Merge lp:~christophe-lyon/gcc-linaro/4.7-backport-vect-cost-model into lp:gcc-linaro/4.7

[Christophe Lyon]

Backport of the vectorizer cost model, from gcc-trunk.
Currently for benchmarking on our 4.7 branch.

[2013-04-02]
Bench results are avaiable at:
http://cbuild.validation.linaro.org/benchmarks/gcc-linaro-4.7%2bbzr115060~christophe-lyon~4.7-backport-vect-cost-model/logs/

No significant difference, as expected.

The build results are at:
http://cbuild.validation.linaro.org/build/gcc-linaro-4.7%2bbzr115060~christophe-lyon~4.7-backport-vect-cost-model/logs/

[2013-04-02]
Fix Changelog.linaro.

[Linaro Toolchain Builder (cbuild)]

cbuild has taken a snapshot of this branch at r115060 and queued it for build.

The diff against the ancestor r115058 is available at:
 http://cbuild.validation.linaro.org/snapshots/gcc-linaro-4.7+bzr115060~christophe-lyon~4.7-backport-vect-cost-model.diff

and will be built on the following builders:
 a9-builder a9hf-builder armv5-builder i686 x86_64 xaarch64 xaarch64_bare xcortexa15hf

You can track the build queue at:
 http://cbuild.validation.linaro.org/helpers/scheduler

cbuild-snapshot: gcc-linaro-4.7+bzr115060~christophe-lyon~4.7-backport-vect-cost-model
cbuild-ancestor: lp:gcc-linaro+bzr115058
cbuild-state: check

[Christophe Lyon (christophe-lyon)]

review: Needs Resubmitting

[Yvan Roux (yvan-roux)]

Do you know why gcc.c-torture/compile/bcopy-1.c moved from FAIL to PASS on i686 after this backport ?

[Christophe Lyon (christophe-lyon)]

> Do you know why gcc.c-torture/compile/bcopy-1.c moved from FAIL to PASS on
> i686 after this backport ?

No. The changes I introduced are ARM-only.

Maybe another of these x86 tests with varying results we observe sometimes?

[Yvan Roux (yvan-roux)]

* Diff matches upstream, but Changelog.linaro doesn't include the removing of the unrelevant 4.8 code:
  - removing of arm_add_stmt_cost, TARGET_VECTORIZE_ADD_STMT_COST and the handling of the enum vect_cost_for_stmt value vec_construct in arm_builtin_vectorization_cost.

* testsuite is fine (with the usual qemu threading issues on cortexa9hf and cortexa15hf)
* benchmarks are ok

[Yvan Roux (yvan-roux)]

Oups, for cortexa9hf it is not qemu related, but we have some changes in thread simulation tests :

-UNSUPPORTED: gcc.dg/simulate-thread/atomic-load-int.c -O2 -g thread simulation test
+PASS: gcc.dg/simulate-thread/atomic-load-int.c -O2 -g thread simulation test
-PASS: gcc.dg/simulate-thread/atomic-load-longlong.c -O0 -g thread simulation test
+UNSUPPORTED: gcc.dg/simulate-thread/atomic-load-longlong.c -O0 -g thread simulation test

[Yvan Roux (yvan-roux)]

review: Needs Fixing

[Linaro Toolchain Builder (cbuild)]

cbuild has taken a snapshot of this branch at r115061 and queued it for build.

The diff against the ancestor r115058 is available at:
 http://cbuild.validation.linaro.org/snapshots/gcc-linaro-4.7+bzr115061~christophe-lyon~4.7-backport-vect-cost-model.diff

and will be built on the following builders:
 a9-builder a9hf-builder armv5-builder i686 x86_64 xaarch64 xaarch64_bare xcortexa15hf

You can track the build queue at:
 http://cbuild.validation.linaro.org/helpers/scheduler

cbuild-snapshot: gcc-linaro-4.7+bzr115061~christophe-lyon~4.7-backport-vect-cost-model
cbuild-ancestor: lp:gcc-linaro+bzr115058
cbuild-state: check

[Yvan Roux (yvan-roux)]

Approved as the changes are only in the Changelog.

review: Approve