/* IRA hard register and memory cost calculation for allocnos or pseudos.
- Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012
- Free Software Foundation, Inc.
+ Copyright (C) 2006-2020 Free Software Foundation, Inc.
Contributed by Vladimir Makarov <vmakarov@redhat.com>.
This file is part of GCC.
#include "config.h"
#include "system.h"
#include "coretypes.h"
-#include "tm.h"
-#include "hard-reg-set.h"
+#include "backend.h"
+#include "target.h"
#include "rtl.h"
-#include "expr.h"
+#include "tree.h"
+#include "predict.h"
+#include "memmodel.h"
#include "tm_p.h"
-#include "flags.h"
-#include "basic-block.h"
+#include "insn-config.h"
#include "regs.h"
+#include "ira.h"
+#include "ira-int.h"
#include "addresses.h"
-#include "insn-config.h"
-#include "recog.h"
#include "reload.h"
-#include "diagnostic-core.h"
-#include "target.h"
-#include "params.h"
-#include "ira-int.h"
/* The flags is set up every time when we calculate pseudo register
classes through function ira_set_pseudo_classes. */
static struct costs *total_allocno_costs;
/* It is the current size of struct costs. */
-static int struct_costs_size;
+static size_t struct_costs_size;
/* Return pointer to structure containing costs of allocno or pseudo
with given NUM in array ARR. */
/* Container of the cost classes. */
enum reg_class classes[N_REG_CLASSES];
/* Map reg class -> index of the reg class in the previous array.
- -1 if it is not a cost classe. */
+ -1 if it is not a cost class. */
int index[N_REG_CLASSES];
/* Map hard regno index of first class in array CLASSES containing
the hard regno, -1 otherwise. */
/* Info about cost classes for each pseudo. */
static cost_classes_t *regno_cost_classes;
-/* Returns hash value for cost classes info V. */
-static hashval_t
-cost_classes_hash (const void *v)
+/* Helper for cost_classes hashing. */
+
+struct cost_classes_hasher : pointer_hash <cost_classes>
{
- const_cost_classes_t hv = (const_cost_classes_t) v;
+ static inline hashval_t hash (const cost_classes *);
+ static inline bool equal (const cost_classes *, const cost_classes *);
+ static inline void remove (cost_classes *);
+};
+/* Returns hash value for cost classes info HV. */
+inline hashval_t
+cost_classes_hasher::hash (const cost_classes *hv)
+{
return iterative_hash (&hv->classes, sizeof (enum reg_class) * hv->num, 0);
}
-/* Compares cost classes info V1 and V2. */
-static int
-cost_classes_eq (const void *v1, const void *v2)
+/* Compares cost classes info HV1 and HV2. */
+inline bool
+cost_classes_hasher::equal (const cost_classes *hv1, const cost_classes *hv2)
{
- const_cost_classes_t hv1 = (const_cost_classes_t) v1;
- const_cost_classes_t hv2 = (const_cost_classes_t) v2;
-
- return hv1->num == hv2->num && memcmp (hv1->classes, hv2->classes,
- sizeof (enum reg_class) * hv1->num);
+ return (hv1->num == hv2->num
+ && memcmp (hv1->classes, hv2->classes,
+ sizeof (enum reg_class) * hv1->num) == 0);
}
/* Delete cost classes info V from the hash table. */
-static void
-cost_classes_del (void *v)
+inline void
+cost_classes_hasher::remove (cost_classes *v)
{
ira_free (v);
}
/* Hash table of unique cost classes. */
-static htab_t cost_classes_htab;
+static hash_table<cost_classes_hasher> *cost_classes_htab;
/* Map allocno class -> cost classes for pseudo of given allocno
class. */
/* Map mode -> cost classes for pseudo of give mode. */
static cost_classes_t cost_classes_mode_cache[MAX_MACHINE_MODE];
+/* Cost classes that include all classes in ira_important_classes. */
+static cost_classes all_cost_classes;
+
+/* Use the array of classes in CLASSES_PTR to fill out the rest of
+ the structure. */
+static void
+complete_cost_classes (cost_classes_t classes_ptr)
+{
+ for (int i = 0; i < N_REG_CLASSES; i++)
+ classes_ptr->index[i] = -1;
+ for (int i = 0; i < FIRST_PSEUDO_REGISTER; i++)
+ classes_ptr->hard_regno_index[i] = -1;
+ for (int i = 0; i < classes_ptr->num; i++)
+ {
+ enum reg_class cl = classes_ptr->classes[i];
+ classes_ptr->index[cl] = i;
+ for (int j = ira_class_hard_regs_num[cl] - 1; j >= 0; j--)
+ {
+ unsigned int hard_regno = ira_class_hard_regs[cl][j];
+ if (classes_ptr->hard_regno_index[hard_regno] < 0)
+ classes_ptr->hard_regno_index[hard_regno] = i;
+ }
+ }
+}
+
/* Initialize info about the cost classes for each pseudo. */
static void
initiate_regno_cost_classes (void)
sizeof (cost_classes_t) * N_REG_CLASSES);
memset (cost_classes_mode_cache, 0,
sizeof (cost_classes_t) * MAX_MACHINE_MODE);
- cost_classes_htab
- = htab_create (200, cost_classes_hash, cost_classes_eq, cost_classes_del);
+ cost_classes_htab = new hash_table<cost_classes_hasher> (200);
+ all_cost_classes.num = ira_important_classes_num;
+ for (int i = 0; i < ira_important_classes_num; i++)
+ all_cost_classes.classes[i] = ira_important_classes[i];
+ complete_cost_classes (&all_cost_classes);
}
/* Create new cost classes from cost classes FROM and set up members
setup_cost_classes (cost_classes_t from)
{
cost_classes_t classes_ptr;
- enum reg_class cl;
- int i, j, hard_regno;
classes_ptr = (cost_classes_t) ira_allocate (sizeof (struct cost_classes));
classes_ptr->num = from->num;
- for (i = 0; i < N_REG_CLASSES; i++)
- classes_ptr->index[i] = -1;
- for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
- classes_ptr->hard_regno_index[i] = -1;
- for (i = 0; i < from->num; i++)
+ for (int i = 0; i < from->num; i++)
+ classes_ptr->classes[i] = from->classes[i];
+ complete_cost_classes (classes_ptr);
+ return classes_ptr;
+}
+
+/* Return a version of FULL that only considers registers in REGS that are
+ valid for mode MODE. Both FULL and the returned class are globally
+ allocated. */
+static cost_classes_t
+restrict_cost_classes (cost_classes_t full, machine_mode mode,
+ const_hard_reg_set regs)
+{
+ static struct cost_classes narrow;
+ int map[N_REG_CLASSES];
+ narrow.num = 0;
+ for (int i = 0; i < full->num; i++)
{
- cl = classes_ptr->classes[i] = from->classes[i];
- classes_ptr->index[cl] = i;
- for (j = ira_class_hard_regs_num[cl] - 1; j >= 0; j--)
+ /* Assume that we'll drop the class. */
+ map[i] = -1;
+
+ /* Ignore classes that are too small for the mode. */
+ enum reg_class cl = full->classes[i];
+ if (!contains_reg_of_mode[cl][mode])
+ continue;
+
+ /* Calculate the set of registers in CL that belong to REGS and
+ are valid for MODE. */
+ HARD_REG_SET valid_for_cl = reg_class_contents[cl] & regs;
+ valid_for_cl &= ~(ira_prohibited_class_mode_regs[cl][mode]
+ | ira_no_alloc_regs);
+ if (hard_reg_set_empty_p (valid_for_cl))
+ continue;
+
+ /* Don't use this class if the set of valid registers is a subset
+ of an existing class. For example, suppose we have two classes
+ GR_REGS and FR_REGS and a union class GR_AND_FR_REGS. Suppose
+ that the mode changes allowed by FR_REGS are not as general as
+ the mode changes allowed by GR_REGS.
+
+ In this situation, the mode changes for GR_AND_FR_REGS could
+ either be seen as the union or the intersection of the mode
+ changes allowed by the two subclasses. The justification for
+ the union-based definition would be that, if you want a mode
+ change that's only allowed by GR_REGS, you can pick a register
+ from the GR_REGS subclass. The justification for the
+ intersection-based definition would be that every register
+ from the class would allow the mode change.
+
+ However, if we have a register that needs to be in GR_REGS,
+ using GR_AND_FR_REGS with the intersection-based definition
+ would be too pessimistic, since it would bring in restrictions
+ that only apply to FR_REGS. Conversely, if we have a register
+ that needs to be in FR_REGS, using GR_AND_FR_REGS with the
+ union-based definition would lose the extra restrictions
+ placed on FR_REGS. GR_AND_FR_REGS is therefore only useful
+ for cases where GR_REGS and FP_REGS are both valid. */
+ int pos;
+ for (pos = 0; pos < narrow.num; ++pos)
{
- hard_regno = ira_class_hard_regs[cl][j];
- if (classes_ptr->hard_regno_index[hard_regno] < 0)
- classes_ptr->hard_regno_index[hard_regno] = i;
+ enum reg_class cl2 = narrow.classes[pos];
+ if (hard_reg_set_subset_p (valid_for_cl, reg_class_contents[cl2]))
+ break;
+ }
+ map[i] = pos;
+ if (pos == narrow.num)
+ {
+ /* If several classes are equivalent, prefer to use the one
+ that was chosen as the allocno class. */
+ enum reg_class cl2 = ira_allocno_class_translate[cl];
+ if (ira_class_hard_regs_num[cl] == ira_class_hard_regs_num[cl2])
+ cl = cl2;
+ narrow.classes[narrow.num++] = cl;
}
}
- return classes_ptr;
+ if (narrow.num == full->num)
+ return full;
+
+ cost_classes **slot = cost_classes_htab->find_slot (&narrow, INSERT);
+ if (*slot == NULL)
+ {
+ cost_classes_t classes = setup_cost_classes (&narrow);
+ /* Map equivalent classes to the representative that we chose above. */
+ for (int i = 0; i < ira_important_classes_num; i++)
+ {
+ enum reg_class cl = ira_important_classes[i];
+ int index = full->index[cl];
+ if (index >= 0)
+ classes->index[cl] = map[index];
+ }
+ *slot = classes;
+ }
+ return *slot;
}
/* Setup cost classes for pseudo REGNO whose allocno class is ACLASS.
cost_classes_t classes_ptr;
enum reg_class cl;
int i;
- PTR *slot;
+ cost_classes **slot;
HARD_REG_SET temp, temp2;
bool exclude_p;
if ((classes_ptr = cost_classes_aclass_cache[aclass]) == NULL)
{
- COPY_HARD_REG_SET (temp, reg_class_contents[aclass]);
- AND_COMPL_HARD_REG_SET (temp, ira_no_alloc_regs);
+ temp = reg_class_contents[aclass] & ~ira_no_alloc_regs;
/* We exclude classes from consideration which are subsets of
ACLASS only if ACLASS is an uniform class. */
exclude_p = ira_uniform_class_p[aclass];
{
/* Exclude non-uniform classes which are subsets of
ACLASS. */
- COPY_HARD_REG_SET (temp2, reg_class_contents[cl]);
- AND_COMPL_HARD_REG_SET (temp2, ira_no_alloc_regs);
+ temp2 = reg_class_contents[cl] & ~ira_no_alloc_regs;
if (hard_reg_set_subset_p (temp2, temp) && cl != aclass)
continue;
}
classes.classes[classes.num++] = cl;
}
- slot = htab_find_slot (cost_classes_htab, &classes, INSERT);
+ slot = cost_classes_htab->find_slot (&classes, INSERT);
if (*slot == NULL)
{
classes_ptr = setup_cost_classes (&classes);
}
classes_ptr = cost_classes_aclass_cache[aclass] = (cost_classes_t) *slot;
}
+ if (regno_reg_rtx[regno] != NULL_RTX)
+ {
+ /* Restrict the classes to those that are valid for REGNO's mode
+ (which might for example exclude singleton classes if the mode
+ requires two registers). Also restrict the classes to those that
+ are valid for subregs of REGNO. */
+ const HARD_REG_SET *valid_regs = valid_mode_changes_for_regno (regno);
+ if (!valid_regs)
+ valid_regs = ®_class_contents[ALL_REGS];
+ classes_ptr = restrict_cost_classes (classes_ptr,
+ PSEUDO_REGNO_MODE (regno),
+ *valid_regs);
+ }
regno_cost_classes[regno] = classes_ptr;
}
/* Setup cost classes for pseudo REGNO with MODE. Usage of MODE can
decrease number of cost classes for the pseudo, if hard registers
- of some important classes can not hold a value of MODE. So the
- pseudo can not get hard register of some important classes and cost
- calculation for such important classes is only waisting CPU
+ of some important classes cannot hold a value of MODE. So the
+ pseudo cannot get hard register of some important classes and cost
+ calculation for such important classes is only wasting CPU
time. */
static void
-setup_regno_cost_classes_by_mode (int regno, enum machine_mode mode)
+setup_regno_cost_classes_by_mode (int regno, machine_mode mode)
{
- static struct cost_classes classes;
- cost_classes_t classes_ptr;
- enum reg_class cl;
- int i;
- PTR *slot;
- HARD_REG_SET temp;
-
- if ((classes_ptr = cost_classes_mode_cache[mode]) == NULL)
+ if (const HARD_REG_SET *valid_regs = valid_mode_changes_for_regno (regno))
+ regno_cost_classes[regno] = restrict_cost_classes (&all_cost_classes,
+ mode, *valid_regs);
+ else
{
- classes.num = 0;
- for (i = 0; i < ira_important_classes_num; i++)
- {
- cl = ira_important_classes[i];
- COPY_HARD_REG_SET (temp, ira_prohibited_class_mode_regs[cl][mode]);
- IOR_HARD_REG_SET (temp, ira_no_alloc_regs);
- if (hard_reg_set_subset_p (reg_class_contents[cl], temp))
- continue;
- classes.classes[classes.num++] = cl;
- }
- slot = htab_find_slot (cost_classes_htab, &classes, INSERT);
- if (*slot == NULL)
- {
- classes_ptr = setup_cost_classes (&classes);
- *slot = classes_ptr;
- }
- else
- classes_ptr = (cost_classes_t) *slot;
- cost_classes_mode_cache[mode] = (cost_classes_t) *slot;
+ if (cost_classes_mode_cache[mode] == NULL)
+ cost_classes_mode_cache[mode]
+ = restrict_cost_classes (&all_cost_classes, mode,
+ reg_class_contents[ALL_REGS]);
+ regno_cost_classes[regno] = cost_classes_mode_cache[mode];
}
- regno_cost_classes[regno] = classes_ptr;
}
-/* Finilize info about the cost classes for each pseudo. */
+/* Finalize info about the cost classes for each pseudo. */
static void
finish_regno_cost_classes (void)
{
ira_free (regno_cost_classes);
- htab_delete (cost_classes_htab);
+ delete cost_classes_htab;
+ cost_classes_htab = NULL;
}
\f
TO_P is FALSE) a register of class RCLASS in mode MODE. X must not
be a pseudo register. */
static int
-copy_cost (rtx x, enum machine_mode mode, reg_class_t rclass, bool to_p,
+copy_cost (rtx x, machine_mode mode, reg_class_t rclass, bool to_p,
secondary_reload_info *prev_sri)
{
secondary_reload_info sri;
copy it. */
sri.prev_sri = prev_sri;
sri.extra_cost = 0;
+ /* PR 68770: Secondary reload might examine the t_icode field. */
+ sri.t_icode = CODE_FOR_nothing;
+
secondary_class = targetm.secondary_reload (to_p, x, rclass, mode, &sri);
if (secondary_class != NO_REGS)
the alternatives. */
static void
record_reg_classes (int n_alts, int n_ops, rtx *ops,
- enum machine_mode *modes, const char **constraints,
- rtx insn, enum reg_class *pref)
+ machine_mode *modes, const char **constraints,
+ rtx_insn *insn, enum reg_class *pref)
{
int alt;
int i, j, k;
- rtx set;
int insn_allows_mem[MAX_RECOG_OPERANDS];
+ move_table *move_in_cost, *move_out_cost;
+ short (*mem_cost)[2];
for (i = 0; i < n_ops; i++)
insn_allows_mem[i] = 0;
/* Process each alternative, each time minimizing an operand's cost
with the cost for each operand in that alternative. */
+ alternative_mask preferred = get_preferred_alternatives (insn);
for (alt = 0; alt < n_alts; alt++)
{
enum reg_class classes[MAX_RECOG_OPERANDS];
int alt_fail = 0;
int alt_cost = 0, op_cost_add;
- if (!recog_data.alternative_enabled_p[alt])
+ if (!TEST_BIT (preferred, alt))
{
for (i = 0; i < recog_data.n_operands; i++)
constraints[i] = skip_alternative (constraints[i]);
unsigned char c;
const char *p = constraints[i];
rtx op = ops[i];
- enum machine_mode mode = modes[i];
+ machine_mode mode = modes[i];
int allows_addr = 0;
int win = 0;
while (*p == '%' || *p == '=' || *p == '+' || *p == '&')
p++;
- if (p[0] >= '0' && p[0] <= '0' + i && (p[1] == ',' || p[1] == 0))
+ if (p[0] >= '0' && p[0] <= '0' + i)
{
/* Copy class and whether memory is allowed from the
matching alternative. Then perform any needed cost
bool in_p = recog_data.operand_type[i] != OP_OUT;
bool out_p = recog_data.operand_type[i] != OP_IN;
enum reg_class op_class = classes[i];
- move_table *move_in_cost, *move_out_cost;
ira_init_register_move_cost_if_necessary (mode);
if (! in_p)
{
ira_assert (out_p);
- move_out_cost = ira_may_move_out_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
{
- rclass = cost_classes[k];
- pp_costs[k]
- = move_out_cost[op_class][rclass] * frequency;
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = mem_cost[rclass][0] * frequency;
+ }
+ }
+ else
+ {
+ move_out_cost = ira_may_move_out_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k]
+ = move_out_cost[op_class][rclass] * frequency;
+ }
}
}
else if (! out_p)
{
ira_assert (in_p);
- move_in_cost = ira_may_move_in_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
+ {
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = mem_cost[rclass][1] * frequency;
+ }
+ }
+ else
{
- rclass = cost_classes[k];
- pp_costs[k]
- = move_in_cost[rclass][op_class] * frequency;
+ move_in_cost = ira_may_move_in_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k]
+ = move_in_cost[rclass][op_class] * frequency;
+ }
}
}
else
{
- move_in_cost = ira_may_move_in_cost[mode];
- move_out_cost = ira_may_move_out_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
{
- rclass = cost_classes[k];
- pp_costs[k] = ((move_in_cost[rclass][op_class]
- + move_out_cost[op_class][rclass])
- * frequency);
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = ((mem_cost[rclass][0]
+ + mem_cost[rclass][1])
+ * frequency);
+ }
+ }
+ else
+ {
+ move_in_cost = ira_may_move_in_cost[mode];
+ move_out_cost = ira_may_move_out_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = ((move_in_cost[rclass][op_class]
+ + move_out_cost[op_class][rclass])
+ * frequency);
+ }
}
}
&& ! find_reg_note (insn, REG_DEAD, op))
alt_cost += 2;
- /* This is in place of ordinary cost computation for
- this operand, so skip to the end of the
- alternative (should be just one character). */
- while (*p && *p++ != ',')
- ;
-
- constraints[i] = p;
- continue;
+ p++;
}
}
{
switch (c)
{
- case ',':
- break;
case '*':
/* Ignore the next letter for this pass. */
c = *++p;
break;
- case '?':
+ case '^':
alt_cost += 2;
- case '!': case '#': case '&':
- case '0': case '1': case '2': case '3': case '4':
- case '5': case '6': case '7': case '8': case '9':
- break;
-
- case 'p':
- allows_addr = 1;
- win = address_operand (op, GET_MODE (op));
- /* We know this operand is an address, so we want it
- to be allocated to a register that can be the
- base of an address, i.e. BASE_REG_CLASS. */
- classes[i]
- = ira_reg_class_subunion[classes[i]]
- [base_reg_class (VOIDmode, ADDR_SPACE_GENERIC,
- ADDRESS, SCRATCH)];
- break;
-
- case 'm': case 'o': case 'V':
- /* It doesn't seem worth distinguishing between
- offsettable and non-offsettable addresses
- here. */
- insn_allows_mem[i] = allows_mem[i] = 1;
- if (MEM_P (op))
- win = 1;
- break;
-
- case '<':
- if (MEM_P (op)
- && (GET_CODE (XEXP (op, 0)) == PRE_DEC
- || GET_CODE (XEXP (op, 0)) == POST_DEC))
- win = 1;
- break;
-
- case '>':
- if (MEM_P (op)
- && (GET_CODE (XEXP (op, 0)) == PRE_INC
- || GET_CODE (XEXP (op, 0)) == POST_INC))
- win = 1;
- break;
-
- case 'E':
- case 'F':
- if (CONST_DOUBLE_AS_FLOAT_P (op)
- || (GET_CODE (op) == CONST_VECTOR
- && (GET_MODE_CLASS (GET_MODE (op))
- == MODE_VECTOR_FLOAT)))
- win = 1;
break;
- case 'G':
- case 'H':
- if (CONST_DOUBLE_AS_FLOAT_P (op)
- && CONST_DOUBLE_OK_FOR_CONSTRAINT_P (op, c, p))
- win = 1;
- break;
-
- case 's':
- if (CONST_INT_P (op) || CONST_DOUBLE_AS_INT_P (op))
- break;
-
- case 'i':
- if (CONSTANT_P (op)
- && (! flag_pic || LEGITIMATE_PIC_OPERAND_P (op)))
- win = 1;
- break;
-
- case 'n':
- if (CONST_INT_P (op) || CONST_DOUBLE_AS_INT_P (op))
- win = 1;
- break;
-
- case 'I':
- case 'J':
- case 'K':
- case 'L':
- case 'M':
- case 'N':
- case 'O':
- case 'P':
- if (CONST_INT_P (op)
- && CONST_OK_FOR_CONSTRAINT_P (INTVAL (op), c, p))
- win = 1;
- break;
-
- case 'X':
- win = 1;
+ case '?':
+ alt_cost += 2;
break;
case 'g':
&& (! flag_pic || LEGITIMATE_PIC_OPERAND_P (op))))
win = 1;
insn_allows_mem[i] = allows_mem[i] = 1;
- case 'r':
classes[i] = ira_reg_class_subunion[classes[i]][GENERAL_REGS];
break;
default:
- if (REG_CLASS_FROM_CONSTRAINT (c, p) != NO_REGS)
- classes[i] = ira_reg_class_subunion[classes[i]]
- [REG_CLASS_FROM_CONSTRAINT (c, p)];
-#ifdef EXTRA_CONSTRAINT_STR
- else if (EXTRA_CONSTRAINT_STR (op, c, p))
- win = 1;
-
- if (EXTRA_MEMORY_CONSTRAINT (c, p))
+ enum constraint_num cn = lookup_constraint (p);
+ enum reg_class cl;
+ switch (get_constraint_type (cn))
{
+ case CT_REGISTER:
+ cl = reg_class_for_constraint (cn);
+ if (cl != NO_REGS)
+ classes[i] = ira_reg_class_subunion[classes[i]][cl];
+ break;
+
+ case CT_CONST_INT:
+ if (CONST_INT_P (op)
+ && insn_const_int_ok_for_constraint (INTVAL (op), cn))
+ win = 1;
+ break;
+
+ case CT_MEMORY:
/* Every MEM can be reloaded to fit. */
insn_allows_mem[i] = allows_mem[i] = 1;
if (MEM_P (op))
win = 1;
- }
- if (EXTRA_ADDRESS_CONSTRAINT (c, p))
- {
+ break;
+
+ case CT_SPECIAL_MEMORY:
+ insn_allows_mem[i] = allows_mem[i] = 1;
+ if (MEM_P (op) && constraint_satisfied_p (op, cn))
+ win = 1;
+ break;
+
+ case CT_ADDRESS:
/* Every address can be reloaded to fit. */
allows_addr = 1;
- if (address_operand (op, GET_MODE (op)))
+ if (address_operand (op, GET_MODE (op))
+ || constraint_satisfied_p (op, cn))
win = 1;
/* We know this operand is an address, so we
want it to be allocated to a hard register
= ira_reg_class_subunion[classes[i]]
[base_reg_class (VOIDmode, ADDR_SPACE_GENERIC,
ADDRESS, SCRATCH)];
+ break;
+
+ case CT_FIXED_FORM:
+ if (constraint_satisfied_p (op, cn))
+ win = 1;
+ break;
}
-#endif
break;
}
p += CONSTRAINT_LEN (c, p);
constraints[i] = p;
+ if (alt_fail)
+ break;
+
/* How we account for this operand now depends on whether it
is a pseudo register or not. If it is, we first check if
any register classes are valid. If not, we ignore this
into that class. */
if (REG_P (op) && REGNO (op) >= FIRST_PSEUDO_REGISTER)
{
- if (classes[i] == NO_REGS)
+ if (classes[i] == NO_REGS && ! allows_mem[i])
{
/* We must always fail if the operand is a REG, but
- we did not find a suitable class.
+ we did not find a suitable class and memory is
+ not allowed.
Otherwise we may perform an uninitialized read
from this_op_costs after the `continue' statement
bool in_p = recog_data.operand_type[i] != OP_OUT;
bool out_p = recog_data.operand_type[i] != OP_IN;
enum reg_class op_class = classes[i];
- move_table *move_in_cost, *move_out_cost;
ira_init_register_move_cost_if_necessary (mode);
if (! in_p)
{
ira_assert (out_p);
- move_out_cost = ira_may_move_out_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
+ {
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = mem_cost[rclass][0] * frequency;
+ }
+ }
+ else
{
- rclass = cost_classes[k];
- pp_costs[k]
- = move_out_cost[op_class][rclass] * frequency;
+ move_out_cost = ira_may_move_out_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k]
+ = move_out_cost[op_class][rclass] * frequency;
+ }
}
}
else if (! out_p)
{
ira_assert (in_p);
- move_in_cost = ira_may_move_in_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
+ {
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = mem_cost[rclass][1] * frequency;
+ }
+ }
+ else
{
- rclass = cost_classes[k];
- pp_costs[k]
- = move_in_cost[rclass][op_class] * frequency;
+ move_in_cost = ira_may_move_in_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k]
+ = move_in_cost[rclass][op_class] * frequency;
+ }
}
}
else
{
- move_in_cost = ira_may_move_in_cost[mode];
- move_out_cost = ira_may_move_out_cost[mode];
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ if (op_class == NO_REGS)
{
- rclass = cost_classes[k];
- pp_costs[k] = ((move_in_cost[rclass][op_class]
- + move_out_cost[op_class][rclass])
- * frequency);
+ mem_cost = ira_memory_move_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = ((mem_cost[rclass][0]
+ + mem_cost[rclass][1])
+ * frequency);
+ }
+ }
+ else
+ {
+ move_in_cost = ira_may_move_in_cost[mode];
+ move_out_cost = ira_may_move_out_cost[mode];
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ pp_costs[k] = ((move_in_cost[rclass][op_class]
+ + move_out_cost[op_class][rclass])
+ * frequency);
+ }
}
}
- /* If the alternative actually allows memory, make
- things a bit cheaper since we won't need an extra
- insn to load it. */
- pp->mem_cost
- = ((out_p ? ira_memory_move_cost[mode][op_class][0] : 0)
- + (in_p ? ira_memory_move_cost[mode][op_class][1] : 0)
- - allows_mem[i]) * frequency;
+ if (op_class == NO_REGS)
+ /* Although we don't need insn to reload from
+ memory, still accessing memory is usually more
+ expensive than a register. */
+ pp->mem_cost = frequency;
+ else
+ /* If the alternative actually allows memory, make
+ things a bit cheaper since we won't need an
+ extra insn to load it. */
+ pp->mem_cost
+ = ((out_p ? ira_memory_move_cost[mode][op_class][0] : 0)
+ + (in_p ? ira_memory_move_cost[mode][op_class][1] : 0)
+ - allows_mem[i]) * frequency;
/* If we have assigned a class to this allocno in
our first pass, add a cost to this alternative
corresponding to what we would add if this
enum reg_class pref_class = pref[COST_INDEX (REGNO (op))];
if (pref_class == NO_REGS)
+ {
+ if (op_class != NO_REGS)
+ alt_cost
+ += ((out_p
+ ? ira_memory_move_cost[mode][op_class][0]
+ : 0)
+ + (in_p
+ ? ira_memory_move_cost[mode][op_class][1]
+ : 0));
+ }
+ else if (op_class == NO_REGS)
alt_cost
+= ((out_p
- ? ira_memory_move_cost[mode][op_class][0] : 0)
+ ? ira_memory_move_cost[mode][pref_class][1]
+ : 0)
+ (in_p
- ? ira_memory_move_cost[mode][op_class][1]
+ ? ira_memory_move_cost[mode][pref_class][0]
: 0));
else if (ira_reg_class_intersect[pref_class][op_class]
== NO_REGS)
- alt_cost += ira_register_move_cost[mode][pref_class][op_class];
+ alt_cost += (ira_register_move_cost
+ [mode][pref_class][op_class]);
}
}
}
alt_cost += ira_memory_move_cost[mode][classes[i]][1];
else
alt_fail = 1;
+
+ if (alt_fail)
+ break;
}
if (alt_fail)
- continue;
+ {
+ /* The loop above might have exited early once the failure
+ was seen. Skip over the constraints for the remaining
+ operands. */
+ i += 1;
+ for (; i < n_ops; ++i)
+ constraints[i] = skip_alternative (constraints[i]);
+ continue;
+ }
op_cost_add = alt_cost * frequency;
/* Finally, update the costs with the information we've
ALLOCNO_BAD_SPILL_P (a) = true;
}
- /* If this insn is a single set copying operand 1 to operand 0 and
- one operand is an allocno with the other a hard reg or an allocno
- that prefers a hard register that is in its own register class
- then we may want to adjust the cost of that register class to -1.
-
- Avoid the adjustment if the source does not die to avoid
- stressing of register allocator by preferrencing two colliding
- registers into single class.
-
- Also avoid the adjustment if a copy between hard registers of the
- class is expensive (ten times the cost of a default copy is
- considered arbitrarily expensive). This avoids losing when the
- preferred class is very expensive as the source of a copy
- instruction. */
- if ((set = single_set (insn)) != 0
- && ops[0] == SET_DEST (set) && ops[1] == SET_SRC (set)
- && REG_P (ops[0]) && REG_P (ops[1])
- && find_regno_note (insn, REG_DEAD, REGNO (ops[1])))
- for (i = 0; i <= 1; i++)
- if (REGNO (ops[i]) >= FIRST_PSEUDO_REGISTER
- && REGNO (ops[!i]) < FIRST_PSEUDO_REGISTER)
- {
- unsigned int regno = REGNO (ops[i]);
- unsigned int other_regno = REGNO (ops[!i]);
- enum machine_mode mode = GET_MODE (ops[!i]);
- cost_classes_t cost_classes_ptr = regno_cost_classes[regno];
- enum reg_class *cost_classes = cost_classes_ptr->classes;
- reg_class_t rclass;
- int nr;
-
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
- {
- rclass = cost_classes[k];
- if (TEST_HARD_REG_BIT (reg_class_contents[rclass], other_regno)
- && (reg_class_size[(int) rclass]
- == ira_reg_class_max_nregs [(int) rclass][(int) mode]))
- {
- if (reg_class_size[rclass] == 1)
- op_costs[i]->cost[k] = -frequency;
- else
- {
- for (nr = 0;
- nr < hard_regno_nregs[other_regno][mode];
- nr++)
- if (! TEST_HARD_REG_BIT (reg_class_contents[rclass],
- other_regno + nr))
- break;
-
- if (nr == hard_regno_nregs[other_regno][mode])
- op_costs[i]->cost[k] = -frequency;
- }
- }
- }
- }
}
\f
pseudo-registers should count as OK. Arguments as for
regno_ok_for_base_p. */
static inline bool
-ok_for_base_p_nonstrict (rtx reg, enum machine_mode mode, addr_space_t as,
+ok_for_base_p_nonstrict (rtx reg, machine_mode mode, addr_space_t as,
enum rtx_code outer_code, enum rtx_code index_code)
{
unsigned regno = REGNO (reg);
SCALE is twice the amount to multiply the cost by (it is twice so
we can represent half-cost adjustments). */
static void
-record_address_regs (enum machine_mode mode, addr_space_t as, rtx x,
+record_address_regs (machine_mode mode, addr_space_t as, rtx x,
int context, enum rtx_code outer_code,
enum rtx_code index_code, int scale)
{
if (code1 == SUBREG)
arg1 = SUBREG_REG (arg1), code1 = GET_CODE (arg1);
- /* If this machine only allows one register per address, it
- must be in the first operand. */
- if (MAX_REGS_PER_ADDRESS == 1)
- record_address_regs (mode, as, arg0, 0, PLUS, code1, scale);
-
- /* If index and base registers are the same on this machine,
+ /* If index registers do not appear, or coincide with base registers,
just record registers in any non-constant operands. We
assume here, as well as in the tests below, that all
addresses are in canonical form. */
- else if (INDEX_REG_CLASS
- == base_reg_class (VOIDmode, as, PLUS, SCRATCH))
+ if (MAX_REGS_PER_ADDRESS == 1
+ || INDEX_REG_CLASS == base_reg_class (VOIDmode, as, PLUS, SCRATCH))
{
record_address_regs (mode, as, arg0, context, PLUS, code1, scale);
if (! CONSTANT_P (arg1))
/* If the second operand is a constant integer, it doesn't
change what class the first operand must be. */
- else if (code1 == CONST_INT || code1 == CONST_DOUBLE)
+ else if (CONST_SCALAR_INT_P (arg1))
record_address_regs (mode, as, arg0, context, PLUS, code1, scale);
/* If the second operand is a symbolic constant, the first
operand must be an index register. */
add_cost = (move_in_cost[i][rclass] * scale) / 2;
if (INT_MAX - add_cost < pp_costs[k])
pp_costs[k] = INT_MAX;
- else
+ else
pp_costs[k] += add_cost;
}
}
/* Calculate the costs of insn operands. */
static void
-record_operand_costs (rtx insn, enum reg_class *pref)
+record_operand_costs (rtx_insn *insn, enum reg_class *pref)
{
const char *constraints[MAX_RECOG_OPERANDS];
- enum machine_mode modes[MAX_RECOG_OPERANDS];
+ machine_mode modes[MAX_RECOG_OPERANDS];
+ rtx set;
int i;
+ if ((set = single_set (insn)) != NULL_RTX
+ /* In rare cases the single set insn might have less 2 operands
+ as the source can be a fixed special reg. */
+ && recog_data.n_operands > 1
+ && recog_data.operand[0] == SET_DEST (set)
+ && recog_data.operand[1] == SET_SRC (set))
+ {
+ int regno, other_regno;
+ rtx dest = SET_DEST (set);
+ rtx src = SET_SRC (set);
+
+ if (GET_CODE (dest) == SUBREG
+ && known_eq (GET_MODE_SIZE (GET_MODE (dest)),
+ GET_MODE_SIZE (GET_MODE (SUBREG_REG (dest)))))
+ dest = SUBREG_REG (dest);
+ if (GET_CODE (src) == SUBREG
+ && known_eq (GET_MODE_SIZE (GET_MODE (src)),
+ GET_MODE_SIZE (GET_MODE (SUBREG_REG (src)))))
+ src = SUBREG_REG (src);
+ if (REG_P (src) && REG_P (dest)
+ && (((regno = REGNO (src)) >= FIRST_PSEUDO_REGISTER
+ && (other_regno = REGNO (dest)) < FIRST_PSEUDO_REGISTER)
+ || ((regno = REGNO (dest)) >= FIRST_PSEUDO_REGISTER
+ && (other_regno = REGNO (src)) < FIRST_PSEUDO_REGISTER)))
+ {
+ machine_mode mode = GET_MODE (SET_SRC (set));
+ cost_classes_t cost_classes_ptr = regno_cost_classes[regno];
+ enum reg_class *cost_classes = cost_classes_ptr->classes;
+ reg_class_t rclass, hard_reg_class, pref_class, bigger_hard_reg_class;
+ int cost, k;
+ move_table *move_costs;
+ bool dead_p = find_regno_note (insn, REG_DEAD, REGNO (src));
+
+ ira_init_register_move_cost_if_necessary (mode);
+ move_costs = ira_register_move_cost[mode];
+ hard_reg_class = REGNO_REG_CLASS (other_regno);
+ bigger_hard_reg_class = ira_pressure_class_translate[hard_reg_class];
+ /* Target code may return any cost for mode which does not
+ fit the hard reg class (e.g. DImode for AREG on
+ i386). Check this and use a bigger class to get the
+ right cost. */
+ if (bigger_hard_reg_class != NO_REGS
+ && ! ira_hard_reg_in_set_p (other_regno, mode,
+ reg_class_contents[hard_reg_class]))
+ hard_reg_class = bigger_hard_reg_class;
+ i = regno == (int) REGNO (src) ? 1 : 0;
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ rclass = cost_classes[k];
+ cost = (i == 0
+ ? move_costs[hard_reg_class][rclass]
+ : move_costs[rclass][hard_reg_class]);
+
+ op_costs[i]->cost[k] = cost * frequency;
+ /* If we have assigned a class to this allocno in our
+ first pass, add a cost to this alternative
+ corresponding to what we would add if this allocno
+ were not in the appropriate class. */
+ if (pref)
+ {
+ if ((pref_class = pref[COST_INDEX (regno)]) == NO_REGS)
+ op_costs[i]->cost[k]
+ += ((i == 0 ? ira_memory_move_cost[mode][rclass][0] : 0)
+ + (i == 1 ? ira_memory_move_cost[mode][rclass][1] : 0)
+ * frequency);
+ else if (ira_reg_class_intersect[pref_class][rclass]
+ == NO_REGS)
+ op_costs[i]->cost[k]
+ += (move_costs[pref_class][rclass]
+ * frequency);
+ }
+ /* If this insn is a single set copying operand 1 to
+ operand 0 and one operand is an allocno with the
+ other a hard reg or an allocno that prefers a hard
+ register that is in its own register class then we
+ may want to adjust the cost of that register class to
+ -1.
+
+ Avoid the adjustment if the source does not die to
+ avoid stressing of register allocator by preferencing
+ two colliding registers into single class. */
+ if (dead_p
+ && TEST_HARD_REG_BIT (reg_class_contents[rclass], other_regno)
+ && (reg_class_size[(int) rclass]
+ == (ira_reg_class_max_nregs
+ [(int) rclass][(int) GET_MODE(src)])))
+ {
+ if (reg_class_size[rclass] == 1)
+ op_costs[i]->cost[k] = -frequency;
+ else if (in_hard_reg_set_p (reg_class_contents[rclass],
+ GET_MODE(src), other_regno))
+ op_costs[i]->cost[k] = -frequency;
+ }
+ }
+ op_costs[i]->mem_cost
+ = ira_memory_move_cost[mode][hard_reg_class][i] * frequency;
+ if (pref && (pref_class = pref[COST_INDEX (regno)]) != NO_REGS)
+ op_costs[i]->mem_cost
+ += ira_memory_move_cost[mode][pref_class][i] * frequency;
+ return;
+ }
+ }
+
for (i = 0; i < recog_data.n_operands; i++)
{
constraints[i] = recog_data.constraints[i];
XEXP (recog_data.operand[i], 0),
0, MEM, SCRATCH, frequency * 2);
else if (constraints[i][0] == 'p'
- || EXTRA_ADDRESS_CONSTRAINT (constraints[i][0],
- constraints[i]))
+ || (insn_extra_address_constraint
+ (lookup_constraint (constraints[i]))))
record_address_regs (VOIDmode, ADDR_SPACE_GENERIC,
recog_data.operand[i], 0, ADDRESS, SCRATCH,
frequency * 2);
}
-
+
/* Check for commutative in a separate loop so everything will have
been initialized. We must do this even if one operand is a
constant--see addsi3 in m68k.md. */
const char *xconstraints[MAX_RECOG_OPERANDS];
int j;
- /* Handle commutative operands by swapping the constraints.
- We assume the modes are the same. */
+ /* Handle commutative operands by swapping the
+ constraints. We assume the modes are the same. */
for (j = 0; j < recog_data.n_operands; j++)
xconstraints[j] = constraints[j];
/* Process one insn INSN. Scan it and record each time it would save
code to put a certain allocnos in a certain class. Return the last
insn processed, so that the scan can be continued from there. */
-static rtx
-scan_one_insn (rtx insn)
+static rtx_insn *
+scan_one_insn (rtx_insn *insn)
{
enum rtx_code pat_code;
rtx set, note;
return insn;
pat_code = GET_CODE (PATTERN (insn));
- if (pat_code == USE || pat_code == CLOBBER || pat_code == ASM_INPUT
- || pat_code == ADDR_VEC || pat_code == ADDR_DIFF_VEC)
+ if (pat_code == ASM_INPUT)
return insn;
+ /* If INSN is a USE/CLOBBER of a pseudo in a mode M then go ahead
+ and initialize the register move costs of mode M.
+
+ The pseudo may be related to another pseudo via a copy (implicit or
+ explicit) and if there are no mode M uses/sets of the original
+ pseudo, then we may leave the register move costs uninitialized for
+ mode M. */
+ if (pat_code == USE || pat_code == CLOBBER)
+ {
+ rtx x = XEXP (PATTERN (insn), 0);
+ if (GET_CODE (x) == REG
+ && REGNO (x) >= FIRST_PSEUDO_REGISTER
+ && have_regs_of_mode[GET_MODE (x)])
+ ira_init_register_move_cost_if_necessary (GET_MODE (x));
+ return insn;
+ }
+
counted_mem = false;
set = single_set (insn);
extract_insn (insn);
/* If this insn loads a parameter from its stack slot, then it
represents a savings, rather than a cost, if the parameter is
- stored in memory. Record this fact.
+ stored in memory. Record this fact.
Similarly if we're loading other constants from memory (constant
pool, TOC references, small data areas, etc) and this is the only
a memory requiring special instructions to load it, decreasing
mem_cost might result in it being loaded using the specialized
instruction into a register, then stored into stack and loaded
- again from the stack. See PR52208. */
+ again from the stack. See PR52208.
+
+ Don't do this if SET_SRC (set) has side effect. See PR56124. */
if (set != 0 && REG_P (SET_DEST (set)) && MEM_P (SET_SRC (set))
&& (note = find_reg_note (insn, REG_EQUIV, NULL_RTX)) != NULL_RTX
- && ((MEM_P (XEXP (note, 0)))
+ && ((MEM_P (XEXP (note, 0))
+ && !side_effects_p (SET_SRC (set)))
|| (CONSTANT_P (XEXP (note, 0))
&& targetm.legitimate_constant_p (GET_MODE (SET_DEST (set)),
XEXP (note, 0))
&& REG_N_SETS (REGNO (SET_DEST (set))) == 1))
- && general_operand (SET_SRC (set), GET_MODE (SET_SRC (set))))
+ && general_operand (SET_SRC (set), GET_MODE (SET_SRC (set)))
+ /* LRA does not use equiv with a symbol for PIC code. */
+ && (! ira_use_lra_p || ! pic_offset_table_rtx
+ || ! contains_symbol_ref_p (XEXP (note, 0))))
{
enum reg_class cl = GENERAL_REGS;
rtx reg = SET_DEST (set);
/* Now add the cost for each operand to the total costs for its
allocno. */
for (i = 0; i < recog_data.n_operands; i++)
- if (REG_P (recog_data.operand[i])
- && REGNO (recog_data.operand[i]) >= FIRST_PSEUDO_REGISTER)
- {
- int regno = REGNO (recog_data.operand[i]);
- struct costs *p = COSTS (costs, COST_INDEX (regno));
- struct costs *q = op_costs[i];
- int *p_costs = p->cost, *q_costs = q->cost;
- cost_classes_t cost_classes_ptr = regno_cost_classes[regno];
- int add_cost;
-
- /* If the already accounted for the memory "cost" above, don't
- do so again. */
- if (!counted_mem)
- {
- add_cost = q->mem_cost;
- if (add_cost > 0 && INT_MAX - add_cost < p->mem_cost)
- p->mem_cost = INT_MAX;
- else
- p->mem_cost += add_cost;
- }
- for (k = cost_classes_ptr->num - 1; k >= 0; k--)
- {
- add_cost = q_costs[k];
- if (add_cost > 0 && INT_MAX - add_cost < p_costs[k])
- p_costs[k] = INT_MAX;
- else
- p_costs[k] += add_cost;
- }
- }
-
+ {
+ rtx op = recog_data.operand[i];
+
+ if (GET_CODE (op) == SUBREG)
+ op = SUBREG_REG (op);
+ if (REG_P (op) && REGNO (op) >= FIRST_PSEUDO_REGISTER)
+ {
+ int regno = REGNO (op);
+ struct costs *p = COSTS (costs, COST_INDEX (regno));
+ struct costs *q = op_costs[i];
+ int *p_costs = p->cost, *q_costs = q->cost;
+ cost_classes_t cost_classes_ptr = regno_cost_classes[regno];
+ int add_cost;
+
+ /* If the already accounted for the memory "cost" above, don't
+ do so again. */
+ if (!counted_mem)
+ {
+ add_cost = q->mem_cost;
+ if (add_cost > 0 && INT_MAX - add_cost < p->mem_cost)
+ p->mem_cost = INT_MAX;
+ else
+ p->mem_cost += add_cost;
+ }
+ for (k = cost_classes_ptr->num - 1; k >= 0; k--)
+ {
+ add_cost = q_costs[k];
+ if (add_cost > 0 && INT_MAX - add_cost < p_costs[k])
+ p_costs[k] = INT_MAX;
+ else
+ p_costs[k] += add_cost;
+ }
+ }
+ }
return insn;
}
for (k = 0; k < cost_classes_ptr->num; k++)
{
rclass = cost_classes[k];
- if (contains_reg_of_mode[rclass][PSEUDO_REGNO_MODE (regno)]
-#ifdef CANNOT_CHANGE_MODE_CLASS
- && ! invalid_mode_change_p (regno, (enum reg_class) rclass)
-#endif
- )
- {
- fprintf (f, " %s:%d", reg_class_names[rclass],
- COSTS (costs, i)->cost[k]);
- if (flag_ira_region == IRA_REGION_ALL
- || flag_ira_region == IRA_REGION_MIXED)
- fprintf (f, ",%d", COSTS (total_allocno_costs, i)->cost[k]);
- }
+ fprintf (f, " %s:%d", reg_class_names[rclass],
+ COSTS (costs, i)->cost[k]);
+ if (flag_ira_region == IRA_REGION_ALL
+ || flag_ira_region == IRA_REGION_MIXED)
+ fprintf (f, ",%d", COSTS (total_allocno_costs, i)->cost[k]);
}
fprintf (f, " MEM:%i", COSTS (costs, i)->mem_cost);
if (flag_ira_region == IRA_REGION_ALL
for (k = 0; k < cost_classes_ptr->num; k++)
{
rclass = cost_classes[k];
- if (contains_reg_of_mode[rclass][PSEUDO_REGNO_MODE (regno)]
-#ifdef CANNOT_CHANGE_MODE_CLASS
- && ! invalid_mode_change_p (regno, (enum reg_class) rclass)
-#endif
- )
- fprintf (f, " %s:%d", reg_class_names[rclass],
- COSTS (costs, regno)->cost[k]);
+ fprintf (f, " %s:%d", reg_class_names[rclass],
+ COSTS (costs, regno)->cost[k]);
}
fprintf (f, " MEM:%i\n", COSTS (costs, regno)->mem_cost);
}
static void
process_bb_for_costs (basic_block bb)
{
- rtx insn;
+ rtx_insn *insn;
frequency = REG_FREQ_FROM_BB (bb);
if (frequency == 0)
int i, k, start, max_cost_classes_num;
int pass;
basic_block bb;
- enum reg_class *regno_best_class;
+ enum reg_class *regno_best_class, new_class;
init_recog ();
regno_best_class
{
basic_block bb;
- FOR_EACH_BB (bb)
+ FOR_EACH_BB_FN (bb, cfun)
process_bb_for_costs (bb);
}
int best_cost, allocno_cost;
enum reg_class best, alt_class;
cost_classes_t cost_classes_ptr = regno_cost_classes[i];
- enum reg_class *cost_classes = cost_classes_ptr->classes;
+ enum reg_class *cost_classes;
int *i_costs = temp_costs->cost;
int i_mem_cost;
int equiv_savings = regno_equiv_gains[i];
continue;
memcpy (temp_costs, COSTS (costs, i), struct_costs_size);
i_mem_cost = temp_costs->mem_cost;
+ cost_classes = cost_classes_ptr->classes;
}
else
{
continue;
memset (temp_costs, 0, struct_costs_size);
i_mem_cost = 0;
+ cost_classes = cost_classes_ptr->classes;
/* Find cost of all allocnos with the same regno. */
for (a = ira_regno_allocno_map[i];
a != NULL;
a = ALLOCNO_NEXT_REGNO_ALLOCNO (a))
{
int *a_costs, *p_costs;
-
+
a_num = ALLOCNO_NUM (a);
if ((flag_ira_region == IRA_REGION_ALL
|| flag_ira_region == IRA_REGION_MIXED)
for (k = 0; k < cost_classes_ptr->num; k++)
{
rclass = cost_classes[k];
- /* Ignore classes that are too small or invalid for this
- operand. */
- if (! contains_reg_of_mode[rclass][PSEUDO_REGNO_MODE (i)]
-#ifdef CANNOT_CHANGE_MODE_CLASS
- || invalid_mode_change_p (i, (enum reg_class) rclass)
-#endif
- )
- continue;
if (i_costs[k] < best_cost)
{
best_cost = i_costs[k];
alt_class = reg_class_subunion[alt_class][rclass];
}
alt_class = ira_allocno_class_translate[alt_class];
- if (best_cost > i_mem_cost)
+ if (best_cost > i_mem_cost
+ && ! non_spilled_static_chain_regno_p (i))
regno_aclass[i] = NO_REGS;
+ else if (!optimize && !targetm.class_likely_spilled_p (best))
+ /* Registers in the alternative class are likely to need
+ longer or slower sequences than registers in the best class.
+ When optimizing we make some effort to use the best class
+ over the alternative class where possible, but at -O0 we
+ effectively give the alternative class equal weight.
+ We then run the risk of using slower alternative registers
+ when plenty of registers from the best class are still free.
+ This is especially true because live ranges tend to be very
+ short in -O0 code and so register pressure tends to be low.
+
+ Avoid that by ignoring the alternative class if the best
+ class has plenty of registers.
+
+ The union class arrays give important classes and only
+ part of it are allocno classes. So translate them into
+ allocno classes. */
+ regno_aclass[i] = ira_allocno_class_translate[best];
else
{
/* Make the common class the biggest class of best and
- alt_class. */
- regno_aclass[i]
- = ira_reg_class_superunion[best][alt_class];
+ alt_class. Translate the common class into an
+ allocno class too. */
+ regno_aclass[i] = (ira_allocno_class_translate
+ [ira_reg_class_superunion[best][alt_class]]);
ira_assert (regno_aclass[i] != NO_REGS
&& ira_reg_allocno_class_p[regno_aclass[i]]);
}
+ if ((new_class
+ = (reg_class) (targetm.ira_change_pseudo_allocno_class
+ (i, regno_aclass[i], best))) != regno_aclass[i])
+ {
+ regno_aclass[i] = new_class;
+ if (hard_reg_set_subset_p (reg_class_contents[new_class],
+ reg_class_contents[best]))
+ best = new_class;
+ if (hard_reg_set_subset_p (reg_class_contents[new_class],
+ reg_class_contents[alt_class]))
+ alt_class = new_class;
+ }
if (pass == flag_expensive_optimizations)
{
- if (best_cost > i_mem_cost)
+ if (best_cost > i_mem_cost
+ /* Do not assign NO_REGS to static chain pointer
+ pseudo when non-local goto is used. */
+ && ! non_spilled_static_chain_regno_p (i))
best = alt_class = NO_REGS;
else if (best == alt_class)
alt_class = NO_REGS;
regno_best_class[i] = best;
if (! allocno_p)
{
- pref[i] = best_cost > i_mem_cost ? NO_REGS : best;
+ pref[i] = (best_cost > i_mem_cost
+ && ! non_spilled_static_chain_regno_p (i)
+ ? NO_REGS : best);
continue;
}
for (a = ira_regno_allocno_map[i];
a != NULL;
a = ALLOCNO_NEXT_REGNO_ALLOCNO (a))
{
- a_num = ALLOCNO_NUM (a);
- if (regno_aclass[i] == NO_REGS)
+ enum reg_class aclass = regno_aclass[i];
+ int a_num = ALLOCNO_NUM (a);
+ int *total_a_costs = COSTS (total_allocno_costs, a_num)->cost;
+ int *a_costs = COSTS (costs, a_num)->cost;
+
+ if (aclass == NO_REGS)
best = NO_REGS;
else
{
- int *total_a_costs = COSTS (total_allocno_costs, a_num)->cost;
- int *a_costs = COSTS (costs, a_num)->cost;
-
/* Finding best class which is subset of the common
class. */
best_cost = (1 << (HOST_BITS_PER_INT - 2)) - 1;
for (k = 0; k < cost_classes_ptr->num; k++)
{
rclass = cost_classes[k];
- if (! ira_class_subset_p[rclass][regno_aclass[i]])
+ if (! ira_class_subset_p[rclass][aclass])
continue;
- /* Ignore classes that are too small or invalid
- for this operand. */
- if (! contains_reg_of_mode[rclass][PSEUDO_REGNO_MODE (i)]
-#ifdef CANNOT_CHANGE_MODE_CLASS
- || invalid_mode_change_p (i, (enum reg_class) rclass)
-#endif
- )
- ;
- else if (total_a_costs[k] < best_cost)
+ if (total_a_costs[k] < best_cost)
{
best_cost = total_a_costs[k];
allocno_cost = a_costs[k];
ALLOCNO_LOOP_TREE_NODE (a)->loop_num);
fprintf (dump_file, ") best %s, allocno %s\n",
reg_class_names[best],
- reg_class_names[regno_aclass[i]]);
+ reg_class_names[aclass]);
}
pref[a_num] = best;
+ if (pass == flag_expensive_optimizations && best != aclass
+ && ira_class_hard_regs_num[best] > 0
+ && (ira_reg_class_max_nregs[best][ALLOCNO_MODE (a)]
+ >= ira_class_hard_regs_num[best]))
+ {
+ int ind = cost_classes_ptr->index[aclass];
+
+ ira_assert (ind >= 0);
+ ira_init_register_move_cost_if_necessary (ALLOCNO_MODE (a));
+ ira_add_allocno_pref (a, ira_class_hard_regs[best][0],
+ (a_costs[ind] - ALLOCNO_CLASS_COST (a))
+ / (ira_register_move_cost
+ [ALLOCNO_MODE (a)][best][aclass]));
+ for (k = 0; k < cost_classes_ptr->num; k++)
+ if (ira_class_subset_p[cost_classes[k]][best])
+ a_costs[k] = a_costs[ind];
+ }
}
}
-
+
if (internal_flag_ira_verbose > 4 && dump_file)
{
if (allocno_p)
/* Process moves involving hard regs to modify allocno hard register
costs. We can do this only after determining allocno class. If a
- hard register forms a register class, than moves with the hard
+ hard register forms a register class, then moves with the hard
register are already taken into account in class costs for the
allocno. */
static void
process_bb_node_for_hard_reg_moves (ira_loop_tree_node_t loop_tree_node)
{
- int i, freq, cost, src_regno, dst_regno, hard_regno;
+ int i, freq, src_regno, dst_regno, hard_regno, a_regno;
bool to_p;
- ira_allocno_t a;
- enum reg_class rclass, hard_reg_class;
- enum machine_mode mode;
+ ira_allocno_t a, curr_a;
+ ira_loop_tree_node_t curr_loop_tree_node;
+ enum reg_class rclass;
basic_block bb;
- rtx insn, set, src, dst;
+ rtx_insn *insn;
+ rtx set, src, dst;
bb = loop_tree_node->bb;
if (bb == NULL)
&& src_regno < FIRST_PSEUDO_REGISTER)
{
hard_regno = src_regno;
- to_p = true;
a = ira_curr_regno_allocno_map[dst_regno];
+ to_p = true;
}
else if (src_regno >= FIRST_PSEUDO_REGISTER
&& dst_regno < FIRST_PSEUDO_REGISTER)
{
hard_regno = dst_regno;
- to_p = false;
a = ira_curr_regno_allocno_map[src_regno];
+ to_p = false;
}
else
continue;
+ if (reg_class_size[(int) REGNO_REG_CLASS (hard_regno)]
+ == (ira_reg_class_max_nregs
+ [REGNO_REG_CLASS (hard_regno)][(int) ALLOCNO_MODE(a)]))
+ /* If the class can provide only one hard reg to the allocno,
+ we processed the insn record_operand_costs already and we
+ actually updated the hard reg cost there. */
+ continue;
rclass = ALLOCNO_CLASS (a);
if (! TEST_HARD_REG_BIT (reg_class_contents[rclass], hard_regno))
continue;
i = ira_class_hard_reg_index[rclass][hard_regno];
if (i < 0)
continue;
- mode = ALLOCNO_MODE (a);
- hard_reg_class = REGNO_REG_CLASS (hard_regno);
- ira_init_register_move_cost_if_necessary (mode);
- cost
- = (to_p ? ira_register_move_cost[mode][hard_reg_class][rclass]
- : ira_register_move_cost[mode][rclass][hard_reg_class]) * freq;
- ira_allocate_and_set_costs (&ALLOCNO_HARD_REG_COSTS (a), rclass,
- ALLOCNO_CLASS_COST (a));
- ira_allocate_and_set_costs (&ALLOCNO_CONFLICT_HARD_REG_COSTS (a),
- rclass, 0);
- ALLOCNO_HARD_REG_COSTS (a)[i] -= cost;
- ALLOCNO_CONFLICT_HARD_REG_COSTS (a)[i] -= cost;
- ALLOCNO_CLASS_COST (a) = MIN (ALLOCNO_CLASS_COST (a),
- ALLOCNO_HARD_REG_COSTS (a)[i]);
+ a_regno = ALLOCNO_REGNO (a);
+ for (curr_loop_tree_node = ALLOCNO_LOOP_TREE_NODE (a);
+ curr_loop_tree_node != NULL;
+ curr_loop_tree_node = curr_loop_tree_node->parent)
+ if ((curr_a = curr_loop_tree_node->regno_allocno_map[a_regno]) != NULL)
+ ira_add_allocno_pref (curr_a, hard_regno, freq);
+ {
+ int cost;
+ enum reg_class hard_reg_class;
+ machine_mode mode;
+
+ mode = ALLOCNO_MODE (a);
+ hard_reg_class = REGNO_REG_CLASS (hard_regno);
+ ira_init_register_move_cost_if_necessary (mode);
+ cost = (to_p ? ira_register_move_cost[mode][hard_reg_class][rclass]
+ : ira_register_move_cost[mode][rclass][hard_reg_class]) * freq;
+ ira_allocate_and_set_costs (&ALLOCNO_HARD_REG_COSTS (a), rclass,
+ ALLOCNO_CLASS_COST (a));
+ ira_allocate_and_set_costs (&ALLOCNO_CONFLICT_HARD_REG_COSTS (a),
+ rclass, 0);
+ ALLOCNO_HARD_REG_COSTS (a)[i] -= cost;
+ ALLOCNO_CONFLICT_HARD_REG_COSTS (a)[i] -= cost;
+ ALLOCNO_CLASS_COST (a) = MIN (ALLOCNO_CLASS_COST (a),
+ ALLOCNO_HARD_REG_COSTS (a)[i]);
+ }
}
}
}
/* Free allocated temporary cost vectors. */
-static void
-free_ira_costs (void)
+void
+target_ira_int::free_ira_costs ()
{
int i;
- free (init_cost);
- init_cost = NULL;
+ free (x_init_cost);
+ x_init_cost = NULL;
for (i = 0; i < MAX_RECOG_OPERANDS; i++)
{
- free (op_costs[i]);
- free (this_op_costs[i]);
- op_costs[i] = this_op_costs[i] = NULL;
+ free (x_op_costs[i]);
+ free (x_this_op_costs[i]);
+ x_op_costs[i] = x_this_op_costs[i] = NULL;
}
- free (temp_costs);
- temp_costs = NULL;
+ free (x_temp_costs);
+ x_temp_costs = NULL;
}
/* This is called each time register related information is
{
int i;
- free_ira_costs ();
+ this_target_ira_int->free_ira_costs ();
max_struct_costs_size
= sizeof (struct costs) + sizeof (int) * (ira_important_classes_num - 1);
/* Don't use ira_allocate because vectors live through several IRA
temp_costs = (struct costs *) xmalloc (max_struct_costs_size);
}
-/* Function called once at the end of compiler work. */
-void
-ira_finish_costs_once (void)
-{
- free_ira_costs ();
-}
-
\f
/* Common initialization function for ira_costs and
ira_free (total_allocno_costs);
}
-/* Entry function which defines classes for pseudos. */
+/* Entry function which defines classes for pseudos.
+ Set pseudo_classes_defined_p only if DEFINE_PSEUDO_CLASSES is true. */
void
-ira_set_pseudo_classes (FILE *dump_file)
+ira_set_pseudo_classes (bool define_pseudo_classes, FILE *dump_file)
{
allocno_p = false;
internal_flag_ira_verbose = flag_ira_verbose;
initiate_regno_cost_classes ();
find_costs_and_classes (dump_file);
finish_regno_cost_classes ();
- pseudo_classes_defined_p = true;
+ if (define_pseudo_classes)
+ pseudo_classes_defined_p = true;
+
finish_costs ();
}
int j, n, regno;
int cost, min_cost, *reg_costs;
enum reg_class aclass, rclass;
- enum machine_mode mode;
+ machine_mode mode;
ira_allocno_t a;
ira_allocno_iterator ai;
ira_allocno_object_iterator oi;
continue;
rclass = REGNO_REG_CLASS (regno);
cost = 0;
- if (ira_hard_reg_set_intersection_p (regno, mode, call_used_reg_set)
- || HARD_REGNO_CALL_PART_CLOBBERED (regno, mode))
+ if (ira_need_caller_save_p (a, regno))
cost += (ALLOCNO_CALL_FREQ (a)
* (ira_memory_move_cost[mode][rclass][0]
+ ira_memory_move_cost[mode][rclass][1]));
else
regno_equiv_gains[regno] += cost;
}
+
+void
+ira_costs_c_finalize (void)
+{
+ this_target_ira_int->free_ira_costs ();
+}