Paolo Carlini [Fri, 27 Nov 2009 10:44:49 +0000 (10:44 +0000)]
re PR c++/35112 (ICE and broken diagnostic with ambiguous class name)
/cp
2009-11-27 Paolo Carlini <paolo.carlini@oracle.com>
PR c++/35112
* pt.c (print_overloaded_functions): New.
(print_candidates): Call the latter.
* parser.c (cp_parser_class_name): Do not duplicate the diagnostics
after the cp_parser_lookup_name call.
/testsuite
2009-11-27 Paolo Carlini <paolo.carlini@oracle.com>
PR c++/35112
* g++.dg/parse/crash53.C: New.
* g++.dg/parse/error17.C: Adjust, error messages are not
duplicated anymore.
* g++.dg/template/dtor7.C: Remove xfail-ed dg-bogus directives.
sse.md (mulv16qi3): Change back from an expander to an insn-and-split.
* config/i386/sse.md (mulv16qi3): Change back from an expander
to an insn-and-split.
(mulv4si): Mention AVX not XOP for AVX exception.
(*sse2_mulv4si3): Likewise.
(mulv2di3): Use vmulld not vpmacsdd for XOP expansion. Tidy.
testsuite/
* gcc.target/i386/xop-imul64-vector.c: Look for vpmulld not vpmacsdd.
Bernd Schmidt [Thu, 26 Nov 2009 21:46:45 +0000 (21:46 +0000)]
cfgexpand (n_stack_vars_conflict): New static variable.
* cfgexpand (n_stack_vars_conflict): New static variable.
(triangular_index): Reduce likelyhood of overflow by performing
the division before the multiplication.
(resize_stack_vars_conflict): Call fatal_error if overflow
occurred in triangular_index. Update n_stack_vars_conflict.
Bernd Schmidt [Thu, 26 Nov 2009 21:41:42 +0000 (21:41 +0000)]
re PR rtl-optimization/38582 (excessive time in rename registers)
PR rtl-opt/38582
* regrename.c (struct du_head): New members id, conflicts,
hard_conflicts and cannot_rename.
(enum scan_actions): Remove terminate_all_read and
terminate_overlapping_read; add mark_all_read.
(scan_actions_name): Likewise.
(du_head_p): New typedef. Define a vector type for it.
(id_to_chain): New static variable.
(note_sets, clear_dead_regs): Delete functions.
(free_chain_data): New function.
(merge_overlapping_regs): Simply walk the conflicts bitmap.
Remove argument B, all callers changed.
(regrename_optimize): Allocate id_to_chain. Ignore chains that have
the cannot_rename bit set. Update regno and nregs of a renamed chain.
Call free_chain_data when done.
(do_replace): Remove death notes when the renamed reg is set in the
last insn; add them if not.
(mark_conflict, note_sets_clobbers): New static function.
(fail_current_block, current_id, open_chains_set, live_hard_regs): New
static variables.
(scan_rtx_reg): Keep track of conflicts between chains, and between
chains and hard regs. Don't terminate chains when we find a read we
can't handle, mark it unrenameable instead. For terminate_write,
terminate chains that are written with an exact match or a superset
of registers. Set fail_current_block if multi-word lifetimes are too
complex to handle.
(scan_rtx_address): Use mark_all_read instead of terminate_all_read.
(build_def_use): Initialize current_id, live_chains and live_hard_regs;
free memory for them when done.
Rearrange the steps so that earlyclobbers are noted before reads
are processed. Add new steps to keep track of hard register lifetimes
outside insn operands.
Bernd Schmidt [Thu, 26 Nov 2009 21:35:35 +0000 (21:35 +0000)]
re PR rtl-optimization/38582 (excessive time in rename registers)
PR rtl-opt/38582
* regrename.c (struct du_chain): Remove member earlyclobber.
(scan_rtx_reg): Don't set it. Remove argument earlyclobber,
all callers changed.
(scan_rtx): Remove argument earlyclobber, all callers changed.
(hide_operands, restore_operands, record_out_operands): New functions,
broken out of build_def_use.
(build_def_use): Call them as necessary.
Janus Weil [Thu, 26 Nov 2009 19:01:02 +0000 (20:01 +0100)]
re PR fortran/42048 ([F03] Erroneous syntax error message on TBP call)
2009-11-26 Janus Weil <janus@gcc.gnu.org>
PR fortran/42048
PR fortran/42167
* gfortran.h (gfc_is_function_return_value): New prototype.
* match.c (gfc_match_call): Use new function
'gfc_is_function_return_value'.
* primary.c (gfc_is_function_return_value): New function to check if a
symbol is the return value of an encompassing function.
(match_actual_arg,gfc_match_rvalue,match_variable): Use new function
'gfc_is_function_return_value'.
* resolve.c (resolve_common_blocks,resolve_actual_arglist): Ditto.
2009-11-26 Janus Weil <janus@gcc.gnu.org>
PR fortran/42048
PR fortran/42167
* gfortran.dg/select_type_10.f03: New test case.
* gfortran.dg/typebound_call_11.f03: Extended test case.
Jason Merrill [Thu, 26 Nov 2009 15:59:52 +0000 (10:59 -0500)]
re PR c++/10690 ([DR 115] Even when used within typeid(), a template-id generating an overload set with only one function should silently decay to a pointer to function)
Nick Clifton [Thu, 26 Nov 2009 10:44:28 +0000 (10:44 +0000)]
rx.c (rx_expand_epilogue): Add checks for sibcalls being used incorrectly.
* config/rx/rx.c (rx_expand_epilogue): Add checks for sibcalls
being used incorrectly.
(rx_function_ok_for_sibcall): New function. Do not allow indirect
sibcalls, or sibcalls from interrupt functions.
(TARGET_FUNCTION_OK_FOR_SIBCALL): Define.
* config/rx/rx.md (sibcall): Convert to a define_expand. Check
for a MEM inside a MEM.
(sibcall_value): Likewise.
(sibcall_internal): New pattern containing old sibcall pattern.
(sibcall_value_internal): Likewise.
Ben Elliston [Thu, 26 Nov 2009 10:06:35 +0000 (10:06 +0000)]
posix-threads.cc (ParkHelper::unpark): Do not initialise result, but assign it instead.
* posix-threads.cc (ParkHelper::unpark): Do not initialise result,
but assign it instead. Eliminates an unused variable warning when
the result == 0 assertion is disabled.
* target.h (targetm.vectorize.builtin_vec_perm_ok): New.
* target-def.h (TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK): New.
* hooks.h, hooks.c (hook_bool_tree_tree_true): New.
* tree-vect-slp.c (vect_create_mask_and_perm): Don't create
the vector constant here...
(vect_transform_slp_perm_load): ... do it here instead. Validate
that the permutation vector is implementable by the target.
Jakub Jelinek [Wed, 25 Nov 2009 20:20:40 +0000 (21:20 +0100)]
re PR target/42165 (-masm=intel doesn't know how to print AVX instructions)
PR target/42165
* config/i386/i386.c (print_operand): For 32-byte memory use
YMMWORD in -masm=intel mode. Use TBYTE instead of XWORD.
* config/i386/i386.md (crc32modesuffix): Expand to nothing
in -masm=intel mode.
(sse4_2_crc32di): Print just crc32 instead of crc32q in
-masm=intel mode.
* config/i386/mmx.md (*mmx_pinsrw): Print correct size of
memory operand in -masm=intel mode.
* config/i386/sse.md (*avx_pinsr<ssevecsize>, *sse4_1_pinsrb,
*sse2_pinsrw): Likewise.
(sse_cvtss2siq, sse_cvtss2siq_2, sse_cvttss2siq): Don't print
q suffix in -masm=intel mode.
Eric Botcazou [Wed, 25 Nov 2009 19:55:11 +0000 (19:55 +0000)]
re PR target/10127 (-fstack-check let's program crash)
PR target/10127
PR ada/20548
PR middle-end/42004
* dwarf2out.c (dwarf2out_args_size_adjust): Rename to...
(dwarf2out_stack_adjust): ...this. Do not adjust the arg size for
ACCUMULATE_OUTGOING_ARGS targets. Rename former version to...
(dwarf2out_notice_stack_adjust): ...this. Adjust for above renaming.
(dwarf2out_frame_debug_expr): Revert previous change and adjust for
above renaming.
(dwarf2out_frame_debug): Add ??? comment. Adjust for above renaming.
Robert Nelson [Wed, 25 Nov 2009 11:23:07 +0000 (11:23 +0000)]
arm.h (enum vfp_reg_type): Add VFP_NONE.
2009-11-25 Robert Nelson <robert.nelson@digikey.com>
Richard Earnshaw <rearnsha@arm.com>
* arm.h (enum vfp_reg_type): Add VFP_NONE.
* arm.c (all_fpus): Use it for floating point units that aren't based
on the VFP architecture.
(aapcs_vfp_is_call_or_return_candidate): Make base_mode a pointer to
enum machine_mode.
Co-Authored-By: Richard Earnshaw <rearnsha@arm.com>
From-SVN: r154647
Sebastian Pop [Wed, 25 Nov 2009 05:30:35 +0000 (05:30 +0000)]
graphite-blocking.c (pbb_strip_mine_profitable_p): The parameter depth is the loop nesting depth in the LST.
2009-11-18 Sebastian Pop <sebastian.pop@amd.com>
* graphite-blocking.c (pbb_strip_mine_profitable_p): The parameter
depth is the loop nesting depth in the LST. Call psct_dynamic_dim
to transform depth into the time_depth before calling
pbb_number_of_iterations_at_time.
(lst_do_strip_mine): Use a scalar variable instead of recomputing
lst_depth.
2009-11-04 Konrad Trifunovic <konrad.trifunovic@inria.fr>
* graphite-interchange.c (build_partial_difference): New.
(memory_stride_in_loop): Refactored the computation of
equality constraints into build_partial_difference function.
Added support for debugging stride computations.
Sebastian Pop [Wed, 25 Nov 2009 05:28:05 +0000 (05:28 +0000)]
graphite-interchange.c (lst_perfect_nestify): Pass 3 parameters for the loops created by the loop distribution.
2009-11-03 Sebastian Pop <sebastian.pop@amd.com>
* graphite-interchange.c (lst_perfect_nestify): Pass 3 parameters
for the loops created by the loop distribution. Do not modify the
input LSTs.
(lst_try_interchange_loops): Same. Use a temporary LST for the
transformed schedule. Call lst_update_scattering before data
dependence analysis.
(lst_try_interchange): Pass an extra parameter INDEX.
(lst_do_interchange_1): New.
(lst_do_interchange): Call lst_do_interchange_1.
(scop_do_interchange): Call lst_update_scattering.
* graphite-poly.c (apply_poly_transforms): Do not call
lst_update_scattering.
* graphite-poly.h (lst_pred): New.
(lst_succ): New.
(lst_find_first_pbb): Return NULL when not found.
(lst_empty_p): New.
(lst_insert_in_sequence): Allow LST1 to be NULL.
(lst_replace): New.
(lst_substitute_3): New.
* gcc.dg/graphite/interchange-1.c: XFail.
* gcc.dg/graphite/interchange-8.c: XFail.
* gcc.dg/graphite/interchange-11.c: XFail.
Sebastian Pop [Wed, 25 Nov 2009 05:25:43 +0000 (05:25 +0000)]
graphite-clast-to-gimple.c (gloog): Do not call sese_reset_aux_in_loops.
2009-10-26 Sebastian Pop <sebastian.pop@amd.com>
* graphite-clast-to-gimple.c (gloog): Do not call
sese_reset_aux_in_loops.
* graphite-sese-to-poly.c (build_loop_iteration_domains): Pass an
extra argument for domains. Do not use loop->aux.
(build_scop_iteration_domain): Initialize and free domains, pass it
to build_loop_iteration_domains and extract the information from
domains. Do not use loop->aux.
* sese.c (sese_reset_aux_in_loops): Removed.
* sese.h (sese_reset_aux_in_loops): Removed.
Sebastian Pop [Wed, 25 Nov 2009 05:25:11 +0000 (05:25 +0000)]
graphite-clast-to-gimple.c (find_pbb_via_hash): Moved up.
2009-10-26 Sebastian Pop <sebastian.pop@amd.com>
* graphite-clast-to-gimple.c (find_pbb_via_hash): Moved up.
(dependency_in_loop_p): Same.
(translate_clast): Do not use loop->aux. Initialize
loop->can_be_parallel.
(mark_loops_parallel): Removed.
* graphite-clast-to-gimple.h (mark_loops_parallel): Removed.
* graphite.c (free_aux_in_new_loops): Removed.
(graphite_finalize): Do not call free_aux_in_new_loops.
(graphite_transform_loops): Do not call mark_loops_parallel.
Sebastian Pop [Wed, 25 Nov 2009 05:22:49 +0000 (05:22 +0000)]
graphite-sese-to-poly.c (create_zero_dim_array): Pass an extra argument base_name.
2009-10-22 Sebastian Pop <sebastian.pop@amd.com>
* graphite-sese-to-poly.c (create_zero_dim_array): Pass an extra
argument base_name.
(rewrite_close_phi_out_of_ssa): Update use of create_zero_dim_array.
(rewrite_phi_out_of_ssa): Same.
(rewrite_cross_bb_scalar_deps): Same.
(translate_scalar_reduction_to_array): Same.
* graphite-sese-to-poly.c (free_data_refs_aux): Fix formatting.
(pdr_add_alias_set): Same.
(build_poly_dr): Same.
(build_alias_set_optimal_p): Same.
(build_base_obj_set_for_drs): Same. Do not store the result of
graphds_dfs.
Sebastian Pop [Wed, 25 Nov 2009 05:18:35 +0000 (05:18 +0000)]
graphite-sese-to-poly.c (pdr_add_data_dimensions): Do not infer loop upper bounds for 1-element arrays at end of structures.
2009-10-20 Sebastian Pop <sebastian.pop@amd.com>
* graphite-sese-to-poly.c (pdr_add_data_dimensions): Do not infer
loop upper bounds for 1-element arrays at end of structures.
* tree-flow.h (array_at_struct_end_p): Declared.
* tree-ssa-loop-niter.c (array_at_struct_end_p): Not static anymore.
graphite-sese-to-poly.c (build_loop_iteration_domains): Generalize construction of parameter constraints from loop iteration analysis.
2009-10-20 Alexander Monakov <amonakov@ispras.ru>
* graphite-sese-to-poly.c (build_loop_iteration_domains): Generalize
construction of parameter constraints from loop iteration analysis.
* gcc.dg/graphite/run-id-2.c: New test.
* graphite-sese-to-poly.c (find_scop_parameters): Initialize
SCOP_CONTEXT.
(build_loop_iteration_domains): Extract bounds on parameters
based on the data size.
(build_scop_context): Do not initialize SCOP_CONTEXT;
add new constraints.
Sebastian Pop [Wed, 25 Nov 2009 05:10:43 +0000 (05:10 +0000)]
graphite-interchange.c (memory_stride_in_loop): The depth argument corresponds to a dynamic time dimension...
2009-10-15 Sebastian Pop <sebastian.pop@amd.com>
* graphite-interchange.c (memory_stride_in_loop): The depth
argument corresponds to a dynamic time dimension, so use
psct_dynamic_dim to refer to the transformed scattering dimension.
Sebastian Pop [Wed, 25 Nov 2009 05:09:36 +0000 (05:09 +0000)]
graphite-blocking.c (lst_do_strip_mine): Avoid strip mining the root of the LST.
2009-10-15 Sebastian Pop <sebastian.pop@amd.com>
* graphite-blocking.c (lst_do_strip_mine): Avoid strip mining the
root of the LST.
* graphite-interchange.c (lst_do_interchange): Avoid interchanging
the root of the LST.
* graphite-poly.c (scop_to_lst): Fix LST sequence in an outermost
fake loop.
(print_lst): Print the root of LST in a different format.
* graphite-poly.h (lst_depth): Adjust to include the root of the LST.
Sebastian Pop [Wed, 25 Nov 2009 05:08:09 +0000 (05:08 +0000)]
re PR middle-end/41193 (slow compilation with graphite / ice with graphite)
2009-10-15 Sebastian Pop <sebastian.pop@amd.com>
* graphite-sese-to-poly.c (insert_copyin): Use gsi_insert_seq_on_edge.
(rewrite_commutative_reductions_out_of_ssa): Call gsi_commit_edge_inserts,
and check the SSA representation.
* gfortran.dg/graphite/id-18.f90: New, reduced from PR41193.
git://gcc.gnu.org / gcc.git / log