Maxim Kuvyrkov [Wed, 7 May 2008 08:09:27 +0000 (08:09 +0000)]
Cleanup ColdFire scheduling support and add V4 pipeline model.
* config/m68k/m68k.md (UNSPEC_TIE): New constant.
(define_attr cpu): Add cfv4 value.
(define_attr type, define_attr type1): Merge into a single 'type'
attribute. Update all uses.
(define_attr opx_type, define_attr opy_type, define_attr opx_access):
Rearrange and update. Rename value 'reg' to 'Rn', add value 'FPn'.
Update all uses.
(define_attr opx_mem, define_attr opy_mem): Remove.
(define_attr op_mem): Clean up, update comment.
(define_attr size): Use specific values instead of general int.
(define_attr guess, define_attr split): Remove. Update all uses.
(movdf_internal, tstsi_internal, tsthi_internal, tstqi_internal,
tst<mode>_68881, pushexthisi_const, movsi_const0_68000_10,
movsi_const0_68040_60, movsi_const0, movsi_cf, movstrictqi_cf,
zero_extendhisi2_cf, zero_extendqisi2_cfv4, cfv4_extendhisi2,
68k_extendhisi2, extendqihi2, cfv4_extendqisi2, 68k_extendqisi2,
floatsi<mode>2_68881, ftrunc<mode>2_68881, ftrunc<mode>2_cf,
fix<mode>qi2_68881, fix<mode>hi2_68881, fix<mode>si2_68881,
adddi_dishl32, addsi3_5200, add<mode>3_floatsi_68881,
add<mode>3_floathi_68881, add<mode>3_floatqi_68881,
add<mode>3_68881, add<mode>3_cf, subdi_dishl32, subsi3,
sub<mode>3_floatsi_68881, sub<mode>3_floathi_68881,
sub<mode>3_floatqi_68881, sub<mode>3_68881, sub<mode>3_cf,
mulhi3, mulhisi3, mulhisisi3_s, mulsi3_68020, mulsi3_cf,
umulhisi3, mulhisisi3_z, mul<mode>3_floatsi_68881,
mul<mode>3_floathi_68881, mul<mode>3_floatqi_68881, fmul<mode>3_cf,
div<mode>3_cf, sqrt<mode>2_cf, abs<mode>2_cf, clzsi2,
one_cmplsi2_5200, subreghi1ashrdi_const32, ashrsi3, lshrsi3,
bsetmemqi, bsetmemqi_ext, bclrmemqi, bclrmemqi_ext,
beq, bne, bgt, blt, bordered, bunordered, buneq, bunge, bungt, bunle,
bunlt, bltgt, tablejump_internal, call, non_symbolic_call_value,
symbolic_call_value_jsr, symbolic_call_value_bsr, link):
Update or set attributes.
(stack_tie): New fake instruction.
* config/m68k/m68k.h (TUNE_CFV4): New macro.
(m68k_sched_attr_size): Update declaration.
(m68k_sched_attr_type2): Remove.
(m68k_sched_address_bypass_p, m68k_sched_indexed_address_bypass_p):
Declare new bypass predicates.
* config/m68k/m68k.c (m68k_sched_issue_rate,
m68k_sched_first_cycle_multipass_dfa_lookahead): Declare hook
implementations.
(TARGET_SCHED_ISSUE_RATE,
TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD): Override hooks.
(override_options): Handle scheduling for ColdFire V4 core.
(m68k_expand_prologue): Emit stack_tie.
(enum attr_op_type): Split value 'OP_TYPE_REG' to 'OP_TYPE_RN' and
'OP_TYPE_FPN'. Update all uses.
(sched_guess_p): Remove.
(sched_address_type): Handle symbolic addresses.
(sched_get_operand): New static function.
(sched_operand_type): Merge into sched_attr_op_type.
(sched_attr_op_type): Handle FP registers, handle quick constants,
update.
(m68k_sched_attr_opx_type, m68k_sched_attr_opy_type): Update.
(m68k_sched_attr_size): Update. Move logic to ...
(sched_get_attr_size_int): New static function.
(sched_get_opxy_mem_type): New static function.
(m68k_sched_attr_op_mem): Update.
(m68k_sched_attr_type2): Remove.
(sched_cfv4_bypass_data): New static variable.
(m68k_sched_adjust_cost): Handle ColdFire V4 bypass.
(m68k_sched_issue_rate): Implement scheduler hook.
(struct _sched_ib: enabled_p): New field.
(m68k_sched_variable_issue): Update. Handle V4.
(SCHED_DUMP_TODO, SCHED_DUMP_DONE, SCHED_DUMP_NOTHING,
sched_dump_class_func_t, sched_dump_split_class,
sched_dump_dfa_guess_unit_code, sched_dump_dfa_state,
sched_dump_dfa_class, m68k_sched_dump): Remove.
(m68k_sched_first_cycle_multipass_dfa_lookahead): Implement scheduler
hook.
(m68k_sched_init_global): Remove statisctics dumping, introduce
sanity check that all instructions have pipeline reservations. Handle
ColdFire V4 core.
(m68k_sched_dfa_pre_advance_cycle, m68k_sched_dfa_post_advance_cycle):
Handle ColdFire V4 core.
(sched_mem_operand_p, sched_get_reg_operand, sched_get_mem_operand):
New static functions.
(m68k_sched_address_bypass_p): New bypass predicate.
(sched_get_indexed_address_scale): New static function.
(m68k_sched_indexed_address_bypass_p): New bypass predicate.
* cf.md: Update comments.
(define_attr type2): Remove. Use 'type' attribute instead.
Update all uses.
(cf_ib): Rename to cfv123_ib. Update all uses.
(cf_oep): Rename to cfv123_oep. Update all uses.
(cf_chr): Rename to cfv123_chr. Update all uses.
(cf_mem): Rename to cfv123_mem. Update all uses.
(cf_mac): Move to more appropriate place.
(cfv123_guess): New automaton and cpu_unit.
(cfv123_*, cfv12_*, cfv1_*, cfv2_*, cfv3_*): Use type attribute.
Update uses of 'size' attribute. Handle before reload scheduling.
(cfv123_guess): New dummy reservation for unhandled instructions.
(cfv4_*): Pipeline description of ColdFire V4 core.
(ignore): New reservation to handle 'ignore' type.
Jakub Jelinek [Wed, 7 May 2008 07:28:14 +0000 (09:28 +0200)]
re PR middle-end/36106 (#pragma omp atomic issues with floating point types)
PR middle-end/36106
* omp-low.c (expand_omp_atomic_pipeline): Load value using the
integral type rather than floating point, then VIEW_CONVERT_EXPR
to the floating point type.
* testsuite/libgomp.c/atomic-5.c: New test.
* testsuite/libgomp.c/atomic-6.c: New test.
* testsuite/libgomp.c/autopar-1.c: New test.
simplify.c (gfc_simplify_bessel_j0,gfc_simplify_bessel_j1, [...]): Mark arguments with ATTRIBUTE_UNUSED.
* simplify.c (gfc_simplify_bessel_j0,gfc_simplify_bessel_j1,
gfc_simplify_bessel_jn,gfc_simplify_bessel_y0,
gfc_simplify_bessel_y1,gfc_simplify_bessel_yn): Mark arguments
with ATTRIBUTE_UNUSED.
Tom Tromey [Tue, 6 May 2008 17:15:07 +0000 (17:15 +0000)]
PR preprocessor/35313, PR preprocessor/36088:
gcc/testsuite
PR preprocessor/35313, PR preprocessor/36088:
* gcc.dg/cpp/pr35313.c: New file.
* gcc.dg/cpp/if-oppr.c: Remove test for ',' in a conditional
expression.
* gcc.dg/cpp/if-oppr2.c: New file.
libcpp
PR preprocessor/35313, PR preprocessor/36088:
* expr.c (optab) <QUERY, COMMA>: Set precedence to 4.
(reduce) <case CPP_QUERY>: Special case CPP_COMMA and CPP_COLON.
H.J. Lu [Tue, 6 May 2008 15:41:08 +0000 (08:41 -0700)]
re PR target/35657 (Alignments of DFP types aren't consistent)
2008-05-06 H.J. Lu <hongjiu.lu@intel.com>
PR target/35657
* config/i386/i386.c (contains_128bit_aligned_vector_p): Renamed
to ...
(contains_aligned_value_p): This. Handle _Decimal128.
(ix86_function_arg_boundary): Only align _Decimal128 to its
natural boundary and handle it properly.
Martin Jambor [Tue, 6 May 2008 14:15:18 +0000 (16:15 +0200)]
ipa-cp.c (ipcp_method_orig_node): Renamed to ipcp_get_orig_node.
2008-05-06 Martin Jambor <mjambor@suse.cz>
* ipa-cp.c (ipcp_method_orig_node): Renamed to ipcp_get_orig_node.
(ipcp_method_is_cloned): Renamed to ipcp_node_is_clone
(ipcp_method_set_orig_node): Removed.
(ipcp_cval_get_cvalue_type): Removed.
(ipcp_method_get_scale): Renamed to ipcp_get_node_scale.
(ipcp_method_set_scale): Renamed to ipcp_set_node_scale.
(ipcp_cval_set_cvalue_type): Removed.
(ipcp_cval_get_cvalue): Removed.
(ipcp_cval_set_cvalue): Removed.
(ipcp_type_is_const): Renamed to ipcp_lat_is_const.
(ipcp_cval_equal_cvalues): Renamed to ipcp_lats_are_equal
(ipcp_lats_are_equal): Changed parameters to two ipcp_lattice's
(ipcp_cval_meet): Renamed to ipa_lattice_meet
(ipcp_cval_changed): Changed to use ipcp_lat_is_const
(ipcp_method_cval): Renamed to ipcp_get_ith_lattice
(ipcp_get_ith_lattice): Changed parameters.
(ipcp_cval_compute): Renamed to ipcp_lattice_from_jfunc
(ipcp_lattice_from_jfunc): Changed parameters.
(ipcp_redirect): Local lattice pointer instead of lattice type variable.
(ipcp_method_cval_print): Added temporary variable info.
(ipcp_redirect): Removed already unused local variable caller.
(ipcp_redirect): New temporary variable orig_callee_info
(ipcp_redirect): Removed newly unused local variable callee.
(ipcp_redirect): Removed (a bit confusing) local variable type.
(ipcp_insert_stage): Added local variable info.
(ipcp_cval_changed): Renamed to ipcp_lattice_changed, parameters
renamed too
(ipcp_formal_create): Removed.
(ipcp_method_cval_set): Removed.
(ipcp_propagate_stage): Renamed lattice variables.
(ipcp_method_cval_set_cvalue_type): Removed.
(ipcp_method_cval_print): Renamed to ipcp_print_all_lattices
(ipcp_print_all_lattices): Changed printed strings to refer to
lattices rather than cvals.
(ipcp_method_cval_init): Renamed to ipcp_initialize_node_lattices
(ipcp_propagate_const): Changed formal parameters.
(build_const_val): Changed formal parameters.
(ipcp_insert_stage): Removed useless variable cvalue
(build_const_val): Changed formal parameters.
(ipcp_method_compute_scale): Renamed to ipcp_compute_node_scale
(ipcp_after_propagate): Renamed to ipcp_change_tops_to_bottom
(ipcp_callsite_param_print): Renamed to ipcp_print_all_jump_functions
(ipcp_profile_mt_count_print): Renamed to ipcp_print_func_profile_counts
(ipcp_print_func_profile_counts): Changed string from "method" to
"function"
(ipcp_profile_cs_count_print): Renamed to ipcp_print_call_profile_counts
(ipcp_profile_edge_print): Renamed to ipcp_print_edge_profiles
(ipcp_profile_bb_print): Renamed to ipcp_print_bb_profiles
(ipcp_structures_print): Renamed to ipcp_print_all_structures
(ipcp_profile_print): Renamed to ipcp_print_profile_data
(ipcp_lat_is_const): Changed parameters and made inline.
(ipcp_replace_map_create): Renamed to ipcp_create_replace_map
(ipcp_redirect): Renamed to ipcp_need_redirect_p
(ipcp_need_redirect_p): Calls ipcp_lat_is_const instead of using
the predicate condition directly
(ipcp_propagate_stage): Added local variable args. Removed local
variable callee. (Both are mere code simplifications.)
(ipcp_method_dont_insert_const): Renamed to
ipcp_node_not_modifiable_p.
(ipcp_node_not_modifiable_p): Made inline.
(ipcp_cloned_create): Renamed to ipcp_init_cloned_node
(ipcp_propagate_const): Renamed to ipcp_propagate_one_const
(ipcp_print_all_lattices): Removed variable cvalue
(ipcp_method_scale_print): Renamed to ipcp_function_scale_print
Updated comments.
Kai Tietz [Tue, 6 May 2008 10:07:13 +0000 (10:07 +0000)]
i386.c (output_set_got): Fix for x86_64 output_emit_asm by using 'q' specifier for instruction.
2008-05-06 Kai Tietz <kai.tietz@onevision.com>
* config/i386/i386.c (output_set_got): Fix for x86_64 output_emit_asm
by using 'q' specifier for instruction.
(ix86_file_end): Replaced case TARGET_64BIT_MS_ABI by TARGET_64BIT.
Anatoly Sokolov [Tue, 6 May 2008 09:30:17 +0000 (13:30 +0400)]
avr.md (*sbrx_branch, [...]): Change mode of zero_extract from QImode to HImode.
* config/avr/avr.md (*sbrx_branch, *sbix_branch, *sbix_branch_tmp):
Change mode of zero_extract from QImode to HImode.
(sign bit tests peepholes): (Ditto.).
* include/parallel/multiway_merge.h:
(multiway_merge_*_unguarded):
Pass sentinel directly, to allow correct determination.
(multiway_merge_loser_tree_unguarded):
Remove over-cautious assertion.
(calls to multiway_merge_*_splitting):
Parametrize with type that is correct in all cases.
* include/parallel/losertree.h:
(delete_min_insert (in many classes)):
Correct and standardize assertions.
Uros Bizjak [Tue, 6 May 2008 08:39:51 +0000 (10:39 +0200)]
mmx.md: Remove double backslashes from asm templates.
* config/i386/mmx.md: Remove double backslashes from asm templates.
(*addv2sf3): Rename from mmx_addv2sf3 insn pattern.
(mmx_addv2sf3): New expander. Use ix86_fixup_binary_operands_no_copy
to handle nonimmediate operands.
(*mulv2sf3): Rename from mmx_mulv2sf3 insn pattern.
(mmx_mulv2sf3): New expander. Use ix86_fixup_binary_operands_no_copy
to handle nonimmediate operands.
(*<code>v2sf3_finite): New insn pattern.
(*<code>v2sf3): Rename from mmx_<code>v2sf3 insn pattern.
(mmx_<code>v2sf3): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(mmx_<plusminus_insn><mode>3): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(*<plusminus_insn><mode>3): New insn pattern.
(mmx_add<mode>3): Removed.
(mmx_ssadd<mode>3): Ditto.
(mmx_usadd<mode>3): Ditto.
(mmx_sub<mode>3): Ditto.
(mmx_sssub<mode>3): Ditto.
(mmx_ussub<mode>3): Ditto.
(*mulv4hi3): Rename from mmx_mulv4hi3 insn pattern.
(mmx_mulv4hi3): New expander. Use ix86_fixup_binary_operands_no_copy
to handle nonimmediate operands.
(*smulv4hi3_highpart): Rename from mmx_smulv4hi3_highpart
insn pattern.
(mmx_smulv4hi3_highpart): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(*umulv4hi3_highpart): Rename from mmx_umulv4hi3_highpart
insn pattern.
(mmx_umulv4hi3_highpart): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(*<code>v4hi3): Rename from mmx_<code>v4hi3 insn pattern.
(mmx_<code>v4hi3): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(*<code>v8qi3): Rename from mmx_<code>v8qi3 insn pattern.
(mmx_<code>v8qi3): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
(*<code><mode>3): Rename from mmx_<code><mode>3 insn pattern.
(mmx_<code><mode>3): New expander. Use
ix86_fixup_binary_operands_no_copy to handle nonimmediate operands.
Danny Smith [Tue, 6 May 2008 02:59:12 +0000 (02:59 +0000)]
target_supports.exp (check_effective_target_4byte_wchar_t): New proc.
* lib/target_supports.exp (check_effective_target_4byte_wchar_t):
New proc.
* gcc.dg/utf16-4.c: Use it.
* gcc.dg/utf32-4.c: Use it.
* g++.dg/ext/utf16-4.C: Use it.
* gcc.dg/ext/utf32-4.C: Use it.
Andrew Pinski [Mon, 5 May 2008 16:10:43 +0000 (09:10 -0700)]
tree-ssa-forwprop.c (forward_propagate_addr_expr_1): If we have the same size types for...
2008-05-05 Andrew Pinski <Andrew.Pinski@playstation.sony.com>
* tree-ssa-forwprop.c (forward_propagate_addr_expr_1): If we have the
same size types for the indirect reference on the rhs, then create a VCE.
2008-05-05 Andrew Pinski <andrew.pinski@playstation.sony.com>
* gcc.dg/tree-ssa/forwprop-5.c: New testcase.
* gcc.dg/tree-ssa/forwprop-6.c: New testcase.
* gcc.dg/tree-ssa/forwprop-7.c: New testcase.
* gcc.dg/tree-ssa/forwprop-8.c: New testcase.
* gcc.dg/tree-ssa/forwprop-9.c: New testcase.
Uros Bizjak [Mon, 5 May 2008 12:56:15 +0000 (14:56 +0200)]
* config/i386/i386.md
(*float<SSEMODEI24:mode><MODEF:mode>2_sse_interunit): Use only
one insn template instead of template series.
(*xordi_1_rex64): Ditto.
(*xordi_2_rex64): Ditto.
David S. Miller [Sun, 4 May 2008 11:25:04 +0000 (11:25 +0000)]
config.gcc (sparc*-*-*): Always set need_64bit_hwint to yes.
gcc/
* config.gcc (sparc*-*-*): Always set need_64bit_hwint to yes.
(sparc*-*-linux*): Use linux.h in tm_file.
(sparc-*-linux*): If 'enabled_targets' is 'all', build a bi-arch
compiler defaulting to 32-bit.
(sparc*-*-*): Remove explicit target settings of need_64bit_hwint,
no longer needed.
* config/sparc/linux.h: Remove definitions now obtained
properly from linux.h
* config/sparc/linux64.h: Likewise.
(ASM_CPU_DEFAULT_SPEC): Change this to ASM_CPU64_DEFAULT_SPEC, we
don't want this setting for 32-bit builds in a biarch compiler.
* doc/install.texi: Add sparc-linux to list of targets
supporting --enable-targets=all.
libcpp/
* configure.ac (sparc*-*-*): Always set need_64bit_hwint to yes.
* configure: Regenerate.
Thomas Koenig [Sun, 4 May 2008 08:06:02 +0000 (08:06 +0000)]
re PR fortran/35990 (run-time abort for PACK of run-time zero sized array)
2008-05-04 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/35990
* intrinsics/pack_generic.c: If an extent of the source
array is less then zero, set it to zero. Set the source
pointer to NULL if the source size is zero. Set the total
number of elements to zero if the vector has an extent
less or equal to zero.
* m4/pack.m4: Set the source pointer to NULL if the
source array is zero-sized. Set the total number of
elemements to zero if the vector has an extent less or
equal to zero.
* generated/pack_i1.c: Regenerated.
* generated/pack_i2.c: Regenerated.
* generated/pack_i4.c: Regenerated.
* generated/pack_i8.c: Regenerated.
* generated/pack_i16.c: Regenerated.
* generated/pack_r4.c: Regenerated.
* generated/pack_r8.c: Regenerated.
* generated/pack_r10.c: Regenerated.
* generated/pack_r16.c: Regenerated.
* generated/pack_c4.c: Regenerated.
* generated/pack_c8.c: Regenerated.
* generated/pack_c10.c: Regenerated.
* generated/pack_c16.c: Regenerated.
2008-05-04 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/35990
* gfortran.dg/intrinsic_pack_4.f90: New test case.
Eric Botcazou [Sat, 3 May 2008 19:35:01 +0000 (19:35 +0000)]
decl.c (maybe_pad_type): Try to get a form of the type with integral mode even if...
* decl.c (maybe_pad_type): Try to get a form of the type with integral
mode even if the alignment is not a factor of the original size. But
make sure to create the inner field with the original size. Reorder.
* trans.c (addressable_p) <COMPONENT_REF>: Treat the field of a padding
record as always addressable.
* utils.c (convert): Deal specially with conversions between original
and packable versions of a record type.
* utils2.c (build_binary_op) <MODIFY_EXPR>: Be more restrictive when
recognizing an assignment between padded objects.
H.J. Lu [Sat, 3 May 2008 16:17:24 +0000 (16:17 +0000)]
i386.c (ix86_builtin_type): Move V4SI_FTYPE_V4SF after V4SI_FTYPE_V8HI.
2008-05-03 H.J. Lu <hongjiu.lu@intel.com>
* config/i386/i386.c (ix86_builtin_type): Move V4SI_FTYPE_V4SF
after V4SI_FTYPE_V8HI.
(ix86_init_mmx_sse_builtins): Move case V4HI_FTYPE_V4HI after
case V4SI_FTYPE_V2DF.
Jerry DeLisle [Sat, 3 May 2008 15:11:33 +0000 (15:11 +0000)]
re PR fortran/33268 (read ('(f3.3)'), a rejected due to the extra (...))
2008-05-03 Jerry DeLisle <jvdelisle@gcc.gnu.org>
PR fortran/33268
* gfortran.h: Add extra_comma pointer to gfc_dt structure. Add iokind to
gfc_expr value union. Add io_kind enum to here from io.c.
* io.c (gfc_free_dt): Free extra_comma.
(gfc_resolve_dt): If an extra comma was encountered and io_unit is type
BT_CHARACTER, resolve to format_expr and set default unit. Error if
io_kind is M_WRITE. (match_io): Match the extra comma and set new
pointer, extra_comma.
Jerry DeLisle [Fri, 2 May 2008 01:33:16 +0000 (01:33 +0000)]
re PR libfortran/36094 (Runtime error show_locus not working correctly)
2008-05-01 Jerry DeLisle <jvdelisle@gcc.gnu.org>
PR libfortran/36094
* runtime/error.c (show_locus): Provide modified error message when
filename has not yet been associated with a unit number.
* io/open.c (encoding_opt[]): Comment out "utf-8" option and add TODO.
Eric Botcazou [Thu, 1 May 2008 10:50:01 +0000 (10:50 +0000)]
decl.c (make_packable_type): Resize the last component to its RM size only if it is of an aggregate type.
* decl.c (make_packable_type): Resize the last component to its RM size
only if it is of an aggregate type.
* trans.c (call_to_gnu): Fix nit in comment.
(gnat_to_gnu): Likewise.
Richard Guenther [Wed, 30 Apr 2008 19:05:12 +0000 (19:05 +0000)]
re PR tree-optimization/21636 (Missed ccp optimization)
2008-04-30 Richard Guenther <rguenther@suse.de>
PR tree-optimization/21636
* tree-ssa-ccp.c (ccp_fold): Handle &p->x with p being a
constant address.
(evaluate_stmt): Print the likely value.
(ccp_visit_stmt): Avoid excessive vertical spacing.
Seongbae Park [Wed, 30 Apr 2008 17:17:03 +0000 (17:17 +0000)]
gcc.c (wrapper_string): New variable.
2008-04-30 Seongbae Park <seongbae.park@gmail.com>
* gcc.c (wrapper_string): New variable.
(insert_wrapper): New function.
(execute): New option -wrapper.
* doc/invoke.texi (Overall Options): New driver option -wrapper.