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. */
98 int start_block_bias; /**< The stack bias at the end of the start block. */
100 pmap *keep_map; /**< mapping blocks to keep nodes. */
102 ir_node **calls; /**< flexible array containing all be_Call nodes */
105 static ir_heights_t *ir_heights;
107 /** Flag: if set, try to omit the frame pointer in all routines. */
108 static int be_omit_fp = 1;
110 static ir_node *be_abi_reg_map_get(pmap *map, const arch_register_t *reg)
112 return (ir_node*)pmap_get(map, reg);
115 static void be_abi_reg_map_set(pmap *map, const arch_register_t* reg,
118 pmap_insert(map, reg, node);
122 * Check if the given register is callee save, ie. will be saved by the callee.
124 static bool arch_register_is_callee_save(
125 const arch_env_t *arch_env,
126 const arch_register_t *reg)
128 if (arch_env->impl->register_saved_by)
129 return arch_env->impl->register_saved_by(reg, /*callee=*/1);
134 * Check if the given register is caller save, ie. must be saved by the caller.
136 static bool arch_register_is_caller_save(
137 const arch_env_t *arch_env,
138 const arch_register_t *reg)
140 if (arch_env->impl->register_saved_by)
141 return arch_env->impl->register_saved_by(reg, /*callee=*/0);
148 _ ____ ___ ____ _ _ _ _
149 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
150 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
151 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
152 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
154 These callbacks are used by the backend to set the parameters
155 for a specific call type.
159 * Set compare function: compares two ABI call object arguments.
161 static int cmp_call_arg(const void *a, const void *b, size_t n)
163 const be_abi_call_arg_t *p = (const be_abi_call_arg_t*)a;
164 const be_abi_call_arg_t *q = (const be_abi_call_arg_t*)b;
166 return !(p->is_res == q->is_res && p->pos == q->pos && p->callee == q->callee);
170 * Get an ABI call object argument.
172 * @param call the abi call
173 * @param is_res true for call results, false for call arguments
174 * @param pos position of the argument
175 * @param callee context type - if we are callee or caller
177 static be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos, int callee)
179 be_abi_call_arg_t arg;
182 memset(&arg, 0, sizeof(arg));
187 hash = is_res * 128 + pos;
189 return (be_abi_call_arg_t*)set_find(call->params, &arg, sizeof(arg), hash);
193 * Set an ABI call object argument.
195 static void remember_call_arg(be_abi_call_arg_t *arg, be_abi_call_t *call, be_abi_context_t context)
197 unsigned hash = arg->is_res * 128 + arg->pos;
198 if (context & ABI_CONTEXT_CALLEE) {
200 set_insert(call->params, arg, sizeof(*arg), hash);
202 if (context & ABI_CONTEXT_CALLER) {
204 set_insert(call->params, arg, sizeof(*arg), hash);
208 /* Set the flags for a call. */
209 void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const be_abi_callbacks_t *cb)
215 /* Sets the number of bytes the stackframe is shrinked by the callee on return */
216 void be_abi_call_set_pop(be_abi_call_t *call, int pop)
222 /* Set register class for call address */
223 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
225 call->cls_addr = cls;
229 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos,
230 ir_mode *load_mode, unsigned alignment,
231 unsigned space_before, unsigned space_after,
232 be_abi_context_t context)
234 be_abi_call_arg_t arg;
235 memset(&arg, 0, sizeof(arg));
236 assert(alignment > 0 && "Alignment must be greater than 0");
238 arg.load_mode = load_mode;
239 arg.alignment = alignment;
240 arg.space_before = space_before;
241 arg.space_after = space_after;
245 remember_call_arg(&arg, call, context);
248 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
250 be_abi_call_arg_t arg;
251 memset(&arg, 0, sizeof(arg));
258 remember_call_arg(&arg, call, context);
261 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
263 be_abi_call_arg_t arg;
264 memset(&arg, 0, sizeof(arg));
271 remember_call_arg(&arg, call, context);
274 /* Get the flags of a ABI call object. */
275 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
281 * Constructor for a new ABI call object.
283 * @param cls_addr register class of the call address
285 * @return the new ABI call object
287 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
289 be_abi_call_t *call = XMALLOCZ(be_abi_call_t);
292 call->params = new_set(cmp_call_arg, 16);
294 call->cls_addr = cls_addr;
296 call->flags.bits.try_omit_fp = be_omit_fp;
302 * Destructor for an ABI call object.
304 static void be_abi_call_free(be_abi_call_t *call)
306 del_set(call->params);
311 * Initializes the frame layout from parts
313 * @param frame the stack layout that will be initialized
314 * @param args the stack argument layout type
315 * @param between the between layout type
316 * @param locals the method frame type
317 * @param param_map an array mapping method argument positions to the stack argument type
319 * @return the initialized stack layout
321 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
322 ir_type *between, ir_type *locals,
323 ir_entity *param_map[])
325 frame->arg_type = args;
326 frame->between_type = between;
327 frame->frame_type = locals;
328 frame->initial_offset = 0;
329 frame->initial_bias = 0;
330 frame->order[1] = between;
331 frame->param_map = param_map;
333 /* typical decreasing stack: locals have the
334 * lowest addresses, arguments the highest */
335 frame->order[0] = locals;
336 frame->order[2] = args;
347 Adjustment of the calls inside a graph.
352 * Transform a call node into a be_Call node.
354 * @param env The ABI environment for the current irg.
355 * @param irn The call node.
356 * @param curr_sp The stack pointer node to use.
357 * @return The stack pointer after the call.
359 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp)
361 ir_graph *irg = get_irn_irg(irn);
362 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
363 ir_type *call_tp = get_Call_type(irn);
364 ir_node *call_ptr = get_Call_ptr(irn);
365 size_t n_params = get_method_n_params(call_tp);
366 ir_node *curr_mem = get_Call_mem(irn);
367 ir_node *bl = get_nodes_block(irn);
369 const arch_register_t *sp = arch_env->sp;
370 be_abi_call_t *call = be_abi_call_new(sp->reg_class);
371 ir_mode *mach_mode = sp->reg_class->mode;
372 int no_alloc = call->flags.bits.frame_is_setup_on_call;
373 int n_res = get_method_n_ress(call_tp);
374 int do_seq = call->flags.bits.store_args_sequential && !no_alloc;
376 ir_node *res_proj = NULL;
377 int n_reg_params = 0;
378 int n_stack_params = 0;
381 const arch_register_t **states = NEW_ARR_F(const arch_register_t*, 0);
382 const arch_register_t **destroyed_regs = NEW_ARR_F(const arch_register_t*, 0);
386 int n_reg_results = 0;
387 const ir_edge_t *edge;
389 int *stack_param_idx;
390 int i, n, destroy_all_regs;
391 int throws_exception;
396 /* Let the isa fill out the abi description for that call node. */
397 arch_env_get_call_abi(arch_env, call_tp, call);
399 /* Insert code to put the stack arguments on the stack. */
400 assert(get_Call_n_params(irn) == n_params);
401 stack_param_idx = ALLOCAN(int, n_params);
402 for (p = 0; p < n_params; ++p) {
403 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
406 int arg_size = get_type_size_bytes(get_method_param_type(call_tp, p));
408 stack_size += round_up2(arg->space_before, arg->alignment);
409 stack_size += round_up2(arg_size, arg->alignment);
410 stack_size += round_up2(arg->space_after, arg->alignment);
412 stack_param_idx[n_stack_params++] = p;
416 /* Collect all arguments which are passed in registers. */
417 reg_param_idxs = ALLOCAN(int, n_params);
418 for (p = 0; p < n_params; ++p) {
419 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
420 if (arg && arg->in_reg) {
421 reg_param_idxs[n_reg_params++] = p;
426 * If the stack is decreasing and we do not want to store sequentially,
427 * or someone else allocated the call frame
428 * we allocate as much space on the stack all parameters need, by
429 * moving the stack pointer along the stack's direction.
431 * Note: we also have to do this for stack_size == 0, because we may have
432 * to adjust stack alignment for the call.
434 if (!do_seq && !no_alloc) {
435 curr_sp = be_new_IncSP(sp, bl, curr_sp, stack_size, 1);
438 dbgi = get_irn_dbg_info(irn);
439 /* If there are some parameters which shall be passed on the stack. */
440 if (n_stack_params > 0) {
442 ir_node **in = ALLOCAN(ir_node*, n_stack_params+1);
445 /* push params in reverse direction because stack grows downwards */
447 for (i = 0; i < n_stack_params >> 1; ++i) {
448 int other = n_stack_params - i - 1;
449 int tmp = stack_param_idx[i];
450 stack_param_idx[i] = stack_param_idx[other];
451 stack_param_idx[other] = tmp;
455 curr_mem = get_Call_mem(irn);
457 in[n_in++] = curr_mem;
460 for (i = 0; i < n_stack_params; ++i) {
461 int p = stack_param_idx[i];
462 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
463 ir_node *param = get_Call_param(irn, p);
464 ir_node *addr = curr_sp;
466 ir_type *param_type = get_method_param_type(call_tp, p);
467 int param_size = get_type_size_bytes(param_type) + arg->space_after;
470 * If we wanted to build the arguments sequentially,
471 * the stack pointer for the next must be incremented,
472 * and the memory value propagated.
476 addr = curr_sp = be_new_IncSP(sp, bl, curr_sp,
477 param_size + arg->space_before, 0);
478 add_irn_dep(curr_sp, curr_mem);
480 curr_ofs += arg->space_before;
481 curr_ofs = round_up2(curr_ofs, arg->alignment);
483 /* Make the expression to compute the argument's offset. */
485 ir_mode *constmode = mach_mode;
486 if (mode_is_reference(mach_mode)) {
489 addr = new_r_Const_long(irg, constmode, curr_ofs);
490 addr = new_r_Add(bl, curr_sp, addr, mach_mode);
494 /* Insert a store for primitive arguments. */
495 if (is_atomic_type(param_type)) {
496 ir_node *nomem = get_irg_no_mem(irg);
497 ir_node *mem_input = do_seq ? curr_mem : nomem;
498 ir_node *store = new_rd_Store(dbgi, bl, mem_input, addr, param, cons_none);
499 mem = new_r_Proj(store, mode_M, pn_Store_M);
501 /* Make a mem copy for compound arguments. */
504 assert(mode_is_reference(get_irn_mode(param)));
505 copy = new_rd_CopyB(dbgi, bl, curr_mem, addr, param, param_type);
506 mem = new_r_Proj(copy, mode_M, pn_CopyB_M);
509 curr_ofs += param_size;
517 /* We need the sync only, if we didn't build the stores sequentially. */
519 if (n_stack_params >= 1) {
520 curr_mem = new_r_Sync(bl, n_in, in);
522 curr_mem = get_Call_mem(irn);
527 /* check for the return_twice property */
528 destroy_all_regs = 0;
529 if (is_SymConst_addr_ent(call_ptr)) {
530 ir_entity *ent = get_SymConst_entity(call_ptr);
532 if (get_entity_additional_properties(ent) & mtp_property_returns_twice)
533 destroy_all_regs = 1;
535 ir_type *call_tp = get_Call_type(irn);
537 if (get_method_additional_properties(call_tp) & mtp_property_returns_twice)
538 destroy_all_regs = 1;
541 /* Put caller save into the destroyed set and state registers in the states
543 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
545 const arch_register_class_t *cls = &arch_env->register_classes[i];
546 for (j = 0; j < cls->n_regs; ++j) {
547 const arch_register_t *reg = arch_register_for_index(cls, j);
549 /* even if destroyed all is specified, neither SP nor FP are
550 * destroyed (else bad things will happen) */
551 if (reg == arch_env->sp || reg == arch_env->bp)
554 if (reg->type & arch_register_type_state) {
555 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
556 ARR_APP1(const arch_register_t*, states, reg);
557 /* we're already in the destroyed set so no need for further
561 if (destroy_all_regs || arch_register_is_caller_save(arch_env, reg)) {
562 if (!(reg->type & arch_register_type_ignore)) {
563 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
569 /* search the largest result proj number */
570 res_projs = ALLOCANZ(ir_node*, n_res);
572 foreach_out_edge(irn, edge) {
573 const ir_edge_t *res_edge;
574 ir_node *irn = get_edge_src_irn(edge);
576 if (!is_Proj(irn) || get_Proj_proj(irn) != pn_Call_T_result)
579 foreach_out_edge(irn, res_edge) {
581 ir_node *res = get_edge_src_irn(res_edge);
583 assert(is_Proj(res));
585 proj = get_Proj_proj(res);
586 assert(proj < n_res);
587 assert(res_projs[proj] == NULL);
588 res_projs[proj] = res;
594 /** TODO: this is not correct for cases where return values are passed
595 * on the stack, but no known ABI does this currently...
597 n_reg_results = n_res;
600 in = ALLOCAN(ir_node*, n_reg_params + ARR_LEN(states));
602 /* make the back end call node and set its register requirements. */
603 for (i = 0; i < n_reg_params; ++i) {
604 in[n_ins++] = get_Call_param(irn, reg_param_idxs[i]);
607 /* add state registers ins */
608 for (s = 0; s < ARR_LEN(states); ++s) {
609 const arch_register_t *reg = states[s];
610 const arch_register_class_t *cls = reg->reg_class;
611 ir_node *regnode = new_r_Unknown(irg, cls->mode);
612 in[n_ins++] = regnode;
614 assert(n_ins == (int) (n_reg_params + ARR_LEN(states)));
616 /* ins collected, build the call */
617 throws_exception = ir_throws_exception(irn);
618 if (env->call->flags.bits.call_has_imm && is_SymConst(call_ptr)) {
620 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, curr_sp,
621 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
622 n_ins, in, get_Call_type(irn));
623 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
626 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, call_ptr,
627 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
628 n_ins, in, get_Call_type(irn));
630 ir_set_throws_exception(low_call, throws_exception);
631 be_Call_set_pop(low_call, call->pop);
633 /* put the call into the list of all calls for later processing */
634 ARR_APP1(ir_node *, env->calls, low_call);
636 /* create new stack pointer */
637 curr_sp = new_r_Proj(low_call, get_irn_mode(curr_sp), pn_be_Call_sp);
638 be_set_constr_single_reg_out(low_call, pn_be_Call_sp, sp,
639 arch_register_req_type_ignore | arch_register_req_type_produces_sp);
640 arch_set_irn_register(curr_sp, sp);
642 /* now handle results */
643 for (i = 0; i < n_res; ++i) {
644 ir_node *proj = res_projs[i];
645 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
646 long pn = i + pn_be_Call_first_res;
648 /* returns values on stack not supported yet */
652 shift the proj number to the right, since we will drop the
653 unspeakable Proj_T from the Call. Therefore, all real argument
654 Proj numbers must be increased by pn_be_Call_first_res
656 pn = i + pn_be_Call_first_res;
659 ir_type *res_type = get_method_res_type(call_tp, i);
660 ir_mode *mode = get_type_mode(res_type);
661 proj = new_r_Proj(low_call, mode, pn);
664 set_Proj_pred(proj, low_call);
665 set_Proj_proj(proj, pn);
669 /* remove register from destroyed regs */
671 size_t n = ARR_LEN(destroyed_regs);
672 for (j = 0; j < n; ++j) {
673 if (destroyed_regs[j] == arg->reg) {
674 destroyed_regs[j] = destroyed_regs[n-1];
675 ARR_SHRINKLEN(destroyed_regs,n-1);
683 Set the register class of the call address to
684 the backend provided class (default: stack pointer class)
686 be_node_set_reg_class_in(low_call, n_be_Call_ptr, call->cls_addr);
688 DBG((dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
690 /* Set the register classes and constraints of the Call parameters. */
691 for (i = 0; i < n_reg_params; ++i) {
692 int index = reg_param_idxs[i];
693 be_abi_call_arg_t *arg = get_call_arg(call, 0, index, 0);
694 assert(arg->reg != NULL);
696 be_set_constr_single_reg_in(low_call, n_be_Call_first_arg + i,
697 arg->reg, arch_register_req_type_none);
700 /* Set the register constraints of the results. */
701 for (i = 0; i < n_res; ++i) {
702 ir_node *proj = res_projs[i];
703 const be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
704 int pn = get_Proj_proj(proj);
707 be_set_constr_single_reg_out(low_call, pn, arg->reg,
708 arch_register_req_type_none);
709 arch_set_irn_register(proj, arg->reg);
711 exchange(irn, low_call);
713 /* kill the ProjT node */
714 if (res_proj != NULL) {
718 /* Make additional projs for the caller save registers
719 and the Keep node which keeps them alive. */
725 int curr_res_proj = pn_be_Call_first_res + n_reg_results;
728 n_ins = ARR_LEN(destroyed_regs) + n_reg_results + 1;
729 in = ALLOCAN(ir_node *, n_ins);
731 /* also keep the stack pointer */
732 set_irn_link(curr_sp, (void*) sp);
735 for (d = 0; d < ARR_LEN(destroyed_regs); ++d) {
736 const arch_register_t *reg = destroyed_regs[d];
737 ir_node *proj = new_r_Proj(low_call, reg->reg_class->mode, curr_res_proj);
739 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
740 be_set_constr_single_reg_out(low_call, curr_res_proj, reg,
741 arch_register_req_type_none);
742 arch_set_irn_register(proj, reg);
744 set_irn_link(proj, (void*) reg);
749 for (i = 0; i < n_reg_results; ++i) {
750 ir_node *proj = res_projs[i];
751 const arch_register_t *reg = arch_get_irn_register(proj);
752 set_irn_link(proj, (void*) reg);
757 /* create the Keep for the caller save registers */
758 keep = be_new_Keep(bl, n, in);
759 for (i = 0; i < n; ++i) {
760 const arch_register_t *reg = (const arch_register_t*)get_irn_link(in[i]);
761 be_node_set_reg_class_in(keep, i, arch_register_get_class(reg));
765 /* Clean up the stack. */
766 assert(stack_size >= call->pop);
767 stack_size -= call->pop;
769 if (stack_size > 0) {
770 ir_node *mem_proj = NULL;
772 foreach_out_edge(low_call, edge) {
773 ir_node *irn = get_edge_src_irn(edge);
774 if (is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
781 mem_proj = new_r_Proj(low_call, mode_M, pn_be_Call_M);
782 keep_alive(mem_proj);
785 /* Clean up the stack frame or revert alignment fixes if we allocated it */
787 curr_sp = be_new_IncSP(sp, bl, curr_sp, -stack_size, 0);
790 be_abi_call_free(call);
793 DEL_ARR_F(destroyed_regs);
799 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
801 * @param alignment the minimum stack alignment
802 * @param size the node containing the non-aligned size
803 * @param block the block where new nodes are allocated on
804 * @param dbg debug info for new nodes
806 * @return a node representing the aligned size
808 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size,
809 ir_node *block, dbg_info *dbg)
811 if (stack_alignment > 1) {
817 assert(is_po2(stack_alignment));
819 mode = get_irn_mode(size);
820 tv = new_tarval_from_long(stack_alignment-1, mode);
821 irg = get_Block_irg(block);
822 mask = new_r_Const(irg, tv);
823 size = new_rd_Add(dbg, block, size, mask, mode);
825 tv = new_tarval_from_long(-(long)stack_alignment, mode);
826 mask = new_r_Const(irg, tv);
827 size = new_rd_And(dbg, block, size, mask, mode);
833 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
835 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp)
837 ir_node *block = get_nodes_block(alloc);
838 ir_graph *irg = get_Block_irg(block);
839 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
840 ir_node *alloc_mem = NULL;
841 ir_node *alloc_res = NULL;
842 ir_type *type = get_Alloc_type(alloc);
845 const ir_edge_t *edge;
850 unsigned stack_alignment;
852 /* all non-stack Alloc nodes should already be lowered before the backend */
853 assert(get_Alloc_where(alloc) == stack_alloc);
855 foreach_out_edge(alloc, edge) {
856 ir_node *irn = get_edge_src_irn(edge);
858 assert(is_Proj(irn));
859 switch (get_Proj_proj(irn)) {
871 /* Beware: currently Alloc nodes without a result might happen,
872 only escape analysis kills them and this phase runs only for object
873 oriented source. We kill the Alloc here. */
874 if (alloc_res == NULL && alloc_mem) {
875 exchange(alloc_mem, get_Alloc_mem(alloc));
879 dbg = get_irn_dbg_info(alloc);
880 count = get_Alloc_count(alloc);
882 /* we might need to multiply the count with the element size */
883 if (type != firm_unknown_type && get_type_size_bytes(type) != 1) {
884 ir_mode *mode = get_irn_mode(count);
885 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type),
887 ir_node *cnst = new_rd_Const(dbg, irg, tv);
888 size = new_rd_Mul(dbg, block, count, cnst, mode);
893 /* The stack pointer will be modified in an unknown manner.
894 We cannot omit it. */
895 env->call->flags.bits.try_omit_fp = 0;
897 stack_alignment = 1 << arch_env->stack_alignment;
898 size = adjust_alloc_size(stack_alignment, size, block, dbg);
899 new_alloc = be_new_AddSP(arch_env->sp, block, curr_sp, size);
900 set_irn_dbg_info(new_alloc, dbg);
902 if (alloc_mem != NULL) {
906 addsp_mem = new_r_Proj(new_alloc, mode_M, pn_be_AddSP_M);
908 /* We need to sync the output mem of the AddSP with the input mem
909 edge into the alloc node. */
910 ins[0] = get_Alloc_mem(alloc);
912 sync = new_r_Sync(block, 2, ins);
914 exchange(alloc_mem, sync);
917 exchange(alloc, new_alloc);
919 /* fix projnum of alloca res */
920 set_Proj_proj(alloc_res, pn_be_AddSP_res);
922 curr_sp = new_r_Proj(new_alloc, get_irn_mode(curr_sp), pn_be_AddSP_sp);
929 * The Free is transformed into a back end free node and connected to the stack nodes.
931 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
933 ir_node *block = get_nodes_block(free);
934 ir_graph *irg = get_irn_irg(free);
935 ir_type *type = get_Free_type(free);
936 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
937 ir_mode *sp_mode = arch_env->sp->reg_class->mode;
938 dbg_info *dbg = get_irn_dbg_info(free);
939 ir_node *subsp, *mem, *res, *size, *sync;
941 unsigned stack_alignment;
943 /* all non-stack-alloc Free nodes should already be lowered before the
945 assert(get_Free_where(free) == stack_alloc);
947 /* we might need to multiply the size with the element size */
948 if (type != firm_unknown_type && get_type_size_bytes(type) != 1) {
949 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
950 ir_node *cnst = new_rd_Const(dbg, irg, tv);
951 ir_node *mul = new_rd_Mul(dbg, block, get_Free_size(free),
955 size = get_Free_size(free);
958 stack_alignment = 1 << arch_env->stack_alignment;
959 size = adjust_alloc_size(stack_alignment, size, block, dbg);
961 /* The stack pointer will be modified in an unknown manner.
962 We cannot omit it. */
963 env->call->flags.bits.try_omit_fp = 0;
964 subsp = be_new_SubSP(arch_env->sp, block, curr_sp, size);
965 set_irn_dbg_info(subsp, dbg);
967 mem = new_r_Proj(subsp, mode_M, pn_be_SubSP_M);
968 res = new_r_Proj(subsp, sp_mode, pn_be_SubSP_sp);
970 /* we need to sync the memory */
971 in[0] = get_Free_mem(free);
973 sync = new_r_Sync(block, 2, in);
975 /* and make the AddSP dependent on the former memory */
976 add_irn_dep(subsp, get_Free_mem(free));
979 exchange(free, sync);
986 * Check if a node is somehow data dependent on another one.
987 * both nodes must be in the same basic block.
988 * @param n1 The first node.
989 * @param n2 The second node.
990 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
992 static int dependent_on(ir_node *n1, ir_node *n2)
994 assert(get_nodes_block(n1) == get_nodes_block(n2));
996 return heights_reachable_in_block(ir_heights, n1, n2);
999 static int cmp_call_dependency(const void *c1, const void *c2)
1001 ir_node *n1 = *(ir_node **) c1;
1002 ir_node *n2 = *(ir_node **) c2;
1006 Classical qsort() comparison function behavior:
1007 0 if both elements are equal
1008 1 if second is "smaller" that first
1009 -1 if first is "smaller" that second
1011 if (dependent_on(n1, n2))
1014 if (dependent_on(n2, n1))
1017 /* The nodes have no depth order, but we need a total order because qsort()
1020 * Additionally, we need to respect transitive dependencies. Consider a
1021 * Call a depending on Call b and an independent Call c.
1022 * We MUST NOT order c > a and b > c. */
1023 h1 = get_irn_height(ir_heights, n1);
1024 h2 = get_irn_height(ir_heights, n2);
1025 if (h1 < h2) return -1;
1026 if (h1 > h2) return 1;
1027 /* Same height, so use a random (but stable) order */
1028 return get_irn_idx(n1) - get_irn_idx(n2);
1032 * Walker: links all Call/Alloc/Free nodes to the Block they are contained.
1033 * Clears the irg_is_leaf flag if a Call is detected.
1035 static void link_ops_in_block_walker(ir_node *irn, void *data)
1037 be_abi_irg_t *env = (be_abi_irg_t*)data;
1038 unsigned code = get_irn_opcode(irn);
1040 if (code == iro_Call ||
1041 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
1042 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
1043 ir_node *bl = get_nodes_block(irn);
1044 void *save = get_irn_link(bl);
1046 if (code == iro_Call)
1047 env->call->flags.bits.irg_is_leaf = 0;
1049 set_irn_link(irn, save);
1050 set_irn_link(bl, irn);
1053 if (code == iro_Builtin && get_Builtin_kind(irn) == ir_bk_return_address) {
1054 ir_node *param = get_Builtin_param(irn, 0);
1055 ir_tarval *tv = get_Const_tarval(param);
1056 unsigned long value = get_tarval_long(tv);
1057 /* use ebp, so the climbframe algo works... */
1059 env->call->flags.bits.try_omit_fp = 0;
1066 * Process all Call/Alloc/Free nodes inside a basic block.
1067 * Note that the link field of the block must contain a linked list of all
1068 * nodes inside the Block. We first order this list according to data dependency
1069 * and that connect the nodes together.
1071 static void process_ops_in_block(ir_node *bl, void *data)
1073 be_abi_irg_t *env = (be_abi_irg_t*)data;
1074 ir_node *curr_sp = env->init_sp;
1081 for (irn = (ir_node*)get_irn_link(bl); irn != NULL;
1082 irn = (ir_node*)get_irn_link(irn)) {
1086 nodes = ALLOCAN(ir_node*, n_nodes);
1087 for (irn = (ir_node*)get_irn_link(bl), n = 0; irn != NULL;
1088 irn = (ir_node*)get_irn_link(irn), ++n) {
1092 /* If there were call nodes in the block. */
1097 /* order the call nodes according to data dependency */
1098 qsort(nodes, n_nodes, sizeof(nodes[0]), cmp_call_dependency);
1100 for (i = n_nodes - 1; i >= 0; --i) {
1101 ir_node *irn = nodes[i];
1103 DBG((dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1104 switch (get_irn_opcode(irn)) {
1107 /* The stack pointer will be modified due to a call. */
1108 env->call->flags.bits.try_omit_fp = 0;
1110 curr_sp = adjust_call(env, irn, curr_sp);
1113 if (get_Alloc_where(irn) == stack_alloc)
1114 curr_sp = adjust_alloc(env, irn, curr_sp);
1117 if (get_Free_where(irn) == stack_alloc)
1118 curr_sp = adjust_free(env, irn, curr_sp);
1121 panic("invalid call");
1125 /* Keep the last stack state in the block by tying it to Keep node,
1126 * the proj from calls is already kept */
1127 if (curr_sp != env->init_sp &&
1128 !(is_Proj(curr_sp) && be_is_Call(get_Proj_pred(curr_sp)))) {
1130 keep = be_new_Keep(bl, 1, nodes);
1131 pmap_insert(env->keep_map, bl, keep);
1135 set_irn_link(bl, curr_sp);
1139 * Adjust all call nodes in the graph to the ABI conventions.
1141 static void process_calls(ir_graph *irg)
1143 be_abi_irg_t *abi = be_get_irg_abi(irg);
1145 abi->call->flags.bits.irg_is_leaf = 1;
1146 irg_walk_graph(irg, firm_clear_link, link_ops_in_block_walker, abi);
1148 ir_heights = heights_new(irg);
1149 irg_block_walk_graph(irg, NULL, process_ops_in_block, abi);
1150 heights_free(ir_heights);
1154 * Computes the stack argument layout type.
1155 * Changes a possibly allocated value param type by moving
1156 * entities to the stack layout type.
1158 * @param call the current call ABI
1159 * @param method_type the method type
1160 * @param param_map an array mapping method arguments to the stack layout
1163 * @return the stack argument layout type
1165 static ir_type *compute_arg_type(ir_graph *irg, be_abi_call_t *call,
1166 ir_type *method_type, ir_entity ***param_map)
1168 struct obstack *obst = be_get_be_obst(irg);
1169 ir_type *frame_type = get_irg_frame_type(irg);
1170 size_t n_params = get_method_n_params(method_type);
1171 size_t n_frame_members = get_compound_n_members(frame_type);
1179 *param_map = map = OALLOCNZ(obst, ir_entity*, n_params);
1180 res = new_type_struct(new_id_from_chars("arg_type", 8));
1182 /* collect existing entities for value_param_types */
1183 for (f = n_frame_members; f > 0; ) {
1184 ir_entity *entity = get_compound_member(frame_type, --f);
1187 set_entity_link(entity, NULL);
1188 if (!is_parameter_entity(entity))
1190 num = get_entity_parameter_number(entity);
1191 assert(num < n_params);
1192 if (map[num] != NULL)
1193 panic("multiple entities for parameter %u in %+F found", f, irg);
1195 if (!get_call_arg(call, 0, num, 1)->on_stack) {
1196 /* don't move this entity */
1201 /* move entity to new arg_type */
1202 set_entity_owner(entity, res);
1205 for (i = 0; i < n_params; ++i) {
1206 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1207 ir_type *param_type = get_method_param_type(method_type, i);
1210 if (!arg->on_stack) {
1214 if (entity == NULL) {
1215 /* create a new entity */
1216 entity = new_parameter_entity(res, i, param_type);
1218 ofs += arg->space_before;
1219 ofs = round_up2(ofs, arg->alignment);
1220 set_entity_offset(entity, ofs);
1221 ofs += arg->space_after;
1222 ofs += get_type_size_bytes(param_type);
1223 arg->stack_ent = entity;
1225 set_type_size_bytes(res, ofs);
1226 set_type_state(res, layout_fixed);
1231 const arch_register_t *reg;
1235 static int cmp_regs(const void *a, const void *b)
1237 const reg_node_map_t *p = (const reg_node_map_t*)a;
1238 const reg_node_map_t *q = (const reg_node_map_t*)b;
1240 if (p->reg->reg_class == q->reg->reg_class)
1241 return p->reg->index - q->reg->index;
1243 return p->reg->reg_class < q->reg->reg_class ? -1 : +1;
1246 static void reg_map_to_arr(reg_node_map_t *res, pmap *reg_map)
1249 size_t n = pmap_count(reg_map);
1252 foreach_pmap(reg_map, ent) {
1253 res[i].reg = (const arch_register_t*)ent->key;
1254 res[i].irn = (ir_node*)ent->value;
1258 qsort(res, n, sizeof(res[0]), cmp_regs);
1262 * Creates a be_Return for a Return node.
1264 * @param @env the abi environment
1265 * @param irn the Return node or NULL if there was none
1266 * @param bl the block where the be_Retun should be placed
1267 * @param mem the current memory
1268 * @param n_res number of return results
1270 static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl,
1271 ir_node *mem, int n_res)
1273 be_abi_call_t *call = env->call;
1274 ir_graph *irg = get_Block_irg(bl);
1275 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1277 pmap *reg_map = pmap_create();
1278 ir_node *keep = (ir_node*)pmap_get(env->keep_map, bl);
1285 const arch_register_t **regs;
1289 get the valid stack node in this block.
1290 If we had a call in that block there is a Keep constructed by process_calls()
1291 which points to the last stack modification in that block. we'll use
1292 it then. Else we use the stack from the start block and let
1293 the ssa construction fix the usage.
1295 stack = be_abi_reg_map_get(env->regs, arch_env->sp);
1297 stack = get_irn_n(keep, 0);
1299 remove_End_keepalive(get_irg_end(irg), keep);
1302 /* Insert results for Return into the register map. */
1303 for (i = 0; i < n_res; ++i) {
1304 ir_node *res = get_Return_res(irn, i);
1305 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1306 assert(arg->in_reg && "return value must be passed in register");
1307 pmap_insert(reg_map, (void *) arg->reg, res);
1310 /* Add uses of the callee save registers. */
1311 foreach_pmap(env->regs, ent) {
1312 const arch_register_t *reg = (const arch_register_t*)ent->key;
1313 if ((reg->type & arch_register_type_ignore) || arch_register_is_callee_save(arch_env, reg))
1314 pmap_insert(reg_map, ent->key, ent->value);
1317 be_abi_reg_map_set(reg_map, arch_env->sp, stack);
1320 Maximum size of the in array for Return nodes is
1321 return args + callee save/ignore registers + memory + stack pointer
1323 in_max = pmap_count(reg_map) + n_res + 2;
1325 in = ALLOCAN(ir_node*, in_max);
1326 regs = ALLOCAN(arch_register_t const*, in_max);
1329 in[1] = be_abi_reg_map_get(reg_map, arch_env->sp);
1331 regs[1] = arch_env->sp;
1334 /* clear SP entry, since it has already been grown. */
1335 pmap_insert(reg_map, (void *) arch_env->sp, NULL);
1336 for (i = 0; i < n_res; ++i) {
1337 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1339 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1340 regs[n++] = arg->reg;
1342 /* Clear the map entry to mark the register as processed. */
1343 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1346 /* grow the rest of the stuff. */
1347 foreach_pmap(reg_map, ent) {
1349 in[n] = (ir_node*)ent->value;
1350 regs[n++] = (const arch_register_t*)ent->key;
1354 /* The in array for the new back end return is now ready. */
1356 dbgi = get_irn_dbg_info(irn);
1360 /* we have to pop the shadow parameter in in case of struct returns */
1362 ret = be_new_Return(dbgi, irg, bl, n_res, pop, n, in);
1364 /* Set the register classes of the return's parameter accordingly. */
1365 for (i = 0; i < n; ++i) {
1366 if (regs[i] == NULL)
1369 be_set_constr_single_reg_in(ret, i, regs[i], arch_register_req_type_none);
1372 /* Free the space of the Epilog's in array and the register <-> proj map. */
1373 pmap_destroy(reg_map);
1378 typedef struct lower_frame_sels_env_t {
1379 ir_node *frame; /**< the current frame */
1380 const arch_register_class_t *sp_class; /**< register class of the stack pointer */
1381 const arch_register_class_t *link_class; /**< register class of the link pointer */
1382 ir_type *frame_tp; /**< the frame type */
1383 int static_link_pos; /**< argument number of the hidden static link */
1384 } lower_frame_sels_env_t;
1387 * Walker: Replaces Sels of frame type and
1388 * value param type entities by FrameAddress.
1389 * Links all used entities.
1391 static void lower_frame_sels_walker(ir_node *irn, void *data)
1393 lower_frame_sels_env_t *ctx = (lower_frame_sels_env_t*)data;
1396 ir_node *ptr = get_Sel_ptr(irn);
1398 if (ptr == ctx->frame) {
1399 ir_entity *ent = get_Sel_entity(irn);
1400 ir_node *bl = get_nodes_block(irn);
1403 nw = be_new_FrameAddr(ctx->sp_class, bl, ctx->frame, ent);
1410 * The start block has no jump, instead it has an initial exec Proj.
1411 * The backend wants to handle all blocks the same way, so we replace
1412 * the out cfg edge with a real jump.
1414 static void fix_start_block(ir_graph *irg)
1416 ir_node *initial_X = get_irg_initial_exec(irg);
1417 ir_node *start_block = get_irg_start_block(irg);
1418 ir_node *jmp = new_r_Jmp(start_block);
1420 assert(is_Proj(initial_X));
1421 exchange(initial_X, jmp);
1422 set_irg_initial_exec(irg, new_r_Bad(irg, mode_X));
1424 /* merge start block with successor if possible */
1426 const ir_edge_t *edge;
1427 foreach_out_edge(jmp, edge) {
1428 ir_node *succ = get_edge_src_irn(edge);
1429 if (!is_Block(succ))
1432 if (get_irn_arity(succ) == 1) {
1433 exchange(succ, start_block);
1441 * Modify the irg itself and the frame type.
1443 static void modify_irg(ir_graph *irg)
1445 be_abi_irg_t *env = be_get_irg_abi(irg);
1446 be_abi_call_t *call = env->call;
1447 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1448 const arch_register_t *sp = arch_env->sp;
1449 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1450 be_irg_t *birg = be_birg_from_irg(irg);
1451 struct obstack *obst = be_get_be_obst(irg);
1452 be_stack_layout_t *stack_layout = be_get_irg_stack_layout(irg);
1455 ir_node *new_mem_proj;
1461 unsigned frame_size;
1464 const arch_register_t *fp_reg;
1465 ir_node *frame_pointer;
1469 const ir_edge_t *edge;
1470 ir_type *arg_type, *bet_type;
1471 lower_frame_sels_env_t ctx;
1472 ir_entity **param_map;
1474 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1476 old_mem = get_irg_initial_mem(irg);
1478 irp_reserve_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1480 arg_type = compute_arg_type(irg, call, method_type, ¶m_map);
1482 /* Convert the Sel nodes in the irg to frame addr nodes: */
1483 ctx.frame = get_irg_frame(irg);
1484 ctx.sp_class = arch_env->sp->reg_class;
1485 ctx.link_class = arch_env->link_class;
1486 ctx.frame_tp = get_irg_frame_type(irg);
1488 /* layout the stackframe now */
1489 if (get_type_state(ctx.frame_tp) == layout_undefined) {
1490 default_layout_compound_type(ctx.frame_tp);
1493 /* align stackframe to 4 byte */
1494 frame_size = get_type_size_bytes(ctx.frame_tp);
1495 if (frame_size % 4 != 0) {
1496 set_type_size_bytes(ctx.frame_tp, frame_size + 4 - (frame_size % 4));
1499 env->regs = pmap_create();
1501 n_params = get_method_n_params(method_type);
1502 args = OALLOCNZ(obst, ir_node*, n_params);
1504 be_add_parameter_entity_stores(irg);
1506 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1508 irp_free_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1510 /* Fill the argument vector */
1511 arg_tuple = get_irg_args(irg);
1512 foreach_out_edge(arg_tuple, edge) {
1513 ir_node *irn = get_edge_src_irn(edge);
1514 if (! is_Anchor(irn)) {
1515 int nr = get_Proj_proj(irn);
1517 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1521 stack_layout->sp_relative = call->flags.bits.try_omit_fp;
1522 bet_type = call->cb->get_between_type(irg);
1523 stack_frame_init(stack_layout, arg_type, bet_type,
1524 get_irg_frame_type(irg), param_map);
1526 /* Count the register params and add them to the number of Projs for the RegParams node */
1527 for (i = 0; i < n_params; ++i) {
1528 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1529 if (arg->in_reg && args[i]) {
1530 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1531 assert(i == get_Proj_proj(args[i]));
1533 /* For now, associate the register with the old Proj from Start representing that argument. */
1534 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1535 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1539 /* Collect all callee-save registers */
1540 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
1541 const arch_register_class_t *cls = &arch_env->register_classes[i];
1542 for (j = 0; j < cls->n_regs; ++j) {
1543 const arch_register_t *reg = &cls->regs[j];
1544 if ((reg->type & arch_register_type_state) || arch_register_is_callee_save(arch_env, reg)) {
1545 pmap_insert(env->regs, (void *) reg, NULL);
1550 fp_reg = call->flags.bits.try_omit_fp ? arch_env->sp : arch_env->bp;
1551 rbitset_clear(birg->allocatable_regs, fp_reg->global_index);
1553 /* handle start block here (place a jump in the block) */
1554 fix_start_block(irg);
1556 pmap_insert(env->regs, (void *) sp, NULL);
1557 pmap_insert(env->regs, (void *) arch_env->bp, NULL);
1558 start_bl = get_irg_start_block(irg);
1559 env->start = be_new_Start(NULL, start_bl, pmap_count(env->regs) + 1);
1560 set_irg_start(irg, env->start);
1563 * make proj nodes for the callee save registers.
1564 * memorize them, since Return nodes get those as inputs.
1566 * Note, that if a register corresponds to an argument, the regs map
1567 * contains the old Proj from start for that argument.
1569 rm = ALLOCAN(reg_node_map_t, pmap_count(env->regs));
1570 reg_map_to_arr(rm, env->regs);
1571 for (i = 0, n = pmap_count(env->regs); i < n; ++i) {
1572 const arch_register_t *reg = rm[i].reg;
1573 ir_mode *mode = reg->reg_class->mode;
1575 arch_register_req_type_t add_type = arch_register_req_type_none;
1579 add_type |= arch_register_req_type_produces_sp;
1580 if (!rbitset_is_set(birg->allocatable_regs, reg->global_index)) {
1581 add_type |= arch_register_req_type_ignore;
1585 proj = new_r_Proj(env->start, mode, nr + 1);
1586 pmap_insert(env->regs, (void *) reg, proj);
1587 be_set_constr_single_reg_out(env->start, nr + 1, reg, add_type);
1588 arch_set_irn_register(proj, reg);
1590 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1593 /* create a new initial memory proj */
1594 assert(is_Proj(old_mem));
1595 arch_set_irn_register_req_out(env->start, 0, arch_no_register_req);
1596 new_mem_proj = new_r_Proj(env->start, mode_M, 0);
1598 set_irg_initial_mem(irg, mem);
1600 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1602 /* set new frame_pointer */
1603 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1604 set_irg_frame(irg, frame_pointer);
1606 /* rewire old mem users to new mem */
1607 exchange(old_mem, mem);
1609 /* keep the mem (for functions with an endless loop = no return) */
1612 set_irg_initial_mem(irg, mem);
1614 /* Now, introduce stack param nodes for all parameters passed on the stack */
1615 for (i = 0; i < n_params; ++i) {
1616 ir_node *arg_proj = args[i];
1617 ir_node *repl = NULL;
1619 if (arg_proj != NULL) {
1620 be_abi_call_arg_t *arg;
1621 ir_type *param_type;
1622 int nr = get_Proj_proj(arg_proj);
1625 nr = MIN(nr, n_params);
1626 arg = get_call_arg(call, 0, nr, 1);
1627 param_type = get_method_param_type(method_type, nr);
1630 repl = (ir_node*)pmap_get(env->regs, arg->reg);
1631 } else if (arg->on_stack) {
1632 ir_node *addr = be_new_FrameAddr(sp->reg_class, start_bl, frame_pointer, arg->stack_ent);
1634 /* For atomic parameters which are actually used, we create a Load node. */
1635 if (is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1636 ir_mode *mode = get_type_mode(param_type);
1637 ir_mode *load_mode = arg->load_mode;
1638 ir_node *nomem = get_irg_no_mem(irg);
1640 ir_node *load = new_r_Load(start_bl, nomem, addr, load_mode, cons_floats);
1641 repl = new_r_Proj(load, load_mode, pn_Load_res);
1643 if (mode != load_mode) {
1644 repl = new_r_Conv(start_bl, repl, mode);
1647 /* The stack parameter is not primitive (it is a struct or array),
1648 * we thus will create a node representing the parameter's address
1654 assert(repl != NULL);
1656 /* Beware: the mode of the register parameters is always the mode of the register class
1657 which may be wrong. Add Conv's then. */
1658 mode = get_irn_mode(args[i]);
1659 if (mode != get_irn_mode(repl)) {
1660 repl = new_r_Conv(get_nodes_block(repl), repl, mode);
1662 exchange(args[i], repl);
1666 /* the arg proj is not needed anymore now and should be only used by the anchor */
1667 assert(get_irn_n_edges(arg_tuple) == 1);
1668 kill_node(arg_tuple);
1669 set_irg_args(irg, new_r_Bad(irg, mode_T));
1671 /* All Return nodes hang on the End node, so look for them there. */
1672 end = get_irg_end_block(irg);
1673 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1674 ir_node *irn = get_Block_cfgpred(end, i);
1676 if (is_Return(irn)) {
1677 ir_node *blk = get_nodes_block(irn);
1678 ir_node *mem = get_Return_mem(irn);
1679 ir_node *ret = create_be_return(env, irn, blk, mem, get_Return_n_ress(irn));
1684 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1685 the code is dead and will never be executed. */
1688 /** Fix the state inputs of calls that still hang on unknowns */
1689 static void fix_call_state_inputs(ir_graph *irg)
1691 be_abi_irg_t *env = be_get_irg_abi(irg);
1692 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1694 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1696 /* Collect caller save registers */
1697 n = arch_env->n_register_classes;
1698 for (i = 0; i < n; ++i) {
1700 const arch_register_class_t *cls = &arch_env->register_classes[i];
1701 for (j = 0; j < cls->n_regs; ++j) {
1702 const arch_register_t *reg = arch_register_for_index(cls, j);
1703 if (reg->type & arch_register_type_state) {
1704 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1709 n = ARR_LEN(env->calls);
1710 n_states = ARR_LEN(stateregs);
1711 for (i = 0; i < n; ++i) {
1713 ir_node *call = env->calls[i];
1715 arity = get_irn_arity(call);
1717 /* the state reg inputs are the last n inputs of the calls */
1718 for (s = 0; s < n_states; ++s) {
1719 int inp = arity - n_states + s;
1720 const arch_register_t *reg = stateregs[s];
1721 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
1723 set_irn_n(call, inp, regnode);
1727 DEL_ARR_F(stateregs);
1731 * Create a trampoline entity for the given method.
1733 static ir_entity *create_trampoline(be_main_env_t *be, ir_entity *method)
1735 ir_type *type = get_entity_type(method);
1736 ident *old_id = get_entity_ld_ident(method);
1737 ident *id = id_mangle3("", old_id, "$stub");
1738 ir_type *parent = be->pic_trampolines_type;
1739 ir_entity *ent = new_entity(parent, old_id, type);
1740 set_entity_ld_ident(ent, id);
1741 set_entity_visibility(ent, ir_visibility_private);
1747 * Returns the trampoline entity for the given method.
1749 static ir_entity *get_trampoline(be_main_env_t *env, ir_entity *method)
1751 ir_entity *result = (ir_entity*)pmap_get(env->ent_trampoline_map, method);
1752 if (result == NULL) {
1753 result = create_trampoline(env, method);
1754 pmap_insert(env->ent_trampoline_map, method, result);
1760 static ir_entity *create_pic_symbol(be_main_env_t *be, ir_entity *entity)
1762 ident *old_id = get_entity_ld_ident(entity);
1763 ident *id = id_mangle3("", old_id, "$non_lazy_ptr");
1764 ir_type *e_type = get_entity_type(entity);
1765 ir_type *type = new_type_pointer(e_type);
1766 ir_type *parent = be->pic_symbols_type;
1767 ir_entity *ent = new_entity(parent, old_id, type);
1768 set_entity_ld_ident(ent, id);
1769 set_entity_visibility(ent, ir_visibility_private);
1774 static ir_entity *get_pic_symbol(be_main_env_t *env, ir_entity *entity)
1776 ir_entity *result = (ir_entity*)pmap_get(env->ent_pic_symbol_map, entity);
1777 if (result == NULL) {
1778 result = create_pic_symbol(env, entity);
1779 pmap_insert(env->ent_pic_symbol_map, entity, result);
1788 * Returns non-zero if a given entity can be accessed using a relative address.
1790 static int can_address_relative(ir_entity *entity)
1792 return get_entity_visibility(entity) != ir_visibility_external
1793 && !(get_entity_linkage(entity) & IR_LINKAGE_MERGE);
1796 static ir_node *get_pic_base(ir_graph *irg)
1798 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1799 if (arch_env->impl->get_pic_base == NULL)
1801 return arch_env->impl->get_pic_base(irg);
1804 /** patches SymConsts to work in position independent code */
1805 static void fix_pic_symconsts(ir_node *node, void *data)
1807 ir_graph *irg = get_irn_irg(node);
1808 be_main_env_t *be = be_get_irg_main_env(irg);
1818 arity = get_irn_arity(node);
1819 for (i = 0; i < arity; ++i) {
1821 ir_node *pred = get_irn_n(node, i);
1823 ir_entity *pic_symbol;
1824 ir_node *pic_symconst;
1826 if (!is_SymConst(pred))
1829 entity = get_SymConst_entity(pred);
1830 block = get_nodes_block(pred);
1832 /* calls can jump to relative addresses, so we can directly jump to
1833 the (relatively) known call address or the trampoline */
1834 if (i == 1 && is_Call(node)) {
1835 ir_entity *trampoline;
1836 ir_node *trampoline_const;
1838 if (can_address_relative(entity))
1841 dbgi = get_irn_dbg_info(pred);
1842 trampoline = get_trampoline(be, entity);
1843 trampoline_const = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1845 set_irn_n(node, i, trampoline_const);
1849 /* everything else is accessed relative to EIP */
1850 mode = get_irn_mode(pred);
1851 pic_base = get_pic_base(irg);
1853 /* all ok now for locally constructed stuff */
1854 if (can_address_relative(entity)) {
1855 ir_node *add = new_r_Add(block, pic_base, pred, mode);
1857 /* make sure the walker doesn't visit this add again */
1858 mark_irn_visited(add);
1859 set_irn_n(node, i, add);
1863 /* get entry from pic symbol segment */
1864 dbgi = get_irn_dbg_info(pred);
1865 pic_symbol = get_pic_symbol(be, entity);
1866 pic_symconst = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1868 add = new_r_Add(block, pic_base, pic_symconst, mode);
1869 mark_irn_visited(add);
1871 /* we need an extra indirection for global data outside our current
1872 module. The loads are always safe and can therefore float
1873 and need no memory input */
1874 load = new_r_Load(block, get_irg_no_mem(irg), add, mode, cons_floats);
1875 load_res = new_r_Proj(load, mode, pn_Load_res);
1877 set_irn_n(node, i, load_res);
1881 void be_abi_introduce(ir_graph *irg)
1883 be_abi_irg_t *env = XMALLOCZ(be_abi_irg_t);
1884 ir_node *old_frame = get_irg_frame(irg);
1885 be_options_t *options = be_get_irg_options(irg);
1886 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1887 ir_entity *entity = get_irg_entity(irg);
1888 ir_type *method_type = get_entity_type(entity);
1889 be_irg_t *birg = be_birg_from_irg(irg);
1890 struct obstack *obst = &birg->obst;
1891 ir_node *dummy = new_r_Dummy(irg,
1892 arch_env->sp->reg_class->mode);
1895 /* determine allocatable registers */
1896 assert(birg->allocatable_regs == NULL);
1897 birg->allocatable_regs = rbitset_obstack_alloc(obst, arch_env->n_registers);
1898 for (r = 0; r < arch_env->n_registers; ++r) {
1899 const arch_register_t *reg = &arch_env->registers[r];
1900 if ( !(reg->type & arch_register_type_ignore)) {
1901 rbitset_set(birg->allocatable_regs, r);
1905 /* break here if backend provides a custom API.
1906 * Note: we shouldn't have to setup any be_abi_irg_t* stuff at all,
1907 * but need more cleanup to make this work
1909 be_set_irg_abi(irg, env);
1911 be_omit_fp = options->omit_fp;
1913 env->keep_map = pmap_create();
1914 env->call = be_abi_call_new(arch_env->sp->reg_class);
1915 arch_env_get_call_abi(arch_env, method_type, env->call);
1917 env->init_sp = dummy;
1918 env->calls = NEW_ARR_F(ir_node*, 0);
1923 irg_walk_graph(irg, fix_pic_symconsts, NULL, env);
1926 /* Lower all call nodes in the IRG. */
1929 /* Process the IRG */
1932 /* fix call inputs for state registers */
1933 fix_call_state_inputs(irg);
1935 /* We don't need the keep map anymore. */
1936 pmap_destroy(env->keep_map);
1937 env->keep_map = NULL;
1939 /* calls array is not needed anymore */
1940 DEL_ARR_F(env->calls);
1943 /* reroute the stack origin of the calls to the true stack origin. */
1944 exchange(dummy, env->init_sp);
1945 exchange(old_frame, get_irg_frame(irg));
1947 pmap_destroy(env->regs);
1951 void be_abi_free(ir_graph *irg)
1953 be_abi_irg_t *env = be_get_irg_abi(irg);
1955 if (env->call != NULL)
1956 be_abi_call_free(env->call);
1957 assert(env->regs == NULL);
1960 be_set_irg_abi(irg, NULL);
1963 void be_put_allocatable_regs(const ir_graph *irg,
1964 const arch_register_class_t *cls, bitset_t *bs)
1966 be_irg_t *birg = be_birg_from_irg(irg);
1967 unsigned *allocatable_regs = birg->allocatable_regs;
1970 assert(bitset_size(bs) == cls->n_regs);
1971 bitset_clear_all(bs);
1972 for (i = 0; i < cls->n_regs; ++i) {
1973 const arch_register_t *reg = &cls->regs[i];
1974 if (rbitset_is_set(allocatable_regs, reg->global_index))
1979 unsigned be_get_n_allocatable_regs(const ir_graph *irg,
1980 const arch_register_class_t *cls)
1982 bitset_t *bs = bitset_alloca(cls->n_regs);
1983 be_put_allocatable_regs(irg, cls, bs);
1984 return bitset_popcount(bs);
1987 void be_set_allocatable_regs(const ir_graph *irg,
1988 const arch_register_class_t *cls,
1989 unsigned *raw_bitset)
1991 be_irg_t *birg = be_birg_from_irg(irg);
1992 unsigned *allocatable_regs = birg->allocatable_regs;
1995 rbitset_clear_all(raw_bitset, cls->n_regs);
1996 for (i = 0; i < cls->n_regs; ++i) {
1997 const arch_register_t *reg = &cls->regs[i];
1998 if (rbitset_is_set(allocatable_regs, reg->global_index))
1999 rbitset_set(raw_bitset, i);
2003 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_abi)
2004 void be_init_abi(void)
2006 FIRM_DBG_REGISTER(dbg, "firm.be.abi");