/* * This file is part of libFirm. * Copyright (C) 2012 University of Karlsruhe. */ /** * @file * @brief Backend node support for generic backend nodes. * @author Sebastian Hack * * Backend node support for generic backend nodes. */ #include "benode.h" #include "array.h" #include "be_t.h" #include "bearch.h" #include "bedump.h" #include "beirg.h" #include "belive.h" #include "besched.h" #include "bitfiddle.h" #include "fourcc.h" #include "irbackedge_t.h" #include "ircons_t.h" #include "irgopt.h" #include "irgwalk.h" #include "irmode_t.h" #include "irnode_t.h" #include "irop_t.h" #include "iropt_t.h" #include "irprintf.h" #include "irverify_t.h" #include "obst.h" #include "panic.h" #include "pmap.h" #include "raw_bitset.h" #include "set.h" #include "target_t.h" #include "util.h" #include /** The be_IncSP attribute type. */ typedef struct { int offset; /**< The offset by which the stack shall be expanded/shrinked. */ bool no_align; } be_incsp_attr_t; typedef struct { ir_entity **in_entities; ir_entity **out_entities; int offset; } be_memperm_attr_t; typedef struct be_relocation_attr_t { ir_entity *entity; unsigned kind; } be_relocation_attr_t; ir_op *op_be_Asm; ir_op *op_be_Copy; ir_op *op_be_CopyKeep; ir_op *op_be_IncSP; ir_op *op_be_Keep; ir_op *op_be_MemPerm; ir_op *op_be_Perm; ir_op *op_be_Relocation; ir_op *op_be_Start; ir_op *op_be_Unknown; #define be_op_tag FOURCC('B', 'E', '\0', '\0') static int be_asm_attr_equal(ir_node const *const a, ir_node const *const b) { be_asm_attr_t const *const attr_a = get_be_asm_attr_const(a); be_asm_attr_t const *const attr_b = get_be_asm_attr_const(b); return attr_a->text == attr_b->text && attr_a->operands == attr_b->operands; } /** * Compare the attributes of two be_IncSP nodes. */ static int be_incsp_attrs_equal(const ir_node *a, const ir_node *b) { const be_incsp_attr_t *attr_a = (const be_incsp_attr_t*)get_irn_generic_attr_const(a); const be_incsp_attr_t *attr_b = (const be_incsp_attr_t*)get_irn_generic_attr_const(b); return attr_a->offset == attr_b->offset && attr_a->no_align == attr_b->no_align && attrs_equal_be_node(a, b); } static int be_relocation_attrs_equal(ir_node const *a, ir_node const *b) { be_relocation_attr_t const *attr_a = (be_relocation_attr_t const*)get_irn_generic_attr_const(a); be_relocation_attr_t const *attr_b = (be_relocation_attr_t const*)get_irn_generic_attr_const(b); return attr_a->entity == attr_b->entity && attr_a->kind == attr_b->kind; } arch_register_req_t const **be_allocate_in_reqs(ir_graph *const irg, unsigned const n) { struct obstack *const obst = be_get_be_obst(irg); return OALLOCN(obst, arch_register_req_t const*, n); } static void be_node_set_register_req_in(ir_node *const node, int const pos, arch_register_req_t const *const req) { backend_info_t *info = be_get_info(node); assert(pos < get_irn_arity(node)); info->in_reqs[pos] = req; } /** * Initializes the generic attribute of all be nodes and return it. */ static void init_node_attr(ir_node *const node, unsigned const n_outputs, arch_irn_flags_t const flags) { ir_graph *irg = get_irn_irg(node); backend_info_t *info = be_get_info(node); unsigned const arity = get_irn_arity(node); arch_register_req_t const **const in_reqs = is_irn_dynamic(node) ? NEW_ARR_F(arch_register_req_t const*, arity) : arity != 0 ? be_allocate_in_reqs(irg, arity) : NULL; for (unsigned i = 0; i < arity; ++i) { in_reqs[i] = arch_no_register_req; } info->in_reqs = in_reqs; struct obstack *const obst = be_get_be_obst(irg); info->out_infos = NEW_ARR_DZ(reg_out_info_t, obst, n_outputs); for (unsigned i = 0; i < n_outputs; ++i) { info->out_infos[i].req = arch_no_register_req; } info->flags = flags; } ir_node *be_new_Perm(ir_node *const block, int const n, ir_node *const *const in) { ir_graph *const irg = get_irn_irg(block); ir_node *const irn = new_ir_node(NULL, irg, block, op_be_Perm, mode_T, n, in); init_node_attr(irn, n, arch_irn_flags_none); for (int i = 0; i < n; ++i) { arch_register_req_t const *const in_req = arch_get_irn_register_req(in[i]); arch_register_req_t const *const slot_req = in_req->width == 1 ? in_req->cls->class_req : be_create_cls_req(irg, in_req->cls, in_req->width); be_node_set_register_req_in( irn, i, slot_req); arch_set_irn_register_req_out(irn, i, slot_req); } return irn; } ir_node *be_new_MemPerm(ir_node *const block, int n, ir_node *const *const in) { ir_graph *const irg = get_irn_irg(block); ir_node *const irn = new_ir_node(NULL, irg, block, op_be_MemPerm, mode_T, n, in); init_node_attr(irn, n, arch_irn_flags_none); for (int i = 0; i < n; ++i) { be_node_set_register_req_in( irn, i, arch_memory_req); arch_set_irn_register_req_out(irn, i, arch_memory_req); } be_memperm_attr_t *attr = (be_memperm_attr_t*)get_irn_generic_attr(irn); attr->in_entities = OALLOCNZ(get_irg_obstack(irg), ir_entity*, n); attr->out_entities = OALLOCNZ(get_irg_obstack(irg), ir_entity*, n); attr->offset = 0; return irn; } static void set_copy_info(ir_node *const irn, ir_graph *const irg, ir_node *const op, arch_irn_flags_t const flags) { init_node_attr(irn, 1, flags); arch_register_req_t const *const op_req = arch_get_irn_register_req(op); arch_register_class_t const *const cls = op_req->cls; be_node_set_register_req_in(irn, 0, cls->class_req); arch_register_req_t *const out_req = be_create_cls_req(irg, cls, op_req->width); out_req->should_be_same = 1U << 0; arch_set_irn_register_req_out(irn, 0, out_req); } ir_node *be_new_d_Copy(dbg_info *const dbgi, ir_node *const block, ir_node *const op) { ir_graph *const irg = get_irn_irg(block); ir_node *const in[] = { op }; ir_node *const res = new_ir_node(dbgi, irg, block, op_be_Copy, get_irn_mode(op), ARRAY_SIZE(in), in); set_copy_info(res, irg, op, arch_irn_flags_none); return res; } ir_node *be_new_Copy(ir_node *const bl, ir_node *const op) { return be_new_d_Copy(NULL, bl, op); } ir_node *be_get_Copy_op(const ir_node *cpy) { return get_irn_n(cpy, n_be_Copy_op); } ir_node *be_new_Copy_before_reg(ir_node *const val, ir_node *const before, arch_register_t const *const reg) { ir_node *const block = get_nodes_block(before); ir_node *const copy = be_new_Copy(block, val); sched_add_before(before, copy); arch_set_irn_register_out(copy, 0, reg); return copy; } ir_node *be_new_Keep(ir_node *const block, int const n, ir_node *const *const in) { ir_graph *irg = get_irn_irg(block); ir_node *res = new_ir_node(NULL, irg, block, op_be_Keep, mode_ANY, n, in); init_node_attr(res, 1, arch_irn_flag_schedule_first); for (int i = 0; i < n; ++i) { arch_register_req_t const *const req = arch_get_irn_register_req(in[i]); be_node_set_register_req_in(res, i, req->cls->class_req); } keep_alive(res); return res; } ir_node *be_new_Keep_one(ir_node *const kept) { ir_node *const in[] = { kept }; ir_node *const block = get_nodes_block(kept); return be_new_Keep(block, ARRAY_SIZE(in), in); } ir_node *be_new_IncSP(ir_node *bl, ir_node *old_sp, int offset, bool no_align) { arch_register_class_t const *const cls = arch_get_irn_register_req(old_sp)->cls; ir_graph *irg = get_irn_irg(bl); ir_node *in[] = { old_sp }; ir_node *irn = new_ir_node(NULL, irg, bl, op_be_IncSP, cls->mode, ARRAY_SIZE(in), in); init_node_attr(irn, 1, arch_irn_flags_none); be_incsp_attr_t *a = (be_incsp_attr_t*)get_irn_generic_attr(irn); a->offset = offset; a->no_align = no_align; /* Set output constraint to stack register. */ be_node_set_register_req_in(irn, 0, cls->class_req); arch_copy_irn_out_info(irn, 0, old_sp); return irn; } ir_node *be_new_CopyKeep(ir_node *const bl, ir_node *const src, int const n, ir_node *const *const in_keep) { ir_mode *mode = get_irn_mode(src); ir_graph *irg = get_irn_irg(bl); int arity = n+1; ir_node **in = ALLOCAN(ir_node*, arity); in[0] = src; MEMCPY(&in[1], in_keep, n); ir_node *irn = new_ir_node(NULL, irg, bl, op_be_CopyKeep, mode, arity, in); set_copy_info(irn, irg, src, arch_irn_flag_schedule_first); for (int i = 0; i < n; ++i) { ir_node *pred = in_keep[i]; arch_register_req_t const *const req = arch_get_irn_register_req(pred); be_node_set_register_req_in(irn, i + 1, req->cls->class_req); } return irn; } ir_node *be_get_CopyKeep_op(const ir_node *cpy) { return get_irn_n(cpy, n_be_CopyKeep_op); } void be_set_MemPerm_in_entity(const ir_node *irn, unsigned n, ir_entity *ent) { assert(n < be_get_MemPerm_entity_arity(irn)); const be_memperm_attr_t *attr = (const be_memperm_attr_t*)get_irn_generic_attr_const(irn); attr->in_entities[n] = ent; } ir_entity* be_get_MemPerm_in_entity(const ir_node* irn, unsigned n) { assert(n < be_get_MemPerm_entity_arity(irn)); const be_memperm_attr_t *attr = (const be_memperm_attr_t*)get_irn_generic_attr_const(irn); return attr->in_entities[n]; } void be_set_MemPerm_out_entity(const ir_node *irn, unsigned n, ir_entity *ent) { assert(n < be_get_MemPerm_entity_arity(irn)); const be_memperm_attr_t *attr = (const be_memperm_attr_t*)get_irn_generic_attr_const(irn); attr->out_entities[n] = ent; } ir_entity* be_get_MemPerm_out_entity(const ir_node* irn, unsigned n) { assert(n < be_get_MemPerm_entity_arity(irn)); const be_memperm_attr_t *attr = (const be_memperm_attr_t*)get_irn_generic_attr_const(irn); return attr->out_entities[n]; } void be_set_MemPerm_offset(ir_node *irn, int offset) { assert(be_is_MemPerm(irn)); be_memperm_attr_t *attr = (be_memperm_attr_t*)get_irn_generic_attr(irn); attr->offset = offset; } int be_get_MemPerm_offset(const ir_node *irn) { assert(be_is_MemPerm(irn)); const be_memperm_attr_t *attr = (const be_memperm_attr_t*)get_irn_generic_attr_const(irn); return attr->offset; } unsigned be_get_MemPerm_entity_arity(const ir_node *irn) { assert(be_is_MemPerm(irn)); return get_irn_arity(irn); } ir_node *be_get_IncSP_pred(ir_node *irn) { assert(be_is_IncSP(irn)); return get_irn_n(irn, n_be_IncSP_pred); } void be_set_IncSP_pred(ir_node *incsp, ir_node *pred) { assert(be_is_IncSP(incsp)); set_irn_n(incsp, n_be_IncSP_pred, pred); } void be_set_IncSP_offset(ir_node *irn, int offset) { assert(be_is_IncSP(irn)); be_incsp_attr_t *a = (be_incsp_attr_t*)get_irn_generic_attr(irn); a->offset = offset; } int be_get_IncSP_offset(const ir_node *irn) { assert(be_is_IncSP(irn)); const be_incsp_attr_t *a = (const be_incsp_attr_t*)get_irn_generic_attr_const(irn); return a->offset; } bool be_get_IncSP_no_align(const ir_node *irn) { assert(be_is_IncSP(irn)); const be_incsp_attr_t *a = (const be_incsp_attr_t*)get_irn_generic_attr_const(irn); return a->no_align; } ir_node *be_new_Phi(ir_node *block, int n_ins, ir_node **ins, arch_register_req_t const *req) { ir_graph *irg = get_irn_irg(block); ir_node *phi = new_ir_node(NULL, irg, block, op_Phi, req->cls->mode, n_ins, ins); phi->attr.phi.u.backedge = new_backedge_arr(get_irg_obstack(irg), n_ins); struct obstack *obst = be_get_be_obst(irg); backend_info_t *info = be_get_info(phi); info->out_infos = NEW_ARR_DZ(reg_out_info_t, obst, 1); info->in_reqs = be_allocate_in_reqs(irg, n_ins); info->out_infos[0].req = req; for (int i = 0; i < n_ins; ++i) { info->in_reqs[i] = req; } verify_new_node(phi); return optimize_node(phi); } ir_node *be_new_Phi0(ir_node *const block, arch_register_req_t const *const req) { ir_graph *const irg = get_irn_irg(block); ir_node *const phi = new_ir_node(NULL, irg, block, op_Phi, req->cls->mode, 0, NULL); struct obstack *const obst = be_get_be_obst(irg); backend_info_t *const info = be_get_info(phi); info->out_infos = NEW_ARR_DZ(reg_out_info_t, obst, 1); info->out_infos[0].req = req; return phi; } ir_node *be_complete_Phi(ir_node *const phi, unsigned const n_ins, ir_node **const ins) { assert(is_Phi(phi) && get_Phi_n_preds(phi) == 0); ir_graph *const irg = get_irn_irg(phi); phi->attr.phi.u.backedge = new_backedge_arr(get_irg_obstack(irg), n_ins); set_irn_in(phi, n_ins, ins); arch_register_req_t const **const in_reqs = be_allocate_in_reqs(irg, n_ins); arch_register_req_t const *const req = arch_get_irn_register_req(phi); for (unsigned i = 0; i < n_ins; ++i) { in_reqs[i] = req; } backend_info_t *const info = be_get_info(phi); info->in_reqs = in_reqs; verify_new_node(phi); return phi; } void be_set_phi_reg_req(ir_node *node, const arch_register_req_t *req) { assert(mode_is_data(get_irn_mode(node))); backend_info_t *info = be_get_info(node); info->out_infos[0].req = req; for (int i = 0, arity = get_irn_arity(node); i < arity; ++i) { info->in_reqs[i] = req; } } ir_node *be_new_Asm(dbg_info *const dbgi, ir_node *const block, int const n_ins, ir_node **const ins, arch_register_req_t const **const in_reqs, int const n_outs, ident *const text, void *const operands) { ir_graph *const irg = get_irn_irg(block); ir_node *const asmn = new_ir_node(dbgi, irg, block, op_be_Asm, mode_T, n_ins, ins); be_info_init_irn(asmn, arch_irn_flags_none, in_reqs, n_outs); be_asm_attr_t *const attr = (be_asm_attr_t*)get_irn_generic_attr(asmn); attr->text = text; attr->operands = operands; return asmn; } ir_node *be_new_Relocation(dbg_info *const dbgi, ir_graph *const irg, unsigned const kind, ir_entity *const entity, ir_mode *const mode) { ir_node *const block = get_irg_start_block(irg); ir_node *const node = new_ir_node(dbgi, irg, block, op_be_Relocation, mode, 0, NULL); be_relocation_attr_t *const attr = (be_relocation_attr_t*)get_irn_generic_attr(node); attr->entity = entity; attr->kind = kind; ir_node *const optimized = optimize_node(node); return optimized; } unsigned be_get_Relocation_kind(ir_node const* const node) { assert(be_is_Relocation(node)); be_relocation_attr_t const *const attr = (be_relocation_attr_t const*)get_irn_generic_attr_const(node); return attr->kind; } ir_entity *be_get_Relocation_entity(ir_node const* const node) { assert(be_is_Relocation(node)); be_relocation_attr_t const *const attr = (be_relocation_attr_t const*)get_irn_generic_attr_const(node); return attr->entity; } ir_node *be_new_Start(ir_graph *const irg, be_start_out const *const outs) { ir_node *const block = get_irg_start_block(irg); ir_node *const start = new_ir_node(NULL, irg, block, op_be_Start, mode_T, 0, NULL); unsigned const n_regs = ir_target.isa->n_registers; /* Count the number of outputs. */ unsigned k = 1; /* +1 for memory */ for (unsigned i = 0; i != n_regs; ++i) { if (outs[i] != BE_START_NO) ++k; } be_info_init_irn(start, arch_irn_flag_schedule_first, NULL, k); /* Set out requirements and registers. */ unsigned l = 0; arch_set_irn_register_req_out(start, l++, arch_memory_req); arch_register_t const *const regs = ir_target.isa->registers; for (unsigned i = 0; i != n_regs; ++i) { if (outs[i] != BE_START_NO) { arch_register_t const *const reg = ®s[i]; bool const ignore = outs[i] == BE_START_IGNORE; arch_register_req_t const *const req = be_create_reg_req(irg, reg, ignore); arch_set_irn_register_req_out(start, l, req); arch_set_irn_register_out( start, l, reg); ++l; } } assert(l == k); return start; } ir_node *be_get_Start_mem(ir_graph *const irg) { ir_node *const start = get_irg_start(irg); return be_get_or_make_Proj_for_pn(start, 0); } ir_node *be_get_Start_proj(ir_graph *const irg, arch_register_t const *const reg) { ir_node *const start = get_irg_start(irg); /* do a naive linear search... */ be_foreach_out(start, i) { arch_register_t const *const out_reg = arch_get_irn_register_out(start, i); if (out_reg == reg) return be_get_or_make_Proj_for_pn(start, i); } panic("tried querying undefined register '%s' at Start", reg->name); } ir_node *be_new_Unknown(ir_node *const block, arch_register_req_t const *const req) { ir_graph *const irg = get_irn_irg(block); ir_mode *const mode = req->cls->mode; ir_node *const node = new_ir_node(NULL, irg, block, op_be_Unknown, mode, 0, NULL); init_node_attr(node, 1, arch_irn_flag_rematerializable); arch_set_irn_register_req_out(node, 0, req); return optimize_node(node); } ir_node *be_new_Proj(ir_node *const pred, unsigned const pos) { arch_register_req_t const *const req = arch_get_irn_register_req_out(pred, pos); return new_r_Proj(pred, req->cls->mode, pos); } ir_node *be_new_Proj_reg(ir_node *const pred, unsigned const pos, arch_register_t const *const reg) { arch_set_irn_register_out(pred, pos, reg); return be_new_Proj(pred, pos); } ir_node *be_get_or_make_Proj_for_pn(ir_node *const irn, unsigned const pn) { ir_node *const proj = get_Proj_for_pn(irn, pn); return proj ? proj : be_new_Proj(irn, pn); } /** * ir_op-Operation: dump a be node to file */ static void be_dump_node(FILE *f, const ir_node *irn, dump_reason_t reason) { assert(is_be_node(irn)); switch (reason) { case dump_node_opcode_txt: fputs(get_irn_opname(irn), f); break; case dump_node_mode_txt: { ir_mode *const mode = get_irn_mode(irn); if (mode != mode_ANY && mode != mode_T) fprintf(f, "%s", get_mode_name(mode)); break; } case dump_node_nodeattr_txt: if (be_is_IncSP(irn)) { const be_incsp_attr_t *attr = (const be_incsp_attr_t*)get_irn_generic_attr_const(irn); fprintf(f, " [%d] ", attr->offset); } break; case dump_node_info_txt: if (be_is_Asm(irn)) { be_asm_attr_t const *const attr = get_be_asm_attr_const(irn); fprintf(f, "assembler text: %s\n", get_id_str(attr->text)); } else if (be_is_IncSP(irn)) { const be_incsp_attr_t *a = (const be_incsp_attr_t*)get_irn_generic_attr_const(irn); fprintf(f, "no_align: %s\n", be_dump_yesno(a->no_align)); fprintf(f, "offset: %d\n", a->offset); } else if (be_is_MemPerm(irn)) { for (unsigned i = 0; i < be_get_MemPerm_entity_arity(irn); ++i) { ir_entity *in = be_get_MemPerm_in_entity(irn, i); ir_entity *out = be_get_MemPerm_out_entity(irn, i); if (in != NULL) ir_fprintf(f, "\nin[%u]: %F\n", i, in); if (out != NULL) ir_fprintf(f, "\nout[%u]: %F\n", i, out); } } break; } } void be_copy_attr(ir_graph *const irg, ir_node const *const old_node, ir_node *const new_node) { default_copy_attr(irg, old_node, new_node); backend_info_t *const old_info = be_get_info(old_node); backend_info_t *const new_info = be_get_info(new_node); *new_info = *old_info; memset(&new_info->sched_info, 0, sizeof(new_info->sched_info)); if (new_info->out_infos) { struct obstack *const obst = be_get_be_obst(irg); new_info->out_infos = DUP_ARR_D(reg_out_info_t, obst, new_info->out_infos); } } void be_switch_attr_init(ir_node *const node, be_switch_attr_t *const attr, ir_switch_table const *const table, ir_entity const *const table_entity) { attr->table = table; attr->table_entity = table_entity; be_foreach_out(node, o) { arch_set_irn_register_req_out(node, o, arch_exec_req); } } bool is_be_node(const ir_node *irn) { return get_op_tag(get_irn_op(irn)) == be_op_tag; } static ir_op *new_be_op(unsigned code, const char *name, op_pin_state p, irop_flags flags, size_t attr_size) { ir_op *const res = new_ir_op(code, name, p, flags, oparity_any, 0, attr_size); set_op_dump(res, be_dump_node); set_op_copy_attr(res, be_copy_attr); set_op_tag(res, be_op_tag); return res; } void be_init_op(void) { assert(op_be_Perm == NULL); /* Acquire all needed opcodes. */ unsigned const o = get_next_ir_opcodes(beo_last + 1); op_be_Asm = new_be_op(o + beo_Asm, "be_Asm", op_pin_state_exc_pinned, irop_flag_fragile | irop_flag_uses_memory, sizeof(be_asm_attr_t)); op_be_Copy = new_be_op(o + beo_Copy, "be_Copy", op_pin_state_floats, irop_flag_none, 0); op_be_CopyKeep = new_be_op(o + beo_CopyKeep, "be_CopyKeep", op_pin_state_floats, irop_flag_keep, 0); op_be_IncSP = new_be_op(o + beo_IncSP, "be_IncSP", op_pin_state_pinned, irop_flag_none, sizeof(be_incsp_attr_t)); op_be_Keep = new_be_op(o + beo_Keep, "be_Keep", op_pin_state_pinned, irop_flag_keep, 0); op_be_MemPerm = new_be_op(o + beo_MemPerm, "be_MemPerm", op_pin_state_pinned, irop_flag_none, sizeof(be_memperm_attr_t)); op_be_Perm = new_be_op(o + beo_Perm, "be_Perm", op_pin_state_pinned, irop_flag_none, 0); op_be_Relocation = new_be_op(o + beo_Relocation, "be_Relocation", op_pin_state_floats, irop_flag_constlike | irop_flag_start_block, sizeof(be_relocation_attr_t)); op_be_Start = new_be_op(o + beo_Start, "be_Start", op_pin_state_pinned, irop_flag_start_block, 0); op_be_Unknown = new_be_op(o + beo_Unknown, "be_Unknown", op_pin_state_floats, irop_flag_constlike, 0); ir_op_set_fragile_indices(op_be_Asm, pn_be_Asm_X_regular, (unsigned)-1); ir_op_set_memory_index(op_be_Asm, pn_be_Asm_M); set_op_attrs_equal(op_be_Asm, be_asm_attr_equal); set_op_attrs_equal(op_be_Copy, attrs_equal_be_node); set_op_attrs_equal(op_be_CopyKeep, attrs_equal_be_node); set_op_attrs_equal(op_be_IncSP, be_incsp_attrs_equal); set_op_attrs_equal(op_be_Keep, attrs_equal_be_node); set_op_attrs_equal(op_be_MemPerm, attrs_equal_be_node); set_op_attrs_equal(op_be_Perm, attrs_equal_be_node); set_op_attrs_equal(op_be_Relocation, be_relocation_attrs_equal); set_op_attrs_equal(op_be_Unknown, attrs_equal_be_node); } void be_finish_op(void) { free_ir_op(op_be_Copy); op_be_Copy = NULL; free_ir_op(op_be_CopyKeep); op_be_CopyKeep = NULL; free_ir_op(op_be_IncSP); op_be_IncSP = NULL; free_ir_op(op_be_Keep); op_be_Keep = NULL; free_ir_op(op_be_MemPerm); op_be_MemPerm = NULL; free_ir_op(op_be_Perm); op_be_Perm = NULL; }