tree-ssa-dom.c
===
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-dom.c,v
retrieving revision 2.127
diff -u -3 -p -r2.127 tree-ssa-dom.c
--- tree-ssa-dom.c 2 Aug 2005 00:12:39 - 2.127
+++ tree-ssa-dom.c 18 Sep 2005 15:57:22 -
@@ -701,7 +701,7 @@ thread_across_edge (struct dom_walk_data
else
{
/* Copy the operands. */
- tree *copy;
+ tree *copy, prefold_lhs;
ssa_op_iter iter;
use_operand_p use_p;
unsigned int num, i = 0;
@@ -725,12 +725,33 @@ thread_across_edge (struct dom_walk_data
/* Try to fold/lookup the new expression. Inserting the
expression into the hash table is unlikely to help
-simplify anything later, so just query the hashtable. */
- cached_lhs = fold (TREE_OPERAND (stmt, 1));
- if (TREE_CODE (cached_lhs) != SSA_NAME
- && !is_gimple_min_invariant (cached_lhs))
- cached_lhs = lookup_avail_expr (stmt, false);
+simplify anything later, so just query the hashtable.
+Sadly, we have to handle conditional assignments specially
+here, because fold expects all the operands of an expression
+to be folded before the expression itself is folded, but we
+can't just substitute the folded condition here. */
+ if (TREE_CODE (TREE_OPERAND (stmt, 1)) == COND_EXPR)
+ {
+ tree cond = COND_EXPR_COND (TREE_OPERAND (stmt, 1));
+ cond = fold (cond);
+ if (cond == boolean_true_node)
+ prefold_lhs = COND_EXPR_THEN (TREE_OPERAND (stmt, 1));
+ else if (cond == boolean_false_node)
+ prefold_lhs = COND_EXPR_ELSE (TREE_OPERAND (stmt, 1));
+ else
+ prefold_lhs = NULL;
+ }
+ else
+ prefold_lhs = TREE_OPERAND (stmt, 1);
+
+ if (prefold_lhs)
+ {
+ cached_lhs = fold (prefold_lhs);
+ if (TREE_CODE (cached_lhs) != SSA_NAME
+ && !is_gimple_min_invariant (cached_lhs))
+ cached_lhs = lookup_avail_expr (stmt, false);
+ }
/* Restore the statement's original uses/defs. */
i = 0;
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From steven at gcc dot gnu dot org 2005-09-18
15:11 ---
Smaller test case:
static unsigned short int crc_table[256];
void AC3_encode_init(void)
{
unsigned int c, n;
for (n = 0; n < 256; n++)
{
c = n << 8;
if (c & 16384)
c = c ^ 49155;
crc_table[n] = c;
}
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From steven at gcc dot gnu dot org 2005-09-18 10:41 --- Re. comment #24 -- I guess Devang didn't update fold_stmt and similar functions for the (still undocumented) modification to GIMPLE to make >> valid. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From law at redhat dot com 2005-09-18 00:37 --- Subject: Re: [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x On Sat, 2005-09-17 at 19:38 +, pinskia at physics dot uc dot edu wrote: > > --- Additional Comments From rguenth at tat dot physik dot > > uni-tuebingen dot de 2005-09-17 19:31 --- > > Please fix the caller who is not folding the condition in the first place > > instead. > > we have "a = b == 0 ? 1 : 2;" before in DOM and then we replace b with > the value that DOM thinks is correct and then we have "0 == 0 ? 1 : 2;" > which is then > passed to fold Is there some reason fold_stmt (which should have been called after the replacement by DOM) didn't fold the conditional? jeff -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From steven at gcc dot gnu dot org 2005-09-17 23:53 --- It is hardly a DOM bug. More an oversight. Probably easily fixed. -- What|Removed |Added AssignedTo|unassigned at gcc dot gnu |steven at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED Last reconfirmed|2005-07-24 15:37:30 |2005-09-17 23:53:02 date|| http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
-- What|Removed |Added BugsThisDependsOn||23936 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-17 23:09 --- I am not going to fix a DOM bug. See http://gcc.gnu.org/ml/gcc-patches/2005-09/msg01084.html for more discussion about this bug. Though I wonder if I would not hit this bug in my tree combiner but I will handle that when I get there. -- What|Removed |Added AssignedTo|pinskia at gcc dot gnu dot |unassigned at gcc dot gnu |org |dot org Status|ASSIGNED|NEW http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at physics dot uc dot edu 2005-09-17 19:38 --- Subject: Re: [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x > > > --- Additional Comments From rguenth at tat dot physik dot uni-tuebingen > dot de 2005-09-17 19:31 --- > Please fix the caller who is not folding the condition in the first place > instead. we have "a = b == 0 ? 1 : 2;" before in DOM and then we replace b with the value that DOM thinks is correct and then we have "0 == 0 ? 1 : 2;" which is then passed to fold -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de 2005-09-17 19:31 --- Please fix the caller who is not folding the condition in the first place instead. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-17 14:54 --- I should note I need to test it still which is why I had not posted it yet. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-17 14:53 --- Here is the patch: Index: fold-const.c === RCS file: /cvs/gcc/gcc/gcc/fold-const.c,v retrieving revision 1.625 diff -u -p -r1.625 fold-const.c --- fold-const.c9 Sep 2005 09:00:33 - 1.625 +++ fold-const.c17 Sep 2005 14:53:00 - @@ -9953,6 +9953,8 @@ fold_ternary (enum tree_code code, tree return NULL_TREE; case COND_EXPR: + /* Fold the condition operand, just in case it needs to be foldded. */ + arg0 = fold (arg0); /* Pedantic ANSI C says that a conditional expression is never an lvalue, so all simple results must be passed through pedantic_non_lvalue. */ if (TREE_CODE (arg0) == INTEGER_CST) -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From drab at kepler dot fjfi dot cvut dot cz 2005-09-17 14:24 --- (In reply to comment #16) > I have a patch. Excellent! Where is it? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-16 23:34 --- I have a patch. -- What|Removed |Added AssignedTo|unassigned at gcc dot gnu |pinskia at gcc dot gnu dot |dot org |org Status|NEW |ASSIGNED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-13 14:05 --- *** Bug 23858 has been marked as a duplicate of this bug. *** -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From janis at gcc dot gnu dot org 2005-09-09 16:52 --- The reduced testcase in comment #5 starts failing with this patch from rakdver: http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00814.html -- What|Removed |Added CC||rakdver at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-09-08 20:14 --- (In reply to comment #12) > in fold but we don't try to fold 0 == 0 first. And to confirm this, this is either a DOM or fold bug, fold is not considered recursive, maybe it should be in this case. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-08-03 00:49 --- (In reply to comment #10) > This is triggered by tree-ifcvt (which is enabled by -ftree-vectorize) But it looks like not a vectorize or tree-ifcvt bug at all but a fold bug. We get: 0 == 0 ? 0 : 3988292384 in fold but we don't try to fold 0 == 0 first. -- What|Removed |Added CC||law at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-08-01 14:01 --- *** Bug 23179 has been marked as a duplicate of this bug. *** -- What|Removed |Added CC||micis at gmx dot de http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From dorit at il dot ibm dot com 2005-07-27 10:38 --- This is triggered by tree-ifcvt (which is enabled by -ftree-vectorize) -- What|Removed |Added CC||dpatel at apple dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From belyshev at depni dot sinp dot msu dot ru 2005-07-24 16:13 --- (In reply to comment #8) > The reduced testcase for the second one: to make it fail on amd64 one needs to change long -> int. and this testcase started to fail at the same time as testcase in comment #5 . -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-07-24
15:56 ---
The reduced testcase for the second one:
unsigned long CRCTab[256];
void InitCRC(void) {
int I, J;
unsigned long C;
for (I=0; I<256; I++)
{
for (C=I,J=0;J<8;J++)
C=(C & 1) ? (C>>1)^0xEDB88320L : (C>>1);
CRCTab[I]=C;
}
}
They do look like the same bug.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From belyshev at depni dot sinp dot msu dot ru 2005-07-24 15:43 --- (In reply to comment #5) > Reduced testcase for the first one: introduced between "2005-05-17 00:20 UTC" and "2005-05-18 00:20 UTC" -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-07-24 15:40 --- Note the backtrace is: #5211 0x003f58d0 in fold_ternary (code=COND_EXPR, type=0x1e0e5b0, op0=0x1ee9d50, op1=0x1e18200, op2=0x1ee2080) at ../../gcc/fold-const.c:9905 #5212 0x003f8368 in fold_build3 (code=COND_EXPR, type=0x1e0e5b0, op0=0x1ee9d50, op1=0x1e18200, op2=0x1ee2080) at ../../gcc/fold-const.c:10375 #5213 0x003bb030 in fold_cond_expr_with_comparison (type=0x1e0e5b0, arg0=0x1ee9d50, arg1=0x1e18200, arg2=0x1ee2080) at ../../gcc/fold-const.c:4 -- What|Removed |Added Target Milestone|--- |4.1.0 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
--- Additional Comments From pinskia at gcc dot gnu dot org 2005-07-24
15:37 ---
Reduced testcase for the first one:
static unsigned short int crc_table[256];
void AC3_encode_init(void)
{
unsigned int c, n, k;
for(n=0; n<256; n++)
{
c = n << 8;
for (k = 0; k < 8; k++)
{
if (c & (1 << 15))
c = ((c << 1) & 0x) ^ (((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
& 0x);
}
crc_table[n] = c;
}
}
--
What|Removed |Added
Status|UNCONFIRMED |NEW
Ever Confirmed||1
Last reconfirmed|-00-00 00:00:00 |2005-07-24 15:37:30
date||
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
[Bug tree-optimization/23049] [4.1 Regression] ICE with -O3 -ftree-vectorize on 4.1.x
-- What|Removed |Added Known to fail||4.1.0 Known to work||4.0.0 Summary|ICE with -O3 -ftree-|[4.1 Regression] ICE with - |vectorize on 4.1.x |O3 -ftree-vectorize on 4.1.x http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23049
