1 /* Integrated Register Allocator (IRA) entry point.
2 Copyright (C) 2006, 2007, 2008
3 Free Software Foundation, Inc.
4 Contributed by Vladimir Makarov <vmakarov@redhat.com>.
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 /* The integrated register allocator (IRA) is a
23 regional register allocator performing graph coloring on a top-down
24 traversal of nested regions. Graph coloring in a region is based
25 on Chaitin-Briggs algorithm. It is called integrated because
26 register coalescing, register live range splitting, and choosing a
27 better hard register are done on-the-fly during coloring. Register
28 coalescing and choosing a cheaper hard register is done by hard
29 register preferencing during hard register assigning. The live
30 range splitting is a byproduct of the regional register allocation.
32 Major IRA notions are:
34 o *Region* is a part of CFG where graph coloring based on
35 Chaitin-Briggs algorithm is done. IRA can work on any set of
36 nested CFG regions forming a tree. Currently the regions are
37 the entire function for the root region and natural loops for
38 the other regions. Therefore data structure representing a
39 region is called loop_tree_node.
41 o *Cover class* is a register class belonging to a set of
42 non-intersecting register classes containing all of the
43 hard-registers available for register allocation. The set of
44 all cover classes for a target is defined in the corresponding
45 machine-description file according some criteria. Such notion
46 is needed because Chaitin-Briggs algorithm works on
47 non-intersected register classes.
49 o *Allocno* represents the live range of a pseudo-register in a
50 region. Besides the obvious attributes like the corresponding
51 pseudo-register number, cover class, conflicting allocnos and
52 conflicting hard-registers, there are a few allocno attributes
53 which are important for understanding the allocation algorithm:
55 - *Live ranges*. This is a list of ranges of *program
56 points* where the allocno lives. Program points represent
57 places where a pseudo can be born or become dead (there are
58 approximately two times more program points than the insns)
59 and they are represented by integers starting with 0. The
60 live ranges are used to find conflicts between allocnos of
61 different cover classes. They also play very important role
62 for the transformation of the IRA internal representation of
63 several regions into a one region representation. The later is
64 used during the reload pass work because each allocno
65 represents all of the corresponding pseudo-registers.
67 - *Hard-register costs*. This is a vector of size equal to the
68 number of available hard-registers of the allocno's cover
69 class. The cost of a callee-clobbered hard-register for an
70 allocno is increased by the cost of save/restore code around
71 the calls through the given allocno's life. If the allocno
72 is a move instruction operand and another operand is a
73 hard-register of the allocno's cover class, the cost of the
74 hard-register is decreased by the move cost.
76 When an allocno is assigned, the hard-register with minimal
77 full cost is used. Initially, a hard-register's full cost is
78 the corresponding value from the hard-register's cost vector.
79 If the allocno is connected by a *copy* (see below) to
80 another allocno which has just received a hard-register, the
81 cost of the hard-register is decreased. Before choosing a
82 hard-register for an allocno, the allocno's current costs of
83 the hard-registers are modified by the conflict hard-register
84 costs of all of the conflicting allocnos which are not
87 - *Conflict hard-register costs*. This is a vector of the same
88 size as the hard-register costs vector. To permit an
89 unassigned allocno to get a better hard-register, IRA uses
90 this vector to calculate the final full cost of the
91 available hard-registers. Conflict hard-register costs of an
92 unassigned allocno are also changed with a change of the
93 hard-register cost of the allocno when a copy involving the
94 allocno is processed as described above. This is done to
95 show other unassigned allocnos that a given allocno prefers
96 some hard-registers in order to remove the move instruction
97 corresponding to the copy.
99 o *Cap*. If a pseudo-register does not live in a region but
100 lives in a nested region, IRA creates a special allocno called
101 a cap in the outer region. A region cap is also created for a
104 o *Copy*. Allocnos can be connected by copies. Copies are used
105 to modify hard-register costs for allocnos during coloring.
106 Such modifications reflects a preference to use the same
107 hard-register for the allocnos connected by copies. Usually
108 copies are created for move insns (in this case it results in
109 register coalescing). But IRA also creates copies for operands
110 of an insn which should be assigned to the same hard-register
111 due to constraints in the machine description (it usually
112 results in removing a move generated in reload to satisfy
113 the constraints) and copies referring to the allocno which is
114 the output operand of an instruction and the allocno which is
115 an input operand dying in the instruction (creation of such
116 copies results in less register shuffling). IRA *does not*
117 create copies between the same register allocnos from different
118 regions because we use another technique for propagating
119 hard-register preference on the borders of regions.
121 Allocnos (including caps) for the upper region in the region tree
122 *accumulate* information important for coloring from allocnos with
123 the same pseudo-register from nested regions. This includes
124 hard-register and memory costs, conflicts with hard-registers,
125 allocno conflicts, allocno copies and more. *Thus, attributes for
126 allocnos in a region have the same values as if the region had no
127 subregions*. It means that attributes for allocnos in the
128 outermost region corresponding to the function have the same values
129 as though the allocation used only one region which is the entire
130 function. It also means that we can look at IRA work as if the
131 first IRA did allocation for all function then it improved the
132 allocation for loops then their subloops and so on.
134 IRA major passes are:
136 o Building IRA internal representation which consists of the
139 * First, IRA builds regions and creates allocnos (file
140 ira-build.c) and initializes most of their attributes.
142 * Then IRA finds a cover class for each allocno and calculates
143 its initial (non-accumulated) cost of memory and each
144 hard-register of its cover class (file ira-cost.c).
146 * IRA creates live ranges of each allocno, calulates register
147 pressure for each cover class in each region, sets up
148 conflict hard registers for each allocno and info about calls
149 the allocno lives through (file ira-lives.c).
151 * IRA removes low register pressure loops from the regions
152 mostly to speed IRA up (file ira-build.c).
154 * IRA propagates accumulated allocno info from lower region
155 allocnos to corresponding upper region allocnos (file
158 * IRA creates all caps (file ira-build.c).
160 * Having live-ranges of allocnos and their cover classes, IRA
161 creates conflicting allocnos of the same cover class for each
162 allocno. Conflicting allocnos are stored as a bit vector or
163 array of pointers to the conflicting allocnos whatever is
164 more profitable (file ira-conflicts.c). At this point IRA
165 creates allocno copies.
167 o Coloring. Now IRA has all necessary info to start graph coloring
168 process. It is done in each region on top-down traverse of the
169 region tree (file ira-color.c). There are following subpasses:
171 * Optional aggressive coalescing of allocnos in the region.
173 * Putting allocnos onto the coloring stack. IRA uses Briggs
174 optimistic coloring which is a major improvement over
175 Chaitin's coloring. Therefore IRA does not spill allocnos at
176 this point. There is some freedom in the order of putting
177 allocnos on the stack which can affect the final result of
178 the allocation. IRA uses some heuristics to improve the order.
180 * Popping the allocnos from the stack and assigning them hard
181 registers. If IRA can not assign a hard register to an
182 allocno and the allocno is coalesced, IRA undoes the
183 coalescing and puts the uncoalesced allocnos onto the stack in
184 the hope that some such allocnos will get a hard register
185 separately. If IRA fails to assign hard register or memory
186 is more profitable for it, IRA spills the allocno. IRA
187 assigns the allocno the hard-register with minimal full
188 allocation cost which reflects the cost of usage of the
189 hard-register for the allocno and cost of usage of the
190 hard-register for allocnos conflicting with given allocno.
192 * After allono assigning in the region, IRA modifies the hard
193 register and memory costs for the corresponding allocnos in
194 the subregions to reflect the cost of possible loads, stores,
195 or moves on the border of the region and its subregions.
196 When default regional allocation algorithm is used
197 (-fira-algorithm=mixed), IRA just propagates the assignment
198 for allocnos if the register pressure in the region for the
199 corresponding cover class is less than number of available
200 hard registers for given cover class.
202 o Spill/restore code moving. When IRA performs an allocation
203 by traversing regions in top-down order, it does not know what
204 happens below in the region tree. Therefore, sometimes IRA
205 misses opportunities to perform a better allocation. A simple
206 optimization tries to improve allocation in a region having
207 subregions and containing in another region. If the
208 corresponding allocnos in the subregion are spilled, it spills
209 the region allocno if it is profitable. The optimization
210 implements a simple iterative algorithm performing profitable
211 transformations while they are still possible. It is fast in
212 practice, so there is no real need for a better time complexity
215 o Code change. After coloring, two allocnos representing the same
216 pseudo-register outside and inside a region respectively may be
217 assigned to different locations (hard-registers or memory). In
218 this case IRA creates and uses a new pseudo-register inside the
219 region and adds code to move allocno values on the region's
220 borders. This is done during top-down traversal of the regions
221 (file ira-emit.c). In some complicated cases IRA can create a
222 new allocno to move allocno values (e.g. when a swap of values
223 stored in two hard-registers is needed). At this stage, the
224 new allocno is marked as spilled. IRA still creates the
225 pseudo-register and the moves on the region borders even when
226 both allocnos were assigned to the same hard-register. If the
227 reload pass spills a pseudo-register for some reason, the
228 effect will be smaller because another allocno will still be in
229 the hard-register. In most cases, this is better then spilling
230 both allocnos. If reload does not change the allocation
231 for the two pseudo-registers, the trivial move will be removed
232 by post-reload optimizations. IRA does not generate moves for
233 allocnos assigned to the same hard register when the default
234 regional allocation algorithm is used and the register pressure
235 in the region for the corresponding allocno cover class is less
236 than number of available hard registers for given cover class.
237 IRA also does some optimizations to remove redundant stores and
238 to reduce code duplication on the region borders.
240 o Flattening internal representation. After changing code, IRA
241 transforms its internal representation for several regions into
242 one region representation (file ira-build.c). This process is
243 called IR flattening. Such process is more complicated than IR
244 rebuilding would be, but is much faster.
246 o After IR flattening, IRA tries to assign hard registers to all
247 spilled allocnos. This is impelemented by a simple and fast
248 priority coloring algorithm (see function
249 ira_reassign_conflict_allocnos::ira-color.c). Here new allocnos
250 created during the code change pass can be assigned to hard
253 o At the end IRA calls the reload pass. The reload pass
254 communicates with IRA through several functions in file
255 ira-color.c to improve its decisions in
257 * sharing stack slots for the spilled pseudos based on IRA info
258 about pseudo-register conflicts.
260 * reassigning hard-registers to all spilled pseudos at the end
261 of each reload iteration.
263 * choosing a better hard-register to spill based on IRA info
264 about pseudo-register live ranges and the register pressure
265 in places where the pseudo-register lives.
267 IRA uses a lot of data representing the target processors. These
268 data are initilized in file ira.c.
270 If function has no loops (or the loops are ignored when
271 -fira-algorithm=CB is used), we have classic Chaitin-Briggs
272 coloring (only instead of separate pass of coalescing, we use hard
273 register preferencing). In such case, IRA works much faster
274 because many things are not made (like IR flattening, the
275 spill/restore optimization, and the code change).
277 Literature is worth to read for better understanding the code:
279 o Preston Briggs, Keith D. Cooper, Linda Torczon. Improvements to
280 Graph Coloring Register Allocation.
282 o David Callahan, Brian Koblenz. Register allocation via
283 hierarchical graph coloring.
285 o Keith Cooper, Anshuman Dasgupta, Jason Eckhardt. Revisiting Graph
286 Coloring Register Allocation: A Study of the Chaitin-Briggs and
287 Callahan-Koblenz Algorithms.
289 o Guei-Yuan Lueh, Thomas Gross, and Ali-Reza Adl-Tabatabai. Global
290 Register Allocation Based on Graph Fusion.
292 o Vladimir Makarov. The Integrated Register Allocator for GCC.
294 o Vladimir Makarov. The top-down register allocator for irregular
295 register file architectures.
302 #include "coretypes.h"
311 #include "hard-reg-set.h"
312 #include "basic-block.h"
317 #include "tree-pass.h"
321 #include "integrate.h"
327 /* A modified value of flag `-fira-verbose' used internally. */
328 int internal_flag_ira_verbose;
330 /* Dump file of the allocator if it is not NULL. */
333 /* Pools for allocnos, copies, allocno live ranges. */
334 alloc_pool allocno_pool, copy_pool, allocno_live_range_pool;
336 /* The number of elements in the following array. */
337 int ira_spilled_reg_stack_slots_num;
339 /* The following array contains info about spilled pseudo-registers
340 stack slots used in current function so far. */
341 struct ira_spilled_reg_stack_slot *ira_spilled_reg_stack_slots;
343 /* Correspondingly overall cost of the allocation, cost of the
344 allocnos assigned to hard-registers, cost of the allocnos assigned
345 to memory, cost of loads, stores and register move insns generated
346 for pseudo-register live range splitting (see ira-emit.c). */
347 int ira_overall_cost;
348 int ira_reg_cost, ira_mem_cost;
349 int ira_load_cost, ira_store_cost, ira_shuffle_cost;
350 int ira_move_loops_num, ira_additional_jumps_num;
352 /* Map: hard regs X modes -> set of hard registers for storing value
353 of given mode starting with given hard register. */
354 HARD_REG_SET ira_reg_mode_hard_regset[FIRST_PSEUDO_REGISTER][NUM_MACHINE_MODES];
356 /* The following two variables are array analogs of the macros
357 MEMORY_MOVE_COST and REGISTER_MOVE_COST. */
358 short int ira_memory_move_cost[MAX_MACHINE_MODE][N_REG_CLASSES][2];
359 move_table *ira_register_move_cost[MAX_MACHINE_MODE];
361 /* Similar to may_move_in_cost but it is calculated in IRA instead of
362 regclass. Another difference is that we take only available hard
363 registers into account to figure out that one register class is a
364 subset of the another one. */
365 move_table *ira_may_move_in_cost[MAX_MACHINE_MODE];
367 /* Similar to may_move_out_cost but it is calculated in IRA instead of
368 regclass. Another difference is that we take only available hard
369 registers into account to figure out that one register class is a
370 subset of the another one. */
371 move_table *ira_may_move_out_cost[MAX_MACHINE_MODE];
373 /* Register class subset relation: TRUE if the first class is a subset
374 of the second one considering only hard registers available for the
376 int ira_class_subset_p[N_REG_CLASSES][N_REG_CLASSES];
378 /* Temporary hard reg set used for a different calculation. */
379 static HARD_REG_SET temp_hard_regset;
383 /* The function sets up the map IRA_REG_MODE_HARD_REGSET. */
385 setup_reg_mode_hard_regset (void)
387 int i, m, hard_regno;
389 for (m = 0; m < NUM_MACHINE_MODES; m++)
390 for (hard_regno = 0; hard_regno < FIRST_PSEUDO_REGISTER; hard_regno++)
392 CLEAR_HARD_REG_SET (ira_reg_mode_hard_regset[hard_regno][m]);
393 for (i = hard_regno_nregs[hard_regno][m] - 1; i >= 0; i--)
394 if (hard_regno + i < FIRST_PSEUDO_REGISTER)
395 SET_HARD_REG_BIT (ira_reg_mode_hard_regset[hard_regno][m],
402 /* Hard registers that can not be used for the register allocator for
403 all functions of the current compilation unit. */
404 static HARD_REG_SET no_unit_alloc_regs;
406 /* Array of the number of hard registers of given class which are
407 available for allocation. The order is defined by the
409 short ira_class_hard_regs[N_REG_CLASSES][FIRST_PSEUDO_REGISTER];
411 /* The number of elements of the above array for given register
413 int ira_class_hard_regs_num[N_REG_CLASSES];
415 /* Index (in ira_class_hard_regs) for given register class and hard
416 register (in general case a hard register can belong to several
417 register classes). The index is negative for hard registers
418 unavailable for the allocation. */
419 short ira_class_hard_reg_index[N_REG_CLASSES][FIRST_PSEUDO_REGISTER];
421 /* The function sets up the three arrays declared above. */
423 setup_class_hard_regs (void)
425 int cl, i, hard_regno, n;
426 HARD_REG_SET processed_hard_reg_set;
428 ira_assert (SHRT_MAX >= FIRST_PSEUDO_REGISTER);
429 /* We could call ORDER_REGS_FOR_LOCAL_ALLOC here (it is usually
430 putting hard callee-used hard registers first). But our
431 heuristics work better. */
432 for (cl = (int) N_REG_CLASSES - 1; cl >= 0; cl--)
434 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
435 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
436 CLEAR_HARD_REG_SET (processed_hard_reg_set);
437 for (n = 0, i = 0; i < FIRST_PSEUDO_REGISTER; i++)
439 #ifdef REG_ALLOC_ORDER
440 hard_regno = reg_alloc_order[i];
444 if (TEST_HARD_REG_BIT (processed_hard_reg_set, hard_regno))
446 SET_HARD_REG_BIT (processed_hard_reg_set, hard_regno);
447 if (! TEST_HARD_REG_BIT (temp_hard_regset, hard_regno))
448 ira_class_hard_reg_index[cl][hard_regno] = -1;
451 ira_class_hard_reg_index[cl][hard_regno] = n;
452 ira_class_hard_regs[cl][n++] = hard_regno;
455 ira_class_hard_regs_num[cl] = n;
459 /* Number of given class hard registers available for the register
460 allocation for given classes. */
461 int ira_available_class_regs[N_REG_CLASSES];
463 /* Set up IRA_AVAILABLE_CLASS_REGS. */
465 setup_available_class_regs (void)
469 memset (ira_available_class_regs, 0, sizeof (ira_available_class_regs));
470 for (i = 0; i < N_REG_CLASSES; i++)
472 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[i]);
473 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
474 for (j = 0; j < FIRST_PSEUDO_REGISTER; j++)
475 if (TEST_HARD_REG_BIT (temp_hard_regset, j))
476 ira_available_class_regs[i]++;
480 /* Set up global variables defining info about hard registers for the
481 allocation. These depend on USE_HARD_FRAME_P whose TRUE value means
482 that we can use the hard frame pointer for the allocation. */
484 setup_alloc_regs (bool use_hard_frame_p)
486 COPY_HARD_REG_SET (no_unit_alloc_regs, fixed_reg_set);
487 if (! use_hard_frame_p)
488 SET_HARD_REG_BIT (no_unit_alloc_regs, HARD_FRAME_POINTER_REGNUM);
489 setup_class_hard_regs ();
490 setup_available_class_regs ();
495 /* Set up IRA_MEMORY_MOVE_COST, IRA_REGISTER_MOVE_COST. */
497 setup_class_subset_and_memory_move_costs (void)
500 enum machine_mode mode;
501 HARD_REG_SET temp_hard_regset2;
503 for (mode = 0; mode < MAX_MACHINE_MODE; mode++)
504 ira_memory_move_cost[mode][NO_REGS][0]
505 = ira_memory_move_cost[mode][NO_REGS][1] = SHRT_MAX;
506 for (cl = (int) N_REG_CLASSES - 1; cl >= 0; cl--)
508 if (cl != (int) NO_REGS)
509 for (mode = 0; mode < MAX_MACHINE_MODE; mode++)
511 ira_memory_move_cost[mode][cl][0] = MEMORY_MOVE_COST (mode, cl, 0);
512 ira_memory_move_cost[mode][cl][1] = MEMORY_MOVE_COST (mode, cl, 1);
513 /* Costs for NO_REGS are used in cost calculation on the
514 1st pass when the preferred register classes are not
515 known yet. In this case we take the best scenario. */
516 if (ira_memory_move_cost[mode][NO_REGS][0]
517 > ira_memory_move_cost[mode][cl][0])
518 ira_memory_move_cost[mode][NO_REGS][0]
519 = ira_memory_move_cost[mode][cl][0];
520 if (ira_memory_move_cost[mode][NO_REGS][1]
521 > ira_memory_move_cost[mode][cl][1])
522 ira_memory_move_cost[mode][NO_REGS][1]
523 = ira_memory_move_cost[mode][cl][1];
525 for (cl2 = (int) N_REG_CLASSES - 1; cl2 >= 0; cl2--)
527 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
528 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
529 COPY_HARD_REG_SET (temp_hard_regset2, reg_class_contents[cl2]);
530 AND_COMPL_HARD_REG_SET (temp_hard_regset2, no_unit_alloc_regs);
531 ira_class_subset_p[cl][cl2]
532 = hard_reg_set_subset_p (temp_hard_regset, temp_hard_regset2);
539 /* Define the following macro if allocation through malloc if
541 #define IRA_NO_OBSTACK
543 #ifndef IRA_NO_OBSTACK
544 /* Obstack used for storing all dynamic data (except bitmaps) of the
546 static struct obstack ira_obstack;
549 /* Obstack used for storing all bitmaps of the IRA. */
550 static struct bitmap_obstack ira_bitmap_obstack;
552 /* Allocate memory of size LEN for IRA data. */
554 ira_allocate (size_t len)
558 #ifndef IRA_NO_OBSTACK
559 res = obstack_alloc (&ira_obstack, len);
566 /* Reallocate memory PTR of size LEN for IRA data. */
568 ira_reallocate (void *ptr, size_t len)
572 #ifndef IRA_NO_OBSTACK
573 res = obstack_alloc (&ira_obstack, len);
575 res = xrealloc (ptr, len);
580 /* Free memory ADDR allocated for IRA data. */
582 ira_free (void *addr ATTRIBUTE_UNUSED)
584 #ifndef IRA_NO_OBSTACK
592 /* Allocate and returns bitmap for IRA. */
594 ira_allocate_bitmap (void)
596 return BITMAP_ALLOC (&ira_bitmap_obstack);
599 /* Free bitmap B allocated for IRA. */
601 ira_free_bitmap (bitmap b ATTRIBUTE_UNUSED)
608 /* Output information about allocation of all allocnos (except for
609 caps) into file F. */
611 ira_print_disposition (FILE *f)
617 fprintf (f, "Disposition:");
618 max_regno = max_reg_num ();
619 for (n = 0, i = FIRST_PSEUDO_REGISTER; i < max_regno; i++)
620 for (a = ira_regno_allocno_map[i];
622 a = ALLOCNO_NEXT_REGNO_ALLOCNO (a))
627 fprintf (f, " %4d:r%-4d", ALLOCNO_NUM (a), ALLOCNO_REGNO (a));
628 if ((bb = ALLOCNO_LOOP_TREE_NODE (a)->bb) != NULL)
629 fprintf (f, "b%-3d", bb->index);
631 fprintf (f, "l%-3d", ALLOCNO_LOOP_TREE_NODE (a)->loop->num);
632 if (ALLOCNO_HARD_REGNO (a) >= 0)
633 fprintf (f, " %3d", ALLOCNO_HARD_REGNO (a));
640 /* Outputs information about allocation of all allocnos into
643 ira_debug_disposition (void)
645 ira_print_disposition (stderr);
650 /* For each reg class, table listing all the classes contained in it
651 (excluding the class itself. Non-allocatable registers are
652 excluded from the consideration). */
653 static enum reg_class alloc_reg_class_subclasses[N_REG_CLASSES][N_REG_CLASSES];
655 /* Initialize the table of subclasses of each reg class. */
657 setup_reg_subclasses (void)
660 HARD_REG_SET temp_hard_regset2;
662 for (i = 0; i < N_REG_CLASSES; i++)
663 for (j = 0; j < N_REG_CLASSES; j++)
664 alloc_reg_class_subclasses[i][j] = LIM_REG_CLASSES;
666 for (i = 0; i < N_REG_CLASSES; i++)
668 if (i == (int) NO_REGS)
671 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[i]);
672 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
673 if (hard_reg_set_equal_p (temp_hard_regset, ira_zero_hard_reg_set))
675 for (j = 0; j < N_REG_CLASSES; j++)
680 COPY_HARD_REG_SET (temp_hard_regset2, reg_class_contents[j]);
681 AND_COMPL_HARD_REG_SET (temp_hard_regset2, no_unit_alloc_regs);
682 if (! hard_reg_set_subset_p (temp_hard_regset,
685 p = &alloc_reg_class_subclasses[j][0];
686 while (*p != LIM_REG_CLASSES) p++;
687 *p = (enum reg_class) i;
694 /* Number of cover classes. Cover classes is non-intersected register
695 classes containing all hard-registers available for the
697 int ira_reg_class_cover_size;
699 /* The array containing cover classes (see also comments for macro
700 IRA_COVER_CLASSES). Only first IRA_REG_CLASS_COVER_SIZE elements are
702 enum reg_class ira_reg_class_cover[N_REG_CLASSES];
704 /* The number of elements in the subsequent array. */
705 int ira_important_classes_num;
707 /* The array containing non-empty classes (including non-empty cover
708 classes) which are subclasses of cover classes. Such classes is
709 important for calculation of the hard register usage costs. */
710 enum reg_class ira_important_classes[N_REG_CLASSES];
712 /* The array containing indexes of important classes in the previous
713 array. The array elements are defined only for important
715 int ira_important_class_nums[N_REG_CLASSES];
717 #ifdef IRA_COVER_CLASSES
719 /* Check IRA_COVER_CLASSES and sets the four global variables defined
722 setup_cover_and_important_classes (void)
726 static enum reg_class classes[] = IRA_COVER_CLASSES;
727 HARD_REG_SET temp_hard_regset2;
729 ira_reg_class_cover_size = 0;
730 for (i = 0; (cl = classes[i]) != LIM_REG_CLASSES; i++)
732 for (j = 0; j < i; j++)
733 if (reg_classes_intersect_p (cl, classes[j]))
735 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
736 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
737 if (! hard_reg_set_equal_p (temp_hard_regset, ira_zero_hard_reg_set))
738 ira_reg_class_cover[ira_reg_class_cover_size++] = cl;
740 ira_important_classes_num = 0;
741 for (cl = 0; cl < N_REG_CLASSES; cl++)
743 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
744 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
745 if (! hard_reg_set_equal_p (temp_hard_regset, ira_zero_hard_reg_set))
746 for (j = 0; j < ira_reg_class_cover_size; j++)
748 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
749 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
750 COPY_HARD_REG_SET (temp_hard_regset2,
751 reg_class_contents[ira_reg_class_cover[j]]);
752 AND_COMPL_HARD_REG_SET (temp_hard_regset2, no_unit_alloc_regs);
753 if (cl == ira_reg_class_cover[j]
754 || (hard_reg_set_subset_p (temp_hard_regset, temp_hard_regset2)
755 && ! hard_reg_set_equal_p (temp_hard_regset,
758 ira_important_class_nums[cl] = ira_important_classes_num;
759 ira_important_classes[ira_important_classes_num++] = cl;
766 /* Map of all register classes to corresponding cover class containing
767 the given class. If given class is not a subset of a cover class,
768 we translate it into the cheapest cover class. */
769 enum reg_class ira_class_translate[N_REG_CLASSES];
771 #ifdef IRA_COVER_CLASSES
773 /* Set up array IRA_CLASS_TRANSLATE. */
775 setup_class_translate (void)
777 enum reg_class cl, cover_class, best_class, *cl_ptr;
778 enum machine_mode mode;
779 int i, cost, min_cost, best_cost;
781 for (cl = 0; cl < N_REG_CLASSES; cl++)
782 ira_class_translate[cl] = NO_REGS;
783 for (i = 0; i < ira_reg_class_cover_size; i++)
785 cover_class = ira_reg_class_cover[i];
786 for (cl_ptr = &alloc_reg_class_subclasses[cover_class][0];
787 (cl = *cl_ptr) != LIM_REG_CLASSES;
790 if (ira_class_translate[cl] == NO_REGS)
791 ira_class_translate[cl] = cover_class;
792 #ifdef ENABLE_IRA_CHECKING
795 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
796 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
797 if (! hard_reg_set_subset_p (temp_hard_regset,
798 ira_zero_hard_reg_set))
803 ira_class_translate[cover_class] = cover_class;
805 /* For classes which are not fully covered by a cover class (in
806 other words covered by more one cover class), use the cheapest
808 for (cl = 0; cl < N_REG_CLASSES; cl++)
810 if (cl == NO_REGS || ira_class_translate[cl] != NO_REGS)
812 best_class = NO_REGS;
814 for (i = 0; i < ira_reg_class_cover_size; i++)
816 cover_class = ira_reg_class_cover[i];
817 COPY_HARD_REG_SET (temp_hard_regset,
818 reg_class_contents[cover_class]);
819 AND_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl]);
820 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
821 if (! hard_reg_set_equal_p (temp_hard_regset, ira_zero_hard_reg_set))
824 for (mode = 0; mode < MAX_MACHINE_MODE; mode++)
826 cost = (ira_memory_move_cost[mode][cl][0]
827 + ira_memory_move_cost[mode][cl][1]);
831 if (best_class == NO_REGS || best_cost > min_cost)
833 best_class = cover_class;
834 best_cost = min_cost;
838 ira_class_translate[cl] = best_class;
843 /* The biggest important reg_class inside of intersection of the two
844 reg_classes (that is calculated taking only hard registers
845 available for allocation into account). If the both reg_classes
846 contain no hard registers available for allocation, the value is
847 calculated by taking all hard-registers including fixed ones into
849 enum reg_class ira_reg_class_intersect[N_REG_CLASSES][N_REG_CLASSES];
851 /* The biggest important reg_class inside of union of the two
852 reg_classes (that is calculated taking only hard registers
853 available for allocation into account). If the both reg_classes
854 contain no hard registers available for allocation, the value is
855 calculated by taking all hard-registers including fixed ones into
856 account. In other words, the value is the corresponding
857 reg_class_subunion value. */
858 enum reg_class ira_reg_class_union[N_REG_CLASSES][N_REG_CLASSES];
860 #ifdef IRA_COVER_CLASSES
862 /* Set up IRA_REG_CLASS_INTERSECT and IRA_REG_CLASS_UNION. */
864 setup_reg_class_intersect_union (void)
866 int i, cl1, cl2, cl3;
867 HARD_REG_SET intersection_set, union_set, temp_set2;
869 for (cl1 = 0; cl1 < N_REG_CLASSES; cl1++)
871 for (cl2 = 0; cl2 < N_REG_CLASSES; cl2++)
873 ira_reg_class_intersect[cl1][cl2] = NO_REGS;
874 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl1]);
875 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
876 COPY_HARD_REG_SET (temp_set2, reg_class_contents[cl2]);
877 AND_COMPL_HARD_REG_SET (temp_set2, no_unit_alloc_regs);
878 if (hard_reg_set_equal_p (temp_hard_regset, ira_zero_hard_reg_set)
879 && hard_reg_set_equal_p (temp_set2, ira_zero_hard_reg_set))
883 cl3 = reg_class_subclasses[cl1][i];
884 if (cl3 == LIM_REG_CLASSES)
886 if (reg_class_subset_p (ira_reg_class_intersect[cl1][cl2],
888 ira_reg_class_intersect[cl1][cl2] = cl3;
890 ira_reg_class_union[cl1][cl2] = reg_class_subunion[cl1][cl2];
893 ira_reg_class_union[cl1][cl2] = NO_REGS;
894 COPY_HARD_REG_SET (intersection_set, reg_class_contents[cl1]);
895 AND_HARD_REG_SET (intersection_set, reg_class_contents[cl2]);
896 AND_COMPL_HARD_REG_SET (intersection_set, no_unit_alloc_regs);
897 COPY_HARD_REG_SET (union_set, reg_class_contents[cl1]);
898 IOR_HARD_REG_SET (union_set, reg_class_contents[cl2]);
899 AND_COMPL_HARD_REG_SET (union_set, no_unit_alloc_regs);
900 for (i = 0; i < ira_important_classes_num; i++)
902 cl3 = ira_important_classes[i];
903 COPY_HARD_REG_SET (temp_hard_regset, reg_class_contents[cl3]);
904 AND_COMPL_HARD_REG_SET (temp_hard_regset, no_unit_alloc_regs);
905 if (hard_reg_set_subset_p (temp_hard_regset, intersection_set))
909 reg_class_contents[(int)
910 ira_reg_class_intersect[cl1][cl2]]);
911 AND_COMPL_HARD_REG_SET (temp_set2, no_unit_alloc_regs);
912 if (! hard_reg_set_subset_p (temp_hard_regset, temp_set2)
913 /* Ignore unavailable hard registers and prefer
914 smallest class for debugging purposes. */
915 || (hard_reg_set_equal_p (temp_hard_regset, temp_set2)
916 && hard_reg_set_subset_p
917 (reg_class_contents[cl3],
919 [(int) ira_reg_class_intersect[cl1][cl2]])))
920 ira_reg_class_intersect[cl1][cl2] = (enum reg_class) cl3;
922 if (hard_reg_set_subset_p (temp_hard_regset, union_set))
926 reg_class_contents[(int) ira_reg_class_union[cl1][cl2]]);
927 AND_COMPL_HARD_REG_SET (temp_set2, no_unit_alloc_regs);
928 if (ira_reg_class_union[cl1][cl2] == NO_REGS
929 || (hard_reg_set_subset_p (temp_set2, temp_hard_regset)
931 && (! hard_reg_set_equal_p (temp_set2,
933 /* Ignore unavailable hard registers and
934 prefer smallest class for debugging
936 || hard_reg_set_subset_p
937 (reg_class_contents[cl3],
939 [(int) ira_reg_class_union[cl1][cl2]]))))
940 ira_reg_class_union[cl1][cl2] = (enum reg_class) cl3;
949 /* Output all cover classes and the translation map into file F. */
951 print_class_cover (FILE *f)
953 static const char *const reg_class_names[] = REG_CLASS_NAMES;
956 fprintf (f, "Class cover:\n");
957 for (i = 0; i < ira_reg_class_cover_size; i++)
958 fprintf (f, " %s", reg_class_names[ira_reg_class_cover[i]]);
959 fprintf (f, "\nClass translation:\n");
960 for (i = 0; i < N_REG_CLASSES; i++)
961 fprintf (f, " %s -> %s\n", reg_class_names[i],
962 reg_class_names[ira_class_translate[i]]);
965 /* Output all cover classes and the translation map into
968 ira_debug_class_cover (void)
970 print_class_cover (stderr);
973 /* Set up different arrays concerning class subsets, cover and
974 important classes. */
976 find_reg_class_closure (void)
978 setup_reg_subclasses ();
979 #ifdef IRA_COVER_CLASSES
980 setup_cover_and_important_classes ();
981 setup_class_translate ();
982 setup_reg_class_intersect_union ();
988 /* Map: register class x machine mode -> number of hard registers of
989 given class needed to store value of given mode. If the number is
990 different, the size will be negative. */
991 int ira_reg_class_nregs[N_REG_CLASSES][MAX_MACHINE_MODE];
993 /* Maximal value of the previous array elements. */
996 /* Form IRA_REG_CLASS_NREGS map. */
998 setup_reg_class_nregs (void)
1004 for (cl = 0; cl < N_REG_CLASSES; cl++)
1005 for (m = 0; m < MAX_MACHINE_MODE; m++)
1007 ira_reg_class_nregs[cl][m] = CLASS_MAX_NREGS (cl, m);
1008 if (ira_max_nregs < ira_reg_class_nregs[cl][m])
1009 ira_max_nregs = ira_reg_class_nregs[cl][m];
1015 /* Array whose values are hard regset of hard registers available for
1016 the allocation of given register class whose HARD_REGNO_MODE_OK
1017 values for given mode are zero. */
1018 HARD_REG_SET prohibited_class_mode_regs[N_REG_CLASSES][NUM_MACHINE_MODES];
1020 /* Set up PROHIBITED_CLASS_MODE_REGS. */
1022 setup_prohibited_class_mode_regs (void)
1024 int i, j, k, hard_regno;
1027 for (i = 0; i < ira_reg_class_cover_size; i++)
1029 cl = ira_reg_class_cover[i];
1030 for (j = 0; j < NUM_MACHINE_MODES; j++)
1032 CLEAR_HARD_REG_SET (prohibited_class_mode_regs[cl][j]);
1033 for (k = ira_class_hard_regs_num[cl] - 1; k >= 0; k--)
1035 hard_regno = ira_class_hard_regs[cl][k];
1036 if (! HARD_REGNO_MODE_OK (hard_regno, j))
1037 SET_HARD_REG_BIT (prohibited_class_mode_regs[cl][j],
1046 /* Allocate and initialize IRA_REGISTER_MOVE_COST,
1047 IRA_MAY_MOVE_IN_COST, and IRA_MAY_MOVE_OUT_COST for MODE if it is
1050 ira_init_register_move_cost (enum machine_mode mode)
1054 ira_assert (ira_register_move_cost[mode] == NULL
1055 && ira_may_move_in_cost[mode] == NULL
1056 && ira_may_move_out_cost[mode] == NULL);
1057 if (move_cost[mode] == NULL)
1058 init_move_cost (mode);
1059 ira_register_move_cost[mode] = move_cost[mode];
1060 /* Don't use ira_allocate because the tables exist out of scope of a
1062 ira_may_move_in_cost[mode]
1063 = (move_table *) xmalloc (sizeof (move_table) * N_REG_CLASSES);
1064 memcpy (ira_may_move_in_cost[mode], may_move_in_cost[mode],
1065 sizeof (move_table) * N_REG_CLASSES);
1066 ira_may_move_out_cost[mode]
1067 = (move_table *) xmalloc (sizeof (move_table) * N_REG_CLASSES);
1068 memcpy (ira_may_move_out_cost[mode], may_move_out_cost[mode],
1069 sizeof (move_table) * N_REG_CLASSES);
1070 for (cl1 = 0; cl1 < N_REG_CLASSES; cl1++)
1072 for (cl2 = 0; cl2 < N_REG_CLASSES; cl2++)
1074 if (ira_class_subset_p[cl1][cl2])
1075 ira_may_move_in_cost[mode][cl1][cl2] = 0;
1076 if (ira_class_subset_p[cl2][cl1])
1077 ira_may_move_out_cost[mode][cl1][cl2] = 0;
1084 /* Hard regsets whose all bits are correspondingly zero or one. */
1085 HARD_REG_SET ira_zero_hard_reg_set;
1086 HARD_REG_SET ira_one_hard_reg_set;
1088 /* This is called once during compiler work. It sets up
1089 different arrays whose values don't depend on the compiled
1092 ira_init_once (void)
1094 enum machine_mode mode;
1096 CLEAR_HARD_REG_SET (ira_zero_hard_reg_set);
1097 SET_HARD_REG_SET (ira_one_hard_reg_set);
1098 for (mode = 0; mode < MAX_MACHINE_MODE; mode++)
1100 ira_register_move_cost[mode] = NULL;
1101 ira_may_move_in_cost[mode] = NULL;
1102 ira_may_move_out_cost[mode] = NULL;
1104 ira_init_costs_once ();
1107 /* Free ira_register_move_cost, ira_may_move_in_cost, and
1108 ira_may_move_out_cost for each mode. */
1110 free_register_move_costs (void)
1112 enum machine_mode mode;
1114 for (mode = 0; mode < MAX_MACHINE_MODE; mode++)
1116 if (ira_may_move_in_cost[mode] != NULL)
1117 free (ira_may_move_in_cost[mode]);
1118 if (ira_may_move_out_cost[mode] != NULL)
1119 free (ira_may_move_out_cost[mode]);
1120 ira_register_move_cost[mode] = NULL;
1121 ira_may_move_in_cost[mode] = NULL;
1122 ira_may_move_out_cost[mode] = NULL;
1126 /* This is called every time when register related information is
1131 free_register_move_costs ();
1132 setup_reg_mode_hard_regset ();
1133 setup_alloc_regs (flag_omit_frame_pointer != 0);
1134 setup_class_subset_and_memory_move_costs ();
1135 find_reg_class_closure ();
1136 setup_reg_class_nregs ();
1137 setup_prohibited_class_mode_regs ();
1141 /* Function called once at the end of compiler work. */
1143 ira_finish_once (void)
1145 ira_finish_costs_once ();
1146 free_register_move_costs ();
1151 /* Array whose values are hard regset of hard registers for which
1152 move of the hard register in given mode into itself is
1154 HARD_REG_SET ira_prohibited_mode_move_regs[NUM_MACHINE_MODES];
1156 /* Flag of that the above array has been initialized. */
1157 static bool ira_prohibited_mode_move_regs_initialized_p = false;
1159 /* Set up IRA_PROHIBITED_MODE_MOVE_REGS. */
1161 setup_prohibited_mode_move_regs (void)
1164 rtx test_reg1, test_reg2, move_pat, move_insn;
1166 if (ira_prohibited_mode_move_regs_initialized_p)
1168 ira_prohibited_mode_move_regs_initialized_p = true;
1169 test_reg1 = gen_rtx_REG (VOIDmode, 0);
1170 test_reg2 = gen_rtx_REG (VOIDmode, 0);
1171 move_pat = gen_rtx_SET (VOIDmode, test_reg1, test_reg2);
1172 move_insn = gen_rtx_INSN (VOIDmode, 0, 0, 0, 0, 0, move_pat, -1, 0);
1173 for (i = 0; i < NUM_MACHINE_MODES; i++)
1175 SET_HARD_REG_SET (ira_prohibited_mode_move_regs[i]);
1176 for (j = 0; j < FIRST_PSEUDO_REGISTER; j++)
1178 if (! HARD_REGNO_MODE_OK (j, i))
1180 SET_REGNO (test_reg1, j);
1181 PUT_MODE (test_reg1, i);
1182 SET_REGNO (test_reg2, j);
1183 PUT_MODE (test_reg2, i);
1184 INSN_CODE (move_insn) = -1;
1185 recog_memoized (move_insn);
1186 if (INSN_CODE (move_insn) < 0)
1188 extract_insn (move_insn);
1189 if (! constrain_operands (1))
1191 CLEAR_HARD_REG_BIT (ira_prohibited_mode_move_regs[i], j);
1198 /* Function specific hard registers that can not be used for the
1199 register allocation. */
1200 HARD_REG_SET ira_no_alloc_regs;
1202 /* Return TRUE if *LOC contains an asm. */
1204 insn_contains_asm_1 (rtx *loc, void *data ATTRIBUTE_UNUSED)
1208 if (GET_CODE (*loc) == ASM_OPERANDS)
1214 /* Return TRUE if INSN contains an ASM. */
1216 insn_contains_asm (rtx insn)
1218 return for_each_rtx (&insn, insn_contains_asm_1, NULL);
1221 /* Set up regs_asm_clobbered. */
1223 compute_regs_asm_clobbered (char *regs_asm_clobbered)
1227 memset (regs_asm_clobbered, 0, sizeof (char) * FIRST_PSEUDO_REGISTER);
1232 FOR_BB_INSNS_REVERSE (bb, insn)
1234 struct df_ref **def_rec;
1236 if (insn_contains_asm (insn))
1237 for (def_rec = DF_INSN_DEFS (insn); *def_rec; def_rec++)
1239 struct df_ref *def = *def_rec;
1240 unsigned int dregno = DF_REF_REGNO (def);
1241 if (dregno < FIRST_PSEUDO_REGISTER)
1244 enum machine_mode mode = GET_MODE (DF_REF_REAL_REG (def));
1245 unsigned int end = dregno
1246 + hard_regno_nregs[dregno][mode] - 1;
1248 for (i = dregno; i <= end; ++i)
1249 regs_asm_clobbered[i] = 1;
1257 /* Set up ELIMINABLE_REGSET, IRA_NO_ALLOC_REGS, and REGS_EVER_LIVE. */
1259 setup_eliminable_regset (void)
1261 /* Like regs_ever_live, but 1 if a reg is set or clobbered from an
1262 asm. Unlike regs_ever_live, elements of this array corresponding
1263 to eliminable regs (like the frame pointer) are set if an asm
1265 char *regs_asm_clobbered
1266 = (char *) alloca (FIRST_PSEUDO_REGISTER * sizeof (char));
1267 #ifdef ELIMINABLE_REGS
1269 static const struct {const int from, to; } eliminables[] = ELIMINABLE_REGS;
1271 /* FIXME: If EXIT_IGNORE_STACK is set, we will not save and restore
1272 sp for alloca. So we can't eliminate the frame pointer in that
1273 case. At some point, we should improve this by emitting the
1274 sp-adjusting insns for this case. */
1276 = (! flag_omit_frame_pointer
1277 || (cfun->calls_alloca && EXIT_IGNORE_STACK)
1278 || crtl->accesses_prior_frames
1279 || crtl->stack_realign_needed
1280 || FRAME_POINTER_REQUIRED);
1282 frame_pointer_needed = need_fp;
1284 COPY_HARD_REG_SET (ira_no_alloc_regs, no_unit_alloc_regs);
1285 CLEAR_HARD_REG_SET (eliminable_regset);
1287 compute_regs_asm_clobbered (regs_asm_clobbered);
1288 /* Build the regset of all eliminable registers and show we can't
1289 use those that we already know won't be eliminated. */
1290 #ifdef ELIMINABLE_REGS
1291 for (i = 0; i < (int) ARRAY_SIZE (eliminables); i++)
1294 = (! CAN_ELIMINATE (eliminables[i].from, eliminables[i].to)
1295 || (eliminables[i].to == STACK_POINTER_REGNUM && need_fp));
1297 if (! regs_asm_clobbered[eliminables[i].from])
1299 SET_HARD_REG_BIT (eliminable_regset, eliminables[i].from);
1302 SET_HARD_REG_BIT (ira_no_alloc_regs, eliminables[i].from);
1304 else if (cannot_elim)
1305 error ("%s cannot be used in asm here",
1306 reg_names[eliminables[i].from]);
1308 df_set_regs_ever_live (eliminables[i].from, true);
1310 #if FRAME_POINTER_REGNUM != HARD_FRAME_POINTER_REGNUM
1311 if (! regs_asm_clobbered[HARD_FRAME_POINTER_REGNUM])
1313 SET_HARD_REG_BIT (eliminable_regset, HARD_FRAME_POINTER_REGNUM);
1315 SET_HARD_REG_BIT (ira_no_alloc_regs, HARD_FRAME_POINTER_REGNUM);
1318 error ("%s cannot be used in asm here",
1319 reg_names[HARD_FRAME_POINTER_REGNUM]);
1321 df_set_regs_ever_live (HARD_FRAME_POINTER_REGNUM, true);
1325 if (! regs_asm_clobbered[FRAME_POINTER_REGNUM])
1327 SET_HARD_REG_BIT (eliminable_regset, FRAME_POINTER_REGNUM);
1329 SET_HARD_REG_BIT (ira_no_alloc_regs, FRAME_POINTER_REGNUM);
1332 error ("%s cannot be used in asm here", reg_names[FRAME_POINTER_REGNUM]);
1334 df_set_regs_ever_live (FRAME_POINTER_REGNUM, true);
1340 /* The length of the following two arrays. */
1341 int ira_reg_equiv_len;
1343 /* The element value is TRUE if the corresponding regno value is
1345 bool *ira_reg_equiv_invariant_p;
1347 /* The element value is equiv constant of given pseudo-register or
1349 rtx *ira_reg_equiv_const;
1351 /* Set up the two arrays declared above. */
1353 find_reg_equiv_invariant_const (void)
1357 rtx list, insn, note, constant, x;
1359 for (i = FIRST_PSEUDO_REGISTER; i < reg_equiv_init_size; i++)
1361 constant = NULL_RTX;
1362 invariant_p = false;
1363 for (list = reg_equiv_init[i]; list != NULL_RTX; list = XEXP (list, 1))
1365 insn = XEXP (list, 0);
1366 note = find_reg_note (insn, REG_EQUIV, NULL_RTX);
1368 if (note == NULL_RTX)
1373 if (! function_invariant_p (x)
1375 /* A function invariant is often CONSTANT_P but may
1376 include a register. We promise to only pass CONSTANT_P
1377 objects to LEGITIMATE_PIC_OPERAND_P. */
1378 || (CONSTANT_P (x) && LEGITIMATE_PIC_OPERAND_P (x)))
1380 /* It can happen that a REG_EQUIV note contains a MEM
1381 that is not a legitimate memory operand. As later
1382 stages of the reload assume that all addresses found
1383 in the reg_equiv_* arrays were originally legitimate,
1384 we ignore such REG_EQUIV notes. */
1385 if (memory_operand (x, VOIDmode))
1386 invariant_p = MEM_READONLY_P (x);
1387 else if (function_invariant_p (x))
1389 if (GET_CODE (x) == PLUS
1390 || x == frame_pointer_rtx || x == arg_pointer_rtx)
1397 ira_reg_equiv_invariant_p[i] = invariant_p;
1398 ira_reg_equiv_const[i] = constant;
1404 /* Set up REG_RENUMBER and CALLER_SAVE_NEEDED (used by reload) from
1405 the allocation found by IRA. */
1407 setup_reg_renumber (void)
1409 int regno, hard_regno;
1411 ira_allocno_iterator ai;
1413 caller_save_needed = 0;
1414 FOR_EACH_ALLOCNO (a, ai)
1416 /* There are no caps at this point. */
1417 ira_assert (ALLOCNO_CAP_MEMBER (a) == NULL);
1418 if (! ALLOCNO_ASSIGNED_P (a))
1419 /* It can happen if A is not referenced but partially anticipated
1420 somewhere in a region. */
1421 ALLOCNO_ASSIGNED_P (a) = true;
1422 ira_free_allocno_updated_costs (a);
1423 hard_regno = ALLOCNO_HARD_REGNO (a);
1424 regno = (int) REGNO (ALLOCNO_REG (a));
1425 reg_renumber[regno] = (hard_regno < 0 ? -1 : hard_regno);
1426 if (hard_regno >= 0 && ALLOCNO_CALLS_CROSSED_NUM (a) != 0
1427 && ! ira_hard_reg_not_in_set_p (hard_regno, ALLOCNO_MODE (a),
1430 ira_assert (!optimize || flag_caller_saves
1431 || regno >= ira_reg_equiv_len
1432 || ira_reg_equiv_const[regno]
1433 || ira_reg_equiv_invariant_p[regno]);
1434 caller_save_needed = 1;
1439 /* Set up allocno assignment flags for further allocation
1442 setup_allocno_assignment_flags (void)
1446 ira_allocno_iterator ai;
1448 FOR_EACH_ALLOCNO (a, ai)
1450 if (! ALLOCNO_ASSIGNED_P (a))
1451 /* It can happen if A is not referenced but partially anticipated
1452 somewhere in a region. */
1453 ira_free_allocno_updated_costs (a);
1454 hard_regno = ALLOCNO_HARD_REGNO (a);
1455 /* Don't assign hard registers to allocnos which are destination
1456 of removed store at the end of loop. It has no sense to keep
1457 the same value in different hard registers. It is also
1458 impossible to assign hard registers correctly to such
1459 allocnos because the cost info and info about intersected
1460 calls are incorrect for them. */
1461 ALLOCNO_ASSIGNED_P (a) = (hard_regno >= 0
1462 || ALLOCNO_MEM_OPTIMIZED_DEST_P (a)
1463 || (ALLOCNO_MEMORY_COST (a)
1464 - ALLOCNO_COVER_CLASS_COST (a)) < 0);
1465 ira_assert (hard_regno < 0
1466 || ! ira_hard_reg_not_in_set_p (hard_regno, ALLOCNO_MODE (a),
1468 [ALLOCNO_COVER_CLASS (a)]));
1472 /* Evaluate overall allocation cost and the costs for using hard
1473 registers and memory for allocnos. */
1475 calculate_allocation_cost (void)
1477 int hard_regno, cost;
1479 ira_allocno_iterator ai;
1481 ira_overall_cost = ira_reg_cost = ira_mem_cost = 0;
1482 FOR_EACH_ALLOCNO (a, ai)
1484 hard_regno = ALLOCNO_HARD_REGNO (a);
1485 ira_assert (hard_regno < 0
1486 || ! ira_hard_reg_not_in_set_p
1487 (hard_regno, ALLOCNO_MODE (a),
1488 reg_class_contents[ALLOCNO_COVER_CLASS (a)]));
1491 cost = ALLOCNO_MEMORY_COST (a);
1492 ira_mem_cost += cost;
1494 else if (ALLOCNO_HARD_REG_COSTS (a) != NULL)
1496 cost = (ALLOCNO_HARD_REG_COSTS (a)
1497 [ira_class_hard_reg_index
1498 [ALLOCNO_COVER_CLASS (a)][hard_regno]]);
1499 ira_reg_cost += cost;
1503 cost = ALLOCNO_COVER_CLASS_COST (a);
1504 ira_reg_cost += cost;
1506 ira_overall_cost += cost;
1509 if (internal_flag_ira_verbose > 0 && ira_dump_file != NULL)
1511 fprintf (ira_dump_file,
1512 "+++Costs: overall %d, reg %d, mem %d, ld %d, st %d, move %d\n",
1513 ira_overall_cost, ira_reg_cost, ira_mem_cost,
1514 ira_load_cost, ira_store_cost, ira_shuffle_cost);
1515 fprintf (ira_dump_file, "+++ move loops %d, new jumps %d\n",
1516 ira_move_loops_num, ira_additional_jumps_num);
1521 #ifdef ENABLE_IRA_CHECKING
1522 /* Check the correctness of the allocation. We do need this because
1523 of complicated code to transform more one region internal
1524 representation into one region representation. */
1526 check_allocation (void)
1528 ira_allocno_t a, conflict_a;
1529 int hard_regno, conflict_hard_regno, nregs, conflict_nregs;
1530 ira_allocno_conflict_iterator aci;
1531 ira_allocno_iterator ai;
1533 FOR_EACH_ALLOCNO (a, ai)
1535 if (ALLOCNO_CAP_MEMBER (a) != NULL
1536 || (hard_regno = ALLOCNO_HARD_REGNO (a)) < 0)
1538 nregs = hard_regno_nregs[hard_regno][ALLOCNO_MODE (a)];
1539 FOR_EACH_ALLOCNO_CONFLICT (a, conflict_a, aci)
1540 if ((conflict_hard_regno = ALLOCNO_HARD_REGNO (conflict_a)) >= 0)
1544 [conflict_hard_regno][ALLOCNO_MODE (conflict_a)]);
1545 if ((conflict_hard_regno <= hard_regno
1546 && hard_regno < conflict_hard_regno + conflict_nregs)
1547 || (hard_regno <= conflict_hard_regno
1548 && conflict_hard_regno < hard_regno + nregs))
1550 fprintf (stderr, "bad allocation for %d and %d\n",
1551 ALLOCNO_REGNO (a), ALLOCNO_REGNO (conflict_a));
1559 /* Fix values of array REG_EQUIV_INIT after live range splitting done
1562 fix_reg_equiv_init (void)
1564 int max_regno = max_reg_num ();
1566 rtx x, prev, next, insn, set;
1568 if (reg_equiv_init_size < max_regno)
1571 = (rtx *) ggc_realloc (reg_equiv_init, max_regno * sizeof (rtx));
1572 while (reg_equiv_init_size < max_regno)
1573 reg_equiv_init[reg_equiv_init_size++] = NULL_RTX;
1574 for (i = FIRST_PSEUDO_REGISTER; i < reg_equiv_init_size; i++)
1575 for (prev = NULL_RTX, x = reg_equiv_init[i]; x != NULL_RTX; x = next)
1579 set = single_set (insn);
1580 ira_assert (set != NULL_RTX
1581 && (REG_P (SET_DEST (set)) || REG_P (SET_SRC (set))));
1582 if (REG_P (SET_DEST (set))
1583 && ((int) REGNO (SET_DEST (set)) == i
1584 || (int) ORIGINAL_REGNO (SET_DEST (set)) == i))
1585 new_regno = REGNO (SET_DEST (set));
1586 else if (REG_P (SET_SRC (set))
1587 && ((int) REGNO (SET_SRC (set)) == i
1588 || (int) ORIGINAL_REGNO (SET_SRC (set)) == i))
1589 new_regno = REGNO (SET_SRC (set));
1596 if (prev == NULL_RTX)
1597 reg_equiv_init[i] = next;
1599 XEXP (prev, 1) = next;
1600 XEXP (x, 1) = reg_equiv_init[new_regno];
1601 reg_equiv_init[new_regno] = x;
1607 #ifdef ENABLE_IRA_CHECKING
1608 /* Print redundant memory-memory copies. */
1610 print_redundant_copies (void)
1614 ira_copy_t cp, next_cp;
1615 ira_allocno_iterator ai;
1617 FOR_EACH_ALLOCNO (a, ai)
1619 if (ALLOCNO_CAP_MEMBER (a) != NULL)
1622 hard_regno = ALLOCNO_HARD_REGNO (a);
1623 if (hard_regno >= 0)
1625 for (cp = ALLOCNO_COPIES (a); cp != NULL; cp = next_cp)
1627 next_cp = cp->next_first_allocno_copy;
1630 next_cp = cp->next_second_allocno_copy;
1631 if (internal_flag_ira_verbose > 4 && ira_dump_file != NULL
1632 && cp->insn != NULL_RTX
1633 && ALLOCNO_HARD_REGNO (cp->first) == hard_regno)
1634 fprintf (ira_dump_file,
1635 " Redundant move from %d(freq %d):%d\n",
1636 INSN_UID (cp->insn), cp->freq, hard_regno);
1642 /* Setup preferred and alternative classes for new pseudo-registers
1643 created by IRA starting with START. */
1645 setup_preferred_alternate_classes_for_new_pseudos (int start)
1648 int max_regno = max_reg_num ();
1650 for (i = start; i < max_regno; i++)
1652 old_regno = ORIGINAL_REGNO (regno_reg_rtx[i]);
1653 ira_assert (i != old_regno);
1654 setup_reg_classes (i, reg_preferred_class (old_regno),
1655 reg_alternate_class (old_regno));
1656 if (internal_flag_ira_verbose > 2 && ira_dump_file != NULL)
1657 fprintf (ira_dump_file,
1658 " New r%d: setting preferred %s, alternative %s\n",
1659 i, reg_class_names[reg_preferred_class (old_regno)],
1660 reg_class_names[reg_alternate_class (old_regno)]);
1666 /* Regional allocation can create new pseudo-registers. This function
1667 expands some arrays for pseudo-registers. */
1669 expand_reg_info (int old_size)
1672 int size = max_reg_num ();
1675 for (i = old_size; i < size; i++)
1677 reg_renumber[i] = -1;
1678 setup_reg_classes (i, GENERAL_REGS, ALL_REGS);
1684 /* This page contains code for sorting the insn chain used by reload.
1685 In the old register allocator, the insn chain order corresponds to
1686 the order of insns in RTL. By putting insns with higher execution
1687 frequency BBs first, reload has a better chance to generate less
1688 expensive operand reloads for such insns. */
1690 /* Map bb index -> order number in the BB chain in RTL code. */
1691 static int *basic_block_order_nums;
1693 /* Map chain insn uid -> order number in the insn chain before sorting
1695 static int *chain_insn_order;
1697 /* The function is used to sort insn chain according insn execution
1700 chain_freq_compare (const void *v1p, const void *v2p)
1702 const struct insn_chain *c1 = *(struct insn_chain * const *)v1p;
1703 const struct insn_chain *c2 = *(struct insn_chain * const *)v2p;
1706 diff = (BASIC_BLOCK (c2->block)->frequency
1707 - BASIC_BLOCK (c1->block)->frequency);
1710 /* Keep the same order in BB scope. */
1711 return (chain_insn_order[INSN_UID(c1->insn)]
1712 - chain_insn_order[INSN_UID(c2->insn)]);
1715 /* Sort the insn chain according insn original order. */
1717 chain_bb_compare (const void *v1p, const void *v2p)
1719 const struct insn_chain *c1 = *(struct insn_chain * const *)v1p;
1720 const struct insn_chain *c2 = *(struct insn_chain * const *)v2p;
1723 diff = (basic_block_order_nums[c1->block]
1724 - basic_block_order_nums[c2->block]);
1727 /* Keep the same order in BB scope. */
1728 return (chain_insn_order[INSN_UID(c1->insn)]
1729 - chain_insn_order[INSN_UID(c2->insn)]);
1732 /* Sort the insn chain according to insn frequencies if
1733 FREQ_P or according to insn original order otherwise. */
1735 ira_sort_insn_chain (bool freq_p)
1737 struct insn_chain *chain, **chain_arr;
1741 chain_insn_order = (int *) ira_allocate (get_max_uid () * sizeof (int));
1742 for (n = 0, chain = reload_insn_chain; chain != 0; chain = chain->next)
1744 chain_insn_order[INSN_UID (chain->insn)] = n;
1750 = (struct insn_chain **) ira_allocate (n * sizeof (struct insn_chain *));
1751 basic_block_order_nums
1752 = (int *) ira_allocate (sizeof (int) * last_basic_block);
1756 basic_block_order_nums[bb->index] = n++;
1758 for (n = 0, chain = reload_insn_chain; chain != 0; chain = chain->next)
1759 chain_arr[n++] = chain;
1760 qsort (chain_arr, n, sizeof (struct insn_chain *),
1761 freq_p ? chain_freq_compare : chain_bb_compare);
1762 ira_free (chain_insn_order);
1763 for (i = 1; i < n - 1; i++)
1765 chain_arr[i]->next = chain_arr[i + 1];
1766 chain_arr[i]->prev = chain_arr[i - 1];
1768 chain_arr[i]->next = NULL;
1769 chain_arr[i]->prev = chain_arr[i - 1];
1770 reload_insn_chain = chain_arr[0];
1771 reload_insn_chain->prev = NULL;
1772 reload_insn_chain->next = chain_arr[1];
1773 ira_free (basic_block_order_nums);
1774 ira_free (chain_arr);
1779 /* All natural loops. */
1780 struct loops ira_loops;
1782 /* This is the main entry of IRA. */
1786 int overall_cost_before, allocated_reg_info_size;
1788 int max_regno_before_ira, ira_max_point_before_emit;
1790 int saved_flag_ira_algorithm;
1793 timevar_push (TV_IRA);
1795 if (flag_ira_verbose < 10)
1797 internal_flag_ira_verbose = flag_ira_verbose;
1802 internal_flag_ira_verbose = flag_ira_verbose - 10;
1803 ira_dump_file = stderr;
1806 setup_prohibited_mode_move_regs ();
1808 df_note_add_problem ();
1812 df_live_add_problem ();
1813 df_live_set_all_dirty ();
1815 #ifdef ENABLE_CHECKING
1816 df->changeable_flags |= DF_VERIFY_SCHEDULED;
1819 df_clear_flags (DF_NO_INSN_RESCAN);
1820 regstat_init_n_sets_and_refs ();
1821 regstat_compute_ri ();
1823 /* If we are not optimizing, then this is the only place before
1824 register allocation where dataflow is done. And that is needed
1825 to generate these warnings. */
1827 generate_setjmp_warnings ();
1829 rebuild_p = update_equiv_regs ();
1831 #ifndef IRA_NO_OBSTACK
1832 gcc_obstack_init (&ira_obstack);
1834 bitmap_obstack_initialize (&ira_bitmap_obstack);
1837 max_regno = max_reg_num ();
1838 ira_reg_equiv_len = max_regno;
1839 ira_reg_equiv_invariant_p
1840 = (bool *) ira_allocate (max_regno * sizeof (bool));
1841 memset (ira_reg_equiv_invariant_p, 0, max_regno * sizeof (bool));
1842 ira_reg_equiv_const = (rtx *) ira_allocate (max_regno * sizeof (rtx));
1843 memset (ira_reg_equiv_const, 0, max_regno * sizeof (rtx));
1844 find_reg_equiv_invariant_const ();
1847 timevar_push (TV_JUMP);
1848 rebuild_jump_labels (get_insns ());
1849 purge_all_dead_edges ();
1850 timevar_pop (TV_JUMP);
1854 max_regno_before_ira = allocated_reg_info_size = max_reg_num ();
1855 allocate_reg_info ();
1856 setup_eliminable_regset ();
1858 ira_overall_cost = ira_reg_cost = ira_mem_cost = 0;
1859 ira_load_cost = ira_store_cost = ira_shuffle_cost = 0;
1860 ira_move_loops_num = ira_additional_jumps_num = 0;
1862 ira_assert (current_loops == NULL);
1863 flow_loops_find (&ira_loops);
1864 current_loops = &ira_loops;
1865 saved_flag_ira_algorithm = flag_ira_algorithm;
1866 if (optimize && number_of_loops () > (unsigned) IRA_MAX_LOOPS_NUM)
1867 flag_ira_algorithm = IRA_ALGORITHM_CB;
1869 if (internal_flag_ira_verbose > 0 && ira_dump_file != NULL)
1870 fprintf (ira_dump_file, "Building IRA IR\n");
1871 loops_p = ira_build (optimize
1872 && (flag_ira_algorithm == IRA_ALGORITHM_REGIONAL
1873 || flag_ira_algorithm == IRA_ALGORITHM_MIXED));
1877 ira_fast_allocation ();
1879 ira_max_point_before_emit = ira_max_point;
1885 max_regno = max_reg_num ();
1888 ira_initiate_assign ();
1891 expand_reg_info (allocated_reg_info_size);
1892 setup_preferred_alternate_classes_for_new_pseudos
1893 (allocated_reg_info_size);
1894 allocated_reg_info_size = max_regno;
1896 if (internal_flag_ira_verbose > 0 && ira_dump_file != NULL)
1897 fprintf (ira_dump_file, "Flattening IR\n");
1898 ira_flattening (max_regno_before_ira, ira_max_point_before_emit);
1899 /* New insns were generated: add notes and recalculate live
1903 flow_loops_find (&ira_loops);
1904 current_loops = &ira_loops;
1906 setup_allocno_assignment_flags ();
1907 ira_initiate_assign ();
1908 ira_reassign_conflict_allocnos (max_regno);
1912 setup_reg_renumber ();
1914 calculate_allocation_cost ();
1916 #ifdef ENABLE_IRA_CHECKING
1918 check_allocation ();
1921 delete_trivially_dead_insns (get_insns (), max_reg_num ());
1922 max_regno = max_reg_num ();
1924 /* Determine if the current function is a leaf before running IRA
1925 since this can impact optimizations done by the prologue and
1926 epilogue thus changing register elimination offsets. */
1927 current_function_is_leaf = leaf_function_p ();
1929 /* And the reg_equiv_memory_loc array. */
1930 VEC_safe_grow (rtx, gc, reg_equiv_memory_loc_vec, max_regno);
1931 memset (VEC_address (rtx, reg_equiv_memory_loc_vec), 0,
1932 sizeof (rtx) * max_regno);
1933 reg_equiv_memory_loc = VEC_address (rtx, reg_equiv_memory_loc_vec);
1935 if (max_regno != max_regno_before_ira)
1937 regstat_free_n_sets_and_refs ();
1939 regstat_init_n_sets_and_refs ();
1940 regstat_compute_ri ();
1943 allocate_initial_values (reg_equiv_memory_loc);
1945 overall_cost_before = ira_overall_cost;
1948 fix_reg_equiv_init ();
1950 #ifdef ENABLE_IRA_CHECKING
1951 print_redundant_copies ();
1954 ira_spilled_reg_stack_slots_num = 0;
1955 ira_spilled_reg_stack_slots
1956 = ((struct ira_spilled_reg_stack_slot *)
1957 ira_allocate (max_regno
1958 * sizeof (struct ira_spilled_reg_stack_slot)));
1959 memset (ira_spilled_reg_stack_slots, 0,
1960 max_regno * sizeof (struct ira_spilled_reg_stack_slot));
1963 timevar_pop (TV_IRA);
1965 timevar_push (TV_RELOAD);
1966 df_set_flags (DF_NO_INSN_RESCAN);
1967 build_insn_chain ();
1970 ira_sort_insn_chain (true);
1972 reload_completed = !reload (get_insns (), optimize > 0);
1974 timevar_pop (TV_RELOAD);
1976 timevar_push (TV_IRA);
1980 ira_free (ira_spilled_reg_stack_slots);
1982 ira_finish_assign ();
1985 if (internal_flag_ira_verbose > 0 && ira_dump_file != NULL
1986 && overall_cost_before != ira_overall_cost)
1987 fprintf (ira_dump_file, "+++Overall after reload %d\n", ira_overall_cost);
1990 flow_loops_free (&ira_loops);
1991 free_dominance_info (CDI_DOMINATORS);
1993 bb->loop_father = NULL;
1994 current_loops = NULL;
1996 flag_ira_algorithm = saved_flag_ira_algorithm;
1999 regstat_free_n_sets_and_refs ();
2003 cleanup_cfg (CLEANUP_EXPENSIVE);
2005 ira_free (ira_reg_equiv_invariant_p);
2006 ira_free (ira_reg_equiv_const);
2009 bitmap_obstack_release (&ira_bitmap_obstack);
2010 #ifndef IRA_NO_OBSTACK
2011 obstack_free (&ira_obstack, NULL);
2014 /* The code after the reload has changed so much that at this point
2015 we might as well just rescan everything. Not that
2016 df_rescan_all_insns is not going to help here because it does not
2017 touch the artificial uses and defs. */
2018 df_finish_pass (true);
2020 df_live_add_problem ();
2021 df_scan_alloc (NULL);
2027 timevar_pop (TV_IRA);
2035 return flag_ira != 0;
2038 /* Run the integrated register allocator. */
2040 rest_of_handle_ira (void)
2046 struct rtl_opt_pass pass_ira =
2051 gate_ira, /* gate */
2052 rest_of_handle_ira, /* execute */
2055 0, /* static_pass_number */
2057 0, /* properties_required */
2058 0, /* properties_provided */
2059 0, /* properties_destroyed */
2060 0, /* todo_flags_start */
2062 TODO_ggc_collect /* todo_flags_finish */