[multiple changes]

2012-11-06  Tristan Gingold  <gingold@adacore.com>

* exp_vfpt.adb: Document VAX float point layout.

2012-11-06  Geert Bosch  <bosch@adacore.com>

* eval_fat.adb (Machine): Don't return -0.0 on targets without
signed zeros.

2012-11-06  Ed Schonberg  <schonberg@adacore.com>

* sem_ch9.adb (Analyze_Entry_Call_Alternative,
Check_Triggering_Statement): Reject properly an indirect call.

From-SVN: r193222

xoscons.adb, [...]: Add support for post-processing.

2012-11-06  Pascal Obry  <obry@adacore.com>

* xoscons.adb, xutil.adb, xutil.ads: Add support for post-processing.

From-SVN: r193221

[multiple changes]

2012-11-06  Yannick Moy  <moy@adacore.com>

* s-bignum.adb (Div_Rem): Fix another bug in step D3.

2012-11-06  Tristan Gingold  <gingold@adacore.com>

* s-tarest.adb (Create_Restricted_Task): Call
Create_Restricted_Task_Sequential in sequential case.

2012-11-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_prag.adb (Expand_Pragma_Loop_Assertion): Do not rewrite the
pragma into a null statement as its presence is desirable in -gnatG
output.

From-SVN: r193219

[multiple changes]

2012-11-06  Ed Schonberg  <schonberg@adacore.com>

* sem_ch8.adb (Check_Constrained_Object): Do nothing if the
renamed object is a limited record.

2012-11-06  Bernard Banner  <banner@adacore.com>

* sysdep.c (_getpagesize): New. Minor reformatting.

From-SVN: r193218

sem_prag.adb: Minor reformatting.

2012-11-06  Robert Dewar  <dewar@adacore.com>

* sem_prag.adb: Minor reformatting.

2012-11-06  Robert Dewar  <dewar@adacore.com>

* s-bignum.adb (Div_Rem): Fix bug in step D3.
* uintp.adb (UI_Div_Rem): Add comment on bug in step D3.

From-SVN: r193217

[multiple changes]

2012-11-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_prag.adb (Expand_Pragma_Loop_Assertion): Update the comment
on intended expansion. Reimplement the logic which expands the
termination variants.
(Process_Increase_Decrease): Update the parameter profile and the
comment related to it. Accommodate the new aggregate-like appearance of
the termination variants.
* sem_prag.adb (Analyze_Pragma): Update the syntax of pragma
Loop_Assertion. Reimplement the semantic analysis of the pragma
to accommodate the new aggregate-like variant.
(Check_Variant): New routine.
* snames.ads-tmpl: Change names Name_Decreases and Name_Increases
to Name_Decreasing and Name_Increasing respectively. Add name
Variant.

2012-11-06  Ed Schonberg  <schonberg@adacore.com>

* sem_eval.adb: Static evaluation of case expressions.

From-SVN: r193216

[multiple changes]

2012-11-06  Robert Dewar  <dewar@adacore.com>

* exp_prag.adb, impunit.adb, exp_ch9.adb, par-ch4.adb,
s-tarest.adb: Minor reformatting.

2012-11-06  Tristan Gingold  <gingold@adacore.com>

* s-tposen.ads: Minor comment update.

2012-11-06  Arnaud Charlet  <charlet@adacore.com>

* sysdep.c: Rename sig* wrappers to use unique names.

2012-11-06  Yannick Moy  <moy@adacore.com>

* exp_dbug.adb (Qualify_Entity_Name): Mark entity as having a qualified
name after being treated, in formal verification mode.

2012-11-06  Fedor Rybin  <frybin@adacore.com>

* gnat_ugn.texi: Updating gnattest section to reflect changes
in default behaviour of the tool.

From-SVN: r193215

[multiple changes]

2012-11-06  Thomas Quinot  <quinot@adacore.com>

* s-oscons-tmplt.c: Interfaces.C now needs to be WITH'd even
on platforms that do not support sockets (for the benefit of
subtype IOCTL_Req_T).

2012-11-06  Ed Schonberg  <schonberg@adacore.com>

* par-ch4.adb (P_Primary): if-expressions, case-expressions,
and quantified expressions are legal if surrounded by parentheses
from an enclosing context, such as a call or an instantiation.

2012-11-06  Yannick Moy  <moy@adacore.com>

* impunit.adb (Get_Kind_Of_Unit): Return appropriate kind for
predefined implementation files, instead of returning
Not_Predefined_Unit on all .adb files.

2012-11-06  Tristan Gingold  <gingold@adacore.com>

* exp_ch9.adb (Build_Activation_Chain_Entity): Return immediately if
partition elaboration policy is sequential.
(Build_Task_Activation_Call): Likewise. Use
Activate_Restricted_Tasks on restricted profile.
(Make_Task_Create_Call): Do not use the _Chain
parameter if elaboration policy is sequential. Call
Create_Restricted_Task_Sequential in that case.
* exp_ch3.adb (Build_Initialization_Call): Change condition to
support concurrent elaboration policy.
(Build_Record_Init_Proc): Likewise.
(Init_Formals): Likewise.
* bindgen.adb (Gen_Adainit): Declare Partition_Elaboration_Policy
and set it in generated code if the elaboration policy is
sequential. The procedure called to activate all tasks is now
named __gnat_activate_all_tasks.
* rtsfind.adb (RE_Activate_Restricted_Task,
RE_Create_Restricted_Task_Sequential): New RE_Id literals.
* s-tarest.adb (Create_Restricted_Task): Added to create a task without
adding it on an activation chain.
(Activate_Tasks): Has now a Chain parameter.
(Activate_All_Tasks_Sequential): Added. Called by the binder to
activate all tasks.
(Activate_Restricted_Tasks): Added. Called during elaboration to
activate tasks of the units.
* s-tarest.ads: Remove pragma Partition_Elaboration_Policy.
(Partition_Elaboration_Policy): New variable (set by the binder).
(Create_Restricted_Task): Revert removal of the chain parameter.
(Create_Restricted_Task_Sequential): New procedure.
(Activate_Restricted_Tasks): Revert removal.
(Activate_All_Tasks_Sequential): New procedure.

From-SVN: r193214

2012-11-06  Bernard Banner  <banner@adacore.com>

* adaint.c Add file macro definitions missing on Android.
* adaint.h Avoid definitions related to task affinity and CPU
sets since this functionality is missing on the Android
* errno.c (__set_errno): Android already contains such a named
procedure so do include again.
* gsocket.h: Sockets not supported on Android.
* init.c: Avoid linux related code not supported on Android.
* sysdep.c (sigismember, sigaddset, sigdelset, sigemptyset,
sigfillset): wrapper functions since sig routines are defined
as inline macros on Android.
* terminals.c: Add stubs for terminal related functions not
supported on Android.

From-SVN: r193213

sem_prag.adb (Analyze_Pragma): Do not output the characters of an illegal argument as it may not have...

2012-11-06  Hristian Kirtchev  <kirtchev@adacore.com>

* sem_prag.adb (Analyze_Pragma): Do not
output the characters of an illegal argument as it may not have
characters to begin with.

2012-11-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_prag.adb (Expand_Pragma_Loop_Assertion):
Change the order of argument processing to avoid disappearing
increase / decrease expressions.

From-SVN: r193212

[multiple changes]

2012-11-06  Hristian Kirtchev  <kirtchev@adacore.com>

* exp_prag.adb: Add with and use clause for Sem_Ch8.
(Expand_N_Pragma): Add a new variant to expand pragma Loop_Assertion.
(Expand_Pragma_Loop_Assertion): New routine.
* par-prag.adb (Prag): The semantic analysis of pragma
Loop_Assertion is carried out by Analyze_Pragma. No need for
checks in the parser.
* sem_prag.adb: Add a reference position value for pragma
Loop_Assertion in Sig_Flags.
(Analyze_Pragma): Add semantic analysis for pragma Loop_Assertion.
* snames.ads-tmpl: Add the following new names:
Name_Decreases Name_Increases Name_Loop_Assertion.
Add new pragma id Pragma_Loop_Assertion.

2012-11-06  Ed Schonberg  <schonberg@adacore.com>

* exp_ch5.adb: Identifier in iterator must have debug
information.

From-SVN: r193211

re PR target/41993 ([sh] ICE in create_pre_exit, at mode-switching.c:399)

PR target/41993
* mode-switching.c (create_pre_exit): Set return_copy to
last_insn when copy_start is a pseudo reg.

Co-Authored-By: Kaz Kojima <kkojima@gcc.gnu.org>
From-SVN: r193210

Makefile.in (osconstool): Fix target.

libada/
* Makefile.in (osconstool): Fix target.
ada/
* gcc-interface/Makefile.in, gcc-interface/Make-lang.in: Remove
duplicate rules handled by Make-generated.in.

From-SVN: r193209

i386.c (bdesc_args): Rename CODE_FOR_avx2_umulhrswv16hi3 to CODE_FOR_avx2_pmulhrswv16hi3.

* config/i386/i386.c (bdesc_args): Rename CODE_FOR_avx2_umulhrswv16hi3 to
CODE_FOR_avx2_pmulhrswv16hi3.
* config/i386/predicates.md (const1_operand): Extend for vectors.
* config/i386/sse.md (ssse3_avx2): Extend.
(ssedoublemode): Ditto.
(<sse2_avx2>_uavg<mode>3): Merge avx2_uavgv32qi3, sse2_uavgv16qi3,
avx2_uavgv16hi3 and sse2_uavgv8hi3 into one.
(*<sse2_avx2>_uavg<mode>3): Merge *avx2_uavgv32qi3, *sse2_uavgv16qi3,
*avx2_uavgv16hi3 and *sse2_uavgv8hi3 into one.
(PMULHRSW): New.
(<ssse3_avx2>_pmulhrsw<mode>3): Merge avx2_umulhrswv16hi3,
ssse3_pmulhrswv8hi3 and ssse3_pmulhrswv4hi3 into one.
(*avx2_pmulhrswv16hi3): Replace const_vector with const1_operand
predicate.
(*ssse3_pmulhrswv8hi3): Ditto.
(*ssse3_pmulhrswv4hi3): Ditto.

From-SVN: r193208

Update Copyright years.

From-SVN: r193207

Fix ChangeLog typo.

From-SVN: r193206

epiphany.c (epiphany_address_cost): Use MODE parameter.

        * config/epiphany/epiphany.c (epiphany_address_cost):
        Use MODE parameter.

From-SVN: r193205

Function Multiversioning
========================

Sriraman Tallam, tmsriram@google.com

Overview of the patch which adds support to specify function versions.  This is
only enabled for target i386.

Example:

int foo ();  /* Default version */
int foo () __attribute__ ((target("avx,popcnt")));/*Specialized for avx and popcnt */
int foo () __attribute__ ((target("arch=core2,ssse3")));/*Specialized for core2 and ssse3*/

int main ()
{
 int (*p)() = &foo;
 return foo () + (*p)();
}

int foo ()
{
 return 0;
}

int __attribute__ ((target("avx,popcnt")))
foo ()
{
 return 0;
}

int __attribute__ ((target("arch=core2,ssse3")))
foo ()
{
 return 0;
}

The above example has foo defined 3 times, but all 3 definitions of foo are
different versions of the same function. The call to foo in main, directly and
via a pointer, are calls to the multi-versioned function foo which is dispatched
to the right foo at run-time.

Front-end changes:

The front-end changes are calls at appropriate places to target hooks that
determine the following:

* Determine if two function decls with the same signature are versions.
* Determine the assembler name of a function version.
* Generate the dispatcher function for a set of function versions.
* Compare versions to see if one has a higher priority over the other.

All the implementation happens in the target-specific config/i386/i386.c.

What does the patch do?

* Tracking decls that correspond to function versions of function
name, say "foo":

When the front-end sees more than one decl for "foo", it calls a target hook to
determine if they are versions. To prevent duplicate definition errors with
other versions of "foo", "decls_match" function in cp/decl.c is made to return
false when 2 decls have are deemed versions by the target. This will make all
function versions of "foo" to be added to the overload list of "foo".

* Change the assembler names of the function versions.

For i386, the target changes the assembler names of the function versions by
 suffixing the sorted list of args to "target" to the function name of "foo".
For example, the assembler name of
 "void foo () __attribute__ ((target ("sse4")))" will
become _Z3foov.sse4.  The target hook mangle_decl_assembler_name is used
for this.

* Overload resolution:

 Function "build_over_call" in cp/call.c sees a call to function
"foo", which is multi-versioned. The overload resolution happens in
function "joust" in "cp/call.c". Here, the call to "foo" has all
possible versions of "foo" as candidates. All the candidates of "foo" are
stored in the cgraph side data structure. Each version of foo is chained in a
doubly-linked list with the default function as the first element.  This allows
any pass to access all the semantically identical versions. A call to a
multi-versioned function will be replaced by a call to a dispatcher function,
determined by a target hook, to execute the right function version at run-time.

Optimization to directly call a version when possible:
Also, in joust, where overload resolution happens, a multiversioned function
resolution is made to return the most specialized version.  This is the version
that will be checked for dispatching first and is determined by the target.
Now, if the caller can inline this function version then a direct call is made
to this function version rather than go through the dispatcher. When a direct
call cannot be made, a call to the dispatcher function is created.

* Creating the dispatcher body.

The dispatcher body, called the resolver is made only when there is a call to a
multiversioned function dispatcher or the address of a function is taken. This
is generated during cgraph_analyze_function. This is done by another target hook.

* Dispatch ordering.

The order in which the function versions are checked during dispatch is based
on a priority value assigned for the ISA that is catered. More specialized
versions are checked for dispatching first.  This is to mitigate the ambiguity
that can arise when more than one function version is valid for execution on
a particular platform.  This is not a perfect solution, and in future the user
should be allowed to assign a dispatching priority value to each version.

Function MV in the Intel compiler:

The intel compiler supports function multiversioning and the syntax is
similar to the patch proposed here.  Here is an example of how to
generate multiple function versions with the intel compiler.

/* Create a stub function to specify the various versions of function that
   will be created, using declspec attribute cpu_dispatch.  */
__declspec (cpu_dispatch (core_i7_sse4_2, atom, generic))
void foo () {};

/* Bodies of each function version.  */

/* Intel Corei7 processor + SSE4.2 version.  */
__declspec (cpu_specific(core_i7_sse4_2))
void foo ()
{
  printf ("corei7 + sse4.2");
}

/* Atom processor.  */
__declspec (cpu_specific(atom))
void foo ()
{
  printf ("atom");
}

/* The generic or the default version.  */
__declspec (cpu_specific(generic))
void foo ()
{
  printf ("This is generic");
}

A new function version is generated by defining a new function with the same
signature but with a different cpu_specific declspec attribute string.  The
set of cpu_specific strings that are allowed is the following:

"core_2nd_gen_avx"
"core_aes_pclmulqdq"
"core_i7_sse4_2"
"core_2_duo_sse4_1"
"core_2_duo_ssse3"
"atom"
"pentium_4_sse3"
"pentium_4"
"pentium_m"
"pentium_iii"
"generic"

Comparison with the GCC MV implementation in this patch:

* Version creation syntax:

The implementation in this patch also has a similar syntax to specify function
versions. The first stub function is not needed.  Here is the code to generate
the function versions with this patch:

/* Intel Corei7 processor + SSE4.2 version.  */
__attribute__ ((target ("arch=corei7, sse4.2")))
void foo ()
{
  printf ("corei7 + sse4.2");
}

/* Atom processor.  */
__attribute__ ((target ("arch=atom")))
void foo ()
{
  printf ("atom");
}

void foo ()
{
}

The target attribute can have one of the following arch names:

"amd"
"intel"
"atom"
"core2"
"corei7"
"nehalem"
"westmere"
"sandybridge"
"amdfam10h"
"barcelona"
"shanghai"
"istanbul"
"amdfam15h"
"bdver1"
"bdver2"

and any number of the following ISA names:

"cmov"
"mmx"
"popcnt"
"sse"
"sse2"
"sse3"
"ssse3"
"sse4.1"
"sse4.2"
"avx"
"avx2"

* doc/tm.texi.in (TARGET_OPTION_FUNCTION_VERSIONS): New hook
description.
* (TARGET_COMPARE_VERSION_PRIORITY): New hook description.
* (TARGET_GET_FUNCTION_VERSIONS_DISPATCHER): New hook description.
* (TARGET_GENERATE_VERSION_DISPATCHER_BODY): New hook description.
* doc/tm.texi: Regenerate.
* target.def (compare_version_priority): New target hook.
* (generate_version_dispatcher_body): New target hook.
* (get_function_versions_dispatcher): New target hook.
* (function_versions): New target hook.
* cgraph.c (cgraph_fnver_htab): New htab.
(cgraph_fn_ver_htab_hash): New function.
(cgraph_fn_ver_htab_eq): New function.
(version_info_node): New pointer.
(insert_new_cgraph_node_version): New function.
(get_cgraph_node_version): New function.
(delete_function_version): New function.
(record_function_versions): New function.
* cgraph.h (cgraph_node): New bitfield dispatcher_function.
(cgraph_function_version_info): New struct.
(get_cgraph_node_version): New function.
(insert_new_cgraph_node_version): New function.
(record_function_versions): New function.
(delete_function_version): New function.
(init_lowered_empty_function): Expose function.
* tree.h (DECL_FUNCTION_VERSIONED): New macro.
(tree_function_decl): New bit-field versioned_function.
* cgraphunit.c (cgraph_analyze_function): Generate body of multiversion
function dispatcher.
(cgraph_analyze_functions): Analyze dispatcher function.
(init_lowered_empty_function): Make non-static. New parameter in_ssa.
(assemble_thunk): Add parameter to call to init_lowered_empty_function.
* config/i386/i386.c (add_condition_to_bb): New function.
(get_builtin_code_for_version): New function.
(ix86_compare_version_priority): New function.
(feature_compare): New function.
(dispatch_function_versions): New function.
(ix86_function_versions): New function.
(attr_strcmp): New function.
(ix86_mangle_function_version_assembler_name): New function.
(ix86_mangle_decl_assembler_name): New function.
(make_name): New function.
(make_dispatcher_decl): New function.
(is_function_default_version): New function.
(ix86_get_function_versions_dispatcher): New function.
(make_attribute): New function.
(make_resolver_func): New function.
(ix86_generate_version_dispatcher_body): New function.
(fold_builtin_cpu): Return integer for cpu builtins.
(TARGET_MANGLE_DECL_ASSEMBLER_NAME): New macro.
(TARGET_COMPARE_VERSION_PRIORITY): New macro.
(TARGET_GENERATE_VERSION_DISPATCHER_BODY): New macro.
(TARGET_GET_FUNCTION_VERSIONS_DISPATCHER): New macro.
(TARGET_OPTION_FUNCTION_VERSIONS): New macro.

* class.c (add_method): Change assembler names of function versions.
(mark_versions_used): New static function.
(resolve_address_of_overloaded_function): Create dispatcher decl and
return address of dispatcher instead.
* decl.c (decls_match): Make decls unmatched for versioned
functions.
(duplicate_decls): Remove ambiguity for versioned functions.
Delete versioned function data for merged decls.
* decl2.c (check_classfn): Check attributes of versioned functions
for match.
* call.c (get_function_version_dispatcher): New function.
(mark_versions_used): New static function.
(build_over_call): Make calls to multiversioned functions
to call the dispatcher.
(joust): For calls to multi-versioned functions, make the most
specialized function version win.

* testsuite/g++.dg/mv1.C: New test.
* testsuite/g++.dg/mv2.C: New test.
* testsuite/g++.dg/mv3.C: New test.
* testsuite/g++.dg/mv4.C: New test.
* testsuite/g++.dg/mv5.C: New test.
* testsuite/g++.dg/mv6.C: New test.

From-SVN: r193204

* recog.c (extract_insn): Enabled alternative defaults to 1.

From-SVN: r193203

* include/profile/map.h (map::emplace_hint): Add missing return.

From-SVN: r193202

Daily bump.

From-SVN: r193201

Replace REX_INT_REG_P with REX_INT_REGNO_P

* config/i386/i386.c (print_reg): Replace REX_INT_REG_P with
REX_INT_REGNO_P.

From-SVN: r193197

deque: Constrain InputIterator parameters.

* include/profile/deque: Constrain InputIterator parameters.
* include/profile/forward_list: Likewise.
* include/profile/list: Likewise.
* include/profile/map.h: Likewise.
* include/profile/multimap.h: Likewise.
* include/profile/set.h: Likewise.
* include/profile/multiset.h: Likewise.
* include/profile/vector: Likewise.

From-SVN: r193196

re PR testsuite/55186 (gcc.dg/const-uniq-1.c fails due to vector expected but not being in the constant pool)

PR testsuite/55186
* gcc.dg/const-uniq-1.c (a): Increase length four times.

From-SVN: r193194

Use -Wl,-undefined,dynamic_lookup on darwin

* gcc.dg/torture/pr53922.c: Use -Wl,-undefined,dynamic_lookup on
darwin.

From-SVN: r193193

re PR tree-optimization/54986 (segfault on constant initialized to object address at -O)

PR tree-optimization/54986
* gimple-fold.c (canonicalize_constructor_val): Strip again all no-op
conversions on entry but add them back on exit if needed.

From-SVN: r193188

final.c (final_scan_insn): Handle all comparison codes in non-jump and cmove insn.

* gcc/final.c (final_scan_insn) [HAVE_cc0]: Handle all comparison
codes in non-jump and cmove insn.

* gcc/testsuite/gcc.dg/torture/fp-compare.c: New testcase.

From-SVN: r193187

forward_list: Update to meet allocator-aware requirements.

* include/profile/forward_list: Update to meet allocator-aware
requirements.
* include/debug/forward_list: Likewise.
* include/debug/vector: Verify allocators are swapped or equal.
* include/debug/macros.h (__glibcxx_check_equal_allocs): Define.
* include/debug/formatter.h: Add new debug message.
* src/c++11/debug.cc: Likewise.
* testsuite/23_containers/forward_list/allocator/swap.cc: Do not
swap containers with non-propagating, non-equal allocators.
* testsuite/23_containers/vector/allocator/swap.cc: Likewise.

From-SVN: r193186

re PR libstdc++/55028 (_GLIBCXX_DEBUG is broken when using v7 namespace)

2012-11-05  Benjamin Kosnik  <bkoz@redhat.com>
            Oleg Smolsky  <oleg@smolsky.net>

PR libstdc++/55028
*  config/abi/pre/gnu-versioned-namespace.ver: Add symbols.
* testsuite/23_containers/unordered_multimap/insert/55028-debug.cc: New.

Co-Authored-By: Oleg Smolsky <oleg@smolsky.net>
From-SVN: r193185

throw_allocator.h (__throw_value_base): Add move semantic, not throwing.

2012-10-05  François Dumont  <fdumont@gcc.gnu.org>

* include/ext/throw_allocator.h (__throw_value_base): Add move
semantic, not throwing.
(__throw_value_limit): Likewise.
(__throw_value_random): Likewise.
* testsuite/util/exception/safety.h: Add validation of C++11
methods emplace/emplace_front/emplace_back/emplace_hint.
* testsuite/util/testsuite_container_traits.h: Signal emplace
support on deque, forward_list, list and vector.
* testsuite/23_containers/deque/requirements/exception/
propagation_consistent.cc: Remove dg-do run fail.

From-SVN: r193184

mode-switching.c (create_pre_exit): Force late switching if __builtin_{apply,return} emitted a load that require...

* mode-switching.c (create_pre_exit): Force late switching if
__builtin_{apply,return} emitted a load that require mode,
other than MODE_EXIT.

Co-Authored-By: Vladimir Yakovlev <vladimir.b.yakovlev@intel.com>
From-SVN: r193182

re PR libstdc++/55215 (Constructor seeding is broken for Mersenne twister)

2012-11-05  Paolo Carlini  <paolo.carlini@oracle.com>

PR libstdc++/55215
* include/bits/random.tcc (mersenne_twister_engine<>::seed(_Sseq&)):
Assign state_size to _M_p.
* testsuite/26_numerics/random/mersenne_twister_engine/cons/55215.cc:
New.
* testsuite/26_numerics/random/independent_bits_engine/cons/55215.cc:
Likewise.
* testsuite/26_numerics/random/shuffle_order_engine/cons/55215.cc:
Likewise.
* testsuite/26_numerics/random/subtract_with_carry_engine/cons/
55215.cc: Likewise.
* testsuite/26_numerics/random/discard_block_engine/cons/55215.cc:
Likewise.
* testsuite/26_numerics/random/linear_congruential_engine/cons/
55215.cc: Likewise.

From-SVN: r193181

re PR target/55204 (ICE: in extract_insn, at recog.c:2140 (unrecognizable insn) with -O --param loop-invariant-max-bbs-in-loop=0)

gcc/
PR target/55204
* config/i386/i386.c (ix86_address_subreg_operand): Remove stack
pointer check.
(print_reg): Use true_regnum rather than REGNO.
(ix86_print_operand_address): Remove SUBREG handling.

From-SVN: r193178

const-1.c: Update.

* gcc.dg/const-1.c: Update.
* gcc.dg/pure-1.c: Update.
* tree-ssa-loop-niter.c (finite_loop_p): Revamp to be just wrapper of
max_loop_iterations.

From-SVN: r193175

Detect overflow in size calculations in __cxa_vec_new{2,3}

2012-11-03  Florian Weimer  <fweimer@redhat.com>

* libsupc++/vec.cc (compute_size): New.
(__cxa_vec_new2, __cxa_vec_new3): Use it.
* testsuite/18_support/cxa_vec.cc: New.

From-SVN: r193174

runtime: provide initcontext and fixcontext for NetBSD

From-SVN: r193173

os: support more OSes

From-SVN: r193172

reorg.c (fill_simple_delay_slots): Avoid calling optimize_skip with a return instruction.

* reorg.c (fill_simple_delay_slots): Avoid calling optimize_skip
with a return instruction.

From-SVN: r193171

re PR rtl-optimization/55151 (ICE: in assign_by_spills, at lra-assigns.c:1217 with -fPIC)

2012-11-05  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/55151
* lra-constraints.c (process_alt_operands): Permit putting reg
value into memory.  Increase reject for this case.

2012-11-05  Vladimir Makarov  <vmakarov@redhat.com>

PR rtl-optimization/55151
* gcc.dg/pr55151.c: New test.

From-SVN: r193170

final.c (reemit_insn_block_notes): Do not change scope if insn location is UNKNOWN_LOCATION.

2012-11-05  Dehao Chen  <dehao@google.com>

* final.c (reemit_insn_block_notes): Do not change scope if insn
location is UNKNOWN_LOCATION.

From-SVN: r193169

md.texi (Defining Attributes): Document that we are defining HAVE_ATTR_name macors as 1 for defined attributes...

* doc/md.texi (Defining Attributes): Document that we are defining
HAVE_ATTR_name macors as 1 for defined attributes, and as 0
for undefined special attributes.
* final.c (asm_insn_count, align_fuzz): Always define.
(insn_current_reference_address): Likewise.
(init_insn_lengths): Use if (HAVE_ATTR_length) instead of
#ifdef HAVE_ATTR_length.
(get_attr_length_1, shorten_branches, final): Likewise.
(final_scan_insn, output_asm_name): Likewise.
* genattr.c (gen_attr): Define HAVE_ATTR_name macros for
defined attributes as 1.
Remove ancient get_attr_alternative compatibility code.
For special purpose attributes not provided, define HAVE_ATTR_name
as 0.
In case no length attribute is given, provide stub definitions
for insn_*_length* functions, and also include insn-addr.h.
In case no enabled attribute is given, provide stub definition.
* genattrtab.c (write_length_unit_log): Always write a definition.
* hooks.c (hook_int_rtx_1, hook_int_rtx_unreachable): New functions.
* hooks.h (hook_int_rtx_1, hook_int_rtx_unreachable): Declare.
* lra-int.h (struct lra_insn_recog_data): Make member
alternative_enabled_p unconditional.
* lra.c (free_insn_recog_data): Use if (HAVE_ATTR_length) instead of
#ifdef HAVE_ATTR_length.
(lra_set_insn_recog_data): Likewise.  Make initialization of
alternative_enabled_p unconditional.
(lra_update_insn_recog_data): Use #if instead of #ifdef for
HAVE_ATTR_enabled.
* recog.c [!HAVE_ATTR_enabled] (get_attr_enabled): Don't define.
(extract_insn): Check HAVE_ATTR_enabled.
(gate_handle_split_before_regstack): Use #if instead of
#if defined for HAVE_ATTR_length.

From-SVN: r193168

re PR debug/54970 (Missing DW_OP_GNU_implicit_pointer in debuginfo)

PR debug/54970
PR debug/54971
* gcc.dg/guality/pr54970.c: Use NOP instead of "NOP" in inline-asm.

From-SVN: r193162

ipa-inline.c (compute_uninlined_call_time, [...]): New functions.

* ipa-inline.c (compute_uninlined_call_time,
compute_inlined_call_time): New functions.
(RELATIVE_TIME_BENEFIT_RANGE): New macro.
(relative_time_benefit): Rewrite.
(edge_badness): Rewrite path with guessed profile and estimated profile.
* ipa-inline.h (INLINE_HINT_declared_inline, INLINE_HINT_cross_module):
New hints.
(struct inline_summary): Add GROWTH filed.
* ipa-inline-analysis.c (dump_inline_hints): Update.
(reset_inline_summary): Update.
(dump_inline_summary): Update.
(will_be_nonconstant_predicate): Cleanup to use gimple_store_p and
gimple_assign_load_p predicates.
(estimate_node_size_and_time): Drop INLINE_HINT_declared_inline hint.
(simple_edge_hints): New function.
(do_estimate_edge_time): Return time of invocation of callee rather
than the time scaled by edge frequency; update hints code.
(do_estimate_edge_hints): Update.
(do_estimate_growth): Cleanup.

From-SVN: r193161

tree-ssa-loop-niter.c (find_loop_niter): Remove just_once_each_iteration_p.

* tree-ssa-loop-niter.c (find_loop_niter): Remove just_once_each_iteration_p.
(maybe_lower_iteration_bound): Initialize not_executed_last_iteration to  NULL
* tree-ssa-loop-ivcanon.c (canonicalize_loop_induction_variables): Skip
just_once_each_iteration_p; record estimated bound when loop has only one
likely exit; test just_once_each_iteration_p before IV canon itself.

From-SVN: r193159

re PR target/55194 (h8300 ICE during conftest in libgcc dwarf2out:7605)

PR target/55194
* dwarf2out.c (value_format) <case dw_val_class_high_pc>: Handle
also DWARF2_ADDR_SIZE 1 and 2.

From-SVN: r193158

ipa-inline.c (leaf_node_p): Rename to ...

* ipa-inline.c (leaf_node_p): Rename to ...
(num_calls) ... this one.
(want_early_inline_function_p): Allow smal growth on non-leafs.

From-SVN: r193157

re PR testsuite/51128 (New LTO failures)

PR testsuite/51128
* gcc.dg/torture/pr55018.c: Skip if -fno-fat-lto-objects was passed.

From-SVN: r193156

* gcc.dg/autopar/pr49960.c: Fix testcase.

From-SVN: r193155

* gcc.dg/tree-ssa/cunroll-9.c: Fix scan file name.

From-SVN: r193154

cunroll-9.c: Dump cunrolli details.

* gcc.dg/tree-ssa/cunroll-9.c: Dump cunrolli details.
Fix scan-tree-dump-times directive.

From-SVN: r193153

re PR debug/54402 (var-tracking does not scale)

PR debug/54402
* var-tracking.c (fp_setter): Return false if there is REG_CFA_RESTORE
hfp note.
(vt_initialize): Look for fp_setter in any bb, not just successor of
entry bb.

From-SVN: r193152

sh.h (TARGET_CACHE32, [...]): Delete macro.

* config/sh/sh.h (TARGET_CACHE32, TARGET_HARVARD): Delete macro.
(TARGET_SUPERSCALAR): Add TARGET_SH2A.
(CACHE_LOG): Use TARGET_HARD_SH4 and TARGET_SH5 instead of
TARGET_CACHE32.
(TRAMPOLINE_ALIGNMENT): Use TARGET_HARD_SH4 and TARGET_SH5 instead of
TARGET_HARVARD.
* config/sh/sh.c (sh_trampoline_init): Likewise.

From-SVN: r193151

* system.h (loc_t): Poison.

From-SVN: r193150

Daily bump.

From-SVN: r193149

fold-const.c (fold_unary_loc): Disable conversion optimization for void type.

2012-11-05  Marc Glisse  <marc.glisse@inria.fr>

* fold-const.c (fold_unary_loc): Disable conversion optimization
for void type.

From-SVN: r193146

acinclude.m4 (LIBAT_CHECK_IFUNC): Clarify message.

libatomic/
* acinclude.m4 (LIBAT_CHECK_IFUNC): Clarify message.
* configure: Regenerate.

From-SVN: r193145

configure: Regenerate.

libatomic/
* configure: Regenerate.

libitm/
* configure: Regenerate.

From-SVN: r193144

bid_functions.h: Check for __GLIBC__ additionally to LINUX when defining format specifiers.

libgcc/config/libbid/
* bid_functions.h: Check for __GLIBC__ additionally to LINUX when
defining format specifiers.

config/
* dfp.m4 (enable_decimal_float): Enable for i?86*-*-gnu*.

gcc/
* configure: Regenerate.

libdecnumber/
* configure: Regenerate.

libgcc/
* configure: Regenerate.

From-SVN: r193143

configure.ac: Add stanza for *-*-gnu* threads configuration.

boehm-gc/
* configure.ac: Add stanza for *-*-gnu* threads configuration.
* configure: Regenerate.
* include/gc_config.h.in: Likewise.
* dyn_load.c (_GNU_SOURCE): Define for __GNU__.
* include/gc_config_macros.h (_REENTRANT, GC_PTHREADS): Define for
GC_GNU_THREADS.
* include/private/gcconfig.h (DATASTART): Don't define for I386 &&
HURD.
(SIG_SUSPEND, SIG_THR_RESTART, SEARCH_FOR_DATA_START): Define for
I386 && HURD.
(UNIX_LIKE, REGISTER_LIBRARIES_EARLY): Define for HURD.
* os_dep.c (GC_init_linux_data_start): Handle for HURD.
* pthread_support.c (GC_thr_init): Add case for GC_GNU_THREADS.
* specific.c: Use for GC_GNU_THREADS.
* threadlibs.c: Add case for GC_GNU_THREADS.

From-SVN: r193142

combine.c: Adjust toplevel comment.

* combine.c: Adjust toplevel comment.
(make_extraction): Adjust head comment and move up canonicalization.

From-SVN: r193141

re PR debug/54693 (VTA guality issues with loops)

PR debug/54693
* tree-ssa-loop-ivopts.c (remove_unused_ivs): Emit debug temps
for dropped IV sets.

From-SVN: r193139

re PR debug/54693 (VTA guality issues with loops)

PR debug/54693
* tree-ssa-threadedge.c (propagate_threaded_block_debug_into):
New, rewritten from debug stmt copying code...
(thread_around_empty_block): ... removed from here.
(thread_across_edge): Call propagate_threaded_block_debug_into.

From-SVN: r193138

expr.c (expand_expr_real_1): Change to not using input_location.

2012-11-01  Dehao Chen  <dehao@google.com>

* expr.c (expand_expr_real_1): Change to not using input_location.

From-SVN: r193137

re PR fortran/55199 ([OOP] Equivalenced variable has wrong type when used with generic member function)

2012-11-04  Janus Weil  <janus@gcc.gnu.org>

PR fortran/55199
* primary.c (gfc_match_varspec): Clear typespec if it cannot be
determined at this point.

2012-11-04  Janus Weil  <janus@gcc.gnu.org>

PR fortran/55199
* gfortran.dg/associate_12.f90: New.

From-SVN: r193136

invoke.texi (profile-use): update documentation.

* invoke.texi (profile-use): update documentation.
* opts.c (common_handle_option): Enable tree-vectorize
and tree-loop-distribute-patterns.
* tree-loop-distribution.c (tree_loop_distribution): Skip loops
optimized for size.
* config/i386/i386.c (ix86_option_override_internal): Enabe
prefetch_lop_arrays for profile-use.

From-SVN: r193135

ChangeLog: Fix whitespace.

* ChangeLog: Fix whitespace.
* testsuite/ChangeLog: Ditto.

From-SVN: r193134

Daily bump.

From-SVN: r193132

Fix for building against Mingw64 w32api

* configure.ac (FLAGS_FOR_TARGET,target=cygwin): Fix for building
against Mingw64 w32api.
* configure: Regenerate.

From-SVN: r193129

Regenerate top-level configure

* configure: Regenerated.

From-SVN: r193128

Add check_effective_target_masm_intel

PR target/54255
* lib/target-supports.exp (check_effective_target_masm_intel): New
proc.
* gcc.target/i386/asm-dialect-1.c: Use dg-require-effective-target
masm_intel.

From-SVN: r193127

Add check_effective_target_maybe_x32

* lib/target-supports.exp (check_effective_target_maybe_x32): New
proc.
* gcc.target/i386/pr54457.c: Use dg-require-effective-target
maybe_x32.
* gcc.target/i386/pr53249.c: Likewise.

Co-Authored-By: Jack Howarth <howarth@bromo.med.uc.edu>
From-SVN: r193126

octeon-bbit-3.c: Change the second call to abort to call abort1 so that cross jumping does not...

2012-11-03  Andrew Pinski  <apinski@cavium.com>

 * gcc.target/mips/octeon-bbit-3.c: Change the second call to abort to
call abort1 so that cross jumping does not happen.

From-SVN: r193125

final patch for enabling VxWorks

From-SVN: r193124

extend.texi: Copy-edit to use present tense except when explicitly describing future or past...

2012-11-03  Sandra Loosemore  <sandra@codesourcery.com>

gcc/
* doc/extend.texi: Copy-edit to use present tense except when
explicitly describing future or past behavior.

From-SVN: r193123

sh.c (output_movedouble, [...]): Use multi-line strings for asm.

* config/sh/sh.c (output_movedouble, output_far_jump,
output_ieee_ccmpeq): Use multi-line strings for asm.
(find_regmode_weight): Wrap lines.
(sh_optimize_target_register_callee_saved): Remove commented out
variable.
(emit_load_ptr): Remove prototype.

From-SVN: r193120

re PR target/51244 ([SH] Inefficient conditional branch and code around T bit)

PR target/51244
* config/sh/sh.md (*cbranch_t): Allow splitting after reload.
Allow going beyond current basic block before reload when looking for
the reg set insn.
* config/sh/sh.c (sh_find_set_of_reg): Don't stop at labels.

PR target/51244
* gcc.target/sh/pr51244-18.c: New.
* gcc.target/sh/pr51244-19.c: New.

From-SVN: r193119

octeon-exts-7.c (bar): Make sure the extraction of b requires a 64-bit operation followed by a truncation.

gcc/testsuite/
* gcc.target/mips/octeon-exts-7.c (bar): Make sure the extraction of
b requires a 64-bit operation followed by a truncation.

From-SVN: r193118

loop-38.c: Correct testcase.

* gcc.dg/tree-ssa/loop-38.c: Correct testcase.

* tree-ssa-loop-niter.c (discover_iteration_bound_by_body_walk): Add
missing sign to unsigned.

From-SVN: r193117

* doc/html/*: Regenerate.

From-SVN: r193116

Daily bump.

From-SVN: r193115

syscall, mksysinfo: Prepare syscall package for 64-bit int.

From-SVN: r193112

re PR rtl-optimization/54524 (Spurious add on sum of bitshifts (forward-propagate issue))

2012-11-02  Andrew Pinski  <apinski@cavium.com>

        PR rtl-opt/54524
        * simplify-rtx.c (simplify_relational_operation_1): Don't simplify
        (LTU/GEU (PLUS a 0) 0) into (GEU/LTU a 0) since they are not equivalent.

From-SVN: r193111

runtime: Fix reflect.Call support for 64-bit ints.

From-SVN: r193110

configure.xml: Adjust GNU libc link.

* doc/xml/manual/configure.xml: Adjust GNU libc link.

* doc/xml/manual/debug.xml: Adjust GDB documentation link.
Use GCC instead of gcc.

* doc/xml/manual/prerequisites.xml: Remove detailed reference
to binutils downloads.

From-SVN: r193109

pr48258-2.c: Simplfy the acceptance conditions to verify the reduction code is vectorized.

2012-11-02  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/pr48258-2.c: Simplfy the acceptance
conditions to verify the reduction code is vectorized.

From-SVN: r193107

Update ChangeLog files as requested for cgraph change to checked down cast.

Index: gcc/cp/ChangeLog

2012-10-31  Lawrence Crowl  <crowl@google.com>

* decl2.c (var_finalized_p): Rename varpool_node to
varpool_node_for_decl.
(maybe_emit_vtables): Likewise.

Index: gcc/ada/ChangeLog

2012-10-31  Lawrence Crowl  <crowl@google.com>

* gcc-interface/utils.c (gnat_write_global_declarations):
Rename varpool_node to varpool_node_for_decl.

Index: gcc/lto/ChangeLog

2012-10-31  Lawrence Crowl  <crowl@google.com>

* lto.c (lto_wpa_write_files): Change symtab checking to a checked
down-cast via dyn_cast.
* lto-partition.c (add_symbol_to_partition_1): Likewise.
(undo_partition): Likewise.
(lto_balanced_map): Likewise.
(get_symbol_class): Likewise and via is_a.
(lto_balanced_map): Change symtab checking to is_a.

From-SVN: r193106

Add a new option --clean_build to validate_failures.py

This is useful when you have two builds of the same compiler.  One with
your changes.  The other one, a clean build at the same revision.
Instead of using a manifest file, --clean_build will compare the
results it gather from the patched build against those it gathers from
the clean build.

Usage

$ cd /top/of/patched/gcc/bld
$ validate_failures.py --clean_build=clean/bld-gcc
Source directory: /usr/local/google/home/dnovillo/gcc/trunk
Build target:     x86_64-unknown-linux-gnu
Getting actual results from build directory .
        ./x86_64-unknown-linux-gnu/libstdc++-v3/testsuite/libstdc++.sum
        ./x86_64-unknown-linux-gnu/libffi/testsuite/libffi.sum
        ./x86_64-unknown-linux-gnu/libgomp/testsuite/libgomp.sum
        ./x86_64-unknown-linux-gnu/libgo/libgo.sum
        ./x86_64-unknown-linux-gnu/boehm-gc/testsuite/boehm-gc.sum
        ./x86_64-unknown-linux-gnu/libatomic/testsuite/libatomic.sum
        ./x86_64-unknown-linux-gnu/libmudflap/testsuite/libmudflap.sum
        ./x86_64-unknown-linux-gnu/libitm/testsuite/libitm.sum
        ./x86_64-unknown-linux-gnu/libjava/testsuite/libjava.sum
        ./gcc/testsuite/g++/g++.sum
        ./gcc/testsuite/gnat/gnat.sum
        ./gcc/testsuite/ada/acats/acats.sum
        ./gcc/testsuite/gcc/gcc.sum
        ./gcc/testsuite/gfortran/gfortran.sum
        ./gcc/testsuite/obj-c++/obj-c++.sum
        ./gcc/testsuite/go/go.sum
        ./gcc/testsuite/objc/objc.sum
Getting actual results from build directory clean/bld-gcc
        clean/bld-gcc/x86_64-unknown-linux-gnu/libstdc++-v3/testsuite/libstdc++.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libffi/testsuite/libffi.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libgomp/testsuite/libgomp.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libgo/libgo.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/boehm-gc/testsuite/boehm-gc.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libatomic/testsuite/libatomic.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libmudflap/testsuite/libmudflap.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libitm/testsuite/libitm.sum
        clean/bld-gcc/x86_64-unknown-linux-gnu/libjava/testsuite/libjava.sum
        clean/bld-gcc/gcc/testsuite/g++/g++.sum
        clean/bld-gcc/gcc/testsuite/gnat/gnat.sum
        clean/bld-gcc/gcc/testsuite/ada/acats/acats.sum
        clean/bld-gcc/gcc/testsuite/gcc/gcc.sum
        clean/bld-gcc/gcc/testsuite/gfortran/gfortran.sum
        clean/bld-gcc/gcc/testsuite/obj-c++/obj-c++.sum
        clean/bld-gcc/gcc/testsuite/go/go.sum
        clean/bld-gcc/gcc/testsuite/objc/objc.sum

SUCCESS: No unexpected failures.

2012-11-02  Diego Novillo  <dnovillo@google.com>

* testsuite-management/validate_failures.py: Add option
--clean_build to compare test results against another
build.

From-SVN: r193105

tree-ssa-loop-niter.c (double_int_cmp, [...]): New functions.

* tree-ssa-loop-niter.c (double_int_cmp, bound_index,
discover_iteration_bound_by_body_walk): New functions.
(discover_iteration_bound_by_body_walk): Use it.

* gcc.dg/tree-ssa/loop-38.c: New testcase.

From-SVN: r193104

Pass -fno-ident to tests making sure "pow" is not called

From-SVN: r193103

predict.c (predict_loops): Predict also exits not dominating latch.

* predict.c (predict_loops): Predict also exits not dominating
latch.

From-SVN: r193102

* predict.c (predict_loops): Do not predict infinite loops.

From-SVN: r193101

fmaq.c (fmaq): Fix build.

2012-11-01  Tobias Burnus  <burnus@net-b.de>

        * math/fmaq.c (fmaq): Fix build.

From-SVN: r193100

fmaq.c (fmaq): Merge from GLIBC.

2012-11-01  Tobias Burnus  <burnus@net-b.de>
            Joseph Myers  <joseph@codesourcery.com>

        * math/fmaq.c (fmaq): Merge from GLIBC. Handle cases
        with small x * y using scaling, not as x * y + z.
        * math/lgammaq.c (lgammaq): Fix signgam handling.

Co-Authored-By: Joseph Myers <joseph@codesourcery.com>
From-SVN: r193099

re PR tree-optimization/55079 (false positive -Warray-bounds (also seen at -O3 bootstrap))

PR middle-end/55079
* tree-ssa-loop-niter.c (number_of_iterations_exit): Update
MAX field if NITER was folded to contant.
(record_estimate): Sanity check.
* tree-ssa-loop-ivcanon.c (remove_exits_and_undefined_stmts): New
function.
(remove_redundant_iv_test): New function.
(loops_to_unloop, loops_to_unloop_nunroll): New static vars.
(unloop_loops): Break out from ...
(try_unroll_loop_completely): ... here; Pass in MAXITER; use
remove_exits_and_undefined_stmts; do not unloop.
(canonicalize_loop_induction_variables): Compute MAXITER;
use remove_redundant_iv_test; remove loop_close_ssa_invalidated
and irred_invalidated arguments.
(canonicalize_induction_variables): Compute fresh bound estimates;
unloop; walk from innermost.
(tree_unroll_loops_completely): Likewise.

* gcc.dg/tree-ssa/cunroll-10.c: New testcase.
* gcc.dg/tree-ssa/cunroll-9.c: New testcase.

From-SVN: r193098

Fix typo in changelog.

From-SVN: r193097

re PR middle-end/55130 (ICE in insn_rhs_dead_pseudo_p, at lra-constraints.c:3224)

2012-11-02  Vladimir Makarov  <vmakarov@redhat.com>

PR middle-end/55130
* lra-constraints.c (debug_loc_equivalence_change_p): Rename to
loc_equivalence_change_p.
(lra_constraints): Check equiv_insn_bitmap for debug insn.  Call
loc_equivalence_change_p for non-transformed insn.

2012-11-02  Vladimir Makarov  <vmakarov@redhat.com>

PR middle-end/55150
* gcc.target/i386/pr55130.c: New test.

From-SVN: r193096

re PR target/55175 (i386/sfp-exceptions.c:52:7: error: impossible constraint in 'asm')

PR target/55175
* config/i386/sfp-exceptions.c: Guard with _SOFT_FLOAT.
* config/i386/sfp-machine.h: Guard exception handling
code with _SOFT_FLOAT.
* config/i386/32/sfp-machine.h: Guard rounding handling
code with _SOFT_FLOAT.
* config/i386/64/sfp-machine.h: Ditto.

From-SVN: r193095

cfg-flags.def (ABNORMAL_CALL): Fix comment.

* cfg-flags.def (ABNORMAL_CALL): Fix comment.
(EH): Likewise.
(SIBCALL): Likewise.
* cfgrtl.c (rtl_verify_flow_info_1): Adjust error messages.
Deal with EDGE_SIBCALL and fix the EDGE_ABNORMAL check.

From-SVN: r193094

re PR libstdc++/55169 (std::discrete_distribution::operator(generator&) makes unnecessary copy of parameter vector)

2012-11-02  Paolo Carlini  <paolo.carlini@oracle.com>

PR libstdc++/55169
* include/bits/random.h: Remove all uses of param().
(chi_squared_distribution<>::__generate_impl(_ForwardIterator,
_ForwardIterator, _UniformRandomNumberGenerator&): Declare
* include/bits/random.tcc: ... define.
* include/ext/random: Remove all uses of param().

From-SVN: r193092

allocator.xml: Update reference to Hoard.

* doc/xml/manual/allocator.xml: Update reference to Hoard.
Update reference to Berger's OOPSLA 2002 paper.

From-SVN: r193091