backport: re PR target/52006 (ARM ICE with incorrectly peepholed tls_load_dot_plus_eight)

Backported from mainline
2012-01-28  Jakub Jelinek  <jakub@redhat.com>

PR target/52006
* config/arm/arm.md (pic_add_dot_plus_eight peephole2): Use
arm_general_register_operand predicate for operand 2 instead of
register_operand.

* gcc.target/arm/pr52006.c: New test.

From-SVN: r184057

backport: re PR libmudflap/40778 (Mudflap instrumentation missing in cloned function.)

Backported from mainline
2012-01-19  Jakub Jelinek  <jakub@redhat.com>

PR libmudflap/40778
* tree-mudflap.c (mf_artificial): New function.
(execute_mudflap_function_ops, execute_mudflap_function_decls,
mx_register_decls, mudflap_enqueue_decl): Use it.

* testsuite/libmudflap.c/fail68-frag.c: New test.

From-SVN: r184056

backport: re PR rtl-optimization/51767 (ICE with degenerated asm goto)

Backported from mainline
2012-01-05  Jakub Jelinek  <jakub@redhat.com>

PR rtl-optimization/51767
* cfgrtl.c (force_nonfallthru_and_redirect): Force addition
of jump_block and add an extra edge for degenerated asm gotos.

* gcc.c-torture/compile/pr51767.c: New test.

From-SVN: r184055

backport: re PR middle-end/51768 (ICE with invalid asm goto)

Backported from mainline
2012-01-05  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/51768
* stmt.c (check_unique_operand_names): Don't ICE during error
reporting if i is from labels chain.

* c-c++-common/pr51768.c: New test.

From-SVN: r184054

backport: re PR middle-end/44777 (ICE: SIGSEGV with -fprofile-use in gcc.c-torture/execute/comp-goto-2.c)

Backported from mainline
2012-01-05  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/44777
* profile.c (branch_prob): Split bbs that have exit edge
and need a fake entry edge too.

* gcc.dg/tree-prof/pr44777.c: New test.

From-SVN: r184053

backport: re PR debug/51695 (NOTE_INSN_VAR_LOCATION notes are sometimes too large)

Backported from mainline
2012-01-04  Jakub Jelinek  <jakub@redhat.com>

PR debug/51695
* dwarf2out.c (output_loc_list): For now drop >= 64KB expressions
in .debug_loc on the floor.

* gcc.dg/pr51695.c: New test.

From-SVN: r184052

backport: re PR c++/51669 (ICE verify-gimple with openmp)

Backported from mainline
2012-01-03  Jakub Jelinek  <jakub@redhat.com>

PR c++/51669
* semantics.c (finish_omp_clauses): Call fold_build_cleanup_point_expr
on OMP_CLAUSE_{IF,NUM_THREADS,SCHEDULE_CHUNK}_EXPR.

* g++.dg/gomp/pr51669.C: New test.

From-SVN: r184051

backport: re PR driver/48306 (presence of gcc subdir with . in PATH causes breakdown)

Backported from mainline
2012-01-26  Jakub Jelinek  <jakub@redhat.com>

* make-relative-prefix.c (make_relative_prefix_1): Avoid warning
about using preprocessor directives inside of macro arguments.

2012-01-02  Jakub Jelinek  <jakub@redhat.com>

* make-relative-prefix.c (make_relative_prefix_1): Avoid
stack overflow if PATH contains just a single entry and
HOST_EXECUTABLE_SUFFIX needs to be used.

PR driver/48306
* make-relative-prefix.c: Include sys/stat.h.
(make_relative_prefix_1): If access succeeds, check also stat
if nstore is a regular file.

From-SVN: r184050

backport: re PR c/51360 (spurious unused-but-set-variable warning for var used in OpenMP pragma)

Backported from mainline
2011-12-15  Jakub Jelinek  <jakub@redhat.com>

PR c/51360
* c-parser.c (c_parser_omp_clause_num_threads,
c_parser_omp_clause_schedule): Call mark_exp_read.

* semantics.c (finish_omp_clauses): For OMP_CLAUSE_NUM_THREADS_EXPR
and OMP_CLAUSE_SCHEDULE_CHUNK_EXPR call mark_rvalue_use.

* gcc.dg/gomp/pr51360.c: New test.
* g++.dg/gomp/pr51360.C: New test.
* g++.dg/gomp/pr51360-2.C: New test.

From-SVN: r184049

backport: re PR debug/51517 (Wrong debug information for pointers with negative strides.)

Backported from mainline
2011-12-15  Jakub Jelinek  <jakub@redhat.com>

PR debug/51517
* tree-ssa-coalesce.c (coalesce_ssa_name): For !optimize, test
!DECL_IGNORED_P instead of !DECL_ARTIFICIAL.

* trans-decl.c (gfc_get_symbol_decl): Don't set DECL_INITAL on span.
(gfc_trans_deferred_vars): Instead add its runtime initialization
here.

From-SVN: r184048

backport: re PR middle-end/52140 (Libdfp testcase encounters ICE when compiled with soft-dfp (no -mcpu).)

Backport from mainline
2012-02-09  Peter Bergner  <bergner@vnet.ibm.com>

gcc/
PR middle-end/52140
* dojump.c (do_compare_rtx_and_jump): Use SCALAR_FLOAT_MODE_P.

gcc/testsuite/
PR middle-end/52140
* gcc.dg/dfp/pr52140.c: New test.

From-SVN: r184046

backport: [multiple changes]

2012-02-09  Andrey Belevantsev  <abel@ispras.ru>

        Backport from mainline
        2012-01-20  Andrey Belevantsev  <abel@ispras.ru>

        PR target/51106
        * function.c (instantiate_virtual_regs_in_insn): Use
        delete_insn_and_edges when removing a wrong asm insn.

        Backport from mainline
        2012-01-20  Jakub Jelinek  <jakub@redhat.com>

        PR target/51106
        * gcc.dg/torture/pr51106-1.c: New test.
        * gcc.dg/torture/pr51106-2.c: New test.

From-SVN: r184038

Daily bump.

From-SVN: r184031

re PR fortran/52151 (Segfault with realloc on assignment and RESHAPE to unallocated LHS)

2012-02-08  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52151
        * trans-expr.c (fcncall_realloc_result): Set also the stride.

2012-02-08  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52151
        * gfortran.dg/realloc_on_assign_12.f90: New.

From-SVN: r184020

Daily bump.

From-SVN: r183990

re PR middle-end/51994 (git-1.7.8.3 miscompiled due to negative bitpos from get_inner_reference)

PR middle-end/51994
* expr.c (get_inner_reference): If there is an offset, add a negative
bit position to it (if any).

From-SVN: r183975

re PR target/40068 (GCC fails to apply dllexport attribute to typeinfo.)

2012-02-07  Kai Tietz  <ktietz@redhat.com>
            Dave Korn  <dave.korn.cygwin@gmail.com>

        PR target/40068
        * config/i386/winnt-cxx.c (i386_pe_adjust_class_at_definition):
        Take care that typinfo gets dllexport-attribute.

Co-Authored-By: Dave Korn <dave.korn.cygwin@gmail.com>
From-SVN: r183963

Daily bump.

From-SVN: r183950

re PR target/52107 (IBM 128bit long double constant loaded inefficiently)

PR target/52107
* config/rs6000/rs6000.c (rs6000_emit_move): Don't create DImode
subregs of TFmode.

From-SVN: r183947

Daily bump.

From-SVN: r183924

Daily bump.

From-SVN: r183899

Daily bump.

From-SVN: r183886

re PR fortran/52093 (Extra parenthesis with Size and Shape functions argument cause internal compiler error: Segmentation fault)

2012-02-03  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52093
        * simplify.c (gfc_simplify_size): Handle INTRINSIC_PARENTHESES.

2012-02-03  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52093
        * gfortran.dg/shape_7.f90: New.

From-SVN: r183877

re PR fortran/52012 (Wrong-code with realloc on assignment and RESHAPE w/ ORDER=)

2012-02-03  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/52012
* trans-expr.c (fcncall_realloc_result): Correct calculation of
result offset. If variable shape is correct, retain the bounds,
whatever they are.

2012-02-03  Paul Thomas  <pault@gcc.gnu.org>

PR fortran/52012
* gfortran.dg/realloc_on_assign_10.f90: New test.
* gfortran.dg/realloc_on_assign_11.f90: New test.

From-SVN: r183874

re PR java/48512 ([4.6] gcj spec files references incorrectly crtmt.o on i686-w64-mingw32 target)

        PR libjava/48512
        * configure.ac (THREADSTARTFILESPEC): Don't add crtmet.o file for
        w64 windows targets.
        * configure: Regenerated.

From-SVN: r183868

backport: re PR tree-optimization/51118 (ICE: tree check: expected tree that contains ‘typed’ structure, have ‘block’ in fold_checksum_tree, at fold-const.c:14160)

Backport from mainline:
2011-11-18  Uros Bizjak  <ubizjak@gmail.com>

PR tree-optimization/51118
* fold-const.c (fold_checksum_tree): Check for TS_TYPED structure
before using TREE_TYPE accessor on expr.

From-SVN: r183864

Daily bump.

From-SVN: r183858

Daily bump.

From-SVN: r183820

Fix file naming convention from r183797.

From-SVN: r183799

backport: re PR rtl-optimization/51374 ([avr] insn combine reorders volatile memory accesses)

gcc/
Backport from mainline r183796
PR rtl-optimization/51374
* combine.c (can_combine_p): Don't allow volatile_refs_p insns
to cross other volatile_refs_p insns.

gcc/testsuite/
Backport from mainline r183796
PR rtl-optimization/51374
* testsuite/gcc.target/avr/torture/pr51374-1.c: New.

From-SVN: r183797

re PR libstdc++/51795 (linear_congruential_engine doesn't work correctly)

2012-02-01  Paolo Carlini  <paolo.carlini@oracle.com>

PR libstdc++/51795
* include/bits/random.h (linear_congruential_generator): Add
static_assert preventing instantiation for values of 'a' and 'm'
currently handled incorrectly but _Mod::__calc.
* include/bits/random.tcc (seed_seq::generate): Avoid unsafe
uses of _Mod::__calc.

From-SVN: r183795

Daily bump.

From-SVN: r183786

Backport from mainline.

2011-01-31  Matthew Gretton-Dann  <matthew.gretton-dann@arm.com>
config/arm/thumb2.md (thumb2_mov_notscc): Use MVN for true
condition.

From-SVN: r183761

s390.md ("*ashr<mode>3_and"): Add missing z196 flag to srak instruction.

2012-01-31  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

* config/s390/s390.md ("*ashr<mode>3_and"): Add missing z196 flag
to srak instruction.

From-SVN: r183760

Daily bump.

From-SVN: r183748

re PR target/51835 (ARM EABI violation when passing arguments to helper floating functions like __aeabi_d2iz)

2012-01-30  Bin Cheng  <bin.cheng@arm.com>

PR target/51835
* config/arm/arm.c (arm_libcall_uses_aapcs_base): Use correct ABI
for __aeabi_d2iz/__aeabi_d2uiz with hard-float.

testcases:
PR target/51835
* gcc.target/arm/pr51835.c: New testcase.

From-SVN: r183734

re PR target/50313 (ARM: PIC code references a non-existant label)

Fix PR target/50313

2012-01-30  Ramana Radhakrishnan  <ramana.radhakrishnan@linaro.org>

Backport from mainline.
        2012-01-20  Ramana Radhakrishnan  <ramana.radhakrishnan@linaro.org>

PR target/50313
* config/arm/arm.c (arm_load_pic_register): Use
gen_pic_load_addr_unified. Delete calls to gen_pic_load_addr_32bit
, gen_pic_add_dot_plus_eight and gen_pic_add_dot_plus_four.
(arm_pic_static_addr): Likewise.
(arm_rtx_costs_1): Adjust cost for UNSPEC_PIC_UNIFIED.
(arm_note_pic_base): Handle UNSPEC_PIC_UNIFIED.
* config/arm/arm.md (UNSPEC_PIC_UNIFIED): Define.
(pic_load_addr_unified): New.

From-SVN: r183727

Daily bump.

From-SVN: r183691

Daily bump.

From-SVN: r183673

re PR fortran/52022 (Wrong-code with procedures passed as actual argument)

2012-01-28  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52022
        * trans-expr.c (gfc_conv_procedure_call): Fix passing
        of functions, which return allocatables.

2012-01-28  Tobias Burnus  <burnus@net-b.de>

        PR fortran/52022
        * gfortran.dg/dummy_procedure_7.f90: New.

From-SVN: r183665

Daily bump.

From-SVN: r183659

Daily bump.

From-SVN: r183597

Daily bump.

From-SVN: r183546

re PR target/51934 (FAIL: g++.dg/torture/pr51344.C  -O0  (test for excess errors)  due to "cdecl attribute ignored" warning)

PR target/51934
* g++.dg/torture/pr51344.C: Limit to x86.

From-SVN: r183526

tree-ssa-sccvn.c (vn_reference_eq): Also compare if both bases are dereferenced.

2012-01-25  Richard Guenther  <rguenther@suse.de>

* tree-ssa-sccvn.c (vn_reference_eq): Also compare if both
bases are dereferenced.

From-SVN: r183523

re PR fortran/51966 (ICE in gfc_conv_array_constructor_expr)

2012-01-25  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51966
        * resolve.c (resolve_structure_cons): Only create an
        array constructors for nonscalars.

2012-01-25  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51966
        * gfortran.dg/derived_constructor_char_3.f90: New.

From-SVN: r183511

Daily bump.

From-SVN: r183506

[multiple changes]

2012-01-24  Richard Guenther  <rguenther@suse.de>

Forward-port to branch
2010-09-21  Jakub Jelinek  <jakub@redhat.com>

PR middle-end/45678
* expr.c (expand_expr_real_1) <case VIEW_CONVERT_EXPR>: If
op0 isn't sufficiently aligned and there is movmisalignM
insn for mode, use it to load op0 into a temporary register.

From-SVN: r183471

re PR fortran/51948 ([OOP] Rejects valid: Function result value in MOVE_ALLOC, nested in SELECT TYPE)

2012-01-24  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51948
        * check.c (variable_check): Fix checking for
        result variables and deeply nested BLOCKs.

2012-01-24  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51948
        * gfortran.dg/move_alloc_12.f90: New.

From-SVN: r183469

Daily bump.

From-SVN: r183461

lang.c (java_init_options_struct): Set frontend_set_flag_trapping_math.

* lang.c (java_init_options_struct): Set
frontend_set_flag_trapping_math.

From-SVN: r183426

Daily bump.

From-SVN: r183401

Daily bump.

From-SVN: r183384

re PR fortran/51913 ([OOP] bug when submitting a class pointer to a subroutine)

2012-01-21  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51913
        * interface.c (compare_parameter): Fix CLASS comparison.

2012-01-21  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51913
        * gfortran.dg/class_47.f90: New.

From-SVN: r183369

re PR ada/46192 (wrong code for renaming of volatile packed array with address clause)

PR ada/46192
* gcc-interface/decl.c (gnat_to_gnu_entity) <object>: In the case of a
renaming, preserve the volatileness through the indirection, if any.

From-SVN: r183366

Daily bump.

From-SVN: r183362

* cfgrtl.c (rtl_dump_bb): Do not dump insns for {ENTRY|EXIT}_BLOCK.

From-SVN: r183353

* g++.dg/torture/pr51344.C: Fix typo.

From-SVN: r183331

Daily bump.

From-SVN: r183323

backport: re PR c++/51344 (cc1plus hangs when compiling)

        Backport from trunk
        PR c++/51344
        * decl2.c (save_template_attributes): Use merge_attributes
        instead of chaining up via TREE_CHAIN.

        * g++.dg/torture/pr51344.C: New test.

From-SVN: r183315

re PR fortran/51904 (ICE on SIZE function evaluation)

2012-01-19  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51904
        *expr.c (gfc_build_intrinsic_call): Also set the symtree.

2012-01-19  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51904
        * gfortran.dg/intrinsic_size_2.f90: New.

From-SVN: r183314

re PR target/48743 (-march=native mis-detects AMD K6-2+ / K6-3 as Athlon - compiled C fails with "illegal instruction")

2012-01-19  Quentin Neill  <quentin.neill@amd.com>

PR target/48743
* config/i386/driver-i386.c (host_detect_local_cpu): Also check
family to distinguish PROCESSOR_ATHLON.

From-SVN: r183309

Daily bump.

From-SVN: r183294

re PR tree-optimization/49642 (constant part of a macro not optimized away as expected due to splitter)

gcc:

2012-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/49642
* ipa-split.c (forbidden_dominators): New variable.
(check_forbidden_calls): New function.
(dominated_by_forbidden): Likewise.
(consider_split): Check for forbidden dominators.
(execute_split_functions): Initialize and free forbidden
dominators info; call check_forbidden_calls.

gcc/testsuite:

2012-01-11  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/49642
* gcc.dg/tree-ssa/pr49642.c: New test.

From-SVN: r183284

rs6000.md (call_value_indirect_aix32): Fix typo in mode of operand[4].

        * config/rs6000/rs6000.md (call_value_indirect_aix32): Fix typo
        in mode of operand[4].

From-SVN: r183282

Daily bump.

From-SVN: r183267

* gcc.dg/pr51821.c (dg-options): Add -std=c99.

From-SVN: r183234

Daily bump.

From-SVN: r183227

re PR c++/51854 (ICE in mangle.c with literal of complex floating type)

PR c++/51854
* mangle.c (write_template_arg_literal): Sorry instead of aborting.

From-SVN: r183224

re PR c++/51868 (Crash when generating the move constructor for a class with a bit field)

PR c++/51868
* typeck.c (build_static_cast_1): Handle bit-fields properly.

From-SVN: r183223

Daily bump.

From-SVN: r183203

re PR rtl-optimization/51821 (64bit > 32bit conversion produces incorrect results with optimizations)

PR rtl-optimization/51821
* recog.c (peep2_find_free_register): Determine clobbered registers
from insn pattern.

testsuite/ChangeLog:

PR rtl-optimization/51821
* gcc.dg/pr51821.c: New test.

From-SVN: r183198

Daily bump.

From-SVN: r183189

re PR fortran/51800 (Error: Automatic array 'dummy' at (1) cannot have an initializer)

2012-01-14  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51800
        * resolve.c (build_default_init_expr): Also initialize
        nonconstant-length strings with -finit-character=<n>.

2012-01-14  Tobias Burnus  <burnus@net-b.de>

        PR fortran/51800
        * gfortran.dg/init_flag_8.f90: New.
        * gfortran.dg/init_flag_9.f90: New.

From-SVN: r183181

Daily bump.

From-SVN: r183176

Daily bump.

From-SVN: r183147

backport: re PR target/51756 (wrong warning: uninitialized variable put into program memory area)

Backport from mainline r183129
PR target/51756
* config/avr/avr.c (avr_encode_section_info): Test for absence of
DECL_EXTERNAL when checking for initializers of progmem variables.

From-SVN: r183131

backport: arm.md (mov_notscc): Use MVN for false condition.

Backport from mainline
2012-01-11  Matthew Gretton-Dann  <matthew.gretton-dann@arm.com>

* gcc/config/arm/arm.md (mov_notscc): Use MVN for false condition.
* gcc/testsuite/testsuite/gcc.c-torture/execute/20120110-1.c: New testcase.

From-SVN: r183127

i386.md (*zero_extendsidi2_rex64): Correct movl template.

* config/i386/i386.md (*zero_extendsidi2_rex64): Correct movl template.
(x86_shift<mode>_adj_1): Remove operand constraint from operand 3.

From-SVN: r183125

Daily bump.

From-SVN: r183116

Daily bump.

From-SVN: r183084

Revert:

2008-09-18  Andrew Pinski  <andrew_pinski@playstation.sony.com>

PR rtl-opt/37451
* loop-doloop.c (doloop_modify): New argument zero_extend_p and
zero extend count after the correction to it is done.
(doloop_optimize): Update call to doloop_modify, don't zero extend
count before call.

2008-11-03  Andrew Pinski  <andrew_pinski@playstation.sony.com>

PR rtl-opt/37782
* loop-doloop.c (doloop_modify): Add from_mode argument that says what
mode count is in.
(doloop_optimize): Update call to doloop_modify.

testsuite:
* gcc.c-torture/execute/doloop-1.c,
gcc.c-torture/execute/doloop-2.c: New tests.

From-SVN: r183071

Daily bump.

From-SVN: r183046

mips.md (loadgp_newabi_<mode>): Add missing earlyclobber.

gcc/
* config/mips/mips.md (loadgp_newabi_<mode>): Add missing
earlyclobber.

From-SVN: r183035

re PR tree-optimization/51759 (miscompile writes past end of bitfield)

2012-01-09  Martin Jambor  <mjambor@suse.cz>

        PR tree-optimization/51759
* g++.dg/ipa/pr51759.C: New test.

From-SVN: r183031

* gcc-interface/trans.c (addressable_p) <COMPONENT_REF>: Fix thinko.

From-SVN: r183026

sol2-unwind.h (sparc64_is_sighandler): Check that the purported sigacthandler address isn't null before dereferencing it.

* config/sparc/sol2-unwind.h (sparc64_is_sighandler): Check that the
purported sigacthandler address isn't null before dereferencing it.
(sparc_is_sighandler): Likewise.

From-SVN: r183021

backport: re PR rtl-optimization/38644 (Optimization flag -O1 -fschedule-insns2 causes wrong code)

2012-01-09  Ramana Radhakrishnan  <ramana.radhakrishnan@linaro.org>

Backport from mainline
2011-11-04  Jiangning Liu  <jiangning.liu@arm.com>

PR rtl-optimization/38644
* config/arm/arm.c (thumb1_expand_epilogue): Add memory barrier
for epilogue having stack adjustment.

2012-01-09  Ramana Radhakrishnan  <ramana.radhakrishnan@linaro.org>

Backport from mainline:
2011-11-04  Jiangning Liu  <jiangning.liu@arm.com>

PR rtl-optimization/38644
* gcc.target/arm/stack-red-zone.c: New.

From-SVN: r183019

backport: headmerge-2.c: Adjust scan pattern.

2012-01-09  Andrew Stubbs  <ams@codesourcery.com>

Backport from mainline:

2012-01-06  Andrew Stubbs  <ams@codesourcery.com>

* gcc.target/arm/headmerge-2.c: Adjust scan pattern.

From-SVN: r183009

re PR ada/41929 (64-bit null_pointer_deref1 gnat.dg test consumes all available memory)

PR ada/41929
* config/sparc/sol2-unwind.h (sparc64_is_sighandler): Remove SAVPC and
add CFA.  Revert back to old code for Solaris 8+ multi-threaded.
(sparc_is_sighandler): Likewise.
(MD_FALLBACK_FRAME_STATE_FOR): Adjust call to IS_SIGHANDLER.

From-SVN: r183006

Daily bump.

From-SVN: r182998

Daily bump.

From-SVN: r182987

Daily bump.

From-SVN: r182978

backport: c-decl.c (ext_block): Moved up.

Backport from mainline
2012-01-06  Arnaud Charlet  <charlet@adacore.com>

* c-decl.c (ext_block): Moved up.
(collect_all_refs, for_each_global_decl): Take ext_block into account.

From-SVN: r182960

re PR middle-end/48660 (ARM ICE in expand_expr_real_1)

gcc/
PR middle-end/48660
* expr.h (copy_blkmode_to_reg): Declare.
* expr.c (copy_blkmode_to_reg): New function.
(expand_assignment): Don't expand register RESULT_DECLs before
the lhs.  Use copy_blkmode_to_reg to copy BLKmode values into a
RESULT_DECL register.
(expand_expr_real_1): Handle BLKmode decls when looking for promotion.

gcc/testsuite/
PR middle-end/48660
* g++.dg/pr48660.C: New test.

From-SVN: r182955

* ada/acats/overflow.lst: Add cb20004.

From-SVN: r182948

Daily bump.

From-SVN: r182941

re PR tree-optimization/51315 (unaligned memory accesses generated with -ftree-sra)

PR tree-optimization/51315
* tree-sra.c (tree_non_aligned_mem_for_access_p): New predicate.
(build_accesses_from_assign): Use it instead of tree_non_aligned_mem_p.

From-SVN: r182933

Daily bump.

From-SVN: r182894

re PR tree-optimization/51624 (Assert_Failure atree.adb:808 during stage 3)

PR tree-optimization/51624
* tree-sra.c (build_ref_for_model): When replicating a chain of
COMPONENT_REFs, stop as soon as the offset would become negative.

From-SVN: r182890