2 * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Backend ABI implementation.
23 * @author Sebastian Hack, Michael Beck
36 #include "irgraph_t.h"
39 #include "iredges_t.h"
42 #include "irprintf_t.h"
48 #include "raw_bitset.h"
55 #include "besched_t.h"
57 #include "bessaconstr.h"
59 typedef struct _be_abi_call_arg_t {
60 unsigned is_res : 1; /**< 1: the call argument is a return value. 0: it's a call parameter. */
61 unsigned in_reg : 1; /**< 1: this argument is transmitted in registers. */
62 unsigned on_stack : 1; /**< 1: this argument is transmitted on the stack. */
65 const arch_register_t *reg;
67 unsigned alignment; /**< stack alignment */
68 unsigned space_before; /**< allocate space before */
69 unsigned space_after; /**< allocate space after */
72 struct _be_abi_call_t {
73 be_abi_call_flags_t flags;
74 const be_abi_callbacks_t *cb;
75 ir_type *between_type;
77 const arch_register_class_t *cls_addr;
80 struct _be_abi_irg_t {
82 be_stack_layout_t *frame; /**< The stack frame model. */
83 be_irg_t *birg; /**< The back end IRG. */
84 const arch_isa_t *isa; /**< The isa. */
85 survive_dce_t *dce_survivor;
87 be_abi_call_t *call; /**< The ABI call information. */
88 ir_type *method_type; /**< The type of the method of the IRG. */
90 ir_node *init_sp; /**< The node representing the stack pointer
91 at the start of the function. */
93 ir_node *start_barrier; /**< The barrier of the start block */
95 ir_node *reg_params; /**< The reg params node. */
96 pmap *regs; /**< A map of all callee-save and ignore regs to
97 their Projs to the RegParams node. */
99 int start_block_bias; /**< The stack bias at the end of the start block. */
101 void *cb; /**< ABI Callback self pointer. */
103 pmap *keep_map; /**< mapping blocks to keep nodes. */
104 pset *ignore_regs; /**< Additional registers which shall be ignored. */
106 ir_node **calls; /**< flexible array containing all be_Call nodes */
108 arch_register_req_t sp_req;
109 arch_register_req_t sp_cls_req;
111 DEBUG_ONLY(firm_dbg_module_t *dbg;) /**< The debugging module. */
114 static heights_t *ir_heights;
116 /* Flag: if set, try to omit the frame pointer if called by the backend */
117 static int be_omit_fp = 1;
120 _ ____ ___ ____ _ _ _ _
121 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
122 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
123 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
124 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
126 These callbacks are used by the backend to set the parameters
127 for a specific call type.
131 * Set compare function: compares two ABI call object arguments.
133 static int cmp_call_arg(const void *a, const void *b, size_t n)
135 const be_abi_call_arg_t *p = a, *q = b;
136 return !(p->is_res == q->is_res && p->pos == q->pos);
140 * Get or set an ABI call object argument.
142 * @param call the abi call
143 * @param is_res true for call results, false for call arguments
144 * @param pos position of the argument
145 * @param do_insert true if the argument is set, false if it's retrieved
147 static be_abi_call_arg_t *get_or_set_call_arg(be_abi_call_t *call, int is_res, int pos, int do_insert)
149 be_abi_call_arg_t arg;
152 memset(&arg, 0, sizeof(arg));
156 hash = is_res * 128 + pos;
159 ? set_insert(call->params, &arg, sizeof(arg), hash)
160 : set_find(call->params, &arg, sizeof(arg), hash);
164 * Retrieve an ABI call object argument.
166 * @param call the ABI call object
167 * @param is_res true for call results, false for call arguments
168 * @param pos position of the argument
170 static INLINE be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos)
172 return get_or_set_call_arg(call, is_res, pos, 0);
175 /* Set the flags for a call. */
176 void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const be_abi_callbacks_t *cb)
183 /* Set register class for call address */
184 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
186 call->cls_addr = cls;
190 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos, unsigned alignment, unsigned space_before, unsigned space_after)
192 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 0, arg_pos, 1);
194 arg->alignment = alignment;
195 arg->space_before = space_before;
196 arg->space_after = space_after;
197 assert(alignment > 0 && "Alignment must be greater than 0");
200 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg)
202 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 0, arg_pos, 1);
207 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg)
209 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 1, arg_pos, 1);
214 /* Get the flags of a ABI call object. */
215 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
221 * Constructor for a new ABI call object.
223 * @return the new ABI call object
225 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
227 be_abi_call_t *call = xmalloc(sizeof(call[0]));
230 call->params = new_set(cmp_call_arg, 16);
232 call->cls_addr = cls_addr;
234 call->flags.bits.try_omit_fp = be_omit_fp;
240 * Destructor for an ABI call object.
242 static void be_abi_call_free(be_abi_call_t *call)
244 del_set(call->params);
250 | ___| __ __ _ _ __ ___ ___ | | | | __ _ _ __ __| | (_)_ __ __ _
251 | |_ | '__/ _` | '_ ` _ \ / _ \ | |_| |/ _` | '_ \ / _` | | | '_ \ / _` |
252 | _|| | | (_| | | | | | | __/ | _ | (_| | | | | (_| | | | | | | (_| |
253 |_| |_| \__,_|_| |_| |_|\___| |_| |_|\__,_|_| |_|\__,_|_|_|_| |_|\__, |
256 Handling of the stack frame. It is composed of three types:
257 1) The type of the arguments which are pushed on the stack.
258 2) The "between type" which consists of stuff the call of the
259 function pushes on the stack (like the return address and
260 the old base pointer for ia32).
261 3) The Firm frame type which consists of all local variables
265 static int get_stack_entity_offset(be_stack_layout_t *frame, ir_entity *ent, int bias)
267 ir_type *t = get_entity_owner(ent);
268 int ofs = get_entity_offset(ent);
272 /* Find the type the entity is contained in. */
273 for(index = 0; index < N_FRAME_TYPES; ++index) {
274 if(frame->order[index] == t)
278 /* Add the size of all the types below the one of the entity to the entity's offset */
279 for(i = 0; i < index; ++i)
280 ofs += get_type_size_bytes(frame->order[i]);
282 /* correct the offset by the initial position of the frame pointer */
283 ofs -= frame->initial_offset;
285 /* correct the offset with the current bias. */
292 * Retrieve the entity with given offset from a frame type.
294 static ir_entity *search_ent_with_offset(ir_type *t, int offset)
298 for(i = 0, n = get_compound_n_members(t); i < n; ++i) {
299 ir_entity *ent = get_compound_member(t, i);
300 if(get_entity_offset(ent) == offset)
307 static int stack_frame_compute_initial_offset(be_stack_layout_t *frame)
309 ir_type *base = frame->stack_dir < 0 ? frame->between_type : frame->frame_type;
310 ir_entity *ent = search_ent_with_offset(base, 0);
312 frame->initial_offset = ent ? get_stack_entity_offset(frame, ent, 0) : 0;
314 return frame->initial_offset;
318 * Initializes the frame layout from parts
320 * @param frame the stack layout that will be initialized
321 * @param args the stack argument layout type
322 * @param between the between layout type
323 * @param locals the method frame type
324 * @param stack_dir the stack direction
325 * @param param_map an array mapping method argument positions to the stack argument type
327 * @return the initialized stack layout
329 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
330 ir_type *between, ir_type *locals, int stack_dir,
331 ir_entity *param_map[])
333 frame->arg_type = args;
334 frame->between_type = between;
335 frame->frame_type = locals;
336 frame->initial_offset = 0;
337 frame->stack_dir = stack_dir;
338 frame->order[1] = between;
339 frame->param_map = param_map;
342 frame->order[0] = args;
343 frame->order[2] = locals;
346 frame->order[0] = locals;
347 frame->order[2] = args;
353 /** Dumps the stack layout to file. */
354 static void stack_layout_dump(FILE *file, be_stack_layout_t *frame)
358 ir_fprintf(file, "initial offset: %d\n", frame->initial_offset);
359 for (j = 0; j < N_FRAME_TYPES; ++j) {
360 ir_type *t = frame->order[j];
362 ir_fprintf(file, "type %d: %F size: %d\n", j, t, get_type_size_bytes(t));
363 for (i = 0, n = get_compound_n_members(t); i < n; ++i) {
364 ir_entity *ent = get_compound_member(t, i);
365 ir_fprintf(file, "\t%F int ofs: %d glob ofs: %d\n", ent, get_entity_offset_bytes(ent), get_stack_entity_offset(frame, ent, 0));
372 * Returns non-zero if the call argument at given position
373 * is transfered on the stack.
375 static INLINE int is_on_stack(be_abi_call_t *call, int pos)
377 be_abi_call_arg_t *arg = get_call_arg(call, 0, pos);
378 return arg && !arg->in_reg;
388 Adjustment of the calls inside a graph.
393 * Transform a call node.
394 * @param env The ABI environment for the current irg.
395 * @param irn The call node.
396 * @param curr_sp The stack pointer node to use.
397 * @return The stack pointer after the call.
399 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, ir_node *alloca_copy)
401 ir_graph *irg = env->birg->irg;
402 const arch_env_t *arch_env = env->birg->main_env->arch_env;
403 const arch_isa_t *isa = arch_env->isa;
404 ir_type *mt = get_Call_type(irn);
405 ir_node *call_ptr = get_Call_ptr(irn);
406 int n_params = get_method_n_params(mt);
407 ir_node *curr_mem = get_Call_mem(irn);
408 ir_node *bl = get_nodes_block(irn);
409 pset *results = pset_new_ptr(8);
410 pset *caller_save = pset_new_ptr(8);
411 pset *states = pset_new_ptr(2);
413 int stack_dir = arch_isa_stack_dir(isa);
414 const arch_register_t *sp = arch_isa_sp(isa);
415 be_abi_call_t *call = be_abi_call_new(sp->reg_class);
416 ir_mode *mach_mode = sp->reg_class->mode;
417 struct obstack *obst = &env->obst;
418 int no_alloc = call->flags.bits.frame_is_setup_on_call;
420 ir_node *res_proj = NULL;
421 int curr_res_proj = pn_Call_max;
429 const arch_register_t *reg;
430 const ir_edge_t *edge;
435 /* Let the isa fill out the abi description for that call node. */
436 arch_isa_get_call_abi(isa, mt, call);
438 /* Insert code to put the stack arguments on the stack. */
439 assert(get_Call_n_params(irn) == n_params);
440 for(i = 0; i < n_params; ++i) {
441 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
444 int arg_size = get_type_size_bytes(get_method_param_type(mt, i));
446 stack_size += round_up2(arg->space_before, arg->alignment);
447 stack_size += round_up2(arg_size, arg->alignment);
448 stack_size += round_up2(arg->space_after, arg->alignment);
449 obstack_int_grow(obst, i);
453 pos = obstack_finish(obst);
455 /* Collect all arguments which are passed in registers. */
456 for(i = 0, n = get_Call_n_params(irn); i < n; ++i) {
457 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
458 if(arg && arg->in_reg) {
459 obstack_int_grow(obst, i);
463 low_args = obstack_finish(obst);
465 /* If there are some parameters which shall be passed on the stack. */
468 int do_seq = call->flags.bits.store_args_sequential && !no_alloc;
471 * Reverse list of stack parameters if call arguments are from left to right.
472 * We must them reverse again if they are pushed (not stored) and the stack
473 * direction is downwards.
475 if (call->flags.bits.left_to_right ^ (do_seq && stack_dir < 0)) {
476 for (i = 0; i < n_pos >> 1; ++i) {
477 int other = n_pos - i - 1;
485 * If the stack is decreasing and we do not want to store sequentially,
486 * or someone else allocated the call frame
487 * we allocate as much space on the stack all parameters need, by
488 * moving the stack pointer along the stack's direction.
490 if(stack_dir < 0 && !do_seq && !no_alloc) {
491 curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, stack_size);
493 add_irn_dep(curr_sp, alloca_copy);
499 obstack_ptr_grow(obst, get_Call_mem(irn));
500 curr_mem = new_NoMem();
502 curr_mem = get_Call_mem(irn);
505 for(i = 0; i < n_pos; ++i) {
507 be_abi_call_arg_t *arg = get_call_arg(call, 0, p);
508 ir_node *param = get_Call_param(irn, p);
509 ir_node *addr = curr_sp;
511 ir_type *param_type = get_method_param_type(mt, p);
512 int param_size = get_type_size_bytes(param_type) + arg->space_after;
515 * If we wanted to build the arguments sequentially,
516 * the stack pointer for the next must be incremented,
517 * and the memory value propagated.
521 addr = curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, param_size + arg->space_before);
523 add_irn_dep(curr_sp, alloca_copy);
526 add_irn_dep(curr_sp, curr_mem);
529 curr_ofs += arg->space_before;
530 curr_ofs = round_up2(curr_ofs, arg->alignment);
532 /* Make the expression to compute the argument's offset. */
534 ir_mode *constmode = mach_mode;
535 if(mode_is_reference(mach_mode)) {
538 addr = new_r_Const_long(irg, bl, constmode, curr_ofs);
539 addr = new_r_Add(irg, bl, curr_sp, addr, mach_mode);
543 /* Insert a store for primitive arguments. */
544 if (is_atomic_type(param_type)) {
546 store = new_r_Store(irg, bl, curr_mem, addr, param);
547 mem = new_r_Proj(irg, bl, store, mode_M, pn_Store_M);
550 /* Make a mem copy for compound arguments. */
554 assert(mode_is_reference(get_irn_mode(param)));
555 copy = new_r_CopyB(irg, bl, curr_mem, addr, param, param_type);
556 mem = new_r_Proj(irg, bl, copy, mode_M, pn_CopyB_M_regular);
559 curr_ofs += param_size;
564 obstack_ptr_grow(obst, mem);
567 in = (ir_node **) obstack_finish(obst);
569 /* We need the sync only, if we didn't build the stores sequentially. */
572 curr_mem = new_r_Sync(irg, bl, n_pos + 1, in);
574 curr_mem = get_Call_mem(irn);
577 obstack_free(obst, in);
580 /* Collect caller save registers */
581 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
583 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
584 for(j = 0; j < cls->n_regs; ++j) {
585 const arch_register_t *reg = arch_register_for_index(cls, j);
586 if(arch_register_type_is(reg, caller_save)) {
587 pset_insert_ptr(caller_save, (void *) reg);
589 if(arch_register_type_is(reg, state)) {
590 pset_insert_ptr(caller_save, (void*) reg);
591 pset_insert_ptr(states, (void*) reg);
596 /* search the greatest result proj number */
598 /* TODO: what if the result is NOT used? Currently there is
599 * no way to detect this later, especially there is no way to
600 * see this in the proj numbers.
601 * While this is ok for the register allocator, it is bad for
602 * backends which need to change the be_Call further (x87 simulator
603 * for instance. However for this particular case the call_type is
606 foreach_out_edge(irn, edge) {
607 const ir_edge_t *res_edge;
608 ir_node *irn = get_edge_src_irn(edge);
610 if(is_Proj(irn) && get_Proj_proj(irn) == pn_Call_T_result) {
612 foreach_out_edge(irn, res_edge) {
614 be_abi_call_arg_t *arg;
615 ir_node *res = get_edge_src_irn(res_edge);
617 assert(is_Proj(res));
619 proj = get_Proj_proj(res);
620 arg = get_call_arg(call, 1, proj);
623 shift the proj number to the right, since we will drop the
624 unspeakable Proj_T from the Call. Therefore, all real argument
625 Proj numbers must be increased by pn_be_Call_first_res
627 proj += pn_be_Call_first_res;
628 set_Proj_proj(res, proj);
629 obstack_ptr_grow(obst, res);
631 if(proj > curr_res_proj)
632 curr_res_proj = proj;
634 pset_remove_ptr(caller_save, arg->reg);
635 //pmap_insert(arg_regs, arg->reg, INT_TO_PTR(proj + 1))
642 obstack_ptr_grow(obst, NULL);
643 res_projs = obstack_finish(obst);
645 /* make the back end call node and set its register requirements. */
646 for(i = 0; i < n_low_args; ++i) {
647 obstack_ptr_grow(obst, get_Call_param(irn, low_args[i]));
649 foreach_pset(states, reg) {
650 const arch_register_class_t *cls = arch_register_get_class(reg);
652 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
653 ir_fprintf(stderr, "Adding %+F\n", regnode);
655 ir_node *regnode = new_rd_Unknown(irg, arch_register_class_mode(cls));
656 obstack_ptr_grow(obst, regnode);
658 count = n_low_args + pset_count(states);
660 in = obstack_finish(obst);
662 if(env->call->flags.bits.call_has_imm && get_irn_opcode(call_ptr) == iro_SymConst) {
663 low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
665 curr_res_proj + pset_count(caller_save), count,
666 in, get_Call_type(irn));
667 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
669 low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
671 curr_res_proj + pset_count(caller_save),
672 count, in, get_Call_type(irn));
674 ARR_APP1(ir_node*, env->calls, low_call);
677 Set the register class of the call address to
678 the backend provided class (default: stack pointer class)
680 be_node_set_reg_class(low_call, be_pos_Call_ptr, call->cls_addr);
682 DBG((env->dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
684 /* Set the register classes and constraints of the Call parameters. */
685 for(i = 0; i < n_low_args; ++i) {
686 int index = low_args[i];
687 be_abi_call_arg_t *arg = get_call_arg(call, 0, index);
688 assert(arg->reg != NULL);
690 be_set_constr_single_reg(low_call, be_pos_Call_first_arg + index, arg->reg);
693 /* Set the register constraints of the results. */
694 for (i = 0; res_projs[i]; ++i) {
695 int pn = get_Proj_proj(res_projs[i]);
697 /* Correct Proj number since it has been adjusted! (see above) */
698 const be_abi_call_arg_t *arg = get_call_arg(call, 1, pn - pn_Call_max);
700 /* Matze: we need the information about the real mode for later
701 * transforms (signed/unsigend compares, stores...), so leave the fixup
702 * for the backend transform phase... */
705 const arch_register_class_t *cls = arch_register_get_class(arg->reg);
706 ir_mode *mode = arch_register_class_mode(cls);
707 set_irn_mode(irn, mode);
711 be_set_constr_single_reg(low_call, BE_OUT_POS(pn), arg->reg);
712 arch_set_irn_register(arch_env, res_projs[i], arg->reg);
714 obstack_free(obst, in);
715 exchange(irn, low_call);
717 /* redirect the result projs to the lowered call instead of the Proj_T */
718 for (i = 0; res_projs[i]; ++i)
719 set_Proj_pred(res_projs[i], low_call);
721 /* set the now unnecessary projT to bad */
722 if(res_proj != NULL) {
723 be_kill_node(res_proj);
726 /* Make additional projs for the caller save registers
727 and the Keep node which keeps them alive. */
728 if (pset_count(caller_save) > 0) {
729 const arch_register_t *reg;
733 for (reg = pset_first(caller_save), n = 0; reg; reg = pset_next(caller_save), ++n) {
734 ir_node *proj = new_r_Proj(irg, bl, low_call, reg->reg_class->mode, curr_res_proj);
736 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
737 be_set_constr_single_reg(low_call, BE_OUT_POS(curr_res_proj), reg);
739 /* a call can produce ignore registers, in this case set the flag and register for the Proj */
740 if (arch_register_type_is(reg, ignore)) {
741 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
742 be_node_set_flags(low_call, BE_OUT_POS(curr_res_proj), arch_irn_flags_ignore);
745 set_irn_link(proj, (void *) reg);
746 obstack_ptr_grow(obst, proj);
750 /* create the Keep for the caller save registers */
751 in = (ir_node **) obstack_finish(obst);
752 keep = be_new_Keep(NULL, irg, bl, n, in);
753 for (i = 0; i < n; ++i) {
754 const arch_register_t *reg = get_irn_link(in[i]);
755 be_node_set_reg_class(keep, i, reg->reg_class);
757 obstack_free(obst, in);
760 /* Clean up the stack. */
762 ir_node *mem_proj = NULL;
764 foreach_out_edge(low_call, edge) {
765 ir_node *irn = get_edge_src_irn(edge);
766 if(is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
773 mem_proj = new_r_Proj(irg, bl, low_call, mode_M, pn_Call_M);
774 keep_alive(mem_proj);
777 /* Clean up the stack frame if we allocated it */
779 curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, -stack_size);
780 add_irn_dep(curr_sp, mem_proj);
782 add_irn_dep(curr_sp, alloca_copy);
788 be_abi_call_free(call);
789 obstack_free(obst, pos);
792 del_pset(caller_save);
798 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
800 * @param alignment the minimum stack alignment
801 * @param size the node containing the non-aligned size
802 * @param irg the irg where new nodes are allocated on
803 * @param irg the block where new nodes are allocated on
804 * @param dbg debug info for new nodes
806 * @return a node representing the aligned size
808 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size, ir_graph *irg, ir_node *block, dbg_info *dbg) {
809 if (stack_alignment > 1) {
810 ir_mode *mode = get_irn_mode(size);
811 tarval *tv = new_tarval_from_long(stack_alignment-1, mode);
812 ir_node *mask = new_r_Const(irg, block, mode, tv);
814 size = new_rd_Add(dbg, irg, block, size, mask, mode);
816 tv = new_tarval_from_long(-(long)stack_alignment, mode);
817 mask = new_r_Const(irg, block, mode, tv);
818 size = new_rd_And(dbg, irg, block, size, mask, mode);
824 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
826 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp, ir_node **result_copy)
835 const ir_edge_t *edge;
836 ir_node *new_alloc, *size, *addr, *copy, *ins[2];
837 unsigned stack_alignment;
839 if (get_Alloc_where(alloc) != stack_alloc) {
844 block = get_nodes_block(alloc);
845 irg = get_irn_irg(block);
848 type = get_Alloc_type(alloc);
850 foreach_out_edge(alloc, edge) {
851 ir_node *irn = get_edge_src_irn(edge);
853 assert(is_Proj(irn));
854 switch(get_Proj_proj(irn)) {
866 /* Beware: currently Alloc nodes without a result might happen,
867 only escape analysis kills them and this phase runs only for object
868 oriented source. We kill the Alloc here. */
869 if (alloc_res == NULL && alloc_mem) {
870 exchange(alloc_mem, get_Alloc_mem(alloc));
874 dbg = get_irn_dbg_info(alloc);
876 /* we might need to multiply the size with the element size */
877 if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
878 tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
879 ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
880 ir_node *mul = new_rd_Mul(dbg, irg, block, get_Alloc_size(alloc),
884 size = get_Alloc_size(alloc);
887 /* The stack pointer will be modified in an unknown manner.
888 We cannot omit it. */
889 env->call->flags.bits.try_omit_fp = 0;
891 /* FIXME: size must be here round up for the stack alignment, but
892 this must be transmitted from the backend. */
894 size = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
895 new_alloc = be_new_AddSP(env->isa->sp, irg, block, curr_sp, size);
896 set_irn_dbg_info(new_alloc, dbg);
898 if(alloc_mem != NULL) {
902 addsp_mem = new_r_Proj(irg, block, new_alloc, mode_M, pn_be_AddSP_M);
904 /* We need to sync the output mem of the AddSP with the input mem
905 edge into the alloc node. */
906 ins[0] = get_Alloc_mem(alloc);
908 sync = new_r_Sync(irg, block, 2, ins);
910 exchange(alloc_mem, sync);
913 exchange(alloc, new_alloc);
915 /* fix projnum of alloca res */
916 set_Proj_proj(alloc_res, pn_be_AddSP_res);
918 addr = env->isa->stack_dir < 0 ? alloc_res : curr_sp;
920 /* copy the address away, since it could be used after further stack pointer modifications. */
921 /* Let it point curr_sp just for the moment, I'll reroute it in a second. */
922 *result_copy = copy = be_new_Copy(env->isa->sp->reg_class, irg, block, curr_sp);
924 /* Let all users of the Alloc() result now point to the copy. */
925 edges_reroute(alloc_res, copy, irg);
927 /* Rewire the copy appropriately. */
928 set_irn_n(copy, be_pos_Copy_op, addr);
937 * The Free is transformed into a back end free node and connected to the stack nodes.
939 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
943 ir_node *subsp, *mem, *res, *size, *sync;
947 unsigned stack_alignment;
950 if (get_Free_where(free) != stack_alloc) {
955 block = get_nodes_block(free);
956 irg = get_irn_irg(block);
957 type = get_Free_type(free);
958 sp_mode = env->isa->sp->reg_class->mode;
959 dbg = get_irn_dbg_info(free);
961 /* we might need to multiply the size with the element size */
962 if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
963 tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
964 ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
965 ir_node *mul = new_rd_Mul(dbg, irg, block, get_Free_size(free),
969 size = get_Free_size(free);
972 /* FIXME: size must be here round up for the stack alignment, but
973 this must be transmitted from the backend. */
975 size = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
977 /* The stack pointer will be modified in an unknown manner.
978 We cannot omit it. */
979 env->call->flags.bits.try_omit_fp = 0;
980 subsp = be_new_SubSP(env->isa->sp, irg, block, curr_sp, size);
981 set_irn_dbg_info(subsp, dbg);
983 mem = new_r_Proj(irg, block, subsp, mode_M, pn_be_SubSP_M);
984 res = new_r_Proj(irg, block, subsp, sp_mode, pn_be_SubSP_res);
986 /* we need to sync the memory */
987 in[0] = get_Free_mem(free);
989 sync = new_r_Sync(irg, block, 2, in);
991 /* and make the AddSP dependent on the former memory */
992 add_irn_dep(subsp, get_Free_mem(free));
995 exchange(free, sync);
1001 /* the following function is replaced by the usage of the heights module */
1004 * Walker for dependent_on().
1005 * This function searches a node tgt recursively from a given node
1006 * but is restricted to the given block.
1007 * @return 1 if tgt was reachable from curr, 0 if not.
1009 static int check_dependence(ir_node *curr, ir_node *tgt, ir_node *bl)
1013 if (get_nodes_block(curr) != bl)
1019 /* Phi functions stop the recursion inside a basic block */
1020 if (! is_Phi(curr)) {
1021 for(i = 0, n = get_irn_arity(curr); i < n; ++i) {
1022 if (check_dependence(get_irn_n(curr, i), tgt, bl))
1032 * Check if a node is somehow data dependent on another one.
1033 * both nodes must be in the same basic block.
1034 * @param n1 The first node.
1035 * @param n2 The second node.
1036 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
1038 static int dependent_on(ir_node *n1, ir_node *n2)
1040 assert(get_nodes_block(n1) == get_nodes_block(n2));
1042 return heights_reachable_in_block(ir_heights, n1, n2);
1045 static int cmp_call_dependecy(const void *c1, const void *c2)
1047 ir_node *n1 = *(ir_node **) c1;
1048 ir_node *n2 = *(ir_node **) c2;
1051 Classical qsort() comparison function behavior:
1052 0 if both elements are equal
1053 1 if second is "smaller" that first
1054 -1 if first is "smaller" that second
1056 if (dependent_on(n1, n2))
1059 if (dependent_on(n2, n1))
1066 * Walker: links all Call/alloc/Free nodes to the Block they are contained.
1068 static void link_calls_in_block_walker(ir_node *irn, void *data)
1070 ir_opcode code = get_irn_opcode(irn);
1072 if (code == iro_Call ||
1073 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
1074 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
1075 be_abi_irg_t *env = data;
1076 ir_node *bl = get_nodes_block(irn);
1077 void *save = get_irn_link(bl);
1079 if (code == iro_Call)
1080 env->call->flags.bits.irg_is_leaf = 0;
1082 set_irn_link(irn, save);
1083 set_irn_link(bl, irn);
1089 * Process all Call nodes inside a basic block.
1090 * Note that the link field of the block must contain a linked list of all
1091 * Call nodes inside the Block. We first order this list according to data dependency
1092 * and that connect the calls together.
1094 static void process_calls_in_block(ir_node *bl, void *data)
1096 be_abi_irg_t *env = data;
1097 ir_node *curr_sp = env->init_sp;
1101 for(irn = get_irn_link(bl), n = 0; irn; irn = get_irn_link(irn), ++n)
1102 obstack_ptr_grow(&env->obst, irn);
1104 /* If there were call nodes in the block. */
1108 ir_node *copy = NULL;
1111 nodes = obstack_finish(&env->obst);
1113 /* order the call nodes according to data dependency */
1114 qsort(nodes, n, sizeof(nodes[0]), cmp_call_dependecy);
1116 for(i = n - 1; i >= 0; --i) {
1117 ir_node *irn = nodes[i];
1119 DBG((env->dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1120 switch(get_irn_opcode(irn)) {
1122 curr_sp = adjust_call(env, irn, curr_sp, copy);
1125 curr_sp = adjust_alloc(env, irn, curr_sp, ©);
1128 curr_sp = adjust_free(env, irn, curr_sp);
1135 obstack_free(&env->obst, nodes);
1137 /* Keep the last stack state in the block by tying it to Keep node */
1139 keep = be_new_Keep(env->isa->sp->reg_class, get_irn_irg(bl), bl, 1, nodes);
1140 pmap_insert(env->keep_map, bl, keep);
1143 set_irn_link(bl, curr_sp);
1144 } /* process_calls_in_block */
1147 * Adjust all call nodes in the graph to the ABI conventions.
1149 static void process_calls(be_abi_irg_t *env)
1151 ir_graph *irg = env->birg->irg;
1153 env->call->flags.bits.irg_is_leaf = 1;
1154 irg_walk_graph(irg, firm_clear_link, link_calls_in_block_walker, env);
1156 ir_heights = heights_new(env->birg->irg);
1157 irg_block_walk_graph(irg, NULL, process_calls_in_block, env);
1158 heights_free(ir_heights);
1162 static ir_node *setup_frame(be_abi_irg_t *env)
1164 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1165 const arch_register_t *sp = isa->sp;
1166 const arch_register_t *bp = isa->bp;
1167 be_abi_call_flags_bits_t flags = env->call->flags.bits;
1168 ir_graph *irg = env->birg->irg;
1169 ir_node *bl = get_irg_start_block(irg);
1170 ir_node *no_mem = get_irg_no_mem(irg);
1171 ir_node *old_frame = get_irg_frame(irg);
1172 ir_node *stack = pmap_get(env->regs, (void *) sp);
1173 ir_node *frame = pmap_get(env->regs, (void *) bp);
1175 int stack_nr = get_Proj_proj(stack);
1177 if(flags.try_omit_fp) {
1178 stack = be_new_IncSP(sp, irg, bl, stack, no_mem, BE_STACK_FRAME_SIZE_EXPAND);
1183 frame = be_new_Copy(bp->reg_class, irg, bl, stack);
1185 be_node_set_flags(frame, -1, arch_irn_flags_dont_spill);
1186 if(!flags.fp_free) {
1187 be_set_constr_single_reg(frame, -1, bp);
1188 be_node_set_flags(frame, -1, arch_irn_flags_ignore);
1189 arch_set_irn_register(env->birg->main_env->arch_env, frame, bp);
1192 stack = be_new_IncSP(sp, irg, bl, stack, frame, BE_STACK_FRAME_SIZE_EXPAND);
1195 be_node_set_flags(env->reg_params, -(stack_nr + 1), arch_irn_flags_ignore);
1196 env->init_sp = stack;
1197 set_irg_frame(irg, frame);
1198 edges_reroute(old_frame, frame, irg);
1203 static void clearup_frame(be_abi_irg_t *env, ir_node *ret, pmap *reg_map, struct obstack *obst)
1205 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1206 const arch_register_t *sp = isa->sp;
1207 const arch_register_t *bp = isa->bp;
1208 ir_graph *irg = env->birg->irg;
1209 ir_node *ret_mem = get_Return_mem(ret);
1210 ir_node *frame = get_irg_frame(irg);
1211 ir_node *bl = get_nodes_block(ret);
1212 ir_node *stack = get_irn_link(bl);
1216 if(env->call->flags.bits.try_omit_fp) {
1217 stack = be_new_IncSP(sp, irg, bl, stack, ret_mem, -BE_STACK_FRAME_SIZE_SHRINK);
1221 stack = be_new_SetSP(sp, irg, bl, stack, frame, ret_mem);
1222 be_set_constr_single_reg(stack, -1, sp);
1223 be_node_set_flags(stack, -1, arch_irn_flags_ignore);
1226 pmap_foreach(env->regs, ent) {
1227 const arch_register_t *reg = ent->key;
1228 ir_node *irn = ent->value;
1231 obstack_ptr_grow(&env->obst, stack);
1233 obstack_ptr_grow(&env->obst, frame);
1234 else if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
1235 obstack_ptr_grow(obst, irn);
1242 * Computes the stack argument layout type.
1243 * Changes a possibly allocated value param type by moving
1244 * entities to the stack layout type.
1246 * @param env the ABI environment
1247 * @param call the current call ABI
1248 * @param method_type the method type
1249 * @param param_map an array mapping method arguments to the stack layout type
1251 * @return the stack argument layout type
1253 static ir_type *compute_arg_type(be_abi_irg_t *env, be_abi_call_t *call, ir_type *method_type, ir_entity ***param_map)
1255 int dir = env->call->flags.bits.left_to_right ? 1 : -1;
1256 int inc = env->birg->main_env->arch_env->isa->stack_dir * dir;
1257 int n = get_method_n_params(method_type);
1258 int curr = inc > 0 ? 0 : n - 1;
1264 ir_type *val_param_tp = get_method_value_param_type(method_type);
1265 ident *id = get_entity_ident(get_irg_entity(env->birg->irg));
1268 *param_map = map = obstack_alloc(&env->obst, n * sizeof(ir_entity *));
1269 res = new_type_struct(mangle_u(id, new_id_from_chars("arg_type", 8)));
1270 for (i = 0; i < n; ++i, curr += inc) {
1271 ir_type *param_type = get_method_param_type(method_type, curr);
1272 be_abi_call_arg_t *arg = get_call_arg(call, 0, curr);
1275 if (arg->on_stack) {
1277 /* the entity was already created, move it to the param type */
1278 arg->stack_ent = get_method_value_param_ent(method_type, i);
1279 remove_struct_member(val_param_tp, arg->stack_ent);
1280 set_entity_owner(arg->stack_ent, res);
1281 add_struct_member(res, arg->stack_ent);
1282 /* must be automatic to set a fixed layout */
1283 set_entity_allocation(arg->stack_ent, allocation_automatic);
1286 snprintf(buf, sizeof(buf), "param_%d", i);
1287 arg->stack_ent = new_entity(res, new_id_from_str(buf), param_type);
1289 ofs += arg->space_before;
1290 ofs = round_up2(ofs, arg->alignment);
1291 set_entity_offset(arg->stack_ent, ofs);
1292 ofs += arg->space_after;
1293 ofs += get_type_size_bytes(param_type);
1294 map[i] = arg->stack_ent;
1297 set_type_size_bytes(res, ofs);
1298 set_type_state(res, layout_fixed);
1303 static void create_register_perms(const arch_isa_t *isa, ir_graph *irg, ir_node *bl, pmap *regs)
1306 struct obstack obst;
1308 obstack_init(&obst);
1310 /* Create a Perm after the RegParams node to delimit it. */
1311 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
1312 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1317 for(n_regs = 0, j = 0; j < cls->n_regs; ++j) {
1318 const arch_register_t *reg = &cls->regs[j];
1319 ir_node *irn = pmap_get(regs, (void *) reg);
1321 if(irn && !arch_register_type_is(reg, ignore)) {
1323 obstack_ptr_grow(&obst, irn);
1324 set_irn_link(irn, (void *) reg);
1328 obstack_ptr_grow(&obst, NULL);
1329 in = obstack_finish(&obst);
1331 perm = be_new_Perm(cls, irg, bl, n_regs, in);
1332 for(j = 0; j < n_regs; ++j) {
1333 ir_node *arg = in[j];
1334 arch_register_t *reg = get_irn_link(arg);
1335 pmap_insert(regs, reg, arg);
1336 be_set_constr_single_reg(perm, BE_OUT_POS(j), reg);
1339 obstack_free(&obst, in);
1342 obstack_free(&obst, NULL);
1347 const arch_register_t *reg;
1351 static int cmp_regs(const void *a, const void *b)
1353 const reg_node_map_t *p = a;
1354 const reg_node_map_t *q = b;
1356 if(p->reg->reg_class == q->reg->reg_class)
1357 return p->reg->index - q->reg->index;
1359 return p->reg->reg_class - q->reg->reg_class;
1362 static reg_node_map_t *reg_map_to_arr(struct obstack *obst, pmap *reg_map)
1365 int n = pmap_count(reg_map);
1367 reg_node_map_t *res = obstack_alloc(obst, n * sizeof(res[0]));
1369 pmap_foreach(reg_map, ent) {
1370 res[i].reg = ent->key;
1371 res[i].irn = ent->value;
1375 qsort(res, n, sizeof(res[0]), cmp_regs);
1380 * Creates a barrier.
1382 static ir_node *create_barrier(be_abi_irg_t *env, ir_node *bl, ir_node **mem, pmap *regs, int in_req)
1384 ir_graph *irg = env->birg->irg;
1385 int n_regs = pmap_count(regs);
1391 rm = reg_map_to_arr(&env->obst, regs);
1393 for(n = 0; n < n_regs; ++n)
1394 obstack_ptr_grow(&env->obst, rm[n].irn);
1397 obstack_ptr_grow(&env->obst, *mem);
1401 in = (ir_node **) obstack_finish(&env->obst);
1402 irn = be_new_Barrier(irg, bl, n, in);
1403 obstack_free(&env->obst, in);
1405 for(n = 0; n < n_regs; ++n) {
1406 const arch_register_t *reg = rm[n].reg;
1408 int pos = BE_OUT_POS(n);
1411 proj = new_r_Proj(irg, bl, irn, get_irn_mode(rm[n].irn), n);
1412 be_node_set_reg_class(irn, n, reg->reg_class);
1414 be_set_constr_single_reg(irn, n, reg);
1415 be_set_constr_single_reg(irn, pos, reg);
1416 be_node_set_reg_class(irn, pos, reg->reg_class);
1417 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
1419 /* if the proj projects a ignore register or a node which is set to ignore, propagate this property. */
1420 if(arch_register_type_is(reg, ignore) || arch_irn_is(env->birg->main_env->arch_env, in[n], ignore))
1421 flags |= arch_irn_flags_ignore;
1423 if(arch_irn_is(env->birg->main_env->arch_env, in[n], modify_sp))
1424 flags |= arch_irn_flags_modify_sp;
1426 be_node_set_flags(irn, pos, flags);
1428 pmap_insert(regs, (void *) reg, proj);
1432 *mem = new_r_Proj(irg, bl, irn, mode_M, n);
1435 obstack_free(&env->obst, rm);
1440 * Creates a be_Return for a Return node.
1442 * @param @env the abi environment
1443 * @param irn the Return node or NULL if there was none
1444 * @param bl the block where the be_Retun should be placed
1445 * @param mem the current memory
1446 * @param n_res number of return results
1448 static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, ir_node *mem, int n_res) {
1449 be_abi_call_t *call = env->call;
1450 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1452 pmap *reg_map = pmap_create();
1453 ir_node *keep = pmap_get(env->keep_map, bl);
1459 const arch_register_t **regs;
1463 get the valid stack node in this block.
1464 If we had a call in that block there is a Keep constructed by process_calls()
1465 which points to the last stack modification in that block. we'll use
1466 it then. Else we use the stack from the start block and let
1467 the ssa construction fix the usage.
1469 stack = be_abi_reg_map_get(env->regs, isa->sp);
1471 ir_node *bad = new_r_Bad(env->birg->irg);
1472 stack = get_irn_n(keep, 0);
1473 set_nodes_block(keep, bad);
1474 set_irn_n(keep, 0, bad);
1475 // exchange(keep, new_r_Bad(env->birg->irg));
1478 /* Insert results for Return into the register map. */
1479 for(i = 0; i < n_res; ++i) {
1480 ir_node *res = get_Return_res(irn, i);
1481 be_abi_call_arg_t *arg = get_call_arg(call, 1, i);
1482 assert(arg->in_reg && "return value must be passed in register");
1483 pmap_insert(reg_map, (void *) arg->reg, res);
1486 /* Add uses of the callee save registers. */
1487 pmap_foreach(env->regs, ent) {
1488 const arch_register_t *reg = ent->key;
1489 if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
1490 pmap_insert(reg_map, ent->key, ent->value);
1493 be_abi_reg_map_set(reg_map, isa->sp, stack);
1495 /* Make the Epilogue node and call the arch's epilogue maker. */
1496 create_barrier(env, bl, &mem, reg_map, 1);
1497 call->cb->epilogue(env->cb, bl, &mem, reg_map);
1500 Maximum size of the in array for Return nodes is
1501 return args + callee save/ignore registers + memory + stack pointer
1503 in_max = pmap_count(reg_map) + n_res + 2;
1505 in = obstack_alloc(&env->obst, in_max * sizeof(in[0]));
1506 regs = obstack_alloc(&env->obst, in_max * sizeof(regs[0]));
1509 in[1] = be_abi_reg_map_get(reg_map, isa->sp);
1514 /* clear SP entry, since it has already been grown. */
1515 pmap_insert(reg_map, (void *) isa->sp, NULL);
1516 for(i = 0; i < n_res; ++i) {
1517 be_abi_call_arg_t *arg = get_call_arg(call, 1, i);
1519 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1520 regs[n++] = arg->reg;
1522 /* Clear the map entry to mark the register as processed. */
1523 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1526 /* grow the rest of the stuff. */
1527 pmap_foreach(reg_map, ent) {
1530 regs[n++] = ent->key;
1534 /* The in array for the new back end return is now ready. */
1535 ret = be_new_Return(irn ? get_irn_dbg_info(irn) : NULL, env->birg->irg, bl, n_res, n, in);
1537 /* Set the register classes of the return's parameter accordingly. */
1538 for(i = 0; i < n; ++i)
1540 be_node_set_reg_class(ret, i, regs[i]->reg_class);
1542 /* Free the space of the Epilog's in array and the register <-> proj map. */
1543 obstack_free(&env->obst, in);
1544 pmap_destroy(reg_map);
1549 typedef struct lower_frame_sels_env_t {
1551 ir_entity *value_param_list; /**< the list of all value param entities */
1552 } lower_frame_sels_env_t;
1555 * Walker: Replaces Sels of frame type and
1556 * value param type entities by FrameAddress.
1558 static void lower_frame_sels_walker(ir_node *irn, void *data)
1560 lower_frame_sels_env_t *ctx = data;
1563 ir_graph *irg = current_ir_graph;
1564 ir_node *frame = get_irg_frame(irg);
1565 ir_node *param_base = get_irg_value_param_base(irg);
1566 ir_node *ptr = get_Sel_ptr(irn);
1568 if (ptr == frame || ptr == param_base) {
1569 be_abi_irg_t *env = ctx->env;
1570 ir_entity *ent = get_Sel_entity(irn);
1571 ir_node *bl = get_nodes_block(irn);
1574 nw = be_new_FrameAddr(env->isa->sp->reg_class, irg, bl, frame, ent);
1577 /* check, if it's a param sel and if have not seen this entity immediatly before */
1578 if (ptr == param_base && ctx->value_param_list != ent) {
1579 set_entity_link(ent, ctx->value_param_list);
1580 ctx->value_param_list = ent;
1587 * Check if a value parameter is transmitted as a register.
1588 * This might happen if the address of an parameter is taken which is
1589 * transmitted in registers.
1591 * Note that on some architectures this case must be handled specially
1592 * because the place of the backing store is determined by their ABI.
1594 * In the default case we move the entity to the frame type and create
1595 * a backing store into the first block.
1597 static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_entity *value_param_list) {
1598 be_abi_call_t *call = env->call;
1599 ir_graph *irg = env->birg->irg;
1600 ir_entity *ent, *next_ent, *new_list;
1602 DEBUG_ONLY(firm_dbg_module_t *dbg = env->dbg;)
1605 for (ent = value_param_list; ent; ent = next_ent) {
1606 int i = get_struct_member_index(get_entity_owner(ent), ent);
1607 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
1609 next_ent = get_entity_link(ent);
1611 DBG((dbg, LEVEL_2, "\targ #%d need backing store\n", i));
1612 set_entity_link(ent, new_list);
1617 /* ok, change the graph */
1618 ir_node *start_bl = get_irg_start_block(irg);
1619 ir_node *first_bl = NULL;
1620 ir_node *frame, *imem, *nmem, *store, *mem, *args, *args_bl;
1621 const ir_edge_t *edge;
1622 optimization_state_t state;
1625 foreach_block_succ(start_bl, edge) {
1626 ir_node *succ = get_edge_src_irn(edge);
1627 if (start_bl != succ) {
1633 /* we had already removed critical edges, so the following
1634 assertion should be always true. */
1635 assert(get_Block_n_cfgpreds(first_bl) == 1);
1637 /* now create backing stores */
1638 frame = get_irg_frame(irg);
1639 imem = get_irg_initial_mem(irg);
1641 save_optimization_state(&state);
1643 nmem = new_r_Proj(irg, first_bl, get_irg_start(irg), mode_M, pn_Start_M);
1644 restore_optimization_state(&state);
1646 /* reroute all edges to the new memory source */
1647 edges_reroute(imem, nmem, irg);
1651 args = get_irg_args(irg);
1652 args_bl = get_nodes_block(args);
1653 for (ent = new_list; ent; ent = get_entity_link(ent)) {
1654 int i = get_struct_member_index(get_entity_owner(ent), ent);
1655 ir_type *tp = get_entity_type(ent);
1656 ir_mode *mode = get_type_mode(tp);
1659 /* address for the backing store */
1660 addr = be_new_FrameAddr(env->isa->sp->reg_class, irg, first_bl, frame, ent);
1663 mem = new_r_Proj(irg, first_bl, store, mode_M, pn_Store_M);
1665 /* the backing store itself */
1666 store = new_r_Store(irg, first_bl, mem, addr,
1667 new_r_Proj(irg, args_bl, args, mode, i));
1669 /* the new memory Proj gets the last Proj from store */
1670 set_Proj_pred(nmem, store);
1671 set_Proj_proj(nmem, pn_Store_M);
1673 /* move all entities to the frame type */
1674 frame_tp = get_irg_frame_type(irg);
1675 offset = get_type_size_bytes(frame_tp);
1676 for (ent = new_list; ent; ent = get_entity_link(ent)) {
1677 ir_type *tp = get_entity_type(ent);
1678 int align = get_type_alignment_bytes(tp);
1680 offset += align - 1;
1682 set_entity_owner(ent, frame_tp);
1683 add_class_member(frame_tp, ent);
1684 /* must be automatic to set a fixed layout */
1685 set_entity_allocation(ent, allocation_automatic);
1686 set_entity_offset(ent, offset);
1687 offset += get_type_size_bytes(tp);
1689 set_type_size_bytes(frame_tp, offset);
1694 * The start block has no jump, instead it has an initial exec Proj.
1695 * The backend wants to handle all blocks the same way, so we replace
1696 * the out cfg edge with a real jump.
1698 static void fix_start_block(ir_node *block, void *env) {
1701 ir_node *start_block;
1704 /* we processed the start block, return */
1708 irg = get_irn_irg(block);
1709 start_block = get_irg_start_block(irg);
1711 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1712 ir_node *pred = get_Block_cfgpred(block, i);
1713 ir_node *pred_block = get_nodes_block(pred);
1715 /* ok, we are in the block, having start as cfg predecessor */
1716 if (pred_block == start_block) {
1717 ir_node *jump = new_r_Jmp(irg, pred_block);
1718 set_Block_cfgpred(block, i, jump);
1725 * Modify the irg itself and the frame type.
1727 static void modify_irg(be_abi_irg_t *env)
1729 be_abi_call_t *call = env->call;
1730 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1731 const arch_register_t *sp = arch_isa_sp(isa);
1732 ir_graph *irg = env->birg->irg;
1733 ir_node *bl = get_irg_start_block(irg);
1734 ir_node *end = get_irg_end_block(irg);
1735 ir_node *old_mem = get_irg_initial_mem(irg);
1736 ir_node *new_mem_proj;
1738 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1739 pset *dont_save = pset_new_ptr(8);
1745 const arch_register_t *fp_reg;
1746 ir_node *frame_pointer;
1748 ir_node *reg_params_bl;
1751 ir_node *value_param_base;
1752 const ir_edge_t *edge;
1753 ir_type *arg_type, *bet_type;
1754 lower_frame_sels_env_t ctx;
1755 ir_entity **param_map;
1757 bitset_t *used_proj_nr;
1758 DEBUG_ONLY(firm_dbg_module_t *dbg = env->dbg;)
1760 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1762 /* Convert the Sel nodes in the irg to frame load/store/addr nodes. */
1764 ctx.value_param_list = NULL;
1765 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1767 /* value_param_base anchor is not needed anymore now */
1768 value_param_base = get_irg_value_param_base(irg);
1769 be_kill_node(value_param_base);
1770 set_irg_value_param_base(irg, new_r_Bad(irg));
1772 env->frame = obstack_alloc(&env->obst, sizeof(env->frame[0]));
1773 env->regs = pmap_create();
1775 used_proj_nr = bitset_alloca(1024);
1776 n_params = get_method_n_params(method_type);
1777 args = obstack_alloc(&env->obst, n_params * sizeof(args[0]));
1778 memset(args, 0, n_params * sizeof(args[0]));
1780 /* Check if a value parameter is transmitted as a register.
1781 * This might happen if the address of an parameter is taken which is
1782 * transmitted in registers.
1784 * Note that on some architectures this case must be handled specially
1785 * because the place of the backing store is determined by their ABI.
1787 * In the default case we move the entity to the frame type and create
1788 * a backing store into the first block.
1790 fix_address_of_parameter_access(env, ctx.value_param_list);
1792 /* Fill the argument vector */
1793 arg_tuple = get_irg_args(irg);
1794 foreach_out_edge(arg_tuple, edge) {
1795 ir_node *irn = get_edge_src_irn(edge);
1796 int nr = get_Proj_proj(irn);
1798 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1801 arg_type = compute_arg_type(env, call, method_type, ¶m_map);
1802 bet_type = call->cb->get_between_type(env->cb);
1803 stack_frame_init(env->frame, arg_type, bet_type, get_irg_frame_type(irg), isa->stack_dir, param_map);
1805 /* Count the register params and add them to the number of Projs for the RegParams node */
1806 for(i = 0; i < n_params; ++i) {
1807 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
1808 if(arg->in_reg && args[i]) {
1809 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1810 assert(i == get_Proj_proj(args[i]));
1812 /* For now, associate the register with the old Proj from Start representing that argument. */
1813 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1814 bitset_set(used_proj_nr, i);
1815 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1819 /* Collect all callee-save registers */
1820 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
1821 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1822 for(j = 0; j < cls->n_regs; ++j) {
1823 const arch_register_t *reg = &cls->regs[j];
1824 if(arch_register_type_is(reg, callee_save) ||
1825 arch_register_type_is(reg, state)) {
1826 pmap_insert(env->regs, (void *) reg, NULL);
1831 pmap_insert(env->regs, (void *) sp, NULL);
1832 pmap_insert(env->regs, (void *) isa->bp, NULL);
1833 reg_params_bl = get_irg_start_block(irg);
1834 env->reg_params = be_new_RegParams(irg, reg_params_bl, pmap_count(env->regs));
1835 add_irn_dep(env->reg_params, get_irg_start(irg));
1838 * make proj nodes for the callee save registers.
1839 * memorize them, since Return nodes get those as inputs.
1841 * Note, that if a register corresponds to an argument, the regs map contains
1842 * the old Proj from start for that argument.
1845 rm = reg_map_to_arr(&env->obst, env->regs);
1846 for(i = 0, n = pmap_count(env->regs); i < n; ++i) {
1847 arch_register_t *reg = (void *) rm[i].reg;
1848 ir_mode *mode = reg->reg_class->mode;
1850 int pos = BE_OUT_POS((int) nr);
1856 bitset_set(used_proj_nr, nr);
1857 proj = new_r_Proj(irg, reg_params_bl, env->reg_params, mode, nr);
1858 pmap_insert(env->regs, (void *) reg, proj);
1859 be_set_constr_single_reg(env->reg_params, pos, reg);
1860 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
1863 * If the register is an ignore register,
1864 * The Proj for that register shall also be ignored during register allocation.
1866 if(arch_register_type_is(reg, ignore))
1867 flags |= arch_irn_flags_ignore;
1870 flags |= arch_irn_flags_modify_sp;
1872 be_node_set_flags(env->reg_params, pos, flags);
1874 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1876 obstack_free(&env->obst, rm);
1878 /* create a new initial memory proj */
1879 assert(is_Proj(old_mem));
1880 new_mem_proj = new_r_Proj(irg, get_nodes_block(old_mem),
1881 new_r_Unknown(irg, mode_T), mode_M,
1882 get_Proj_proj(old_mem));
1885 /* Generate the Prologue */
1886 fp_reg = call->cb->prologue(env->cb, &mem, env->regs);
1888 /* do the stack allocation BEFORE the barrier, or spill code
1889 might be added before it */
1890 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1891 env->init_sp = be_new_IncSP(sp, irg, bl, env->init_sp, BE_STACK_FRAME_SIZE_EXPAND);
1892 be_abi_reg_map_set(env->regs, sp, env->init_sp);
1894 env->start_barrier = barrier = create_barrier(env, bl, &mem, env->regs, 0);
1896 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1897 arch_set_irn_register(env->birg->main_env->arch_env, env->init_sp, sp);
1899 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1900 set_irg_frame(irg, frame_pointer);
1901 pset_insert_ptr(env->ignore_regs, fp_reg);
1903 /* rewire old mem users to new mem */
1904 set_Proj_pred(new_mem_proj, get_Proj_pred(old_mem));
1905 exchange(old_mem, mem);
1907 set_irg_initial_mem(irg, mem);
1909 /* Now, introduce stack param nodes for all parameters passed on the stack */
1910 for(i = 0; i < n_params; ++i) {
1911 ir_node *arg_proj = args[i];
1912 ir_node *repl = NULL;
1914 if(arg_proj != NULL) {
1915 be_abi_call_arg_t *arg;
1916 ir_type *param_type;
1917 int nr = get_Proj_proj(arg_proj);
1919 nr = MIN(nr, n_params);
1920 arg = get_call_arg(call, 0, nr);
1921 param_type = get_method_param_type(method_type, nr);
1924 repl = pmap_get(env->regs, (void *) arg->reg);
1927 else if(arg->on_stack) {
1928 /* For atomic parameters which are actually used, we create a StackParam node. */
1929 if(is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1930 ir_mode *mode = get_type_mode(param_type);
1931 const arch_register_class_t *cls = arch_isa_get_reg_class_for_mode(isa, mode);
1932 repl = be_new_StackParam(cls, isa->bp->reg_class, irg, reg_params_bl, mode, frame_pointer, arg->stack_ent);
1935 /* The stack parameter is not primitive (it is a struct or array),
1936 we thus will create a node representing the parameter's address
1939 repl = be_new_FrameAddr(sp->reg_class, irg, reg_params_bl, frame_pointer, arg->stack_ent);
1943 assert(repl != NULL);
1944 exchange(args[i], repl);
1948 /* the arg proj is not needed anymore now */
1949 assert(get_irn_n_edges(arg_tuple) == 0);
1950 be_kill_node(arg_tuple);
1951 set_irg_args(irg, new_rd_Bad(irg));
1953 /* All Return nodes hang on the End node, so look for them there. */
1954 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1955 ir_node *irn = get_Block_cfgpred(end, i);
1957 if (is_Return(irn)) {
1958 ir_node *ret = create_be_return(env, irn, get_nodes_block(irn), get_Return_mem(irn), get_Return_n_ress(irn));
1962 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1963 the code is dead and will never be executed. */
1965 del_pset(dont_save);
1966 obstack_free(&env->obst, args);
1968 /* handle start block here (place a jump in the block) */
1970 irg_block_walk_graph(irg, fix_start_block, NULL, &temp);
1973 /** Fix the state inputs of calls that still hang on unknowns */
1975 void fix_call_state_inputs(be_abi_irg_t *env)
1977 const arch_isa_t *isa = env->isa;
1979 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1981 /* Collect caller save registers */
1982 n = arch_isa_get_n_reg_class(isa);
1983 for(i = 0; i < n; ++i) {
1985 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1986 for(j = 0; j < cls->n_regs; ++j) {
1987 const arch_register_t *reg = arch_register_for_index(cls, j);
1988 if(arch_register_type_is(reg, state)) {
1989 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1994 n = ARR_LEN(env->calls);
1995 n_states = ARR_LEN(stateregs);
1996 for(i = 0; i < n; ++i) {
1998 ir_node *call = env->calls[i];
2000 arity = get_irn_arity(call);
2002 /* the statereg inputs are the last n inputs of the calls */
2003 for(s = 0; s < n_states; ++s) {
2004 int inp = arity - n_states + s;
2005 const arch_register_t *reg = stateregs[s];
2006 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
2008 set_irn_n(call, inp, regnode);
2013 be_abi_irg_t *be_abi_introduce(be_irg_t *birg)
2015 be_abi_irg_t *env = xmalloc(sizeof(env[0]));
2016 ir_node *old_frame = get_irg_frame(birg->irg);
2017 ir_graph *irg = birg->irg;
2021 optimization_state_t state;
2022 unsigned *limited_bitset;
2024 be_omit_fp = birg->main_env->options->omit_fp;
2026 obstack_init(&env->obst);
2028 env->isa = birg->main_env->arch_env->isa;
2029 env->method_type = get_entity_type(get_irg_entity(irg));
2030 env->call = be_abi_call_new(env->isa->sp->reg_class);
2031 arch_isa_get_call_abi(env->isa, env->method_type, env->call);
2033 env->ignore_regs = pset_new_ptr_default();
2034 env->keep_map = pmap_create();
2035 env->dce_survivor = new_survive_dce();
2038 env->sp_req.type = arch_register_req_type_limited;
2039 env->sp_req.cls = arch_register_get_class(env->isa->sp);
2040 limited_bitset = rbitset_obstack_alloc(&env->obst, env->sp_req.cls->n_regs);
2041 rbitset_set(limited_bitset, arch_register_get_index(env->isa->sp));
2042 env->sp_req.limited = limited_bitset;
2044 env->sp_cls_req.type = arch_register_req_type_normal;
2045 env->sp_cls_req.cls = arch_register_get_class(env->isa->sp);
2047 /* Beware: later we replace this node by the real one, ensure it is not CSE'd
2048 to another Unknown or the stack pointer gets used */
2049 save_optimization_state(&state);
2051 env->init_sp = dummy = new_r_Unknown(irg, env->isa->sp->reg_class->mode);
2052 restore_optimization_state(&state);
2053 FIRM_DBG_REGISTER(env->dbg, "firm.be.abi");
2055 env->calls = NEW_ARR_F(ir_node*, 0);
2057 /* Lower all call nodes in the IRG. */
2061 Beware: init backend abi call object after processing calls,
2062 otherwise some information might be not yet available.
2064 env->cb = env->call->cb->init(env->call, birg->main_env->arch_env, irg);
2066 /* Process the IRG */
2069 /* fix call inputs for state registers */
2070 fix_call_state_inputs(env);
2072 /* We don't need the keep map anymore. */
2073 pmap_destroy(env->keep_map);
2075 /* calls array is not needed anymore */
2076 DEL_ARR_F(env->calls);
2078 /* reroute the stack origin of the calls to the true stack origin. */
2079 exchange(dummy, env->init_sp);
2080 exchange(old_frame, get_irg_frame(irg));
2082 /* Make some important node pointers survive the dead node elimination. */
2083 survive_dce_register_irn(env->dce_survivor, &env->init_sp);
2084 pmap_foreach(env->regs, ent) {
2085 survive_dce_register_irn(env->dce_survivor, (ir_node **) &ent->value);
2088 env->call->cb->done(env->cb);
2093 void be_abi_free(be_abi_irg_t *env)
2095 be_abi_call_free(env->call);
2096 free_survive_dce(env->dce_survivor);
2097 del_pset(env->ignore_regs);
2098 pmap_destroy(env->regs);
2099 obstack_free(&env->obst, NULL);
2103 void be_abi_put_ignore_regs(be_abi_irg_t *abi, const arch_register_class_t *cls, bitset_t *bs)
2105 arch_register_t *reg;
2107 for(reg = pset_first(abi->ignore_regs); reg; reg = pset_next(abi->ignore_regs))
2108 if(reg->reg_class == cls)
2109 bitset_set(bs, reg->index);
2112 /* Returns the stack layout from a abi environment. */
2113 const be_stack_layout_t *be_abi_get_stack_layout(const be_abi_irg_t *abi) {
2120 | ___(_)_ __ / ___|| |_ __ _ ___| | __
2121 | |_ | \ \/ / \___ \| __/ _` |/ __| |/ /
2122 | _| | |> < ___) | || (_| | (__| <
2123 |_| |_/_/\_\ |____/ \__\__,_|\___|_|\_\
2127 typedef ir_node **node_array;
2129 typedef struct fix_stack_walker_env_t {
2130 node_array sp_nodes;
2131 const arch_env_t *arch_env;
2132 } fix_stack_walker_env_t;
2135 * Walker. Collect all stack modifying nodes.
2137 static void collect_stack_nodes_walker(ir_node *node, void *data)
2139 fix_stack_walker_env_t *env = data;
2141 if (arch_irn_is(env->arch_env, node, modify_sp)) {
2142 assert(get_irn_mode(node) != mode_M && get_irn_mode(node) != mode_T);
2143 ARR_APP1(ir_node*, env->sp_nodes, node);
2147 void be_abi_fix_stack_nodes(be_abi_irg_t *env)
2149 be_ssa_construction_env_t senv;
2152 be_irg_t *birg = env->birg;
2153 be_lv_t *lv = be_get_birg_liveness(birg);
2154 fix_stack_walker_env_t walker_env;
2157 walker_env.sp_nodes = NEW_ARR_F(ir_node*, 0);
2158 walker_env.arch_env = birg->main_env->arch_env;
2159 isa = walker_env.arch_env->isa;
2161 irg_walk_graph(birg->irg, collect_stack_nodes_walker, NULL, &walker_env);
2163 /* nothing to be done if we didn't find any node, in fact we mustn't
2164 * continue, as for endless loops incsp might have had no users and is bad
2167 len = ARR_LEN(walker_env.sp_nodes);
2169 DEL_ARR_F(walker_env.sp_nodes);
2173 be_ssa_construction_init(&senv, birg);
2174 be_ssa_construction_add_copies(&senv, walker_env.sp_nodes,
2175 ARR_LEN(walker_env.sp_nodes));
2176 be_ssa_construction_fix_users_array(&senv, walker_env.sp_nodes,
2177 ARR_LEN(walker_env.sp_nodes));
2180 len = ARR_LEN(walker_env.sp_nodes);
2181 for(i = 0; i < len; ++i) {
2182 be_liveness_update(lv, walker_env.sp_nodes[i]);
2184 be_ssa_construction_update_liveness_phis(&senv, lv);
2187 phis = be_ssa_construction_get_new_phis(&senv);
2189 /* set register requirements for stack phis */
2190 len = ARR_LEN(phis);
2191 for(i = 0; i < len; ++i) {
2192 ir_node *phi = phis[i];
2193 be_set_phi_reg_req(walker_env.arch_env, phi, &env->sp_req);
2194 be_set_phi_flags(walker_env.arch_env, phi, arch_irn_flags_ignore | arch_irn_flags_modify_sp);
2195 arch_set_irn_register(walker_env.arch_env, phi, env->isa->sp);
2197 be_ssa_construction_destroy(&senv);
2199 DEL_ARR_F(walker_env.sp_nodes);
2202 static int process_stack_bias(be_abi_irg_t *env, ir_node *bl, int bias)
2204 const arch_env_t *arch_env = env->birg->main_env->arch_env;
2205 int omit_fp = env->call->flags.bits.try_omit_fp;
2208 sched_foreach(bl, irn) {
2211 Check, if the node relates to an entity on the stack frame.
2212 If so, set the true offset (including the bias) for that
2215 ir_entity *ent = arch_get_frame_entity(arch_env, irn);
2217 int offset = get_stack_entity_offset(env->frame, ent, bias);
2218 arch_set_frame_offset(arch_env, irn, offset);
2219 DBG((env->dbg, LEVEL_2, "%F has offset %d (including bias %d)\n", ent, offset, bias));
2223 If the node modifies the stack pointer by a constant offset,
2224 record that in the bias.
2226 if(arch_irn_is(arch_env, irn, modify_sp)) {
2227 int ofs = arch_get_sp_bias(arch_env, irn);
2229 if(be_is_IncSP(irn)) {
2230 if(ofs == BE_STACK_FRAME_SIZE_EXPAND) {
2231 ofs = get_type_size_bytes(get_irg_frame_type(env->birg->irg));
2232 be_set_IncSP_offset(irn, ofs);
2233 } else if(ofs == BE_STACK_FRAME_SIZE_SHRINK) {
2234 ofs = - get_type_size_bytes(get_irg_frame_type(env->birg->irg));
2235 be_set_IncSP_offset(irn, ofs);
2248 * A helper struct for the bias walker.
2251 be_abi_irg_t *env; /**< The ABI irg environment. */
2252 int start_block_bias; /**< The bias at the end of the start block. */
2253 ir_node *start_block; /**< The start block of the current graph. */
2257 * Block-Walker: fix all stack offsets
2259 static void stack_bias_walker(ir_node *bl, void *data)
2261 struct bias_walk *bw = data;
2262 if (bl != bw->start_block) {
2263 process_stack_bias(bw->env, bl, bw->start_block_bias);
2267 void be_abi_fix_stack_bias(be_abi_irg_t *env)
2269 ir_graph *irg = env->birg->irg;
2270 struct bias_walk bw;
2272 stack_frame_compute_initial_offset(env->frame);
2273 // stack_layout_dump(stdout, env->frame);
2275 /* Determine the stack bias at the end of the start block. */
2276 bw.start_block_bias = process_stack_bias(env, get_irg_start_block(irg), 0);
2278 /* fix the bias is all other blocks */
2280 bw.start_block = get_irg_start_block(irg);
2281 irg_block_walk_graph(irg, stack_bias_walker, NULL, &bw);
2284 ir_node *be_abi_get_callee_save_irn(be_abi_irg_t *abi, const arch_register_t *reg)
2286 assert(arch_register_type_is(reg, callee_save));
2287 assert(pmap_contains(abi->regs, (void *) reg));
2288 return pmap_get(abi->regs, (void *) reg);
2291 ir_node *be_abi_get_ignore_irn(be_abi_irg_t *abi, const arch_register_t *reg)
2293 assert(arch_register_type_is(reg, ignore));
2294 assert(pmap_contains(abi->regs, (void *) reg));
2295 return pmap_get(abi->regs, (void *) reg);
2298 ir_node *be_abi_get_start_barrier(be_abi_irg_t *abi)
2300 return abi->start_barrier;
2304 * Returns non-zero if the ABI has omitted the frame pointer in
2305 * the current graph.
2307 int be_abi_omit_fp(const be_abi_irg_t *abi) {
2308 return abi->call->flags.bits.try_omit_fp;