2 * Copyright (C) 1995-2008 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 Implements several optimizations for IA32.
23 * @author Matthias Braun, Christian Wuerdig
31 #include "firm_types.h"
45 #include "bepeephole.h"
47 #include "ia32_new_nodes.h"
48 #include "ia32_optimize.h"
49 #include "bearch_ia32_t.h"
50 #include "gen_ia32_regalloc_if.h"
51 #include "ia32_common_transform.h"
52 #include "ia32_transform.h"
53 #include "ia32_dbg_stat.h"
54 #include "ia32_architecture.h"
56 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
58 static void copy_mark(const ir_node *old, ir_node *newn)
60 if (is_ia32_is_reload(old))
61 set_ia32_is_reload(newn);
62 if (is_ia32_is_spill(old))
63 set_ia32_is_spill(newn);
64 if (is_ia32_is_remat(old))
65 set_ia32_is_remat(newn);
68 typedef enum produces_flag_t {
71 produces_zero_in_carry
75 * Return which usable flag the given node produces about the result.
76 * That is zero (ZF) and sign(SF).
77 * We do not check for carry (CF) or overflow (OF).
79 * @param node the node to check
80 * @param pn the projection number of the used result
82 static produces_flag_t check_produces_zero_sign(ir_node *node, int pn)
85 const ia32_immediate_attr_t *imm_attr;
87 if (!is_ia32_irn(node))
88 return produces_no_flag;
90 switch (get_ia32_irn_opcode(node)) {
105 assert((int)n_ia32_ShlD_count == (int)n_ia32_ShrD_count);
106 count = get_irn_n(node, n_ia32_ShlD_count);
107 goto check_shift_amount;
112 assert((int)n_ia32_Shl_count == (int)n_ia32_Shr_count
113 && (int)n_ia32_Shl_count == (int)n_ia32_Sar_count);
114 count = get_irn_n(node, n_ia32_Shl_count);
116 /* when shift count is zero the flags are not affected, so we can only
117 * do this for constants != 0 */
118 if (!is_ia32_Immediate(count))
119 return produces_no_flag;
121 imm_attr = get_ia32_immediate_attr_const(count);
122 if (imm_attr->symconst != NULL)
123 return produces_no_flag;
124 if ((imm_attr->offset & 0x1f) == 0)
125 return produces_no_flag;
129 return pn == pn_ia32_Mul_res_high ?
130 produces_zero_in_carry : produces_no_flag;
133 return produces_no_flag;
136 return pn == pn_ia32_res ? produces_zero_sign : produces_no_flag;
140 * Replace Cmp(x, 0) by a Test(x, x)
142 static void peephole_ia32_Cmp(ir_node *const node)
146 ia32_immediate_attr_t const *imm;
152 ia32_attr_t const *attr;
155 arch_register_t const *reg;
157 if (get_ia32_op_type(node) != ia32_Normal)
160 right = get_irn_n(node, n_ia32_Cmp_right);
161 if (!is_ia32_Immediate(right))
164 imm = get_ia32_immediate_attr_const(right);
165 if (imm->symconst != NULL || imm->offset != 0)
168 dbgi = get_irn_dbg_info(node);
169 irg = get_irn_irg(node);
170 block = get_nodes_block(node);
171 noreg = ia32_new_NoReg_gp(irg);
172 nomem = get_irg_no_mem(current_ir_graph);
173 op = get_irn_n(node, n_ia32_Cmp_left);
174 attr = get_ia32_attr(node);
175 ins_permuted = attr->data.ins_permuted;
177 if (is_ia32_Cmp(node)) {
178 test = new_bd_ia32_Test(dbgi, block, noreg, noreg, nomem,
179 op, op, ins_permuted);
181 test = new_bd_ia32_Test8Bit(dbgi, block, noreg, noreg, nomem,
182 op, op, ins_permuted);
184 set_ia32_ls_mode(test, get_ia32_ls_mode(node));
186 reg = arch_get_irn_register_out(node, pn_ia32_Cmp_eflags);
187 arch_set_irn_register_out(test, pn_ia32_Test_eflags, reg);
189 foreach_out_edge_safe(node, edge) {
190 ir_node *const user = get_edge_src_irn(edge);
193 exchange(user, test);
196 sched_add_before(node, test);
197 copy_mark(node, test);
198 be_peephole_exchange(node, test);
202 * Peephole optimization for Test instructions.
203 * - Remove the Test, if an appropriate flag was produced which is still live
204 * - Change a Test(x, c) to 8Bit, if 0 <= c < 256 (3 byte shorter opcode)
206 static void peephole_ia32_Test(ir_node *node)
208 ir_node *left = get_irn_n(node, n_ia32_Test_left);
209 ir_node *right = get_irn_n(node, n_ia32_Test_right);
211 assert((int)n_ia32_Test_left == (int)n_ia32_Test8Bit_left
212 && (int)n_ia32_Test_right == (int)n_ia32_Test8Bit_right);
214 if (left == right) { /* we need a test for 0 */
215 ir_node *block = get_nodes_block(node);
216 int pn = pn_ia32_res;
222 produces_flag_t produced;
224 if (get_nodes_block(left) != block)
228 pn = get_Proj_proj(op);
229 op = get_Proj_pred(op);
232 /* walk schedule up and abort when we find left or some other node
233 * destroys the flags */
236 schedpoint = sched_prev(schedpoint);
237 if (schedpoint == op)
239 if (arch_irn_is(schedpoint, modify_flags))
241 if (schedpoint == block)
242 panic("couldn't find left");
245 produced = check_produces_zero_sign(op, pn);
246 if (produced == produces_no_flag)
249 /* make sure users only look at the sign/zero flag */
250 foreach_out_edge(node, edge) {
251 ir_node *user = get_edge_src_irn(edge);
252 ia32_condition_code_t cc = get_ia32_condcode(user);
254 if (cc == ia32_cc_equal || cc == ia32_cc_not_equal)
256 if (produced == produces_zero_sign
257 && (cc == ia32_cc_sign || cc == ia32_cc_not_sign)) {
263 op_mode = get_ia32_ls_mode(op);
265 op_mode = get_irn_mode(op);
267 /* Make sure we operate on the same bit size */
268 if (get_mode_size_bits(op_mode) != get_mode_size_bits(get_ia32_ls_mode(node)))
271 if (produced == produces_zero_in_carry) {
272 /* patch users to look at the carry instead of the zero flag */
273 foreach_out_edge(node, edge) {
274 ir_node *user = get_edge_src_irn(edge);
275 ia32_condition_code_t cc = get_ia32_condcode(user);
278 case ia32_cc_equal: cc = ia32_cc_above_equal; break;
279 case ia32_cc_not_equal: cc = ia32_cc_below; break;
280 default: panic("unexpected pn");
282 set_ia32_condcode(user, cc);
286 if (get_irn_mode(op) != mode_T) {
287 set_irn_mode(op, mode_T);
289 /* If there are other users, reroute them to result proj */
290 if (get_irn_n_edges(op) != 2) {
291 ir_node *res = new_r_Proj(op, mode_Iu, pn_ia32_res);
292 edges_reroute_except(op, res, res);
295 if (get_irn_n_edges(left) == 2)
299 flags_mode = ia32_reg_classes[CLASS_ia32_flags].mode;
300 flags_proj = new_r_Proj(op, flags_mode, pn_ia32_flags);
301 arch_set_irn_register(flags_proj, &ia32_registers[REG_EFLAGS]);
303 assert(get_irn_mode(node) != mode_T);
305 be_peephole_exchange(node, flags_proj);
306 } else if (is_ia32_Immediate(right)) {
307 ia32_immediate_attr_t const *const imm = get_ia32_immediate_attr_const(right);
310 /* A test with a symconst is rather strange, but better safe than sorry */
311 if (imm->symconst != NULL)
314 offset = imm->offset;
315 if (get_ia32_op_type(node) == ia32_AddrModeS) {
316 ia32_attr_t *const attr = get_ia32_attr(node);
318 if ((offset & 0xFFFFFF00) == 0) {
319 /* attr->am_offs += 0; */
320 } else if ((offset & 0xFFFF00FF) == 0) {
321 ir_node *imm_node = ia32_create_Immediate(NULL, 0, offset>>8);
322 set_irn_n(node, n_ia32_Test_right, imm_node);
324 } else if ((offset & 0xFF00FFFF) == 0) {
325 ir_node *imm_node = ia32_create_Immediate(NULL, 0, offset>>16);
326 set_irn_n(node, n_ia32_Test_right, imm_node);
328 } else if ((offset & 0x00FFFFFF) == 0) {
329 ir_node *imm_node = ia32_create_Immediate(NULL, 0, offset>>24);
330 set_irn_n(node, n_ia32_Test_right, imm_node);
335 } else if (offset < 256) {
336 arch_register_t const* const reg = arch_get_irn_register(left);
338 if (reg != &ia32_registers[REG_EAX] &&
339 reg != &ia32_registers[REG_EBX] &&
340 reg != &ia32_registers[REG_ECX] &&
341 reg != &ia32_registers[REG_EDX]) {
348 /* Technically we should build a Test8Bit because of the register
349 * constraints, but nobody changes registers at this point anymore. */
350 set_ia32_ls_mode(node, mode_Bu);
355 * AMD Athlon works faster when RET is not destination of
356 * conditional jump or directly preceded by other jump instruction.
357 * Can be avoided by placing a Rep prefix before the return.
359 static void peephole_ia32_Return(ir_node *node)
361 if (!ia32_cg_config.use_pad_return)
364 /* check if this return is the first on the block */
365 sched_foreach_reverse_from(node, irn) {
366 switch (get_irn_opcode(irn)) {
368 /* the return node itself, ignore */
372 /* ignore no code generated */
375 /* arg, IncSP 0 nodes might occur, ignore these */
376 if (be_get_IncSP_offset(irn) == 0)
386 /* ensure, that the 3 byte return is generated */
387 be_Return_set_emit_pop(node, 1);
390 /* only optimize up to 48 stores behind IncSPs */
391 #define MAXPUSH_OPTIMIZE 48
394 * Tries to create Push's from IncSP, Store combinations.
395 * The Stores are replaced by Push's, the IncSP is modified
396 * (possibly into IncSP 0, but not removed).
398 static void peephole_IncSP_Store_to_push(ir_node *irn)
404 ir_node *stores[MAXPUSH_OPTIMIZE];
409 ir_node *first_push = NULL;
411 memset(stores, 0, sizeof(stores));
413 assert(be_is_IncSP(irn));
415 inc_ofs = be_get_IncSP_offset(irn);
420 * We first walk the schedule after the IncSP node as long as we find
421 * suitable Stores that could be transformed to a Push.
422 * We save them into the stores array which is sorted by the frame offset/4
423 * attached to the node
426 for (node = sched_next(irn); !sched_is_end(node); node = sched_next(node)) {
431 /* it has to be a Store */
432 if (!is_ia32_Store(node))
435 /* it has to use our sp value */
436 if (get_irn_n(node, n_ia32_base) != irn)
438 /* Store has to be attached to NoMem */
439 mem = get_irn_n(node, n_ia32_mem);
443 /* unfortunately we can't support the full AMs possible for push at the
444 * moment. TODO: fix this */
445 if (!is_ia32_NoReg_GP(get_irn_n(node, n_ia32_index)))
448 offset = get_ia32_am_offs_int(node);
449 /* we should NEVER access uninitialized stack BELOW the current SP */
452 /* storing at half-slots is bad */
453 if ((offset & 3) != 0)
456 if (inc_ofs - 4 < offset || offset >= MAXPUSH_OPTIMIZE * 4)
458 storeslot = offset >> 2;
460 /* storing into the same slot twice is bad (and shouldn't happen...) */
461 if (stores[storeslot] != NULL)
464 stores[storeslot] = node;
465 if (storeslot > maxslot)
471 for (i = -1; i < maxslot; ++i) {
472 if (stores[i + 1] == NULL)
476 /* walk through the Stores and create Pushs for them */
477 block = get_nodes_block(irn);
478 spmode = get_irn_mode(irn);
479 irg = get_irn_irg(irn);
480 for (; i >= 0; --i) {
481 const arch_register_t *spreg;
483 ir_node *val, *mem, *mem_proj;
484 ir_node *store = stores[i];
485 ir_node *noreg = ia32_new_NoReg_gp(irg);
487 val = get_irn_n(store, n_ia32_unary_op);
488 mem = get_irn_n(store, n_ia32_mem);
489 spreg = arch_get_irn_register(curr_sp);
491 push = new_bd_ia32_Push(get_irn_dbg_info(store), block, noreg, noreg,
493 copy_mark(store, push);
495 if (first_push == NULL)
498 sched_add_after(skip_Proj(curr_sp), push);
500 /* create stackpointer Proj */
501 curr_sp = new_r_Proj(push, spmode, pn_ia32_Push_stack);
502 arch_set_irn_register(curr_sp, spreg);
504 /* create memory Proj */
505 mem_proj = new_r_Proj(push, mode_M, pn_ia32_Push_M);
507 /* rewire Store Projs */
508 foreach_out_edge_safe(store, edge) {
509 ir_node *proj = get_edge_src_irn(edge);
512 switch (get_Proj_proj(proj)) {
513 case pn_ia32_Store_M:
514 exchange(proj, mem_proj);
517 panic("unexpected Proj on Store->IncSp");
521 /* use the memproj now */
522 be_peephole_exchange(store, push);
527 foreach_out_edge_safe(irn, edge) {
528 ir_node *const src = get_edge_src_irn(edge);
529 int const pos = get_edge_src_pos(edge);
531 if (src == first_push)
534 set_irn_n(src, pos, curr_sp);
537 be_set_IncSP_offset(irn, inc_ofs);
542 * Creates a Push instruction before the given schedule point.
544 * @param dbgi debug info
545 * @param block the block
546 * @param stack the previous stack value
547 * @param schedpoint the new node is added before this node
548 * @param reg the register to pop
550 * @return the new stack value
552 static ir_node *create_push(dbg_info *dbgi, ir_node *block,
553 ir_node *stack, ir_node *schedpoint)
555 const arch_register_t *esp = &ia32_registers[REG_ESP];
557 ir_node *val = ia32_new_NoReg_gp(cg);
558 ir_node *noreg = ia32_new_NoReg_gp(cg);
559 ir_graph *irg = get_irn_irg(block);
560 ir_node *nomem = get_irg_no_mem(irg);
561 ir_node *push = new_bd_ia32_Push(dbgi, block, noreg, noreg, nomem, val, stack);
562 sched_add_before(schedpoint, push);
564 stack = new_r_Proj(push, mode_Iu, pn_ia32_Push_stack);
565 arch_set_irn_register(stack, esp);
570 static void peephole_store_incsp(ir_node *store)
580 ir_node *am_base = get_irn_n(store, n_ia32_Store_base);
581 if (!be_is_IncSP(am_base)
582 || get_nodes_block(am_base) != get_nodes_block(store))
584 mem = get_irn_n(store, n_ia32_Store_mem);
585 if (!is_ia32_NoReg_GP(get_irn_n(store, n_ia32_Store_index))
589 int incsp_offset = be_get_IncSP_offset(am_base);
590 if (incsp_offset <= 0)
593 /* we have to be at offset 0 */
594 int my_offset = get_ia32_am_offs_int(store);
595 if (my_offset != 0) {
596 /* TODO here: find out whether there is a store with offset 0 before
597 * us and whether we can move it down to our place */
600 ir_mode *ls_mode = get_ia32_ls_mode(store);
601 int my_store_size = get_mode_size_bytes(ls_mode);
603 if (my_offset + my_store_size > incsp_offset)
606 /* correctness checking:
607 - noone else must write to that stackslot
608 (because after translation incsp won't allocate it anymore)
610 sched_foreach_reverse_from(store, node) {
616 /* make sure noone else can use the space on the stack */
617 arity = get_irn_arity(node);
618 for (i = 0; i < arity; ++i) {
619 ir_node *pred = get_irn_n(node, i);
623 if (i == n_ia32_base &&
624 (get_ia32_op_type(node) == ia32_AddrModeS
625 || get_ia32_op_type(node) == ia32_AddrModeD)) {
626 int node_offset = get_ia32_am_offs_int(node);
627 ir_mode *node_ls_mode = get_ia32_ls_mode(node);
628 int node_size = get_mode_size_bytes(node_ls_mode);
629 /* overlapping with our position? abort */
630 if (node_offset < my_offset + my_store_size
631 && node_offset + node_size >= my_offset)
633 /* otherwise it's fine */
637 /* strange use of esp: abort */
642 /* all ok, change to push */
643 dbgi = get_irn_dbg_info(store);
644 block = get_nodes_block(store);
645 noreg = ia32_new_NoReg_gp(cg);
646 val = get_irn_n(store, n_ia32_Store_val);
648 push = new_bd_ia32_Push(dbgi, block, noreg, noreg, mem,
650 create_push(dbgi, current_ir_graph, block, am_base, store);
655 * Return true if a mode can be stored in the GP register set
657 static inline int mode_needs_gp_reg(ir_mode *mode)
659 if (mode == ia32_mode_fpcw)
661 if (get_mode_size_bits(mode) > 32)
663 return mode_is_int(mode) || mode_is_reference(mode) || mode == mode_b;
667 * Tries to create Pops from Load, IncSP combinations.
668 * The Loads are replaced by Pops, the IncSP is modified
669 * (possibly into IncSP 0, but not removed).
671 static void peephole_Load_IncSP_to_pop(ir_node *irn)
673 const arch_register_t *esp = &ia32_registers[REG_ESP];
674 int i, maxslot, inc_ofs, ofs;
675 ir_node *node, *pred_sp, *block;
676 ir_node *loads[MAXPUSH_OPTIMIZE];
677 unsigned regmask = 0;
678 unsigned copymask = ~0;
680 memset(loads, 0, sizeof(loads));
681 assert(be_is_IncSP(irn));
683 inc_ofs = -be_get_IncSP_offset(irn);
688 * We first walk the schedule before the IncSP node as long as we find
689 * suitable Loads that could be transformed to a Pop.
690 * We save them into the stores array which is sorted by the frame offset/4
691 * attached to the node
694 pred_sp = be_get_IncSP_pred(irn);
695 for (node = sched_prev(irn); !sched_is_end(node); node = sched_prev(node)) {
698 const arch_register_t *sreg, *dreg;
700 /* it has to be a Load */
701 if (!is_ia32_Load(node)) {
702 if (be_is_Copy(node)) {
703 if (!mode_needs_gp_reg(get_irn_mode(node))) {
704 /* not a GP copy, ignore */
707 dreg = arch_get_irn_register(node);
708 sreg = arch_get_irn_register(be_get_Copy_op(node));
709 if (regmask & copymask & (1 << sreg->index)) {
712 if (regmask & copymask & (1 << dreg->index)) {
715 /* we CAN skip Copies if neither the destination nor the source
716 * is not in our regmask, ie none of our future Pop will overwrite it */
717 regmask |= (1 << dreg->index) | (1 << sreg->index);
718 copymask &= ~((1 << dreg->index) | (1 << sreg->index));
724 /* we can handle only GP loads */
725 if (!mode_needs_gp_reg(get_ia32_ls_mode(node)))
728 /* it has to use our predecessor sp value */
729 if (get_irn_n(node, n_ia32_base) != pred_sp) {
730 /* it would be ok if this load does not use a Pop result,
731 * but we do not check this */
735 /* should have NO index */
736 if (!is_ia32_NoReg_GP(get_irn_n(node, n_ia32_index)))
739 offset = get_ia32_am_offs_int(node);
740 /* we should NEVER access uninitialized stack BELOW the current SP */
743 /* storing at half-slots is bad */
744 if ((offset & 3) != 0)
747 if (offset < 0 || offset >= MAXPUSH_OPTIMIZE * 4)
749 /* ignore those outside the possible windows */
750 if (offset > inc_ofs - 4)
752 loadslot = offset >> 2;
754 /* loading from the same slot twice is bad (and shouldn't happen...) */
755 if (loads[loadslot] != NULL)
758 dreg = arch_get_irn_register_out(node, pn_ia32_Load_res);
759 if (regmask & (1 << dreg->index)) {
760 /* this register is already used */
763 regmask |= 1 << dreg->index;
765 loads[loadslot] = node;
766 if (loadslot > maxslot)
773 /* find the first slot */
774 for (i = maxslot; i >= 0; --i) {
775 ir_node *load = loads[i];
781 ofs = inc_ofs - (maxslot + 1) * 4;
784 /* create a new IncSP if needed */
785 block = get_nodes_block(irn);
787 pred_sp = be_new_IncSP(esp, block, pred_sp, -inc_ofs, be_get_IncSP_align(irn));
788 sched_add_before(irn, pred_sp);
791 /* walk through the Loads and create Pops for them */
792 for (++i; i <= maxslot; ++i) {
793 ir_node *load = loads[i];
795 const arch_register_t *reg;
797 mem = get_irn_n(load, n_ia32_mem);
798 reg = arch_get_irn_register_out(load, pn_ia32_Load_res);
800 pop = new_bd_ia32_Pop(get_irn_dbg_info(load), block, mem, pred_sp);
801 arch_set_irn_register_out(pop, pn_ia32_Load_res, reg);
803 copy_mark(load, pop);
805 /* create stackpointer Proj */
806 pred_sp = new_r_Proj(pop, mode_Iu, pn_ia32_Pop_stack);
807 arch_set_irn_register(pred_sp, esp);
809 sched_add_before(irn, pop);
812 foreach_out_edge_safe(load, edge) {
813 ir_node *proj = get_edge_src_irn(edge);
815 set_Proj_pred(proj, pop);
818 /* we can remove the Load now */
823 be_set_IncSP_offset(irn, -ofs);
824 be_set_IncSP_pred(irn, pred_sp);
829 * Find a free GP register if possible, else return NULL.
831 static const arch_register_t *get_free_gp_reg(ir_graph *irg)
833 be_irg_t *birg = be_birg_from_irg(irg);
836 for (i = 0; i < N_ia32_gp_REGS; ++i) {
837 const arch_register_t *reg = &ia32_reg_classes[CLASS_ia32_gp].regs[i];
838 if (!rbitset_is_set(birg->allocatable_regs, reg->global_index))
841 if (be_peephole_get_value(reg->global_index) == NULL)
849 * Creates a Pop instruction before the given schedule point.
851 * @param dbgi debug info
852 * @param block the block
853 * @param stack the previous stack value
854 * @param schedpoint the new node is added before this node
855 * @param reg the register to pop
857 * @return the new stack value
859 static ir_node *create_pop(dbg_info *dbgi, ir_node *block,
860 ir_node *stack, ir_node *schedpoint,
861 const arch_register_t *reg)
863 const arch_register_t *esp = &ia32_registers[REG_ESP];
864 ir_graph *irg = get_irn_irg(block);
870 pop = new_bd_ia32_Pop(dbgi, block, get_irg_no_mem(irg), stack);
872 stack = new_r_Proj(pop, mode_Iu, pn_ia32_Pop_stack);
873 arch_set_irn_register(stack, esp);
874 val = new_r_Proj(pop, mode_Iu, pn_ia32_Pop_res);
875 arch_set_irn_register(val, reg);
877 sched_add_before(schedpoint, pop);
880 keep = be_new_Keep(block, 1, in);
881 sched_add_before(schedpoint, keep);
887 * Optimize an IncSp by replacing it with Push/Pop.
889 static void peephole_be_IncSP(ir_node *node)
891 const arch_register_t *esp = &ia32_registers[REG_ESP];
892 const arch_register_t *reg;
898 /* first optimize incsp->incsp combinations */
899 node = be_peephole_IncSP_IncSP(node);
901 /* transform IncSP->Store combinations to Push where possible */
902 peephole_IncSP_Store_to_push(node);
904 /* transform Load->IncSP combinations to Pop where possible */
905 peephole_Load_IncSP_to_pop(node);
907 if (arch_get_irn_register(node) != esp)
910 /* replace IncSP -4 by Pop freereg when possible */
911 offset = be_get_IncSP_offset(node);
912 if ((offset != -8 || ia32_cg_config.use_add_esp_8) &&
913 (offset != -4 || ia32_cg_config.use_add_esp_4) &&
914 (offset != +4 || ia32_cg_config.use_sub_esp_4) &&
915 (offset != +8 || ia32_cg_config.use_sub_esp_8))
919 /* we need a free register for pop */
920 reg = get_free_gp_reg(get_irn_irg(node));
924 dbgi = get_irn_dbg_info(node);
925 block = get_nodes_block(node);
926 stack = be_get_IncSP_pred(node);
928 stack = create_pop(dbgi, block, stack, node, reg);
931 stack = create_pop(dbgi, block, stack, node, reg);
934 dbgi = get_irn_dbg_info(node);
935 block = get_nodes_block(node);
936 stack = be_get_IncSP_pred(node);
937 stack = new_bd_ia32_PushEax(dbgi, block, stack);
938 arch_set_irn_register(stack, esp);
939 sched_add_before(node, stack);
942 stack = new_bd_ia32_PushEax(dbgi, block, stack);
943 arch_set_irn_register(stack, esp);
944 sched_add_before(node, stack);
948 be_peephole_exchange(node, stack);
952 * Peephole optimisation for ia32_Const's
954 static void peephole_ia32_Const(ir_node *node)
956 const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(node);
957 const arch_register_t *reg;
962 /* try to transform a mov 0, reg to xor reg reg */
963 if (attr->offset != 0 || attr->symconst != NULL)
965 if (ia32_cg_config.use_mov_0)
967 /* xor destroys the flags, so no-one must be using them */
968 if (be_peephole_get_value(REG_EFLAGS) != NULL)
971 reg = arch_get_irn_register(node);
972 assert(be_peephole_get_reg_value(reg) == NULL);
974 /* create xor(produceval, produceval) */
975 block = get_nodes_block(node);
976 dbgi = get_irn_dbg_info(node);
977 xorn = new_bd_ia32_Xor0(dbgi, block);
978 arch_set_irn_register(xorn, reg);
980 sched_add_before(node, xorn);
982 copy_mark(node, xorn);
983 be_peephole_exchange(node, xorn);
986 static inline int is_noreg(const ir_node *node)
988 return is_ia32_NoReg_GP(node);
991 ir_node *ia32_immediate_from_long(long val)
993 ir_graph *irg = current_ir_graph;
994 ir_node *start_block = get_irg_start_block(irg);
996 = new_bd_ia32_Immediate(NULL, start_block, NULL, 0, 0, val);
997 arch_set_irn_register(immediate, &ia32_registers[REG_GP_NOREG]);
1002 static ir_node *create_immediate_from_am(const ir_node *node)
1004 ir_node *block = get_nodes_block(node);
1005 int offset = get_ia32_am_offs_int(node);
1006 int sc_sign = is_ia32_am_sc_sign(node);
1007 const ia32_attr_t *attr = get_ia32_attr_const(node);
1008 int sc_no_pic_adjust = attr->data.am_sc_no_pic_adjust;
1009 ir_entity *entity = get_ia32_am_sc(node);
1012 res = new_bd_ia32_Immediate(NULL, block, entity, sc_sign, sc_no_pic_adjust,
1014 arch_set_irn_register(res, &ia32_registers[REG_GP_NOREG]);
1018 static int is_am_one(const ir_node *node)
1020 int offset = get_ia32_am_offs_int(node);
1021 ir_entity *entity = get_ia32_am_sc(node);
1023 return offset == 1 && entity == NULL;
1026 static int is_am_minus_one(const ir_node *node)
1028 int offset = get_ia32_am_offs_int(node);
1029 ir_entity *entity = get_ia32_am_sc(node);
1031 return offset == -1 && entity == NULL;
1035 * Transforms a LEA into an Add or SHL if possible.
1037 static void peephole_ia32_Lea(ir_node *node)
1041 const arch_register_t *base_reg;
1042 const arch_register_t *index_reg;
1043 const arch_register_t *out_reg;
1052 assert(is_ia32_Lea(node));
1054 /* we can only do this if it is allowed to clobber the flags */
1055 if (be_peephole_get_value(REG_EFLAGS) != NULL)
1058 base = get_irn_n(node, n_ia32_Lea_base);
1059 index = get_irn_n(node, n_ia32_Lea_index);
1061 if (is_noreg(base)) {
1065 base_reg = arch_get_irn_register(base);
1067 if (is_noreg(index)) {
1071 index_reg = arch_get_irn_register(index);
1074 if (base == NULL && index == NULL) {
1075 /* we shouldn't construct these in the first place... */
1076 #ifdef DEBUG_libfirm
1077 ir_fprintf(stderr, "Optimisation warning: found immediate only lea\n");
1082 out_reg = arch_get_irn_register(node);
1083 scale = get_ia32_am_scale(node);
1084 assert(!is_ia32_need_stackent(node) || get_ia32_frame_ent(node) != NULL);
1085 /* check if we have immediates values (frame entities should already be
1086 * expressed in the offsets) */
1087 if (get_ia32_am_offs_int(node) != 0 || get_ia32_am_sc(node) != NULL) {
1093 /* we can transform leas where the out register is the same as either the
1094 * base or index register back to an Add or Shl */
1095 if (out_reg == base_reg) {
1096 if (index == NULL) {
1097 #ifdef DEBUG_libfirm
1098 if (!has_immediates) {
1099 ir_fprintf(stderr, "Optimisation warning: found lea which is "
1104 goto make_add_immediate;
1106 if (scale == 0 && !has_immediates) {
1111 /* can't create an add */
1113 } else if (out_reg == index_reg) {
1115 if (has_immediates && scale == 0) {
1117 goto make_add_immediate;
1118 } else if (!has_immediates && scale > 0) {
1120 op2 = ia32_immediate_from_long(scale);
1122 } else if (!has_immediates) {
1123 #ifdef DEBUG_libfirm
1124 ir_fprintf(stderr, "Optimisation warning: found lea which is "
1128 } else if (scale == 0 && !has_immediates) {
1133 /* can't create an add */
1136 /* can't create an add */
1141 if (ia32_cg_config.use_incdec) {
1142 if (is_am_one(node)) {
1143 dbgi = get_irn_dbg_info(node);
1144 block = get_nodes_block(node);
1145 res = new_bd_ia32_Inc(dbgi, block, op1);
1146 arch_set_irn_register(res, out_reg);
1149 if (is_am_minus_one(node)) {
1150 dbgi = get_irn_dbg_info(node);
1151 block = get_nodes_block(node);
1152 res = new_bd_ia32_Dec(dbgi, block, op1);
1153 arch_set_irn_register(res, out_reg);
1157 op2 = create_immediate_from_am(node);
1160 dbgi = get_irn_dbg_info(node);
1161 block = get_nodes_block(node);
1162 ir_graph *irg = get_irn_irg(node);
1163 ir_node *noreg = ia32_new_NoReg_gp(irg);
1164 ir_node *nomem = get_irg_no_mem(irg);
1165 res = new_bd_ia32_Add(dbgi, block, noreg, noreg, nomem, op1, op2);
1166 arch_set_irn_register(res, out_reg);
1167 set_ia32_commutative(res);
1171 dbgi = get_irn_dbg_info(node);
1172 block = get_nodes_block(node);
1173 res = new_bd_ia32_Shl(dbgi, block, op1, op2);
1174 arch_set_irn_register(res, out_reg);
1178 SET_IA32_ORIG_NODE(res, node);
1180 /* add new ADD/SHL to schedule */
1181 DBG_OPT_LEA2ADD(node, res);
1183 /* exchange the Add and the LEA */
1184 sched_add_before(node, res);
1185 copy_mark(node, res);
1186 be_peephole_exchange(node, res);
1190 * Split a Imul mem, imm into a Load mem and Imul reg, imm if possible.
1192 static void peephole_ia32_Imul_split(ir_node *imul)
1194 const ir_node *right = get_irn_n(imul, n_ia32_IMul_right);
1195 const arch_register_t *reg;
1198 if (!is_ia32_Immediate(right) || get_ia32_op_type(imul) != ia32_AddrModeS) {
1199 /* no memory, imm form ignore */
1202 /* we need a free register */
1203 reg = get_free_gp_reg(get_irn_irg(imul));
1207 /* fine, we can rebuild it */
1208 res = ia32_turn_back_am(imul);
1209 arch_set_irn_register(res, reg);
1213 * Replace xorps r,r and xorpd r,r by pxor r,r
1215 static void peephole_ia32_xZero(ir_node *xorn)
1217 set_irn_op(xorn, op_ia32_xPzero);
1221 * Replace 16bit sign extension from ax to eax by shorter cwtl
1223 static void peephole_ia32_Conv_I2I(ir_node *node)
1225 const arch_register_t *eax = &ia32_registers[REG_EAX];
1226 ir_mode *smaller_mode = get_ia32_ls_mode(node);
1227 ir_node *val = get_irn_n(node, n_ia32_Conv_I2I_val);
1232 if (get_mode_size_bits(smaller_mode) != 16 ||
1233 !mode_is_signed(smaller_mode) ||
1234 eax != arch_get_irn_register(val) ||
1235 eax != arch_get_irn_register_out(node, pn_ia32_Conv_I2I_res))
1238 dbgi = get_irn_dbg_info(node);
1239 block = get_nodes_block(node);
1240 cwtl = new_bd_ia32_Cwtl(dbgi, block, val);
1241 arch_set_irn_register(cwtl, eax);
1242 sched_add_before(node, cwtl);
1243 be_peephole_exchange(node, cwtl);
1247 * Register a peephole optimisation function.
1249 static void register_peephole_optimisation(ir_op *op, peephole_opt_func func)
1251 assert(op->ops.generic == NULL);
1252 op->ops.generic = (op_func)func;
1255 /* Perform peephole-optimizations. */
1256 void ia32_peephole_optimization(ir_graph *irg)
1258 /* we currently do it in 2 passes because:
1259 * Lea -> Add could be usefull as flag producer for Test later
1263 ir_clear_opcodes_generic_func();
1264 register_peephole_optimisation(op_ia32_Cmp, peephole_ia32_Cmp);
1265 register_peephole_optimisation(op_ia32_Cmp8Bit, peephole_ia32_Cmp);
1266 register_peephole_optimisation(op_ia32_Lea, peephole_ia32_Lea);
1267 if (ia32_cg_config.use_short_sex_eax)
1268 register_peephole_optimisation(op_ia32_Conv_I2I, peephole_ia32_Conv_I2I);
1269 if (ia32_cg_config.use_pxor)
1270 register_peephole_optimisation(op_ia32_xZero, peephole_ia32_xZero);
1271 if (! ia32_cg_config.use_imul_mem_imm32)
1272 register_peephole_optimisation(op_ia32_IMul, peephole_ia32_Imul_split);
1273 be_peephole_opt(irg);
1276 ir_clear_opcodes_generic_func();
1277 register_peephole_optimisation(op_ia32_Const, peephole_ia32_Const);
1278 register_peephole_optimisation(op_be_IncSP, peephole_be_IncSP);
1279 register_peephole_optimisation(op_ia32_Test, peephole_ia32_Test);
1280 register_peephole_optimisation(op_ia32_Test8Bit, peephole_ia32_Test);
1281 register_peephole_optimisation(op_be_Return, peephole_ia32_Return);
1282 be_peephole_opt(irg);
1286 * Removes node from schedule if it is not used anymore. If irn is a mode_T node
1287 * all its Projs are removed as well.
1288 * @param irn The irn to be removed from schedule
1290 static inline void try_kill(ir_node *node)
1292 if (get_irn_mode(node) == mode_T) {
1293 foreach_out_edge_safe(node, edge) {
1294 ir_node *proj = get_edge_src_irn(edge);
1299 if (get_irn_n_edges(node) != 0)
1302 if (sched_is_scheduled(node)) {
1309 static void optimize_conv_store(ir_node *node)
1314 ir_mode *store_mode;
1316 if (!is_ia32_Store(node) && !is_ia32_Store8Bit(node))
1319 assert((int)n_ia32_Store_val == (int)n_ia32_Store8Bit_val);
1320 pred_proj = get_irn_n(node, n_ia32_Store_val);
1321 if (is_Proj(pred_proj)) {
1322 pred = get_Proj_pred(pred_proj);
1326 if (!is_ia32_Conv_I2I(pred) && !is_ia32_Conv_I2I8Bit(pred))
1328 if (get_ia32_op_type(pred) != ia32_Normal)
1331 /* the store only stores the lower bits, so we only need the conv
1332 * it it shrinks the mode */
1333 conv_mode = get_ia32_ls_mode(pred);
1334 store_mode = get_ia32_ls_mode(node);
1335 if (get_mode_size_bits(conv_mode) < get_mode_size_bits(store_mode))
1338 ir_fprintf(stderr, "Optimisation warning: unoptimized ia32 Store(Conv) (%+F, %+F)\n", node, pred);
1339 set_irn_n(node, n_ia32_Store_val, get_irn_n(pred, n_ia32_Conv_I2I_val));
1340 if (get_irn_n_edges(pred_proj) == 0) {
1341 kill_node(pred_proj);
1342 if (pred != pred_proj)
1347 static void optimize_load_conv(ir_node *node)
1349 ir_node *pred, *predpred;
1353 if (!is_ia32_Conv_I2I(node) && !is_ia32_Conv_I2I8Bit(node))
1356 assert((int)n_ia32_Conv_I2I_val == (int)n_ia32_Conv_I2I8Bit_val);
1357 pred = get_irn_n(node, n_ia32_Conv_I2I_val);
1361 predpred = get_Proj_pred(pred);
1362 if (!is_ia32_Load(predpred))
1365 /* the load is sign extending the upper bits, so we only need the conv
1366 * if it shrinks the mode */
1367 load_mode = get_ia32_ls_mode(predpred);
1368 conv_mode = get_ia32_ls_mode(node);
1369 if (get_mode_size_bits(conv_mode) < get_mode_size_bits(load_mode))
1372 if (get_mode_sign(conv_mode) != get_mode_sign(load_mode)) {
1373 /* change the load if it has only 1 user */
1374 if (get_irn_n_edges(pred) == 1) {
1376 if (get_mode_sign(conv_mode)) {
1377 newmode = find_signed_mode(load_mode);
1379 newmode = find_unsigned_mode(load_mode);
1381 assert(newmode != NULL);
1382 set_ia32_ls_mode(predpred, newmode);
1384 /* otherwise we have to keep the conv */
1390 ir_fprintf(stderr, "Optimisation warning: unoptimized ia32 Conv(Load) (%+F, %+F)\n", node, predpred);
1391 exchange(node, pred);
1394 static void optimize_conv_conv(ir_node *node)
1396 ir_node *pred_proj, *pred, *result_conv;
1397 ir_mode *pred_mode, *conv_mode;
1401 if (!is_ia32_Conv_I2I(node) && !is_ia32_Conv_I2I8Bit(node))
1404 assert((int)n_ia32_Conv_I2I_val == (int)n_ia32_Conv_I2I8Bit_val);
1405 pred_proj = get_irn_n(node, n_ia32_Conv_I2I_val);
1406 if (is_Proj(pred_proj))
1407 pred = get_Proj_pred(pred_proj);
1411 if (!is_ia32_Conv_I2I(pred) && !is_ia32_Conv_I2I8Bit(pred))
1414 /* we know that after a conv, the upper bits are sign extended
1415 * so we only need the 2nd conv if it shrinks the mode */
1416 conv_mode = get_ia32_ls_mode(node);
1417 conv_mode_bits = get_mode_size_bits(conv_mode);
1418 pred_mode = get_ia32_ls_mode(pred);
1419 pred_mode_bits = get_mode_size_bits(pred_mode);
1421 if (conv_mode_bits == pred_mode_bits
1422 && get_mode_sign(conv_mode) == get_mode_sign(pred_mode)) {
1423 result_conv = pred_proj;
1424 } else if (conv_mode_bits <= pred_mode_bits) {
1425 /* if 2nd conv is smaller then first conv, then we can always take the
1427 if (get_irn_n_edges(pred_proj) == 1) {
1428 result_conv = pred_proj;
1429 set_ia32_ls_mode(pred, conv_mode);
1431 /* Argh:We must change the opcode to 8bit AND copy the register constraints */
1432 if (get_mode_size_bits(conv_mode) == 8) {
1433 const arch_register_req_t **reqs = arch_get_irn_register_reqs_in(node);
1434 set_irn_op(pred, op_ia32_Conv_I2I8Bit);
1435 arch_set_irn_register_reqs_in(pred, reqs);
1438 /* we don't want to end up with 2 loads, so we better do nothing */
1439 if (get_irn_mode(pred) == mode_T) {
1443 result_conv = exact_copy(pred);
1444 set_ia32_ls_mode(result_conv, conv_mode);
1446 /* Argh:We must change the opcode to 8bit AND copy the register constraints */
1447 if (get_mode_size_bits(conv_mode) == 8) {
1448 const arch_register_req_t **reqs = arch_get_irn_register_reqs_in(node);
1449 set_irn_op(result_conv, op_ia32_Conv_I2I8Bit);
1450 arch_set_irn_register_reqs_in(result_conv, reqs);
1454 /* if both convs have the same sign, then we can take the smaller one */
1455 if (get_mode_sign(conv_mode) == get_mode_sign(pred_mode)) {
1456 result_conv = pred_proj;
1458 /* no optimisation possible if smaller conv is sign-extend */
1459 if (mode_is_signed(pred_mode)) {
1462 /* we can take the smaller conv if it is unsigned */
1463 result_conv = pred_proj;
1467 ir_fprintf(stderr, "Optimisation warning: unoptimized ia32 Conv(Conv) (%+F, %+F)\n", node, pred);
1468 /* Some user (like Phis) won't be happy if we change the mode. */
1469 set_irn_mode(result_conv, get_irn_mode(node));
1472 exchange(node, result_conv);
1474 if (get_irn_n_edges(pred_proj) == 0) {
1475 kill_node(pred_proj);
1476 if (pred != pred_proj)
1479 optimize_conv_conv(result_conv);
1482 static void optimize_node(ir_node *node, void *env)
1486 optimize_load_conv(node);
1487 optimize_conv_store(node);
1488 optimize_conv_conv(node);
1492 * Performs conv and address mode optimization.
1494 void ia32_optimize_graph(ir_graph *irg)
1496 irg_walk_blkwise_graph(irg, NULL, optimize_node, NULL);
1499 void ia32_init_optimize(void)
1501 FIRM_DBG_REGISTER(dbg, "firm.be.ia32.optimize");