[PATCH 00/10] Make -floop-interchange catch almost all testcases of -ftree-loop-linear
Jack Howarth
howarth@bromo.med.uc.edu
Mon Jan 17 06:31:00 GMT 2011
On Sat, Jan 15, 2011 at 03:05:05AM -0600, Sebastian Pop wrote:
> Hi Richi,
>
> Here is the first part of the patch to remove the lambda framework.
> These patches un-xfail most of the transforms that we wanted to see
> with -floop-interchange and that were part of the testsuite of
> -ftree-loop-linear:
>
> gcc.dg/graphite/block-0.c
> gcc.dg/graphite/block-1.c
> gcc.dg/graphite/block-4.c
> gcc.dg/graphite/block-7.c
> gcc.dg/graphite/block-8.c
> gcc.dg/graphite/interchange-1.c
> gcc.dg/graphite/interchange-11.c
> gcc.dg/graphite/interchange-12.c
> gcc.dg/graphite/interchange-13.c
> gcc.dg/graphite/interchange-14.c
> gcc.dg/graphite/interchange-15.c
> gcc.dg/graphite/interchange-8.c
> gcc.dg/graphite/interchange-mvt.c
>
> The only xfails remaining to be fixed before removing the lambda
> framework are:
> - interchange-2.c: some problem linked to the scop detection
> - interchange-3.f90: data dependence analysis does not validate the interchange.
> I will continue my investigation on these two xfails.
>
> With the patches, the Graphite testsuite has the following XFAILs:
>
> block-4.c:59:/* { dg-final { scan-tree-dump-times "will be loop blocked" 1 "graphite" { xfail *-*-* } } } */
> interchange-2.c:55:/* { dg-final { scan-tree-dump-times "will be interchanged" 1 "graphite" { xfail *-*-* } } } */
> pr35356-3.c:39:/* { dg-final { scan-tree-dump-times "loop_1" 0 "graphite" { xfail *-*-* } } } */
> pr37485.c:31:/* { dg-final { scan-tree-dump-times "Loop blocked" 1 "graphite" { xfail *-*-* }} } */
>
> block-1.f90:11:! { dg-final { scan-tree-dump-times "will be loop blocked" 1 "graphite" { xfail *-*-* } } }
> block-2.f:20:! { dg-final { scan-tree-dump-times "will be loop blocked" 2 "graphite" { xfail *-*-* } } }
> block-3.f90:15:! { dg-final { scan-tree-dump-times "number of SCoPs: 1" 1 "graphite" { xfail *-*-* } } }
> block-3.f90:16:! { dg-final { scan-tree-dump-times "will be loop blocked" 1 "graphite" { xfail *-*-* } } }
> block-4.f90:18:! { dg-final { scan-tree-dump-times "number of SCoPs: 1" 1 "graphite" { xfail *-*-* } } }
> block-4.f90:19:! { dg-final { scan-tree-dump-times "will be loop blocked" 1 "graphite" { xfail *-*-* } } }
> interchange-1.f:44:! { dg-final { scan-tree-dump-times "will be interchanged" 1 "graphite" { xfail *-*-* } } }
> interchange-3.f90:27:! { dg-final { scan-tree-dump-times "will be interchanged" 1 "graphite" { xfail *-*-* } } }
> pr14741.f90:27:! { dg-final { scan-tree-dump-times "number of SCoPs: 1" 1 "graphite" { xfail *-*-* } } }
> pr14741.f90:28:! { dg-final { scan-tree-dump-times "will be loop blocked" 1 "graphite" { xfail *-*-* } } }
> scop-1.f:12:! { dg-final { scan-tree-dump-times "number of SCoPs: 1" 1 "graphite" { xfail *-*-* } } }
>
> I'm committing these patches to the graphite branch for regstrap and
> SPEC testing.
Regression tests for these 10 patches on x86_64-apple-darwin10 shows no graphite regressions...
http://gcc.gnu.org/ml/gcc-testresults/2011-01/msg01391.html
>
> Sebastian Pop (10):
> Add debug_gmp_value.
> Print the data dependence polyhedron in the PPL format.
> Test the profitability of interchange on the perfect nest.
> Fix pbb_remove_duplicate_pdrs.
> speedup compilation
> Correct the precedence relation.
> Use PIP to determine the integer feasibility of a constraint system.
> Minimize the number of expensive calls to ppl_powerset_is_empty.
> Expect at least the version 0.11 of PPL.
> Remove the temporary array for reductions written to memory.
>
> ChangeLog.graphite | 5 +
> configure | 6 +-
> configure.ac | 4 +-
> gcc/ChangeLog.graphite | 96 ++++++++++
> gcc/doc/install.texi | 2 +-
> gcc/graphite-dependences.c | 226 +++++++++--------------
> gcc/graphite-interchange.c | 16 +-
> gcc/graphite-poly.c | 13 +-
> gcc/graphite-ppl.c | 76 ++++++++
> gcc/graphite-ppl.h | 3 +
> gcc/graphite-sese-to-poly.c | 43 ++++-
> gcc/testsuite/gcc.dg/graphite/block-0.c | 3 +-
> gcc/testsuite/gcc.dg/graphite/block-1.c | 5 +-
> gcc/testsuite/gcc.dg/graphite/block-4.c | 2 +
> gcc/testsuite/gcc.dg/graphite/block-7.c | 3 +-
> gcc/testsuite/gcc.dg/graphite/block-8.c | 58 ++++++
> gcc/testsuite/gcc.dg/graphite/interchange-1.c | 4 +-
> gcc/testsuite/gcc.dg/graphite/interchange-11.c | 3 +-
> gcc/testsuite/gcc.dg/graphite/interchange-12.c | 3 +-
> gcc/testsuite/gcc.dg/graphite/interchange-13.c | 54 ++++++
> gcc/testsuite/gcc.dg/graphite/interchange-14.c | 58 ++++++
> gcc/testsuite/gcc.dg/graphite/interchange-15.c | 53 ++++++
> gcc/testsuite/gcc.dg/graphite/interchange-8.c | 5 +-
> gcc/testsuite/gcc.dg/graphite/interchange-mvt.c | 3 +-
> gcc/tree-data-ref.c | 4 +-
> 25 files changed, 573 insertions(+), 175 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/graphite/block-8.c
> create mode 100644 gcc/testsuite/gcc.dg/graphite/interchange-13.c
> create mode 100644 gcc/testsuite/gcc.dg/graphite/interchange-14.c
> create mode 100644 gcc/testsuite/gcc.dg/graphite/interchange-15.c
More information about the Gcc-patches
mailing list