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"
50 #include "beabihelper.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 1: in registers, 0: on 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;
84 * The ABI information for the current graph.
87 be_abi_call_t *call; /**< The ABI call information. */
89 ir_node *init_sp; /**< The node representing the stack pointer
90 at the start of the function. */
92 pmap *regs; /**< A map of all callee-save and ignore regs to
93 their Projs to the RegParams node. */
94 pmap *keep_map; /**< mapping blocks to keep nodes. */
96 ir_node **calls; /**< flexible array containing all be_Call nodes */
99 static ir_heights_t *ir_heights;
101 static ir_node *be_abi_reg_map_get(pmap *map, const arch_register_t *reg)
103 return pmap_get(ir_node, map, reg);
106 static void be_abi_reg_map_set(pmap *map, const arch_register_t* reg,
109 pmap_insert(map, reg, node);
113 * Check if the given register is callee save, ie. will be saved by the callee.
115 static bool arch_register_is_callee_save(
116 const arch_env_t *arch_env,
117 const arch_register_t *reg)
119 if (arch_env->impl->register_saved_by)
120 return arch_env->impl->register_saved_by(reg, /*callee=*/1);
125 * Check if the given register is caller save, ie. must be saved by the caller.
127 static bool arch_register_is_caller_save(
128 const arch_env_t *arch_env,
129 const arch_register_t *reg)
131 if (arch_env->impl->register_saved_by)
132 return arch_env->impl->register_saved_by(reg, /*callee=*/0);
139 _ ____ ___ ____ _ _ _ _
140 / \ | __ )_ _| / ___|__ _| | | |__ __ _ ___| | _____
141 / _ \ | _ \| | | | / _` | | | '_ \ / _` |/ __| |/ / __|
142 / ___ \| |_) | | | |__| (_| | | | |_) | (_| | (__| <\__ \
143 /_/ \_\____/___| \____\__,_|_|_|_.__/ \__,_|\___|_|\_\___/
145 These callbacks are used by the backend to set the parameters
146 for a specific call type.
150 * Set compare function: compares two ABI call object arguments.
152 static int cmp_call_arg(const void *a, const void *b, size_t n)
154 const be_abi_call_arg_t *p = (const be_abi_call_arg_t*)a;
155 const be_abi_call_arg_t *q = (const be_abi_call_arg_t*)b;
157 return !(p->is_res == q->is_res && p->pos == q->pos && p->callee == q->callee);
161 * Get an ABI call object argument.
163 * @param call the abi call
164 * @param is_res true for call results, false for call arguments
165 * @param pos position of the argument
166 * @param callee context type - if we are callee or caller
168 static be_abi_call_arg_t *get_call_arg(be_abi_call_t *call, int is_res, int pos, int callee)
170 be_abi_call_arg_t arg;
173 memset(&arg, 0, sizeof(arg));
178 hash = is_res * 128 + pos;
180 return set_find(be_abi_call_arg_t, call->params, &arg, sizeof(arg), hash);
184 * Set an ABI call object argument.
186 static void remember_call_arg(be_abi_call_arg_t *arg, be_abi_call_t *call, be_abi_context_t context)
188 unsigned hash = arg->is_res * 128 + arg->pos;
189 if (context & ABI_CONTEXT_CALLEE) {
191 (void)set_insert(be_abi_call_arg_t, call->params, arg, sizeof(*arg), hash);
193 if (context & ABI_CONTEXT_CALLER) {
195 (void)set_insert(be_abi_call_arg_t, call->params, arg, sizeof(*arg), hash);
199 /* Set the flags for a call. */
200 void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const be_abi_callbacks_t *cb)
206 /* Sets the number of bytes the stackframe is shrinked by the callee on return */
207 void be_abi_call_set_pop(be_abi_call_t *call, int pop)
213 void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos,
214 ir_mode *load_mode, unsigned alignment,
215 unsigned space_before, unsigned space_after,
216 be_abi_context_t context)
218 be_abi_call_arg_t arg;
219 memset(&arg, 0, sizeof(arg));
220 assert(alignment > 0 && "Alignment must be greater than 0");
221 arg.load_mode = load_mode;
222 arg.alignment = alignment;
223 arg.space_before = space_before;
224 arg.space_after = space_after;
228 remember_call_arg(&arg, call, context);
231 void be_abi_call_param_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
233 be_abi_call_arg_t arg;
234 memset(&arg, 0, sizeof(arg));
241 remember_call_arg(&arg, call, context);
244 void be_abi_call_res_reg(be_abi_call_t *call, int arg_pos, const arch_register_t *reg, be_abi_context_t context)
246 be_abi_call_arg_t arg;
247 memset(&arg, 0, sizeof(arg));
254 remember_call_arg(&arg, call, context);
257 /* Get the flags of a ABI call object. */
258 be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
264 * Constructor for a new ABI call object.
266 * @return the new ABI call object
268 static be_abi_call_t *be_abi_call_new(void)
270 be_abi_call_t *call = XMALLOCZ(be_abi_call_t);
272 call->params = new_set(cmp_call_arg, 16);
274 call->flags.try_omit_fp = be_options.omit_fp;
280 * Destructor for an ABI call object.
282 static void be_abi_call_free(be_abi_call_t *call)
284 del_set(call->params);
289 * Initializes the frame layout from parts
291 * @param frame the stack layout that will be initialized
292 * @param args the stack argument layout type
293 * @param between the between layout type
294 * @param locals the method frame type
296 * @return the initialized stack layout
298 static be_stack_layout_t *stack_frame_init(be_stack_layout_t *frame, ir_type *args,
299 ir_type *between, ir_type *locals)
301 frame->arg_type = args;
302 frame->between_type = between;
303 frame->frame_type = locals;
304 frame->initial_offset = 0;
305 frame->initial_bias = 0;
306 frame->order[1] = between;
308 /* typical decreasing stack: locals have the
309 * lowest addresses, arguments the highest */
310 frame->order[0] = locals;
311 frame->order[2] = args;
322 Adjustment of the calls inside a graph.
327 * Transform a call node into a be_Call node.
329 * @param env The ABI environment for the current irg.
330 * @param irn The call node.
331 * @param curr_sp The stack pointer node to use.
332 * @return The stack pointer after the call.
334 static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp)
336 ir_graph *irg = get_irn_irg(irn);
337 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
338 ir_type *call_tp = get_Call_type(irn);
339 ir_node *call_ptr = get_Call_ptr(irn);
340 size_t n_params = get_method_n_params(call_tp);
341 ir_node *curr_mem = get_Call_mem(irn);
342 ir_node *bl = get_nodes_block(irn);
344 const arch_register_t *sp = arch_env->sp;
345 be_abi_call_t *call = be_abi_call_new();
346 ir_mode *mach_mode = sp->reg_class->mode;
347 int n_res = get_method_n_ress(call_tp);
349 ir_node *res_proj = NULL;
350 int n_reg_params = 0;
351 int n_stack_params = 0;
354 const arch_register_t **states = NEW_ARR_F(const arch_register_t*, 0);
355 const arch_register_t **destroyed_regs = NEW_ARR_F(const arch_register_t*, 0);
359 int n_reg_results = 0;
361 int *stack_param_idx;
363 int throws_exception;
368 /* Let the isa fill out the abi description for that call node. */
369 arch_env_get_call_abi(arch_env, call_tp, call);
371 /* Insert code to put the stack arguments on the stack. */
372 assert((size_t)get_Call_n_params(irn) == n_params);
373 stack_param_idx = ALLOCAN(int, n_params);
374 for (p = 0; p < n_params; ++p) {
375 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
378 int arg_size = get_type_size_bytes(get_method_param_type(call_tp, p));
380 stack_size += round_up2(arg->space_before, arg->alignment);
381 stack_size += round_up2(arg_size, arg->alignment);
382 stack_size += round_up2(arg->space_after, arg->alignment);
384 stack_param_idx[n_stack_params++] = p;
388 /* Collect all arguments which are passed in registers. */
389 reg_param_idxs = ALLOCAN(int, n_params);
390 for (p = 0; p < n_params; ++p) {
391 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
392 if (arg && arg->in_reg) {
393 reg_param_idxs[n_reg_params++] = p;
398 * If the stack is decreasing and we do not want to store sequentially,
399 * or someone else allocated the call frame
400 * we allocate as much space on the stack all parameters need, by
401 * moving the stack pointer along the stack's direction.
403 * Note: we also have to do this for stack_size == 0, because we may have
404 * to adjust stack alignment for the call.
406 curr_sp = be_new_IncSP(sp, bl, curr_sp, stack_size, 1);
408 dbgi = get_irn_dbg_info(irn);
409 /* If there are some parameters which shall be passed on the stack. */
410 if (n_stack_params > 0) {
412 ir_node **in = ALLOCAN(ir_node*, n_stack_params+1);
415 curr_mem = get_Call_mem(irn);
416 in[n_in++] = curr_mem;
418 for (i = 0; i < n_stack_params; ++i) {
419 int p = stack_param_idx[i];
420 be_abi_call_arg_t *arg = get_call_arg(call, 0, p, 0);
421 ir_node *param = get_Call_param(irn, p);
422 ir_node *addr = curr_sp;
424 ir_type *param_type = get_method_param_type(call_tp, p);
425 int param_size = get_type_size_bytes(param_type) + arg->space_after;
428 * If we wanted to build the arguments sequentially,
429 * the stack pointer for the next must be incremented,
430 * and the memory value propagated.
432 curr_ofs += arg->space_before;
433 curr_ofs = round_up2(curr_ofs, arg->alignment);
435 /* Make the expression to compute the argument's offset. */
437 ir_mode *constmode = mach_mode;
438 if (mode_is_reference(mach_mode)) {
441 addr = new_r_Const_long(irg, constmode, curr_ofs);
442 addr = new_r_Add(bl, curr_sp, addr, mach_mode);
445 /* Insert a store for primitive arguments. */
446 if (is_atomic_type(param_type)) {
447 ir_node *nomem = get_irg_no_mem(irg);
448 ir_node *mem_input = nomem;
449 ir_node *store = new_rd_Store(dbgi, bl, mem_input, addr, param, cons_none);
450 mem = new_r_Proj(store, mode_M, pn_Store_M);
452 /* Make a mem copy for compound arguments. */
455 assert(mode_is_reference(get_irn_mode(param)));
456 copy = new_rd_CopyB(dbgi, bl, curr_mem, addr, param, param_type);
457 mem = new_r_Proj(copy, mode_M, pn_CopyB_M);
460 curr_ofs += param_size;
465 /* We need the sync only, if we didn't build the stores sequentially. */
466 if (n_stack_params >= 1) {
467 curr_mem = new_r_Sync(bl, n_in, in);
469 curr_mem = get_Call_mem(irn);
473 /* Put caller save into the destroyed set and state registers in the states
475 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
477 const arch_register_class_t *cls = &arch_env->register_classes[i];
478 for (j = 0; j < cls->n_regs; ++j) {
479 const arch_register_t *reg = arch_register_for_index(cls, j);
481 /* even if destroyed all is specified, neither SP nor FP are
482 * destroyed (else bad things will happen) */
483 if (reg == arch_env->sp || reg == arch_env->bp)
486 if (reg->type & arch_register_type_state) {
487 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
488 ARR_APP1(const arch_register_t*, states, reg);
489 /* we're already in the destroyed set so no need for further
493 if (arch_register_is_caller_save(arch_env, reg)) {
494 if (!(reg->type & arch_register_type_ignore)) {
495 ARR_APP1(const arch_register_t*, destroyed_regs, reg);
501 /* search the largest result proj number */
502 res_projs = ALLOCANZ(ir_node*, n_res);
504 foreach_out_edge(irn, edge) {
505 ir_node *irn = get_edge_src_irn(edge);
507 if (!is_Proj(irn) || get_Proj_proj(irn) != pn_Call_T_result)
510 foreach_out_edge(irn, res_edge) {
512 ir_node *res = get_edge_src_irn(res_edge);
514 assert(is_Proj(res));
516 proj = get_Proj_proj(res);
517 assert(proj < n_res);
518 assert(res_projs[proj] == NULL);
519 res_projs[proj] = res;
525 /** TODO: this is not correct for cases where return values are passed
526 * on the stack, but no known ABI does this currently...
528 n_reg_results = n_res;
531 in = ALLOCAN(ir_node*, n_reg_params + ARR_LEN(states));
533 /* make the back end call node and set its register requirements. */
534 for (i = 0; i < n_reg_params; ++i) {
535 in[n_ins++] = get_Call_param(irn, reg_param_idxs[i]);
538 /* add state registers ins */
539 for (s = 0; s < ARR_LEN(states); ++s) {
540 const arch_register_t *reg = states[s];
541 const arch_register_class_t *cls = reg->reg_class;
542 ir_node *regnode = new_r_Unknown(irg, cls->mode);
543 in[n_ins++] = regnode;
545 assert(n_ins == (int) (n_reg_params + ARR_LEN(states)));
547 /* ins collected, build the call */
548 throws_exception = ir_throws_exception(irn);
549 if (env->call->flags.call_has_imm && is_SymConst(call_ptr)) {
551 low_call = be_new_Call(dbgi, irg, bl, curr_mem, sp->single_req, curr_sp,
552 sp->single_req, curr_sp,
553 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
554 n_ins, in, get_Call_type(irn));
555 be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
558 low_call = be_new_Call(dbgi, irg, bl, curr_mem, sp->single_req, curr_sp,
559 sp->reg_class->class_req, call_ptr,
560 n_reg_results + pn_be_Call_first_res + ARR_LEN(destroyed_regs),
561 n_ins, in, get_Call_type(irn));
563 ir_set_throws_exception(low_call, throws_exception);
564 be_Call_set_pop(low_call, call->pop);
566 /* put the call into the list of all calls for later processing */
567 ARR_APP1(ir_node *, env->calls, low_call);
569 /* create new stack pointer */
570 curr_sp = new_r_Proj(low_call, get_irn_mode(curr_sp), pn_be_Call_sp);
571 be_set_constr_single_reg_out(low_call, pn_be_Call_sp, sp,
572 arch_register_req_type_ignore | arch_register_req_type_produces_sp);
573 arch_set_irn_register(curr_sp, sp);
575 /* now handle results */
576 for (i = 0; i < n_res; ++i) {
577 ir_node *proj = res_projs[i];
578 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
580 /* returns values on stack not supported yet */
584 shift the proj number to the right, since we will drop the
585 unspeakable Proj_T from the Call. Therefore, all real argument
586 Proj numbers must be increased by pn_be_Call_first_res
588 long pn = i + pn_be_Call_first_res;
591 ir_type *res_type = get_method_res_type(call_tp, i);
592 ir_mode *mode = get_type_mode(res_type);
593 proj = new_r_Proj(low_call, mode, pn);
596 set_Proj_pred(proj, low_call);
597 set_Proj_proj(proj, pn);
601 /* remove register from destroyed regs */
603 size_t n = ARR_LEN(destroyed_regs);
604 for (j = 0; j < n; ++j) {
605 if (destroyed_regs[j] == arg->reg) {
606 destroyed_regs[j] = destroyed_regs[n-1];
607 ARR_SHRINKLEN(destroyed_regs,n-1);
614 DBG((dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
616 /* Set the register classes and constraints of the Call parameters. */
617 for (i = 0; i < n_reg_params; ++i) {
618 int index = reg_param_idxs[i];
619 be_abi_call_arg_t *arg = get_call_arg(call, 0, index, 0);
620 assert(arg->reg != NULL);
622 be_set_constr_single_reg_in(low_call, n_be_Call_first_arg + i,
623 arg->reg, arch_register_req_type_none);
626 /* Set the register constraints of the results. */
627 for (i = 0; i < n_res; ++i) {
628 ir_node *proj = res_projs[i];
629 const be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 0);
630 int pn = get_Proj_proj(proj);
633 be_set_constr_single_reg_out(low_call, pn, arg->reg,
634 arch_register_req_type_none);
635 arch_set_irn_register(proj, arg->reg);
637 exchange(irn, low_call);
639 /* kill the ProjT node */
640 if (res_proj != NULL) {
644 /* Make additional projs for the caller save registers
645 and the Keep node which keeps them alive. */
651 int curr_res_proj = pn_be_Call_first_res + n_reg_results;
654 n_ins = ARR_LEN(destroyed_regs) + n_reg_results + 1;
655 in = ALLOCAN(ir_node *, n_ins);
657 /* also keep the stack pointer */
658 set_irn_link(curr_sp, (void*) sp);
661 for (d = 0; d < ARR_LEN(destroyed_regs); ++d) {
662 const arch_register_t *reg = destroyed_regs[d];
663 ir_node *proj = new_r_Proj(low_call, reg->reg_class->mode, curr_res_proj);
665 /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
666 be_set_constr_single_reg_out(low_call, curr_res_proj, reg,
667 arch_register_req_type_none);
668 arch_set_irn_register(proj, reg);
670 set_irn_link(proj, (void*) reg);
675 for (i = 0; i < n_reg_results; ++i) {
676 ir_node *proj = res_projs[i];
677 const arch_register_t *reg = arch_get_irn_register(proj);
678 set_irn_link(proj, (void*) reg);
683 /* create the Keep for the caller save registers */
684 keep = be_new_Keep(bl, n, in);
685 for (i = 0; i < n; ++i) {
686 const arch_register_t *reg = (const arch_register_t*)get_irn_link(in[i]);
687 be_node_set_reg_class_in(keep, i, reg->reg_class);
691 /* Clean up the stack. */
692 assert(stack_size >= call->pop);
693 stack_size -= call->pop;
695 if (stack_size > 0) {
696 ir_node *mem_proj = NULL;
698 foreach_out_edge(low_call, edge) {
699 ir_node *irn = get_edge_src_irn(edge);
700 if (is_Proj(irn) && get_Proj_proj(irn) == pn_Call_M) {
707 mem_proj = new_r_Proj(low_call, mode_M, pn_be_Call_M);
708 keep_alive(mem_proj);
711 /* Clean up the stack frame or revert alignment fixes if we allocated it */
712 curr_sp = be_new_IncSP(sp, bl, curr_sp, -stack_size, 0);
714 be_abi_call_free(call);
717 DEL_ARR_F(destroyed_regs);
723 * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
725 * @param alignment the minimum stack alignment
726 * @param size the node containing the non-aligned size
727 * @param block the block where new nodes are allocated on
728 * @param dbg debug info for new nodes
730 * @return a node representing the aligned size
732 static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size,
733 ir_node *block, dbg_info *dbg)
735 if (stack_alignment > 1) {
741 assert(is_po2(stack_alignment));
743 mode = get_irn_mode(size);
744 tv = new_tarval_from_long(stack_alignment-1, mode);
745 irg = get_Block_irg(block);
746 mask = new_r_Const(irg, tv);
747 size = new_rd_Add(dbg, block, size, mask, mode);
749 tv = new_tarval_from_long(-(long)stack_alignment, mode);
750 mask = new_r_Const(irg, tv);
751 size = new_rd_And(dbg, block, size, mask, mode);
757 * The alloca is transformed into a back end alloca node and connected to the stack nodes.
759 static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp)
761 ir_node *block = get_nodes_block(alloc);
762 ir_graph *irg = get_Block_irg(block);
763 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
764 ir_node *alloc_mem = NULL;
765 ir_node *alloc_res = NULL;
766 ir_type *type = get_Alloc_type(alloc);
773 unsigned stack_alignment;
775 /* all non-stack Alloc nodes should already be lowered before the backend */
776 assert(get_Alloc_where(alloc) == stack_alloc);
778 foreach_out_edge(alloc, edge) {
779 ir_node *irn = get_edge_src_irn(edge);
781 assert(is_Proj(irn));
782 switch (get_Proj_proj(irn)) {
794 /* Beware: currently Alloc nodes without a result might happen,
795 only escape analysis kills them and this phase runs only for object
796 oriented source. We kill the Alloc here. */
797 if (alloc_res == NULL && alloc_mem) {
798 exchange(alloc_mem, get_Alloc_mem(alloc));
802 dbg = get_irn_dbg_info(alloc);
803 count = get_Alloc_count(alloc);
805 /* we might need to multiply the count with the element size */
806 if (!is_unknown_type(type) && get_type_size_bytes(type) != 1) {
807 ir_mode *mode = get_irn_mode(count);
808 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type),
810 ir_node *cnst = new_rd_Const(dbg, irg, tv);
811 size = new_rd_Mul(dbg, block, count, cnst, mode);
816 /* The stack pointer will be modified in an unknown manner.
817 We cannot omit it. */
818 env->call->flags.try_omit_fp = 0;
820 stack_alignment = 1 << arch_env->stack_alignment;
821 size = adjust_alloc_size(stack_alignment, size, block, dbg);
822 new_alloc = be_new_AddSP(arch_env->sp, block, curr_sp, size);
823 set_irn_dbg_info(new_alloc, dbg);
825 if (alloc_mem != NULL) {
829 addsp_mem = new_r_Proj(new_alloc, mode_M, pn_be_AddSP_M);
831 /* We need to sync the output mem of the AddSP with the input mem
832 edge into the alloc node. */
833 ins[0] = get_Alloc_mem(alloc);
835 sync = new_r_Sync(block, 2, ins);
837 exchange(alloc_mem, sync);
840 exchange(alloc, new_alloc);
842 /* fix projnum of alloca res */
843 set_Proj_proj(alloc_res, pn_be_AddSP_res);
845 curr_sp = new_r_Proj(new_alloc, get_irn_mode(curr_sp), pn_be_AddSP_sp);
852 * The Free is transformed into a back end free node and connected to the stack nodes.
854 static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
856 ir_node *block = get_nodes_block(free);
857 ir_graph *irg = get_irn_irg(free);
858 ir_type *type = get_Free_type(free);
859 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
860 ir_mode *sp_mode = arch_env->sp->reg_class->mode;
861 dbg_info *dbg = get_irn_dbg_info(free);
862 ir_node *subsp, *mem, *res, *size, *sync;
864 unsigned stack_alignment;
866 /* all non-stack-alloc Free nodes should already be lowered before the
868 assert(get_Free_where(free) == stack_alloc);
870 /* we might need to multiply the size with the element size */
871 if (!is_unknown_type(type) && get_type_size_bytes(type) != 1) {
872 ir_tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
873 ir_node *cnst = new_rd_Const(dbg, irg, tv);
874 ir_node *mul = new_rd_Mul(dbg, block, get_Free_count(free),
878 size = get_Free_count(free);
881 stack_alignment = 1 << arch_env->stack_alignment;
882 size = adjust_alloc_size(stack_alignment, size, block, dbg);
884 /* The stack pointer will be modified in an unknown manner.
885 We cannot omit it. */
886 env->call->flags.try_omit_fp = 0;
887 subsp = be_new_SubSP(arch_env->sp, block, curr_sp, size);
888 set_irn_dbg_info(subsp, dbg);
890 mem = new_r_Proj(subsp, mode_M, pn_be_SubSP_M);
891 res = new_r_Proj(subsp, sp_mode, pn_be_SubSP_sp);
893 /* we need to sync the memory */
894 in[0] = get_Free_mem(free);
896 sync = new_r_Sync(block, 2, in);
898 /* and make the AddSP dependent on the former memory */
899 add_irn_dep(subsp, get_Free_mem(free));
902 exchange(free, sync);
909 * Check if a node is somehow data dependent on another one.
910 * both nodes must be in the same basic block.
911 * @param n1 The first node.
912 * @param n2 The second node.
913 * @return 1, if n1 is data dependent (transitively) on n2, 0 if not.
915 static int dependent_on(ir_node *n1, ir_node *n2)
917 assert(get_nodes_block(n1) == get_nodes_block(n2));
919 return heights_reachable_in_block(ir_heights, n1, n2);
922 static int cmp_call_dependency(const void *c1, const void *c2)
924 ir_node *n1 = *(ir_node **) c1;
925 ir_node *n2 = *(ir_node **) c2;
929 Classical qsort() comparison function behavior:
930 0 if both elements are equal
931 1 if second is "smaller" that first
932 -1 if first is "smaller" that second
934 if (dependent_on(n1, n2))
937 if (dependent_on(n2, n1))
940 /* The nodes have no depth order, but we need a total order because qsort()
943 * Additionally, we need to respect transitive dependencies. Consider a
944 * Call a depending on Call b and an independent Call c.
945 * We MUST NOT order c > a and b > c. */
946 h1 = get_irn_height(ir_heights, n1);
947 h2 = get_irn_height(ir_heights, n2);
948 if (h1 < h2) return -1;
949 if (h1 > h2) return 1;
950 /* Same height, so use a random (but stable) order */
951 return get_irn_idx(n1) - get_irn_idx(n2);
955 * Walker: links all Call/Alloc/Free nodes to the Block they are contained.
957 static void link_ops_in_block_walker(ir_node *irn, void *data)
959 be_abi_irg_t *env = (be_abi_irg_t*)data;
960 unsigned code = get_irn_opcode(irn);
962 if (code == iro_Call ||
963 (code == iro_Alloc && get_Alloc_where(irn) == stack_alloc) ||
964 (code == iro_Free && get_Free_where(irn) == stack_alloc)) {
965 ir_node *bl = get_nodes_block(irn);
966 void *save = get_irn_link(bl);
968 set_irn_link(irn, save);
969 set_irn_link(bl, irn);
972 if (code == iro_Builtin && get_Builtin_kind(irn) == ir_bk_return_address) {
973 ir_node *param = get_Builtin_param(irn, 0);
974 ir_tarval *tv = get_Const_tarval(param);
975 unsigned long value = get_tarval_long(tv);
976 /* use ebp, so the climbframe algo works... */
978 env->call->flags.try_omit_fp = 0;
985 * Process all Call/Alloc/Free nodes inside a basic block.
986 * Note that the link field of the block must contain a linked list of all
987 * nodes inside the Block. We first order this list according to data dependency
988 * and that connect the nodes together.
990 static void process_ops_in_block(ir_node *bl, void *data)
992 be_abi_irg_t *env = (be_abi_irg_t*)data;
993 ir_node *curr_sp = env->init_sp;
1000 for (irn = (ir_node*)get_irn_link(bl); irn != NULL;
1001 irn = (ir_node*)get_irn_link(irn)) {
1005 nodes = ALLOCAN(ir_node*, n_nodes);
1006 for (irn = (ir_node*)get_irn_link(bl), n = 0; irn != NULL;
1007 irn = (ir_node*)get_irn_link(irn), ++n) {
1011 /* If there were call nodes in the block. */
1016 /* order the call nodes according to data dependency */
1017 qsort(nodes, n_nodes, sizeof(nodes[0]), cmp_call_dependency);
1019 for (i = n_nodes - 1; i >= 0; --i) {
1020 ir_node *irn = nodes[i];
1022 DBG((dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
1023 switch (get_irn_opcode(irn)) {
1025 curr_sp = adjust_call(env, irn, curr_sp);
1028 if (get_Alloc_where(irn) == stack_alloc)
1029 curr_sp = adjust_alloc(env, irn, curr_sp);
1032 if (get_Free_where(irn) == stack_alloc)
1033 curr_sp = adjust_free(env, irn, curr_sp);
1036 panic("invalid call");
1040 /* Keep the last stack state in the block by tying it to Keep node,
1041 * the proj from calls is already kept */
1042 if (curr_sp != env->init_sp &&
1043 !(is_Proj(curr_sp) && be_is_Call(get_Proj_pred(curr_sp)))) {
1045 keep = be_new_Keep(bl, 1, nodes);
1046 pmap_insert(env->keep_map, bl, keep);
1050 set_irn_link(bl, curr_sp);
1054 * Adjust all call nodes in the graph to the ABI conventions.
1056 static void process_calls(ir_graph *const irg, be_abi_irg_t *const abi)
1058 irg_walk_graph(irg, firm_clear_link, link_ops_in_block_walker, abi);
1060 ir_heights = heights_new(irg);
1061 irg_block_walk_graph(irg, NULL, process_ops_in_block, abi);
1062 heights_free(ir_heights);
1066 * Computes the stack argument layout type.
1067 * Changes a possibly allocated value param type by moving
1068 * entities to the stack layout type.
1070 * @param call the current call ABI
1071 * @param method_type the method type
1073 * @return the stack argument layout type
1075 static ir_type *compute_arg_type(ir_graph *irg, be_abi_call_t *call,
1076 ir_type *method_type)
1078 struct obstack *obst = be_get_be_obst(irg);
1079 ir_type *frame_type = get_irg_frame_type(irg);
1080 size_t n_params = get_method_n_params(method_type);
1081 size_t n_frame_members = get_compound_n_members(frame_type);
1082 ir_entity *va_start_entity = NULL;
1088 ir_entity **map = OALLOCNZ(obst, ir_entity*, n_params);
1089 res = new_type_struct(new_id_from_chars("arg_type", 8));
1091 /* collect existing entities for value_param_types */
1092 for (f = n_frame_members; f > 0; ) {
1093 ir_entity *entity = get_compound_member(frame_type, --f);
1096 set_entity_link(entity, NULL);
1097 if (!is_parameter_entity(entity))
1099 num = get_entity_parameter_number(entity);
1100 if (num == IR_VA_START_PARAMETER_NUMBER) {
1101 /* move entity to new arg_type */
1102 set_entity_owner(entity, res);
1103 va_start_entity = entity;
1106 assert(num < n_params);
1107 if (map[num] != NULL)
1108 panic("multiple entities for parameter %u in %+F found", f, irg);
1110 if (num != n_params && get_call_arg(call, 0, num, 1)->in_reg) {
1111 /* don't move this entity */
1116 /* move entity to new arg_type */
1117 set_entity_owner(entity, res);
1120 for (i = 0; i < n_params; ++i) {
1121 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1122 ir_type *param_type = get_method_param_type(method_type, i);
1128 if (entity == NULL) {
1129 /* create a new entity */
1130 entity = new_parameter_entity(res, i, param_type);
1132 ofs += arg->space_before;
1133 ofs = round_up2(ofs, arg->alignment);
1134 set_entity_offset(entity, ofs);
1135 ofs += arg->space_after;
1136 ofs += get_type_size_bytes(param_type);
1137 arg->stack_ent = entity;
1139 if (va_start_entity != NULL) {
1140 set_entity_offset(va_start_entity, ofs);
1142 set_type_size_bytes(res, ofs);
1143 set_type_state(res, layout_fixed);
1149 const arch_register_t *reg;
1153 static int cmp_regs(const void *a, const void *b)
1155 const reg_node_map_t *p = (const reg_node_map_t*)a;
1156 const reg_node_map_t *q = (const reg_node_map_t*)b;
1158 if (p->reg->reg_class == q->reg->reg_class)
1159 return p->reg->index - q->reg->index;
1161 return p->reg->reg_class < q->reg->reg_class ? -1 : +1;
1164 static void reg_map_to_arr(reg_node_map_t *res, pmap *reg_map)
1167 size_t n = pmap_count(reg_map);
1170 foreach_pmap(reg_map, ent) {
1171 res[i].reg = (const arch_register_t*)ent->key;
1172 res[i].irn = (ir_node*)ent->value;
1176 qsort(res, n, sizeof(res[0]), cmp_regs);
1180 * Creates a be_Return for a Return node.
1182 * @param @env the abi environment
1183 * @param irn the Return node
1185 static ir_node *create_be_return(be_abi_irg_t *const env, ir_node *const irn)
1187 ir_node *const bl = get_nodes_block(irn);
1188 be_abi_call_t *call = env->call;
1189 ir_graph *irg = get_Block_irg(bl);
1190 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1191 pmap *reg_map = pmap_create();
1192 ir_node *keep = pmap_get(ir_node, env->keep_map, bl);
1197 const arch_register_t **regs;
1201 get the valid stack node in this block.
1202 If we had a call in that block there is a Keep constructed by process_calls()
1203 which points to the last stack modification in that block. we'll use
1204 it then. Else we use the stack from the start block and let
1205 the ssa construction fix the usage.
1207 stack = be_abi_reg_map_get(env->regs, arch_env->sp);
1209 stack = get_irn_n(keep, 0);
1211 remove_End_keepalive(get_irg_end(irg), keep);
1214 int const n_res = get_Return_n_ress(irn);
1215 /* Insert results for Return into the register map. */
1216 for (i = 0; i < n_res; ++i) {
1217 ir_node *res = get_Return_res(irn, i);
1218 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1219 assert(arg->in_reg && "return value must be passed in register");
1220 pmap_insert(reg_map, (void *) arg->reg, res);
1223 /* Add uses of the callee save registers. */
1224 foreach_pmap(env->regs, ent) {
1225 const arch_register_t *reg = (const arch_register_t*)ent->key;
1226 if ((reg->type & arch_register_type_ignore) || arch_register_is_callee_save(arch_env, reg))
1227 pmap_insert(reg_map, ent->key, ent->value);
1230 be_abi_reg_map_set(reg_map, arch_env->sp, stack);
1233 Maximum size of the in array for Return nodes is
1234 return args + callee save/ignore registers + memory + stack pointer
1236 in_max = pmap_count(reg_map) + n_res + 2;
1238 in = ALLOCAN(ir_node*, in_max);
1239 regs = ALLOCAN(arch_register_t const*, in_max);
1241 in[0] = get_Return_mem(irn);
1242 in[1] = be_abi_reg_map_get(reg_map, arch_env->sp);
1244 regs[1] = arch_env->sp;
1247 /* clear SP entry, since it has already been grown. */
1248 pmap_insert(reg_map, (void *) arch_env->sp, NULL);
1249 for (i = 0; i < n_res; ++i) {
1250 be_abi_call_arg_t *arg = get_call_arg(call, 1, i, 1);
1252 in[n] = be_abi_reg_map_get(reg_map, arg->reg);
1253 regs[n++] = arg->reg;
1255 /* Clear the map entry to mark the register as processed. */
1256 be_abi_reg_map_set(reg_map, arg->reg, NULL);
1259 /* grow the rest of the stuff. */
1260 foreach_pmap(reg_map, ent) {
1262 in[n] = (ir_node*)ent->value;
1263 regs[n++] = (const arch_register_t*)ent->key;
1267 /* The in array for the new back end return is now ready. */
1268 dbg_info *const dbgi = get_irn_dbg_info(irn);
1269 ir_node *const ret = be_new_Return(dbgi, irg, bl, n_res, call->pop, n, in);
1271 /* Set the register classes of the return's parameter accordingly. */
1272 for (i = 0; i < n; ++i) {
1273 if (regs[i] == NULL)
1276 be_set_constr_single_reg_in(ret, i, regs[i], arch_register_req_type_none);
1279 /* Free the space of the Epilog's in array and the register <-> proj map. */
1280 pmap_destroy(reg_map);
1285 typedef struct lower_frame_sels_env_t {
1286 ir_node *frame; /**< the current frame */
1287 const arch_register_class_t *sp_class; /**< register class of the stack pointer */
1288 } lower_frame_sels_env_t;
1291 * Walker: Replaces Sels of frame type and
1292 * value param type entities by FrameAddress.
1293 * Links all used entities.
1295 static void lower_frame_sels_walker(ir_node *irn, void *data)
1297 lower_frame_sels_env_t *ctx = (lower_frame_sels_env_t*)data;
1300 ir_node *ptr = get_Sel_ptr(irn);
1302 if (ptr == ctx->frame) {
1303 ir_entity *ent = get_Sel_entity(irn);
1304 ir_node *bl = get_nodes_block(irn);
1307 nw = be_new_FrameAddr(ctx->sp_class, bl, ctx->frame, ent);
1314 * The start block has no jump, instead it has an initial exec Proj.
1315 * The backend wants to handle all blocks the same way, so we replace
1316 * the out cfg edge with a real jump.
1318 static void fix_start_block(ir_graph *irg)
1320 ir_node *initial_X = get_irg_initial_exec(irg);
1321 ir_node *start_block = get_irg_start_block(irg);
1322 ir_node *jmp = new_r_Jmp(start_block);
1324 assert(is_Proj(initial_X));
1325 exchange(initial_X, jmp);
1326 set_irg_initial_exec(irg, new_r_Bad(irg, mode_X));
1328 /* merge start block with successor if possible */
1330 foreach_out_edge(jmp, edge) {
1331 ir_node *succ = get_edge_src_irn(edge);
1332 if (!is_Block(succ))
1335 if (get_irn_arity(succ) == 1) {
1336 exchange(succ, start_block);
1344 * Modify the irg itself and the frame type.
1346 static void modify_irg(ir_graph *const irg, be_abi_irg_t *const env)
1348 be_abi_call_t *call = env->call;
1349 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1350 const arch_register_t *sp = arch_env->sp;
1351 ir_type *method_type = get_entity_type(get_irg_entity(irg));
1352 be_irg_t *birg = be_birg_from_irg(irg);
1353 struct obstack *obst = be_get_be_obst(irg);
1354 be_stack_layout_t *stack_layout = be_get_irg_stack_layout(irg);
1357 ir_node *new_mem_proj;
1365 const arch_register_t *fp_reg;
1366 ir_node *frame_pointer;
1370 ir_type *arg_type, *bet_type;
1371 lower_frame_sels_env_t ctx;
1373 DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
1375 old_mem = get_irg_initial_mem(irg);
1377 irp_reserve_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1379 arg_type = compute_arg_type(irg, call, method_type);
1381 /* Convert the Sel nodes in the irg to frame addr nodes: */
1382 ctx.frame = get_irg_frame(irg);
1383 ctx.sp_class = arch_env->sp->reg_class;
1385 ir_type *const frame_tp = get_irg_frame_type(irg);
1386 /* layout the stackframe now */
1387 if (get_type_state(frame_tp) == layout_undefined) {
1388 default_layout_compound_type(frame_tp);
1391 /* align stackframe */
1392 unsigned const alignment = 1U << arch_env->stack_alignment;
1393 unsigned const frame_size = round_up2(get_type_size_bytes(frame_tp), alignment);
1394 set_type_size_bytes(frame_tp, frame_size);
1396 env->regs = pmap_create();
1398 n_params = get_method_n_params(method_type);
1399 args = OALLOCNZ(obst, ir_node*, n_params);
1401 be_add_parameter_entity_stores(irg);
1403 irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
1405 irp_free_resources(irp, IRP_RESOURCE_ENTITY_LINK);
1407 /* Fill the argument vector */
1408 arg_tuple = get_irg_args(irg);
1409 foreach_out_edge(arg_tuple, edge) {
1410 ir_node *irn = get_edge_src_irn(edge);
1411 if (! is_Anchor(irn)) {
1412 int nr = get_Proj_proj(irn);
1414 DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
1418 stack_layout->sp_relative = call->flags.try_omit_fp;
1419 bet_type = call->cb->get_between_type(irg);
1420 stack_frame_init(stack_layout, arg_type, bet_type,
1421 get_irg_frame_type(irg));
1423 /* Count the register params and add them to the number of Projs for the RegParams node */
1424 for (i = 0; i < n_params; ++i) {
1425 be_abi_call_arg_t *arg = get_call_arg(call, 0, i, 1);
1426 if (arg->in_reg && args[i]) {
1427 assert(arg->reg != sp && "cannot use stack pointer as parameter register");
1428 assert(i == get_Proj_proj(args[i]));
1430 /* For now, associate the register with the old Proj from Start representing that argument. */
1431 pmap_insert(env->regs, (void *) arg->reg, args[i]);
1432 DBG((dbg, LEVEL_2, "\targ #%d -> reg %s\n", i, arg->reg->name));
1436 /* Collect all callee-save registers */
1437 for (i = 0, n = arch_env->n_register_classes; i < n; ++i) {
1438 const arch_register_class_t *cls = &arch_env->register_classes[i];
1439 for (j = 0; j < cls->n_regs; ++j) {
1440 const arch_register_t *reg = &cls->regs[j];
1441 if ((reg->type & arch_register_type_state) || arch_register_is_callee_save(arch_env, reg)) {
1442 pmap_insert(env->regs, (void *) reg, NULL);
1447 fp_reg = call->flags.try_omit_fp ? arch_env->sp : arch_env->bp;
1448 rbitset_clear(birg->allocatable_regs, fp_reg->global_index);
1450 /* handle start block here (place a jump in the block) */
1451 fix_start_block(irg);
1453 pmap_insert(env->regs, (void *) sp, NULL);
1454 pmap_insert(env->regs, (void *) arch_env->bp, NULL);
1455 start_bl = get_irg_start_block(irg);
1456 ir_node *const start = be_new_Start(NULL, start_bl, pmap_count(env->regs) + 1);
1457 set_irg_start(irg, start);
1460 * make proj nodes for the callee save registers.
1461 * memorize them, since Return nodes get those as inputs.
1463 * Note, that if a register corresponds to an argument, the regs map
1464 * contains the old Proj from start for that argument.
1466 rm = ALLOCAN(reg_node_map_t, pmap_count(env->regs));
1467 reg_map_to_arr(rm, env->regs);
1468 for (i = 0, n = pmap_count(env->regs); i < n; ++i) {
1469 const arch_register_t *reg = rm[i].reg;
1470 ir_mode *mode = reg->reg_class->mode;
1472 arch_register_req_type_t add_type = arch_register_req_type_none;
1476 add_type |= arch_register_req_type_produces_sp;
1477 if (!rbitset_is_set(birg->allocatable_regs, reg->global_index)) {
1478 add_type |= arch_register_req_type_ignore;
1482 proj = new_r_Proj(start, mode, nr + 1);
1483 pmap_insert(env->regs, (void *) reg, proj);
1484 be_set_constr_single_reg_out(start, nr + 1, reg, add_type);
1485 arch_set_irn_register(proj, reg);
1487 DBG((dbg, LEVEL_2, "\tregister save proj #%d -> reg %s\n", nr, reg->name));
1490 /* create a new initial memory proj */
1491 assert(is_Proj(old_mem));
1492 arch_set_irn_register_req_out(start, 0, arch_no_register_req);
1493 new_mem_proj = new_r_Proj(start, mode_M, 0);
1495 set_irg_initial_mem(irg, mem);
1497 env->init_sp = be_abi_reg_map_get(env->regs, sp);
1499 /* set new frame_pointer */
1500 frame_pointer = be_abi_reg_map_get(env->regs, fp_reg);
1501 set_irg_frame(irg, frame_pointer);
1503 /* rewire old mem users to new mem */
1504 exchange(old_mem, mem);
1506 /* keep the mem (for functions with an endless loop = no return) */
1509 set_irg_initial_mem(irg, mem);
1511 /* Now, introduce stack param nodes for all parameters passed on the stack */
1512 for (i = 0; i < n_params; ++i) {
1513 ir_node *arg_proj = args[i];
1514 ir_node *repl = NULL;
1516 if (arg_proj != NULL) {
1517 be_abi_call_arg_t *arg;
1518 ir_type *param_type;
1519 int nr = get_Proj_proj(arg_proj);
1522 nr = MIN(nr, n_params);
1523 arg = get_call_arg(call, 0, nr, 1);
1524 param_type = get_method_param_type(method_type, nr);
1527 repl = pmap_get(ir_node, env->regs, arg->reg);
1529 ir_node *addr = be_new_FrameAddr(sp->reg_class, start_bl, frame_pointer, arg->stack_ent);
1531 /* For atomic parameters which are actually used, we create a Load node. */
1532 if (is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
1533 ir_mode *mode = get_type_mode(param_type);
1534 ir_mode *load_mode = arg->load_mode;
1535 ir_node *nomem = get_irg_no_mem(irg);
1537 ir_node *load = new_r_Load(start_bl, nomem, addr, load_mode, cons_floats);
1538 repl = new_r_Proj(load, load_mode, pn_Load_res);
1540 if (mode != load_mode) {
1541 repl = new_r_Conv(start_bl, repl, mode);
1544 /* The stack parameter is not primitive (it is a struct or array),
1545 * we thus will create a node representing the parameter's address
1551 assert(repl != NULL);
1553 /* Beware: the mode of the register parameters is always the mode of the register class
1554 which may be wrong. Add Conv's then. */
1555 mode = get_irn_mode(args[i]);
1556 if (mode != get_irn_mode(repl)) {
1557 repl = new_r_Conv(get_nodes_block(repl), repl, mode);
1559 exchange(args[i], repl);
1563 /* the arg proj is not needed anymore now and should be only used by the anchor */
1564 assert(get_irn_n_edges(arg_tuple) == 1);
1565 kill_node(arg_tuple);
1566 set_irg_args(irg, new_r_Bad(irg, mode_T));
1568 /* All Return nodes hang on the End node, so look for them there. */
1569 end = get_irg_end_block(irg);
1570 for (i = 0, n = get_Block_n_cfgpreds(end); i < n; ++i) {
1571 ir_node *irn = get_Block_cfgpred(end, i);
1573 if (is_Return(irn)) {
1574 ir_node *const ret = create_be_return(env, irn);
1579 /* if we have endless loops here, n might be <= 0. Do NOT create a be_Return then,
1580 the code is dead and will never be executed. */
1583 /** Fix the state inputs of calls that still hang on unknowns */
1584 static void fix_call_state_inputs(ir_graph *const irg, be_abi_irg_t *const env)
1586 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1588 arch_register_t **stateregs = NEW_ARR_F(arch_register_t*, 0);
1590 /* Collect caller save registers */
1591 n = arch_env->n_register_classes;
1592 for (i = 0; i < n; ++i) {
1594 const arch_register_class_t *cls = &arch_env->register_classes[i];
1595 for (j = 0; j < cls->n_regs; ++j) {
1596 const arch_register_t *reg = arch_register_for_index(cls, j);
1597 if (reg->type & arch_register_type_state) {
1598 ARR_APP1(arch_register_t*, stateregs, (arch_register_t *)reg);
1603 n = ARR_LEN(env->calls);
1604 n_states = ARR_LEN(stateregs);
1605 for (i = 0; i < n; ++i) {
1607 ir_node *call = env->calls[i];
1609 arity = get_irn_arity(call);
1611 /* the state reg inputs are the last n inputs of the calls */
1612 for (s = 0; s < n_states; ++s) {
1613 int inp = arity - n_states + s;
1614 const arch_register_t *reg = stateregs[s];
1615 ir_node *regnode = be_abi_reg_map_get(env->regs, reg);
1617 set_irn_n(call, inp, regnode);
1621 DEL_ARR_F(stateregs);
1625 * Create a trampoline entity for the given method.
1627 static ir_entity *create_trampoline(be_main_env_t *be, ir_entity *method)
1629 ir_type *type = get_entity_type(method);
1630 ident *old_id = get_entity_ld_ident(method);
1631 ident *id = id_mangle3("", old_id, "$stub");
1632 ir_type *parent = be->pic_trampolines_type;
1633 ir_entity *ent = new_entity(parent, old_id, type);
1634 set_entity_ld_ident(ent, id);
1635 set_entity_visibility(ent, ir_visibility_private);
1641 * Returns the trampoline entity for the given method.
1643 static ir_entity *get_trampoline(be_main_env_t *env, ir_entity *method)
1645 ir_entity *result = pmap_get(ir_entity, env->ent_trampoline_map, method);
1646 if (result == NULL) {
1647 result = create_trampoline(env, method);
1648 pmap_insert(env->ent_trampoline_map, method, result);
1654 static ir_entity *create_pic_symbol(be_main_env_t *be, ir_entity *entity)
1656 ident *old_id = get_entity_ld_ident(entity);
1657 ident *id = id_mangle3("", old_id, "$non_lazy_ptr");
1658 ir_type *e_type = get_entity_type(entity);
1659 ir_type *type = new_type_pointer(e_type);
1660 ir_type *parent = be->pic_symbols_type;
1661 ir_entity *ent = new_entity(parent, old_id, type);
1662 set_entity_ld_ident(ent, id);
1663 set_entity_visibility(ent, ir_visibility_private);
1668 static ir_entity *get_pic_symbol(be_main_env_t *env, ir_entity *entity)
1670 ir_entity *result = pmap_get(ir_entity, env->ent_pic_symbol_map, entity);
1671 if (result == NULL) {
1672 result = create_pic_symbol(env, entity);
1673 pmap_insert(env->ent_pic_symbol_map, entity, result);
1682 * Returns non-zero if a given entity can be accessed using a relative address.
1684 static int can_address_relative(ir_entity *entity)
1686 return entity_has_definition(entity) && !(get_entity_linkage(entity) & IR_LINKAGE_MERGE);
1689 static ir_node *get_pic_base(ir_graph *irg)
1691 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1692 if (arch_env->impl->get_pic_base == NULL)
1694 return arch_env->impl->get_pic_base(irg);
1697 /** patches SymConsts to work in position independent code */
1698 static void fix_pic_symconsts(ir_node *node, void *data)
1700 ir_graph *irg = get_irn_irg(node);
1701 be_main_env_t *be = be_get_irg_main_env(irg);
1711 arity = get_irn_arity(node);
1712 for (i = 0; i < arity; ++i) {
1714 ir_node *pred = get_irn_n(node, i);
1716 ir_entity *pic_symbol;
1717 ir_node *pic_symconst;
1719 if (!is_SymConst(pred))
1722 entity = get_SymConst_entity(pred);
1723 block = get_nodes_block(pred);
1725 /* calls can jump to relative addresses, so we can directly jump to
1726 the (relatively) known call address or the trampoline */
1727 if (i == 1 && is_Call(node)) {
1728 ir_entity *trampoline;
1729 ir_node *trampoline_const;
1731 if (can_address_relative(entity))
1734 dbgi = get_irn_dbg_info(pred);
1735 trampoline = get_trampoline(be, entity);
1736 trampoline_const = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1738 set_irn_n(node, i, trampoline_const);
1742 /* everything else is accessed relative to EIP */
1743 mode = get_irn_mode(pred);
1744 pic_base = get_pic_base(irg);
1746 /* all ok now for locally constructed stuff */
1747 if (can_address_relative(entity)) {
1748 ir_node *add = new_r_Add(block, pic_base, pred, mode);
1750 /* make sure the walker doesn't visit this add again */
1751 mark_irn_visited(add);
1752 set_irn_n(node, i, add);
1756 /* get entry from pic symbol segment */
1757 dbgi = get_irn_dbg_info(pred);
1758 pic_symbol = get_pic_symbol(be, entity);
1759 pic_symconst = new_rd_SymConst_addr_ent(dbgi, irg, mode_P_code,
1761 add = new_r_Add(block, pic_base, pic_symconst, mode);
1762 mark_irn_visited(add);
1764 /* we need an extra indirection for global data outside our current
1765 module. The loads are always safe and can therefore float
1766 and need no memory input */
1767 load = new_r_Load(block, get_irg_no_mem(irg), add, mode, cons_floats);
1768 load_res = new_r_Proj(load, mode, pn_Load_res);
1770 set_irn_n(node, i, load_res);
1774 void be_abi_introduce(ir_graph *irg)
1776 ir_node *old_frame = get_irg_frame(irg);
1777 const arch_env_t *arch_env = be_get_irg_arch_env(irg);
1778 ir_entity *entity = get_irg_entity(irg);
1779 ir_type *method_type = get_entity_type(entity);
1780 be_irg_t *birg = be_birg_from_irg(irg);
1781 struct obstack *obst = &birg->obst;
1782 ir_node *dummy = new_r_Dummy(irg,
1783 arch_env->sp->reg_class->mode);
1786 /* determine allocatable registers */
1787 assert(birg->allocatable_regs == NULL);
1788 birg->allocatable_regs = rbitset_obstack_alloc(obst, arch_env->n_registers);
1789 for (r = 0; r < arch_env->n_registers; ++r) {
1790 const arch_register_t *reg = &arch_env->registers[r];
1791 if ( !(reg->type & arch_register_type_ignore)) {
1792 rbitset_set(birg->allocatable_regs, r);
1796 /* Break here if backend provides a custom API. */
1799 env.keep_map = pmap_create();
1800 env.call = be_abi_call_new();
1801 arch_env_get_call_abi(arch_env, method_type, env.call);
1803 env.init_sp = dummy;
1804 env.calls = NEW_ARR_F(ir_node*, 0);
1808 if (be_options.pic) {
1809 irg_walk_graph(irg, fix_pic_symconsts, NULL, NULL);
1812 /* Lower all call nodes in the IRG. */
1813 process_calls(irg, &env);
1815 /* Process the IRG */
1816 modify_irg(irg, &env);
1818 /* fix call inputs for state registers */
1819 fix_call_state_inputs(irg, &env);
1821 be_abi_call_free(env.call);
1823 /* We don't need the keep map anymore. */
1824 pmap_destroy(env.keep_map);
1826 /* calls array is not needed anymore */
1827 DEL_ARR_F(env.calls);
1829 /* reroute the stack origin of the calls to the true stack origin. */
1830 exchange(dummy, env.init_sp);
1831 exchange(old_frame, get_irg_frame(irg));
1833 pmap_destroy(env.regs);
1836 void be_put_allocatable_regs(const ir_graph *irg,
1837 const arch_register_class_t *cls, bitset_t *bs)
1839 be_irg_t *birg = be_birg_from_irg(irg);
1840 unsigned *allocatable_regs = birg->allocatable_regs;
1843 assert(bitset_size(bs) == cls->n_regs);
1844 bitset_clear_all(bs);
1845 for (i = 0; i < cls->n_regs; ++i) {
1846 const arch_register_t *reg = &cls->regs[i];
1847 if (rbitset_is_set(allocatable_regs, reg->global_index))
1852 unsigned be_get_n_allocatable_regs(const ir_graph *irg,
1853 const arch_register_class_t *cls)
1855 bitset_t *bs = bitset_alloca(cls->n_regs);
1856 be_put_allocatable_regs(irg, cls, bs);
1857 return bitset_popcount(bs);
1860 void be_set_allocatable_regs(const ir_graph *irg,
1861 const arch_register_class_t *cls,
1862 unsigned *raw_bitset)
1864 be_irg_t *birg = be_birg_from_irg(irg);
1865 unsigned *allocatable_regs = birg->allocatable_regs;
1868 rbitset_clear_all(raw_bitset, cls->n_regs);
1869 for (i = 0; i < cls->n_regs; ++i) {
1870 const arch_register_t *reg = &cls->regs[i];
1871 if (rbitset_is_set(allocatable_regs, reg->global_index))
1872 rbitset_set(raw_bitset, i);
1876 BE_REGISTER_MODULE_CONSTRUCTOR(be_init_abi)
1877 void be_init_abi(void)
1879 FIRM_DBG_REGISTER(dbg, "firm.be.abi");