[Bug c/39500] New: autopar fails to parallel
autopar fails auto-parallelization with the code below:
int X[1000];
int main(void)
{
int i;
for (i = 0; i < 100; i++)
X[i] = X[i+100];
return 0;
}
--
Summary: autopar fails to parallel
Product: gcc
Version: 4.4.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: nemokingdom at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39500
[Bug c/39500] autopar fails to parallel
--- Comment #1 from nemokingdom at gmail dot com 2009-03-19 01:27 ---
(In reply to comment #0)
> autopar fails auto-parallelization with the code below:
>
> int X[1000];
>
> int main(void)
> {
> int i;
>
> for (i = 0; i < 100; i++)
> X[i] = X[i+100];
>
> return 0;
> }
>
This bug basically about x1 and y1 not allowed to be in between zero and niter
- 1. This seems to be an typical off by one error.
x1 and y1 not allowed to be in between zero and niter - 1. This seems to
be an typical off by one error.
Index: tree-data-ref.c
===
--- tree-data-ref.c (revision 144721)
+++ tree-data-ref.c (working copy)
@@ -2292,7 +2292,7 @@
/* If the overlap occurs outside of the bounds of the
loop, there is no dependence. */
- if (x1 > niter || y1 > niter)
+ if (x1 >= niter || y1 >= niter)
{
*overlaps_a = conflict_fn_no_dependence ();
*overlaps_b = conflict_fn_no_dependence ();
--
nemokingdom at gmail dot com changed:
What|Removed |Added
Status|UNCONFIRMED |RESOLVED
Resolution||FIXED
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39500
[Bug c/39500] autopar fails to parallel
--- Comment #4 from nemokingdom at gmail dot com 2009-03-19 02:46 ---
(In reply to comment #3)
> I will take care of this one.
>
I add test case for this situation.
Index: testsuite/gcc.dg/autopar/pr39500-1.c
===
--- testsuite/gcc.dg/autopar/pr39500-1.c(revision 0)
+++ testsuite/gcc.dg/autopar/pr39500-1.c(revision 0)
@@ -0,0 +1,29 @@
+/* pr39500: autopar fails to parallel */
+/* origin: nemoking...@gmail.com(LiFeng)*/
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-parallelize-loops=4 -fdump-tree-parloops-details
-fdump-tree-final_cleanup" } */
+
+void abort (void);
+
+int main(void)
+{
+ int i;
+ int x[100];
+
+ for (i = 0; i < 100; i++)
+ x[i] = x[i+100];
+
+ for (i = 0; i < 100; i++)
+{
+ if (x[i] != x[i+100])
+ abort ();
+}
+
+ return 0;
+}
+
+/* Check that the first loop in parloop got parallelized. */
+
+/* { dg-final { scan-tree-dump-times "SUCCESS: may be parallelized" 1
"parloops" } } */
+/* { dg-final { cleanup-tree-dump "parloops" } } */
+/* { dg-final { cleanup-tree-dump "final_cleanup" } } */
Index: testsuite/gcc.dg/autopar/pr39500-2.c
===
--- testsuite/gcc.dg/autopar/pr39500-2.c(revision 0)
+++ testsuite/gcc.dg/autopar/pr39500-2.c(revision 0)
@@ -0,0 +1,21 @@
+/* pr39500: autopar fails to parallel */
+/* origin: nemoking...@gmail.com(LiFeng)*/
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftree-parallelize-loops=4 -fdump-tree-parloops-details
-fdump-tree-final_cleanup" } */
+
+int main(void)
+{
+ int i;
+ int x[100];
+
+ for (i = 0; i < 101; i++)
+ x[i] = x[i+100];
+
+ return 0;
+}
+
+/* Check that the first loop in parloop got parallelized. */
+
+/* { dg-final { scan-tree-dump-times "SUCCESS: may be parallelized" 0
"parloops" } } */
+/* { dg-final { cleanup-tree-dump "parloops" } } */
+/* { dg-final { cleanup-tree-dump "final_cleanup" } } */
Index: tree-data-ref.c
===
--- tree-data-ref.c (revision 144648)
+++ tree-data-ref.c (working copy)
@@ -2302,7 +2302,7 @@
/* If the overlap occurs outside of the bounds of the
loop, there is no dependence. */
- if (x1 > niter || y1 > niter)
+ if (x1 >= niter || y1 >= niter)
{
*overlaps_a = conflict_fn_no_dependence ();
*overlaps_b = conflict_fn_no_dependence ();
--
nemokingdom at gmail dot com changed:
What|Removed |Added
Component|middle-end |c
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39500
[Bug middle-end/39575] New: after build_sese_loop_nests, loops still unordered
in build_sese_loop_nests, the loops in sese->loop_nest can be: loop3 loop2 loop1(test case: gcc.dg/graphite/block-3.c), after the algorithms(sort the loops) there, it will still unordered as loop2 loop1 loop3. I'm not sure if this is the bug. but if this bug is confirmed, I will take care of this. -- Summary: after build_sese_loop_nests, loops still unordered Product: gcc Version: 4.4.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: nemokingdom at gmail dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39575
[Bug middle-end/39568] [graphite] Remove GBB_LOOPS
--- Comment #2 from nemokingdom at gmail dot com 2009-03-30 11:30 ---
I test with bootstrap enabled and success.
Here is the patch:
Index: sese.h
===
--- sese.h (revision 145190)
+++ sese.h (working copy)
@@ -341,54 +341,6 @@
CASE_LABEL_EXPR. */
VEC (gimple, heap) *conditions;
VEC (gimple, heap) *condition_cases;
-
- /* LOOPS contains for every column in the graphite domain the corresponding
- gimple loop. If there exists no corresponding gimple loop LOOPS contains
- NULL.
-
- Example:
-
- Original code:
-
- for (i = 0; i <= 20; i++)
- for (j = 5; j <= 10; j++)
- A
-
- Original domain:
-
- | i >= 0
- | i <= 20
- | j >= 0
- | j <= 10
-
- This is a two dimensional domain with "Loop i" represented in
- dimension 0, and "Loop j" represented in dimension 1. Original
- loops vector:
-
- | 0 1
- | Loop iLoop j
-
- After some changes (Exchange i and j, strip-mine i), the domain
- is:
-
- | i >= 0
- | i <= 20
- | j >= 0
- | j <= 10
- | ii <= i
- | ii + 1 >= i
- | ii <= 2k
- | ii >= 2k
-
- Iterator vector:
- | 012 3
- | Loop j NULL Loop iNULL
-
- Means the original loop i is now on dimension 2 of the domain and
- loop j in the original loop nest is now on dimension 0.
- Dimensions 1 and 3 represent the newly created loops. */
- VEC (loop_p, heap) *loops;
-
VEC (data_reference_p, heap) *data_refs;
htab_t cloog_iv_types;
} *gimple_bb_p;
@@ -397,10 +349,9 @@
#define GBB_DATA_REFS(GBB) GBB->data_refs
#define GBB_CONDITIONS(GBB) GBB->conditions
#define GBB_CONDITION_CASES(GBB) GBB->condition_cases
-#define GBB_LOOPS(GBB) (GBB->loops)
#define GBB_CLOOG_IV_TYPES(GBB) GBB->cloog_iv_types
-/* Return the loop that contains the basic block GBB. */
+/* Return the innermost loop that contains the basic block GBB. */
static inline struct loop *
gbb_loop (struct gimple_bb *gbb)
@@ -412,24 +363,26 @@
If there is no corresponding gimple loop, we return NULL. */
static inline loop_p
-gbb_loop_at_index (gimple_bb_p gbb, int index)
+gbb_loop_at_index (gimple_bb_p gbb, sese region, int index)
{
- return VEC_index (loop_p, GBB_LOOPS (gbb), index);
-}
+ loop_p loop = gbb_loop (gbb);
+ int depth = sese_loop_depth (region, loop);
-/* Returns the index of LOOP in the loop nest around GBB. */
+ while (--depth > index)
+{
+ loop = loop_outer (loop);
+}
+ gcc_assert (sese_contains_loop (region, loop));
+
+ return loop;
+}
+
+/* Returns the index of LOOP in the region, 0 based index */
static inline int
-gbb_loop_index (gimple_bb_p gbb, loop_p loop)
+gbb_loop_index (sese region, loop_p loop)
{
- int i;
- loop_p l;
-
- for (i = 0; VEC_iterate (loop_p, GBB_LOOPS (gbb), i, l); i++)
-if (loop == l)
- return i;
-
- gcc_unreachable();
+ return sese_loop_depth (region, loop) - 1;
}
/* The number of common loops in REGION for GBB1 and GBB2. */
Index: graphite-clast-to-gimple.c
===
--- graphite-clast-to-gimple.c (revision 145190)
+++ graphite-clast-to-gimple.c (working copy)
@@ -316,6 +316,7 @@
int index = 0;
CloogStatement *cs = user_stmt->statement;
poly_bb_p pbb = (poly_bb_p) cloog_statement_usr (cs);
+ sese region = SCOP_REGION (PBB_SCOP (pbb));
for (t = user_stmt->substitutions; t; t = t->next)
{
@@ -327,8 +328,8 @@
if (expr->type == expr_term
&& !term->var)
{
- loop_p loop = gbb_loop_at_index (PBB_BLACK_BOX (pbb), index);
- tree oldiv = oldiv_for_loop (SCOP_REGION (PBB_SCOP (pbb)), loop);
+ loop_p loop = gbb_loop_at_index (PBB_BLACK_BOX (pbb), region, index);
+ tree oldiv = oldiv_for_loop (region, loop);
tree type = oldiv ? TREE_TYPE (oldiv) : integer_type_node;
tree value = gmp_cst_to_tree (type, term->val);
loop_iv_stack_insert_constant (stack, index, value);
@@ -507,7 +508,8 @@
free (*slot);
new_name = loop_iv_stack_get_iv
- (ivstack, gbb_loop_index (PBB_BLACK_BOX (pbb), iv->loop));
+ (ivstack, gbb_loop_index (region, iv->loop));
+
*slot = new_rename_map_elt (old_name, new_name);
}
}
@@ -685,6 +687,7 @@
{
gimple_bb_p gbb = PBB_BLACK_BOX (pbb);
struct clast_stmt *t;
+ sese region = SCOP_REGION (PBB_SCOP (pbb));
int index = 0;
for (t = user_stmt->substitutions; t; t = t->next, index++)
@@ -703,8 +706,8 @@
if (!*slot)
{
- loop_p loop = gbb_loop_at_index (PBB_BLACK_BOX (pbb), index);
- tree oldiv = oldiv_for_loop (SCOP_REGION (PBB_SCOP (pbb)), loop);
+ loop_p loop = gbb_loop_at_index (
[Bug middle-end/42298] New: GCC fails validity in mdbx in polyhedral benchmark.
t= 0.00 Tolerance=0.100E-11
> Value=0.100E-13 Target= 0.00 Tolerance=0.100E-11
= Value=100 Target=100
> Value=0.2897000 Target=0.2897000 Tolerance=0.500E-03
> Value=-3.341300 Target=-3.341300 Tolerance=0.100E-02
> Value=-3.051700 Target=-3.051700 Tolerance=0.500E-03
> Value= 1.100700 Target= 1.100700 Tolerance=0.100E-02
> Value=0.500E-13 Target= 0.00 Tolerance=0.100E-11
> Value=-.100E-13 Target= 0.00 Tolerance=0.100E-11
> Value=-.300E-13 Target= 0.00 Tolerance=0.100E-11
mdbx FAILED1 fails and 39 passes
Error executing R/mdbx/ - End of Instruction Stream
--
Summary: GCC fails validity in mdbx in polyhedral benchmark.
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: nemokingdom at gmail dot com
GCC build triplet: i686-pc-linux-gnu
GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42298
[Bug middle-end/42298] GCC fails validity in mdbx in polyhedral benchmark.
--- Comment #1 from nemokingdom at gmail dot com 2009-12-05 07:04 --- The polyhedral benchmark is available in: http://www.polyhedron.com/web_images/documents/pb05.zip > One fail (of 40 checkings) in mdbx in polyhedral benchmark with option -O2 > -fno-loop-block -fno-loop-strip-mine. > > git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/tr...@155007 > > $ ../configure --prefix=/home/li/source/gcc.trunk/install > --disable-bootstrap > --enable-languages=c,fortran,c++ --with-cloog=/usr/lib/cloog-0.15.7 > --with-ppl=/usr/lib/ppl-0.10.2 > > ## - ## > ## Platform. ## > ## - ## > > hostname = laptop > uname -m = i686 > uname -r = 2.6.27-7-generic > uname -s = Linux > > > > Date & Time : 5 Dec 2009 14:42:10 > Test Name : airgraphite > Compile Command : /home/li/source/gcc.trunk/install/bin/gfortran -O2 > -fno-loop-block -fno-loop-strip-mine %n.f90 -o %n > Benchmarks : -ac -aermod -air -capacita -channel -doduc -fatigue -gas_dyn > -induct -linpk mdbx -nf -protein -rnflow -test_fpu -tfft > Maximum Times : 200.0 > Target Error % : 0.100 > Minimum Repeats : 2 > Maximum Repeats : 2 > >Benchmark Compile Executable Ave Run Number Estim > Name(secs) (bytes)(secs) Repeats Err % >- --- -- --- --- -- > -ac 0.00 0 -1.00 0. > -aermod 0.00 0 -1.00 0. > -air 0.00 0 -1.00 0. >-capacita 0.00 0 -1.00 0. > -channel 0.00 0 -1.00 0. > -doduc 0.00 0 -1.00 0. > -fatigue 0.00 0 -1.00 0. > -gas_dyn 0.00 0 -1.00 0. > -induct 0.00 0 -1.00 0. > -linpk 0.00 0 -1.00 0. > mdbx 4.13 40691 18.09 2 1.5863 > -nf 0.00 0 -1.00 2 1.5863 > -protein 0.00 0 -1.00 2 1.5863 > -rnflow 0.00 0 -1.00 2 1.5863 >-test_fpu 0.00 0 -1.00 2 1.5863 >-tfft 0.00 0 -1.00 2 1.5863 > > Geometric Mean Execution Time = Infinity seconds > > > Polyhedron Benchmark Validator > Copyright (C) Polyhedron Software Ltd - 2004 - All rights reserved > > Test File: airgraphite.run > Instruction File : air_valid.in > > = Value=1000 Target=1000 > > Value=0.2804000 Target=0.270 Tolerance=0.100E-01 > FAIL <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< > > Value=-3.328200 Target=-3.32 Tolerance=0.200E-01 > > Value=-3.047800 Target=-3.045000 Tolerance=0.500E-02 > > Value= 17.45580 Target= 17.1 Tolerance= 1.00 > > Value=-.400E-13 Target= 0.00 Tolerance=0.100E-11 > > Value=-.200E-13 Target= 0.00 Tolerance=0.100E-11 > > Value=-.200E-13 Target= 0.00 Tolerance=0.100E-11 > = Value=100 Target=100 > > Value=0.2962000 Target=0.2962000 Tolerance=0.200E-03 > > Value=-3.344000 Target=-3.344000 Tolerance=0.200E-03 > > Value=-3.047800 Target=-3.047800 Tolerance=0.500E-03 > > Value= 1.001200 Target= 1.001200 Tolerance=0.200E-03 > > Value=0.100E-13 Target= 0.00 Tolerance=0.100E-11 > > Value=-.700E-14 Target= 0.00 Tolerance=0.100E-11 > > Value=0.100E-13 Target= 0.00 Tolerance=0.100E-11 > = Value=100 Target=100 > > Value=0.2962000 Target=0.2962000 Tolerance=0.200E-03 > > Value=-3.344000 Target=-3.344000 Tolerance=0.200E-03 > > Value=-3.047800 Target=-3.047800 Tolerance=0.500E-03 > > Value= 1.001200 Target= 1.001200 Tolerance=0.200E-03 > > Value=-.200E-13 Target= 0.00 Tolerance=0.100E-11 > > Value=0.600E
[Bug middle-end/42298] [4.5 Regression] GCC fails validity in mdbx in polyhedral benchmark.
--- Comment #3 from nemokingdom at gmail dot com 2009-12-07 13:01 --- Subject: Re: [4.5 Regression] GCC fails validity in mdbx in polyhedral benchmark. Hi Rguenth, On Mon, Dec 7, 2009 at 7:14 PM, rguenth at gcc dot gnu dot org wrote: > > > --- Comment #2 from rguenth at gcc dot gnu dot org 2009-12-07 11:14 > --- > Huh? Does plain -O2 work? Do I understand correctly that -O2 -fno-loop-block > -fno-loop-strip-mine miscompiles? Em, this benchmark can be compiled, but failed at validity of the benchmark testing (passed 39 of 40). So I think this caused a runtime error. -O2 also has this fail. I disable Graphite by -fno-loop-strip-mine -fno-loop-block. So I think it's not related to Graphite. > > Sebastian, how can disabling graphite options but not enabling graphite > miscompile anything?? P1, please fix / comment. > > > -- > > rguenth at gcc dot gnu dot org changed: > > What |Removed |Added > > CC| |spop at gcc dot gnu dot org > Keywords| |wrong-code > Priority|P3 |P1 > Summary|GCC fails validity in mdbx |[4.5 Regression] GCC fails > |in polyhedral benchmark. |validity in mdbx in > | |polyhedral benchmark. > Target Milestone|--- |4.5.0 > > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42298 > > --- You are receiving this mail because: --- > You reported the bug, or are watching the reporter. > Li -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42298
