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
35 #include "irgraph_t.h"
38 #include "iredges_t.h"
41 #include "irprintf_t.h"
47 #include "raw_bitset.h"
54 #include "besched_t.h"
56 #include "bessaconstr.h"
58 typedef struct _be_abi_call_arg_t {
59 unsigned is_res : 1; /**< 1: the call argument is a return value. 0: it's a call parameter. */
60 unsigned in_reg : 1; /**< 1: this argument is transmitted in registers. */
61 unsigned on_stack : 1; /**< 1: this argument is transmitted on the stack. */
64 const arch_register_t *reg;
66 unsigned alignment; /**< stack alignment */
67 unsigned space_before; /**< allocate space before */
68 unsigned space_after; /**< allocate space after */
71 struct _be_abi_call_t {
72 be_abi_call_flags_t flags;
73 const be_abi_callbacks_t *cb;
74 ir_type *between_type;
76 const arch_register_class_t *cls_addr;
79 struct _be_abi_irg_t {
81 be_stack_layout_t *frame; /**< The stack frame model. */
82 be_irg_t *birg; /**< The back end IRG. */
83 const arch_isa_t *isa; /**< The isa. */
84 survive_dce_t *dce_survivor;
86 be_abi_call_t *call; /**< The ABI call information. */
87 ir_type *method_type; /**< The type of the method of the IRG. */
89 ir_node *init_sp; /**< The node representing the stack pointer
90 at the start of the function. */
92 ir_node *start_barrier; /**< The barrier of the start block */
94 ir_node *reg_params; /**< The reg params node. */
95 pmap *regs; /**< A map of all callee-save and ignore regs to
96 their Projs to the RegParams node. */
98 int start_block_bias; /**< The stack bias at the end of the start block. */
100 void *cb; /**< ABI Callback self pointer. */
102 pmap *keep_map; /**< mapping blocks to keep nodes. */
103 pset *ignore_regs; /**< Additional registers which shall be ignored. */
105 ir_node **calls; /**< flexible array containing all be_Call nodes */
107 arch_register_req_t sp_req;
108 arch_register_req_t sp_cls_req;
110 DEBUG_ONLY(firm_dbg_module_t *dbg;) /**< The debugging module. */
113 static heights_t *ir_heights;
115 /* Flag: if set, try to omit the frame pointer if called by the backend */
116 static int be_omit_fp = 1;
119 _ ____ ___ ____ _ _ _ _
120 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
121 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
122 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
123 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
125 These callbacks are used by the backend to set the parameters
126 for a specific call type.
130 * Set compare function: compares two ABI call object arguments.
132 static int cmp_call_arg(const void *a, const void *b, size_t n)
134 const be_abi_call_arg_t *p = a, *q = b;
135 return !(p->is_res == q->is_res && p->pos == q->pos);
139 * Get or set an ABI call object argument.
141 * @param call the abi call
142 * @param is_res true for call results, false for call arguments
143 * @param pos position of the argument
144 * @param do_insert true if the argument is set, false if it's retrieved
146 static be_abi_call_arg_t *get_or_set_call_arg(be_abi_call_t *call, int is_res, int pos, int do_insert)
148 be_abi_call_arg_t arg;
151 memset(&arg, 0, sizeof(arg));
155 hash = is_res * 128 + pos;
158 ? set_insert(call->params, &arg, sizeof(arg), hash)
159 : set_find(call->params, &arg, sizeof(arg), hash);
163 * Retrieve an ABI call object argument.
165 * @param call the ABI call object
166 * @param is_res true for call results, false for call arguments
167 * @param pos position of the argument
169 static INLINE be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos)
171 return get_or_set_call_arg(call, is_res, pos, 0);
174 /* Set the flags for a call. */
175 void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const be_abi_callbacks_t *cb)
182 /* Set register class for call address */
183 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
185 call->cls_addr = cls;
189 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos, unsigned alignment, unsigned space_before, unsigned space_after)
191 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 0, arg_pos, 1);
193 arg->alignment = alignment;
194 arg->space_before = space_before;
195 arg->space_after = space_after;
196 assert(alignment > 0 && "Alignment must be greater than 0");
199 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg)
201 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 0, arg_pos, 1);
206 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg)
208 be_abi_call_arg_t *arg = get_or_set_call_arg(call, 1, arg_pos, 1);
213 /* Get the flags of a ABI call object. */
214 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
220 * Constructor for a new ABI call object.
222 * @return the new ABI call object
224 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
226 be_abi_call_t *call = xmalloc(sizeof(call[0]));
229 call->params = new_set(cmp_call_arg, 16);
231 call->cls_addr = cls_addr;
233 call->flags.bits.try_omit_fp = be_omit_fp;
239 * Destructor for an ABI call object.
241 static void be_abi_call_free(be_abi_call_t *call)
243 del_set(call->params);
249 | ___| __ __ _ _ __ ___ ___ | | | | __ _ _ __ __| | (_)_ __ __ _
250 | |_ | '__/ _` | '_ ` _ \ / _ \ | |_| |/ _` | '_ \ / _` | | | '_ \ / _` |
251 | _|| | | (_| | | | | | | __/ | _ | (_| | | | | (_| | | | | | | (_| |
252 |_| |_| \__,_|_| |_| |_|\___| |_| |_|\__,_|_| |_|\__,_|_|_|_| |_|\__, |
255 Handling of the stack frame. It is composed of three types:
256 1) The type of the arguments which are pushed on the stack.
257 2) The "between type" which consists of stuff the call of the
258 function pushes on the stack (like the return address and
259 the old base pointer for ia32).
260 3) The Firm frame type which consists of all local variables
264 static int get_stack_entity_offset(be_stack_layout_t *frame, ir_entity *ent, int bias)
266 ir_type *t = get_entity_owner(ent);
267 int ofs = get_entity_offset(ent);
271 /* Find the type the entity is contained in. */
272 for(index = 0; index < N_FRAME_TYPES; ++index) {
273 if(frame->order[index] == t)
277 /* Add the size of all the types below the one of the entity to the entity's offset */
278 for(i = 0; i < index; ++i)
279 ofs += get_type_size_bytes(frame->order[i]);
281 /* correct the offset by the initial position of the frame pointer */
282 ofs -= frame->initial_offset;
284 /* correct the offset with the current bias. */
291 * Retrieve the entity with given offset from a frame type.
293 static ir_entity *search_ent_with_offset(ir_type *t, int offset)
297 for(i = 0, n = get_compound_n_members(t); i < n; ++i) {
298 ir_entity *ent = get_compound_member(t, i);
299 if(get_entity_offset(ent) == offset)
306 static int stack_frame_compute_initial_offset(be_stack_layout_t *frame)
308 ir_type *base = frame->stack_dir < 0 ? frame->between_type : frame->frame_type;
309 ir_entity *ent = search_ent_with_offset(base, 0);
311 frame->initial_offset = ent ? get_stack_entity_offset(frame, ent, 0) : 0;
313 return frame->initial_offset;
317 * Initializes the frame layout from parts
319 * @param frame the stack layout that will be initialized
320 * @param args the stack argument layout type
321 * @param between the between layout type
322 * @param locals the method frame type
323 * @param stack_dir the stack direction
324 * @param param_map an array mapping method argument positions to the stack argument type
326 * @return the initialized stack layout
328 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
329 ir_type *between, ir_type *locals, int stack_dir,
330 ir_entity *param_map[])
332 frame->arg_type = args;
333 frame->between_type = between;
334 frame->frame_type = locals;
335 frame->initial_offset = 0;
336 frame->stack_dir = stack_dir;
337 frame->order[1] = between;
338 frame->param_map = param_map;
341 frame->order[0] = args;
342 frame->order[2] = locals;
345 frame->order[0] = locals;
346 frame->order[2] = args;
352 /** Dumps the stack layout to file. */
353 static void stack_layout_dump(FILE *file, be_stack_layout_t *frame)
357 ir_fprintf(file, "initial offset: %d\n", frame->initial_offset);
358 for (j = 0; j < N_FRAME_TYPES; ++j) {
359 ir_type *t = frame->order[j];
361 ir_fprintf(file, "type %d: %F size: %d\n", j, t, get_type_size_bytes(t));
362 for (i = 0, n = get_compound_n_members(t); i < n; ++i) {
363 ir_entity *ent = get_compound_member(t, i);
364 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));
371 * Returns non-zero if the call argument at given position
372 * is transfered on the stack.
374 static INLINE int is_on_stack(be_abi_call_t *call, int pos)
376 be_abi_call_arg_t *arg = get_call_arg(call, 0, pos);
377 return arg && !arg->in_reg;
387 Adjustment of the calls inside a graph.
392 * Transform a call node.
393 * @param env The ABI environment for the current irg.
394 * @param irn The call node.
395 * @param curr_sp The stack pointer node to use.
396 * @return The stack pointer after the call.
398 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, ir_node *alloca_copy)
400 ir_graph *irg = env->birg->irg;
401 const arch_env_t *arch_env = env->birg->main_env->arch_env;
402 const arch_isa_t *isa = arch_env->isa;
403 ir_type *mt = get_Call_type(irn);
404 ir_node *call_ptr = get_Call_ptr(irn);
405 int n_params = get_method_n_params(mt);
406 ir_node *curr_mem = get_Call_mem(irn);
407 ir_node *bl = get_nodes_block(irn);
408 pset *results = pset_new_ptr(8);
409 pset *caller_save = pset_new_ptr(8);
410 pset *states = pset_new_ptr(2);
412 int stack_dir = arch_isa_stack_dir(isa);
413 const arch_register_t *sp = arch_isa_sp(isa);
414 be_abi_call_t *call = be_abi_call_new(sp->reg_class);
415 ir_mode *mach_mode = sp->reg_class->mode;
416 struct obstack *obst = &env->obst;
417 int no_alloc = call->flags.bits.frame_is_setup_on_call;
419 ir_node *res_proj = NULL;
420 int curr_res_proj = pn_Call_max;
428 const arch_register_t *reg;
429 const ir_edge_t *edge;
434 /* Let the isa fill out the abi description for that call node. */
435 arch_isa_get_call_abi(isa, mt, call);
437 /* Insert code to put the stack arguments on the stack. */
438 assert(get_Call_n_params(irn) == n_params);
439 for(i = 0; i < n_params; ++i) {
440 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
443 int arg_size = get_type_size_bytes(get_method_param_type(mt, i));
445 stack_size += round_up2(arg->space_before, arg->alignment);
446 stack_size += round_up2(arg_size, arg->alignment);
447 stack_size += round_up2(arg->space_after, arg->alignment);
448 obstack_int_grow(obst, i);
452 pos = obstack_finish(obst);
454 /* Collect all arguments which are passed in registers. */
455 for(i = 0, n = get_Call_n_params(irn); i < n; ++i) {
456 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
457 if(arg && arg->in_reg) {
458 obstack_int_grow(obst, i);
462 low_args = obstack_finish(obst);
464 /* If there are some parameters which shall be passed on the stack. */
467 int do_seq = call->flags.bits.store_args_sequential && !no_alloc;
470 * Reverse list of stack parameters if call arguments are from left to right.
471 * We must them reverse again if they are pushed (not stored) and the stack
472 * direction is downwards.
474 if (call->flags.bits.left_to_right ^ (do_seq && stack_dir < 0)) {
475 for (i = 0; i < n_pos >> 1; ++i) {
476 int other = n_pos - i - 1;
484 * If the stack is decreasing and we do not want to store sequentially,
485 * or someone else allocated the call frame
486 * we allocate as much space on the stack all parameters need, by
487 * moving the stack pointer along the stack's direction.
489 if(stack_dir < 0 && !do_seq && !no_alloc) {
490 curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, stack_size);
492 add_irn_dep(curr_sp, alloca_copy);
498 obstack_ptr_grow(obst, get_Call_mem(irn));
499 curr_mem = new_NoMem();
501 curr_mem = get_Call_mem(irn);
504 for(i = 0; i < n_pos; ++i) {
506 be_abi_call_arg_t *arg = get_call_arg(call, 0, p);
507 ir_node *param = get_Call_param(irn, p);
508 ir_node *addr = curr_sp;
510 ir_type *param_type = get_method_param_type(mt, p);
511 int param_size = get_type_size_bytes(param_type) + arg->space_after;
514 * If we wanted to build the arguments sequentially,
515 * the stack pointer for the next must be incremented,
516 * and the memory value propagated.
520 addr = curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, param_size + arg->space_before);
522 add_irn_dep(curr_sp, alloca_copy);
525 add_irn_dep(curr_sp, curr_mem);
528 curr_ofs += arg->space_before;
529 curr_ofs = round_up2(curr_ofs, arg->alignment);
531 /* Make the expression to compute the argument's offset. */
533 ir_mode *constmode = mach_mode;
534 if(mode_is_reference(mach_mode)) {
537 addr = new_r_Const_long(irg, bl, constmode, curr_ofs);
538 addr = new_r_Add(irg, bl, curr_sp, addr, mach_mode);
542 /* Insert a store for primitive arguments. */
543 if (is_atomic_type(param_type)) {
545 store = new_r_Store(irg, bl, curr_mem, addr, param);
546 mem = new_r_Proj(irg, bl, store, mode_M, pn_Store_M);
549 /* Make a mem copy for compound arguments. */
553 assert(mode_is_reference(get_irn_mode(param)));
554 copy = new_r_CopyB(irg, bl, curr_mem, addr, param, param_type);
555 mem = new_r_Proj(irg, bl, copy, mode_M, pn_CopyB_M_regular);
558 curr_ofs += param_size;
563 obstack_ptr_grow(obst, mem);
566 in = (ir_node **) obstack_finish(obst);
568 /* We need the sync only, if we didn't build the stores sequentially. */
571 curr_mem = new_r_Sync(irg, bl, n_pos + 1, in);
573 curr_mem = get_Call_mem(irn);
576 obstack_free(obst, in);
579 /* Collect caller save registers */
580 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
582 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
583 for(j = 0; j < cls->n_regs; ++j) {
584 const arch_register_t *reg = arch_register_for_index(cls, j);
585 if(arch_register_type_is(reg, caller_save)) {
586 pset_insert_ptr(caller_save, (void *) reg);
588 if(arch_register_type_is(reg, state)) {
589 pset_insert_ptr(caller_save, (void*) reg);
590 pset_insert_ptr(states, (void*) reg);
595 /* search the greatest result proj number */
597 /* TODO: what if the result is NOT used? Currently there is
598 * no way to detect this later, especially there is no way to
599 * see this in the proj numbers.
600 * While this is ok for the register allocator, it is bad for
601 * backends which need to change the be_Call further (x87 simulator
602 * for instance. However for this particular case the call_type is
605 foreach_out_edge(irn, edge) {
606 const ir_edge_t *res_edge;
607 ir_node *irn = get_edge_src_irn(edge);
609 if(is_Proj(irn) && get_Proj_proj(irn) == pn_Call_T_result) {
611 foreach_out_edge(irn, res_edge) {
613 be_abi_call_arg_t *arg;
614 ir_node *res = get_edge_src_irn(res_edge);
616 assert(is_Proj(res));
618 proj = get_Proj_proj(res);
619 arg = get_call_arg(call, 1, proj);
622 shift the proj number to the right, since we will drop the
623 unspeakable Proj_T from the Call. Therefore, all real argument
624 Proj numbers must be increased by pn_be_Call_first_res
626 proj += pn_be_Call_first_res;
627 set_Proj_proj(res, proj);
628 obstack_ptr_grow(obst, res);
630 if(proj > curr_res_proj)
631 curr_res_proj = proj;
633 pset_remove_ptr(caller_save, arg->reg);
634 //pmap_insert(arg_regs, arg->reg, INT_TO_PTR(proj + 1))
641 obstack_ptr_grow(obst, NULL);
642 res_projs = obstack_finish(obst);
644 /* make the back end call node and set its register requirements. */
645 for(i = 0; i < n_low_args; ++i) {
646 obstack_ptr_grow(obst, get_Call_param(irn, low_args[i]));
648 foreach_pset(states, reg) {
649 const arch_register_class_t *cls = arch_register_get_class(reg);
651 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
652 ir_fprintf(stderr, "Adding %+F\n", regnode);
654 ir_node *regnode = new_rd_Unknown(irg, arch_register_class_mode(cls));
655 obstack_ptr_grow(obst, regnode);
657 count = n_low_args + pset_count(states);
659 in = obstack_finish(obst);
661 if(env->call->flags.bits.call_has_imm && get_irn_opcode(call_ptr) == iro_SymConst) {
662 low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
664 curr_res_proj + pset_count(caller_save), count,
665 in, get_Call_type(irn));
666 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
668 low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
670 curr_res_proj + pset_count(caller_save),
671 count, in, get_Call_type(irn));
673 ARR_APP1(ir_node*, env->calls, low_call);
676 Set the register class of the call address to
677 the backend provided class (default: stack pointer class)
679 be_node_set_reg_class(low_call, be_pos_Call_ptr, call->cls_addr);
681 DBG((env->dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
683 /* Set the register classes and constraints of the Call parameters. */
684 for(i = 0; i < n_low_args; ++i) {
685 int index = low_args[i];
686 be_abi_call_arg_t *arg = get_call_arg(call, 0, index);
687 assert(arg->reg != NULL);
689 be_set_constr_single_reg(low_call, be_pos_Call_first_arg + index, arg->reg);
692 /* Set the register constraints of the results. */
693 for (i = 0; res_projs[i]; ++i) {
694 int pn = get_Proj_proj(res_projs[i]);
696 /* Correct Proj number since it has been adjusted! (see above) */
697 const be_abi_call_arg_t *arg = get_call_arg(call, 1, pn - pn_Call_max);
699 /* Matze: we need the information about the real mode for later
700 * transforms (signed/unsigend compares, stores...), so leave the fixup
701 * for the backend transform phase... */
704 const arch_register_class_t *cls = arch_register_get_class(arg->reg);
705 ir_mode *mode = arch_register_class_mode(cls);
706 set_irn_mode(irn, mode);
710 be_set_constr_single_reg(low_call, BE_OUT_POS(pn), arg->reg);
711 arch_set_irn_register(arch_env, res_projs[i], arg->reg);
713 obstack_free(obst, in);
714 exchange(irn, low_call);
716 /* redirect the result projs to the lowered call instead of the Proj_T */
717 for (i = 0; res_projs[i]; ++i)
718 set_Proj_pred(res_projs[i], low_call);
720 /* set the now unnecessary projT to bad */
721 if(res_proj != NULL) {
722 be_kill_node(res_proj);
725 /* Make additional projs for the caller save registers
726 and the Keep node which keeps them alive. */
727 if (pset_count(caller_save) > 0) {
728 const arch_register_t *reg;
732 for (reg = pset_first(caller_save), n = 0; reg; reg = pset_next(caller_save), ++n) {
733 ir_node *proj = new_r_Proj(irg, bl, low_call, reg->reg_class->mode, curr_res_proj);
735 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
736 be_set_constr_single_reg(low_call, BE_OUT_POS(curr_res_proj), reg);
738 /* a call can produce ignore registers, in this case set the flag and register for the Proj */
739 if (arch_register_type_is(reg, ignore)) {
740 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
741 be_node_set_flags(low_call, BE_OUT_POS(curr_res_proj), arch_irn_flags_ignore);
744 set_irn_link(proj, (void *) reg);
745 obstack_ptr_grow(obst, proj);
749 /* create the Keep for the caller save registers */
750 in = (ir_node **) obstack_finish(obst);
751 keep = be_new_Keep(NULL, irg, bl, n, in);
752 for (i = 0; i < n; ++i) {
753 const arch_register_t *reg = get_irn_link(in[i]);
754 be_node_set_reg_class(keep, i, reg->reg_class);
756 obstack_free(obst, in);
759 /* Clean up the stack. */
761 ir_node *mem_proj = NULL;
763 foreach_out_edge(low_call, edge) {
764 ir_node *irn = get_edge_src_irn(edge);
765 if(is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
772 mem_proj = new_r_Proj(irg, bl, low_call, mode_M, pn_Call_M);
773 keep_alive(mem_proj);
776 /* Clean up the stack frame if we allocated it */
778 curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, -stack_size);
779 add_irn_dep(curr_sp, mem_proj);
781 add_irn_dep(curr_sp, alloca_copy);
787 be_abi_call_free(call);
788 obstack_free(obst, pos);
791 del_pset(caller_save);
797 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
799 * @param alignment the minimum stack alignment
800 * @param size the node containing the non-aligned size
801 * @param irg the irg where new nodes are allocated on
802 * @param irg the block where new nodes are allocated on
803 * @param dbg debug info for new nodes
805 * @return a node representing the aligned size
807 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size, ir_graph *irg, ir_node *block, dbg_info *dbg) {
808 if (stack_alignment > 1) {
809 ir_mode *mode = get_irn_mode(size);
810 tarval *tv = new_tarval_from_long(stack_alignment-1, mode);
811 ir_node *mask = new_r_Const(irg, block, mode, tv);
813 size = new_rd_Add(dbg, irg, block, size, mask, mode);
815 tv = new_tarval_from_long(-(long)stack_alignment, mode);
816 mask = new_r_Const(irg, block, mode, tv);
817 size = new_rd_And(dbg, irg, block, size, mask, mode);
823 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
825 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp, ir_node **result_copy)
834 const ir_edge_t *edge;
835 ir_node *new_alloc, *size, *addr, *copy, *ins[2];
836 unsigned stack_alignment;
838 if (get_Alloc_where(alloc) != stack_alloc) {
843 block = get_nodes_block(alloc);
844 irg = get_irn_irg(block);
847 type = get_Alloc_type(alloc);
849 foreach_out_edge(alloc, edge) {
850 ir_node *irn = get_edge_src_irn(edge);
852 assert(is_Proj(irn));
853 switch(get_Proj_proj(irn)) {
865 /* Beware: currently Alloc nodes without a result might happen,
866 only escape analysis kills them and this phase runs only for object
867 oriented source. We kill the Alloc here. */
868 if (alloc_res == NULL && alloc_mem) {
869 exchange(alloc_mem, get_Alloc_mem(alloc));
873 dbg = get_irn_dbg_info(alloc);
875 /* we might need to multiply the size with the element size */
876 if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
877 tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
878 ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
879 ir_node *mul = new_rd_Mul(dbg, irg, block, get_Alloc_size(alloc),
883 size = get_Alloc_size(alloc);
886 /* The stack pointer will be modified in an unknown manner.
887 We cannot omit it. */
888 env->call->flags.bits.try_omit_fp = 0;
890 /* FIXME: size must be here round up for the stack alignment, but
891 this must be transmitted from the backend. */
893 size = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
894 new_alloc = be_new_AddSP(env->isa->sp, irg, block, curr_sp, size);
895 set_irn_dbg_info(new_alloc, dbg);
897 if(alloc_mem != NULL) {
901 addsp_mem = new_r_Proj(irg, block, new_alloc, mode_M, pn_be_AddSP_M);
903 /* We need to sync the output mem of the AddSP with the input mem
904 edge into the alloc node. */
905 ins[0] = get_Alloc_mem(alloc);
907 sync = new_r_Sync(irg, block, 2, ins);
909 exchange(alloc_mem, sync);
912 exchange(alloc, new_alloc);
914 /* fix projnum of alloca res */
915 set_Proj_proj(alloc_res, pn_be_AddSP_res);
917 addr = env->isa->stack_dir < 0 ? alloc_res : curr_sp;
919 /* copy the address away, since it could be used after further stack pointer modifications. */
920 /* Let it point curr_sp just for the moment, I'll reroute it in a second. */
921 *result_copy = copy = be_new_Copy(env->isa->sp->reg_class, irg, block, curr_sp);
923 /* Let all users of the Alloc() result now point to the copy. */
924 edges_reroute(alloc_res, copy, irg);
926 /* Rewire the copy appropriately. */
927 set_irn_n(copy, be_pos_Copy_op, addr);
936 * The Free is transformed into a back end free node and connected to the stack nodes.
938 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
942 ir_node *subsp, *mem, *res, *size, *sync;
946 unsigned stack_alignment;
949 if (get_Free_where(free) != stack_alloc) {
954 block = get_nodes_block(free);
955 irg = get_irn_irg(block);
956 type = get_Free_type(free);
957 sp_mode = env->isa->sp->reg_class->mode;
958 dbg = get_irn_dbg_info(free);
960 /* we might need to multiply the size with the element size */
961 if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
962 tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
963 ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
964 ir_node *mul = new_rd_Mul(dbg, irg, block, get_Free_size(free),
968 size = get_Free_size(free);
971 /* FIXME: size must be here round up for the stack alignment, but
972 this must be transmitted from the backend. */
974 size = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
976 /* The stack pointer will be modified in an unknown manner.
977 We cannot omit it. */
978 env->call->flags.bits.try_omit_fp = 0;
979 subsp = be_new_SubSP(env->isa->sp, irg, block, curr_sp, size);
980 set_irn_dbg_info(subsp, dbg);
982 mem = new_r_Proj(irg, block, subsp, mode_M, pn_be_SubSP_M);
983 res = new_r_Proj(irg, block, subsp, sp_mode, pn_be_SubSP_res);
985 /* we need to sync the memory */
986 in[0] = get_Free_mem(free);
988 sync = new_r_Sync(irg, block, 2, in);
990 /* and make the AddSP dependent on the former memory */
991 add_irn_dep(subsp, get_Free_mem(free));
994 exchange(free, sync);
1000 /* the following function is replaced by the usage of the heights module */
1003 * Walker for dependent_on().
1004 * This function searches a node tgt recursively from a given node
1005 * but is restricted to the given block.
1006 * @return 1 if tgt was reachable from curr, 0 if not.
1008 static int check_dependence(ir_node *curr, ir_node *tgt, ir_node *bl)
1012 if (get_nodes_block(curr) != bl)
1018 /* Phi functions stop the recursion inside a basic block */
1019 if (! is_Phi(curr)) {
1020 for(i = 0, n = get_irn_arity(curr); i < n; ++i) {
1021 if (check_dependence(get_irn_n(curr, i), tgt, bl))
1031 * Check if a node is somehow data dependent on another one.
1032 * both nodes must be in the same basic block.
1033 * @param n1 The first node.
1034 * @param n2 The second node.
1035 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
1037 static int dependent_on(ir_node *n1, ir_node *n2)
1039 assert(get_nodes_block(n1) == get_nodes_block(n2));
1041 return heights_reachable_in_block(ir_heights, n1, n2);
1044 static int cmp_call_dependency(const void *c1, const void *c2)
1046 ir_node *n1 = *(ir_node **) c1;
1047 ir_node *n2 = *(ir_node **) c2;
1050 Classical qsort() comparison function behavior:
1051 0 if both elements are equal
1052 1 if second is "smaller" that first
1053 -1 if first is "smaller" that second
1055 if (dependent_on(n1, n2))
1058 if (dependent_on(n2, n1))
1065 * Walker: links all Call/alloc/Free nodes to the Block they are contained.
1067 static void link_calls_in_block_walker(ir_node *irn, void *data)
1069 ir_opcode code = get_irn_opcode(irn);
1071 if (code == iro_Call ||
1072 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
1073 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
1074 be_abi_irg_t *env = data;
1075 ir_node *bl = get_nodes_block(irn);
1076 void *save = get_irn_link(bl);
1078 if (code == iro_Call)
1079 env->call->flags.bits.irg_is_leaf = 0;
1081 set_irn_link(irn, save);
1082 set_irn_link(bl, irn);
1088 * Process all Call nodes inside a basic block.
1089 * Note that the link field of the block must contain a linked list of all
1090 * Call nodes inside the Block. We first order this list according to data dependency
1091 * and that connect the calls together.
1093 static void process_calls_in_block(ir_node *bl, void *data)
1095 be_abi_irg_t *env = data;
1096 ir_node *curr_sp = env->init_sp;
1100 for(irn = get_irn_link(bl), n = 0; irn; irn = get_irn_link(irn), ++n)
1101 obstack_ptr_grow(&env->obst, irn);
1103 /* If there were call nodes in the block. */
1107 ir_node *copy = NULL;
1110 nodes = obstack_finish(&env->obst);
1112 /* order the call nodes according to data dependency */
1113 qsort(nodes, n, sizeof(nodes[0]), cmp_call_dependency);
1115 for(i = n - 1; i >= 0; --i) {
1116 ir_node *irn = nodes[i];
1118 DBG((env->dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1119 switch(get_irn_opcode(irn)) {
1121 curr_sp = adjust_call(env, irn, curr_sp, copy);
1124 curr_sp = adjust_alloc(env, irn, curr_sp, ©);
1127 curr_sp = adjust_free(env, irn, curr_sp);
1134 obstack_free(&env->obst, nodes);
1136 /* Keep the last stack state in the block by tying it to Keep node */
1138 keep = be_new_Keep(env->isa->sp->reg_class, get_irn_irg(bl), bl, 1, nodes);
1139 pmap_insert(env->keep_map, bl, keep);
1142 set_irn_link(bl, curr_sp);
1143 } /* process_calls_in_block */
1146 * Adjust all call nodes in the graph to the ABI conventions.
1148 static void process_calls(be_abi_irg_t *env)
1150 ir_graph *irg = env->birg->irg;
1152 env->call->flags.bits.irg_is_leaf = 1;
1153 irg_walk_graph(irg, firm_clear_link, link_calls_in_block_walker, env);
1155 ir_heights = heights_new(env->birg->irg);
1156 irg_block_walk_graph(irg, NULL, process_calls_in_block, env);
1157 heights_free(ir_heights);
1161 static ir_node *setup_frame(be_abi_irg_t *env)
1163 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1164 const arch_register_t *sp = isa->sp;
1165 const arch_register_t *bp = isa->bp;
1166 be_abi_call_flags_bits_t flags = env->call->flags.bits;
1167 ir_graph *irg = env->birg->irg;
1168 ir_node *bl = get_irg_start_block(irg);
1169 ir_node *no_mem = get_irg_no_mem(irg);
1170 ir_node *old_frame = get_irg_frame(irg);
1171 ir_node *stack = pmap_get(env->regs, (void *) sp);
1172 ir_node *frame = pmap_get(env->regs, (void *) bp);
1174 int stack_nr = get_Proj_proj(stack);
1176 if(flags.try_omit_fp) {
1177 stack = be_new_IncSP(sp, irg, bl, stack, no_mem, BE_STACK_FRAME_SIZE_EXPAND);
1182 frame = be_new_Copy(bp->reg_class, irg, bl, stack);
1184 be_node_set_flags(frame, -1, arch_irn_flags_dont_spill);
1185 if(!flags.fp_free) {
1186 be_set_constr_single_reg(frame, -1, bp);
1187 be_node_set_flags(frame, -1, arch_irn_flags_ignore);
1188 arch_set_irn_register(env->birg->main_env->arch_env, frame, bp);
1191 stack = be_new_IncSP(sp, irg, bl, stack, frame, BE_STACK_FRAME_SIZE_EXPAND);
1194 be_node_set_flags(env->reg_params, -(stack_nr + 1), arch_irn_flags_ignore);
1195 env->init_sp = stack;
1196 set_irg_frame(irg, frame);
1197 edges_reroute(old_frame, frame, irg);
1202 static void clearup_frame(be_abi_irg_t *env, ir_node *ret, pmap *reg_map, struct obstack *obst)
1204 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1205 const arch_register_t *sp = isa->sp;
1206 const arch_register_t *bp = isa->bp;
1207 ir_graph *irg = env->birg->irg;
1208 ir_node *ret_mem = get_Return_mem(ret);
1209 ir_node *frame = get_irg_frame(irg);
1210 ir_node *bl = get_nodes_block(ret);
1211 ir_node *stack = get_irn_link(bl);
1215 if(env->call->flags.bits.try_omit_fp) {
1216 stack = be_new_IncSP(sp, irg, bl, stack, ret_mem, -BE_STACK_FRAME_SIZE_SHRINK);
1220 stack = be_new_SetSP(sp, irg, bl, stack, frame, ret_mem);
1221 be_set_constr_single_reg(stack, -1, sp);
1222 be_node_set_flags(stack, -1, arch_irn_flags_ignore);
1225 pmap_foreach(env->regs, ent) {
1226 const arch_register_t *reg = ent->key;
1227 ir_node *irn = ent->value;
1230 obstack_ptr_grow(&env->obst, stack);
1232 obstack_ptr_grow(&env->obst, frame);
1233 else if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
1234 obstack_ptr_grow(obst, irn);
1241 * Computes the stack argument layout type.
1242 * Changes a possibly allocated value param type by moving
1243 * entities to the stack layout type.
1245 * @param env the ABI environment
1246 * @param call the current call ABI
1247 * @param method_type the method type
1248 * @param param_map an array mapping method arguments to the stack layout type
1250 * @return the stack argument layout type
1252 static ir_type *compute_arg_type(be_abi_irg_t *env, be_abi_call_t *call, ir_type *method_type, ir_entity ***param_map)
1254 int dir = env->call->flags.bits.left_to_right ? 1 : -1;
1255 int inc = env->birg->main_env->arch_env->isa->stack_dir * dir;
1256 int n = get_method_n_params(method_type);
1257 int curr = inc > 0 ? 0 : n - 1;
1263 ir_type *val_param_tp = get_method_value_param_type(method_type);
1264 ident *id = get_entity_ident(get_irg_entity(env->birg->irg));
1267 *param_map = map = obstack_alloc(&env->obst, n * sizeof(ir_entity *));
1268 res = new_type_struct(mangle_u(id, new_id_from_chars("arg_type", 8)));
1269 for (i = 0; i < n; ++i, curr += inc) {
1270 ir_type *param_type = get_method_param_type(method_type, curr);
1271 be_abi_call_arg_t *arg = get_call_arg(call, 0, curr);
1274 if (arg->on_stack) {
1276 /* the entity was already created, move it to the param type */
1277 arg->stack_ent = get_method_value_param_ent(method_type, i);
1278 remove_struct_member(val_param_tp, arg->stack_ent);
1279 set_entity_owner(arg->stack_ent, res);
1280 add_struct_member(res, arg->stack_ent);
1281 /* must be automatic to set a fixed layout */
1282 set_entity_allocation(arg->stack_ent, allocation_automatic);
1285 snprintf(buf, sizeof(buf), "param_%d", i);
1286 arg->stack_ent = new_entity(res, new_id_from_str(buf), param_type);
1288 ofs += arg->space_before;
1289 ofs = round_up2(ofs, arg->alignment);
1290 set_entity_offset(arg->stack_ent, ofs);
1291 ofs += arg->space_after;
1292 ofs += get_type_size_bytes(param_type);
1293 map[i] = arg->stack_ent;
1296 set_type_size_bytes(res, ofs);
1297 set_type_state(res, layout_fixed);
1302 static void create_register_perms(const arch_isa_t *isa, ir_graph *irg, ir_node *bl, pmap *regs)
1305 struct obstack obst;
1307 obstack_init(&obst);
1309 /* Create a Perm after the RegParams node to delimit it. */
1310 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
1311 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1316 for(n_regs = 0, j = 0; j < cls->n_regs; ++j) {
1317 const arch_register_t *reg = &cls->regs[j];
1318 ir_node *irn = pmap_get(regs, (void *) reg);
1320 if(irn && !arch_register_type_is(reg, ignore)) {
1322 obstack_ptr_grow(&obst, irn);
1323 set_irn_link(irn, (void *) reg);
1327 obstack_ptr_grow(&obst, NULL);
1328 in = obstack_finish(&obst);
1330 perm = be_new_Perm(cls, irg, bl, n_regs, in);
1331 for(j = 0; j < n_regs; ++j) {
1332 ir_node *arg = in[j];
1333 arch_register_t *reg = get_irn_link(arg);
1334 pmap_insert(regs, reg, arg);
1335 be_set_constr_single_reg(perm, BE_OUT_POS(j), reg);
1338 obstack_free(&obst, in);
1341 obstack_free(&obst, NULL);
1346 const arch_register_t *reg;
1350 static int cmp_regs(const void *a, const void *b)
1352 const reg_node_map_t *p = a;
1353 const reg_node_map_t *q = b;
1355 if(p->reg->reg_class == q->reg->reg_class)
1356 return p->reg->index - q->reg->index;
1358 return p->reg->reg_class - q->reg->reg_class;
1361 static reg_node_map_t *reg_map_to_arr(struct obstack *obst, pmap *reg_map)
1364 int n = pmap_count(reg_map);
1366 reg_node_map_t *res = obstack_alloc(obst, n * sizeof(res[0]));
1368 pmap_foreach(reg_map, ent) {
1369 res[i].reg = ent->key;
1370 res[i].irn = ent->value;
1374 qsort(res, n, sizeof(res[0]), cmp_regs);
1379 * Creates a barrier.
1381 static ir_node *create_barrier(be_abi_irg_t *env, ir_node *bl, ir_node **mem, pmap *regs, int in_req)
1383 ir_graph *irg = env->birg->irg;
1384 int n_regs = pmap_count(regs);
1390 rm = reg_map_to_arr(&env->obst, regs);
1392 for(n = 0; n < n_regs; ++n)
1393 obstack_ptr_grow(&env->obst, rm[n].irn);
1396 obstack_ptr_grow(&env->obst, *mem);
1400 in = (ir_node **) obstack_finish(&env->obst);
1401 irn = be_new_Barrier(irg, bl, n, in);
1402 obstack_free(&env->obst, in);
1404 for(n = 0; n < n_regs; ++n) {
1405 const arch_register_t *reg = rm[n].reg;
1407 int pos = BE_OUT_POS(n);
1410 proj = new_r_Proj(irg, bl, irn, get_irn_mode(rm[n].irn), n);
1411 be_node_set_reg_class(irn, n, reg->reg_class);
1413 be_set_constr_single_reg(irn, n, reg);
1414 be_set_constr_single_reg(irn, pos, reg);
1415 be_node_set_reg_class(irn, pos, reg->reg_class);
1416 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
1418 /* if the proj projects a ignore register or a node which is set to ignore, propagate this property. */
1419 if(arch_register_type_is(reg, ignore) || arch_irn_is(env->birg->main_env->arch_env, in[n], ignore))
1420 flags |= arch_irn_flags_ignore;
1422 if(arch_irn_is(env->birg->main_env->arch_env, in[n], modify_sp))
1423 flags |= arch_irn_flags_modify_sp;
1425 be_node_set_flags(irn, pos, flags);
1427 pmap_insert(regs, (void *) reg, proj);
1431 *mem = new_r_Proj(irg, bl, irn, mode_M, n);
1434 obstack_free(&env->obst, rm);
1439 * Creates a be_Return for a Return node.
1441 * @param @env the abi environment
1442 * @param irn the Return node or NULL if there was none
1443 * @param bl the block where the be_Retun should be placed
1444 * @param mem the current memory
1445 * @param n_res number of return results
1447 static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, ir_node *mem, int n_res) {
1448 be_abi_call_t *call = env->call;
1449 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1451 pmap *reg_map = pmap_create();
1452 ir_node *keep = pmap_get(env->keep_map, bl);
1458 const arch_register_t **regs;
1462 get the valid stack node in this block.
1463 If we had a call in that block there is a Keep constructed by process_calls()
1464 which points to the last stack modification in that block. we'll use
1465 it then. Else we use the stack from the start block and let
1466 the ssa construction fix the usage.
1468 stack = be_abi_reg_map_get(env->regs, isa->sp);
1470 ir_node *bad = new_r_Bad(env->birg->irg);
1471 stack = get_irn_n(keep, 0);
1472 set_nodes_block(keep, bad);
1473 set_irn_n(keep, 0, bad);
1474 // exchange(keep, new_r_Bad(env->birg->irg));
1477 /* Insert results for Return into the register map. */
1478 for(i = 0; i < n_res; ++i) {
1479 ir_node *res = get_Return_res(irn, i);
1480 be_abi_call_arg_t *arg = get_call_arg(call, 1, i);
1481 assert(arg->in_reg && "return value must be passed in register");
1482 pmap_insert(reg_map, (void *) arg->reg, res);
1485 /* Add uses of the callee save registers. */
1486 pmap_foreach(env->regs, ent) {
1487 const arch_register_t *reg = ent->key;
1488 if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
1489 pmap_insert(reg_map, ent->key, ent->value);
1492 be_abi_reg_map_set(reg_map, isa->sp, stack);
1494 /* Make the Epilogue node and call the arch's epilogue maker. */
1495 create_barrier(env, bl, &mem, reg_map, 1);
1496 call->cb->epilogue(env->cb, bl, &mem, reg_map);
1499 Maximum size of the in array for Return nodes is
1500 return args + callee save/ignore registers + memory + stack pointer
1502 in_max = pmap_count(reg_map) + n_res + 2;
1504 in = obstack_alloc(&env->obst, in_max * sizeof(in[0]));
1505 regs = obstack_alloc(&env->obst, in_max * sizeof(regs[0]));
1508 in[1] = be_abi_reg_map_get(reg_map, isa->sp);
1513 /* clear SP entry, since it has already been grown. */
1514 pmap_insert(reg_map, (void *) isa->sp, NULL);
1515 for(i = 0; i < n_res; ++i) {
1516 be_abi_call_arg_t *arg = get_call_arg(call, 1, i);
1518 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1519 regs[n++] = arg->reg;
1521 /* Clear the map entry to mark the register as processed. */
1522 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1525 /* grow the rest of the stuff. */
1526 pmap_foreach(reg_map, ent) {
1529 regs[n++] = ent->key;
1533 /* The in array for the new back end return is now ready. */
1534 ret = be_new_Return(irn ? get_irn_dbg_info(irn) : NULL, env->birg->irg, bl, n_res, n, in);
1536 /* Set the register classes of the return's parameter accordingly. */
1537 for(i = 0; i < n; ++i)
1539 be_node_set_reg_class(ret, i, regs[i]->reg_class);
1541 /* Free the space of the Epilog's in array and the register <-> proj map. */
1542 obstack_free(&env->obst, in);
1543 pmap_destroy(reg_map);
1548 typedef struct lower_frame_sels_env_t {
1550 ir_entity *value_param_list; /**< the list of all value param entities */
1551 } lower_frame_sels_env_t;
1554 * Walker: Replaces Sels of frame type and
1555 * value param type entities by FrameAddress.
1557 static void lower_frame_sels_walker(ir_node *irn, void *data)
1559 lower_frame_sels_env_t *ctx = data;
1562 ir_graph *irg = current_ir_graph;
1563 ir_node *frame = get_irg_frame(irg);
1564 ir_node *param_base = get_irg_value_param_base(irg);
1565 ir_node *ptr = get_Sel_ptr(irn);
1567 if (ptr == frame || ptr == param_base) {
1568 be_abi_irg_t *env = ctx->env;
1569 ir_entity *ent = get_Sel_entity(irn);
1570 ir_node *bl = get_nodes_block(irn);
1573 nw = be_new_FrameAddr(env->isa->sp->reg_class, irg, bl, frame, ent);
1576 /* check, if it's a param sel and if have not seen this entity immediatly before */
1577 if (ptr == param_base && ctx->value_param_list != ent) {
1578 set_entity_link(ent, ctx->value_param_list);
1579 ctx->value_param_list = ent;
1586 * Check if a value parameter is transmitted as a register.
1587 * This might happen if the address of an parameter is taken which is
1588 * transmitted in registers.
1590 * Note that on some architectures this case must be handled specially
1591 * because the place of the backing store is determined by their ABI.
1593 * In the default case we move the entity to the frame type and create
1594 * a backing store into the first block.
1596 static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_entity *value_param_list) {
1597 be_abi_call_t *call = env->call;
1598 ir_graph *irg = env->birg->irg;
1599 ir_entity *ent, *next_ent, *new_list;
1601 DEBUG_ONLY(firm_dbg_module_t *dbg = env->dbg;)
1604 for (ent = value_param_list; ent; ent = next_ent) {
1605 int i = get_struct_member_index(get_entity_owner(ent), ent);
1606 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
1608 next_ent = get_entity_link(ent);
1610 DBG((dbg, LEVEL_2, "\targ #%d need backing store\n", i));
1611 set_entity_link(ent, new_list);
1616 /* ok, change the graph */
1617 ir_node *start_bl = get_irg_start_block(irg);
1618 ir_node *first_bl = NULL;
1619 ir_node *frame, *imem, *nmem, *store, *mem, *args, *args_bl;
1620 const ir_edge_t *edge;
1621 optimization_state_t state;
1624 foreach_block_succ(start_bl, edge) {
1625 ir_node *succ = get_edge_src_irn(edge);
1626 if (start_bl != succ) {
1632 /* we had already removed critical edges, so the following
1633 assertion should be always true. */
1634 assert(get_Block_n_cfgpreds(first_bl) == 1);
1636 /* now create backing stores */
1637 frame = get_irg_frame(irg);
1638 imem = get_irg_initial_mem(irg);
1640 save_optimization_state(&state);
1642 nmem = new_r_Proj(irg, first_bl, get_irg_start(irg), mode_M, pn_Start_M);
1643 restore_optimization_state(&state);
1645 /* reroute all edges to the new memory source */
1646 edges_reroute(imem, nmem, irg);
1650 args = get_irg_args(irg);
1651 args_bl = get_nodes_block(args);
1652 for (ent = new_list; ent; ent = get_entity_link(ent)) {
1653 int i = get_struct_member_index(get_entity_owner(ent), ent);
1654 ir_type *tp = get_entity_type(ent);
1655 ir_mode *mode = get_type_mode(tp);
1658 /* address for the backing store */
1659 addr = be_new_FrameAddr(env->isa->sp->reg_class, irg, first_bl, frame, ent);
1662 mem = new_r_Proj(irg, first_bl, store, mode_M, pn_Store_M);
1664 /* the backing store itself */
1665 store = new_r_Store(irg, first_bl, mem, addr,
1666 new_r_Proj(irg, args_bl, args, mode, i));
1668 /* the new memory Proj gets the last Proj from store */
1669 set_Proj_pred(nmem, store);
1670 set_Proj_proj(nmem, pn_Store_M);
1672 /* move all entities to the frame type */
1673 frame_tp = get_irg_frame_type(irg);
1674 offset = get_type_size_bytes(frame_tp);
1675 for (ent = new_list; ent; ent = get_entity_link(ent)) {
1676 ir_type *tp = get_entity_type(ent);
1677 int align = get_type_alignment_bytes(tp);
1679 offset += align - 1;
1681 set_entity_owner(ent, frame_tp);
1682 add_class_member(frame_tp, ent);
1683 /* must be automatic to set a fixed layout */
1684 set_entity_allocation(ent, allocation_automatic);
1685 set_entity_offset(ent, offset);
1686 offset += get_type_size_bytes(tp);
1688 set_type_size_bytes(frame_tp, offset);
1693 * The start block has no jump, instead it has an initial exec Proj.
1694 * The backend wants to handle all blocks the same way, so we replace
1695 * the out cfg edge with a real jump.
1697 static void fix_start_block(ir_node *block, void *env) {
1700 ir_node *start_block;
1703 /* we processed the start block, return */
1707 irg = get_irn_irg(block);
1708 start_block = get_irg_start_block(irg);
1710 for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
1711 ir_node *pred = get_Block_cfgpred(block, i);
1712 ir_node *pred_block = get_nodes_block(pred);
1714 /* ok, we are in the block, having start as cfg predecessor */
1715 if (pred_block == start_block) {
1716 ir_node *jump = new_r_Jmp(irg, pred_block);
1717 set_Block_cfgpred(block, i, jump);
1724 * Modify the irg itself and the frame type.
1726 static void modify_irg(be_abi_irg_t *env)
1728 be_abi_call_t *call = env->call;
1729 const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
1730 const arch_register_t *sp = arch_isa_sp(isa);
1731 ir_graph *irg = env->birg->irg;
1732 ir_node *bl = get_irg_start_block(irg);
1733 ir_node *end = get_irg_end_block(irg);
1734 ir_node *old_mem = get_irg_initial_mem(irg);
1735 ir_node *new_mem_proj;
1737 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1738 pset *dont_save = pset_new_ptr(8);
1744 const arch_register_t *fp_reg;
1745 ir_node *frame_pointer;
1747 ir_node *reg_params_bl;
1750 ir_node *value_param_base;
1751 const ir_edge_t *edge;
1752 ir_type *arg_type, *bet_type;
1753 lower_frame_sels_env_t ctx;
1754 ir_entity **param_map;
1756 bitset_t *used_proj_nr;
1757 DEBUG_ONLY(firm_dbg_module_t *dbg = env->dbg;)
1759 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1761 /* Convert the Sel nodes in the irg to frame load/store/addr nodes. */
1763 ctx.value_param_list = NULL;
1764 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1766 /* value_param_base anchor is not needed anymore now */
1767 value_param_base = get_irg_value_param_base(irg);
1768 be_kill_node(value_param_base);
1769 set_irg_value_param_base(irg, new_r_Bad(irg));
1771 env->frame = obstack_alloc(&env->obst, sizeof(env->frame[0]));
1772 env->regs = pmap_create();
1774 used_proj_nr = bitset_alloca(1024);
1775 n_params = get_method_n_params(method_type);
1776 args = obstack_alloc(&env->obst, n_params * sizeof(args[0]));
1777 memset(args, 0, n_params * sizeof(args[0]));
1779 /* Check if a value parameter is transmitted as a register.
1780 * This might happen if the address of an parameter is taken which is
1781 * transmitted in registers.
1783 * Note that on some architectures this case must be handled specially
1784 * because the place of the backing store is determined by their ABI.
1786 * In the default case we move the entity to the frame type and create
1787 * a backing store into the first block.
1789 fix_address_of_parameter_access(env, ctx.value_param_list);
1791 /* Fill the argument vector */
1792 arg_tuple = get_irg_args(irg);
1793 foreach_out_edge(arg_tuple, edge) {
1794 ir_node *irn = get_edge_src_irn(edge);
1795 int nr = get_Proj_proj(irn);
1797 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1800 arg_type = compute_arg_type(env, call, method_type, ¶m_map);
1801 bet_type = call->cb->get_between_type(env->cb);
1802 stack_frame_init(env->frame, arg_type, bet_type, get_irg_frame_type(irg), isa->stack_dir, param_map);
1804 /* Count the register params and add them to the number of Projs for the RegParams node */
1805 for(i = 0; i < n_params; ++i) {
1806 be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
1807 if(arg->in_reg && args[i]) {
1808 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1809 assert(i == get_Proj_proj(args[i]));
1811 /* For now, associate the register with the old Proj from Start representing that argument. */
1812 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1813 bitset_set(used_proj_nr, i);
1814 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1818 /* Collect all callee-save registers */
1819 for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
1820 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1821 for(j = 0; j < cls->n_regs; ++j) {
1822 const arch_register_t *reg = &cls->regs[j];
1823 if(arch_register_type_is(reg, callee_save) ||
1824 arch_register_type_is(reg, state)) {
1825 pmap_insert(env->regs, (void *) reg, NULL);
1830 pmap_insert(env->regs, (void *) sp, NULL);
1831 pmap_insert(env->regs, (void *) isa->bp, NULL);
1832 reg_params_bl = get_irg_start_block(irg);
1833 env->reg_params = be_new_RegParams(irg, reg_params_bl, pmap_count(env->regs));
1834 add_irn_dep(env->reg_params, get_irg_start(irg));
1837 * make proj nodes for the callee save registers.
1838 * memorize them, since Return nodes get those as inputs.
1840 * Note, that if a register corresponds to an argument, the regs map contains
1841 * the old Proj from start for that argument.
1844 rm = reg_map_to_arr(&env->obst, env->regs);
1845 for(i = 0, n = pmap_count(env->regs); i < n; ++i) {
1846 arch_register_t *reg = (void *) rm[i].reg;
1847 ir_mode *mode = reg->reg_class->mode;
1849 int pos = BE_OUT_POS((int) nr);
1855 bitset_set(used_proj_nr, nr);
1856 proj = new_r_Proj(irg, reg_params_bl, env->reg_params, mode, nr);
1857 pmap_insert(env->regs, (void *) reg, proj);
1858 be_set_constr_single_reg(env->reg_params, pos, reg);
1859 arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
1862 * If the register is an ignore register,
1863 * The Proj for that register shall also be ignored during register allocation.
1865 if(arch_register_type_is(reg, ignore))
1866 flags |= arch_irn_flags_ignore;
1869 flags |= arch_irn_flags_modify_sp;
1871 be_node_set_flags(env->reg_params, pos, flags);
1873 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1875 obstack_free(&env->obst, rm);
1877 /* create a new initial memory proj */
1878 assert(is_Proj(old_mem));
1879 new_mem_proj = new_r_Proj(irg, get_nodes_block(old_mem),
1880 new_r_Unknown(irg, mode_T), mode_M,
1881 get_Proj_proj(old_mem));
1884 /* Generate the Prologue */
1885 fp_reg = call->cb->prologue(env->cb, &mem, env->regs);
1887 /* do the stack allocation BEFORE the barrier, or spill code
1888 might be added before it */
1889 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1890 env->init_sp = be_new_IncSP(sp, irg, bl, env->init_sp, BE_STACK_FRAME_SIZE_EXPAND);
1891 be_abi_reg_map_set(env->regs, sp, env->init_sp);
1893 env->start_barrier = barrier = create_barrier(env, bl, &mem, env->regs, 0);
1895 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1896 arch_set_irn_register(env->birg->main_env->arch_env, env->init_sp, sp);
1898 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1899 set_irg_frame(irg, frame_pointer);
1900 pset_insert_ptr(env->ignore_regs, fp_reg);
1902 /* rewire old mem users to new mem */
1903 set_Proj_pred(new_mem_proj, get_Proj_pred(old_mem));
1904 exchange(old_mem, mem);
1906 set_irg_initial_mem(irg, mem);
1908 /* Now, introduce stack param nodes for all parameters passed on the stack */
1909 for(i = 0; i < n_params; ++i) {
1910 ir_node *arg_proj = args[i];
1911 ir_node *repl = NULL;
1913 if(arg_proj != NULL) {
1914 be_abi_call_arg_t *arg;
1915 ir_type *param_type;
1916 int nr = get_Proj_proj(arg_proj);
1918 nr = MIN(nr, n_params);
1919 arg = get_call_arg(call, 0, nr);
1920 param_type = get_method_param_type(method_type, nr);
1923 repl = pmap_get(env->regs, (void *) arg->reg);
1926 else if(arg->on_stack) {
1927 /* For atomic parameters which are actually used, we create a StackParam node. */
1928 if(is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1929 ir_mode *mode = get_type_mode(param_type);
1930 const arch_register_class_t *cls = arch_isa_get_reg_class_for_mode(isa, mode);
1931 repl = be_new_StackParam(cls, isa->bp->reg_class, irg, reg_params_bl, mode, frame_pointer, arg->stack_ent);
1934 /* The stack parameter is not primitive (it is a struct or array),
1935 we thus will create a node representing the parameter's address
1938 repl = be_new_FrameAddr(sp->reg_class, irg, reg_params_bl, frame_pointer, arg->stack_ent);
1942 assert(repl != NULL);
1943 exchange(args[i], repl);
1947 /* the arg proj is not needed anymore now */
1948 assert(get_irn_n_edges(arg_tuple) == 0);
1949 be_kill_node(arg_tuple);
1950 set_irg_args(irg, new_rd_Bad(irg));
1952 /* All Return nodes hang on the End node, so look for them there. */
1953 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1954 ir_node *irn = get_Block_cfgpred(end, i);
1956 if (is_Return(irn)) {
1957 ir_node *ret = create_be_return(env, irn, get_nodes_block(irn), get_Return_mem(irn), get_Return_n_ress(irn));
1961 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1962 the code is dead and will never be executed. */
1964 del_pset(dont_save);
1965 obstack_free(&env->obst, args);
1967 /* handle start block here (place a jump in the block) */
1969 irg_block_walk_graph(irg, fix_start_block, NULL, &temp);
1972 /** Fix the state inputs of calls that still hang on unknowns */
1974 void fix_call_state_inputs(be_abi_irg_t *env)
1976 const arch_isa_t *isa = env->isa;
1978 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1980 /* Collect caller save registers */
1981 n = arch_isa_get_n_reg_class(isa);
1982 for(i = 0; i < n; ++i) {
1984 const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
1985 for(j = 0; j < cls->n_regs; ++j) {
1986 const arch_register_t *reg = arch_register_for_index(cls, j);
1987 if(arch_register_type_is(reg, state)) {
1988 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1993 n = ARR_LEN(env->calls);
1994 n_states = ARR_LEN(stateregs);
1995 for(i = 0; i < n; ++i) {
1997 ir_node *call = env->calls[i];
1999 arity = get_irn_arity(call);
2001 /* the statereg inputs are the last n inputs of the calls */
2002 for(s = 0; s < n_states; ++s) {
2003 int inp = arity - n_states + s;
2004 const arch_register_t *reg = stateregs[s];
2005 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
2007 set_irn_n(call, inp, regnode);
2012 be_abi_irg_t *be_abi_introduce(be_irg_t *birg)
2014 be_abi_irg_t *env = xmalloc(sizeof(env[0]));
2015 ir_node *old_frame = get_irg_frame(birg->irg);
2016 ir_graph *irg = birg->irg;
2020 optimization_state_t state;
2021 unsigned *limited_bitset;
2023 be_omit_fp = birg->main_env->options->omit_fp;
2025 obstack_init(&env->obst);
2027 env->isa = birg->main_env->arch_env->isa;
2028 env->method_type = get_entity_type(get_irg_entity(irg));
2029 env->call = be_abi_call_new(env->isa->sp->reg_class);
2030 arch_isa_get_call_abi(env->isa, env->method_type, env->call);
2032 env->ignore_regs = pset_new_ptr_default();
2033 env->keep_map = pmap_create();
2034 env->dce_survivor = new_survive_dce();
2037 env->sp_req.type = arch_register_req_type_limited;
2038 env->sp_req.cls = arch_register_get_class(env->isa->sp);
2039 limited_bitset = rbitset_obstack_alloc(&env->obst, env->sp_req.cls->n_regs);
2040 rbitset_set(limited_bitset, arch_register_get_index(env->isa->sp));
2041 env->sp_req.limited = limited_bitset;
2043 env->sp_cls_req.type = arch_register_req_type_normal;
2044 env->sp_cls_req.cls = arch_register_get_class(env->isa->sp);
2046 /* Beware: later we replace this node by the real one, ensure it is not CSE'd
2047 to another Unknown or the stack pointer gets used */
2048 save_optimization_state(&state);
2050 env->init_sp = dummy = new_r_Unknown(irg, env->isa->sp->reg_class->mode);
2051 restore_optimization_state(&state);
2052 FIRM_DBG_REGISTER(env->dbg, "firm.be.abi");
2054 env->calls = NEW_ARR_F(ir_node*, 0);
2056 /* Lower all call nodes in the IRG. */
2060 Beware: init backend abi call object after processing calls,
2061 otherwise some information might be not yet available.
2063 env->cb = env->call->cb->init(env->call, birg->main_env->arch_env, irg);
2065 /* Process the IRG */
2068 /* fix call inputs for state registers */
2069 fix_call_state_inputs(env);
2071 /* We don't need the keep map anymore. */
2072 pmap_destroy(env->keep_map);
2074 /* calls array is not needed anymore */
2075 DEL_ARR_F(env->calls);
2077 /* reroute the stack origin of the calls to the true stack origin. */
2078 exchange(dummy, env->init_sp);
2079 exchange(old_frame, get_irg_frame(irg));
2081 /* Make some important node pointers survive the dead node elimination. */
2082 survive_dce_register_irn(env->dce_survivor, &env->init_sp);
2083 pmap_foreach(env->regs, ent) {
2084 survive_dce_register_irn(env->dce_survivor, (ir_node **) &ent->value);
2087 env->call->cb->done(env->cb);
2092 void be_abi_free(be_abi_irg_t *env)
2094 be_abi_call_free(env->call);
2095 free_survive_dce(env->dce_survivor);
2096 del_pset(env->ignore_regs);
2097 pmap_destroy(env->regs);
2098 obstack_free(&env->obst, NULL);
2102 void be_abi_put_ignore_regs(be_abi_irg_t *abi, const arch_register_class_t *cls, bitset_t *bs)
2104 arch_register_t *reg;
2106 for(reg = pset_first(abi->ignore_regs); reg; reg = pset_next(abi->ignore_regs))
2107 if(reg->reg_class == cls)
2108 bitset_set(bs, reg->index);
2111 /* Returns the stack layout from a abi environment. */
2112 const be_stack_layout_t *be_abi_get_stack_layout(const be_abi_irg_t *abi) {
2119 | ___(_)_ __ / ___|| |_ __ _ ___| | __
2120 | |_ | \ \/ / \___ \| __/ _` |/ __| |/ /
2121 | _| | |> < ___) | || (_| | (__| <
2122 |_| |_/_/\_\ |____/ \__\__,_|\___|_|\_\
2126 typedef ir_node **node_array;
2128 typedef struct fix_stack_walker_env_t {
2129 node_array sp_nodes;
2130 const arch_env_t *arch_env;
2131 } fix_stack_walker_env_t;
2134 * Walker. Collect all stack modifying nodes.
2136 static void collect_stack_nodes_walker(ir_node *node, void *data)
2138 fix_stack_walker_env_t *env = data;
2140 if (arch_irn_is(env->arch_env, node, modify_sp)) {
2141 assert(get_irn_mode(node) != mode_M && get_irn_mode(node) != mode_T);
2142 ARR_APP1(ir_node*, env->sp_nodes, node);
2146 void be_abi_fix_stack_nodes(be_abi_irg_t *env)
2148 be_ssa_construction_env_t senv;
2151 be_irg_t *birg = env->birg;
2152 be_lv_t *lv = be_get_birg_liveness(birg);
2153 fix_stack_walker_env_t walker_env;
2156 walker_env.sp_nodes = NEW_ARR_F(ir_node*, 0);
2157 walker_env.arch_env = birg->main_env->arch_env;
2158 isa = walker_env.arch_env->isa;
2160 irg_walk_graph(birg->irg, collect_stack_nodes_walker, NULL, &walker_env);
2162 /* nothing to be done if we didn't find any node, in fact we mustn't
2163 * continue, as for endless loops incsp might have had no users and is bad
2166 len = ARR_LEN(walker_env.sp_nodes);
2168 DEL_ARR_F(walker_env.sp_nodes);
2172 be_ssa_construction_init(&senv, birg);
2173 be_ssa_construction_add_copies(&senv, walker_env.sp_nodes,
2174 ARR_LEN(walker_env.sp_nodes));
2175 be_ssa_construction_fix_users_array(&senv, walker_env.sp_nodes,
2176 ARR_LEN(walker_env.sp_nodes));
2179 len = ARR_LEN(walker_env.sp_nodes);
2180 for(i = 0; i < len; ++i) {
2181 be_liveness_update(lv, walker_env.sp_nodes[i]);
2183 be_ssa_construction_update_liveness_phis(&senv, lv);
2186 phis = be_ssa_construction_get_new_phis(&senv);
2188 /* set register requirements for stack phis */
2189 len = ARR_LEN(phis);
2190 for(i = 0; i < len; ++i) {
2191 ir_node *phi = phis[i];
2192 be_set_phi_reg_req(walker_env.arch_env, phi, &env->sp_req);
2193 be_set_phi_flags(walker_env.arch_env, phi, arch_irn_flags_ignore | arch_irn_flags_modify_sp);
2194 arch_set_irn_register(walker_env.arch_env, phi, env->isa->sp);
2196 be_ssa_construction_destroy(&senv);
2198 DEL_ARR_F(walker_env.sp_nodes);
2201 static int process_stack_bias(be_abi_irg_t *env, ir_node *bl, int bias)
2203 const arch_env_t *arch_env = env->birg->main_env->arch_env;
2204 int omit_fp = env->call->flags.bits.try_omit_fp;
2207 sched_foreach(bl, irn) {
2210 Check, if the node relates to an entity on the stack frame.
2211 If so, set the true offset (including the bias) for that
2214 ir_entity *ent = arch_get_frame_entity(arch_env, irn);
2216 int offset = get_stack_entity_offset(env->frame, ent, bias);
2217 arch_set_frame_offset(arch_env, irn, offset);
2218 DBG((env->dbg, LEVEL_2, "%F has offset %d (including bias %d)\n", ent, offset, bias));
2222 If the node modifies the stack pointer by a constant offset,
2223 record that in the bias.
2225 if(arch_irn_is(arch_env, irn, modify_sp)) {
2226 int ofs = arch_get_sp_bias(arch_env, irn);
2228 if(be_is_IncSP(irn)) {
2229 if(ofs == BE_STACK_FRAME_SIZE_EXPAND) {
2230 ofs = get_type_size_bytes(get_irg_frame_type(env->birg->irg));
2231 be_set_IncSP_offset(irn, ofs);
2232 } else if(ofs == BE_STACK_FRAME_SIZE_SHRINK) {
2233 ofs = - get_type_size_bytes(get_irg_frame_type(env->birg->irg));
2234 be_set_IncSP_offset(irn, ofs);
2247 * A helper struct for the bias walker.
2250 be_abi_irg_t *env; /**< The ABI irg environment. */
2251 int start_block_bias; /**< The bias at the end of the start block. */
2252 ir_node *start_block; /**< The start block of the current graph. */
2256 * Block-Walker: fix all stack offsets
2258 static void stack_bias_walker(ir_node *bl, void *data)
2260 struct bias_walk *bw = data;
2261 if (bl != bw->start_block) {
2262 process_stack_bias(bw->env, bl, bw->start_block_bias);
2266 void be_abi_fix_stack_bias(be_abi_irg_t *env)
2268 ir_graph *irg = env->birg->irg;
2269 struct bias_walk bw;
2271 stack_frame_compute_initial_offset(env->frame);
2272 // stack_layout_dump(stdout, env->frame);
2274 /* Determine the stack bias at the end of the start block. */
2275 bw.start_block_bias = process_stack_bias(env, get_irg_start_block(irg), 0);
2277 /* fix the bias is all other blocks */
2279 bw.start_block = get_irg_start_block(irg);
2280 irg_block_walk_graph(irg, stack_bias_walker, NULL, &bw);
2283 ir_node *be_abi_get_callee_save_irn(be_abi_irg_t *abi, const arch_register_t *reg)
2285 assert(arch_register_type_is(reg, callee_save));
2286 assert(pmap_contains(abi->regs, (void *) reg));
2287 return pmap_get(abi->regs, (void *) reg);
2290 ir_node *be_abi_get_ignore_irn(be_abi_irg_t *abi, const arch_register_t *reg)
2292 assert(arch_register_type_is(reg, ignore));
2293 assert(pmap_contains(abi->regs, (void *) reg));
2294 return pmap_get(abi->regs, (void *) reg);
2297 ir_node *be_abi_get_start_barrier(be_abi_irg_t *abi)
2299 return abi->start_barrier;
2303 * Returns non-zero if the ABI has omitted the frame pointer in
2304 * the current graph.
2306 int be_abi_omit_fp(const be_abi_irg_t *abi) {
2307 return abi->call->flags.bits.try_omit_fp;