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
31 #include "irgraph_t.h"
34 #include "iredges_t.h"
37 #include "irprintf_t.h"
44 #include "raw_bitset.h"
55 #include "bessaconstr.h"
57 #include "betranshlp.h"
59 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
61 typedef struct be_abi_call_arg_t {
62 unsigned is_res : 1; /**< 1: the call argument is a return value. 0: it's a call parameter. */
63 unsigned in_reg : 1; /**< 1: this argument is transmitted in registers. */
64 unsigned on_stack : 1; /**< 1: this argument is transmitted on the stack. */
65 unsigned callee : 1; /**< 1: someone called us. 0: We call another function */
68 const arch_register_t *reg;
71 unsigned alignment; /**< stack alignment */
72 unsigned space_before; /**< allocate space before */
73 unsigned space_after; /**< allocate space after */
76 struct be_abi_call_t {
77 be_abi_call_flags_t flags; /**< Flags describing the ABI behavior on calls */
78 int pop; /**< number of bytes the stack frame is shrinked by the callee on return. */
79 const be_abi_callbacks_t *cb;
80 ir_type *between_type;
82 const arch_register_class_t *cls_addr; /**< register class of the call address */
86 * The ABI information for the current graph.
89 be_abi_call_t *call; /**< The ABI call information. */
91 ir_node *init_sp; /**< The node representing the stack pointer
92 at the start of the function. */
94 ir_node *start; /**< The be_Start params node. */
95 pmap *regs; /**< A map of all callee-save and ignore regs to
96 their Projs to the RegParams node. */
97 int start_block_bias; /**< The stack bias at the end of the start block. */
99 pmap *keep_map; /**< mapping blocks to keep nodes. */
101 ir_node **calls; /**< flexible array containing all be_Call nodes */
104 static ir_heights_t *ir_heights;
106 /** Flag: if set, try to omit the frame pointer in all routines. */
107 static int be_omit_fp = 1;
109 static ir_node *be_abi_reg_map_get(pmap *map, const arch_register_t *reg)
111 return pmap_get(ir_node, map, reg);
114 static void be_abi_reg_map_set(pmap *map, const arch_register_t* reg,
117 pmap_insert(map, reg, node);
121 * Check if the given register is callee save, ie. will be saved by the callee.
123 static bool arch_register_is_callee_save(
124 const arch_env_t *arch_env,
125 const arch_register_t *reg)
127 if (arch_env->impl->register_saved_by)
128 return arch_env->impl->register_saved_by(reg, /*callee=*/1);
133 * Check if the given register is caller save, ie. must be saved by the caller.
135 static bool arch_register_is_caller_save(
136 const arch_env_t *arch_env,
137 const arch_register_t *reg)
139 if (arch_env->impl->register_saved_by)
140 return arch_env->impl->register_saved_by(reg, /*callee=*/0);
147 _ ____ ___ ____ _ _ _ _
148 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
149 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
150 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
151 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
153 These callbacks are used by the backend to set the parameters
154 for a specific call type.
158 * Set compare function: compares two ABI call object arguments.
160 static int cmp_call_arg(const void *a, const void *b, size_t n)
162 const be_abi_call_arg_t *p = (const be_abi_call_arg_t*)a;
163 const be_abi_call_arg_t *q = (const be_abi_call_arg_t*)b;
165 return !(p->is_res == q->is_res && p->pos == q->pos && p->callee == q->callee);
169 * Get an ABI call object argument.
171 * @param call the abi call
172 * @param is_res true for call results, false for call arguments
173 * @param pos position of the argument
174 * @param callee context type - if we are callee or caller
176 static be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos, int callee)
178 be_abi_call_arg_t arg;
181 memset(&arg, 0, sizeof(arg));
186 hash = is_res * 128 + pos;
188 return set_find(be_abi_call_arg_t, call->params, &arg, sizeof(arg), hash);
192 * Set an ABI call object argument.
194 static void remember_call_arg(be_abi_call_arg_t *arg, be_abi_call_t *call, be_abi_context_t context)
196 unsigned hash = arg->is_res * 128 + arg->pos;
197 if (context & ABI_CONTEXT_CALLEE) {
199 (void)set_insert(be_abi_call_arg_t, call->params, arg, sizeof(*arg), hash);
201 if (context & ABI_CONTEXT_CALLER) {
203 (void)set_insert(be_abi_call_arg_t, call->params, arg, sizeof(*arg), hash);
207 /* Set the flags for a call. */
208 void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const be_abi_callbacks_t *cb)
214 /* Sets the number of bytes the stackframe is shrinked by the callee on return */
215 void be_abi_call_set_pop(be_abi_call_t *call, int pop)
221 /* Set register class for call address */
222 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
224 call->cls_addr = cls;
228 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos,
229 ir_mode *load_mode, unsigned alignment,
230 unsigned space_before, unsigned space_after,
231 be_abi_context_t context)
233 be_abi_call_arg_t arg;
234 memset(&arg, 0, sizeof(arg));
235 assert(alignment > 0 && "Alignment must be greater than 0");
237 arg.load_mode = load_mode;
238 arg.alignment = alignment;
239 arg.space_before = space_before;
240 arg.space_after = space_after;
244 remember_call_arg(&arg, call, context);
247 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
249 be_abi_call_arg_t arg;
250 memset(&arg, 0, sizeof(arg));
257 remember_call_arg(&arg, call, context);
260 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
262 be_abi_call_arg_t arg;
263 memset(&arg, 0, sizeof(arg));
270 remember_call_arg(&arg, call, context);
273 /* Get the flags of a ABI call object. */
274 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
280 * Constructor for a new ABI call object.
282 * @param cls_addr register class of the call address
284 * @return the new ABI call object
286 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
288 be_abi_call_t *call = XMALLOCZ(be_abi_call_t);
291 call->params = new_set(cmp_call_arg, 16);
293 call->cls_addr = cls_addr;
295 call->flags.bits.try_omit_fp = be_omit_fp;
301 * Destructor for an ABI call object.
303 static void be_abi_call_free(be_abi_call_t *call)
305 del_set(call->params);
310 * Initializes the frame layout from parts
312 * @param frame the stack layout that will be initialized
313 * @param args the stack argument layout type
314 * @param between the between layout type
315 * @param locals the method frame type
317 * @return the initialized stack layout
319 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
320 ir_type *between, ir_type *locals)
322 frame->arg_type = args;
323 frame->between_type = between;
324 frame->frame_type = locals;
325 frame->initial_offset = 0;
326 frame->initial_bias = 0;
327 frame->order[1] = between;
329 /* typical decreasing stack: locals have the
330 * lowest addresses, arguments the highest */
331 frame->order[0] = locals;
332 frame->order[2] = args;
343 Adjustment of the calls inside a graph.
348 * Transform a call node into a be_Call node.
350 * @param env The ABI environment for the current irg.
351 * @param irn The call node.
352 * @param curr_sp The stack pointer node to use.
353 * @return The stack pointer after the call.
355 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp)
357 ir_graph *irg = get_irn_irg(irn);
358 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
359 ir_type *call_tp = get_Call_type(irn);
360 ir_node *call_ptr = get_Call_ptr(irn);
361 size_t n_params = get_method_n_params(call_tp);
362 ir_node *curr_mem = get_Call_mem(irn);
363 ir_node *bl = get_nodes_block(irn);
365 const arch_register_t *sp = arch_env->sp;
366 be_abi_call_t *call = be_abi_call_new(sp->reg_class);
367 ir_mode *mach_mode = sp->reg_class->mode;
368 int n_res = get_method_n_ress(call_tp);
370 ir_node *res_proj = NULL;
371 int n_reg_params = 0;
372 int n_stack_params = 0;
375 const arch_register_t **states = NEW_ARR_F(const arch_register_t*, 0);
376 const arch_register_t **destroyed_regs = NEW_ARR_F(const arch_register_t*, 0);
380 int n_reg_results = 0;
382 int *stack_param_idx;
384 int throws_exception;
389 /* Let the isa fill out the abi description for that call node. */
390 arch_env_get_call_abi(arch_env, call_tp, call);
392 /* Insert code to put the stack arguments on the stack. */
393 assert((size_t)get_Call_n_params(irn) == n_params);
394 stack_param_idx = ALLOCAN(int, n_params);
395 for (p = 0; p < n_params; ++p) {
396 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
399 int arg_size = get_type_size_bytes(get_method_param_type(call_tp, p));
401 stack_size += round_up2(arg->space_before, arg->alignment);
402 stack_size += round_up2(arg_size, arg->alignment);
403 stack_size += round_up2(arg->space_after, arg->alignment);
405 stack_param_idx[n_stack_params++] = p;
409 /* Collect all arguments which are passed in registers. */
410 reg_param_idxs = ALLOCAN(int, n_params);
411 for (p = 0; p < n_params; ++p) {
412 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
413 if (arg && arg->in_reg) {
414 reg_param_idxs[n_reg_params++] = p;
419 * If the stack is decreasing and we do not want to store sequentially,
420 * or someone else allocated the call frame
421 * we allocate as much space on the stack all parameters need, by
422 * moving the stack pointer along the stack's direction.
424 * Note: we also have to do this for stack_size == 0, because we may have
425 * to adjust stack alignment for the call.
427 curr_sp = be_new_IncSP(sp, bl, curr_sp, stack_size, 1);
429 dbgi = get_irn_dbg_info(irn);
430 /* If there are some parameters which shall be passed on the stack. */
431 if (n_stack_params > 0) {
433 ir_node **in = ALLOCAN(ir_node*, n_stack_params+1);
436 curr_mem = get_Call_mem(irn);
437 in[n_in++] = curr_mem;
439 for (i = 0; i < n_stack_params; ++i) {
440 int p = stack_param_idx[i];
441 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
442 ir_node *param = get_Call_param(irn, p);
443 ir_node *addr = curr_sp;
445 ir_type *param_type = get_method_param_type(call_tp, p);
446 int param_size = get_type_size_bytes(param_type) + arg->space_after;
449 * If we wanted to build the arguments sequentially,
450 * the stack pointer for the next must be incremented,
451 * and the memory value propagated.
453 curr_ofs += arg->space_before;
454 curr_ofs = round_up2(curr_ofs, arg->alignment);
456 /* Make the expression to compute the argument's offset. */
458 ir_mode *constmode = mach_mode;
459 if (mode_is_reference(mach_mode)) {
462 addr = new_r_Const_long(irg, constmode, curr_ofs);
463 addr = new_r_Add(bl, curr_sp, addr, mach_mode);
466 /* Insert a store for primitive arguments. */
467 if (is_atomic_type(param_type)) {
468 ir_node *nomem = get_irg_no_mem(irg);
469 ir_node *mem_input = nomem;
470 ir_node *store = new_rd_Store(dbgi, bl, mem_input, addr, param, cons_none);
471 mem = new_r_Proj(store, mode_M, pn_Store_M);
473 /* Make a mem copy for compound arguments. */
476 assert(mode_is_reference(get_irn_mode(param)));
477 copy = new_rd_CopyB(dbgi, bl, curr_mem, addr, param, param_type);
478 mem = new_r_Proj(copy, mode_M, pn_CopyB_M);
481 curr_ofs += param_size;
486 /* 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);
494 /* Put caller save into the destroyed set and state registers in the states
496 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
498 const arch_register_class_t *cls = &arch_env->register_classes[i];
499 for (j = 0; j < cls->n_regs; ++j) {
500 const arch_register_t *reg = arch_register_for_index(cls, j);
502 /* even if destroyed all is specified, neither SP nor FP are
503 * destroyed (else bad things will happen) */
504 if (reg == arch_env->sp || reg == arch_env->bp)
507 if (reg->type & arch_register_type_state) {
508 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
509 ARR_APP1(const arch_register_t*, states, reg);
510 /* we're already in the destroyed set so no need for further
514 if (arch_register_is_caller_save(arch_env, reg)) {
515 if (!(reg->type & arch_register_type_ignore)) {
516 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
522 /* search the largest result proj number */
523 res_projs = ALLOCANZ(ir_node*, n_res);
525 foreach_out_edge(irn, edge) {
526 ir_node *irn = get_edge_src_irn(edge);
528 if (!is_Proj(irn) || get_Proj_proj(irn) != pn_Call_T_result)
531 foreach_out_edge(irn, res_edge) {
533 ir_node *res = get_edge_src_irn(res_edge);
535 assert(is_Proj(res));
537 proj = get_Proj_proj(res);
538 assert(proj < n_res);
539 assert(res_projs[proj] == NULL);
540 res_projs[proj] = res;
546 /** TODO: this is not correct for cases where return values are passed
547 * on the stack, but no known ABI does this currently...
549 n_reg_results = n_res;
552 in = ALLOCAN(ir_node*, n_reg_params + ARR_LEN(states));
554 /* make the back end call node and set its register requirements. */
555 for (i = 0; i < n_reg_params; ++i) {
556 in[n_ins++] = get_Call_param(irn, reg_param_idxs[i]);
559 /* add state registers ins */
560 for (s = 0; s < ARR_LEN(states); ++s) {
561 const arch_register_t *reg = states[s];
562 const arch_register_class_t *cls = reg->reg_class;
563 ir_node *regnode = new_r_Unknown(irg, cls->mode);
564 in[n_ins++] = regnode;
566 assert(n_ins == (int) (n_reg_params + ARR_LEN(states)));
568 /* ins collected, build the call */
569 throws_exception = ir_throws_exception(irn);
570 if (env->call->flags.bits.call_has_imm && is_SymConst(call_ptr)) {
572 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, curr_sp,
573 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
574 n_ins, in, get_Call_type(irn));
575 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
578 low_call = be_new_Call(dbgi, irg, bl, curr_mem, curr_sp, call_ptr,
579 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
580 n_ins, in, get_Call_type(irn));
582 ir_set_throws_exception(low_call, throws_exception);
583 be_Call_set_pop(low_call, call->pop);
585 /* put the call into the list of all calls for later processing */
586 ARR_APP1(ir_node *, env->calls, low_call);
588 /* create new stack pointer */
589 curr_sp = new_r_Proj(low_call, get_irn_mode(curr_sp), pn_be_Call_sp);
590 be_set_constr_single_reg_out(low_call, pn_be_Call_sp, sp,
591 arch_register_req_type_ignore | arch_register_req_type_produces_sp);
592 arch_set_irn_register(curr_sp, sp);
594 /* now handle results */
595 for (i = 0; i < n_res; ++i) {
596 ir_node *proj = res_projs[i];
597 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
599 /* returns values on stack not supported yet */
603 shift the proj number to the right, since we will drop the
604 unspeakable Proj_T from the Call. Therefore, all real argument
605 Proj numbers must be increased by pn_be_Call_first_res
607 long pn = i + pn_be_Call_first_res;
610 ir_type *res_type = get_method_res_type(call_tp, i);
611 ir_mode *mode = get_type_mode(res_type);
612 proj = new_r_Proj(low_call, mode, pn);
615 set_Proj_pred(proj, low_call);
616 set_Proj_proj(proj, pn);
620 /* remove register from destroyed regs */
622 size_t n = ARR_LEN(destroyed_regs);
623 for (j = 0; j < n; ++j) {
624 if (destroyed_regs[j] == arg->reg) {
625 destroyed_regs[j] = destroyed_regs[n-1];
626 ARR_SHRINKLEN(destroyed_regs,n-1);
634 Set the register class of the call address to
635 the backend provided class (default: stack pointer class)
637 be_node_set_reg_class_in(low_call, n_be_Call_ptr, call->cls_addr);
639 DBG((dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
641 /* Set the register classes and constraints of the Call parameters. */
642 for (i = 0; i < n_reg_params; ++i) {
643 int index = reg_param_idxs[i];
644 be_abi_call_arg_t *arg = get_call_arg(call, 0, index, 0);
645 assert(arg->reg != NULL);
647 be_set_constr_single_reg_in(low_call, n_be_Call_first_arg + i,
648 arg->reg, arch_register_req_type_none);
651 /* Set the register constraints of the results. */
652 for (i = 0; i < n_res; ++i) {
653 ir_node *proj = res_projs[i];
654 const be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
655 int pn = get_Proj_proj(proj);
658 be_set_constr_single_reg_out(low_call, pn, arg->reg,
659 arch_register_req_type_none);
660 arch_set_irn_register(proj, arg->reg);
662 exchange(irn, low_call);
664 /* kill the ProjT node */
665 if (res_proj != NULL) {
669 /* Make additional projs for the caller save registers
670 and the Keep node which keeps them alive. */
676 int curr_res_proj = pn_be_Call_first_res + n_reg_results;
679 n_ins = ARR_LEN(destroyed_regs) + n_reg_results + 1;
680 in = ALLOCAN(ir_node *, n_ins);
682 /* also keep the stack pointer */
683 set_irn_link(curr_sp, (void*) sp);
686 for (d = 0; d < ARR_LEN(destroyed_regs); ++d) {
687 const arch_register_t *reg = destroyed_regs[d];
688 ir_node *proj = new_r_Proj(low_call, reg->reg_class->mode, curr_res_proj);
690 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
691 be_set_constr_single_reg_out(low_call, curr_res_proj, reg,
692 arch_register_req_type_none);
693 arch_set_irn_register(proj, reg);
695 set_irn_link(proj, (void*) reg);
700 for (i = 0; i < n_reg_results; ++i) {
701 ir_node *proj = res_projs[i];
702 const arch_register_t *reg = arch_get_irn_register(proj);
703 set_irn_link(proj, (void*) reg);
708 /* create the Keep for the caller save registers */
709 keep = be_new_Keep(bl, n, in);
710 for (i = 0; i < n; ++i) {
711 const arch_register_t *reg = (const arch_register_t*)get_irn_link(in[i]);
712 be_node_set_reg_class_in(keep, i, arch_register_get_class(reg));
716 /* Clean up the stack. */
717 assert(stack_size >= call->pop);
718 stack_size -= call->pop;
720 if (stack_size > 0) {
721 ir_node *mem_proj = NULL;
723 foreach_out_edge(low_call, edge) {
724 ir_node *irn = get_edge_src_irn(edge);
725 if (is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
732 mem_proj = new_r_Proj(low_call, mode_M, pn_be_Call_M);
733 keep_alive(mem_proj);
736 /* Clean up the stack frame or revert alignment fixes if we allocated it */
737 curr_sp = be_new_IncSP(sp, bl, curr_sp, -stack_size, 0);
739 be_abi_call_free(call);
742 DEL_ARR_F(destroyed_regs);
748 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
750 * @param alignment the minimum stack alignment
751 * @param size the node containing the non-aligned size
752 * @param block the block where new nodes are allocated on
753 * @param dbg debug info for new nodes
755 * @return a node representing the aligned size
757 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size,
758 ir_node *block, dbg_info *dbg)
760 if (stack_alignment > 1) {
766 assert(is_po2(stack_alignment));
768 mode = get_irn_mode(size);
769 tv = new_tarval_from_long(stack_alignment-1, mode);
770 irg = get_Block_irg(block);
771 mask = new_r_Const(irg, tv);
772 size = new_rd_Add(dbg, block, size, mask, mode);
774 tv = new_tarval_from_long(-(long)stack_alignment, mode);
775 mask = new_r_Const(irg, tv);
776 size = new_rd_And(dbg, block, size, mask, mode);
782 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
784 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp)
786 ir_node *block = get_nodes_block(alloc);
787 ir_graph *irg = get_Block_irg(block);
788 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
789 ir_node *alloc_mem = NULL;
790 ir_node *alloc_res = NULL;
791 ir_type *type = get_Alloc_type(alloc);
798 unsigned stack_alignment;
800 /* all non-stack Alloc nodes should already be lowered before the backend */
801 assert(get_Alloc_where(alloc) == stack_alloc);
803 foreach_out_edge(alloc, edge) {
804 ir_node *irn = get_edge_src_irn(edge);
806 assert(is_Proj(irn));
807 switch (get_Proj_proj(irn)) {
819 /* Beware: currently Alloc nodes without a result might happen,
820 only escape analysis kills them and this phase runs only for object
821 oriented source. We kill the Alloc here. */
822 if (alloc_res == NULL && alloc_mem) {
823 exchange(alloc_mem, get_Alloc_mem(alloc));
827 dbg = get_irn_dbg_info(alloc);
828 count = get_Alloc_count(alloc);
830 /* we might need to multiply the count with the element size */
831 if (!is_unknown_type(type) && get_type_size_bytes(type) != 1) {
832 ir_mode *mode = get_irn_mode(count);
833 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type),
835 ir_node *cnst = new_rd_Const(dbg, irg, tv);
836 size = new_rd_Mul(dbg, block, count, cnst, mode);
841 /* The stack pointer will be modified in an unknown manner.
842 We cannot omit it. */
843 env->call->flags.bits.try_omit_fp = 0;
845 stack_alignment = 1 << arch_env->stack_alignment;
846 size = adjust_alloc_size(stack_alignment, size, block, dbg);
847 new_alloc = be_new_AddSP(arch_env->sp, block, curr_sp, size);
848 set_irn_dbg_info(new_alloc, dbg);
850 if (alloc_mem != NULL) {
854 addsp_mem = new_r_Proj(new_alloc, mode_M, pn_be_AddSP_M);
856 /* We need to sync the output mem of the AddSP with the input mem
857 edge into the alloc node. */
858 ins[0] = get_Alloc_mem(alloc);
860 sync = new_r_Sync(block, 2, ins);
862 exchange(alloc_mem, sync);
865 exchange(alloc, new_alloc);
867 /* fix projnum of alloca res */
868 set_Proj_proj(alloc_res, pn_be_AddSP_res);
870 curr_sp = new_r_Proj(new_alloc, get_irn_mode(curr_sp), pn_be_AddSP_sp);
877 * The Free is transformed into a back end free node and connected to the stack nodes.
879 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
881 ir_node *block = get_nodes_block(free);
882 ir_graph *irg = get_irn_irg(free);
883 ir_type *type = get_Free_type(free);
884 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
885 ir_mode *sp_mode = arch_env->sp->reg_class->mode;
886 dbg_info *dbg = get_irn_dbg_info(free);
887 ir_node *subsp, *mem, *res, *size, *sync;
889 unsigned stack_alignment;
891 /* all non-stack-alloc Free nodes should already be lowered before the
893 assert(get_Free_where(free) == stack_alloc);
895 /* we might need to multiply the size with the element size */
896 if (!is_unknown_type(type) && get_type_size_bytes(type) != 1) {
897 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
898 ir_node *cnst = new_rd_Const(dbg, irg, tv);
899 ir_node *mul = new_rd_Mul(dbg, block, get_Free_count(free),
903 size = get_Free_count(free);
906 stack_alignment = 1 << arch_env->stack_alignment;
907 size = adjust_alloc_size(stack_alignment, size, block, dbg);
909 /* The stack pointer will be modified in an unknown manner.
910 We cannot omit it. */
911 env->call->flags.bits.try_omit_fp = 0;
912 subsp = be_new_SubSP(arch_env->sp, block, curr_sp, size);
913 set_irn_dbg_info(subsp, dbg);
915 mem = new_r_Proj(subsp, mode_M, pn_be_SubSP_M);
916 res = new_r_Proj(subsp, sp_mode, pn_be_SubSP_sp);
918 /* we need to sync the memory */
919 in[0] = get_Free_mem(free);
921 sync = new_r_Sync(block, 2, in);
923 /* and make the AddSP dependent on the former memory */
924 add_irn_dep(subsp, get_Free_mem(free));
927 exchange(free, sync);
934 * Check if a node is somehow data dependent on another one.
935 * both nodes must be in the same basic block.
936 * @param n1 The first node.
937 * @param n2 The second node.
938 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
940 static int dependent_on(ir_node *n1, ir_node *n2)
942 assert(get_nodes_block(n1) == get_nodes_block(n2));
944 return heights_reachable_in_block(ir_heights, n1, n2);
947 static int cmp_call_dependency(const void *c1, const void *c2)
949 ir_node *n1 = *(ir_node **) c1;
950 ir_node *n2 = *(ir_node **) c2;
954 Classical qsort() comparison function behavior:
955 0 if both elements are equal
956 1 if second is "smaller" that first
957 -1 if first is "smaller" that second
959 if (dependent_on(n1, n2))
962 if (dependent_on(n2, n1))
965 /* The nodes have no depth order, but we need a total order because qsort()
968 * Additionally, we need to respect transitive dependencies. Consider a
969 * Call a depending on Call b and an independent Call c.
970 * We MUST NOT order c > a and b > c. */
971 h1 = get_irn_height(ir_heights, n1);
972 h2 = get_irn_height(ir_heights, n2);
973 if (h1 < h2) return -1;
974 if (h1 > h2) return 1;
975 /* Same height, so use a random (but stable) order */
976 return get_irn_idx(n1) - get_irn_idx(n2);
980 * Walker: links all Call/Alloc/Free nodes to the Block they are contained.
982 static void link_ops_in_block_walker(ir_node *irn, void *data)
984 be_abi_irg_t *env = (be_abi_irg_t*)data;
985 unsigned code = get_irn_opcode(irn);
987 if (code == iro_Call ||
988 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
989 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
990 ir_node *bl = get_nodes_block(irn);
991 void *save = get_irn_link(bl);
993 set_irn_link(irn, save);
994 set_irn_link(bl, irn);
997 if (code == iro_Builtin && get_Builtin_kind(irn) == ir_bk_return_address) {
998 ir_node *param = get_Builtin_param(irn, 0);
999 ir_tarval *tv = get_Const_tarval(param);
1000 unsigned long value = get_tarval_long(tv);
1001 /* use ebp, so the climbframe algo works... */
1003 env->call->flags.bits.try_omit_fp = 0;
1010 * Process all Call/Alloc/Free nodes inside a basic block.
1011 * Note that the link field of the block must contain a linked list of all
1012 * nodes inside the Block. We first order this list according to data dependency
1013 * and that connect the nodes together.
1015 static void process_ops_in_block(ir_node *bl, void *data)
1017 be_abi_irg_t *env = (be_abi_irg_t*)data;
1018 ir_node *curr_sp = env->init_sp;
1025 for (irn = (ir_node*)get_irn_link(bl); irn != NULL;
1026 irn = (ir_node*)get_irn_link(irn)) {
1030 nodes = ALLOCAN(ir_node*, n_nodes);
1031 for (irn = (ir_node*)get_irn_link(bl), n = 0; irn != NULL;
1032 irn = (ir_node*)get_irn_link(irn), ++n) {
1036 /* If there were call nodes in the block. */
1041 /* order the call nodes according to data dependency */
1042 qsort(nodes, n_nodes, sizeof(nodes[0]), cmp_call_dependency);
1044 for (i = n_nodes - 1; i >= 0; --i) {
1045 ir_node *irn = nodes[i];
1047 DBG((dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1048 switch (get_irn_opcode(irn)) {
1051 /* The stack pointer will be modified due to a call. */
1052 env->call->flags.bits.try_omit_fp = 0;
1054 curr_sp = adjust_call(env, irn, curr_sp);
1057 if (get_Alloc_where(irn) == stack_alloc)
1058 curr_sp = adjust_alloc(env, irn, curr_sp);
1061 if (get_Free_where(irn) == stack_alloc)
1062 curr_sp = adjust_free(env, irn, curr_sp);
1065 panic("invalid call");
1069 /* Keep the last stack state in the block by tying it to Keep node,
1070 * the proj from calls is already kept */
1071 if (curr_sp != env->init_sp &&
1072 !(is_Proj(curr_sp) && be_is_Call(get_Proj_pred(curr_sp)))) {
1074 keep = be_new_Keep(bl, 1, nodes);
1075 pmap_insert(env->keep_map, bl, keep);
1079 set_irn_link(bl, curr_sp);
1083 * Adjust all call nodes in the graph to the ABI conventions.
1085 static void process_calls(ir_graph *irg)
1087 be_abi_irg_t *abi = be_get_irg_abi(irg);
1089 irg_walk_graph(irg, firm_clear_link, link_ops_in_block_walker, abi);
1091 ir_heights = heights_new(irg);
1092 irg_block_walk_graph(irg, NULL, process_ops_in_block, abi);
1093 heights_free(ir_heights);
1097 * Computes the stack argument layout type.
1098 * Changes a possibly allocated value param type by moving
1099 * entities to the stack layout type.
1101 * @param call the current call ABI
1102 * @param method_type the method type
1104 * @return the stack argument layout type
1106 static ir_type *compute_arg_type(ir_graph *irg, be_abi_call_t *call,
1107 ir_type *method_type)
1109 struct obstack *obst = be_get_be_obst(irg);
1110 ir_type *frame_type = get_irg_frame_type(irg);
1111 size_t n_params = get_method_n_params(method_type);
1112 size_t n_frame_members = get_compound_n_members(frame_type);
1113 ir_entity *va_start_entity = NULL;
1119 ir_entity **map = OALLOCNZ(obst, ir_entity*, n_params);
1120 res = new_type_struct(new_id_from_chars("arg_type", 8));
1122 /* collect existing entities for value_param_types */
1123 for (f = n_frame_members; f > 0; ) {
1124 ir_entity *entity = get_compound_member(frame_type, --f);
1127 set_entity_link(entity, NULL);
1128 if (!is_parameter_entity(entity))
1130 num = get_entity_parameter_number(entity);
1131 if (num == IR_VA_START_PARAMETER_NUMBER) {
1132 /* move entity to new arg_type */
1133 set_entity_owner(entity, res);
1134 va_start_entity = entity;
1137 assert(num < n_params);
1138 if (map[num] != NULL)
1139 panic("multiple entities for parameter %u in %+F found", f, irg);
1141 if (num != n_params && !get_call_arg(call, 0, num, 1)->on_stack) {
1142 /* don't move this entity */
1147 /* move entity to new arg_type */
1148 set_entity_owner(entity, res);
1151 for (i = 0; i < n_params; ++i) {
1152 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1153 ir_type *param_type = get_method_param_type(method_type, i);
1156 if (!arg->on_stack) {
1160 if (entity == NULL) {
1161 /* create a new entity */
1162 entity = new_parameter_entity(res, i, param_type);
1164 ofs += arg->space_before;
1165 ofs = round_up2(ofs, arg->alignment);
1166 set_entity_offset(entity, ofs);
1167 ofs += arg->space_after;
1168 ofs += get_type_size_bytes(param_type);
1169 arg->stack_ent = entity;
1171 if (va_start_entity != NULL) {
1172 set_entity_offset(va_start_entity, ofs);
1174 set_type_size_bytes(res, ofs);
1175 set_type_state(res, layout_fixed);
1181 const arch_register_t *reg;
1185 static int cmp_regs(const void *a, const void *b)
1187 const reg_node_map_t *p = (const reg_node_map_t*)a;
1188 const reg_node_map_t *q = (const reg_node_map_t*)b;
1190 if (p->reg->reg_class == q->reg->reg_class)
1191 return p->reg->index - q->reg->index;
1193 return p->reg->reg_class < q->reg->reg_class ? -1 : +1;
1196 static void reg_map_to_arr(reg_node_map_t *res, pmap *reg_map)
1199 size_t n = pmap_count(reg_map);
1202 foreach_pmap(reg_map, ent) {
1203 res[i].reg = (const arch_register_t*)ent->key;
1204 res[i].irn = (ir_node*)ent->value;
1208 qsort(res, n, sizeof(res[0]), cmp_regs);
1212 * Creates a be_Return for a Return node.
1214 * @param @env the abi environment
1215 * @param irn the Return node or NULL if there was none
1216 * @param bl the block where the be_Retun should be placed
1217 * @param mem the current memory
1218 * @param n_res number of return results
1220 static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl,
1221 ir_node *mem, int n_res)
1223 be_abi_call_t *call = env->call;
1224 ir_graph *irg = get_Block_irg(bl);
1225 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1227 pmap *reg_map = pmap_create();
1228 ir_node *keep = pmap_get(ir_node, env->keep_map, bl);
1235 const arch_register_t **regs;
1239 get the valid stack node in this block.
1240 If we had a call in that block there is a Keep constructed by process_calls()
1241 which points to the last stack modification in that block. we'll use
1242 it then. Else we use the stack from the start block and let
1243 the ssa construction fix the usage.
1245 stack = be_abi_reg_map_get(env->regs, arch_env->sp);
1247 stack = get_irn_n(keep, 0);
1249 remove_End_keepalive(get_irg_end(irg), keep);
1252 /* Insert results for Return into the register map. */
1253 for (i = 0; i < n_res; ++i) {
1254 ir_node *res = get_Return_res(irn, i);
1255 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1256 assert(arg->in_reg && "return value must be passed in register");
1257 pmap_insert(reg_map, (void *) arg->reg, res);
1260 /* Add uses of the callee save registers. */
1261 foreach_pmap(env->regs, ent) {
1262 const arch_register_t *reg = (const arch_register_t*)ent->key;
1263 if ((reg->type & arch_register_type_ignore) || arch_register_is_callee_save(arch_env, reg))
1264 pmap_insert(reg_map, ent->key, ent->value);
1267 be_abi_reg_map_set(reg_map, arch_env->sp, stack);
1270 Maximum size of the in array for Return nodes is
1271 return args + callee save/ignore registers + memory + stack pointer
1273 in_max = pmap_count(reg_map) + n_res + 2;
1275 in = ALLOCAN(ir_node*, in_max);
1276 regs = ALLOCAN(arch_register_t const*, in_max);
1279 in[1] = be_abi_reg_map_get(reg_map, arch_env->sp);
1281 regs[1] = arch_env->sp;
1284 /* clear SP entry, since it has already been grown. */
1285 pmap_insert(reg_map, (void *) arch_env->sp, NULL);
1286 for (i = 0; i < n_res; ++i) {
1287 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1289 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1290 regs[n++] = arg->reg;
1292 /* Clear the map entry to mark the register as processed. */
1293 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1296 /* grow the rest of the stuff. */
1297 foreach_pmap(reg_map, ent) {
1299 in[n] = (ir_node*)ent->value;
1300 regs[n++] = (const arch_register_t*)ent->key;
1304 /* The in array for the new back end return is now ready. */
1306 dbgi = get_irn_dbg_info(irn);
1310 /* we have to pop the shadow parameter in in case of struct returns */
1312 ret = be_new_Return(dbgi, irg, bl, n_res, pop, n, in);
1314 /* Set the register classes of the return's parameter accordingly. */
1315 for (i = 0; i < n; ++i) {
1316 if (regs[i] == NULL)
1319 be_set_constr_single_reg_in(ret, i, regs[i], arch_register_req_type_none);
1322 /* Free the space of the Epilog's in array and the register <-> proj map. */
1323 pmap_destroy(reg_map);
1328 typedef struct lower_frame_sels_env_t {
1329 ir_node *frame; /**< the current frame */
1330 const arch_register_class_t *sp_class; /**< register class of the stack pointer */
1331 const arch_register_class_t *link_class; /**< register class of the link pointer */
1332 ir_type *frame_tp; /**< the frame type */
1333 int static_link_pos; /**< argument number of the hidden static link */
1334 } lower_frame_sels_env_t;
1337 * Walker: Replaces Sels of frame type and
1338 * value param type entities by FrameAddress.
1339 * Links all used entities.
1341 static void lower_frame_sels_walker(ir_node *irn, void *data)
1343 lower_frame_sels_env_t *ctx = (lower_frame_sels_env_t*)data;
1346 ir_node *ptr = get_Sel_ptr(irn);
1348 if (ptr == ctx->frame) {
1349 ir_entity *ent = get_Sel_entity(irn);
1350 ir_node *bl = get_nodes_block(irn);
1353 nw = be_new_FrameAddr(ctx->sp_class, bl, ctx->frame, ent);
1360 * The start block has no jump, instead it has an initial exec Proj.
1361 * The backend wants to handle all blocks the same way, so we replace
1362 * the out cfg edge with a real jump.
1364 static void fix_start_block(ir_graph *irg)
1366 ir_node *initial_X = get_irg_initial_exec(irg);
1367 ir_node *start_block = get_irg_start_block(irg);
1368 ir_node *jmp = new_r_Jmp(start_block);
1370 assert(is_Proj(initial_X));
1371 exchange(initial_X, jmp);
1372 set_irg_initial_exec(irg, new_r_Bad(irg, mode_X));
1374 /* merge start block with successor if possible */
1376 foreach_out_edge(jmp, edge) {
1377 ir_node *succ = get_edge_src_irn(edge);
1378 if (!is_Block(succ))
1381 if (get_irn_arity(succ) == 1) {
1382 exchange(succ, start_block);
1390 * Modify the irg itself and the frame type.
1392 static void modify_irg(ir_graph *irg)
1394 be_abi_irg_t *env = be_get_irg_abi(irg);
1395 be_abi_call_t *call = env->call;
1396 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1397 const arch_register_t *sp = arch_env->sp;
1398 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1399 be_irg_t *birg = be_birg_from_irg(irg);
1400 struct obstack *obst = be_get_be_obst(irg);
1401 be_stack_layout_t *stack_layout = be_get_irg_stack_layout(irg);
1404 ir_node *new_mem_proj;
1410 unsigned frame_size;
1413 const arch_register_t *fp_reg;
1414 ir_node *frame_pointer;
1418 ir_type *arg_type, *bet_type;
1419 lower_frame_sels_env_t ctx;
1421 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1423 old_mem = get_irg_initial_mem(irg);
1425 irp_reserve_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1427 arg_type = compute_arg_type(irg, call, method_type);
1429 /* Convert the Sel nodes in the irg to frame addr nodes: */
1430 ctx.frame = get_irg_frame(irg);
1431 ctx.sp_class = arch_env->sp->reg_class;
1432 ctx.link_class = arch_env->link_class;
1433 ctx.frame_tp = get_irg_frame_type(irg);
1435 /* layout the stackframe now */
1436 if (get_type_state(ctx.frame_tp) == layout_undefined) {
1437 default_layout_compound_type(ctx.frame_tp);
1440 /* align stackframe to 4 byte */
1441 frame_size = get_type_size_bytes(ctx.frame_tp);
1442 if (frame_size % 4 != 0) {
1443 set_type_size_bytes(ctx.frame_tp, frame_size + 4 - (frame_size % 4));
1446 env->regs = pmap_create();
1448 n_params = get_method_n_params(method_type);
1449 args = OALLOCNZ(obst, ir_node*, n_params);
1451 be_add_parameter_entity_stores(irg);
1453 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1455 irp_free_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1457 /* Fill the argument vector */
1458 arg_tuple = get_irg_args(irg);
1459 foreach_out_edge(arg_tuple, edge) {
1460 ir_node *irn = get_edge_src_irn(edge);
1461 if (! is_Anchor(irn)) {
1462 int nr = get_Proj_proj(irn);
1464 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1468 stack_layout->sp_relative = call->flags.bits.try_omit_fp;
1469 bet_type = call->cb->get_between_type(irg);
1470 stack_frame_init(stack_layout, arg_type, bet_type,
1471 get_irg_frame_type(irg));
1473 /* Count the register params and add them to the number of Projs for the RegParams node */
1474 for (i = 0; i < n_params; ++i) {
1475 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1476 if (arg->in_reg && args[i]) {
1477 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1478 assert(i == get_Proj_proj(args[i]));
1480 /* For now, associate the register with the old Proj from Start representing that argument. */
1481 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1482 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1486 /* Collect all callee-save registers */
1487 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
1488 const arch_register_class_t *cls = &arch_env->register_classes[i];
1489 for (j = 0; j < cls->n_regs; ++j) {
1490 const arch_register_t *reg = &cls->regs[j];
1491 if ((reg->type & arch_register_type_state) || arch_register_is_callee_save(arch_env, reg)) {
1492 pmap_insert(env->regs, (void *) reg, NULL);
1497 fp_reg = call->flags.bits.try_omit_fp ? arch_env->sp : arch_env->bp;
1498 rbitset_clear(birg->allocatable_regs, fp_reg->global_index);
1500 /* handle start block here (place a jump in the block) */
1501 fix_start_block(irg);
1503 pmap_insert(env->regs, (void *) sp, NULL);
1504 pmap_insert(env->regs, (void *) arch_env->bp, NULL);
1505 start_bl = get_irg_start_block(irg);
1506 env->start = be_new_Start(NULL, start_bl, pmap_count(env->regs) + 1);
1507 set_irg_start(irg, env->start);
1510 * make proj nodes for the callee save registers.
1511 * memorize them, since Return nodes get those as inputs.
1513 * Note, that if a register corresponds to an argument, the regs map
1514 * contains the old Proj from start for that argument.
1516 rm = ALLOCAN(reg_node_map_t, pmap_count(env->regs));
1517 reg_map_to_arr(rm, env->regs);
1518 for (i = 0, n = pmap_count(env->regs); i < n; ++i) {
1519 const arch_register_t *reg = rm[i].reg;
1520 ir_mode *mode = reg->reg_class->mode;
1522 arch_register_req_type_t add_type = arch_register_req_type_none;
1526 add_type |= arch_register_req_type_produces_sp;
1527 if (!rbitset_is_set(birg->allocatable_regs, reg->global_index)) {
1528 add_type |= arch_register_req_type_ignore;
1532 proj = new_r_Proj(env->start, mode, nr + 1);
1533 pmap_insert(env->regs, (void *) reg, proj);
1534 be_set_constr_single_reg_out(env->start, nr + 1, reg, add_type);
1535 arch_set_irn_register(proj, reg);
1537 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1540 /* create a new initial memory proj */
1541 assert(is_Proj(old_mem));
1542 arch_set_irn_register_req_out(env->start, 0, arch_no_register_req);
1543 new_mem_proj = new_r_Proj(env->start, mode_M, 0);
1545 set_irg_initial_mem(irg, mem);
1547 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1549 /* set new frame_pointer */
1550 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1551 set_irg_frame(irg, frame_pointer);
1553 /* rewire old mem users to new mem */
1554 exchange(old_mem, mem);
1556 /* keep the mem (for functions with an endless loop = no return) */
1559 set_irg_initial_mem(irg, mem);
1561 /* Now, introduce stack param nodes for all parameters passed on the stack */
1562 for (i = 0; i < n_params; ++i) {
1563 ir_node *arg_proj = args[i];
1564 ir_node *repl = NULL;
1566 if (arg_proj != NULL) {
1567 be_abi_call_arg_t *arg;
1568 ir_type *param_type;
1569 int nr = get_Proj_proj(arg_proj);
1572 nr = MIN(nr, n_params);
1573 arg = get_call_arg(call, 0, nr, 1);
1574 param_type = get_method_param_type(method_type, nr);
1577 repl = pmap_get(ir_node, env->regs, arg->reg);
1578 } else if (arg->on_stack) {
1579 ir_node *addr = be_new_FrameAddr(sp->reg_class, start_bl, frame_pointer, arg->stack_ent);
1581 /* For atomic parameters which are actually used, we create a Load node. */
1582 if (is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1583 ir_mode *mode = get_type_mode(param_type);
1584 ir_mode *load_mode = arg->load_mode;
1585 ir_node *nomem = get_irg_no_mem(irg);
1587 ir_node *load = new_r_Load(start_bl, nomem, addr, load_mode, cons_floats);
1588 repl = new_r_Proj(load, load_mode, pn_Load_res);
1590 if (mode != load_mode) {
1591 repl = new_r_Conv(start_bl, repl, mode);
1594 /* The stack parameter is not primitive (it is a struct or array),
1595 * we thus will create a node representing the parameter's address
1601 assert(repl != NULL);
1603 /* Beware: the mode of the register parameters is always the mode of the register class
1604 which may be wrong. Add Conv's then. */
1605 mode = get_irn_mode(args[i]);
1606 if (mode != get_irn_mode(repl)) {
1607 repl = new_r_Conv(get_nodes_block(repl), repl, mode);
1609 exchange(args[i], repl);
1613 /* the arg proj is not needed anymore now and should be only used by the anchor */
1614 assert(get_irn_n_edges(arg_tuple) == 1);
1615 kill_node(arg_tuple);
1616 set_irg_args(irg, new_r_Bad(irg, mode_T));
1618 /* All Return nodes hang on the End node, so look for them there. */
1619 end = get_irg_end_block(irg);
1620 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1621 ir_node *irn = get_Block_cfgpred(end, i);
1623 if (is_Return(irn)) {
1624 ir_node *blk = get_nodes_block(irn);
1625 ir_node *mem = get_Return_mem(irn);
1626 ir_node *ret = create_be_return(env, irn, blk, mem, get_Return_n_ress(irn));
1631 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1632 the code is dead and will never be executed. */
1635 /** Fix the state inputs of calls that still hang on unknowns */
1636 static void fix_call_state_inputs(ir_graph *irg)
1638 be_abi_irg_t *env = be_get_irg_abi(irg);
1639 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1641 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1643 /* Collect caller save registers */
1644 n = arch_env->n_register_classes;
1645 for (i = 0; i < n; ++i) {
1647 const arch_register_class_t *cls = &arch_env->register_classes[i];
1648 for (j = 0; j < cls->n_regs; ++j) {
1649 const arch_register_t *reg = arch_register_for_index(cls, j);
1650 if (reg->type & arch_register_type_state) {
1651 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1656 n = ARR_LEN(env->calls);
1657 n_states = ARR_LEN(stateregs);
1658 for (i = 0; i < n; ++i) {
1660 ir_node *call = env->calls[i];
1662 arity = get_irn_arity(call);
1664 /* the state reg inputs are the last n inputs of the calls */
1665 for (s = 0; s < n_states; ++s) {
1666 int inp = arity - n_states + s;
1667 const arch_register_t *reg = stateregs[s];
1668 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
1670 set_irn_n(call, inp, regnode);
1674 DEL_ARR_F(stateregs);
1678 * Create a trampoline entity for the given method.
1680 static ir_entity *create_trampoline(be_main_env_t *be, ir_entity *method)
1682 ir_type *type = get_entity_type(method);
1683 ident *old_id = get_entity_ld_ident(method);
1684 ident *id = id_mangle3("", old_id, "$stub");
1685 ir_type *parent = be->pic_trampolines_type;
1686 ir_entity *ent = new_entity(parent, old_id, type);
1687 set_entity_ld_ident(ent, id);
1688 set_entity_visibility(ent, ir_visibility_private);
1694 * Returns the trampoline entity for the given method.
1696 static ir_entity *get_trampoline(be_main_env_t *env, ir_entity *method)
1698 ir_entity *result = pmap_get(ir_entity, env->ent_trampoline_map, method);
1699 if (result == NULL) {
1700 result = create_trampoline(env, method);
1701 pmap_insert(env->ent_trampoline_map, method, result);
1707 static ir_entity *create_pic_symbol(be_main_env_t *be, ir_entity *entity)
1709 ident *old_id = get_entity_ld_ident(entity);
1710 ident *id = id_mangle3("", old_id, "$non_lazy_ptr");
1711 ir_type *e_type = get_entity_type(entity);
1712 ir_type *type = new_type_pointer(e_type);
1713 ir_type *parent = be->pic_symbols_type;
1714 ir_entity *ent = new_entity(parent, old_id, type);
1715 set_entity_ld_ident(ent, id);
1716 set_entity_visibility(ent, ir_visibility_private);
1721 static ir_entity *get_pic_symbol(be_main_env_t *env, ir_entity *entity)
1723 ir_entity *result = pmap_get(ir_entity, env->ent_pic_symbol_map, entity);
1724 if (result == NULL) {
1725 result = create_pic_symbol(env, entity);
1726 pmap_insert(env->ent_pic_symbol_map, entity, result);
1735 * Returns non-zero if a given entity can be accessed using a relative address.
1737 static int can_address_relative(ir_entity *entity)
1739 return entity_has_definition(entity) && !(get_entity_linkage(entity) & IR_LINKAGE_MERGE);
1742 static ir_node *get_pic_base(ir_graph *irg)
1744 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1745 if (arch_env->impl->get_pic_base == NULL)
1747 return arch_env->impl->get_pic_base(irg);
1750 /** patches SymConsts to work in position independent code */
1751 static void fix_pic_symconsts(ir_node *node, void *data)
1753 ir_graph *irg = get_irn_irg(node);
1754 be_main_env_t *be = be_get_irg_main_env(irg);
1764 arity = get_irn_arity(node);
1765 for (i = 0; i < arity; ++i) {
1767 ir_node *pred = get_irn_n(node, i);
1769 ir_entity *pic_symbol;
1770 ir_node *pic_symconst;
1772 if (!is_SymConst(pred))
1775 entity = get_SymConst_entity(pred);
1776 block = get_nodes_block(pred);
1778 /* calls can jump to relative addresses, so we can directly jump to
1779 the (relatively) known call address or the trampoline */
1780 if (i == 1 && is_Call(node)) {
1781 ir_entity *trampoline;
1782 ir_node *trampoline_const;
1784 if (can_address_relative(entity))
1787 dbgi = get_irn_dbg_info(pred);
1788 trampoline = get_trampoline(be, entity);
1789 trampoline_const = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1791 set_irn_n(node, i, trampoline_const);
1795 /* everything else is accessed relative to EIP */
1796 mode = get_irn_mode(pred);
1797 pic_base = get_pic_base(irg);
1799 /* all ok now for locally constructed stuff */
1800 if (can_address_relative(entity)) {
1801 ir_node *add = new_r_Add(block, pic_base, pred, mode);
1803 /* make sure the walker doesn't visit this add again */
1804 mark_irn_visited(add);
1805 set_irn_n(node, i, add);
1809 /* get entry from pic symbol segment */
1810 dbgi = get_irn_dbg_info(pred);
1811 pic_symbol = get_pic_symbol(be, entity);
1812 pic_symconst = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1814 add = new_r_Add(block, pic_base, pic_symconst, mode);
1815 mark_irn_visited(add);
1817 /* we need an extra indirection for global data outside our current
1818 module. The loads are always safe and can therefore float
1819 and need no memory input */
1820 load = new_r_Load(block, get_irg_no_mem(irg), add, mode, cons_floats);
1821 load_res = new_r_Proj(load, mode, pn_Load_res);
1823 set_irn_n(node, i, load_res);
1827 void be_abi_introduce(ir_graph *irg)
1829 be_abi_irg_t *env = XMALLOCZ(be_abi_irg_t);
1830 ir_node *old_frame = get_irg_frame(irg);
1831 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1832 ir_entity *entity = get_irg_entity(irg);
1833 ir_type *method_type = get_entity_type(entity);
1834 be_irg_t *birg = be_birg_from_irg(irg);
1835 struct obstack *obst = &birg->obst;
1836 ir_node *dummy = new_r_Dummy(irg,
1837 arch_env->sp->reg_class->mode);
1840 /* determine allocatable registers */
1841 assert(birg->allocatable_regs == NULL);
1842 birg->allocatable_regs = rbitset_obstack_alloc(obst, arch_env->n_registers);
1843 for (r = 0; r < arch_env->n_registers; ++r) {
1844 const arch_register_t *reg = &arch_env->registers[r];
1845 if ( !(reg->type & arch_register_type_ignore)) {
1846 rbitset_set(birg->allocatable_regs, r);
1850 /* break here if backend provides a custom API.
1851 * Note: we shouldn't have to setup any be_abi_irg_t* stuff at all,
1852 * but need more cleanup to make this work
1854 be_set_irg_abi(irg, env);
1856 be_omit_fp = be_options.omit_fp;
1858 env->keep_map = pmap_create();
1859 env->call = be_abi_call_new(arch_env->sp->reg_class);
1860 arch_env_get_call_abi(arch_env, method_type, env->call);
1862 env->init_sp = dummy;
1863 env->calls = NEW_ARR_F(ir_node*, 0);
1867 if (be_options.pic) {
1868 irg_walk_graph(irg, fix_pic_symconsts, NULL, env);
1871 /* Lower all call nodes in the IRG. */
1874 /* Process the IRG */
1877 /* fix call inputs for state registers */
1878 fix_call_state_inputs(irg);
1880 /* We don't need the keep map anymore. */
1881 pmap_destroy(env->keep_map);
1882 env->keep_map = NULL;
1884 /* calls array is not needed anymore */
1885 DEL_ARR_F(env->calls);
1888 /* reroute the stack origin of the calls to the true stack origin. */
1889 exchange(dummy, env->init_sp);
1890 exchange(old_frame, get_irg_frame(irg));
1892 pmap_destroy(env->regs);
1896 void be_abi_free(ir_graph *irg)
1898 be_abi_irg_t *env = be_get_irg_abi(irg);
1900 if (env->call != NULL)
1901 be_abi_call_free(env->call);
1902 assert(env->regs == NULL);
1905 be_set_irg_abi(irg, NULL);
1908 void be_put_allocatable_regs(const ir_graph *irg,
1909 const arch_register_class_t *cls, bitset_t *bs)
1911 be_irg_t *birg = be_birg_from_irg(irg);
1912 unsigned *allocatable_regs = birg->allocatable_regs;
1915 assert(bitset_size(bs) == cls->n_regs);
1916 bitset_clear_all(bs);
1917 for (i = 0; i < cls->n_regs; ++i) {
1918 const arch_register_t *reg = &cls->regs[i];
1919 if (rbitset_is_set(allocatable_regs, reg->global_index))
1924 unsigned be_get_n_allocatable_regs(const ir_graph *irg,
1925 const arch_register_class_t *cls)
1927 bitset_t *bs = bitset_alloca(cls->n_regs);
1928 be_put_allocatable_regs(irg, cls, bs);
1929 return bitset_popcount(bs);
1932 void be_set_allocatable_regs(const ir_graph *irg,
1933 const arch_register_class_t *cls,
1934 unsigned *raw_bitset)
1936 be_irg_t *birg = be_birg_from_irg(irg);
1937 unsigned *allocatable_regs = birg->allocatable_regs;
1940 rbitset_clear_all(raw_bitset, cls->n_regs);
1941 for (i = 0; i < cls->n_regs; ++i) {
1942 const arch_register_t *reg = &cls->regs[i];
1943 if (rbitset_is_set(allocatable_regs, reg->global_index))
1944 rbitset_set(raw_bitset, i);
1948 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_abi)
1949 void be_init_abi(void)
1951 FIRM_DBG_REGISTER(dbg, "firm.be.abi");