Danny Smith [Mon, 16 Jul 2007 09:19:26 +0000 (09:19 +0000)]
ctype_base.h: Use the generic libstdc++ masks rather than MSVCRT defines.
* config/os/mingw32/ctype_base.h: Use the generic libstdc++ masks
rather than MSVCRT defines.
* config/os/mingw32/ctype_noninline.h (classic_table): Construct
and return a static classic table based on generic libstdc++ masks.
Janne Blomqvist [Sun, 15 Jul 2007 21:33:32 +0000 (00:33 +0300)]
re PR target/32748 (gfortran.dg/array_constructor_6.f90)
2007-07-16 Janne Blomqvist <jb@gcc.gnu.org>
PR fortran/32748
* trans-decl.c (gfc_build_builtin_function_decls): Remove
DECL_IS_MALLOC attribute from internal_realloc, thus reverting
part of my 2007-07-03 patch.
Jerry DeLisle [Sun, 15 Jul 2007 16:29:19 +0000 (16:29 +0000)]
re PR libfortran/32611 (Print sign of negative zero)
2007-07-15 Jerry DeLisle <jvdelisle@gcc.gnu.org>
Francois-Xavier Coudert <fxcoudert@gcc.gnu.org>
PR fortran/32611
* gfortran.h (gfc_option_t): Add flag_sign_zero field.
* lang.opt (-fsign-zero): New option.
* trans.h: Rename gfor_fndecl_set_std into gfor_fndecl_set_options.
* trans-decl.c (gfc_build_builtin_function_decls): Build the function
declaration to pass an array containing the options to be used by the
runtime library. (gfc_generate_function_code): Build an array that
contains option values to be passed to the runtime library and the call
to the function.
* options.c (gfc_init_options): Initialize the flag_sign_zero field.
(gfc_handle_option): Handle the -fsign-zero option.
Jerry DeLisle [Sun, 15 Jul 2007 16:26:22 +0000 (16:26 +0000)]
re PR libfortran/32611 (Print sign of negative zero)
2007-07-15 Jerry DeLisle <jvdelisle@gcc.gnu.org>
Francois-Xavier Coudert <fxcoudert@gcc.gnu.org>
PR fortran/32611
* runtime/compile_options.c (set_std): Remove.
(set_options): New function.
(init_compile_options): Add initialization for -fsign-zero option.
* gfortran.map (GFORTRAN_1.0): Rename _gfortran_set_std into
_gfortran_set_options.
* libgfortran.h (compile_options_t): Add sign_zero field.
* io/write.c (output_float): Use the sign bit of the value to determine
if a negative sign should be emitted for zero values. Do not emit the
negative sign for zero if -fno-sign-zero was set during compile.
Jerry DeLisle [Sun, 15 Jul 2007 05:29:29 +0000 (05:29 +0000)]
re PR libfortran/32752 (Segfault on WRITE with modified unix_stream structure)
2007-07-14 Jerry DeLisle <jvdelisle@gcc.gnu.org>
PR libgfortran/32752
* io/unix.c (unix_stream): Move buffer pointer adjacent to small_buffer.
* io/transfer.c (formatted_transfer_scalar): If stream I/O, set
bytes_used to zero. Fix off by one error in calculation of pos and
skips. Eliminate duplicate pending_spaces check.
Thomas Koenig [Sat, 14 Jul 2007 20:39:10 +0000 (20:39 +0000)]
re PR libfortran/32731 (pack/unpack with kind=1 or kind=2 mask)
2007-07-14 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/32731
* iresolve.c(gfc_resolve_pack): A scalar mask has
to be kind=4, an array mask with kind<4 is converted
to gfc_default_logical_kind automatically.
(gfc_resolve_unpack): Convert mask to gfc_default_lotical_kind
if it has a kind<4.
2007-07-14 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/32731
* gfortran.dg/pack_mask_1.f90: New test.
* gfortran.dg/unpack_mask_1.f90: New test.
target-supports.exp (check_effective_target_vect_aligned_arrays): New procedure to check if arrays are naturally aligned to the vector alignment...
* lib/target-supports.exp (check_effective_target_vect_aligned_arrays):
New procedure to check if arrays are naturally aligned to the vector
alignment boundary.
* gcc.dg/vect/pr25413a.c: Do not include stdarg.h. Use __SIZE_TYPE__
in size_t typedef. Adjust tree dump scan results according to
vect_aligned_arrays target check.
* gcc.dg/vect/pr25413.c: Adjust tree dump scan results according to
vect_aligned_arrays target check.
* gcc.dg/vect/pr31699.c: Ditto.
Add ability to track uninitialized variables, and mark uninitialized variables in the Dwarf debug info.
Add ability to track uninitialized variables, and mark uninitialized
variables in the Dwarf debug info. Controlled by compile option
-fvar-tracking-uninit
Sa Liu [Fri, 13 Jul 2007 18:31:08 +0000 (18:31 +0000)]
config.gcc: Add options for arch and tune on SPU.
2007-07-13 Sa Liu <saliu@de.ibm.com>
* config.gcc: Add options for arch and tune on SPU.
* config/spu/predicates.md: Add constant operands 0 and 1.
* config/spu/spu-builtins.def: Add builtins for double precision
floating point comparison: si_dfceq, si_dfcmeq, si_dfcgt, si_dfcmgt,
si_dftsv, spu_cmpeq_13, spu_cmpabseq_1, spu_cmpgt_13, spu_cmpabsgt_1,
spu_testsv.
* config/spu/spu-c.c: Define __SPU_EDP__ when builtins invoked with
a CELLEDP target.
* config/spu/spu-protos.h: Add new function prototypes.
* config/spu/spu.c (spu_override_options): Check options -march and
-mtune.
(spu_comp_icode): Add comparison code for DFmode and vector mode.
(spu_emit_branch_or_set): Use the new code for DFmode and vector
mode comparison.
(spu_const_from_int): New. Create a vector constant from 4 ints.
(get_vec_cmp_insn): New. Get insn index of vector compare instruction.
(spu_emit_vector_compare): New. Emit vector compare.
(spu_emit_vector_cond_expr): New. Emit vector conditional expression.
* config/spu/spu.h: Add options -march and -mtune. Define processor
types PROCESSOR_CELL and PROCESSOR_CELLEDP. Define macro
CANONICALIZE_COMPARISON.
* config/spu/spu.md: Add new insns for double precision compare
and double precision vector compare. Add vcond and smax/smin patterns
to enable DFmode vector conditional expression.
* config/spu/spu.opt: Add options -march and -mtune.
* config/spu/spu_internals.h: Add builtins for CELLEDP target:
si_dfceq, si_dfcmeq, si_dfcgt, si_dfcmgt, si_dftsv. Add builtin for
both CELL and CELLEDP targets: spu_testsv.
* config/spu/spu_intrinsics.h: Add flag mnemonics for test special
values.
testsuite/
* gcc.dg/vect/fast-math-vect-reduc-7.c: Switch on test
for V2DFmode vector conditional expression.
* gcc.target/spu/dfcmeq.c: New. Test combination of abs
and dfceq patterns.
* gcc.target/spu/dfcmgt.c: New. Test combination of abs
and dfcgt patterns.
* gcc.target/spu/intrinsics-2.c: New. Test intrinsics for
V2DFmode comparison and test special values.
* lib/target-supports.exp: Switch on test for V2DFmode
vector conditional expression.
Richard Guenther [Fri, 13 Jul 2007 15:41:02 +0000 (15:41 +0000)]
re PR tree-optimization/32721 (CCP removes volatile qualifiers.)
2007-07-13 Richard Guenther <rguenther@suse.de>
PR tree-optimization/32721
* tree-ssa-ccp.c (maybe_fold_stmt_indirect): Preserve
TREE_THIS_VOLATILE on the folded reference.
* tree-ssa-operands.c (get_expr_operands): Set has_volatile_ops
if the array reference has TREE_THIS_VOLATILE set.
Andrew Haley [Fri, 13 Jul 2007 15:36:39 +0000 (15:36 +0000)]
libgcj.ver: Add __gcj_personality_sj0.
2007-07-13 Andrew Haley <aph@redhat.com>
* libgcj.ver: Add __gcj_personality_sj0.
* testsuite/libjava.jvmti/jvmti-interp.exp: Likewise.
* testsuite/libjava.jni/jni.exp: Use -fdollars-in-identifiers.
* testsuite/libjava.jni/cni.exp: Use -fdollars-in-identifiers.
* testsuite/libjava.jvmti/jvmti.exp (gcj_jvmti_compile_cxx_to_o): Likewise.
* gnu/classpath/natVMStackWalker.cc (getCallingClassLoader): Check
klass is non-null.
* java/lang/reflect/natField.cc (getAddr): Call
_Jv_StackTrace::GetCallingClass only if CALLER is non-null.
* java/lang/reflect/natVMProxy.cc (run_proxy): Use
_Jv_getFieldInternal to get field proxyClass.m.
(_Jv_getFieldInternal): New function.
Andrew Haley [Fri, 13 Jul 2007 14:07:16 +0000 (14:07 +0000)]
[multiple changes]
2007-07-13 Andrew Haley <aph@redhat.com>
* libgcj.ver: Add __gcj_personality_sj0.
* testsuite/libjava.jvmti/jvmti-interp.exp: Likewise.
* testsuite/libjava.jni/jni.exp: Use -fdollars-in-identifiers.
* testsuite/libjava.jni/cni.exp: Use -fdollars-in-identifiers.
* testsuite/libjava.jvmti/jvmti.exp (gcj_jvmti_compile_cxx_to_o): Likewise.
* gnu/classpath/natVMStackWalker.cc (getCallingClassLoader): Check
klass is non-null.
* java/lang/reflect/natField.cc (getAddr): Call
_Jv_StackTrace::GetCallingClass only if CALLER is non-null.
* java/lang/reflect/natVMProxy.cc (run_proxy): Use
_Jv_getFieldInternal to get field proxyClass.m.
(_Jv_getFieldInternal): New function.
Daniel Franke [Thu, 12 Jul 2007 22:15:11 +0000 (18:15 -0400)]
re PR fortran/31639 ([4.1/4.2/4.3] ICE in gfc_conv_constant, at fortran/trans-const.c:348 with len)
gcc/fortran:
2007-07-12 Daniel Franke <franke.daniel@gmail.com>
PR fortran/31639
* decl.c (gfc_match_suffix): Removed surplus general error that hides
a more specific message.
* resolve.c (resolve_fl_variable): Reject illegal initializiers only
if not already done.
(resolve_fl_procedure): Added check for initializers of functions.
gcc/testsuite:
2007-07-12 Daniel Franke <franke.daniel@gmail.com>
PR fortran/31639
* gfortran.dg/func_decl_4.f90: New test.
re PR fortran/32599 ([ISO C Binding] Accepts character with len /= 1)
2007-07-12 Christopher D. Rickett <crickett@lanl.gov>
PR fortran/32599
* decl.c (verify_c_interop_param): Require character string dummy
args to BIND(C) procedures to have length 1.
* resolve.c (resolve_fl_procedure): Modify parameter checking for
BIND(C) procedures.
PR fortran/32601
* resolve.c (gfc_iso_c_func_interface): Verify that a valid
expression is given as an argument to C_LOC and C_ASSOCIATED.
* trans-io.c (transfer_expr): Add argument for code block. Add
standards check to determine if an error message should be
reported for printing C_PTR or C_FUNPTR.
(transfer_array_component): Update arguments to transfer_expr.
(gfc_trans_transfer): Ditto.
* symbol.c (gen_cptr_param): Fix whitespace.
2007-07-12 Christopher D. Rickett <crickett@lanl.gov>
PR fortran/32599
* gfortran.dg/32599.f03: New test case.
PR fortran/32601
* gfortran.dg/32601.f03: New test case.
* gfortran.dg/32601_1.f03: Ditto.
* gfortran.dg/c_ptr_tests_9.f03: Updated dg-options.
* gfortran.dg/c_ptr_tests_10.f03: Ditto.
builtins.c (get_pointer_alignment): Honor DECL_ALIGN on a FUNCTION_DECL.
* builtins.c (get_pointer_alignment): Honor DECL_ALIGN on a
FUNCTION_DECL.
* tree.c (build_decl_stat): Move code from here...
(make_node_stat): ... to here. Don't uselessly clear DECL_USER_ALIGN.
(expr_align): Honor DECL_ALIGN on a FUNCTION_DECL. Add comment
about using DECL_ALIGN of LABEL_DECL and CONST_DECL.
* tree.h (DECL_USER_ALIGN): Fix misplaced comment.
* varasm.c (assemble_start_function): Use DECL_ALIGN instead of
FUNCTION_BOUNDARY.
Jakub Jelinek [Thu, 12 Jul 2007 12:16:54 +0000 (14:16 +0200)]
re PR fortran/32550 (openmp: COPYPRIVATE of pointer variables fails)
PR fortran/32550
* trans.h (GFC_POINTER_TYPE_P): Define.
* trans-types.c (gfc_sym_type): Set it for types on attr->sym.pointer.
* trans-openmp.c (gfc_omp_privatize_by_reference): Return false
if GFC_POINTER_TYPE_P is set on the type.
* testsuite/libgomp.fortran/pr32550.f90: New test.
* testsuite/libgomp.fortran/crayptr2.f90: New test.
Richard Guenther [Thu, 12 Jul 2007 10:27:51 +0000 (10:27 +0000)]
gimplify.c (gimplify_conversion): Make sure that the result from maybe_fold_offset_to_reference is trivially...
2007-07-12 Richard Guenther <rguenther@suse.de>
* gimplify.c (gimplify_conversion): Make sure that the result
from maybe_fold_offset_to_reference is trivially convertible
to the desired type before doing the simplification.
(gimplify_expr): Likewise.
* fold-const.c (fold_binary): Use the correct types for
building the simplified expression.
Richard Guenther [Thu, 12 Jul 2007 09:30:27 +0000 (09:30 +0000)]
trans-intrinsic.c (gfc_conv_intrinsic_repeat): Convert arguments to gfc_charlen_type_node.
2007-07-12 Richard Guenther <rguenther@suse.de>
* trans-intrinsic.c (gfc_conv_intrinsic_repeat): Convert
arguments to gfc_charlen_type_node.
* trans-io.c (gfc_convert_array_to_string): Convert type
size to gfc_array_index_type.
Richard Guenther [Thu, 12 Jul 2007 09:29:26 +0000 (09:29 +0000)]
expr.c (expand_java_return): RETURN_EXPR has void type.
2007-07-12 Richard Guenther <rguenther@suse.de>
* expr.c (expand_java_return): RETURN_EXPR has void type.
(build_jni_stub): Likewise. Use a comparison against zero
for null-pointer test in COND_EXPR.
(build_field_ref): Build POINTER_PLUS_EXPR with correct
type. Convert result instead.
(build_invokevirtual): Likewise.
Kai Tietz [Thu, 12 Jul 2007 07:22:06 +0000 (07:22 +0000)]
splay-tree.h (libi_uhostptr_t, [...]): New types, needed for WIN64 when a long is not wide enough for a pointer.
* splay-tree.h (libi_uhostptr_t, libi_shostptr_t): New types,
needed for WIN64 when a long is not wide enough for a pointer.
(splay_tree_key, splay_tree_value): Use the new types.
maint-tool (deps): Output stamp-h instead of config.h.
* maint-tool (deps): Output stamp-h instead of config.h.
* Makefile.in: Rebuild deps.
(maintainer-clean-subdir): Depend on stamp-h rather than config.h.
Keith Seitz [Wed, 11 Jul 2007 17:37:24 +0000 (17:37 +0000)]
interpret.cc (get_local_var_table): Make sure the method is compiled.
* interpret.cc (get_local_var_table) [DIRECT_THREADED]: Make sure the
method is compiled.
* testsuite/libjava.jvmti/interp/getlocalvartable.out: Revise
for compiled methods.
* params.def (PARAM_VERIFY_CANONICAL_TYPES): Remove.
(PARAM_USE_CANONICAL_TYPES): New; decides whether to use canonical
types or not.
* params.h (VERIFY_CANONICAL_TYPES): Remove.
(USE_CANONICAL_TYPES): New.
* doc/invoke.texi (verify-canonical-types): Remove.
(use-canonical-types): Add.
2007-07-11 Douglas Gregor <doug.gregor@gmail.com>
* typeck.c (comptypes): When USE_CANONICAL_TYPES, use the
canonical types; otherwise, fall back to structural type
comparisons. If ENABLE_CHECKING and USE_CANONICAL_TYPES, give an
internal compiler error if the canonical types are wrong.