2006-03-27 Maxim Kuvyrkov <mkuvyrkov@ispras.ru>
* haifa-sched.c (Copyright): Update date.
(try_ready): Change condition to restore non-speculative pattern
of the instruction.
(process_insn_depend_be_in_spec): Code to keep probability of the
speculative dependence non-decreasing.
From-SVN: r112427
2006-03-27 Andreas Krebbel <krebbel1@de.ibm.com>
* config/s390/s390-protos.h (s390_hard_regno_mode_ok,
s390_class_max_nregs): New function prototypes.
* config/s390/s390.c (REGNO_PAIR_OK): New macro.
(s390_hard_regno_mode_ok, s390_class_max_nregs): New functions.
* config/s390/s390.h (HARD_REGNO_NREGS, HARD_REGNO_MODE_OK,
CLASS_MAX_NREGS): Macro bodies replaced by function calls.
From-SVN: r112418
* io.c (check_io_constraints): Don't look at
dt->advance->value.charater.string, unless it is a CHARACTER
constant.
* gfortran.dg/advance_2.f90: New test.
* gfortran.dg/advance_3.f90: New test.
From-SVN: r112417
* dwarf2out.c (add_location_or_const_value_attribute): Call
tree_add_const_value_attribute if nothing else works.
(reference_to_unused): New.
(rtl_for_decl_init): Use reference_to_unused to decide whether
to output an initializer.
From-SVN: r112408
* tree-data-ref.c: Rename DDR_SIZE_VECT to DDR_NB_LOOPS.
(subscript_dependence_tester_1): Declared.
(print_dir_vectors, print_dist_vectors): New.
(debug_data_dependence_relation): New.
(dump_data_dependence_relation): Print more details.
(initialize_data_dependence_relation): Initialize DDR_LOOP_NEST.
(analyze_subscript_affine_affine): Don't ICE when gcd_alpha_beta is 0.
(save_dist_v, save_dir_v, add_outer_distances,
build_classic_dist_vector_1): New.
(build_classic_dist_vector): Rewrite to work on DDR_LOOP_NEST.
Don't test for lambda_vector_lexico_pos.
(same_access_functions, add_multivariate_self_dist,
add_other_self_distances, dir_from_dist): New.
(build_classic_dir_vector): Replace implementation almost identical to
build_classic_dist_vector with a walk of DDR_DIST_VECTS with a call to
dir_from_dist.
(subscript_dependence_tester_1): New.
(subscript_dependence_tester): Handle the lexicographically negative
distance vectors by recomputing the dependence relation.
(compute_affine_dependence): Remove parameter loop_nest_depth.
(compute_self_dependence): Don't call compute_subscript_distance.
(compute_all_dependences): Remove parameters nb_loops, loop_nest_depth.
Add a parameter for the loop_nest.
(find_loop_nest_1, find_loop_nest): New.
(compute_data_dependences_for_loop): Compute the loop nest, and give
up if the nest is not well formed.
* tree-data-ref.h (loop_p): New.
(struct data_dependence_relation): Replace size_vect field with
loop_nest, a vec of loops.
(DDR_SIZE_VECT): Renamed DDR_NB_LOOPS.
(DDR_LOOP_NEST): New.
(print_dir_vectors, print_dist_vectors,
debug_data_dependence_relation): Declared.
(index_in_loop_nest): New.
* tree-vect-analyze.c (vect_analyze_data_ref_dependence): Use
DDR_LOOP_NEST and index_in_loop_nest to determine the dependence
distance.
From-SVN: r112399
2006-03-25 Jerry DeLisle <jvdelisle@gcc.gnu.org>
PR libgfortran/26661
* io/io.h: Add read_sf so it can be used by read_x.
* io/transfer.c (read_sf): Pass no_error flag to read_sf. Use it to
break out rather than error on EOF or EOR conditions.
(read_block): Update call to read_sf.
(read_block_direct): Ditto.
* io/read.c (read_x): Use the modified read_sf instead of read_block.
From-SVN: r112390
2006-03-25 Thomas Koenig <Thomas.Koenig@online.de>
PR libfortran/26735
* io/transfer.c (data_transfer_init): Set u_flags.convert
on an unopened unit if specified by environment variable
(via get_unformatted_convert) or by compile-time option.
2006-03-25 Thomas Koenig <Thomas.Koenig@online.de>
PR libfortran/26735
* gfortran.dg/convert_implied_open.f90: New test case.
From-SVN: r112382
2006-03-25 Thomas Koenig <Thomas.Koenig@online.de>
PR fortran/26769
* iresolve.c (gfc_resolve_reshape): Remove doubling of
kind for complex. For real(kind=10), call reshape_r10.
(gfc_resolve_transpose): For real(kind=10), call
transpose_r10.
2006-03-25 Thomas Koenig <Thomas.Koenig@online.de>
PR fortran/26769
* Makefile.am: Add transpose_r10.c and reshape_r10.c.
* aclocal.m4: Regenerate using aclocal 1.9.3.
* Makefile.in: Regenerate using automake 1.9.3.
* m4/iparm.m4 (rtype_ccode): If rtype_letter is `i',
evaluate to rtype_kind, otherwise to rtype_code.
* generated/transpose_r10.c: Add.
* generated/reshape_r10.c: Add.
2006-03-25 Thomas Koenig <Thomas.Koenig@online.de>
PR fortran/26769
* gfortran.dg/transpose_reshape_r10.f90: New test case.
From-SVN: r112381
2006-03-25 Daniel Berlin <dberlin@dberlin.org>
PR tree-optimization/26804
* tree.h (DECL_CALL_CLOBBERED): New macro.
(tree_decl_common): Add call_clobbered_flag.
* tree-flow-inline.h (is_call_clobbered): Use DECL_CALL_CLOBBERED.
(mark_call_clobbered): Set DECL_CALL_CLOBBERED.
(clear_call_clobbered): Clear DECL_CALL_CLOBBERED.
(mark_non_addressable): Ditto.
* tree-ssa.c (verify_call_clobbered): New function.
(verify_alias_info): Use it.
* tree-pass.h (pass_reset_cc_flags): New prototype.
* tree-ssa-alias.c (pass_reset_cc_flags): New structure.
(reset_cc_flags): New function.
* passes.c (init_optimization_passes): Call reset_cc_flags after
initializing referenced_vars.
From-SVN: r112380
PR middle-end/26717
* fold-const.c (fold_binary) [RDIV_EXPR]: Do not optimize A / A
to 1.0 for non-real operands. Implement A / A optimization for
complex operands.
Co-Authored-By: Roger Sayle <roger@eyesopen.com>
From-SVN: r112379
* dependency.c (gfc_check_dependency): Improve handling of pointers;
Two variables of different types can't have a dependency, and two
variables with the same symbol are equal, even if pointers.
* gfortran.dg/dependency_12.f90: New test case.
From-SVN: r112377
* gfortran.h (gfc_symbol): Add a new "forall_index" bit field.
* match.c (match_forall_iterator): Set forall_index field on
the iteration variable's symbol.
* dependency.c (contains_forall_index_p): New function to
traverse a gfc_expr to check whether it contains a variable
with forall_index set in it's symbol.
(gfc_check_element_vs_element): Return GFC_DEP_EQUAL for scalar
constant expressions that don't variables used as FORALL indices.
* gfortran.dg/dependency_9.f90: New (resurected) test case.
From-SVN: r112373
2006-03-24 Geoffrey Keating <geoffk@apple.com>
* dwarf2out.c (is_c_family): Understand new DWARF3 language types.
(is_cxx): Likewise.
(gen_compile_unit_die): Use new language types for Objective-C and
Objective-C++.
(dwarf2out_decl): Use is_cxx rather than testing explicitly against
DW_LANG_C_plus_plus.
Index: testsuite/ChangeLog
2006-03-24 Geoffrey Keating <geoffk@apple.com>
* objc.dg/dwarf-2.m: New.
* obj-c++.dg/dwarf-2.mm: New.
From-SVN: r112372
2006-03-24 Andreas Krebbel <krebbel1@de.ibm.com>
* config/s390/s390.c (s390_decompose_address): Allow SImode for
index and base register.
(s390_expand_plug_operand, legitimate_address_p, preferred_la_operand_p,
print_operand_address, print_operand): Replaced REG_OK_FOR_BASE_STRICT_P
with REGNO_OK_FOR_BASE_P and REG_OK_FOR_INDEX_STRICT_P with
REGNO_OK_FOR_INDEX_P.
* config/s390/s390.h (REGNO_OK_FOR_INDEX_P): Replaced check with
ADDR_REGNO_P.
(REG_OK_FOR_INDEX_NONSTRICT_P, REG_OK_FOR_BASE_NONSTRICT_P,
REG_OK_FOR_INDEX_STRICT_P, REG_OK_FOR_BASE_STRICT_P, REG_OK_FOR_INDEX_P,
REG_OK_FOR_BASE_P): Definitions removed.
From-SVN: r112357
2006-03-24 Andreas Tobler <a.tobler@schweiz.ch>
John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
* configure.ac (THREADS): Add REENTRANT for *-*-hpux11*.
Warn about POSIX threads not being supported for *-*-hpux10*.
* configure: Regenerate.
Co-Authored-By: John David Anglin <dave.anglin@nrc-cnrc.gc.ca>
From-SVN: r112355
PR middle-end/26611
* gimplify.c (gimplify_bind_expr): Only call omp_add_variable on
non-global variables that don't have DECL_SEEN_IN_BIND_EXPR_P bit
set yet or weren't marked as local yet.
* g++.dg/gomp/pr26611-1.C: New test.
* g++.dg/gomp/pr26611-2.C: New test.
From-SVN: r112351
* config/rs6000/rs6000-protos.h (rs6000_offsettable_memref_p): Declare.
(rs6000_legitimate_small_data_p): Delete.
* config/rs6000/rs6000.c (rs6000_offsettable_memref_p): New predicate.
(rs6000_legitimate_small_data_p): Rename to legitimate_small_data_p
and make static. Add forward declaration. Update uses.
(rs6000_split_multireg_move): Use rs6000_offsettable_memref_p instead
of offsettable_memref_p.
* config/rs6000/rs6000.md (movdf_hardfloat32): Revert 2005-08-23 fix.
Use rs6000_offsettable_memref_p instead of a less accurate predicate.
From-SVN: r112349
* tree-ssa-dom.c (propagate_rhs_into_lhs): Temporarily work
around bug in immediate-use iterator.
* gcc.c-torture/compile/pr26833.c: New test.
* gfortran.fortran-torture/compile/pr26806.f90: New test.
From-SVN: r112348
* src/powerpc/ffitarget.h (enum ffi_abi): Add FFI_LINUX. Default
for 32-bit using IBM extended double format. Fix FFI_LAST_ABI.
* src/powerpc/ffi.c (ffi_prep_args_SYSV): Handle linux variant of
FFI_TYPE_LONGDOUBLE.
(ffi_prep_args64): Assert using IBM extended double.
(ffi_prep_cif_machdep): Don't munge FFI_TYPE_LONGDOUBLE type.
Handle FFI_LINUX FFI_TYPE_LONGDOUBLE return and args.
(ffi_call): Handle FFI_LINUX.
(ffi_closure_helper_SYSV): Non FFI_LINUX long double return needs
gpr3 return pointer as for struct return. Handle FFI_LINUX
FFI_TYPE_LONGDOUBLE return and args. Don't increment "nf"
unnecessarily.
* src/powerpc/ppc_closure.S (ffi_closure_SYSV): Load both f1 and f2
for FFI_TYPE_LONGDOUBLE. Move epilogue insns into case table.
Don't use r6 as pointer to results, instead use sp offset. Don't
make a special call to load lr with case table address, instead
use offset from previous call.
* src/powerpc/sysv.S (ffi_call_SYSV): Save long double return.
* src/powerpc/linux64.S (ffi_call_LINUX64): Simplify long double
return.
From-SVN: r112340
