re PR target/56720 (ICE when expanding vcond with floating point unordered comparisons)

2013-03-25  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
PR target/56720
* config/arm/iterators.md (v_cmp_result): New mode attribute.
* config/arm/neon.md (vcond<mode><mode>): Handle unordered cases.

2013-03-25  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
PR target/56720
* gcc.target/arm/neon-vcond-gt.c: New test.
* gcc.target/arm/neon-vcond-ltgt.c: Likewise.
* gcc.target/arm/neon-vcond-unordered.c: Likewise.

From-SVN: r197040

re PR bootstrap/56689 (internal compiler error: in get_loop_body, at cfgloop.c:841)

2013-03-25  Richard Biener  <rguenther@suse.de>

PR tree-optimization/56689
* tree-vrp.c (execute_vrp): Mark loops for fixup if we removed
any edge.

* gcc.dg/torture/pr56689.c: New testcase.

From-SVN: r197039

tree-ssa-loop-im.c (struct mem_ref): Use bitmap_head instead of bitmap.

2013-03-25  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (struct mem_ref): Use bitmap_head instead
of bitmap.
(memory_references): Likewise.
(outermost_indep_loop, mem_ref_alloc, mark_ref_stored,
gather_mem_refs_stmt, record_dep_loop, ref_indep_loop_p_1,
ref_indep_loop_p_2, find_refs_for_sm): Adjust.
(gather_mem_refs_in_loops): Fold into ...
(analyze_memory_references): ... this.  Move initialization
to tree_ssa_lim_initialize.
(fill_always_executed_in): Rename to ...
(fill_always_executed_in_1): ... this.
(fill_always_executed_in): Move contains_call computation to
this new function from ...
(tree_ssa_lim_initialize): ... here.
(tree_ssa_lim): Call fill_always_executed_in.

From-SVN: r197038

* postreload.c (reload_combine): Fix code detecting returns.

From-SVN: r197037

function.c (emit_use_return_register_into_block): On cc0 targets, do not emit the sequence between cc0 setter and user.

* function.c (emit_use_return_register_into_block): On cc0 targets,
do not emit the sequence between cc0 setter and user.

From-SVN: r197036

* g++.dg/torture/20121105-1.C: Adjust for LLP64 targets.

From-SVN: r197034

predicates.md (local_symbolic_operand): Interpret dll-imported symbols as none-local.

        * config/i386/predicates.md (local_symbolic_operand): Interpret
        dll-imported symbols as none-local.

From-SVN: r197033

50594.cc: For mingw-targets use only static libstdc++-version.

* testsuite/18_support/50594.cc: For mingw-targets use only static
libstdc++-version.
* testsuite/19_diagnostics/error_category/operators/equal.cc
* testsuite/19_diagnostics/error_code/cons/1.cc
* testsuite/19_diagnostics/error_code/operators/bool.cc
* testsuite/19_diagnostics/error_code/operators/equal.cc
* testsuite/19_diagnostics/error_code/operators/not_equal.cc
* testsuite/19_diagnostics/error_condition/cons/1.cc
* testsuite/19_diagnostics/error_condition/operators/bool.cc
* testsuite/19_diagnostics/error_condition/operators/equal.cc
* testsuite/19_diagnostics/error_condition/operators/not_equal.cc
* testsuite/23_containers/set/requirements/exception/basic.cc
* testsuite/26_numerics/headers/cmath/c99_classification_macros_c.cc
* testsuite/lib/dg-options.exp (dg-additional-options): New option.

From-SVN: r197032

tree-ssa-loop-im.c (struct depend): Remove.

2013-03-25  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (struct depend): Remove.
(struct lim_aux_data): Make depends a vec of gimples.
(free_lim_aux_data): Adjust.
(add_dependency): Likewise.
(set_level): Likewise.

From-SVN: r197031

re PR rtl-optimization/56434 (document that __attribute__((__malloc__)) assumes returned pointer has BIGGEST_ALIGNMENT)

2013-03-25  Richard Biener  <rguenther@suse.de>

PR middle-end/56434
* calls.c (expand_call): Use MALLOC_ABI_ALIGNMENT to annotate
the pointer returned by calls with ECF_MALLOC set.

From-SVN: r197030

Daily bump.

From-SVN: r197029

MAINTAINERS (Write After Approval): Add myself.

2013-02-24  Tilo Schwarz  <tilo@tilo-schwarz.de>

* MAINTAINERS (Write After Approval): Add myself.

From-SVN: r197025

mmx.md (mov<mode>): Add ?!Ym,r and r,?!Ym alternatives.

* config/i386/mmx.md (mov<mode>): Add ?!Ym,r and r,?!Ym alternatives.

From-SVN: r197024

re PR libstdc++/56170 (Extension debug_allocator seems non-compliant w.r.t. rebind)

PR libstdc++/56170
* include/ext/debug_allocator.h (debug_allocator): Add missing members
to meet allocator requirements.
* testsuite/ext/debug_allocator/56170.cc: New.

From-SVN: r197023

sse.md (mov<mode>): Merge with movv2sf expander using MMXMODE mode iterator.

* config/i386/sse.md (mov<mode>): Merge with movv2sf expander
using MMXMODE mode iterator.
(*move<mode>_internal): Merge with *movv2sf_internal and
*movv2sf_internal_rex64 using MMXMODE mode iterator.

From-SVN: r197021

re PR libfortran/56696 (Formatted (list-directed)  input fails to signal end of record)

2013-03-24  Tobias Burnus  <burnus@net-b.de>

        PR fortran/56696
        * io/list_read.c (read_real): Fix EOF diagnostic.

2013-03-24  Tobias Burnus  <burnus@net-b.de>

        PR fortran/56696
        * gfortran.dg/eof_5.f90: New.

From-SVN: r197019

Avoid nonconst memmodels in libitm's local outdated copy of <atomic> too

This avoids warnings in libitm for non constant memory models,
fixing the bootstrap with -Werror

Passed bootstrap and test on x86_64-linux.

libitm/:

2013-03-23  Andi Kleen  <andi@my.domain.org>

* local_atomic (__always_inline): Add.
(__calculate_memory_order, atomic_thread_fence,
 atomic_signal_fence, test_and_set, clear, store, load,
         exchange, compare_exchange_weak, compare_exchange_strong,
         fetch_add, fetch_sub, fetch_and, fetch_or, fetch_xor):
Add __always_inline to force inlining.

From-SVN: r197018

Avoid non constant memory model uses in libatomic

x86 ends up using non constant memory models for some of the libatomic
functions. These all end up as __ATOMIC_SEQ_CST. Just use this
directly. This avoids a new warning for non constant memory
models, which broke the bootstrap with -Werror

Passed bootstrap and test on x86_64-linux.

libatomic/:

2013-03-23  Andi Kleen  <ak@linux.intel.com>

* gcas.c: (EXACT_INLINE): Use __ATOMIC_SEQ_CST.
* gexch.c: (EXACT_INLINE): Use __ATOMIC_SEQ_CST.
* gload.c: (EXACT_INLINE): Use __ATOMIC_SEQ_CST.
* gstore.c: (EXACT_INLINE): Use __ATOMIC_SEQ_CST.

diff --git a/libatomic/gcas.c b/libatomic/gcas.c
index edbf611..e3d77f3 100644
--- a/libatomic/gcas.c
+++ b/libatomic/gcas.c
@@ -32,7 +32,7 @@
 # define EXACT_INLINE(N) \
   if (C2(HAVE_ATOMIC_CAS_,N)) \
     return __atomic_compare_exchange_n \
-      (PTR(N,mptr), PTR(N,eptr), *PTR(N,dptr), false, smodel, fmodel)
+      (PTR(N,mptr), PTR(N,eptr), *PTR(N,dptr), false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)
 #else
 # define EXACT_INLINE(N)
 #endif
diff --git a/libatomic/gexch.c b/libatomic/gexch.c
index 1999067..c8c8658 100644
--- a/libatomic/gexch.c
+++ b/libatomic/gexch.c
@@ -33,7 +33,7 @@
   if (C2(HAVE_ATOMIC_EXCHANGE_,N)) \
     { \
       *PTR(N,rptr) = __atomic_exchange_n \
- (PTR(N,mptr), *PTR(N,vptr), smodel); \
+ (PTR(N,mptr), *PTR(N,vptr), __ATOMIC_SEQ_CST); \
       return; \
     }
 #else
diff --git a/libatomic/gload.c b/libatomic/gload.c
index df318d5..85865bd 100644
--- a/libatomic/gload.c
+++ b/libatomic/gload.c
@@ -32,7 +32,7 @@
 # define EXACT_INLINE(N, DEST, SRC, DONE) \
   if (C2(HAVE_ATOMIC_LDST_,N)) \
     { \
-      DEST = __atomic_load_n (SRC, smodel); \
+      DEST = __atomic_load_n (SRC, __ATOMIC_SEQ_CST); \
       DONE; \
     }
 #else
diff --git a/libatomic/gstore.c b/libatomic/gstore.c
index d571e58..84f9a8d 100644
--- a/libatomic/gstore.c
+++ b/libatomic/gstore.c
@@ -32,7 +32,7 @@
 # define EXACT_INLINE(N) \
   if (C2(HAVE_ATOMIC_LDST_,N)) \
     { \
-      __atomic_store_n (PTR(N,mptr), *PTR(N,vptr), smodel); \
+      __atomic_store_n (PTR(N,mptr), *PTR(N,vptr), __ATOMIC_SEQ_CST); \
       return; \
     }
 #else

From-SVN: r197017

Daily bump.

From-SVN: r197016

gcse.c (oprs_unchanged_p): Respect flag_gcse_lm.

* gcse.c (oprs_unchanged_p): Respect flag_gcse_lm.
(record_last_mem_set_info): Likewise.

From-SVN: r197012

df-core.c (rest_of_handle_df_initialize): Use XCNEWVEC instead of XNEWVEC followed by memset.

* df-core.c (rest_of_handle_df_initialize): Use XCNEWVEC instead
of XNEWVEC followed by memset.
(df_worklist_dataflow): Use XNEWVEC instead of xmalloc with a cast.

From-SVN: r197011

builtins.exp: Sort targets alphabetically.

2013-03-23  Sebastian Huber  <sebastian.huber@embedded-brains.de>

* gcc.c-torture/execute/builtins/builtins.exp: Sort targets
alphabetically.

From-SVN: r197007

avr.c, [...]: Where applicable...

* config/avr/avr.c, config/bfin/bfin.c, config/c6x/c6x.c,
config/epiphany/epiphany.c, config/frv/frv.c, config/ia64/ia64.c,
config/iq2000/iq2000.c, config/mcore/mcore.c, config/mep/mep.c,
config/mmix/mmix.c, config/pa/pa.c, config/rs6000/rs6000.c,
config/s390/s390.c, config/sparc/sparc.c, config/spu/spu.c,
config/stormy16/stormy16.c, config/v850/v850.c, config/xtensa/xtensa.c,
dwarf2out.c, hw-doloop.c, resource.c, rtl.h : Where applicable, use
the predicates NOTE_P, NONJUMP_INSN_P, JUMP_P, CALL_P, LABEL_P, and
BARRIER_P instead of GET_CODE.

From-SVN: r197005

sparc.c (sparc_emit_probe_stack_range): Fix small inaccuracy in the probing code.

* config/sparc/sparc.c (sparc_emit_probe_stack_range): Fix small
inaccuracy in the probing code.

* config/sparc/sparc.md (ctrapsi4): Add predicate for operand #3.
(ctrapdi4): Likewise.

From-SVN: r197004

calls.c (expand_call): Add missing guard to code handling return of non-BLKmode structures in MSB.

* calls.c (expand_call): Add missing guard to code handling return
of non-BLKmode structures in MSB.
* function.c (expand_function_end): Likewise.

From-SVN: r197003

combine.c (try_combine): Adjust comment.

* combine.c (try_combine): Adjust comment.  Do not add the set of
insn #0 if the destination indirectly is set or dies in insn #2.
Tidy up code to distribute a new note.

From-SVN: r197002

Daily bump.

From-SVN: r196989

re PR c++/56684 ([C++0x] ICE: unexpected expression 'T' of kind template_parm_index)

PR c++/56684
* pt.c (instantiation_dependent_r): Check DECL_INITIAL of VAR_DECL
and CONST_DECL.

From-SVN: r196983

i386.md (*movdi_internal): Set prefix_rex attribute also for alternatives 16 and 17.

* config/i386/i386.md (*movdi_internal): Set prefix_rex attribute
also for alternatives 16 and 17.

From-SVN: r196982

sse.md (*mov<mode>_internal): Merge with *mov<mode>_internal_rex64.

* config/i386/sse.md (*mov<mode>_internal): Merge with
*mov<mode>_internal_rex64.  Use x64 and nox64 isa attributes.
Emit insn template depending on type attribute.  Use
HAVE_AS_IX86_INTERUNIT_MOVQ to handle broken assemblers that require
movd instead of movq mnemonic for interunit moves.  Rewrite mode
attribute calculation.  Remove unit attribute calculation.
Set prefix attribute to maybe_vex for sselog1 and ssemov types.
Set prefix_data16 attribute for DImode ssemov types.
Use Ym instead of y for SSE-MMX conversion alternatives.
Reorder operand constraints.

testsuite/ChangeLog:

* gcc.target/i386/pr22152.c (dg-options): Add -mtune=core2.

From-SVN: r196981

df.h (df_insn_delete): Adjust prototype.

* df.h (df_insn_delete): Adjust prototype.
* emit-rtl.c (remove_insn): Pass a basic block to df_insn_delete
and let it decide whether mark the basic block dirty.
(set_insn_deleted): Only pass INSN_P insns to df_insn_delete.
* df-scan.c (df_insn_info_delete): New helper function, split
off from df_insn_delete.
(df_scan_free_bb_info): Use it.
(df_insn_rescan, df_insn_rescan_all, df_process_deferred_rescans):
Likewise.
(df_insn_delete): Likewise.  Take insn rtx as argument.  Verify
that the insn is actually an insn and it has a non-NULL basic block.
Do not mark basic block dirty if only deleting a DEBUG_INSN.

From-SVN: r196978

df.h (df_insn_delete): Adjust prototype.

* df.h (df_insn_delete): Adjust prototype.
* emit-rtl.c (remove_insn): Pass a basic block to df_insn_delete
and let it decide whether mark the basic block dirty.
(set_insn_deleted): Only pass INSN_P insns to df_insn_delete.
* df-scan.c (df_insn_info_delete): New helper function, split
off from df_insn_delete.
(df_scan_free_bb_info): Use it.
(df_insn_rescan, df_insn_rescan_all, df_process_deferred_rescans):
Likewise.
(df_insn_delete): Likewise.  Take insn rtx as argument.  Verify
that the insn is actually an insn and it has a non-NULL basic block.
Do not mark basic block dirty if only deleting a DEBUG_INSN.

From-SVN: r196977

re PR testsuite/55994 (multiple definition or memset or strlen for builtins tests with LTO options)

2013-03-22  Sebastian Huber  <sebastian.huber@embedded-brains.de>

PR testsuite/55994
* gcc.c-torture/execute/builtins/builtins.exp: Add
-Wl,--allow-multiple-definition for RTEMS targets.

From-SVN: r196970

Fix up ChangeLog entry date.

From-SVN: r196965

re PR other/43620 ([4.3 Release Blocker] Uploading to gnu.org will fail due to automake security issue)

PR other/43620
* configure.ac (AM_INIT_AUTOMAKE): Add no-dist.
* Makefile.in: Regenerated.
* asan/Makefile.in: Regenerated.
* interception/Makefile.in: Regenerated.
* sanitizer_common/Makefile.in: Regenerated.
* tsan/Makefile.in: Regenerated.

From-SVN: r196964

tree-ssa-loop-im.c (struct mem_ref): Remove indep_ref and dep_ref members.

2013-03-22  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (struct mem_ref): Remove indep_ref and
dep_ref members.
(mem_ref_alloc): Do not allocate them.
(refs_independent_p): Do not query or maintain a cache.

From-SVN: r196961

tree-ssa-loop-im.c (memory_references): Drop all_refs_in_loop.

2013-03-22  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (memory_references): Drop all_refs_in_loop.
(gather_mem_refs_in_loops): Do not compute it.
(analyze_memory_references): Do not allocate it.
(tree_ssa_lim_finalize): Do not free it.
(for_all_locs_in_loop): Do not query all_refs_in_loop.

From-SVN: r196960

is-a.h (as_a): Use gcc_checking_assert.

2013-03-22  Richard Biener  <rguenther@suse.de>

* is-a.h (as_a): Use gcc_checking_assert.

From-SVN: r196959

AArch64 MOVK fix to operand 2

From-SVN: r196958

Testcases that should have been committed with previous patch commits

From-SVN: r196957

tree-ssa-loop-im.c (memory_references): Add refs_stored_in_loop bitmaps.

2013-03-22  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (memory_references): Add refs_stored_in_loop
bitmaps.
(gather_mem_refs_in_loops): Perform store accumulation here.
(create_vop_ref_mapping_loop): Remove.
(create_vop_ref_mapping): Likewise.
(analyze_memory_references): Initialize refs_stored_in_loop.
(LOOP_DEP_BIT): New define to map to bits in (in)dep_loop
bitmaps.
(record_indep_loop): Remove.
(record_dep_loop): New function.
(ref_indep_loop_p_1): Adjust to only walk over references
in the loop, not its subloops.
(ref_indep_loop_p): Rename to ...
(ref_indep_loop_p_2): ... this and recurse over the loop tree,
maintaining a more fine-grained cache.
(ref_indep_loop_p): Wrap ref_indep_loop_p_2.
(tree_ssa_lim_finalize): Free refs_stored_in_loop.

From-SVN: r196956

re PR libstdc++/56678 ([c++11] non-uglified name in <chrono>)

2013-03-22  Paolo Carlini  <paolo.carlini@oracle.com>

PR libstdc++/56678
* include/std/chrono (__is_ratio): Uglify T.

From-SVN: r196954

* gennews (files): Add files for GCC 4.8.

From-SVN: r196902

* lang.c (put_decl_node): Don't iterate over end_params_node.

From-SVN: r196901

win64.S: Make use of ffi_closure_win64_inner symbol pc-relative.

* src/x86/win64.S: Make use of ffi_closure_win64_inner
symbol pc-relative.

From-SVN: r196900

tree-ssa-loop-im.c (struct mem_ref_locs): Remove.

2013-03-22  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (struct mem_ref_locs): Remove.
(struct mem_ref): Make accesses_in_loop a vec of a vec of
aggregate mem_ref_loc.
(free_mem_ref_locs): Inline into ...
(memref_free): ... this and adjust.
(mem_ref_alloc): Adjust.
(mem_ref_locs_alloc): Remove.
(record_mem_ref_loc): Adjust.
(get_all_locs_in_loop): Rewrite into ...
(for_all_locs_in_loop): ... this iterator.
(rewrite_mem_ref_loc): New functor.
(rewrite_mem_refs): Use for_all_locs_in_loop.
(sm_set_flag_if_changed): New functor.
(execute_sm_if_changed_flag_set): Use for_all_locs_in_loop.
(ref_always_accessed): New functor.
(ref_always_accessed_p): Use for_all_locs_in_loop.

From-SVN: r196899

cp-tree.h (identifier_p): New.

* cp-tree.h (identifier_p): New.
* call.c: Throughout, call identifier_p insstead of direct
comparaison of TREE_CODE against IDENTIFIER_NODE.
* decl.c: Likewisse.
* decl2.c: Likewise.
* init.c: Likewise.
* mangle.c: Likewise.
* name-lookup.c: Likewise.
* parser.c: Likewise.
* pt.c: Likewise.
* search.c: Likewise.
* semantics.c: Likewise.
* tree.c: Likewise.
* typeck.c: Likewise.
* typeck2.c: Likewise.

From-SVN: r196897

Daily bump.

From-SVN: r196896

tree-pass.h (PROP_gimple_lvec): New.

2013-03-21  Marc Glisse  <marc.glisse@inria.fr>

* tree-pass.h (PROP_gimple_lvec): New.
* passes.c (dump_properties): Handle PROP_gimple_lvec.
(init_optimization_passes): Move pass_lower_vector.
* tree-vect-generic.c (gate_expand_vector_operations_ssa): Test
PROP_gimple_lvec.
(pass_lower_vector): Provide PROP_gimple_lvec.
(pass_lower_vector_ssa): Likewise.
* cfgexpand.c (pass_expand): Require PROP_gimple_lvec.

From-SVN: r196890

dwarf2out.c (size_of_aranges): Skip DECL_IGNORED_P functions.

From-SVN: r196889

i386.md (*movdi_internal): Disparage slightly all MMX moves to/from memory.

* config/i386/i386.md (*movdi_internal): Disparage slightly
all MMX moves to/from memory.  Use Yi instead of x for SSE-MMX
conversion alternatives.

From-SVN: r196888

re PR middle-end/48087 (-Wall -Werror adds warnings over and above those generated by -Wall)

PR middle-end/48087
* diagnostic.def (DK_WERROR): New kind.
* diagnostic.h (werrorcount): Define.
* diagnostic.c (diagnostic_report_diagnostic): For DK_WARNING
promoted to DK_ERROR, increment DK_WERROR counter instead of
DK_ERROR counter.
* toplev.c (toplev_main): Call print_ignored_options even if
just werrorcount is non-zero.  Exit with FATAL_EXIT_CODE
even if just werrorcount is non-zero.

* pt.c (convert_nontype_argument): Count werrorcount as warnings.
* call.c (build_temp): Likewise.
* method.c (synthesize_method): Likewise.
* typeck.c (convert_for_initialization): Likewise.

From-SVN: r196887

re PR debug/55608 (Debug info quality regressions with file scope vars)

PR debug/55608
* dwarf2out.c (tree_add_const_value_attribute): Call ggc_free (array)
on failure.
(resolve_one_addr): Fail if referenced STRING_CST hasn't been written.
(string_cst_pool_decl): New function.
(optimize_one_addr_into_implicit_ptr): New function.
(resolve_addr_in_expr): Optimize DWARF location expression
DW_OP_addr DW_OP_stack_value where DW_OP_addr refers to some variable
which doesn't live in memory, but has DW_AT_location or
DW_AT_const_value, or refers to a string literal, into
DW_OP_GNU_implicit_pointer.
(optimize_location_into_implicit_ptr): New function.
(resolve_addr): If removing DW_AT_location of a variable because
it was DW_OP_addr of address of the variable, but the variable doesn't
live in memory, try to emit const value attribute for the initializer.

From-SVN: r196886

correct changelog

From-SVN: r196885

tree.h (VECTOR_TYPE_P): New macro.

2013-03-21  Marc Glisse  <marc.glisse@inria.fr>

gcc/
* tree.h (VECTOR_TYPE_P): New macro.
(VECTOR_INTEGER_TYPE_P, VECTOR_FLOAT_TYPE_P, FLOAT_TYPE_P,
TYPE_MODE): Use it.
* fold-const.c (fold_cond_expr_with_comparison): Use build_zero_cst.
VEC_COND_EXPR cannot be lvalues.
(fold_ternary_loc) <VEC_COND_EXPR>: Merge with the COND_EXPR case.

gcc/cp/
* call.c (build_conditional_expr_1): Fold VEC_COND_EXPR.

gcc/testsuite/
* g++.dg/ext/vector21.C: New testcase.

From-SVN: r196884

simplify-rtx.c (simplify_binary_operation_1): Restrict the transformation to equal modes.

2013-03-21  Marc Glisse  <marc.glisse@inria.fr>

* simplify-rtx.c (simplify_binary_operation_1) <VEC_CONCAT>:
Restrict the transformation to equal modes.

From-SVN: r196882

Add forgotten ChangeLog line.

From-SVN: r196879

re PR middle-end/39326 (Segmentation fault with -O1, out of memory with -O2)

2013-03-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/39326
* tree-ssa-loop-im.c (UNANALYZABLE_MEM_ID): New define.
(MEM_ANALYZABLE): Adjust.
(record_mem_ref_loc): Move bitmap ops ...
(gather_mem_refs_stmt): ... here.  Use the shared mem-ref for
unanalyzable refs, do not record locations for it.
(analyze_memory_references): Allocate ref zero as shared
unanalyzable ref.
(refs_independent_p): Do not test for unanalyzed mems here.
(ref_indep_loop_p_1): Special-case disambiguation against
the unanalyzed ref.
(ref_indep_loop_p): Assert we are not queried for the
unanalyzed mem.

From-SVN: r196878

[ARM] Turning off 64bits ops in Neon

2013-03-21  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/arm/arm-protos.h (tune_params): Add
prefer_neon_for_64bits field.
* config/arm/arm.c (prefer_neon_for_64bits): New variable.
(arm_slowmul_tune): Default prefer_neon_for_64bits to false.
(arm_fastmul_tune, arm_strongarm_tune, arm_xscale_tune): Ditto.
(arm_9e_tune, arm_v6t2_tune, arm_cortex_tune): Ditto.
(arm_cortex_a15_tune, arm_cortex_a5_tune): Ditto.
(arm_cortex_a9_tune, arm_v6m_tune, arm_fa726te_tune): Ditto.
(arm_option_override): Handle -mneon-for-64bits new option.
* config/arm/arm.h (TARGET_PREFER_NEON_64BITS): New macro.
(prefer_neon_for_64bits): Declare new variable.
* config/arm/arm.md (arch): Rename neon_onlya8 and neon_nota8 to
avoid_neon_for_64bits and neon_for_64bits. Remove onlya8 and
nota8.
(arch_enabled): Handle new arch types. Remove support for onlya8
and nota8.
(one_cmpldi2): Use new arch names.
* config/arm/arm.opt (mneon-for-64bits): Add option.
* config/arm/neon.md (adddi3_neon, subdi3_neon, iordi3_neon)
(anddi3_neon, xordi3_neon, ashldi3_neon, <shift>di3_neon): Use
neon_for_64bits instead of nota8 and avoid_neon_for_64bits instead
of onlya8.
* doc/invoke.texi (-mneon-for-64bits): Document.

gcc/testsuite:
* gcc.target/arm/neon-for-64bits-1.c: New tests.
* gcc.target/arm/neon-for-64bits-2.c: Likewise.

From-SVN: r196876

re PR middle-end/39326 (Segmentation fault with -O1, out of memory with -O2)

2013-03-21  Richard Biener  <rguenther@suse.de>

PR tree-optimization/39326
* tree-ssa-loop-im.c (bb_loop_postorder): New global static.
(sort_bbs_in_loop_postorder_cmp): New function.
(gather_mem_refs_in_loops): Assign mem-ref IDs in loop
postorder.

From-SVN: r196874

tree-vect-data-refs.c (vect_update_interleaving_chain): Remove.

2013-03-21  Richard Biener  <rguenther@suse.de>

* tree-vect-data-refs.c (vect_update_interleaving_chain): Remove.
(vect_insert_into_interleaving_chain): Likewise.
(vect_drs_dependent_in_basic_block): Inline ...
(vect_slp_analyze_data_ref_dependence): ... here.  New function,
split out from ...
(vect_analyze_data_ref_dependence): ... here.  Simplify.
(vect_check_interleaving): Simplify.
(vect_analyze_data_ref_dependences): Likewise.  Split out ...
(vect_slp_analyze_data_ref_dependences): ... this new function.
(dr_group_sort_cmp): New function.
(vect_analyze_data_ref_accesses): Compute data-reference groups
here instead of in vect_analyze_data_ref_dependence.  Use
a more efficient algorithm.
* tree-vect-slp.c (vect_slp_analyze_bb_1): Use
vect_slp_analyze_data_ref_dependences.  Call
vect_analyze_data_ref_accesses earlier.
* tree-vect-loop.c (vect_analyze_loop_2): Likewise.
* tree-vectorizer.h (vect_analyze_data_ref_dependences): Adjust.
(vect_slp_analyze_data_ref_dependences): New prototype.

* gcc.dg/vect/vect-outer-3a-big-array.c: Adjust.
* gcc.dg/vect/vect-outer-3a.c: Likewise.

From-SVN: r196872

* ChangeLog: Fix whitespace.

From-SVN: r196871

tree-ssa-loop-im.c (can_sm_ref_p): Do not test whether ref is stored in the loop.

2013-03-21  Richard Biener  <rguenther@suse.de>

* tree-ssa-loop-im.c (can_sm_ref_p): Do not test whether
ref is stored in the loop.
(find_refs_for_sm): Walk only over all stores.
(store_motion_loop): Allocate from lim_bitmap_obstack.
(store_motion): Likewise.

From-SVN: r196870

tree-vect-loop-manip.c (slpeel_tree_peel_loop_to_edge): Update virtual SSA form.

2013-03-21  Richard Biener  <rguenther@suse.de>

* tree-vect-loop-manip.c (slpeel_tree_peel_loop_to_edge):
Update virtual SSA form.

From-SVN: r196868

Default to DWARF 4 on Solaris if linker supports CIEv3

* configure.ac (gcc_cv_ld_eh_frame_ciev3): New test.
* configure: Regenerate.
* config.in: Regenerate.
* config/sol2.c (solaris_override_options): Only enforce DWARF 2
if !HAVE_LD_EH_FRAME_CIEV3.

From-SVN: r196866

tree-cfg.c (verify_expr_no_block): New function.

2013-03-21  Richard Biener  <rguenther@suse.de>

* tree-cfg.c (verify_expr_no_block): New function.
(verify_expr_location_1): Verify that neither DECL_DEBUG_EXPR
nor DECL_VALUE_EXPR have locations with associated blocks.
* tree-ssa-live.c (clear_unused_block_pointer_1): Remove.
(clear_unused_block_pointer): Remove code dealing with
blocks in DECL_DEBUG_EXPR locations.

From-SVN: r196865

tree.h (DECL_DEBUG_EXPR_IS_FROM): Rename to ...

2013-03-21  Richard Biener  <rguenther@suse.de>

* tree.h (DECL_DEBUG_EXPR_IS_FROM): Rename to ...
(DECL_HAS_DEBUG_EXPR_P): ... this.  Guard properly.
* tree.c (copy_node_stat): Do not copy DECL_HAS_DEBUG_EXPR_P.
* dwarf2out.c (add_var_loc_to_decl): Use DECL_HAS_DEBUG_EXPR_P
instead of DECL_DEBUG_EXPR_IS_FROM.
* gimplify.c (gimplify_modify_expr): Likewise.
* tree-cfg.c (verify_expr_location_1): Likewise.
* tree-complex.c (create_one_component_var): Likewise.
* tree-sra.c (create_access_replacement): Likewise.
* tree-ssa-live.c (clear_unused_block_pointer_1): Likewise.
(clear_unused_block_pointer): Likewise.
* tree-streamer-in.c (unpack_ts_decl_common_value_fields): Likewise.
* tree-streamer-out.c (pack_ts_decl_common_value_fields): Likewise.
* var-tracking.c (var_debug_decl): Likewise.
(track_expr_p): Likewise.
* tree-inline.c (add_local_variables): Likewise.  Set
DECL_HAS_DEBUG_EXPR_P after copying it.
* tree-diagnostic.c (default_tree_printer): Use DECL_HAS_DEBUG_EXPR_P
instead of DECL_DEBUG_EXPR_IS_FROM.  Guard properly.

c/
* c-objc-common.c (c_tree_printer): Use DECL_HAS_DEBUG_EXPR_P
instead of DECL_DEBUG_EXPR_IS_FROM.  Guard properly.

cp/
* error.c (cp_printer): Use DECL_HAS_DEBUG_EXPR_P instead of
DECL_DEBUG_EXPR_IS_FROM.  Guard properly.

From-SVN: r196864

re PR bootstrap/56656 (Suffix or operands invalid for 'movq')

PR bootstrap/56656
* configure.ac (HAVE_AS_IX86_INTERUNIT_MOVQ): New test.
* configure: Regenerate.
* config.in: Regenerate.
* config/i386/i386.md (*movdf_internal): Use
HAVE_AS_IX86_INTERUNIT_MOVQ to handle broken assemblers that require
movd instead of movq mnemonic for interunit moves.
(*movdi_internal): Ditto.

From-SVN: r196861

aarch64-simd.md (simd_fabd): New Attribute.

2013-03-21   Naveen H.S  <Naveen.Hurugalawadi@caviumnetworks.com>

* config/aarch64/aarch64-simd.md (simd_fabd): New Attribute.
(abd<mode>_3): New pattern.
(aba<mode>_3): New pattern.
(fabd<mode>_3): New pattern.

2013-03-21   Naveen H.S  <Naveen.Hurugalawadi@caviumnetworks.com>

* gcc.target/aarch64/vect.c: Test and result vector added
for sabd and saba instructions.
* gcc.target/aarch64/vect-compile.c: Check for sabd and saba
instructions in assembly.
* gcc.target/aarch64/vect.x: Add sabd and saba test functions.
* gcc.target/aarch64/vect-fp.c: Test and result vector added
for fabd instruction.
* gcc.target/aarch64/vect-fp-compile.c: Check for fabd
instruction in assembly.
* gcc.target/aarch64/vect-fp.x: Add fabd test function.

From-SVN: r196858

aarch64-elf.h (REGISTER_PREFIX): Remove.

2013-03-21   Naveen H.S  <Naveen.Hurugalawadi@caviumnetworks.com>

* config/aarch64/aarch64-elf.h (REGISTER_PREFIX): Remove.
* config/aarch64/aarch64.c (aarch64_print_operand): Remove all
occurrence of REGISTER_PREFIX as its empty string.

From-SVN: r196856

tree-ssa-dom.c (record_equivalences_from_incoming_edge): Record addititional equivalences for equality comparisons between an SSA_NAME...

* tree-ssa-dom.c (record_equivalences_from_incoming_edge): Record
addititional equivalences for equality comparisons between an SSA_NAME
and a constant where the SSA_NAME was set from a widening conversion.

* g++.dg/tree-ssa/ssa-dom.C: New test.

From-SVN: r196855

re PR c++/56646 (ICE: in cp_parser_late_return_type_opt, at cp/parser.c:16970)

PR c++/56646
* parser.c (cp_parser_late_return_type_opt): Save and restore
current_class_ptr/ref.

From-SVN: r196853

re PR c++/54532 ([C++0x][constexpr] internal error when initializing static constexpr with pointer to non-static member variable)

PR c++/54532
* expr.c (cplus_expand_constant): Do nothing if the class is
incomplete.
* semantics.c (reduced_constant_expression_p): Allow PTRMEM_CST.
* typeck2.c (store_init_value): Use reduced_constant_expression_p.
* decl.c (maybe_register_incomplete_var): Handle PTRMEM_CST.
(complete_vars): Likewise.

From-SVN: r196852

name-lookup.c (get_anonymous_namespace_name): Never use get_file_function_name.

* name-lookup.c (get_anonymous_namespace_name): Never use
get_file_function_name.

From-SVN: r196851

* pt.c (retrieve_specialization): Handle null tmpl argument.

From-SVN: r196850

re PR c++/17232 ([DR 1640] classes and class template specializations treated differently w.r.t. core issue #337)

PR c++/17232
PR c++/56642
* pt.c (tsubst_decl): Check return value of register_specialization.
* typeck2.c (abstract_virtuals_error_sfinae): Re-apply complete_type
change.

From-SVN: r196849

Add an atomic test and set pattern on tilegx.

      * config/tilegx/sync.md (atomic_test_and_set): New pattern.

From-SVN: r196848

Daily bump.

From-SVN: r196847

2013-03-20  Robert Mason  <rbmj@verizon.net>

* config/vxlib-tls.c (__gthread_get_tsd_data,)
(__gthread_set_tsd_data, __gthread_enter_tsd_dtor_context,)
(__gthread_leave_tsd_dtor_context): Add prototypes.
(tls_delete_hook): Update.

From-SVN: r196842

i386.md (*movoi_internal_avx): Emit insn template depending on type attribute.

* config/i386/i386.md (*movoi_internal_avx): Emit insn template
depending on type attribute.
(*movti_internal): Ditto.
(*movtf_internal): Ditto.
(*movxf_internal): Ditto.
(*movdf_internal): Ditto.
(*movsf_internal): Ditto.

From-SVN: r196841

i386.md (*movti_internal): Set prefix attribute to maybe_vex for sselog1 and ssemov types.

* config/i386/i386.md (*movti_internal): Set prefix attribute to
maybe_vex for sselog1 and ssemov types.
(*movdi_internal): Reorder operand constraints.
(*movsi_internal): Ditto.  Set prefix attribute to
maybe_vex for sselog1 and ssemov types.
(*movtf_internal): Set prefix attribute to maybe_vex
for sselog1 and ssemov types.
(*movdf_internal): Ditto.  Set prefix_data16 attribute for
DImode ssemov types.  Reorder operand constraints.
(*movsf_internal): Set type of alternatives 3,4 to imov.  Set prefix
attribute to maybe_vex for sselog1 and ssemov types.  Set prefix_data16
attribute for SImode ssemov types.  Reorder operand constraints.

From-SVN: r196834

params.def (PARAM_IPA_CP_ARRAY_INDEX_HINT_BONUS): New parameter.

2013-03-20  Martin Jambor  <mjambor@suse.cz>

* params.def (PARAM_IPA_CP_ARRAY_INDEX_HINT_BONUS): New parameter.
* ipa-cp.c (hint_time_bonus): Add abonus for known array indices.

From-SVN: r196832

predicates.md (indexed_address, [...]): New predicates.

[gcc]
2013-03-20  Pat Haugen <pthaugen@us.ibm.com>

* config/rs6000/predicates.md (indexed_address, update_address_mem
update_indexed_address_mem): New predicates.
* config/rs6000/vsx.md (vsx_extract_<mode>_zero): Set correct "type"
attribute for load/store instructions.
* config/rs6000/dfp.md (movsd_store): Likewise.
(movsd_load): Likewise.
* config/rs6000/rs6000.md (zero_extend<mode>di2_internal1): Likewise.
(unnamed HI->DI extend define_insn): Likewise.
(unnamed SI->DI extend define_insn): Likewise.
(unnamed QI->SI extend define_insn): Likewise.
(unnamed QI->HI extend define_insn): Likewise.
(unnamed HI->SI extend define_insn): Likewise.
(unnamed HI->SI extend define_insn): Likewise.
(extendsfdf2_fpr): Likewise.
(movsi_internal1): Likewise.
(movsi_internal1_single): Likewise.
(movhi_internal): Likewise.
(movqi_internal): Likewise.
(movcc_internal1): Correct mnemonic for stw insn. Set correct "type"
attribute for load/store instructions.
(mov<mode>_hardfloat): Set correct "type" attribute for load/store
instructions.
(mov<mode>_softfloat): Likewise.
(mov<mode>_hardfloat32): Likewise.
(mov<mode>_hardfloat64): Likewise.
(mov<mode>_softfloat64): Likewise.
(movdi_internal32): Likewise.
(movdi_internal64): Likewise.
(probe_stack_<mode>): Likewise.

2013-03-20  Michael Meissner  <meissner@linux.vnet.ibm.com>

* config/rs6000/vector.md (VEC_R): Add 32-bit integer, binary
floating point, and decimal floating point to reload iterator.

* config/rs6000/constraints.md (wl constraint): New constraints to
return FLOAT_REGS if certain options are used to reduce the number
of separate patterns that exist in the file.
(wx constraint): Likewise.
(wz constraint): Likewise.

* config/rs6000/rs6000.c (rs6000_debug_reg_global): If
-mdebug=reg, print wg, wl, wx, and wz constraints.
(rs6000_init_hard_regno_mode_ok): Initialize new constraints.
Initialize the reload functions for 64-bit binary/decimal floating
point types.
(reg_offset_addressing_ok_p): If we are on a power7 or later, use
LFIWZX and STFIWX to load/store 32-bit decimal types, and don't
create the buffer on the stack to overcome not having a 32-bit
load and store.
(rs6000_emit_move): Likewise.
(rs6000_secondary_memory_needed_rtx): Likewise.
(rs6000_alloc_sdmode_stack_slot): Likewise.
(rs6000_preferred_reload_class): On VSX, we can create SFmode 0.0f
via xxlxor, just like DFmode 0.0.

* config/rs6000/rs6000.h (TARGET_NO_SDMODE_STACK): New macro,
define as 1 if we are running on a power7 or newer.
(enum r6000_reg_class_enum): Add new constraints.

* config/rs6000/dfp.md (movsd): Delete, combine with binary
floating point moves in rs6000.md.  Combine power6x (mfpgpr) moves
with other moves by using conditional constraits (wg).  Use LFIWZX
and STFIWX for loading SDmode on power7.  Use xxlxor to create
0.0f.
(movsd splitter): Likewise.
(movsd_hardfloat): Likewise.
(movsd_softfloat): Likewise.

* config/rs6000/rs6000.md (FMOVE32): New iterators to combine
binary and decimal floating point moves.
(fmove_ok): New attributes to combine binary and decimal floating
point moves, and to combine power6x (mfpgpr) moves along normal
floating moves.
(real_value_to_target): Likewise.
(f32_lr): Likewise.
(f32_lm): Likewise.
(f32_li): Likewise.
(f32_sr): Likewise.
(f32_sm): Likewise.
(f32_si): Likewise.
(movsf): Combine binary and decimal floating point moves.  Combine
power6x (mfpgpr) moves with other moves by using conditional
constraits (wg).  Use LFIWZX and STFIWX for loading SDmode on
power7.
(mov<mode> for SFmode/SDmode); Likewise.
(SFmode/SDmode splitters): Likewise.
(movsf_hardfloat): Likewise.
(mov<mode>_hardfloat for SFmode/SDmode): Likewise.
(movsf_softfloat): Likewise.
(mov<mode>_softfloat for SFmode/SDmode): Likewise.

* doc/md.texi (PowerPC and IBM RS6000 constraints): Document wl,
wx and wz constraints.

* config/rs6000/constraints.md (wg constraint): New constraint to
return FLOAT_REGS if -mmfpgpr (power6x) was used.

* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Add wg
constraint.

* config/rs6000/rs6000.c (rs6000_debug_reg_global): If
-mdebug=reg, print wg, wl, wx, and wz constraints.
(rs6000_init_hard_regno_mode_ok): Initialize new constraints.
Initialize the reload functions for 64-bit binary/decimal floating
point types.
(reg_offset_addressing_ok_p): If we are on a power7 or later, use
LFIWZX and STFIWX to load/store 32-bit decimal types, and don't
create the buffer on the stack to overcome not having a 32-bit
load and store.
(rs6000_emit_move): Likewise.
(rs6000_secondary_memory_needed_rtx): Likewise.
(rs6000_alloc_sdmode_stack_slot): Likewise.
(rs6000_preferred_reload_class): On VSX, we can create SFmode 0.0f
via xxlxor, just like DFmode 0.0.

* config/rs6000/dfp.md (movdd): Delete, combine with binary
floating point moves in rs6000.md.  Combine power6x (mfpgpr) moves
with other moves by using conditional constraits (wg).  Use LFIWZX
and STFIWX for loading SDmode on power7.
(movdd splitters): Likewise.
(movdd_hardfloat32): Likewise.
(movdd_softfloat32): Likewise.
(movdd_hardfloat64_mfpgpr): Likewise.
(movdd_hardfloat64): Likewise.
(movdd_softfloat64): Likewise.

* config/rs6000/rs6000.md (FMOVE64): New iterators to combine
64-bit binary and decimal floating point moves.
(FMOVE64X): Likewise.
(movdf): Combine 64-bit binary and decimal floating point moves.
Combine power6x (mfpgpr) moves with other moves by using
conditional constraits (wg).
(mov<mode> for DFmode/DDmode): Likewise.
(DFmode/DDmode splitters): Likewise.
(movdf_hardfloat32): Likewise.
(mov<mode>_hardfloat32 for DFmode/DDmode): Likewise.
(movdf_softfloat32): Likewise.
(movdf_hardfloat64_mfpgpr): Likewise.
(movdf_hardfloat64): Likewise.
(mov<mode>_hardfloat64 for DFmode/DDmode): Likewise.
(movdf_softfloat64): Likewise.
(mov<mode>_softfloat64 for DFmode/DDmode): Likewise.
(reload_<mode>_load): Move to later in the file so they aren't in
the middle of the floating point move insns.
(reload_<mode>_store): Likewise.

* doc/md.texi (PowerPC and IBM RS6000 constraints): Document wg
constraint.

* config/rs6000/rs6000.c (rs6000_debug_reg_global): Print out wg
constraint if -mdebug=reg.
(rs6000_initi_hard_regno_mode_ok): Enable wg constraint if
-mfpgpr.  Enable using dd reload support if needed.

* config/rs6000/dfp.md (movtd): Delete, combine with 128-bit
binary and decimal floating point moves in rs6000.md.
(movtd_internal): Likewise.

* config/rs6000/rs6000.md (FMOVE128): Combine 128-bit binary and
decimal floating point moves.
(movtf): Likewise.
(movtf_internal): Likewise.
(mov<mode>_internal, TDmode/TFmode): Likewise.
(movtf_softfloat): Likewise.
(mov<mode>_softfloat, TDmode/TFmode): Likewise.

* config/rs6000/rs6000.md (movdi_mfpgpr): Delete, combine with
movdi_internal64, using wg constraint for move direct operations.
(movdi_internal64): Likewise.

* config/rs6000/rs6000.c (rs6000_debug_reg_global): Print
MODES_TIEABLE_P for selected modes.  Print the numerical value of
the various virtual registers. Use GPR/FPR first/last values,
instead of hard coding the register numbers.  Print which modes
have reload functions registered.
(rs6000_option_override_internal): If -mdebug=reg, trace the
options settings before/after setting cpu, target and subtarget
settings.
(rs6000_secondary_reload_trace): Improve the RTL dump for
-mdebug=addr and for secondary reload failures in
rs6000_secondary_reload_inner.
(rs6000_secondary_reload_fail): Likewise.
(rs6000_secondary_reload_inner): Likewise.

* config/rs6000/rs6000.md (FIRST_GPR_REGNO): Add convenience
macros for first/last GPR and FPR registers.
(LAST_GPR_REGNO): Likewise.
(FIRST_FPR_REGNO): Likewise.
(LAST_FPR_REGNO): Likewise.

* config/rs6000/vector.md (mul<mode>3): Use the combined macro
VECTOR_UNIT_ALTIVEC_OR_VSX_P instead of separate calls to
VECTOR_UNIT_ALTIVEC_P and VECTOR_UNIT_VSX_P.
(vcond<mode><mode>): Likewise.
(vcondu<mode><mode>): Likewise.
(vector_gtu<mode>): Likewise.
(vector_gte<mode>): Likewise.
(xor<mode>3): Don't allow logical operations on TImode in 32-bit
to prevent the compiler from converting DImode operations to
TImode.
(ior<mode>3): Likewise.
(and<mode>3): Likewise.
(one_cmpl<mode>2): Likewise.
(nor<mode>3): Likewise.
(andc<mode>3): Likewise.

* config/rs6000/constraints.md (wt constraint): New constraint
that returns VSX_REGS if TImode is allowed in VSX registers.

* config/rs6000/predicates.md (easy_fp_constant): 0.0f is an easy
constant under VSX.

* config/rs6000/rs6000-modes.def (PTImode): Define, PTImode is
similar to TImode, but it is restricted to being in the GPRs.

* config/rs6000/rs6000.opt (-mvsx-timode): New switch to allow
TImode to occupy a single VSX register.

* config/rs6000/rs6000-cpus.def (ISA_2_6_MASKS_SERVER): Default to
-mvsx-timode for power7/power8.
(power7 cpu): Likewise.
(power8 cpu): Likewise.

* config/rs6000/rs6000.c (rs6000_hard_regno_nregs_internal): Make
sure that TFmode/TDmode take up two registers if they are ever
allowed in the upper VSX registers.
(rs6000_hard_regno_mode_ok): If -mvsx-timode, allow TImode in VSX
registers.
(rs6000_init_hard_regno_mode_ok): Likewise.
(rs6000_debug_reg_global): Add debugging for PTImode and wt
constraint.  Print if LRA is turned on.
(rs6000_option_override_internal): Give an error if -mvsx-timode
and VSX is not enabled.
(invalid_e500_subreg): Handle PTImode, restricting it to GPRs.  If
-mvsx-timode, restrict TImode to reg+reg addressing, and PTImode
to reg+offset addressing.  Use PTImode when checking offset
addresses for validity.
(reg_offset_addressing_ok_p): Likewise.
(rs6000_legitimate_offset_address_p): Likewise.
(rs6000_legitimize_address): Likewise.
(rs6000_legitimize_reload_address): Likewise.
(rs6000_legitimate_address_p): Likewise.
(rs6000_eliminate_indexed_memrefs): Likewise.
(rs6000_emit_move): Likewise.
(rs6000_secondary_reload): Likewise.
(rs6000_secondary_reload_inner): Handle PTImode.  Allow 64-bit
reloads to fpr registers to continue to use reg+offset addressing,
but 64-bit reloads to altivec registers need reg+reg addressing.
Drop test for PRE_MODIFY, since VSX loads/stores no longer support
it.  Treat LO_SUM like a PLUS operation.
(rs6000_secondary_reload_class): If type is 64-bit, prefer to use
FLOAT_REGS instead of VSX_RGS to allow use of reg+offset
addressing.
(rs6000_cannot_change_mode_class): Do not allow TImode in VSX
registers to share a register with a smaller sized type, since VSX
puts scalars in the upper 64-bits.
(print_operand): Add support for PTImode.
(rs6000_register_move_cost): Use VECTOR_MEM_VSX_P instead of
VECTOR_UNIT_VSX_P to catch types that can be loaded in VSX
registers, but don't have arithmetic support.
(rs6000_memory_move_cost): Add test for VSX.
(rs6000_opt_masks): Add -mvsx-timode.

* config/rs6000/vsx.md (VSm): Change to use 64-bit aligned moves
for TImode.
(VSs): Likewise.
(VSr): Use wt constraint for TImode.
(VSv): Drop TImode support.
(vsx_movti): Delete, replace with versions for 32-bit and 64-bit.
(vsx_movti_64bit): Likewise.
(vsx_movti_32bit): Likewise.
(vec_store_<mode>): Use VSX iterator instead of vector iterator.
(vsx_and<mode>3): Delete use of '?' constraint on inputs, just put
one '?' on the appropriate output constraint.  Do not allow TImode
logical operations on 32-bit systems.
(vsx_ior<mode>3): Likewise.
(vsx_xor<mode>3): Likewise.
(vsx_one_cmpl<mode>2): Likewise.
(vsx_nor<mode>3): Likewise.
(vsx_andc<mode>3): Likewise.
(vsx_concat_<mode>): Likewise.
(vsx_xxpermdi_<mode>): Fix thinko for non V2DF/V2DI modes.

* config/rs6000/rs6000.h (MASK_VSX_TIMODE): Map from
OPTION_MASK_VSX_TIMODE.
(enum rs6000_reg_class_enum): Add RS6000_CONSTRAINT_wt.
(STACK_SAVEAREA_MODE): Use PTImode instead of TImode.

* config/rs6000/rs6000.md (INT mode attribute): Add PTImode.
(TI2 iterator): New iterator for TImode, PTImode.
(wd mode attribute): Add values for vector types.
(movti_string): Replace TI move operations with operations for
TImode and PTImode.  Add support for TImode being allowed in VSX
registers.
(mov<mode>_string, TImode/PTImode): Likewise.
(movti_ppc64): Likewise.
(mov<mode>_ppc64, TImode/PTImode): Likewise.
(TI mode splitters): Likewise.

* doc/md.texi (PowerPC and IBM RS6000 constraints): Document wt
constraint.

[gcc/testsuite]
2013-03-20  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/mmfpgpr.c: New test.
* gcc.target/powerpc/sd-vsx.c: Likewise.
* gcc.target/powerpc/sd-pwr6.c: Likewise.
* gcc.target/powerpc/vsx-float0.c: Likewise.

From-SVN: r196831

re PR tree-optimization/56355 (abs and multiplication)

2013-03-20  Marc Glisse  <marc.glisse@inria.fr>

PR tree-optimization/56355
gcc/
* fold-const.c (tree_binary_nonnegative_warnv_p) <MULT_EXPR>:
Also handle integers with undefined overflow.

gcc/testsuite/
* gcc.dg/pr56355-1.c: New file.

From-SVN: r196829

extend.texi: (micromips, nomicromips, nocompression): Document new function attributes.

gcc/
2013-03-20  Catherine Moore  <clm@codesourcery.com>
    Maciej W. Rozycki  <macro@codesourcery.com>
    Tom de Vries  <tom@codesourcery.com>
    Nathan Sidwell <nathan@codesourcery.com>
    Iain Sandoe  <iain@codesourcery.com>
    Nathan Froyd  <froydnj@codesourcery.com>
    Chao-ying Fu <fu@mips.com>

* doc/extend.texi: (micromips, nomicromips, nocompression):
Document new function attributes.
* doc/invoke.texi (minterlink-compressed, mmicromips,
m14k, m14ke, m14kec): Document new options.
(minterlink-mips16): Update documentation.
* doc/md.texi (ZC, ZD): Document new constraints.
* configure.ac (gcc_cv_as_micromips): Check if linker
supports the .set micromips directive.
* configure: Regenerate.
* config.in: Regenerate.
* config/mips/mips-tables.opt: Regenerate.
* config/mips/micromips.md: New file.
* constraints.md (ZC, ZD): New constraints.
* config/mips/predicates.md (movep_src_register): New predicate.
(movep_src_operand): New predicate.
(non_volatile_mem_operand): New predicate.
* config/mips/mips.md (multimem): New type.
(length): Differentiate between 17-bit and 18-bit branch offsets.
(MOVEP1, MOVEP2): New mode iterator.
  (mov_<load>l): Use ZC constraint.
(mov_<load>r): Likewise.
(mov_<store>l): Likewise.
(mov_<store>r): Likewise.
(*branch_equality<mode>_inverted): Add microMIPS support.
(*branch_equality<mode>): Likewise.
(*jump_absolute): Likewise.
(indirect_jump_<mode>): Likewise.
(tablejump_<mode>): Likewise.
(<optab>_internal): Likewise.
(sibcall_internal): Likewise.
(sibcall_value_internal): Likewise.
(prefetch): Use constraint ZD.
* config/mips/mips.opt (minterlink-compressed): New option.
(minterlink-mips16): Now an alias for minterlink-compressed.
(mmicromips): New option.
* config/mips/sync.md (sync_compare_and_swap<mode>): Use ZR constraint.
(compare_and_swap_12): Likewise.
(sync_add<mode>): Likewise.
(sync_<optab>_12): Likewise.
(sync_old_<optab>_12): Likewise.
(sync_new_<optab>_12): Likewise.
(sync_nand_12): Likewise.
(sync_old_nand_12): Likewise.
(sync_new_nand_12): Likewise.
(sync_sub<mode>): Likewise.
(sync_old_add<mode>): Likewise.
(sync_old_sub<mode>): Likewise.
(sync_new_add<mode>): Likewise.
(sync_new_sub<mode>): Likewise.
(sync_<optab><mode>): Likewise.
(sync_old_<optab><mode>): Likewise.
(sync_new_<optab><mode>): Likewise.
(sync_nand<mode>): Likewise.
(sync_old_nand<mode>): Likewise.
(sync_new_nand<mode>): Likewise.
(sync_lock_test_and_set<mode>): Likewise.
(test_and_set_12): Likewise.
(atomic_compare_and_swap<mode>): Likewise.
(atomic_exchange<mode>_llsc): Likewise.
(atomic_fetch_add<mode>_llsc): Likewise.
* config/mips/mips-cpus.def (m14kc, m14k): New processors.
* config/mips/mips-protos.h (umips_output_save_restore): New prototype.
(umips_save_restore_pattern_p): Likewise.
(umips_load_store_pair_p): Likewise.
(umips_output_load_store_pair): Likewise.
(umips_movep_target_p): Likewise.
(umips_12bit_offset_address_p): Likewise.
* config/mips/mips.c (MIPS_MAX_FIRST_STEP): Update for microMIPS.
(mips_base_mips16): Rename this...
(mips_base_compression_flags): ...to this. Update all uses.
(mips_attribute_table): Add micromips, nomicromips and nocompression.
(mips_mips16_decl_p): Delete.
(mips_nomips16_decl_p): Delete.
  (mips_get_compress_on_flags): New function.
(mips_get_compress_off_flags): New function.
(mips_get_compress_mode): New function.
(mips_get_compress_on_name): New function.
(mips_get_compress_off_name): New function.
(mips_insert_attributes): Support multiple compression types.
(mips_merge_decl_attributes): Likewise.
(umips_12bit_offset_address_p): New function.
(mips_start_function_definition): Emit .set micromips directive.
(mips_call_may_need_jalx_p): New function.
(mips_function_ok_for_sibcall): Add microMIPS support.
(mips_print_operand_punctuation): Support short delay slots and
compact jumps.
(umips_swm_mask, umips_swm_encoding): New.
(umips_build_save_restore): New function.
(mips_for_each_saved_gpr_and_fpr): Add microMIPS support.
(was_mips16_p): Remove.
(old_compression_mode): New.
(mips_set_compression_mode): New function.
(mips_set_current_function): Add microMIPS support.
(mips_option_override): Likewise.
(umips_save_restore_pattern_p): New function.
(umips_output_save_restore): New function.
(umips_load_store_pair_p_1): New function.
(umips_load_store_pair_p): New function.
(umips_output_load_store_pair_1): New function.
(umips_output_load_store_pair): New function.
(umips_movep_target_p) New function.
(mips_prepare_pch_save): Add microMIPS support.
* config/mips/mips.h (TARGET_COMPRESSION): New.
(TARGET_CPU_CPP_BUILTINS): Update macro
to use new compression flags and to support microMIPS.
(MIPS_ISA_LEVEL_SPEC): Add m14k processors.
(MIPS_ARCH_FLOAT_SPEC): Likewise.
(ISA_HAS_LWXS): Include TARGET_MICROMIPS.
(ISA_HAS_LOAD_DELAY): Exclude TARGET_MICROMIPS.
(ASM_SPEC): Support mmicromips and mno-micromips.
(M16STORE_REG_P): New macro.
(MIPS_CALL): Support TARGET_MICROMIPS.
(MICROMIPS_J): New macro.
(mips_base_mips16): Rename this...
(mips_base_compression_flags): ...to this.
(UMIPS_12BIT_OFFSET_P): New macro.
* config/mips/t-sde: (MULTILIB_OPTIONS): Add microMIPS.
(MULTILIB_DIRNAMES): Likewise.
libgcc/
2013-03-20  Catherine Moore  <clm@codesourcery.com>
    Joseph Myers  <joseph@codesourcery.com>
    Chao-ying Fu  <fu@mips.com>

* config/mips/mips16.S:  Don't build for microMIPS.
* config/mips/linux-unwind.h: Handle microMIPS frame.
* config/mips/crtn.S (fini, init): New labels.
gcc/testsuite/
2013-03-20  Catherine Moore  <clm@codesourcery.com>
    Richard Sandiford <rdsandiford@googlemail.com>

* gcc.target/mips/mips.exp: Add microMIPS support.
* gcc.target/mips/umips-movep-2.c: New test.
* gcc.target/mips/umips-lwp-2.c: New test.
* gcc.target/mips/umips-swp-5.c: New test.
* gcc.target/mips/umips-constraints-1.c: New test.
* gcc.target/mips/umips-lwp-3.c: New test.
* gcc.target/mips/umips-swp-6.c: New test.
* gcc.target/mips/umips-constraints-2.c: New test.
* gcc.target/mips/umips-save-restore-1.c: New test.
* gcc.target/mips/umips-lwp-4.c: New test.
* gcc.target/mips/umips-swp-7.c: New test.
* gcc.target/mips/umips-save-restore-2.c: New test.
* gcc.target/mips/umips-lwp-swp-volatile.c: New test.
* gcc.target/mips/umips-lwp-5.c: New test.
* gcc.target/mips/umips-save-restore-3.c: New test.
* gcc.target/mips/umips-lwp-6.c: New test.
* gcc.target/mips/umips-swp-1.c: New test.
* gcc.target/mips/umips-lwp-7.c: New test.
* gcc.target/mips/umips-swp-2.c: New test.
* gcc.target/mips/umips-lwp-8.c: New test.
* gcc.target/mips/umips-swp-3.c: New test.
* gcc.target/mips/umips-movep-1.c: New test.
* gcc.target/mips/umips-lwp-1.c: New test.
* gcc.target/mips/umips-swp-4.c: New test.

Co-Authored-By: Chao-ying Fu <fu@mips.com>
Co-Authored-By: Iain Sandoe <iain@codesourcery.com>
Co-Authored-By: Joseph Myers <joseph@codesourcery.com>
Co-Authored-By: Maciej W. Rozycki <macro@codesourcery.com>
Co-Authored-By: Nathan Froyd <froydnj@codesourcery.com>
Co-Authored-By: Nathan Sidwell <nathan@codesourcery.com>
Co-Authored-By: Richard Sandiford <rdsandiford@googlemail.com>
Co-Authored-By: Tom de Vries <tom@codesourcery.com>
From-SVN: r196828

re PR tree-optimization/56661 (Incorrect code with -O1 -ftree-pre)

2013-03-20  Richard Biener  <rguenther@suse.de>

PR tree-optimization/56661
* tree-ssa-sccvn.c (visit_use): Only value-number calls if
the result does not have to be distinct.

* gcc.dg/torture/pr56661.c: New testcase.

From-SVN: r196825

tree-inline.c (copy_tree_body_r): Sync MEM_REF code with remap_gimple_op_r.

2013-03-20  Richard Biener  <rguenther@suse.de>

* tree-inline.c (copy_tree_body_r): Sync MEM_REF code with
remap_gimple_op_r.

From-SVN: r196824

re PR rtl-optimization/56605 (Redundant branch introduced during loop2 phases)

gcc:

2013-03-13  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>
    Steven Bosscher <steven@gcc.gnu.org>

PR rtl-optimization/56605
* loop-iv.c (implies_p): Handle equal RTXs and subregs.

gcc/testsuite:

2013-03-13  Bill Schmidt  wschmidt@linux.vnet.ibm.com>

PR rtl-optimization/56605
* gcc.target/powerpc/pr56605.c: New.

Co-Authored-By: Steven Bosscher <steven@gcc.gnu.org>
From-SVN: r196823

Don't XFAIL gfortran.dg/do_1.f90 (PR fortran/54932)

PR fortran/54932
* gfortran.dg/do_1.f90: Don't xfail.

From-SVN: r196821

i-fortra.ads: Update comment...

2013-03-20  Tobias Burnus  <burnus@net-b.de>

        * i-fortra.ads: Update comment, add Ada 2012's optional
        Star and Kind data types for enhanced interoperability.

From-SVN: r196814

re PR bootstrap/56656 (Suffix or operands invalid for 'movq')

PR bootstrap/56656
* config/i386/i386.md (*movdi_internal): Handle broken assemblers
that require movd instead of movq.

From-SVN: r196813

tree-ssa-structalias.c (struct variable_info): Add pointer to the first field of an aggregate with sub-vars.

2013-03-20  Richard Biener  <rguenther@suse.de>

* tree-ssa-structalias.c (struct variable_info): Add pointer
to the first field of an aggregate with sub-vars.  Make
this and the pointer to the next subfield its ID.
(vi_next): New function.
(nothing_id, anything_id, readonly_id, escaped_id, nonlocal_id,
storedanything_id, integer_id): Increment by one.
(new_var_info, get_call_vi, lookup_call_clobber_vi,
get_call_clobber_vi): Adjust.
(solution_set_expand): Simplify and speedup.
(solution_set_add): Inline into ...
(set_union_with_increment): ... this.  Adjust accordingly.
(do_sd_constraint): Likewise.
(do_ds_constraint): Likewise.
(do_complex_constraint): Simplify.
(build_pred_graph): Adjust.
(solve_graph): Likewise.  Simplify and speedup.
(get_constraint_for_ssa_var, get_constraint_for_ptr_offset,
get_constraint_for_component_ref, get_constraint_for_1,
first_vi_for_offset, first_or_preceding_vi_for_offset,
create_function_info_for, create_variable_info_for_1,
create_variable_info_for, intra_create_variable_infos): Adjust.
(init_base_vars): Push NULL for ID zero.
(compute_points_to_sets): Adjust.

From-SVN: r196812

cfgloop.c (verify_loop_structure): Streamline and avoid ICEing on corrupt loop tree.

2013-03-20  Richard Biener  <rguenther@suse.de>

* cfgloop.c (verify_loop_structure): Streamline and avoid
ICEing on corrupt loop tree.
* graph.c (draw_cfg_nodes_for_loop): Avoid ICEing on corrupt
loop tree.

From-SVN: r196811

tree-vect-loop-manip.c (slpeel_can_duplicate_loop_p): Do not check whether an SSA update is needed.

2013-03-20  Richard Biener  <rguenther@suse.de>

* tree-vect-loop-manip.c (slpeel_can_duplicate_loop_p): Do not
check whether an SSA update is needed.

From-SVN: r196810

constraints.md (T): Rename to...

gcc/
* config/mips/constraints.md (T): Rename to...
(Yf): ...this.
(U): Rename to...
(Yd): ...this.
* config/mips/mips.md (*movdi_64bit, *movdi_64bit_mips16)
(*mov<mode>_internal, *mov<mode>_mips16): Update accordingly.

From-SVN: r196807

re PR fortran/51825 (Fortran runtime error: Cannot match namelist object name)

2013-03-20  Tilo Schwarz  <tilo@tilo-schwarz.de>

        PR libfortran/51825
        * io/list_read.c (nml_read_obj): Don't end the component loop on
        a nested derived type, but continue with the next loop iteration.
        (nml_get_obj_data): Don't move the first_nl pointer further in
        the list if a qualifier was found.

2013-03-20  Tilo Schwarz  <tilo@tilo-schwarz.de>

        PR libfortran/51825
        * gcc/testsuite/gfortran.dg/namelist_77.f90: New.
        * gcc/testsuite/gfortran.dg/namelist_78.f90: New.

From-SVN: r196806

re PR libfortran/48618 (Negative unit number in OPEN(...) is sometimes allowed)

2013-03-20  Tilo Schwarz  <tilo@tilo-schwarz.de>

        PR libfortran/48618
        * io/open.c (st_open): Raise error for unit number < 0 only if
        unit number does not exist already.

2013-03-20  Tilo Schwarz  <tilo@tilo-schwarz.de>

        PR libfortran/48618
        * gfortran.dg/open_negative_unit_1.f90: New.

From-SVN: r196805

Daily bump.

From-SVN: r196804

Yet another ChangeLog fix.

From-SVN: r196800

Another ChangeLog entry fix.

From-SVN: r196799