2007-06-12 Seongbae Park <seongbae.park@gmail.com>
* opts.c (common_handle_option): Handle new option -fdbg-cnt-list.
* dbgcnt.c (dbg_cnt_set_limit_by_name): Return value
to indicate an error.
(dbg_cnt_process_single_pair, dbg_cnt_list_all_counters): New functions
(dbg_cnt_process_opt): Print an error on a bad argument.
* dbgcnt.h (dbg_cnt_list_all_counters): New function declaration.
* common.opt (-fdbg-cnt-list): New.
* doc/invoke.texi (-fdbg-cnt-list,-fdbg-cnt=): New.
From-SVN: r125657
2007-06-12 Andrew Pinski <andrew_pinski@playstation.sony.com>
PR middle-end/31579
* expr.c (expand_expr_addr_expr_1): Call expand_expr
for the offset with the modifier as EXPAND_INITIALIZER
if the modifier is EXPAND_INITIALIZER.
(expand_expr_real_1 <case INTEGER_CST>): Don't force to
a register if we had an overflow.
2007-06-12 Andrew Pinski <andrew_pinski@playstation.sony.com>
PR middle-end/31579
* g++.dg/torture/pr31579.C: New testcase.
From-SVN: r125655
* gcov.c: Comments updated.
(source_info): Add file_time field.
(source_index): New variable.
(mutiple_files): New variable.
(generate_results): New function extracted from process_file.
(process_file): Save and restore chain of functions, generate
results and free structures only if not merging results.
(release_structures): File names are now freed in create_file_names
(create_file_names): Free previous file names.
(find_source): File date is now read here and modifications in
source files is checked here.
(read_graph_file): Only reverse order of functions for the current
object file.
(make_gcov_file_name): Do not generate long names if input_name is
NULL.
(output_lines): If merging results do not display graph, data and
runs informations.
Checking source file modification is done in find_source.
* doc/gcov.texi: Append an s to sourcefile.
From-SVN: r125649
* config/bfin/bfin.md (UNSPEC_NOP): New constant.
(forced_nop): New pattern.
* config/bfin/bfin.c: Include "df.h".
(add_to_reg): Use df_regs_ever_live_p instead of regs_ever_live.
(bfin_discover_loop): Use df_get_live_in instead of
global_live_at_start.
(bfin_reorder_loops): Pass 0 to cfg_layout_initialize. Call
df_analyze when done.
(gen_one_bundle): Don't generate SEQUENCE insns, just put modes on
the insns. Use QImode for the final insn in a bundle. Call
df_insn_rescan on generated NOPs; use gen_forced_nop instead of
gen_nop.
(reorder_var_tracking_notes): New function.
(bfin_reorg): Pass no argument to split_all_insns. Don't call
update_life_info. Call df_analyze after scheduling and bundle
generation. Call reorder_var_tracking_notes if generating these notes.
Call df_finish_pass at the end.
From-SVN: r125648
2007-06-07 Dirk Mueller <dmueller@suse.de>
* trans-stmt.c (gfc_trans_call): fix gcc_assert to
a comparison, not an assignment.
* optabls.c (debug_optab_libfuncs): fix gcc_assert to
a comparison, not an assignment.
From-SVN: r125647
* fold-const (fold_binary) [RDIV_EXPR]: Also optimize a/cbrt(b/c)
into a*cbrt(c/b) if flag_unsafe_math_optimizations is set.
testuite/ChangeLog:
* gcc.dg/builtins-11.c: Also check folding of a/cbrt(b/c).
From-SVN: r125641
2007-06-09 Daniel Berlin <dberlin@dberlin.org>
* tree-ssa-structalias.c (set_uids_in_ptset): Add is_deref'd
parameter, use it.
(find_what_p_points_to): Pass new parameter to set_uids_in_ptset.
From-SVN: r125603
2007-06-09 Daniel Berlin <dberlin@dberlin.org>
* tree-data-ref.c (dr_may_alias_p): Check that decl_a != decl_b,
and allow DECL_P here.
From-SVN: r125601
* tree-scalar-evolution.c (follow_ssa_edge_in_rhs,
follow_ssa_edge_in_condition_phi, follow_ssa_edge): Keep more precise
track of the size of the expression.
* cfghooks.c (merge_blocks): Remove block from loops structure only
after call of the merge_blocks hook.
* gcc.dg/tree-ssa/loop-28.c: New testcase.
From-SVN: r125595
./: PR tree-optimization/32169
* tree-vrp.c (extract_range_from_unary_expr): For NOP_EXPR and
CONVERT_EXPR, check whether min and max both converted to an
overflow infinity representation.
testsuite/:
PR tree-optimization/32169
* gcc.c-torture/compile/pr32169.c: New test.
From-SVN: r125591
gcc/ChangeLog:
2007-06-08 Harsha Jagasia <harsha.jagasia@amd.com>
Tony Linthicum <tony.linthicum@amd.com>
* doc/extend.texi: Add fvect-cost-model flag.
* common.opt (fvect-cost-model): New flag.
* tree-vectorizer.c (new_stmt_vec_info): Initialize inside and outside
cost fields in stmt_vec_info struct for STMT.
* tree-vectorizer.h (stmt_vec_info): Define inside and outside cost
fields in stmt_vec_info struct and access functions for the same.
(TARG_COND_BRANCH_COST): Define cost of conditional branch.
(TARG_VEC_STMT_COST): Define cost of any vector operation, excluding
load, store and vector to scalar operation.
(TARG_VEC_TO_SCALAR_COST): Define cost of vector to scalar operation.
(TARG_VEC_LOAD_COST): Define cost of aligned vector load.
(TARG_VEC_UNALIGNED_LOAD_COST): Define cost of misasligned vector load.
(TARG_VEC_STORE_COST): Define cost of vector store.
(vect_estimate_min_profitable_iters): Define new function.
* tree-vect-analyze.c (vect_analyze_operations): Add a compile-time
check to evaluate if loop iterations are less than minimum profitable
iterations determined by cost model or minimum vect loop bound defined
by user, whichever is more conservative.
* tree-vect-transform.c (vect_do_peeling_for_loop_bound): Add a
run-time check to evaluate if loop iterations are less than minimum
profitable iterations determined by cost model or minimum vect loop
bound defined by user, whichever is more conservative.
(vect_estimate_min_profitable_iterations): New function to estimate
mimimimum iterartions required for vector version of loop to be
profitable over scalar version.
(vect_model_reduction_cost): New function.
(vect_model_induction_cost): New function.
(vect_model_simple_cost): New function.
(vect_cost_strided_group_size): New function.
(vect_model_store_cost): New function.
(vect_model_load_cost): New function.
(vectorizable_reduction): Call vect_model_reduction_cost during
analysis phase.
(vectorizable_induction): Call vect_model_induction_cost during
analysis phase.
(vectorizable_load): Call vect_model_load_cost during analysis phase.
(vectorizable_store): Call vect_model_store_cost during analysis phase.
(vectorizable_call, vectorizable_assignment, vectorizable_operation,
vectorizable_promotion, vectorizable_demotion): Call
vect_model_simple_cost during analysis phase.
gcc/testsuite/ChangeLog:
2007-06-08 Harsha Jagasia <harsha.jagasia@amd.com>
* gcc.dg/vect/costmodel: New directory.
* gcc.dg/vect/costmodel/i386: New directory.
* gcc.dg/vect/costmodel/i386/i386-costmodel-vect.exp: New testsuite.
* gcc.dg/vect/costmodel/i386/costmodel-fast-math-vect-pr29925.c:
New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-33.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-68.c: New test.
* gcc.dg/vect/costmodel/i386/costmodel-vect-reduc-1char.c: New test.
* gcc.dg/vect/costmodel/x86_64: New directory.
* gcc.dg/vect/costmodel/x86_64/x86_64-costmodel-vect.exp:
New testsuite.
* gcc.dg/vect/costmodel/x86_64/costmodel-fast-math-vect-pr29925.c:
New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-31.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-33.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-68.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-vect-reduc-1char.c: New test.
* gcc.dg/vect/costmodel/x86_64/costmodel-pr30843.c: New test.
Co-Authored-By: Tony Linthicum <tony.linthicum@amd.com>
From-SVN: r125575
PR tree-optimization/32243
* tree-vect-transform.c (vectorizable_type_promotion): Move check
for ncopies after ratio check between nunits_out and nunits_in.
(vectorizable_type_demotion): Remove single-use variable "scalar_type".
testsuite/ChangeLog:
PR tree-optimization/32243
* gcc.dg/vect/vect.exp: Add support for -O3 tests. Reset default
flags for -Os tests.
* gcc.dg/vect/03-vect-pr32243.c: New test.
From-SVN: r125567
PR tree-optimization/32224
* tree-vect-analyze.c (vect_determine_vectorization_factor): Fail
vectorization upon a non GIMPLE_MODIFY_STMT.
From-SVN: r125566
gcc/
2007-06-07 Simon Martin <simartin@users.sourceforge.net>
PR c++/30759
* c-common.h (flag_cpp0x): Replaced by...
(cxx_dialect): ... this new variable specifying the C++ dialect that
is used.
* c-common.c (flag_cpp0x): Removed.
(cxx_dialect): Defined.
* c-cppbuiltin.c (c_cpp_builtins): flag_cpp0x rewritten in terms of
cxx_dialect.
* c-opts.c (c_common_post_options): Likewise.
(set_std_cxx98): Set cxx_dialect to cxx98.
(set_std_cxx0x): Set cxx_dialect to cxx0x.
gcc/cp/
2007-06-07 Simon Martin <simartin@users.sourceforge.net>
PR c++/30759
* decl.c (check_initializer): Report an error when a brace enclosed
initializer is used for a non-aggregate type in C++98.
(redeclaration_error_message): Rewrote flag_cpp0x in terms of
cxx_dialect.
(grokdeclarator): Likewise.
(move_fn_p): Likewise.
* typeck.c (check_return_expr): Likewise.
* call.c (reference_binding): Likewise.
* error.c (cp_cpp_error): Likewise.
* pt.c (check_default_tmpl_args): Likewise.
(tsubst): Likewise.
* lex.c (init_reswords): Likewise.
* parser.c (p_parser_primary_expression): Likewise.
(TOKEN_PRECEDENCE): Likewise.
(cp_parser_init_declarator): Likewise.
(cp_parser_ptr_operator): Likewise.
(cp_parser_parameter_declaration): Likewise.
(cp_parser_enclosed_template_argument_list): Likewise.
(cp_parser_skip_to_end_of_template_parameter_list): Likewise.
(cp_parser_next_token_ends_template_argument_p): Likewise.
gcc/testsuite/
2007-06-07 Simon Martin <simartin@users.sourceforge.net>
PR c++/30759
* g++.dg/init/brace6.C: New test.
From-SVN: r125539
PR tree-optimization/32220
* tree-predcom.c (eliminate_temp_copies): Handle the case that loop
phi node is reached before defining statement.
* gfortran.dg/predcom-2.f: New testcase.
From-SVN: r125531
2007-06-06 Andrew Pinski <andrew_pinski@playstation.sony.com>
PR tree-opt/32231
* tree-vect-transform.c (vectorizable_call): Call update_stmt
after changing the right hand side of the assignment.
2007-06-06 Andrew Pinski <andrew_pinski@playstation.sony.com>
PR tree-opt/32231
* gcc.target/i386/vectorize4.c: New test.
From-SVN: r125521
* tree-ssa-alias-warnings.c (maybe_add_match): Cast according to the
coding conventions.
(add_key): Likewise.
* tree-ssa.c (init_tree_ssa): Use type safe memory macros.
* tree-ssa-ccp.c (ccp_fold_builtin): Avoid using C++ keywords as
variable names.
* tree-ssa-coalesce.c (find_coalesce_pair): Use type safe memory macros.
(add_cost_one_coalesce): Likewise.
* tree-ssa-copy.c (merge_alias_info): Avoid using C++ keywords as
variable names. Rename orig to orig_name for consistency.
* tree-ssa-dom.c (dom_thread_across_edge): Cast according to the coding
conventions.
(cprop_into_successor_phis): Avoid using C++ keywords as variable names.
(record_equivalences_from_stmt): Likewise.
* tree-ssa-dse.c (dse_initialize_block_local_data): Cast according to
the coding conventions.
(memory_ssa_name_same): Likewise.
(dse_optimize_stmt): Likewise.
(dse_record_phis): Likewise.
(dse_finalize_block): Likewise.
* tree-ssa-loop-im.c (outermost_invariant_loop_expr): Avoid using C++
keywords as variable names.
(may_move_till): Cast according to the coding conventions.
(force_move_till_expr): Avoid using C++ keywords as variable names.
(force_move_till): Cast according to the coding conventions.
(memref_hash): Likewise.
(memref_eq): Likewise.
(gather_mem_refs_stmt): Likewise.
* tree-ssa-loop-ivopts.c (contains_abnormal_ssa_name_p): Avoid using C++
keywords as variable names.
(idx_find_step): Cast according to the coding conventions.
(idx_record_use): Likewise.
(find_depends): Likewise.
(prepare_decl_rtl): Likewise.
(mbc_entry_hash): Likewise.
(mbc_entry_eq): Likewise.
* tree-ssa-loop-niter.c (SWAP): Use the correct the type for tmp.
(simplify_replace_tree): Avoid using C++ keywords as variable names.
(idx_infer_loop_bounds): Cast according to the coding conventions.
* tree-ssa-loop-prefetch.c (idx_analyze_ref): Likewise.
* tree-ssa-math-opts.c (occ_new ): Likwise.
* tree-ssanames.c (duplicate_ssa_name_ptr_info): Use type safe memory
macros.
* tree-ssa-operands.c (add_def_op): Avoid using C++ keywords as variable
names.
(add_use_op): Likewise.
(add_vop): Likewise.
(add_vuse_op): Likewise.
(add_vdef_op): Likewise.
(get_expr_operands): Likewise.
(push_stmt_changes): Use type safe memory macros.
* tree-ssa-phiopt.c (replace_phi_edge_with_variable): Avoid using C++
keywords as variable names.
(conditional_replacement): Likewise.
(minmax_replacement): Likewise.
(abs_replacement): Likewise.
* tree-ssa-pre.c (create_expression_by_pieces): Cast according to the
coding conventions.
(insert_fake_stores): Avoid using C++ keywords as variable names.
* tree-ssa-reassoc.c (add_to_ops_vec): Cast according to the coding
conventions.
* tree-ssa-structalias.c (heapvar_lookup): Likewise.
(heapvar_insert): Use type safe memory macros.
(new_var_info): Cast according to the coding conventions.
(new_constraint): Likewise.
(remove_preds_and_fake_succs): Use type safe memory macros.
* tree-ssa-threadupdate.c (thread_block): Cast according to the coding
conventions.
(thread_single_edge): Likewise.
From-SVN: r125504
2007-06-06 Eric Christopher <echristo@apple.com>
* config/i386/i386.c (override_options): Move handling
of TARGET_SUBTARGET* earlier.
From-SVN: r125497
* haifa-sched.c (restore_bb_notes): Clear bb field of the notes
emited outside of basic block.
* cfgbuild.c (find_bb_boundaries): Clear bb field for insns between
the created blocks.
* rtl.h (delete_insn_chain): Declaration changed.
* cfgrtl.c (delete_insn_chain): Add option to clear bb field for
non-removed insns.
(rtl_delete_block, rtl_merge_blocks): Pass true to delete_insn_chain.
(delete_insn_chain_and_edges, try_redirect_by_replacing_jump,
rtl_tidy_fallthru_edge, cfg_layout_merge_blocks): Pass false
to delete_insn_chain.
(rtl_verify_flow_info_1): Verify that the insns in header and footer
do not have bb field set.
(rtl_verify_flow_info): Verify that insns between basic blocks do not
have bb field set.
* recog.c (peephole2_optimize): Add argument to delete_insn_chain call.
* cfgcleanup.c (try_optimize_cfg): Ditto.
From-SVN: r125492
* lambda-code.c (struct lambda_lattice_s): Add a name to the struct.
(lambda_body_vector_new): Use type safe memory macros.
(lambda_linear_expression_new): Likewise.
(lambda_loopnest_new): Likewise.
(lambda_lattice_new): Likewise.
(replace_uses_equiv_to_x_with_y): Cast according to the coding
conventions. Use type safe memory macros.
* lambda.h (struct lambda_trans_matrix_s): Add a name to the struct.
(lambda_body_vector_s): Likewise.
* lambda-mat.c (lambda_matrix_new): Use type safe memory macros.
* lambda-trans.c (lambda_trans_matrix_new): Likewise.
From-SVN: r125491
2007-06-06 Richard Guenther <rguenther@suse.de>
* tree-ssa-forwprop.c (forward_propagate_into_cond): Return 2
if we need to schedule cfg_cleanup.
(tree_ssa_forward_propagate_single_use_vars): Do so.
From-SVN: r125490
./: * fold-const.c (merge_ranges): If range_successor or
range_predecessor fail, just return 0.
testsuite/:
* g++.dg/conversion/enum1.C: New test.
From-SVN: r125486
* config/i386/i386.c (enum pta_flags): Move out of struct scope...
(struct pta): ...from here. Change flags to unsigned to avoid excessive
casting (as it is used as a bit mask).
(override_options): Add casts according to the coding convenventions.
(x86_64_elf_unique_section): Likewise.
(examine_argument): Avoid using C++ keywords as variable names.
(construct_container): Likewise.
(legitimize_pic_address): Likewise.
(get_dllimport_decl): Cast according to the coding conventions. Use
type safe memory macros.
(legitimize_address): Cast according to the coding conventions.
(emit_i387_cw_initialization): Corrected the type of slot to enum
ix86_stack_slot.
(ix86_init_machine_status): Use type safe memory macros.
(bdesc_pcmpestr): Use UNKNOWN instead of integer 0.
(bdesc_pcmpistr): Likewise.
(bdesc_crc32): Likewise.
(bdesc_sse_3arg): Likewise.
(bdesc_2arg): Likewise.
(bdesc_1arg): Likewise.
(ix86_expand_sse_pcmpestr): Cast according to the coding conventions.
(ix86_expand_sse_pcmpistr): Likewise.
(ix86_expand_vec_set_builtin): Use EXPAND_NORMAL instead of integer 0.
(ix86_builtin_vectorized_function): Change the type of fn to unsigned
int to match the langhook definition.
(ix86_builtin_conversion): Change the type of code to unsigned init to
match the langhook definition.
(ix86_preferred_reload_class): Avoid using C++ keywords as variable
names.
(ix86_preferred_output_reload_class): Likewise.
(ix86_cannot_change_mode_class): Likewise.
(ix86_memory_move_cost): Likewise.
(ix86_rtx_costs): Cast the outer_code parameter to enum rtx_code to
avoid excessive casting later on.
(x86_output_mi_thunk): Avoid using C++ keywords as variable names.
From-SVN: r125357
* config/i386/sse.md (sse4_2_pcmpestr_cconly): Prefer pcmpestrm
as flags setting insn.
(sse4_2_pcmpistr_cconly): Prefer pcmpistrm as flags setting insn.
* config/i386/i386.md (UNSPEC_ROUNDP, UNSPEC_ROUNDS): Remove.
(UNSPEC_ROUND): New.
("sse4_1_round<mode>2"): New insn pattern.
("rint<mode>2"): Expand using "sse4_1_round<mode>2" pattern for
SSE4.1 targets.
("floor<mode>2"): Rename from floordf2 and floorsf2. Macroize
expander using SSEMODEF mode macro. Expand using
"sse4_1_round<mode>2" pattern for SSE4.1 targets.
("ceil<mode>2"): Rename from ceildf2 and ceilsf2. Macroize
expander using SSEMODEF mode macro. Expand using
"sse4_1_round<mode>2" pattern for SSE4.1 targets.
("btrunc<mode>2"): Rename from btruncdf2 and btruncsf2. Macroize
expander using SSEMODEF mode macro. Expand using
"sse4_1_round<mode>2" pattern for SSE4.1 targets.
* config/i386/sse.md ("sse4_1_roundpd", "sse4_1_roundps"): Use
UNSPEC_ROUND instead of UNSPEC_ROUNDP.
("sse4_1_roundsd", "sse4_1_roundss"): Use UNSPEC_ROUND instead of
UNSPEC_ROUNDS.
From-SVN: r125356
* lambda.h (build_linear_expr): New.
* lambda-code.c (lbv_to_gcc_expression, lle_to_gcc_expression):
Use build_linear_expr, call fold and force_gimple_operand.
(lambda_loopnest_to_gcc_loopnest): Check that there is
something to insert.
* testsuite/gcc.dg/tree-ssa/ltrans-6.c: New.
Co-Authored-By: Sebastian Pop <sebpop@gmail.com>
From-SVN: r125355
PR tree-optimization/32215
* tree-vectorizer.c (supportable_widening_operation): Return false
for unsupported FIX_TRUNC_EXPR tree code.
(supportable_narrowing_operation): Ditto for FLOAT_EXPR tree code.
From-SVN: r125343
* config/rs6000/rs6000.h (FIXED_SCRATCH): Use r0 as a scratch
register on SPE targets. Change documentation to reflect
reality.
* config/rs6000/rs6000.c (rs6000_conditional_register_usage):
Change FIXED_SCRATCH to 14 and document why we're keeping r14
out of the register allocation pool.
(rs6000_reg_live_or_pic_offset_p): New function.
(rs6000_emit_prologue): Move the actual saving of LR up to free
r0 for holding r11. Split saving of SPE 64-bit registers into
its own case. Ensure that offsets will always be in-range for
'evstdd' by using r11 as a scratch register to point at the start
of the SPE save area. Save r11 if necessary, as it is the static
chain register.
(rs6000_emit_epilogue): Split restoring of SPE 64-bit registers
into its own case. Ensure that offsets will always be in-range
for 'evldd' by using r11 as a scratch register to point at the
start of the SPE save area. Also adjust r11 when restoring
the stack pointer to compensate for pre-loading r11.
From-SVN: r125340
./: * tree-vrp.c (compare_values_warnv): Check TREE_NO_WARNING on a
PLUS_EXPR or MINUS_EXPR node before setting *strict_overflow_p.
(extract_range_from_assert): Set TREE_NO_WARNING when creating an
expression.
(test_for_singularity): Likewise.
testsuite/:
* gcc.dg/Wstrict-overflow-19.c: New test.
From-SVN: r125334
* config/i386/predicates/md (reg_not_xmm0_operand): New predicate.
(nonimm_not_xmm0_operand): Ditto.
* config/i386/sse.md ("sse4_1_blendvpd"): Use "reg_not_xmm0_operand"
as operand[0] and operand[1] predicate. Use "nonimm_not_xmm0_operand"
as operand[2] predicate. Require "z" class XMM register for
operand[3]. Adjust asm template.
("sse4_1_blendvpd"): Ditto.
("sse4_1_pblendvb"): Ditto.
* config/i386/i386.c (ix86_expand_sse_4_operands_builtin): Do not
force op2 into xmm0 register for variable blend instructions.
From-SVN: r125327
./: * tree-vrp.c (adjust_range_with_scev): When loop is not expected
to overflow, reduce overflow infinity to regular infinity.
(vrp_var_may_overflow): New static function.
(vrp_visit_phi_node): Check vrp_var_may_overflow.
testsuite/:
* gcc.dg/Wstrict-overflow-18.c: New test.
From-SVN: r125319
* config/m68k/m68k.c (override_options): Don't override
REAL_MODE_FORMAT.
* config/m68k/m68k-modes.def (SF, DF): Define to use
motorola_single_format and motorola_double_format, resp.
* real.c (motorola_single_format): Renamed from
coldfire_single_format.
(motorola_double_format): Renamed from coldfire_double_format.
(encode_ieee_extended): Generate a proper canonical NaN image
respecting canonical_nan_lsbs_set.
(ieee_extended_motorola_format): Set canonical_nan_lsbs_set to
true.
* real.h: Adjust declarations.
From-SVN: r125295
* config/i386/i386.md ("sse4_1_blendvpd"): Require "z" class XMM
register for operand[3]. Adjust asm template.
("sse4_1_blendvpd"): Ditto.
("sse4_1_pblendvb"): Ditto.
* config/i386/i386.c (ix86_expand_sse_4_operands_builtin): Call
safe_vector_operand() if input operand is VECTOR_MODE_P operand. Do not
force operands[3] into xmm0 register for variable blend instructions.
(ix86_expand_sse_pcmpestr): Do not check operands for
"register_operand", when insn operand predicate is "register_operand".
(ix86_expand_sse_pcmpistr): Ditto.
From-SVN: r125280