2 * Copyright (C) 1995-2011 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
32 #include "irgraph_t.h"
35 #include "iredges_t.h"
38 #include "irprintf_t.h"
45 #include "raw_bitset.h"
56 #include "bessaconstr.h"
58 #include "betranshlp.h"
60 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
62 typedef struct be_abi_call_arg_t {
63 unsigned is_res : 1; /**< 1: the call argument is a return value. 0: it's a call parameter. */
64 unsigned in_reg : 1; /**< 1: this argument is transmitted in registers. */
65 unsigned on_stack : 1; /**< 1: this argument is transmitted on the stack. */
66 unsigned callee : 1; /**< 1: someone called us. 0: We call another function */
69 const arch_register_t *reg;
72 unsigned alignment; /**< stack alignment */
73 unsigned space_before; /**< allocate space before */
74 unsigned space_after; /**< allocate space after */
77 struct be_abi_call_t {
78 be_abi_call_flags_t flags; /**< Flags describing the ABI behavior on calls */
79 int pop; /**< number of bytes the stack frame is shrinked by the callee on return. */
80 const be_abi_callbacks_t *cb;
81 ir_type *between_type;
83 const arch_register_class_t *cls_addr; /**< register class of the call address */
87 * The ABI information for the current graph.
90 be_abi_call_t *call; /**< The ABI call information. */
92 ir_node *init_sp; /**< The node representing the stack pointer
93 at the start of the function. */
95 ir_node *start; /**< The be_Start 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. */
105 ir_node **calls; /**< flexible array containing all be_Call nodes */
108 static ir_heights_t *ir_heights;
110 /** Flag: if set, try to omit the frame pointer in all routines. */
111 static int be_omit_fp = 1;
114 _ ____ ___ ____ _ _ _ _
115 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
116 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
117 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
118 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
120 These callbacks are used by the backend to set the parameters
121 for a specific call type.
125 * Set compare function: compares two ABI call object arguments.
127 static int cmp_call_arg(const void *a, const void *b, size_t n)
129 const be_abi_call_arg_t *p = (const be_abi_call_arg_t*)a;
130 const be_abi_call_arg_t *q = (const be_abi_call_arg_t*)b;
132 return !(p->is_res == q->is_res && p->pos == q->pos && p->callee == q->callee);
136 * Get an ABI call object argument.
138 * @param call the abi call
139 * @param is_res true for call results, false for call arguments
140 * @param pos position of the argument
141 * @param callee context type - if we are callee or caller
143 static be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos, int callee)
145 be_abi_call_arg_t arg;
148 memset(&arg, 0, sizeof(arg));
153 hash = is_res * 128 + pos;
155 return (be_abi_call_arg_t*)set_find(call->params, &arg, sizeof(arg), hash);
159 * Set an ABI call object argument.
161 static void remember_call_arg(be_abi_call_arg_t *arg, be_abi_call_t *call, be_abi_context_t context)
163 unsigned hash = arg->is_res * 128 + arg->pos;
164 if (context & ABI_CONTEXT_CALLEE) {
166 set_insert(call->params, arg, sizeof(*arg), hash);
168 if (context & ABI_CONTEXT_CALLER) {
170 set_insert(call->params, arg, sizeof(*arg), hash);
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)
181 /* Sets the number of bytes the stackframe is shrinked by the callee on return */
182 void be_abi_call_set_pop(be_abi_call_t *call, int pop)
188 /* Set register class for call address */
189 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
191 call->cls_addr = cls;
195 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos,
196 ir_mode *load_mode, unsigned alignment,
197 unsigned space_before, unsigned space_after,
198 be_abi_context_t context)
200 be_abi_call_arg_t arg;
201 memset(&arg, 0, sizeof(arg));
202 assert(alignment > 0 && "Alignment must be greater than 0");
204 arg.load_mode = load_mode;
205 arg.alignment = alignment;
206 arg.space_before = space_before;
207 arg.space_after = space_after;
211 remember_call_arg(&arg, call, context);
214 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
216 be_abi_call_arg_t arg;
217 memset(&arg, 0, sizeof(arg));
224 remember_call_arg(&arg, call, context);
227 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
229 be_abi_call_arg_t arg;
230 memset(&arg, 0, sizeof(arg));
237 remember_call_arg(&arg, call, context);
240 /* Get the flags of a ABI call object. */
241 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
247 * Constructor for a new ABI call object.
249 * @param cls_addr register class of the call address
251 * @return the new ABI call object
253 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
255 be_abi_call_t *call = XMALLOCZ(be_abi_call_t);
258 call->params = new_set(cmp_call_arg, 16);
260 call->cls_addr = cls_addr;
262 call->flags.bits.try_omit_fp = be_omit_fp;
268 * Destructor for an ABI call object.
270 static void be_abi_call_free(be_abi_call_t *call)
272 del_set(call->params);
277 * Initializes the frame layout from parts
279 * @param frame the stack layout that will be initialized
280 * @param args the stack argument layout type
281 * @param between the between layout type
282 * @param locals the method frame type
283 * @param param_map an array mapping method argument positions to the stack argument type
285 * @return the initialized stack layout
287 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
288 ir_type *between, ir_type *locals,
289 ir_entity *param_map[])
291 frame->arg_type = args;
292 frame->between_type = between;
293 frame->frame_type = locals;
294 frame->initial_offset = 0;
295 frame->initial_bias = 0;
296 frame->order[1] = between;
297 frame->param_map = param_map;
299 /* typical decreasing stack: locals have the
300 * lowest addresses, arguments the highest */
301 frame->order[0] = locals;
302 frame->order[2] = args;
313 Adjustment of the calls inside a graph.
318 * Transform a call node into a be_Call node.
320 * @param env The ABI environment for the current irg.
321 * @param irn The call node.
322 * @param curr_sp The stack pointer node to use.
323 * @return The stack pointer after the call.
325 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp)
327 ir_graph *irg = get_irn_irg(irn);
328 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
329 ir_type *call_tp = get_Call_type(irn);
330 ir_node *call_ptr = get_Call_ptr(irn);
331 size_t n_params = get_method_n_params(call_tp);
332 ir_node *curr_mem = get_Call_mem(irn);
333 ir_node *bl = get_nodes_block(irn);
335 const arch_register_t *sp = arch_env->sp;
336 be_abi_call_t *call = be_abi_call_new(sp->reg_class);
337 ir_mode *mach_mode = sp->reg_class->mode;
338 int no_alloc = call->flags.bits.frame_is_setup_on_call;
339 int n_res = get_method_n_ress(call_tp);
340 int do_seq = call->flags.bits.store_args_sequential && !no_alloc;
342 ir_node *res_proj = NULL;
343 int n_reg_params = 0;
344 int n_stack_params = 0;
347 const arch_register_t **states = NEW_ARR_F(const arch_register_t*, 0);
348 const arch_register_t **destroyed_regs = NEW_ARR_F(const arch_register_t*, 0);
352 int n_reg_results = 0;
353 const ir_edge_t *edge;
355 int *stack_param_idx;
356 int i, n, destroy_all_regs;
361 /* Let the isa fill out the abi description for that call node. */
362 arch_env_get_call_abi(arch_env, call_tp, call);
364 /* Insert code to put the stack arguments on the stack. */
365 assert(get_Call_n_params(irn) == n_params);
366 stack_param_idx = ALLOCAN(int, n_params);
367 for (p = 0; p < n_params; ++p) {
368 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
371 int arg_size = get_type_size_bytes(get_method_param_type(call_tp, p));
373 stack_size += round_up2(arg->space_before, arg->alignment);
374 stack_size += round_up2(arg_size, arg->alignment);
375 stack_size += round_up2(arg->space_after, arg->alignment);
377 stack_param_idx[n_stack_params++] = p;
381 /* Collect all arguments which are passed in registers. */
382 reg_param_idxs = ALLOCAN(int, n_params);
383 for (p = 0; p < n_params; ++p) {
384 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
385 if (arg && arg->in_reg) {
386 reg_param_idxs[n_reg_params++] = p;
391 * If the stack is decreasing and we do not want to store sequentially,
392 * or someone else allocated the call frame
393 * we allocate as much space on the stack all parameters need, by
394 * moving the stack pointer along the stack's direction.
396 * Note: we also have to do this for stack_size == 0, because we may have
397 * to adjust stack alignment for the call.
399 if (!do_seq && !no_alloc) {
400 curr_sp = be_new_IncSP(sp, bl, curr_sp, stack_size, 1);
403 dbgi = get_irn_dbg_info(irn);
404 /* If there are some parameters which shall be passed on the stack. */
405 if (n_stack_params > 0) {
407 ir_node **in = ALLOCAN(ir_node*, n_stack_params+1);
411 * Reverse list of stack parameters if call arguments are from left to right.
412 * We must them reverse again if they are pushed (not stored) and the stack
413 * direction is downwards.
415 if (call->flags.bits.left_to_right ^ do_seq) {
416 for (i = 0; i < n_stack_params >> 1; ++i) {
417 int other = n_stack_params - i - 1;
418 int tmp = stack_param_idx[i];
419 stack_param_idx[i] = stack_param_idx[other];
420 stack_param_idx[other] = tmp;
424 curr_mem = get_Call_mem(irn);
426 in[n_in++] = curr_mem;
429 for (i = 0; i < n_stack_params; ++i) {
430 int p = stack_param_idx[i];
431 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
432 ir_node *param = get_Call_param(irn, p);
433 ir_node *addr = curr_sp;
435 ir_type *param_type = get_method_param_type(call_tp, p);
436 int param_size = get_type_size_bytes(param_type) + arg->space_after;
439 * If we wanted to build the arguments sequentially,
440 * the stack pointer for the next must be incremented,
441 * and the memory value propagated.
445 addr = curr_sp = be_new_IncSP(sp, bl, curr_sp,
446 param_size + arg->space_before, 0);
447 add_irn_dep(curr_sp, curr_mem);
449 curr_ofs += arg->space_before;
450 curr_ofs = round_up2(curr_ofs, arg->alignment);
452 /* Make the expression to compute the argument's offset. */
454 ir_mode *constmode = mach_mode;
455 if (mode_is_reference(mach_mode)) {
458 addr = new_r_Const_long(irg, constmode, curr_ofs);
459 addr = new_r_Add(bl, curr_sp, addr, mach_mode);
463 /* Insert a store for primitive arguments. */
464 if (is_atomic_type(param_type)) {
465 ir_node *mem_input = do_seq ? curr_mem : new_r_NoMem(irg);
466 ir_node *store = new_rd_Store(dbgi, bl, mem_input, addr, param, cons_none);
467 mem = new_r_Proj(store, mode_M, pn_Store_M);
469 /* Make a mem copy for compound arguments. */
472 assert(mode_is_reference(get_irn_mode(param)));
473 copy = new_rd_CopyB(dbgi, bl, curr_mem, addr, param, param_type);
474 mem = new_r_Proj(copy, mode_M, pn_CopyB_M);
477 curr_ofs += param_size;
485 /* We need the sync only, if we didn't build the stores sequentially. */
487 if (n_stack_params >= 1) {
488 curr_mem = new_r_Sync(bl, n_in, in);
490 curr_mem = get_Call_mem(irn);
495 /* check for the return_twice property */
496 destroy_all_regs = 0;
497 if (is_SymConst_addr_ent(call_ptr)) {
498 ir_entity *ent = get_SymConst_entity(call_ptr);
500 if (get_entity_additional_properties(ent) & mtp_property_returns_twice)
501 destroy_all_regs = 1;
503 ir_type *call_tp = get_Call_type(irn);
505 if (get_method_additional_properties(call_tp) & mtp_property_returns_twice)
506 destroy_all_regs = 1;
509 /* Put caller save into the destroyed set and state registers in the states
511 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
513 const arch_register_class_t *cls = &arch_env->register_classes[i];
514 for (j = 0; j < cls->n_regs; ++j) {
515 const arch_register_t *reg = arch_register_for_index(cls, j);
517 /* even if destroyed all is specified, neither SP nor FP are
518 * destroyed (else bad things will happen) */
519 if (reg == arch_env->sp || reg == arch_env->bp)
522 if (reg->type & arch_register_type_state) {
523 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
524 ARR_APP1(const arch_register_t*, states, reg);
525 /* we're already in the destroyed set so no need for further
529 if (destroy_all_regs || (reg->type & arch_register_type_caller_save)) {
530 if (!(reg->type & arch_register_type_ignore)) {
531 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
537 /* search the largest result proj number */
538 res_projs = ALLOCANZ(ir_node*, n_res);
540 foreach_out_edge(irn, edge) {
541 const ir_edge_t *res_edge;
542 ir_node *irn = get_edge_src_irn(edge);
544 if (!is_Proj(irn) || get_Proj_proj(irn) != pn_Call_T_result)
547 foreach_out_edge(irn, res_edge) {
549 ir_node *res = get_edge_src_irn(res_edge);
551 assert(is_Proj(res));
553 proj = get_Proj_proj(res);
554 assert(proj < n_res);
555 assert(res_projs[proj] == NULL);
556 res_projs[proj] = res;
562 /** TODO: this is not correct for cases where return values are passed
563 * on the stack, but no known ABI does this currently...
565 n_reg_results = n_res;
568 in = ALLOCAN(ir_node*, n_reg_params + ARR_LEN(states));
570 /* make the back end call node and set its register requirements. */
571 for (i = 0; i < n_reg_params; ++i) {
572 in[n_ins++] = get_Call_param(irn, reg_param_idxs[i]);
575 /* add state registers ins */
576 for (s = 0; s < ARR_LEN(states); ++s) {
577 const arch_register_t *reg = states[s];
578 const arch_register_class_t *cls = arch_register_get_class(reg);
580 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
581 ir_fprintf(stderr, "Adding %+F\n", regnode);
583 ir_node *regnode = new_r_Unknown(irg, arch_register_class_mode(cls));
584 in[n_ins++] = regnode;
586 assert(n_ins == (int) (n_reg_params + ARR_LEN(states)));
588 /* ins collected, build the call */
589 if (env->call->flags.bits.call_has_imm && is_SymConst(call_ptr)) {
591 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, curr_sp,
592 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
593 n_ins, in, get_Call_type(irn));
594 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
597 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, call_ptr,
598 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
599 n_ins, in, get_Call_type(irn));
601 be_Call_set_pop(low_call, call->pop);
603 /* put the call into the list of all calls for later processing */
604 ARR_APP1(ir_node *, env->calls, low_call);
606 /* create new stack pointer */
607 curr_sp = new_r_Proj(low_call, get_irn_mode(curr_sp), pn_be_Call_sp);
608 be_set_constr_single_reg_out(low_call, pn_be_Call_sp, sp,
609 arch_register_req_type_ignore | arch_register_req_type_produces_sp);
610 arch_set_irn_register(curr_sp, sp);
612 /* now handle results */
613 for (i = 0; i < n_res; ++i) {
615 ir_node *proj = res_projs[i];
616 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
618 /* returns values on stack not supported yet */
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 pn = i + pn_be_Call_first_res;
629 ir_type *res_type = get_method_res_type(call_tp, i);
630 ir_mode *mode = get_type_mode(res_type);
631 proj = new_r_Proj(low_call, mode, pn);
634 set_Proj_pred(proj, low_call);
635 set_Proj_proj(proj, pn);
639 /* remove register from destroyed regs */
641 size_t n = ARR_LEN(destroyed_regs);
642 for (j = 0; j < n; ++j) {
643 if (destroyed_regs[j] == arg->reg) {
644 destroyed_regs[j] = destroyed_regs[n-1];
645 ARR_SHRINKLEN(destroyed_regs,n-1);
653 Set the register class of the call address to
654 the backend provided class (default: stack pointer class)
656 be_node_set_reg_class_in(low_call, n_be_Call_ptr, call->cls_addr);
658 DBG((dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
660 /* Set the register classes and constraints of the Call parameters. */
661 for (i = 0; i < n_reg_params; ++i) {
662 int index = reg_param_idxs[i];
663 be_abi_call_arg_t *arg = get_call_arg(call, 0, index, 0);
664 assert(arg->reg != NULL);
666 be_set_constr_single_reg_in(low_call, n_be_Call_first_arg + i,
667 arg->reg, arch_register_req_type_none);
670 /* Set the register constraints of the results. */
671 for (i = 0; i < n_res; ++i) {
672 ir_node *proj = res_projs[i];
673 const be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
674 int pn = get_Proj_proj(proj);
677 be_set_constr_single_reg_out(low_call, pn, arg->reg,
678 arch_register_req_type_none);
679 arch_set_irn_register(proj, arg->reg);
681 exchange(irn, low_call);
683 /* kill the ProjT node */
684 if (res_proj != NULL) {
688 /* Make additional projs for the caller save registers
689 and the Keep node which keeps them alive. */
695 int curr_res_proj = pn_be_Call_first_res + n_reg_results;
698 n_ins = ARR_LEN(destroyed_regs) + n_reg_results + 1;
699 in = ALLOCAN(ir_node *, n_ins);
701 /* also keep the stack pointer */
702 set_irn_link(curr_sp, (void*) sp);
705 for (d = 0; d < ARR_LEN(destroyed_regs); ++d) {
706 const arch_register_t *reg = destroyed_regs[d];
707 ir_node *proj = new_r_Proj(low_call, reg->reg_class->mode, curr_res_proj);
709 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
710 be_set_constr_single_reg_out(low_call, curr_res_proj, reg,
711 arch_register_req_type_none);
712 arch_set_irn_register(proj, reg);
714 set_irn_link(proj, (void*) reg);
719 for (i = 0; i < n_reg_results; ++i) {
720 ir_node *proj = res_projs[i];
721 const arch_register_t *reg = arch_get_irn_register(proj);
722 set_irn_link(proj, (void*) reg);
727 /* create the Keep for the caller save registers */
728 keep = be_new_Keep(bl, n, in);
729 for (i = 0; i < n; ++i) {
730 const arch_register_t *reg = (const arch_register_t*)get_irn_link(in[i]);
731 be_node_set_reg_class_in(keep, i, reg->reg_class);
735 /* Clean up the stack. */
736 assert(stack_size >= call->pop);
737 stack_size -= call->pop;
739 if (stack_size > 0) {
740 ir_node *mem_proj = NULL;
742 foreach_out_edge(low_call, edge) {
743 ir_node *irn = get_edge_src_irn(edge);
744 if (is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
751 mem_proj = new_r_Proj(low_call, mode_M, pn_be_Call_M_regular);
752 keep_alive(mem_proj);
755 /* Clean up the stack frame or revert alignment fixes if we allocated it */
757 curr_sp = be_new_IncSP(sp, bl, curr_sp, -stack_size, 0);
760 be_abi_call_free(call);
763 DEL_ARR_F(destroyed_regs);
769 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
771 * @param alignment the minimum stack alignment
772 * @param size the node containing the non-aligned size
773 * @param block the block where new nodes are allocated on
774 * @param dbg debug info for new nodes
776 * @return a node representing the aligned size
778 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size,
779 ir_node *block, dbg_info *dbg)
781 if (stack_alignment > 1) {
787 assert(is_po2(stack_alignment));
789 mode = get_irn_mode(size);
790 tv = new_tarval_from_long(stack_alignment-1, mode);
791 irg = get_Block_irg(block);
792 mask = new_r_Const(irg, tv);
793 size = new_rd_Add(dbg, block, size, mask, mode);
795 tv = new_tarval_from_long(-(long)stack_alignment, mode);
796 mask = new_r_Const(irg, tv);
797 size = new_rd_And(dbg, block, size, mask, mode);
803 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
805 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp)
807 ir_node *block = get_nodes_block(alloc);
808 ir_graph *irg = get_Block_irg(block);
809 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
810 ir_node *alloc_mem = NULL;
811 ir_node *alloc_res = NULL;
812 ir_type *type = get_Alloc_type(alloc);
815 const ir_edge_t *edge;
820 unsigned stack_alignment;
822 /* all non-stack Alloc nodes should already be lowered before the backend */
823 assert(get_Alloc_where(alloc) == stack_alloc);
825 foreach_out_edge(alloc, edge) {
826 ir_node *irn = get_edge_src_irn(edge);
828 assert(is_Proj(irn));
829 switch (get_Proj_proj(irn)) {
841 /* Beware: currently Alloc nodes without a result might happen,
842 only escape analysis kills them and this phase runs only for object
843 oriented source. We kill the Alloc here. */
844 if (alloc_res == NULL && alloc_mem) {
845 exchange(alloc_mem, get_Alloc_mem(alloc));
849 dbg = get_irn_dbg_info(alloc);
850 count = get_Alloc_count(alloc);
852 /* we might need to multiply the count with the element size */
853 if (type != firm_unknown_type && get_type_size_bytes(type) != 1) {
854 ir_mode *mode = get_irn_mode(count);
855 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type),
857 ir_node *cnst = new_rd_Const(dbg, irg, tv);
858 size = new_rd_Mul(dbg, block, count, cnst, mode);
863 /* The stack pointer will be modified in an unknown manner.
864 We cannot omit it. */
865 env->call->flags.bits.try_omit_fp = 0;
867 stack_alignment = 1 << arch_env->stack_alignment;
868 size = adjust_alloc_size(stack_alignment, size, block, dbg);
869 new_alloc = be_new_AddSP(arch_env->sp, block, curr_sp, size);
870 set_irn_dbg_info(new_alloc, dbg);
872 if (alloc_mem != NULL) {
876 addsp_mem = new_r_Proj(new_alloc, mode_M, pn_be_AddSP_M);
878 /* We need to sync the output mem of the AddSP with the input mem
879 edge into the alloc node. */
880 ins[0] = get_Alloc_mem(alloc);
882 sync = new_r_Sync(block, 2, ins);
884 exchange(alloc_mem, sync);
887 exchange(alloc, new_alloc);
889 /* fix projnum of alloca res */
890 set_Proj_proj(alloc_res, pn_be_AddSP_res);
892 curr_sp = new_r_Proj(new_alloc, get_irn_mode(curr_sp), pn_be_AddSP_sp);
899 * The Free is transformed into a back end free node and connected to the stack nodes.
901 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
903 ir_node *block = get_nodes_block(free);
904 ir_graph *irg = get_irn_irg(free);
905 ir_type *type = get_Free_type(free);
906 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
907 ir_mode *sp_mode = arch_env->sp->reg_class->mode;
908 dbg_info *dbg = get_irn_dbg_info(free);
909 ir_node *subsp, *mem, *res, *size, *sync;
911 unsigned stack_alignment;
913 /* all non-stack-alloc Free nodes should already be lowered before the
915 assert(get_Free_where(free) == stack_alloc);
917 /* we might need to multiply the size with the element size */
918 if (type != firm_unknown_type && get_type_size_bytes(type) != 1) {
919 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
920 ir_node *cnst = new_rd_Const(dbg, irg, tv);
921 ir_node *mul = new_rd_Mul(dbg, block, get_Free_size(free),
925 size = get_Free_size(free);
928 stack_alignment = 1 << arch_env->stack_alignment;
929 size = adjust_alloc_size(stack_alignment, size, block, dbg);
931 /* The stack pointer will be modified in an unknown manner.
932 We cannot omit it. */
933 env->call->flags.bits.try_omit_fp = 0;
934 subsp = be_new_SubSP(arch_env->sp, block, curr_sp, size);
935 set_irn_dbg_info(subsp, dbg);
937 mem = new_r_Proj(subsp, mode_M, pn_be_SubSP_M);
938 res = new_r_Proj(subsp, sp_mode, pn_be_SubSP_sp);
940 /* we need to sync the memory */
941 in[0] = get_Free_mem(free);
943 sync = new_r_Sync(block, 2, in);
945 /* and make the AddSP dependent on the former memory */
946 add_irn_dep(subsp, get_Free_mem(free));
949 exchange(free, sync);
956 * Check if a node is somehow data dependent on another one.
957 * both nodes must be in the same basic block.
958 * @param n1 The first node.
959 * @param n2 The second node.
960 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
962 static int dependent_on(ir_node *n1, ir_node *n2)
964 assert(get_nodes_block(n1) == get_nodes_block(n2));
966 return heights_reachable_in_block(ir_heights, n1, n2);
969 static int cmp_call_dependency(const void *c1, const void *c2)
971 ir_node *n1 = *(ir_node **) c1;
972 ir_node *n2 = *(ir_node **) c2;
976 Classical qsort() comparison function behavior:
977 0 if both elements are equal
978 1 if second is "smaller" that first
979 -1 if first is "smaller" that second
981 if (dependent_on(n1, n2))
984 if (dependent_on(n2, n1))
987 /* The nodes have no depth order, but we need a total order because qsort()
990 * Additionally, we need to respect transitive dependencies. Consider a
991 * Call a depending on Call b and an independent Call c.
992 * We MUST NOT order c > a and b > c. */
993 h1 = get_irn_height(ir_heights, n1);
994 h2 = get_irn_height(ir_heights, n2);
995 if (h1 < h2) return -1;
996 if (h1 > h2) return 1;
997 /* Same height, so use a random (but stable) order */
998 return get_irn_idx(n1) - get_irn_idx(n2);
1002 * Walker: links all Call/Alloc/Free nodes to the Block they are contained.
1003 * Clears the irg_is_leaf flag if a Call is detected.
1005 static void link_ops_in_block_walker(ir_node *irn, void *data)
1007 be_abi_irg_t *env = (be_abi_irg_t*)data;
1008 unsigned code = get_irn_opcode(irn);
1010 if (code == iro_Call ||
1011 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
1012 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
1013 ir_node *bl = get_nodes_block(irn);
1014 void *save = get_irn_link(bl);
1016 if (code == iro_Call)
1017 env->call->flags.bits.irg_is_leaf = 0;
1019 set_irn_link(irn, save);
1020 set_irn_link(bl, irn);
1023 if (code == iro_Builtin && get_Builtin_kind(irn) == ir_bk_return_address) {
1024 ir_node *param = get_Builtin_param(irn, 0);
1025 ir_tarval *tv = get_Const_tarval(param);
1026 unsigned long value = get_tarval_long(tv);
1027 /* use ebp, so the climbframe algo works... */
1029 env->call->flags.bits.try_omit_fp = 0;
1036 * Process all Call/Alloc/Free nodes inside a basic block.
1037 * Note that the link field of the block must contain a linked list of all
1038 * nodes inside the Block. We first order this list according to data dependency
1039 * and that connect the nodes together.
1041 static void process_ops_in_block(ir_node *bl, void *data)
1043 be_abi_irg_t *env = (be_abi_irg_t*)data;
1044 ir_node *curr_sp = env->init_sp;
1051 for (irn = (ir_node*)get_irn_link(bl); irn != NULL;
1052 irn = (ir_node*)get_irn_link(irn)) {
1056 nodes = ALLOCAN(ir_node*, n_nodes);
1057 for (irn = (ir_node*)get_irn_link(bl), n = 0; irn != NULL;
1058 irn = (ir_node*)get_irn_link(irn), ++n) {
1062 /* If there were call nodes in the block. */
1067 /* order the call nodes according to data dependency */
1068 qsort(nodes, n_nodes, sizeof(nodes[0]), cmp_call_dependency);
1070 for (i = n_nodes - 1; i >= 0; --i) {
1071 ir_node *irn = nodes[i];
1073 DBG((dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1074 switch (get_irn_opcode(irn)) {
1077 /* The stack pointer will be modified due to a call. */
1078 env->call->flags.bits.try_omit_fp = 0;
1080 curr_sp = adjust_call(env, irn, curr_sp);
1083 if (get_Alloc_where(irn) == stack_alloc)
1084 curr_sp = adjust_alloc(env, irn, curr_sp);
1087 if (get_Free_where(irn) == stack_alloc)
1088 curr_sp = adjust_free(env, irn, curr_sp);
1091 panic("invalid call");
1095 /* Keep the last stack state in the block by tying it to Keep node,
1096 * the proj from calls is already kept */
1097 if (curr_sp != env->init_sp &&
1098 !(is_Proj(curr_sp) && be_is_Call(get_Proj_pred(curr_sp)))) {
1100 keep = be_new_Keep(bl, 1, nodes);
1101 pmap_insert(env->keep_map, bl, keep);
1105 set_irn_link(bl, curr_sp);
1109 * Adjust all call nodes in the graph to the ABI conventions.
1111 static void process_calls(ir_graph *irg)
1113 be_abi_irg_t *abi = be_get_irg_abi(irg);
1115 abi->call->flags.bits.irg_is_leaf = 1;
1116 irg_walk_graph(irg, firm_clear_link, link_ops_in_block_walker, abi);
1118 ir_heights = heights_new(irg);
1119 irg_block_walk_graph(irg, NULL, process_ops_in_block, abi);
1120 heights_free(ir_heights);
1124 * Computes the stack argument layout type.
1125 * Changes a possibly allocated value param type by moving
1126 * entities to the stack layout type.
1128 * @param env the ABI environment
1129 * @param call the current call ABI
1130 * @param method_type the method type
1131 * @param val_param_tp the value parameter type, will be destroyed
1132 * @param param_map an array mapping method arguments to the stack layout type
1134 * @return the stack argument layout type
1136 static ir_type *compute_arg_type(be_abi_irg_t *env, ir_graph *irg,
1137 be_abi_call_t *call,
1138 ir_type *method_type, ir_type *val_param_tp,
1139 ir_entity ***param_map)
1141 int dir = env->call->flags.bits.left_to_right ? 1 : -1;
1143 int n = get_method_n_params(method_type);
1144 int curr = inc > 0 ? 0 : n - 1;
1145 struct obstack *obst = be_get_be_obst(irg);
1151 ident *id = get_entity_ident(get_irg_entity(irg));
1154 *param_map = map = OALLOCN(obst, ir_entity*, n);
1155 res = new_type_struct(id_mangle_u(id, new_id_from_chars("arg_type", 8)));
1156 for (i = 0; i < n; ++i, curr += inc) {
1157 ir_type *param_type = get_method_param_type(method_type, curr);
1158 be_abi_call_arg_t *arg = get_call_arg(call, 0, curr, 1);
1161 if (arg->on_stack) {
1162 if (val_param_tp != NULL) {
1163 /* the entity was already created, create a copy in the param type */
1164 ir_entity *val_ent = get_method_value_param_ent(method_type, i);
1165 arg->stack_ent = copy_entity_own(val_ent, res);
1166 set_entity_link(val_ent, arg->stack_ent);
1167 set_entity_link(arg->stack_ent, NULL);
1169 /* create a new entity */
1170 snprintf(buf, sizeof(buf), "param_%d", i);
1171 arg->stack_ent = new_entity(res, new_id_from_str(buf), param_type);
1173 ofs += arg->space_before;
1174 ofs = round_up2(ofs, arg->alignment);
1175 set_entity_offset(arg->stack_ent, ofs);
1176 ofs += arg->space_after;
1177 ofs += get_type_size_bytes(param_type);
1178 map[i] = arg->stack_ent;
1181 set_type_size_bytes(res, ofs);
1182 set_type_state(res, layout_fixed);
1187 const arch_register_t *reg;
1191 static int cmp_regs(const void *a, const void *b)
1193 const reg_node_map_t *p = (const reg_node_map_t*)a;
1194 const reg_node_map_t *q = (const reg_node_map_t*)b;
1196 if (p->reg->reg_class == q->reg->reg_class)
1197 return p->reg->index - q->reg->index;
1199 return p->reg->reg_class < q->reg->reg_class ? -1 : +1;
1202 static void reg_map_to_arr(reg_node_map_t *res, pmap *reg_map)
1205 size_t n = pmap_count(reg_map);
1208 foreach_pmap(reg_map, ent) {
1209 res[i].reg = (const arch_register_t*)ent->key;
1210 res[i].irn = (ir_node*)ent->value;
1214 qsort(res, n, sizeof(res[0]), cmp_regs);
1218 * Creates a be_Return for a Return node.
1220 * @param @env the abi environment
1221 * @param irn the Return node or NULL if there was none
1222 * @param bl the block where the be_Retun should be placed
1223 * @param mem the current memory
1224 * @param n_res number of return results
1226 static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl,
1227 ir_node *mem, int n_res)
1229 be_abi_call_t *call = env->call;
1230 ir_graph *irg = get_Block_irg(bl);
1231 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1233 pmap *reg_map = pmap_create();
1234 ir_node *keep = (ir_node*)pmap_get(env->keep_map, bl);
1241 const arch_register_t **regs;
1245 get the valid stack node in this block.
1246 If we had a call in that block there is a Keep constructed by process_calls()
1247 which points to the last stack modification in that block. we'll use
1248 it then. Else we use the stack from the start block and let
1249 the ssa construction fix the usage.
1251 stack = be_abi_reg_map_get(env->regs, arch_env->sp);
1253 stack = get_irn_n(keep, 0);
1255 remove_End_keepalive(get_irg_end(irg), keep);
1258 /* Insert results for Return into the register map. */
1259 for (i = 0; i < n_res; ++i) {
1260 ir_node *res = get_Return_res(irn, i);
1261 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1262 assert(arg->in_reg && "return value must be passed in register");
1263 pmap_insert(reg_map, (void *) arg->reg, res);
1266 /* Add uses of the callee save registers. */
1267 foreach_pmap(env->regs, ent) {
1268 const arch_register_t *reg = (const arch_register_t*)ent->key;
1269 if (reg->type & (arch_register_type_callee_save | arch_register_type_ignore))
1270 pmap_insert(reg_map, ent->key, ent->value);
1273 be_abi_reg_map_set(reg_map, arch_env->sp, stack);
1275 /* Make the Epilogue node and call the arch's epilogue maker. */
1276 call->cb->epilogue(env->cb, bl, &mem, reg_map);
1279 Maximum size of the in array for Return nodes is
1280 return args + callee save/ignore registers + memory + stack pointer
1282 in_max = pmap_count(reg_map) + n_res + 2;
1284 in = ALLOCAN(ir_node*, in_max);
1285 regs = ALLOCAN(arch_register_t const*, in_max);
1288 in[1] = be_abi_reg_map_get(reg_map, arch_env->sp);
1290 regs[1] = arch_env->sp;
1293 /* clear SP entry, since it has already been grown. */
1294 pmap_insert(reg_map, (void *) arch_env->sp, NULL);
1295 for (i = 0; i < n_res; ++i) {
1296 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1298 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1299 regs[n++] = arg->reg;
1301 /* Clear the map entry to mark the register as processed. */
1302 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1305 /* grow the rest of the stuff. */
1306 foreach_pmap(reg_map, ent) {
1308 in[n] = (ir_node*)ent->value;
1309 regs[n++] = (const arch_register_t*)ent->key;
1313 /* The in array for the new back end return is now ready. */
1315 dbgi = get_irn_dbg_info(irn);
1319 /* we have to pop the shadow parameter in in case of struct returns */
1321 ret = be_new_Return(dbgi, irg, bl, n_res, pop, n, in);
1322 arch_irn_add_flags(ret, arch_irn_flags_epilog);
1324 /* Set the register classes of the return's parameter accordingly. */
1325 for (i = 0; i < n; ++i) {
1326 if (regs[i] == NULL)
1329 be_set_constr_single_reg_in(ret, i, regs[i], arch_register_req_type_none);
1332 /* Free the space of the Epilog's in array and the register <-> proj map. */
1333 pmap_destroy(reg_map);
1338 typedef struct ent_pos_pair ent_pos_pair;
1339 struct ent_pos_pair {
1340 ir_entity *ent; /**< a value param entity */
1341 int pos; /**< its parameter number */
1342 ent_pos_pair *next; /**< for linking */
1345 typedef struct lower_frame_sels_env_t {
1346 ent_pos_pair *value_param_list; /**< the list of all value param entities */
1347 ir_node *frame; /**< the current frame */
1348 const arch_register_class_t *sp_class; /**< register class of the stack pointer */
1349 const arch_register_class_t *link_class; /**< register class of the link pointer */
1350 ir_type *value_tp; /**< the value type if any */
1351 ir_type *frame_tp; /**< the frame type */
1352 int static_link_pos; /**< argument number of the hidden static link */
1353 } lower_frame_sels_env_t;
1356 * Return an entity from the backend for an value param entity.
1358 * @param ent an value param type entity
1359 * @param ctx context
1361 static ir_entity *get_argument_entity(ir_entity *ent, lower_frame_sels_env_t *ctx)
1363 ir_entity *argument_ent = (ir_entity*)get_entity_link(ent);
1365 if (argument_ent == NULL) {
1366 /* we have NO argument entity yet: This is bad, as we will
1367 * need one for backing store.
1370 ir_type *frame_tp = ctx->frame_tp;
1371 unsigned offset = get_type_size_bytes(frame_tp);
1372 ir_type *tp = get_entity_type(ent);
1373 unsigned align = get_type_alignment_bytes(tp);
1375 offset += align - 1;
1376 offset &= ~(align - 1);
1378 argument_ent = copy_entity_own(ent, frame_tp);
1380 /* must be automatic to set a fixed layout */
1381 set_entity_offset(argument_ent, offset);
1382 offset += get_type_size_bytes(tp);
1384 set_type_size_bytes(frame_tp, offset);
1385 set_entity_link(ent, argument_ent);
1387 return argument_ent;
1390 * Walker: Replaces Sels of frame type and
1391 * value param type entities by FrameAddress.
1392 * Links all used entities.
1394 static void lower_frame_sels_walker(ir_node *irn, void *data)
1396 lower_frame_sels_env_t *ctx = (lower_frame_sels_env_t*)data;
1399 ir_node *ptr = get_Sel_ptr(irn);
1401 if (ptr == ctx->frame) {
1402 ir_entity *ent = get_Sel_entity(irn);
1403 ir_node *bl = get_nodes_block(irn);
1406 int is_value_param = 0;
1408 if (get_entity_owner(ent) == ctx->value_tp) {
1411 /* replace by its copy from the argument type */
1412 pos = get_struct_member_index(ctx->value_tp, ent);
1413 ent = get_argument_entity(ent, ctx);
1416 nw = be_new_FrameAddr(ctx->sp_class, bl, ctx->frame, ent);
1419 /* check, if it's a param Sel and if have not seen this entity before */
1420 if (is_value_param && get_entity_link(ent) == NULL) {
1426 ARR_APP1(ent_pos_pair, ctx->value_param_list, pair);
1428 set_entity_link(ent, ctx->value_param_list);
1435 * Check if a value parameter is transmitted as a register.
1436 * This might happen if the address of an parameter is taken which is
1437 * transmitted in registers.
1439 * Note that on some architectures this case must be handled specially
1440 * because the place of the backing store is determined by their ABI.
1442 * In the default case we move the entity to the frame type and create
1443 * a backing store into the first block.
1445 static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_graph *irg,
1446 ent_pos_pair *value_param_list)
1448 be_abi_call_t *call = env->call;
1449 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1450 ent_pos_pair *entry, *new_list;
1452 int i, n = ARR_LEN(value_param_list);
1455 for (i = 0; i < n; ++i) {
1456 int pos = value_param_list[i].pos;
1457 be_abi_call_arg_t *arg = get_call_arg(call, 0, pos, 1);
1460 DBG((dbg, LEVEL_2, "\targ #%d need backing store\n", pos));
1461 value_param_list[i].next = new_list;
1462 new_list = &value_param_list[i];
1465 if (new_list != NULL) {
1466 /* ok, change the graph */
1467 ir_node *start_bl = get_irg_start_block(irg);
1468 ir_node *first_bl = get_first_block_succ(start_bl);
1469 ir_node *frame, *imem, *nmem, *store, *mem, *args;
1470 optimization_state_t state;
1473 assert(first_bl && first_bl != start_bl);
1474 /* we had already removed critical edges, so the following
1475 assertion should be always true. */
1476 assert(get_Block_n_cfgpreds(first_bl) == 1);
1478 /* now create backing stores */
1479 frame = get_irg_frame(irg);
1480 imem = get_irg_initial_mem(irg);
1482 save_optimization_state(&state);
1484 nmem = new_r_Proj(get_irg_start(irg), mode_M, pn_Start_M);
1485 restore_optimization_state(&state);
1487 /* reroute all edges to the new memory source */
1488 edges_reroute(imem, nmem);
1492 args = get_irg_args(irg);
1493 for (entry = new_list; entry != NULL; entry = entry->next) {
1495 ir_type *tp = get_entity_type(entry->ent);
1496 ir_mode *mode = get_type_mode(tp);
1499 /* address for the backing store */
1500 addr = be_new_FrameAddr(arch_env->sp->reg_class, first_bl, frame, entry->ent);
1503 mem = new_r_Proj(store, mode_M, pn_Store_M);
1505 /* the backing store itself */
1506 store = new_r_Store(first_bl, mem, addr,
1507 new_r_Proj(args, mode, i), cons_none);
1509 /* the new memory Proj gets the last Proj from store */
1510 set_Proj_pred(nmem, store);
1511 set_Proj_proj(nmem, pn_Store_M);
1513 /* move all entities to the frame type */
1514 frame_tp = get_irg_frame_type(irg);
1515 offset = get_type_size_bytes(frame_tp);
1517 /* we will add new entities: set the layout to undefined */
1518 assert(get_type_state(frame_tp) == layout_fixed);
1519 set_type_state(frame_tp, layout_undefined);
1520 for (entry = new_list; entry != NULL; entry = entry->next) {
1521 ir_entity *ent = entry->ent;
1523 /* If the entity is still on the argument type, move it to the
1525 * This happens if the value_param type was build due to compound
1527 if (get_entity_owner(ent) != frame_tp) {
1528 ir_type *tp = get_entity_type(ent);
1529 unsigned align = get_type_alignment_bytes(tp);
1531 offset += align - 1;
1532 offset &= ~(align - 1);
1533 set_entity_owner(ent, frame_tp);
1534 /* must be automatic to set a fixed layout */
1535 set_entity_offset(ent, offset);
1536 offset += get_type_size_bytes(tp);
1539 set_type_size_bytes(frame_tp, offset);
1540 /* fix the layout again */
1541 set_type_state(frame_tp, layout_fixed);
1546 * The start block has no jump, instead it has an initial exec Proj.
1547 * The backend wants to handle all blocks the same way, so we replace
1548 * the out cfg edge with a real jump.
1550 static void fix_start_block(ir_graph *irg)
1552 ir_node *initial_X = get_irg_initial_exec(irg);
1553 ir_node *start_block = get_irg_start_block(irg);
1554 ir_node *jmp = new_r_Jmp(start_block);
1556 assert(is_Proj(initial_X));
1557 exchange(initial_X, jmp);
1558 set_irg_initial_exec(irg, new_r_Bad(irg));
1562 * Update the entity of Sels to the outer value parameters.
1564 static void update_outer_frame_sels(ir_node *irn, void *env)
1566 lower_frame_sels_env_t *ctx = (lower_frame_sels_env_t*)env;
1573 ptr = get_Sel_ptr(irn);
1574 if (! is_arg_Proj(ptr))
1576 if (get_Proj_proj(ptr) != ctx->static_link_pos)
1578 ent = get_Sel_entity(irn);
1580 if (get_entity_owner(ent) == ctx->value_tp) {
1581 /* replace by its copy from the argument type */
1582 pos = get_struct_member_index(ctx->value_tp, ent);
1583 ent = get_argument_entity(ent, ctx);
1584 set_Sel_entity(irn, ent);
1586 /* check, if we have not seen this entity before */
1587 if (get_entity_link(ent) == NULL) {
1593 ARR_APP1(ent_pos_pair, ctx->value_param_list, pair);
1595 set_entity_link(ent, ctx->value_param_list);
1601 * Fix access to outer local variables.
1603 static void fix_outer_variable_access(be_abi_irg_t *env,
1604 lower_frame_sels_env_t *ctx)
1610 for (i = get_class_n_members(ctx->frame_tp) - 1; i >= 0; --i) {
1611 ir_entity *ent = get_class_member(ctx->frame_tp, i);
1613 if (! is_method_entity(ent))
1616 irg = get_entity_irg(ent);
1621 * FIXME: find the number of the static link parameter
1622 * for now we assume 0 here
1624 ctx->static_link_pos = 0;
1626 irg_walk_graph(irg, NULL, update_outer_frame_sels, ctx);
1631 * Modify the irg itself and the frame type.
1633 static void modify_irg(ir_graph *irg)
1635 be_abi_irg_t *env = be_get_irg_abi(irg);
1636 be_abi_call_t *call = env->call;
1637 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1638 const arch_register_t *sp = arch_env->sp;
1639 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1640 be_irg_t *birg = be_birg_from_irg(irg);
1641 struct obstack *obst = be_get_be_obst(irg);
1642 be_stack_layout_t *stack_layout = be_get_irg_stack_layout(irg);
1645 ir_node *new_mem_proj;
1651 unsigned frame_size;
1654 const arch_register_t *fp_reg;
1655 ir_node *frame_pointer;
1659 const ir_edge_t *edge;
1660 ir_type *arg_type, *bet_type, *tp;
1661 lower_frame_sels_env_t ctx;
1662 ir_entity **param_map;
1664 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1666 old_mem = get_irg_initial_mem(irg);
1668 irp_reserve_resources(irp, IR_RESOURCE_ENTITY_LINK);
1670 /* set the links of all frame entities to NULL, we use it
1671 to detect if an entity is already linked in the value_param_list */
1672 tp = get_method_value_param_type(method_type);
1675 /* clear the links of the clone type, let the
1676 original entities point to its clones */
1677 for (i = get_struct_n_members(tp) - 1; i >= 0; --i) {
1678 ir_entity *mem = get_struct_member(tp, i);
1679 set_entity_link(mem, NULL);
1683 arg_type = compute_arg_type(env, irg, call, method_type, tp, ¶m_map);
1685 /* Convert the Sel nodes in the irg to frame addr nodes: */
1686 ctx.value_param_list = NEW_ARR_F(ent_pos_pair, 0);
1687 ctx.frame = get_irg_frame(irg);
1688 ctx.sp_class = arch_env->sp->reg_class;
1689 ctx.link_class = arch_env->link_class;
1690 ctx.frame_tp = get_irg_frame_type(irg);
1692 /* layout the stackframe now */
1693 if (get_type_state(ctx.frame_tp) == layout_undefined) {
1694 default_layout_compound_type(ctx.frame_tp);
1697 /* we will possible add new entities to the frame: set the layout to undefined */
1698 assert(get_type_state(ctx.frame_tp) == layout_fixed);
1699 set_type_state(ctx.frame_tp, layout_undefined);
1701 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1703 /* fix the frame type layout again */
1704 set_type_state(ctx.frame_tp, layout_fixed);
1705 /* align stackframe to 4 byte */
1706 frame_size = get_type_size_bytes(ctx.frame_tp);
1707 if (frame_size % 4 != 0) {
1708 set_type_size_bytes(ctx.frame_tp, frame_size + 4 - (frame_size % 4));
1711 env->regs = pmap_create();
1713 n_params = get_method_n_params(method_type);
1714 args = OALLOCNZ(obst, ir_node*, n_params);
1717 * for inner function we must now fix access to outer frame entities.
1719 fix_outer_variable_access(env, &ctx);
1721 /* Check if a value parameter is transmitted as a register.
1722 * This might happen if the address of an parameter is taken which is
1723 * transmitted in registers.
1725 * Note that on some architectures this case must be handled specially
1726 * because the place of the backing store is determined by their ABI.
1728 * In the default case we move the entity to the frame type and create
1729 * a backing store into the first block.
1731 fix_address_of_parameter_access(env, irg, ctx.value_param_list);
1733 DEL_ARR_F(ctx.value_param_list);
1734 irp_free_resources(irp, IR_RESOURCE_ENTITY_LINK);
1736 /* Fill the argument vector */
1737 arg_tuple = get_irg_args(irg);
1738 foreach_out_edge(arg_tuple, edge) {
1739 ir_node *irn = get_edge_src_irn(edge);
1740 if (! is_Anchor(irn)) {
1741 int nr = get_Proj_proj(irn);
1743 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1747 bet_type = call->cb->get_between_type(env->cb);
1748 stack_frame_init(stack_layout, arg_type, bet_type,
1749 get_irg_frame_type(irg), param_map);
1750 stack_layout->sp_relative = call->flags.bits.try_omit_fp;
1752 /* Count the register params and add them to the number of Projs for the RegParams node */
1753 for (i = 0; i < n_params; ++i) {
1754 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1755 if (arg->in_reg && args[i]) {
1756 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1757 assert(i == get_Proj_proj(args[i]));
1759 /* For now, associate the register with the old Proj from Start representing that argument. */
1760 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1761 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1765 /* Collect all callee-save registers */
1766 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
1767 const arch_register_class_t *cls = &arch_env->register_classes[i];
1768 for (j = 0; j < cls->n_regs; ++j) {
1769 const arch_register_t *reg = &cls->regs[j];
1770 if (reg->type & (arch_register_type_callee_save | arch_register_type_state)) {
1771 pmap_insert(env->regs, (void *) reg, NULL);
1776 /* handle start block here (place a jump in the block) */
1777 fix_start_block(irg);
1779 pmap_insert(env->regs, (void *) sp, NULL);
1780 pmap_insert(env->regs, (void *) arch_env->bp, NULL);
1781 start_bl = get_irg_start_block(irg);
1782 env->start = be_new_Start(NULL, start_bl, pmap_count(env->regs) + 1);
1783 arch_irn_add_flags(env->start, arch_irn_flags_prolog);
1784 set_irg_start(irg, env->start);
1787 * make proj nodes for the callee save registers.
1788 * memorize them, since Return nodes get those as inputs.
1790 * Note, that if a register corresponds to an argument, the regs map contains
1791 * the old Proj from start for that argument.
1794 rm = ALLOCAN(reg_node_map_t, pmap_count(env->regs));
1795 reg_map_to_arr(rm, env->regs);
1796 for (i = 0, n = pmap_count(env->regs); i < n; ++i) {
1797 const arch_register_t *reg = rm[i].reg;
1798 ir_mode *mode = reg->reg_class->mode;
1800 arch_register_req_type_t add_type = arch_register_req_type_none;
1804 add_type |= arch_register_req_type_produces_sp | arch_register_req_type_ignore;
1807 proj = new_r_Proj(env->start, mode, nr + 1);
1808 pmap_insert(env->regs, (void *) reg, proj);
1809 be_set_constr_single_reg_out(env->start, nr + 1, reg, add_type);
1810 arch_set_irn_register(proj, reg);
1812 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1815 /* create a new initial memory proj */
1816 assert(is_Proj(old_mem));
1817 arch_set_out_register_req(env->start, 0, arch_no_register_req);
1818 new_mem_proj = new_r_Proj(env->start, mode_M, 0);
1820 set_irg_initial_mem(irg, mem);
1822 /* Generate the Prologue */
1823 fp_reg = call->cb->prologue(env->cb, &mem, env->regs, &stack_layout->initial_bias);
1825 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1826 env->init_sp = be_new_IncSP(sp, start_bl, env->init_sp, BE_STACK_FRAME_SIZE_EXPAND, 0);
1827 arch_irn_add_flags(env->init_sp, arch_irn_flags_prolog);
1828 be_abi_reg_map_set(env->regs, sp, env->init_sp);
1830 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1831 arch_set_irn_register(env->init_sp, sp);
1833 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1834 set_irg_frame(irg, frame_pointer);
1835 rbitset_clear(birg->allocatable_regs, fp_reg->global_index);
1837 /* rewire old mem users to new mem */
1838 exchange(old_mem, mem);
1840 /* keep the mem (for functions with an endless loop = no return) */
1843 set_irg_initial_mem(irg, mem);
1845 /* Now, introduce stack param nodes for all parameters passed on the stack */
1846 for (i = 0; i < n_params; ++i) {
1847 ir_node *arg_proj = args[i];
1848 ir_node *repl = NULL;
1850 if (arg_proj != NULL) {
1851 be_abi_call_arg_t *arg;
1852 ir_type *param_type;
1853 int nr = get_Proj_proj(arg_proj);
1856 nr = MIN(nr, n_params);
1857 arg = get_call_arg(call, 0, nr, 1);
1858 param_type = get_method_param_type(method_type, nr);
1861 repl = (ir_node*)pmap_get(env->regs, arg->reg);
1862 } else if (arg->on_stack) {
1863 ir_node *addr = be_new_FrameAddr(sp->reg_class, start_bl, frame_pointer, arg->stack_ent);
1865 /* For atomic parameters which are actually used, we create a Load node. */
1866 if (is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1867 ir_mode *mode = get_type_mode(param_type);
1868 ir_mode *load_mode = arg->load_mode;
1870 ir_node *load = new_r_Load(start_bl, new_r_NoMem(irg), addr, load_mode, cons_floats);
1871 repl = new_r_Proj(load, load_mode, pn_Load_res);
1873 if (mode != load_mode) {
1874 repl = new_r_Conv(start_bl, repl, mode);
1877 /* The stack parameter is not primitive (it is a struct or array),
1878 * we thus will create a node representing the parameter's address
1884 assert(repl != NULL);
1886 /* Beware: the mode of the register parameters is always the mode of the register class
1887 which may be wrong. Add Conv's then. */
1888 mode = get_irn_mode(args[i]);
1889 if (mode != get_irn_mode(repl)) {
1890 repl = new_r_Conv(get_nodes_block(repl), repl, mode);
1892 exchange(args[i], repl);
1896 /* the arg proj is not needed anymore now and should be only used by the anchor */
1897 assert(get_irn_n_edges(arg_tuple) == 1);
1898 kill_node(arg_tuple);
1899 set_irg_args(irg, new_r_Bad(irg));
1901 /* All Return nodes hang on the End node, so look for them there. */
1902 end = get_irg_end_block(irg);
1903 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1904 ir_node *irn = get_Block_cfgpred(end, i);
1906 if (is_Return(irn)) {
1907 ir_node *blk = get_nodes_block(irn);
1908 ir_node *mem = get_Return_mem(irn);
1909 ir_node *ret = create_be_return(env, irn, blk, mem, get_Return_n_ress(irn));
1914 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1915 the code is dead and will never be executed. */
1918 /** Fix the state inputs of calls that still hang on unknowns */
1919 static void fix_call_state_inputs(ir_graph *irg)
1921 be_abi_irg_t *env = be_get_irg_abi(irg);
1922 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1924 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1926 /* Collect caller save registers */
1927 n = arch_env->n_register_classes;
1928 for (i = 0; i < n; ++i) {
1930 const arch_register_class_t *cls = &arch_env->register_classes[i];
1931 for (j = 0; j < cls->n_regs; ++j) {
1932 const arch_register_t *reg = arch_register_for_index(cls, j);
1933 if (reg->type & arch_register_type_state) {
1934 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1939 n = ARR_LEN(env->calls);
1940 n_states = ARR_LEN(stateregs);
1941 for (i = 0; i < n; ++i) {
1943 ir_node *call = env->calls[i];
1945 arity = get_irn_arity(call);
1947 /* the state reg inputs are the last n inputs of the calls */
1948 for (s = 0; s < n_states; ++s) {
1949 int inp = arity - n_states + s;
1950 const arch_register_t *reg = stateregs[s];
1951 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
1953 set_irn_n(call, inp, regnode);
1957 DEL_ARR_F(stateregs);
1961 * Create a trampoline entity for the given method.
1963 static ir_entity *create_trampoline(be_main_env_t *be, ir_entity *method)
1965 ir_type *type = get_entity_type(method);
1966 ident *old_id = get_entity_ld_ident(method);
1967 ident *id = id_mangle3("", old_id, "$stub");
1968 ir_type *parent = be->pic_trampolines_type;
1969 ir_entity *ent = new_entity(parent, old_id, type);
1970 set_entity_ld_ident(ent, id);
1971 set_entity_visibility(ent, ir_visibility_private);
1977 * Returns the trampoline entity for the given method.
1979 static ir_entity *get_trampoline(be_main_env_t *env, ir_entity *method)
1981 ir_entity *result = (ir_entity*)pmap_get(env->ent_trampoline_map, method);
1982 if (result == NULL) {
1983 result = create_trampoline(env, method);
1984 pmap_insert(env->ent_trampoline_map, method, result);
1990 static ir_entity *create_pic_symbol(be_main_env_t *be, ir_entity *entity)
1992 ident *old_id = get_entity_ld_ident(entity);
1993 ident *id = id_mangle3("", old_id, "$non_lazy_ptr");
1994 ir_type *e_type = get_entity_type(entity);
1995 ir_type *type = new_type_pointer(e_type);
1996 ir_type *parent = be->pic_symbols_type;
1997 ir_entity *ent = new_entity(parent, old_id, type);
1998 set_entity_ld_ident(ent, id);
1999 set_entity_visibility(ent, ir_visibility_private);
2004 static ir_entity *get_pic_symbol(be_main_env_t *env, ir_entity *entity)
2006 ir_entity *result = (ir_entity*)pmap_get(env->ent_pic_symbol_map, entity);
2007 if (result == NULL) {
2008 result = create_pic_symbol(env, entity);
2009 pmap_insert(env->ent_pic_symbol_map, entity, result);
2018 * Returns non-zero if a given entity can be accessed using a relative address.
2020 static int can_address_relative(ir_entity *entity)
2022 return get_entity_visibility(entity) != ir_visibility_external
2023 && !(get_entity_linkage(entity) & IR_LINKAGE_MERGE);
2026 static ir_node *get_pic_base(ir_graph *irg)
2028 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
2029 if (arch_env->impl->get_pic_base == NULL)
2031 return arch_env->impl->get_pic_base(irg);
2034 /** patches SymConsts to work in position independent code */
2035 static void fix_pic_symconsts(ir_node *node, void *data)
2037 ir_graph *irg = get_irn_irg(node);
2038 be_main_env_t *be = be_get_irg_main_env(irg);
2048 arity = get_irn_arity(node);
2049 for (i = 0; i < arity; ++i) {
2051 ir_node *pred = get_irn_n(node, i);
2053 ir_entity *pic_symbol;
2054 ir_node *pic_symconst;
2056 if (!is_SymConst(pred))
2059 entity = get_SymConst_entity(pred);
2060 block = get_nodes_block(pred);
2062 /* calls can jump to relative addresses, so we can directly jump to
2063 the (relatively) known call address or the trampoline */
2064 if (i == 1 && is_Call(node)) {
2065 ir_entity *trampoline;
2066 ir_node *trampoline_const;
2068 if (can_address_relative(entity))
2071 dbgi = get_irn_dbg_info(pred);
2072 trampoline = get_trampoline(be, entity);
2073 trampoline_const = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
2075 set_irn_n(node, i, trampoline_const);
2079 /* everything else is accessed relative to EIP */
2080 mode = get_irn_mode(pred);
2081 pic_base = get_pic_base(irg);
2083 /* all ok now for locally constructed stuff */
2084 if (can_address_relative(entity)) {
2085 ir_node *add = new_r_Add(block, pic_base, pred, mode);
2087 /* make sure the walker doesn't visit this add again */
2088 mark_irn_visited(add);
2089 set_irn_n(node, i, add);
2093 /* get entry from pic symbol segment */
2094 dbgi = get_irn_dbg_info(pred);
2095 pic_symbol = get_pic_symbol(be, entity);
2096 pic_symconst = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
2098 add = new_r_Add(block, pic_base, pic_symconst, mode);
2099 mark_irn_visited(add);
2101 /* we need an extra indirection for global data outside our current
2102 module. The loads are always safe and can therefore float
2103 and need no memory input */
2104 load = new_r_Load(block, new_r_NoMem(irg), add, mode, cons_floats);
2105 load_res = new_r_Proj(load, mode, pn_Load_res);
2107 set_irn_n(node, i, load_res);
2111 be_abi_irg_t *be_abi_introduce(ir_graph *irg)
2113 be_abi_irg_t *env = XMALLOCZ(be_abi_irg_t);
2114 ir_node *old_frame = get_irg_frame(irg);
2115 be_options_t *options = be_get_irg_options(irg);
2116 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
2117 ir_entity *entity = get_irg_entity(irg);
2118 ir_type *method_type = get_entity_type(entity);
2119 be_irg_t *birg = be_birg_from_irg(irg);
2120 struct obstack *obst = &birg->obst;
2125 /* determine allocatable registers */
2126 assert(birg->allocatable_regs == NULL);
2127 birg->allocatable_regs = rbitset_obstack_alloc(obst, arch_env->n_registers);
2128 for (r = 0; r < arch_env->n_registers; ++r) {
2129 const arch_register_t *reg = &arch_env->registers[r];
2130 if ( !(reg->type & arch_register_type_ignore)) {
2131 rbitset_set(birg->allocatable_regs, r);
2135 /* break here if backend provides a custom API.
2136 * Note: we shouldn't have to setup any be_abi_irg_t* stuff at all,
2137 * but need more cleanup to make this work
2139 be_set_irg_abi(irg, env);
2141 be_omit_fp = options->omit_fp;
2143 env->keep_map = pmap_create();
2144 env->call = be_abi_call_new(arch_env->sp->reg_class);
2145 arch_env_get_call_abi(arch_env, method_type, env->call);
2147 env->init_sp = dummy = new_r_Dummy(irg, arch_env->sp->reg_class->mode);
2148 env->calls = NEW_ARR_F(ir_node*, 0);
2151 irg_walk_graph(irg, fix_pic_symconsts, NULL, env);
2154 /* Lower all call nodes in the IRG. */
2158 Beware: init backend abi call object after processing calls,
2159 otherwise some information might be not yet available.
2161 env->cb = env->call->cb->init(env->call, irg);
2163 /* Process the IRG */
2166 /* fix call inputs for state registers */
2167 fix_call_state_inputs(irg);
2169 /* We don't need the keep map anymore. */
2170 pmap_destroy(env->keep_map);
2171 env->keep_map = NULL;
2173 /* calls array is not needed anymore */
2174 DEL_ARR_F(env->calls);
2177 /* reroute the stack origin of the calls to the true stack origin. */
2178 exchange(dummy, env->init_sp);
2179 exchange(old_frame, get_irg_frame(irg));
2181 env->call->cb->done(env->cb);
2186 void be_abi_free(ir_graph *irg)
2188 be_abi_irg_t *env = be_get_irg_abi(irg);
2190 if (env->call != NULL)
2191 be_abi_call_free(env->call);
2192 if (env->regs != NULL)
2193 pmap_destroy(env->regs);
2196 be_set_irg_abi(irg, NULL);
2200 * called after nodes have been transformed so some node references can be
2201 * replaced with new nodes
2203 void be_abi_transform_fixup(ir_graph *irg)
2205 be_abi_irg_t *abi = be_get_irg_abi(irg);
2208 if (abi == NULL || abi->regs == NULL)
2211 new_regs = pmap_create();
2212 foreach_pmap(abi->regs, entry) {
2213 ir_node *value = (ir_node*)entry->value;
2214 ir_node *transformed = be_transform_node(value);
2215 pmap_insert(new_regs, entry->key, transformed);
2217 pmap_destroy(abi->regs);
2218 abi->regs = new_regs;
2221 void be_put_allocatable_regs(const ir_graph *irg,
2222 const arch_register_class_t *cls, bitset_t *bs)
2224 be_irg_t *birg = be_birg_from_irg(irg);
2225 unsigned *allocatable_regs = birg->allocatable_regs;
2228 assert(bitset_size(bs) == cls->n_regs);
2229 bitset_clear_all(bs);
2230 for (i = 0; i < cls->n_regs; ++i) {
2231 const arch_register_t *reg = &cls->regs[i];
2232 if (rbitset_is_set(allocatable_regs, reg->global_index))
2237 unsigned be_get_n_allocatable_regs(const ir_graph *irg,
2238 const arch_register_class_t *cls)
2240 bitset_t *bs = bitset_alloca(cls->n_regs);
2241 be_put_allocatable_regs(irg, cls, bs);
2242 return bitset_popcount(bs);
2245 void be_set_allocatable_regs(const ir_graph *irg,
2246 const arch_register_class_t *cls,
2247 unsigned *raw_bitset)
2249 be_irg_t *birg = be_birg_from_irg(irg);
2250 unsigned *allocatable_regs = birg->allocatable_regs;
2253 rbitset_clear_all(raw_bitset, cls->n_regs);
2254 for (i = 0; i < cls->n_regs; ++i) {
2255 const arch_register_t *reg = &cls->regs[i];
2256 if (rbitset_is_set(allocatable_regs, reg->global_index))
2257 rbitset_set(raw_bitset, i);
2261 ir_node *be_abi_get_callee_save_irn(ir_graph *irg, const arch_register_t *reg)
2263 const be_abi_irg_t *abi = be_get_irg_abi(irg);
2264 assert(reg->type & arch_register_type_callee_save);
2265 assert(pmap_contains(abi->regs, (void *) reg));
2266 return (ir_node*)pmap_get(abi->regs, (void *) reg);
2269 ir_node *be_abi_get_ignore_irn(ir_graph *irg, const arch_register_t *reg)
2271 const be_abi_irg_t *abi = be_get_irg_abi(irg);
2272 assert(reg->type & arch_register_type_ignore);
2273 assert(pmap_contains(abi->regs, (void *) reg));
2274 return (ir_node*)pmap_get(abi->regs, (void *) reg);
2277 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_abi);
2278 void be_init_abi(void)
2280 FIRM_DBG_REGISTER(dbg, "firm.be.abi");