Re: Cloog and PPL 0.10

Re: Cloog and PPL 0.10

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 706 bytes --]

Hi,

CLooG and Graphite are now expecting PPL 0.10 to be installed.
You will have to git-pull again the code for CLooG and install a
new PPL version.  Without that, the current head of the graphite
branch won't compile anymore.  Instructions on how to build the
branch are on http://gcc.gnu.org/wiki/Graphite

I recommend that you use the latest snapshots of PPL, right now
ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre22.tar.bz2
as I will clean up the code of the CLooG-PPL backend and it will
rely on new functionality that is present only on beta22.

Attached is the patch that checks for the version 0.10 of PPL.
Paolo, does this patch look good to you?

Thanks,
Sebastian Pop
--
AMD - GNU Tools

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 1157_graphite_ppl-0.10.diff --]
[-- Type: text/x-diff; name=1157_graphite_ppl-0.10.diff, Size: 3220 bytes --]

Index: configure
===================================================================
--- configure	(revision 139047)
+++ configure	(working copy)
@@ -4795,6 +4795,8 @@ fi
 
 
 # Check for PPL
+ppl_major_version=0
+ppl_minor_version=10
 ppllibs=
 pplinc=
 
@@ -4884,8 +4886,8 @@ fi
 saved_CFLAGS="$CFLAGS"
 CFLAGS="$CFLAGS $pplinc $gmpinc"
 
-echo "$as_me:$LINENO: checking for correct version of PPL" >&5
-echo $ECHO_N "checking for correct version of PPL... $ECHO_C" >&6
+echo "$as_me:$LINENO: checking for version $ppl_major_version.$ppl_minor_version of PPL" >&5
+echo $ECHO_N "checking for version $ppl_major_version.$ppl_minor_version of PPL... $ECHO_C" >&6
 cat >conftest.$ac_ext <<_ACEOF
 /* confdefs.h.  */
 _ACEOF
@@ -4897,7 +4899,7 @@ int
 main ()
 {
 
-#if PPL_VERSION_MAJOR != 0 || PPL_VERSION_MINOR != 9
+#if PPL_VERSION_MAJOR != $ppl_major_version || PPL_VERSION_MINOR != $ppl_minor_version
 choke me
 #endif
 
Index: gcc/graphite.c
===================================================================
--- gcc/graphite.c	(revision 139327)
+++ gcc/graphite.c	(working copy)
@@ -4730,6 +4730,7 @@ graphite_transform_loops (void)
   if (dump_file && (dump_flags & TDF_DETAILS))
     fprintf (dump_file, "Graphite loop transformations \n");
 
+  cloog_initialize ();
   build_scops ();
   limit_scops ();
 
@@ -4788,6 +4789,7 @@ graphite_transform_loops (void)
     }
 
   free_scops (current_scops);
+  cloog_finalize ();
 }
 
 #else /* If Cloog is not available: #ifndef HAVE_cloog.  */
Index: gcc/ChangeLog.graphite
===================================================================
--- gcc/ChangeLog.graphite	(revision 139327)
+++ gcc/ChangeLog.graphite	(working copy)
@@ -1,5 +1,10 @@
 2008-08-20  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite.c (graphite_transform_loops): Call cloog_initialize
+	and cloog_finalize.
+
+2008-08-20  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* tree-data-ref.c (stmt_simple_memref_p): New.
 	* tree-data-ref.h (stmt_simple_memref_p): Declared.
 	* graphite.c (stmt_simple_memref_for_scop_p): Removed.
Index: configure.ac
===================================================================
--- configure.ac	(revision 139047)
+++ configure.ac	(working copy)
@@ -1304,6 +1304,8 @@ AC_SUBST(gmpinc)
 
 
 # Check for PPL
+ppl_major_version=0
+ppl_minor_version=10
 ppllibs=
 pplinc=
 
@@ -1367,9 +1369,9 @@ fi
 saved_CFLAGS="$CFLAGS"
 CFLAGS="$CFLAGS $pplinc $gmpinc"
 
-AC_MSG_CHECKING([for correct version of PPL])
+AC_MSG_CHECKING([for version $ppl_major_version.$ppl_minor_version of PPL])
 AC_TRY_COMPILE([#include "ppl_c.h"],[
-#if PPL_VERSION_MAJOR != 0 || PPL_VERSION_MINOR != 9
+#if PPL_VERSION_MAJOR != $ppl_major_version || PPL_VERSION_MINOR != $ppl_minor_version
 choke me
 #endif
 ], [AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no]); clooglibs= ; clooginc= ])
Index: ChangeLog.graphite
===================================================================
--- ChangeLog.graphite	(revision 139047)
+++ ChangeLog.graphite	(working copy)
@@ -1,3 +1,8 @@
+2008-08-20  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* configure.ac: Request PPL version 0.10.
+	* configure: Regenerated.
+
 2008-08-13  Tobias Grosser  <grosser@fim.uni-passau.de>
 
 	* configure: Rebuilt.

Re: Cloog and PPL 0.10

[Paolo Bonzini]

Sebastian Pop wrote:
> Hi,
> 
> CLooG and Graphite are now expecting PPL 0.10 to be installed.
> You will have to git-pull again the code for CLooG and install a
> new PPL version.  Without that, the current head of the graphite
> branch won't compile anymore.  Instructions on how to build the
> branch are on http://gcc.gnu.org/wiki/Graphite
> 
> I recommend that you use the latest snapshots of PPL, right now
> ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre22.tar.bz2
> as I will clean up the code of the CLooG-PPL backend and it will
> rely on new functionality that is present only on beta22.
> 
> Attached is the patch that checks for the version 0.10 of PPL.
> Paolo, does this patch look good to you?

Yes, thanks.

Paolo

Re: Cloog and PPL 0.10

[Jack Howarth]

Sebastian,
   Do you have a patch for allowing graphite to build against
polylib as well as ppl yet? The current graphite branch patches
and ppl 1.0pre22 still appears unable to do any actual loop
optimizations so I would like to be able to go back to using
polylib for some polyhedron 2005 benchmarks of the new
loop optimizations. Thanks in advance.
                  Jack

On Thu, Aug 21, 2008 at 03:20:34PM +0200, Paolo Bonzini wrote:
> Sebastian Pop wrote:
>> Hi,
>>
>> CLooG and Graphite are now expecting PPL 0.10 to be installed.
>> You will have to git-pull again the code for CLooG and install a
>> new PPL version.  Without that, the current head of the graphite
>> branch won't compile anymore.  Instructions on how to build the
>> branch are on http://gcc.gnu.org/wiki/Graphite
>>
>> I recommend that you use the latest snapshots of PPL, right now
>> ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre22.tar.bz2
>> as I will clean up the code of the CLooG-PPL backend and it will
>> rely on new functionality that is present only on beta22.
>>
>> Attached is the patch that checks for the version 0.10 of PPL.
>> Paolo, does this patch look good to you?
>
> Yes, thanks.
>
> Paolo

Re: [PPL-devel] Cloog and PPL 0.10

[Roberto Bagnara]

Jack Howarth wrote:
> The current graphite branch patches
> and ppl 1.0pre22 still appears unable to do any actual loop
> optimizations [...]

Hi there,

I don't know if the reason of this was the lack of a simplification
operation in the PPL.  This has now be implemented and is available
in the snapshot

     ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre23.tar.bz2

All the best,

    Roberto

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara@cs.unipr.it

Re: [PPL-devel] Cloog and PPL 0.10

[Sebastian Pop]

On Thu, Aug 21, 2008 at 12:09 PM, Roberto Bagnara <bagnara@cs.unipr.it> wrote:
> I don't know if the reason of this was the lack of a simplification
> operation in the PPL.  This has now be implemented and is available
> in the snapshot
>
>    ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre23.tar.bz2

Excellent!  Many thanks Roberto for this quick fix ;-) I'm going to
modify CLooG to call this function and see what happens to CLooG's
testsuite.

Sebastian

Re: [PPL-devel] Cloog and PPL 0.10

[Jack Howarth]

On Thu, Aug 21, 2008 at 12:40:08PM -0500, Sebastian Pop wrote:
> On Thu, Aug 21, 2008 at 12:09 PM, Roberto Bagnara <bagnara@cs.unipr.it> wrote:
> > I don't know if the reason of this was the lack of a simplification
> > operation in the PPL.  This has now be implemented and is available
> > in the snapshot
> >
> >    ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre23.tar.bz2
> 
> Excellent!  Many thanks Roberto for this quick fix ;-) I'm going to
> modify CLooG to call this function and see what happens to CLooG's
> testsuite.
> 
> Sebastian

Sebastian,
    Are there options in graphite to dump the loop optimization details
for each loop detected; similar to how vectorization will report...

 a.f90:8: note: not vectorized: data ref analysis failed D.1462_61 =
 (*aa_60(D))[D.1461_59]

for code like...

subroutine sub(aa,bb,n,m)
implicit none
integer, intent(in) :: n,m
real, intent(inout) :: aa(n,m)
real, intent(in)    :: bb(n,m)
integer :: i,j
do i = 1,m
   do j= 2,n
      aa(i,j)= aa(i,j-1)+bb(i,j-1)
   enddo
enddo
do j= 2,n
   do i = 1,m
      aa(i,j)= aa(i,j-1)+bb(i,j-1)
   enddo
enddo
end subroutine

that can't currently be vectorized.
              Jack

Re: [PPL-devel] Cloog and PPL 0.10

[Sebastian Pop]

On Thu, Aug 21, 2008 at 2:29 PM, Jack Howarth <howarth@bromo.msbb.uc.edu> wrote:
>    Are there options in graphite to dump the loop optimization details
> for each loop detected; similar to how vectorization will report...

You can try -fdump-tree-graphite-all
Though, the messages are probably not yet as precise as for the vectorizer.

Sebastian

Re: [PPL-devel] Cloog and PPL 0.10

[Jack Howarth]

Robert,
    Is the external ABI of ppl the same between 1.0pre22 and 1.0pre23?
I am wondering if I should just be able to install the new 1.0pre23
or if I need to rebuild both cloog-ppl git and gcc against it as well.
                  Jack

On Thu, Aug 21, 2008 at 07:09:40PM +0200, Roberto Bagnara wrote:
> Jack Howarth wrote:
>> The current graphite branch patches
>> and ppl 1.0pre22 still appears unable to do any actual loop
>> optimizations [...]
>
> Hi there,
>
> I don't know if the reason of this was the lack of a simplification
> operation in the PPL.  This has now be implemented and is available
> in the snapshot
>
>     ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre23.tar.bz2
>
> All the best,
>
>    Roberto
>
> -- 
> Prof. Roberto Bagnara
> Computer Science Group
> Department of Mathematics, University of Parma, Italy
> http://www.cs.unipr.it/~bagnara/
> mailto:bagnara@cs.unipr.it

Re: [PPL-devel] Cloog and PPL 0.10

[Sebastian Pop]

On Thu, Aug 21, 2008 at 2:18 PM, Jack Howarth <howarth@bromo.msbb.uc.edu> wrote:
>    Is the external ABI of ppl the same between 1.0pre22 and 1.0pre23?
> I am wondering if I should just be able to install the new 1.0pre23
> or if I need to rebuild both cloog-ppl git and gcc against it as well.

pre23 contains more functions for domain_simplify, but these are not
yet used in cloog.git.  I will send a note when CLooG is using these
and when you'll have to update cloog and ppl.  For the moment either
pre22 or pre23 should just work.

Sebastian

Re: [PPL-devel] Cloog and PPL 0.10

[Roberto Bagnara]

Jack Howarth wrote:
>     Is the external ABI of ppl the same between 1.0pre22 and 1.0pre23?

Hi Jack,

the answer is negative.

> I am wondering if I should just be able to install the new 1.0pre23
> or if I need to rebuild both cloog-ppl git and gcc against it as well.

In general, until we make an officiao release, we do not worry about
the ABI.  So recompilation is almost always necessary.
Cheers,

    Roberto

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara@cs.unipr.it

Re: Cloog and PPL 0.10

[Roberto Bagnara]

Hi there,

a new snapshot release of PPL 0.10 has been uploaded:

     ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre24.tar.bz2

This contains the latest fixes and improvements to the
"simplify using context" operation.  Please let us know if, whith this
one, the graphite branch performs the expected loop optimizations.
Let us also know if you want us to add other functionalities
to PPL 0.10.
All the best,

    Roberto

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara@cs.unipr.it

Re: Cloog and PPL 0.10

[Sebastian Pop]

On Fri, Aug 22, 2008 at 11:19 AM, Roberto Bagnara <bagnara@cs.unipr.it> wrote:
> a new snapshot release of PPL 0.10 has been uploaded:
>
>    ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre24.tar.bz2
>

As well, you should git-pull again the Cloog changes.

Sebastian

Re: Cloog and PPL 0.10

[Jack Howarth]

On Fri, Aug 22, 2008 at 11:55:05AM -0500, Sebastian Pop wrote:
> On Fri, Aug 22, 2008 at 11:19 AM, Roberto Bagnara <bagnara@cs.unipr.it> wrote:
> > a new snapshot release of PPL 0.10 has been uploaded:
> >
> >    ftp://ftp.cs.unipr.it/pub/ppl/snapshots/ppl-0.10pre24.tar.bz2
> >
> 
> As well, you should git-pull again the Cloog changes.
> 
> Sebastian

Sebastian,
    So cloog-ppl git is now modified to use the new simplification
routines and graphite will now actually perform optimizations on
the generated code?
                Jack

Re: Cloog and PPL 0.10

[Sebastian Pop]

On Fri, Aug 22, 2008 at 1:09 PM, Jack Howarth <howarth@bromo.msbb.uc.edu> wrote:
>    So cloog-ppl git is now modified to use the new simplification
> routines and graphite will now actually perform optimizations on
> the generated code?

Wait till end of today to try graphite branch, otherwise
you'll get all the errors we saw.  We're fixing these right now.

Thanks,
Sebastian

Re: Cloog and PPL 0.10

[Jack Howarth]

On Fri, Aug 22, 2008 at 01:47:20PM -0500, Sebastian Pop wrote:
> On Fri, Aug 22, 2008 at 1:09 PM, Jack Howarth <howarth@bromo.msbb.uc.edu> wrote:
> >    So cloog-ppl git is now modified to use the new simplification
> > routines and graphite will now actually perform optimizations on
> > the generated code?
> 
> Wait till end of today to try graphite branch, otherwise
> you'll get all the errors we saw.  We're fixing these right now.
> 
> Thanks,
> Sebastian

Sebastian,
     With tonight's graphite branch changes added to the rest on top
of gcc trunk and using current clogg-ppl git with ppl-1.0rc24, I see
two gcc.dg/graphite/graphite.exp testsuite failures on i686-apple-darwin9...

FAIL: gcc.dg/graphite/block-1.c (internal compiler error)
FAIL: gcc.dg/graphite/block-1.c (test for excess errors)
FAIL: gcc.dg/graphite/block-1.c scan-tree-dump-times graphite "Loop blocked" 3
FAIL: gcc.dg/graphite/scop-16.c (internal compiler error)
FAIL: gcc.dg/graphite/scop-16.c (test for excess errors)

which are...

Executing on host: /sw/src/fink.build/gcc44-4.3.999-20080822/darwin_objdir/gcc/xgcc -B/sw/src/fink.build/gcc44-4.3.999-20080822/darwin_objdir/gc
c/ /sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/block-1.c   -O2 -floop-block -fdump-tree-graphite-al
l -S  -o block-1.s    (timeout = 300)
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/block-1.c: In function 'main':
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/block-1.c:6: error: found a real definition for a non-r
egister
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/block-1.c:6: error: in statement
D.1950_13 = A_68 + graphiteIV.24_4;
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/block-1.c:6: internal compiler error: verify_ssa failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

xecuting on host: /sw/src/fink.build/gcc44-4.3.999-20080822/darwin_objdir/gcc/xgcc -B/sw/src/fink.build/gcc44-4.3.999-20080822/darwin_objdir/gc
c/ /sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/scop-16.c   -O2 -floop-block -fdump-tree-graphite-al
l -S  -o scop-16.s    (timeout = 300)
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/scop-16.c: In function 'test':
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/scop-16.c:5: error: found a real definition for a non-r
egister
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/scop-16.c:5: error: in statement
# VUSE <a_83> { a }
D.1955_21 = a[graphiteIV.37_78][a_83];
/sw/src/fink.build/gcc44-4.3.999-20080822/gcc-4.4-20080822/gcc/testsuite/gcc.dg/graphite/scop-16.c:5: internal compiler error: verify_ssa failed
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

I assume these are the remaining errors unfixed from...

http://gcc.gnu.org/ml/gcc-patches/2008-08/msg01715.html

Is that correct? Thanks for any clarifications.
                Jack

Re: Cloog and PPL 0.10

[Sebastian Pop]

On Fri, Aug 22, 2008 at 10:24 PM, Jack Howarth
<howarth@bromo.msbb.uc.edu> wrote:
> FAIL: gcc.dg/graphite/block-1.c (internal compiler error)
> FAIL: gcc.dg/graphite/block-1.c (test for excess errors)
> FAIL: gcc.dg/graphite/block-1.c scan-tree-dump-times graphite "Loop blocked" 3
> FAIL: gcc.dg/graphite/scop-16.c (internal compiler error)
> FAIL: gcc.dg/graphite/scop-16.c (test for excess errors)
>
> I assume these are the remaining errors unfixed from...
>
> http://gcc.gnu.org/ml/gcc-patches/2008-08/msg01715.html
>
> Is that correct? Thanks for any clarifications.

Jan's patch fixes block-0.c.  The remaining errors are out of graphite
ICEs.  These should be fixed in next patches.

Thanks for the test,
Sebastian

Re: Cloog and PPL 0.10

[Jack Howarth]

On Sat, Aug 23, 2008 at 04:21:48PM -0500, Sebastian Pop wrote:
> 
> Jan's patch fixes block-0.c.  The remaining errors are out of graphite
> ICEs.  These should be fixed in next patches.
> 
> Thanks for the test,
> Sebastian

Sebastian,
    Is current graphite branch, current cloog-git and ppl 1.0rc24 producing
code from graphite and the new loop optimizations on the backend yet? I don't
seem to be seeing any differences in the assembly generated with...

-funroll-loops -O3 -floop-block -floop-interchange -floop-strip-mine -fgraphite

compared to

-funroll-loops -O3 -fgraphite

...in gfortran compiling MDBNCH (http://www.fisica.uniud.it/~ercolessi/mdbnch.html).
          Jack

Re: Cloog and PPL 0.10

[Jack Howarth]

Sebastian,
    Which version of cloog should one use with polylib and
graphite? I assume the cloog-ppl git should still build fine
against polylib for use with graphite, correct? Are there
non-ppl related fixes for graphite in the cloog-ppl git
which would be applicable to polylib as well?
                    Jack