PR tree-optimization/33860
* tree-vect-transform.c (vect_analyze_data_ref_access): Don't allow
interleaved accesses in case the dr is inside the inner-loop during
outer-loop vectorization.
From-SVN: r129587
* config/rs6000/rs6000.c (rs6000_override_options): Set
rs6000_altivec_abi on AIX.
(conditional_register_usage): Set VR20:31 fixed on AIX.
From-SVN: r129582
gcc/:
2007-10-22 Seongbae Park <seongbae.park@gmail.com>
David S. Miller <davem@davemloft.net>
* df-scan.c (df_get_call_refs): Mark global registers as both a
DF_REF_REG_USE and a non-clobber DF_REF_REG_DEF.
gcc/testsuite/:
2007-10-22 David S. Miller <davem@davemloft.net>
* gcc.dg/globalreg-1.c: New test.
From-SVN: r129572
PR tree-optimization/33834
PR tree-optimization/33835
* tree-vect-analyze.c (vect_analyze_operations): RELEVANT and LIVE stmts
need to be checked for success seperately.
* tree-vect-transform.c (vectorizable_call, vectorizable_conversion):
Remove the check that stmt is not LIVE.
(vectorizable_assignment, vectorizable_induction): Likewise.
(vectorizable_operation, vectorizable_type_demotion): Likewise.
(vectorizable_type_promotion, vectorizable_load, vectorizable_store):
Likewise.
(vectorizable_live_operation): Check that op is not NULL.
From-SVN: r129571
* doc/invoke.texi (Optimization Options): In the summary, show
that the value for options -falign-jumps, -falign-labels,
-falign-loops, -falign-functions, -fsched-stalled-insns, and
-fsched-stalled-insns-dep is optional. In the summary, show that
-fsched-stalled-insns takes an optional argument and that
-fparallelize-loops takes an argument. Add opindex for several
optimization options, and fix typos in the opindex commands for
several others. In the descriptions for -fsched-stalled-insns
and -fsched-stalled-insns-dep, show the options with and without
values.
From-SVN: r129565
2007-10-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/31244
* gfortran.h (gfc_data_value): Change repeat from unsigned int
to mpz_t.
* decl.c(top_val_list): Remove msg variable. Use mpz_t for
repeat count.
* resolve.c (values): Change left from unsigned int to mpz_t.
(next_data_value): Change for mpz_t.
(check_data_variable): Change ??? to FIXME in a comment. Use
"mpz_t left".
(resolve_data ): Use "mpz_t left".
From-SVN: r129561
PR rtl-optimization/33644
* cfgcleanup.c: Do not include dce.h.
* cfgrtl.c (delete_insn_chain_and_edges): Resurrect.
* combine.c (distribute_notes): Delete REG_LIBCALL_ID case.
* dce.c (something_changed): Delete.
(libcall_dead_p): New predicate.
(delete_unmarked_insns): Use it to delete dead libcalls.
Deal with REG_LIBCALL and REG_RETVAL notes.
(prescan_libcall_for_dce): New function.
(prescan_insns_for_dce): Use it to deal with libcalls.
(mark_reg_dependencies): Do nothing special for libcalls.
(dce_process_block): Likewise.
(fast_dce): Delete unused local variable.
(run_fast_dce): Do not return a value.
* dce.h (struct df): Delete.
(run_fast_dce): Adjust prototype.
* optabs.c (libcall_id): Delete.
(maybe_encapsulate_block): Do not emit REG_LIBCALL_ID notes.
(emit_no_conflict_block): Do not look for REG_LIBCALL_ID notes.
* reload1.c (reload): Delete REG_LIBCALL_ID case.
* rtl.h (delete_insn_chain_and_edges): Resurrect prototype.
* see.c (see_update_relevancy): Look for REG_LIBCALL and REG_RETVAL
notes instead of REG_LIBCALL_ID notes.
* reg-notes.def (LIBCALL_ID): Delete.
* Makefile.in (see.o): Add dce.h dependency.
(cfgcleanup.o): Remove dce.h dependency.
From-SVN: r129556
PR c++/33372
* semantics.c (finish_omp_clauses): Check !type_dependent_expression_p
before checking if its type is integral.
* g++.dg/gomp/pr33372-1.C: New test.
* g++.dg/gomp/pr33372-2.C: New test.
* g++.dg/gomp/pr33372-3.C: New test.
From-SVN: r129555
PR tree-optimization/33854
* tree-vect-analyze.c (vect_determine_vectorization_factor): Add
FLOAT_EXPR to the list of promotion operations.
From-SVN: r129549
2007-10-21 Paul Thomas <pault@gcc.gnu.org>
PR fortran/33749
* resolve.c (resolve_ordinary_assign): New function that takes
the code to resolve an assignment from resolve_code. In
addition, it makes a temporary of any vector index, on the
lhs, using gfc_get_parentheses.
(resolve_code): On EXEC_ASSIGN call the new function.
2007-10-21 Paul Thomas <pault@gcc.gnu.org>
PR fortran/33749
* gfortran.dg/assign_9.f90: New test.
From-SVN: r129539
gcc/
* config/mips/mips-protos.h (mips_regno_mode_ok_for_base_p): Give
the STRICT_P argument type "bool" rather than "int".
(mips_legitimate_address_p): Likewise.
(fp_register_operand, lo_operand): Delete.
(mips_subword): Give the HIGH_P argument type "bool" rather than "int".
(mips_emit_scc): Rename to...
(mips_expand_scc): ...this.
(gen_conditional_branch): Rename to...
(mips_expand_conditional_branch): ...this.
(gen_conditional_move): Rename to...
(mips_expand_conditional_move): ...this.
(mips_gen_conditional_trap): Rename to...
(mips_expand_conditional_trap): ...this and take an rtx code instead
of an operands array.
(mips_expand_call): Give the SIBCALL_P argument type "bool"
rather than "int".
(mips_emit_fcc_reload): Rename to...
(mips_expand_fcc_reload): ...this.
(init_cumulative_args): Rename to...
(mips_init_cumulative_args): ...this and remove the libname argument.
(function_arg_advance): Rename to...
(mips_function_arg_advance): ...this.
(function_arg): Rename to...
(mips_function_arg): ...this and use "rtx" instead of
"struct rtx_def *".
(function_arg_boundary): Rename to...
(mips_function_arg_boundary): ...this.
(mips_expand_unaligned_load): Rename to...
(mips_expand_ext_as_unaligned_load): ...this and give the WIDTH
and BITPOS arguments type "HOST_WIDE_INT".
(mips_expand_unaligned_store): Rename to...
(mips_expand_ins_as_unaligned_store): ...this and give the WIDTH
and BITPOS arguments type "HOST_WIDE_INT".
(override_options): Rename to...
(mips_override_options): ...this.
(print_operand): Rename to...
(mips_print_operand): ...this.
(print_operand_address): Rename to...
(mips_print_operand_address): ...this.
(mips_output_ascii): Remove the PREFIX argument.
(mips_expand_epilogue): Give the SIBCALL_P argument type "bool"
rather than "int".
(mips_can_use_return_insn): Return a "bool" rather than an "int".
(mips_function_value): Remove the FUNC argument. Use "rtx" instead
of "struct rtx_def *".
(mips_secondary_reload_class): Give the IN_P argument type "bool"
rather than "int".
(build_mips16_call_stub): Delete.
(mips_store_data_bypass_p): Return a "bool" rather than an "int".
(mips_use_ins_ext_p): Give the second and third arguments type
"HOST_WIDE_INT" rather than "rtx".
* config/mips/mips.h: Update after the above changes. Tweak comments
and formatting.
(SECONDARY_INPUT_RELOAD_CLASS): Pass a boolean IN_P argument.
(SECONDARY_OUTPUT_RELOAD_CLASS): Likewise.
(mips_print_operand_punct, mips_hard_regno_mode_ok): Change the
element type from "char" to "bool".
(mips_section_threshold, sym_lineno, set_noat, mips_branch_likely)
(mips_cpu_info_table): Delete.
* config/mips/mips.md: Update after the above changes. Tweak some
formatting.
* config/mips/mips-ps-3d.md: Likewise.
* config/mips/mips.c: Tweak comments and formatting.
Use NULL-like macros rather than 0 in various places.
(mips_section_threshold): Rename to...
(mips_small_data_threshold): ...this and give it an unsigned type.
(num_source_filenames): Remove explicit initialization.
(sdb_label_count): Likewise.
(sym_lineno): Delete.
(set_noat): Make static.
(mips_branch_likely): Likewise, and give it type "bool".
(mips_isa_info): Rename to...
(mips_isa_option_info): ...this.
(mips_flag_delayed_branch): Rename to...
(mips_base_delayed_branch): ...this.
(mips_split_addresses): Delete.
(mips_print_operand_punct, mips_hard_regno_mode_ok): Change the
element type from "char" to "bool".
(mips_regno_to_class): Add an explicit array size.
(mips16_flipper): Change from type "int" to type "bool".
(mips_comp_type_attributes): Remove redundant FUNCTION_TYPE check.
(mips_tls_operand_p): Rename to...
(mips_tls_symbol_p): ...this.
(mips_global_symbol_p): Remove redundant "const".
(mips_offset_within_alignment_p): Simplify.
(mips_regno_mode_ok_for_base_p): Change the name of the last argument
to STRICT_P and give it type bool.
(mips_valid_base_register_p): Likewise.
(mips_classify_address): Likewise.
(mips_legitimate_address_p): Likewise.
(mips16_unextended_reference_p): Take the offset as an
"unsigned HOST_WIDE_INT" rather than an "rtx". Simplify.
(mips_const_insns): Use IN_RANGE.
(mips_force_temporary): Remove an unnecessary copy_rtx.
(mips16_gp_pseudo_reg): Simplify.
(mips_split_symbol): Remove an unnecessary copy_rtx.
(mips_legitimize_tls_address): Only create registers if we need them.
(mips_legitimize_address): Use mips_split_plus.
(mips_move_integer): Swap DEST and TEMP arguments. Rename local
variable COST to NUM_OPS.
(mips_legitimize_move): Use gen_lowpart.
(mips_rewrite_small_data): Rename the OP parameter to PATTERN.
(m16_check_op): Use IN_RANGE.
(mips_subword): Give the HIGH_P argument type "bool" rather than "int".
(mips_split_doubleword_move): Pass booleans to mips_subword.
(mips_output_move): Store the mode in a local variable.
Explicitly assert that this function does not deal with
MFLO and MFHI. Use SMALL_OPERAND_UNSIGNED. Make order of
cases more consistent.
(mips_relational_operand_ok_p): Rename to...
(mips_int_order_operand_ok_p): ...this.
(mips_canonicalize_comparison): Rename to...
(mips_canonicalize_int_order_test): ...this.
(mips_emit_int_relational): Rename to...
(mips_emit_int_order_test): ...this. Store the mode in a local
variable.
(mips_reverse_fp_cond_p): Rename to...
(mips_reversed_fp_cond): ...this.
(mips_emit_compare): Make the order of the statements more consistent.
(mips_emit_scc): Rename to...
(mips_expand_scc): ...this.
(gen_conditional_branch): Rename to...
(mips_expand_conditional_branch): ...this.
(gen_conditional_move): Rename to...
(mips_expand_conditional_move): ...this. Build the condition
separately from the main pattern.
(mips_gen_conditional_trap): Rename to...
(mips_expand_conditional_trap): ...this and take an rtx code instead
of an operands array. Simplify.
(init_cumulative_args): Rename to...
(mips_init_cumulative_args): ...this. Use memset rather than a
copy from a zeroed static. Use prototype_p and stdarg_p.
(mips_arg_info): Rename to...
(mips_get_arg_info): ...this and put the INFO argument first.
(function_arg): Rename to...
(mips_function_arg): ...this and use "rtx" instead of
"struct rtx_def *". Only split structures into chunks if
TARGET_HARD_FLOAT, rather than checking !TARGET_SOFT_FLOAT on
a chunk-by-chunk basis. Use SCALAR_FLOAT_TYPE_P instead of a
comparison with REAL_TYPE. Rename local variable REG to REGNO
and make it unsigned.
(function_arg_advance): Rename to...
(mips_function_arg_advance): ...this.
(function_arg_boundary): Rename to...
(mips_function_arg_boundary): ...this.
(mips_pad_arg_upward): Use SCALAR_INT_MODE_P instead of a comparison
with MODE_INT.
(mips_fpr_return_fields): Use SCALAR_FLOAT_TYPE_P instead of a
comparison with REAL_TYPE.
(mips_function_value): Remove the FUNC argument. Rename local
variable UNSIGNEDP to UNSIGNED_P.
(mips_return_in_memory): Use IN_RANGE.
(mips_setup_incoming_varargs): Pass a boolean rather than an int
to FUNCTION_ARG_ADVANCE.
(mips_gimplify_va_arg_expr): Rename local variable INDIRECT
to INDIRECT_P. Remove a redundant mips_abi check. Only calculate
SIZE and RSIZE if needed.
(mips16_stub): Rename FPRET to FP_RET_P and give it type "bool"
rather than "int".
(build_mips16_function_stub): Rename to...
(mips16_build_function_stub): ...this.
(build_mips16_call_stub): Rename to...
(mips16_build_call_stub): ...this. Make static, and return a
"bool" rather than an "int". Rename the ARG_SIZE argument to
ARGS_SIZE. Rename the local variable FPRET to FP_RET_P and give
it type "bool". Use "jr" rather than "j". Use XNEW. Simplify.
(mips_load_call_address): Give the SIBCALL_P argument type "bool"
rather than "int".
(mips_expand_call): Likewise. Assert that MIPS16 calls via stubs
are not sibling ones.
(mips_emit_fcc_reload): Rename to...
(mips_expand_fcc_reload): ...this.
(mips_get_unaligned_mem): Give the WIDTH and BITPOS arguments type
"HOST_WIDE_INT".
(mips_expand_unaligned_load): Rename to...
(mips_expand_ext_as_unaligned_load): ...this and give the WIDTH
and BITPOS arguments type "HOST_WIDE_INT". Remove redundant
SUBREG_BYTE check.
(mips_expand_unaligned_store): Rename to...
(mips_expand_ins_as_unaligned_store): ...this and give the WIDTH
and BITPOS arguments type "HOST_WIDE_INT".
(mips_use_ins_ext_p): Rename second and third arguments to WIDTH
and BITPOS and give them type "HOST_WIDE_INT" rather than "rtx".
Use IN_RANGE.
(mips_init_split_addresses): Replace with...
(mips_split_addresses_p): ...this new function.
(mips_init_relocs): Check mips_split_addresses_p () rather than
mips_split_addresses. Always initialize mips_split_p with booleans.
(print_operand_reloc): Rename to...
(mips_print_operand_reloc): ...this. Use a gcc_assert rather than
a fatal_error.
(print_operand): Rename to...
(mips_print_operand): ...this.
(print_operand_address): Rename to...
(mips_print_operand_address): ...this. Use GP_REG_FIRST.
(mips_in_small_data_p): Simplify.
(mips_output_ascii): Remove the PREFIX argument and "register"
keywords. Use the STRING parameter directly and cast to
"unsigned char".
(mips16e_collect_argument_save_p): Use IN_RANGE.
(mips_function_has_gp_insn): Simplify.
(mips_set_return_address): Use BITSET_P.
(mips_restore_gp): Use a separate statement to pick the base
register.
(mips_output_function_prologue): Use a local FRAME variable.
(mips_save_reg): Pass booleans to mips_subword.
(mips_expand_prologue): Use a local FRAME variable. Use
cprestore_size to detect when .cprestore is needed.
(mips_restore_reg): Use GP_REG_FIRST.
(mips_expand_epilogue): Give the SIBCALL_P argument type "bool"
rather than "int". Use local FRAME and REGNO variables.
(mips_can_use_return_insn): Return a "bool" rather than an "int".
(mips_secondary_reload_class): Give the IN_P argument type "bool"
rather than "int".
(mips_linked_madd_p): Rename arguments to OUT_INSN and IN_INSN.
(mips_store_data_bypass_p): Return a "bool" rather than an "int".
(vr4130_swap_insns_p): Rename local variables to DEP1_P and DEP2_P.
(mips_74k_agen_init): Assume INSN satisfies USEFUL_INSN_P.
(mips_variable_issue): Do nothing unless INSN satisfies USEFUL_INSN_P.
(builtin_description): Rename to...
(mips_builtin_description): ...this.
(mips_bdesc): Rename to...
(mips_ps_bdesc): ...this.
(sb1_bdesc): Rename to...
(mips_sb1_bdesc): ...this.
(dsp_bdesc): Rename to...
(mips_dsp_bdesc): ...this.
(dsp_32only_bdesc): Rename to...
(mips_dsp_32only_bdesc): ...this.
(bdesc_map): Rename to...
(mips_bdesc_map): ...this.
(bdesc_arrays): Rename to...
(mips_bdesc_arrays): ...this.
(mips_prepare_builtin_arg): Rename OP to OPNO and ARGNUM to ARGNO.
Tweak error message.
(mips_expand_builtin_direct): Rename HAS_TARGET to HAS_TARGET_P.
Rename local variables I and J to OPNO and ARGNO respectively.
Assert that the number of arguments is no more than the number of
input operands.
(mips_expand_builtin_compare): Use an array of arguments rather
than an array of operands. Assert that the number of operands is
as expected.
(add_constant): Rename to...
(mips16_add_constant): ...this. Use XNEW.
(dump_constants_1): Rename to...
(mips16_emit_constants_1): ...this. Use ALL_SCALAR_FIXED_POINT_MODE_P.
(dump_constants): Rename to...
(mips16_emit_constants): ...this.
(mips_sim_wait_reg, mips_sim_record_set): Use END_REGNO.
(mips_set_mips16_mode): Don't call mips_init_split_addresses.
Assign mips16_p to was_mips16_p.
(mips_parse_cpu): Tweak warning message.
(override_options): Rename to...
(mips_override_options): ...this. Move the mips_cost initialization
out of the "shared with GAS" block. Quote option names in error
messages.
(mips_conditional_register_usage): Make local variable REGNO unsigned.
(mips_order_regs_for_local_alloc): Remove register keyword.
From-SVN: r129532
* config/mips/mips.c (mips_set_mips16_mode) Say sorry for hard-float
MIPS16 code unless using o32 or o64.
gcc/testsuite/
* gcc.target/mips/mips.exp (setup_mips_tests): Set mips_mips16.
(dg-mips-options): Skip tests that specify an ABI other than o32
and o64 if generating MIPS16 hard-float code.
From-SVN: r129530
gcc/
* config/mips/mips.c (mips_function_ok_for_sibcall): Only forbid
sibling calls to "mips16" functions if the function satisfies
const_call_insn_operand.
From-SVN: r129526
gcc/
* doc/invoke.texi (-mpaired-single): Don't say that the option
requires 64-bit code.
* config/mips/mips-protos.h (mips_modes_tieable_p): Declare.
* config/mips/mips.h (ISA_HAS_PAIRED_SINGLE): New macro.
(ISA_HAS_NMADD_NMSUB): Add a mode argument. Return true for
V2SF if ISA_MIPS32R2.
(MODES_TIEABLE_P): Use mips_modes_tieable_p.
* config/mips/mips.c (mips_rtx_costs): Pass a mode argument
to ISA_HAS_NMADD_NMSUB.
(mips_split_doubleword_move): Handle V2SF.
(mips_modes_tieable_p): New function.
(override_options): Report a warning rather than an error when
-mpaired-single is used on ISAs that don't support it; use
ISA_HAS_PAIRED_SINGLE to check that case.
* config/mips/mips.md (MOVE64): New mode iterator. Replace DI
and DF move splitters with a single MOVE64 splitter, thereby adding
a V2SF splitter too.
(SPLITF): Add TARGET_DOUBLE_FLOAT conditions to DI and DF.
Add a TARGET_FLOAT64 condition to TF. Add V2SF to the iterator.
(HALFMODE): Add V2SF.
(*nmadd<mode>, *nmadd<mode>_fastmath, *nmsub<mode>)
(*nmsub<mode>_fastmath): Add a mode argument to ISA_HAS_NMADD_NMSUB.
(movv2sf_hardfloat_64bit): Tweak ordering of conditions.
(movv2sf_hardfloat_32bit): New pattern.
(load_low<mode>, load_high<mode>, store_word<mode>): Remove
TARGET_DOUBLE_FLOAT conditions.
gcc/testsuite/
* gcc.dg/vect/vect.exp: Extend -mpaired-single handling to all
MIPS targets.
* g++.dg/vect/vect.exp: Likewise.
* lib/fortran-torture.exp: Likewise.
* gcc.target/mips/mips-ps-1.c: Use mpaired_single rather than
mipsisa64*-*-* as the target selector. Remove -mips64,
-mhard-float and -mgp64 from the options list.
* gcc.target/mips/mips-ps-2.c: Likewise.
* gcc.target/mips/mips-ps-3.c: Likewise.
* gcc.target/mips/mips-ps-4.c: Likewise.
* gcc.target/mips/mips-ps-6.c: Likewise.
* gcc.target/mips/mips-ps-5.c: Remove -mhard-float from the
options list.
* gcc.target/mips/sb1-1.c: Likewise.
* gcc.target/mips/mips-ps-type.c: Likewise.
* gcc.target/mips/mips-ps-7.c: New test.
* gcc.target/mips/mips-ps-type-2.c: Likewise.
* gcc.target/mips/fpr-moves-6.c: Remove XFAIL.
* gcc.target/mips/mips.exp (setup_mips_tests): Set mips_fp and
mips_gp instead of mips_fp64 and mips_gp64. Treat -mgp32 -mfp64
as forcing an ABI and an architecture.
(is_gp32_flag, is_gp64_flag): Fold into...
(dg-mips-options): ...here. Make -mpaired-single imply -mfp64,
then -mfp64 imply -mhard-float. Apply register rules after the
loop. Handle -march=mipsN like -mipsN.
From-SVN: r129522
* config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Point
saved CR2 offset to low 32 bits of regs->ccr rather than the whole
64-bit register in 64-bit libgcc.
From-SVN: r129520
2007-10-20 Tobias Burnus <burnus@net-b.de>
PR fortran/33818
* resolve.c (resolve_variable): Check that symbol is in the same
namespace as the entry function.
2007-10-20 Tobias Burnus <burnus@net-b.de>
PR fortran/33818
* gfortran.dg/entry_dummy_ref_3.f90: New.
From-SVN: r129510
