/* Induction variable optimizations.
- Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009
+ Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
Free Software Foundation, Inc.
-
+
This file is part of GCC.
-
+
GCC is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 3, or (at your option) any
later version.
-
+
GCC is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
-
+
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
4) The trees are transformed to use the new variables, the dead code is
removed.
-
+
All of this is done loop by loop. Doing it globally is theoretically
possible, it might give a better performance and it might enable us
to decide costs more precisely, but getting all the interactions right
#include "tree-dump.h"
#include "timevar.h"
#include "cfgloop.h"
-#include "varray.h"
#include "expr.h"
#include "tree-pass.h"
#include "ggc.h"
struct iv *iv; /* Induction variable description. */
bool has_nonlin_use; /* For a loop-level invariant, whether it is used in
an expression that is not an induction variable. */
- unsigned inv_id; /* Id of an invariant. */
bool preserve_biv; /* For the original biv, whether to preserve it. */
+ unsigned inv_id; /* Id of an invariant. */
};
/* Types of uses. */
unsigned id; /* The number of the candidate. */
bool important; /* Whether this is an "important" candidate, i.e. such
that it should be considered by all uses. */
- enum iv_position pos; /* Where it is computed. */
+ ENUM_BITFIELD(iv_position) pos : 8; /* Where it is computed. */
gimple incremented_at;/* For original biv, the statement where it is
incremented. */
tree var_before; /* The variable used for it before increment. */
print_generic_expr (dump_file, niter, TDF_SLIM);
fprintf (dump_file, "\n\n");
};
-
+
fprintf (dump_file, "Induction variables:\n\n");
EXECUTE_IF_SET_IN_BITMAP (data->relevant, 0, i, bi)
iv = get_iv (data, op);
if (!iv)
return NULL;
-
+
if (iv->have_use_for)
{
use = iv_use (data, iv->use_id);
|| bitpos % GET_MODE_ALIGNMENT (mode) != 0
|| bitpos % BITS_PER_UNIT != 0)
return true;
-
+
if (!constant_multiple_of (step, al, &mul))
return true;
}
while (handled_component_p (*ref))
ref = &TREE_OPERAND (*ref, 0);
if (TREE_CODE (*ref) == INDIRECT_REF)
- *ref = fold_indirect_ref (*ref);
+ {
+ tree tem = gimple_fold_indirect_ref (TREE_OPERAND (*ref, 0));
+ if (tem)
+ *ref = tem;
+ }
}
}
unsigned i;
struct iv_cand *cand = NULL;
tree type, orig_type;
-
+
if (base)
{
orig_type = TREE_TYPE (base);
it. The candidate computation is scheduled on all available positions. */
static void
-add_candidate (struct ivopts_data *data,
+add_candidate (struct ivopts_data *data,
tree base, tree step, bool important, struct iv_use *use)
{
if (ip_normal_pos (data->current_loop))
return ret;
}
-
+
/* n_map_members is a power of two, so this computes modulo. */
s = cand->id & (use->n_map_members - 1);
for (i = s; i < use->n_map_members; i++)
static rtx
produce_memory_decl_rtl (tree obj, int *regno)
{
+ addr_space_t as = TYPE_ADDR_SPACE (TREE_TYPE (obj));
+ enum machine_mode address_mode = targetm.addr_space.address_mode (as);
rtx x;
-
+
gcc_assert (obj);
if (TREE_STATIC (obj) || DECL_EXTERNAL (obj))
{
const char *name = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (obj));
- x = gen_rtx_SYMBOL_REF (Pmode, name);
+ x = gen_rtx_SYMBOL_REF (address_mode, name);
SET_SYMBOL_REF_DECL (x, obj);
x = gen_rtx_MEM (DECL_MODE (obj), x);
+ set_mem_addr_space (x, as);
targetm.encode_section_info (obj, x, true);
}
else
{
- x = gen_raw_REG (Pmode, (*regno)++);
+ x = gen_raw_REG (address_mode, (*regno)++);
x = gen_rtx_MEM (DECL_MODE (obj), x);
+ set_mem_addr_space (x, as);
}
return x;
unsigned cost;
/* Avoid using hard regs in ways which may be unsupported. */
int regno = LAST_VIRTUAL_REGISTER + 1;
- enum function_frequency real_frequency = cfun->function_frequency;
+ struct cgraph_node *node = cgraph_node (current_function_decl);
+ enum node_frequency real_frequency = node->frequency;
- cfun->function_frequency = FUNCTION_FREQUENCY_NORMAL;
+ node->frequency = NODE_FREQUENCY_NORMAL;
crtl->maybe_hot_insn_p = speed;
walk_tree (&expr, prepare_decl_rtl, ®no, NULL);
start_sequence ();
seq = get_insns ();
end_sequence ();
default_rtl_profile ();
- cfun->function_frequency = real_frequency;
+ node->frequency = real_frequency;
cost = seq_cost (seq, speed);
if (MEM_P (rslt))
- cost += address_cost (XEXP (rslt, 0), TYPE_MODE (type), speed);
+ cost += address_cost (XEXP (rslt, 0), TYPE_MODE (type),
+ TYPE_ADDR_SPACE (type), speed);
return cost;
}
if (stmt_after_increment (loop, cand, at))
{
aff_tree cstep_aff;
-
+
if (common_type != uutype)
cstep_common = fold_convert (common_type, cstep);
else
cost = 1;
costs[mode] = cost;
-
+
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Addition in %s costs %d\n",
GET_MODE_NAME (mode), cost);
gen_int_mode (cst, mode), NULL_RTX, 0);
seq = get_insns ();
end_sequence ();
-
+
cost = seq_cost (seq, speed);
if (dump_file && (dump_flags & TDF_DETAILS))
}
/* Returns true if multiplying by RATIO is allowed in an address. Test the
- validity for a memory reference accessing memory of mode MODE. */
+ validity for a memory reference accessing memory of mode MODE in
+ address space AS. */
+
+DEF_VEC_P (sbitmap);
+DEF_VEC_ALLOC_P (sbitmap, heap);
bool
-multiplier_allowed_in_address_p (HOST_WIDE_INT ratio, enum machine_mode mode)
+multiplier_allowed_in_address_p (HOST_WIDE_INT ratio, enum machine_mode mode,
+ addr_space_t as)
{
#define MAX_RATIO 128
- static sbitmap valid_mult[MAX_MACHINE_MODE];
-
- if (!valid_mult[mode])
+ unsigned int data_index = (int) as * MAX_MACHINE_MODE + (int) mode;
+ static VEC (sbitmap, heap) *valid_mult_list;
+ sbitmap valid_mult;
+
+ if (data_index >= VEC_length (sbitmap, valid_mult_list))
+ VEC_safe_grow_cleared (sbitmap, heap, valid_mult_list, data_index + 1);
+
+ valid_mult = VEC_index (sbitmap, valid_mult_list, data_index);
+ if (!valid_mult)
{
- rtx reg1 = gen_raw_REG (Pmode, LAST_VIRTUAL_REGISTER + 1);
+ enum machine_mode address_mode = targetm.addr_space.address_mode (as);
+ rtx reg1 = gen_raw_REG (address_mode, LAST_VIRTUAL_REGISTER + 1);
rtx addr;
HOST_WIDE_INT i;
- valid_mult[mode] = sbitmap_alloc (2 * MAX_RATIO + 1);
- sbitmap_zero (valid_mult[mode]);
- addr = gen_rtx_fmt_ee (MULT, Pmode, reg1, NULL_RTX);
+ valid_mult = sbitmap_alloc (2 * MAX_RATIO + 1);
+ sbitmap_zero (valid_mult);
+ addr = gen_rtx_fmt_ee (MULT, address_mode, reg1, NULL_RTX);
for (i = -MAX_RATIO; i <= MAX_RATIO; i++)
{
- XEXP (addr, 1) = gen_int_mode (i, Pmode);
- if (memory_address_p (mode, addr))
- SET_BIT (valid_mult[mode], i + MAX_RATIO);
+ XEXP (addr, 1) = gen_int_mode (i, address_mode);
+ if (memory_address_addr_space_p (mode, addr, as))
+ SET_BIT (valid_mult, i + MAX_RATIO);
}
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " allowed multipliers:");
for (i = -MAX_RATIO; i <= MAX_RATIO; i++)
- if (TEST_BIT (valid_mult[mode], i + MAX_RATIO))
+ if (TEST_BIT (valid_mult, i + MAX_RATIO))
fprintf (dump_file, " %d", (int) i);
fprintf (dump_file, "\n");
fprintf (dump_file, "\n");
}
+
+ VEC_replace (sbitmap, valid_mult_list, data_index, valid_mult);
}
if (ratio > MAX_RATIO || ratio < -MAX_RATIO)
return false;
- return TEST_BIT (valid_mult[mode], ratio + MAX_RATIO);
+ return TEST_BIT (valid_mult, ratio + MAX_RATIO);
}
/* Returns cost of address in shape symbol + var + OFFSET + RATIO * index.
If SYMBOL_PRESENT is false, symbol is omitted. If VAR_PRESENT is false,
variable is omitted. Compute the cost for a memory reference that accesses
- a memory location of mode MEM_MODE.
+ a memory location of mode MEM_MODE in address space AS.
MAY_AUTOINC is set to true if the autoincrement (increasing index by
size of MEM_MODE / RATIO) is available. To make this determination, we
TODO -- there must be some better way. This all is quite crude. */
+typedef struct
+{
+ HOST_WIDE_INT min_offset, max_offset;
+ unsigned costs[2][2][2][2];
+} *address_cost_data;
+
+DEF_VEC_P (address_cost_data);
+DEF_VEC_ALLOC_P (address_cost_data, heap);
+
static comp_cost
get_address_cost (bool symbol_present, bool var_present,
unsigned HOST_WIDE_INT offset, HOST_WIDE_INT ratio,
- HOST_WIDE_INT cstep, enum machine_mode mem_mode, bool speed,
+ HOST_WIDE_INT cstep, enum machine_mode mem_mode,
+ addr_space_t as, bool speed,
bool stmt_after_inc, bool *may_autoinc)
{
- static bool initialized[MAX_MACHINE_MODE];
- static HOST_WIDE_INT rat[MAX_MACHINE_MODE], off[MAX_MACHINE_MODE];
- static HOST_WIDE_INT min_offset[MAX_MACHINE_MODE], max_offset[MAX_MACHINE_MODE];
- static unsigned costs[MAX_MACHINE_MODE][2][2][2][2];
+ enum machine_mode address_mode = targetm.addr_space.address_mode (as);
+ static VEC(address_cost_data, heap) *address_cost_data_list;
+ unsigned int data_index = (int) as * MAX_MACHINE_MODE + (int) mem_mode;
+ address_cost_data data;
static bool has_preinc[MAX_MACHINE_MODE], has_postinc[MAX_MACHINE_MODE];
static bool has_predec[MAX_MACHINE_MODE], has_postdec[MAX_MACHINE_MODE];
unsigned cost, acost, complexity;
unsigned HOST_WIDE_INT mask;
unsigned bits;
- if (!initialized[mem_mode])
+ if (data_index >= VEC_length (address_cost_data, address_cost_data_list))
+ VEC_safe_grow_cleared (address_cost_data, heap, address_cost_data_list,
+ data_index + 1);
+
+ data = VEC_index (address_cost_data, address_cost_data_list, data_index);
+ if (!data)
{
HOST_WIDE_INT i;
HOST_WIDE_INT start = BIGGEST_ALIGNMENT / BITS_PER_UNIT;
+ HOST_WIDE_INT rat, off;
int old_cse_not_expected;
unsigned sym_p, var_p, off_p, rat_p, add_c;
rtx seq, addr, base;
rtx reg0, reg1;
- initialized[mem_mode] = true;
+ data = (address_cost_data) xcalloc (1, sizeof (*data));
- reg1 = gen_raw_REG (Pmode, LAST_VIRTUAL_REGISTER + 1);
+ reg1 = gen_raw_REG (address_mode, LAST_VIRTUAL_REGISTER + 1);
- addr = gen_rtx_fmt_ee (PLUS, Pmode, reg1, NULL_RTX);
+ addr = gen_rtx_fmt_ee (PLUS, address_mode, reg1, NULL_RTX);
for (i = start; i <= 1 << 20; i <<= 1)
{
- XEXP (addr, 1) = gen_int_mode (i, Pmode);
- if (!memory_address_p (mem_mode, addr))
+ XEXP (addr, 1) = gen_int_mode (i, address_mode);
+ if (!memory_address_addr_space_p (mem_mode, addr, as))
break;
}
- max_offset[mem_mode] = i == start ? 0 : i >> 1;
- off[mem_mode] = max_offset[mem_mode];
+ data->max_offset = i == start ? 0 : i >> 1;
+ off = data->max_offset;
for (i = start; i <= 1 << 20; i <<= 1)
{
- XEXP (addr, 1) = gen_int_mode (-i, Pmode);
- if (!memory_address_p (mem_mode, addr))
+ XEXP (addr, 1) = gen_int_mode (-i, address_mode);
+ if (!memory_address_addr_space_p (mem_mode, addr, as))
break;
}
- min_offset[mem_mode] = i == start ? 0 : -(i >> 1);
+ data->min_offset = i == start ? 0 : -(i >> 1);
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "get_address_cost:\n");
fprintf (dump_file, " min offset %s %d\n",
GET_MODE_NAME (mem_mode),
- (int) min_offset[mem_mode]);
+ (int) data->min_offset);
fprintf (dump_file, " max offset %s %d\n",
GET_MODE_NAME (mem_mode),
- (int) max_offset[mem_mode]);
+ (int) data->max_offset);
}
- rat[mem_mode] = 1;
+ rat = 1;
for (i = 2; i <= MAX_RATIO; i++)
- if (multiplier_allowed_in_address_p (i, mem_mode))
+ if (multiplier_allowed_in_address_p (i, mem_mode, as))
{
- rat[mem_mode] = i;
+ rat = i;
break;
}
/* Compute the cost of various addressing modes. */
acost = 0;
- reg0 = gen_raw_REG (Pmode, LAST_VIRTUAL_REGISTER + 1);
- reg1 = gen_raw_REG (Pmode, LAST_VIRTUAL_REGISTER + 2);
+ reg0 = gen_raw_REG (address_mode, LAST_VIRTUAL_REGISTER + 1);
+ reg1 = gen_raw_REG (address_mode, LAST_VIRTUAL_REGISTER + 2);
if (HAVE_PRE_DECREMENT)
{
- addr = gen_rtx_PRE_DEC (Pmode, reg0);
- has_predec[mem_mode] = memory_address_p (mem_mode, addr);
+ addr = gen_rtx_PRE_DEC (address_mode, reg0);
+ has_predec[mem_mode]
+ = memory_address_addr_space_p (mem_mode, addr, as);
}
if (HAVE_POST_DECREMENT)
{
- addr = gen_rtx_POST_DEC (Pmode, reg0);
- has_postdec[mem_mode] = memory_address_p (mem_mode, addr);
+ addr = gen_rtx_POST_DEC (address_mode, reg0);
+ has_postdec[mem_mode]
+ = memory_address_addr_space_p (mem_mode, addr, as);
}
if (HAVE_PRE_INCREMENT)
{
- addr = gen_rtx_PRE_INC (Pmode, reg0);
- has_preinc[mem_mode] = memory_address_p (mem_mode, addr);
+ addr = gen_rtx_PRE_INC (address_mode, reg0);
+ has_preinc[mem_mode]
+ = memory_address_addr_space_p (mem_mode, addr, as);
}
if (HAVE_POST_INCREMENT)
{
- addr = gen_rtx_POST_INC (Pmode, reg0);
- has_postinc[mem_mode] = memory_address_p (mem_mode, addr);
+ addr = gen_rtx_POST_INC (address_mode, reg0);
+ has_postinc[mem_mode]
+ = memory_address_addr_space_p (mem_mode, addr, as);
}
for (i = 0; i < 16; i++)
{
addr = reg0;
if (rat_p)
- addr = gen_rtx_fmt_ee (MULT, Pmode, addr,
- gen_int_mode (rat[mem_mode], Pmode));
+ addr = gen_rtx_fmt_ee (MULT, address_mode, addr,
+ gen_int_mode (rat, address_mode));
if (var_p)
- addr = gen_rtx_fmt_ee (PLUS, Pmode, addr, reg1);
+ addr = gen_rtx_fmt_ee (PLUS, address_mode, addr, reg1);
if (sym_p)
{
- base = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (""));
+ base = gen_rtx_SYMBOL_REF (address_mode, ggc_strdup (""));
/* ??? We can run into trouble with some backends by presenting
it with symbols which haven't been properly passed through
targetm.encode_section_info. By setting the local bit, we
SYMBOL_REF_FLAGS (base) = SYMBOL_FLAG_LOCAL;
if (off_p)
- base = gen_rtx_fmt_e (CONST, Pmode,
- gen_rtx_fmt_ee (PLUS, Pmode,
- base,
- gen_int_mode (off[mem_mode],
- Pmode)));
+ base = gen_rtx_fmt_e (CONST, address_mode,
+ gen_rtx_fmt_ee
+ (PLUS, address_mode, base,
+ gen_int_mode (off, address_mode)));
}
else if (off_p)
- base = gen_int_mode (off[mem_mode], Pmode);
+ base = gen_int_mode (off, address_mode);
else
base = NULL_RTX;
-
+
if (base)
- addr = gen_rtx_fmt_ee (PLUS, Pmode, addr, base);
+ addr = gen_rtx_fmt_ee (PLUS, address_mode, addr, base);
start_sequence ();
/* To avoid splitting addressing modes, pretend that no cse will
follow. */
old_cse_not_expected = cse_not_expected;
cse_not_expected = true;
- addr = memory_address (mem_mode, addr);
+ addr = memory_address_addr_space (mem_mode, addr, as);
cse_not_expected = old_cse_not_expected;
seq = get_insns ();
end_sequence ();
acost = seq_cost (seq, speed);
- acost += address_cost (addr, mem_mode, speed);
+ acost += address_cost (addr, mem_mode, as, speed);
if (!acost)
acost = 1;
- costs[mem_mode][sym_p][var_p][off_p][rat_p] = acost;
+ data->costs[sym_p][var_p][off_p][rat_p] = acost;
}
/* On some targets, it is quite expensive to load symbol to a register,
If VAR_PRESENT is true, try whether the mode with
SYMBOL_PRESENT = false is cheaper even with cost of addition, and
if this is the case, use it. */
- add_c = add_cost (Pmode, speed);
+ add_c = add_cost (address_mode, speed);
for (i = 0; i < 8; i++)
{
var_p = i & 1;
off_p = (i >> 1) & 1;
rat_p = (i >> 2) & 1;
- acost = costs[mem_mode][0][1][off_p][rat_p] + 1;
+ acost = data->costs[0][1][off_p][rat_p] + 1;
if (var_p)
acost += add_c;
- if (acost < costs[mem_mode][1][var_p][off_p][rat_p])
- costs[mem_mode][1][var_p][off_p][rat_p] = acost;
+ if (acost < data->costs[1][var_p][off_p][rat_p])
+ data->costs[1][var_p][off_p][rat_p] = acost;
}
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "Address costs:\n");
-
+
for (i = 0; i < 16; i++)
{
sym_p = i & 1;
if (rat_p)
fprintf (dump_file, "rat * ");
- acost = costs[mem_mode][sym_p][var_p][off_p][rat_p];
+ acost = data->costs[sym_p][var_p][off_p][rat_p];
fprintf (dump_file, "index costs %d\n", acost);
}
if (has_predec[mem_mode] || has_postdec[mem_mode]
fprintf (dump_file, " May include autoinc/dec\n");
fprintf (dump_file, "\n");
}
+
+ VEC_replace (address_cost_data, address_cost_data_list,
+ data_index, data);
}
- bits = GET_MODE_BITSIZE (Pmode);
+ bits = GET_MODE_BITSIZE (address_mode);
mask = ~(~(unsigned HOST_WIDE_INT) 0 << (bits - 1) << 1);
offset &= mask;
if ((offset >> (bits - 1) & 1))
cost = 0;
offset_p = (s_offset != 0
- && min_offset[mem_mode] <= s_offset
- && s_offset <= max_offset[mem_mode]);
+ && data->min_offset <= s_offset
+ && s_offset <= data->max_offset);
ratio_p = (ratio != 1
- && multiplier_allowed_in_address_p (ratio, mem_mode));
+ && multiplier_allowed_in_address_p (ratio, mem_mode, as));
if (ratio != 1 && !ratio_p)
- cost += multiply_by_cost (ratio, Pmode, speed);
+ cost += multiply_by_cost (ratio, address_mode, speed);
if (s_offset && !offset_p && !symbol_present)
- cost += add_cost (Pmode, speed);
+ cost += add_cost (address_mode, speed);
if (may_autoinc)
*may_autoinc = autoinc;
- acost = costs[mem_mode][symbol_present][var_present][offset_p][ratio_p];
+ acost = data->costs[symbol_present][var_present][offset_p][ratio_p];
complexity = (symbol_present != 0) + (var_present != 0) + offset_p + ratio_p;
return new_cost (cost + acost, complexity);
}
case MULT_EXPR:
if (cst_and_fits_in_hwi (op0))
cost = new_cost (multiply_by_cost (int_cst_value (op0), mode, speed), 0);
- else if (cst_and_fits_in_hwi (op1))
+ else if (cst_and_fits_in_hwi (op1))
cost = new_cost (multiply_by_cost (int_cst_value (op1), mode, speed), 0);
else
return new_cost (target_spill_cost [speed], 0);
tree toffset;
enum machine_mode mode;
int unsignedp, volatilep;
-
+
core = get_inner_reference (addr, &bitsize, &bitpos, &toffset, &mode,
&unsignedp, &volatilep, false);
*var_present = false;
return zero_cost;
}
-
+
*symbol_present = false;
*var_present = true;
return zero_cost;
if (!constant_multiple_of (ustep, cstep, &rat))
return infinite_cost;
-
+
if (double_int_fits_in_shwi_p (rat))
ratio = double_int_to_shwi (rat);
else
/* use = ubase + ratio * (var - cbase). If either cbase is a constant
or ratio == 1, it is better to handle this like
-
+
ubase - ratio * cbase + ratio * var
-
+
(also holds in the case ratio == -1, TODO. */
if (cst_and_fits_in_hwi (cbase))
{
- offset = - ratio * int_cst_value (cbase);
+ offset = - ratio * int_cst_value (cbase);
cost = difference_cost (data,
ubase, build_int_cst (utype, 0),
&symbol_present, &var_present, &offset,
}
else if (address_p
&& !POINTER_TYPE_P (ctype)
- && multiplier_allowed_in_address_p (ratio,
- TYPE_MODE (TREE_TYPE (utype))))
+ && multiplier_allowed_in_address_p
+ (ratio, TYPE_MODE (TREE_TYPE (utype)),
+ TYPE_ADDR_SPACE (TREE_TYPE (utype))))
{
cbase
= fold_build2 (MULT_EXPR, ctype, cbase, build_int_cst (ctype, ratio));
get_address_cost (symbol_present, var_present,
offset, ratio, cstepi,
TYPE_MODE (TREE_TYPE (utype)),
+ TYPE_ADDR_SPACE (TREE_TYPE (utype)),
speed, stmt_is_after_inc,
can_autoinc));
bitmap depends_on_elim = NULL, depends_on_express = NULL, depends_on;
comp_cost elim_cost, express_cost, cost;
bool ok;
+ tree *control_var, *bound_cst;
/* Only consider real candidates. */
if (!cand->iv)
/* Try expressing the original giv. If it is compared with an invariant,
note that we cannot get rid of it. */
- ok = extract_cond_operands (data, use->stmt, NULL, NULL, NULL, &cmp_iv);
+ ok = extract_cond_operands (data, use->stmt, &control_var, &bound_cst,
+ NULL, &cmp_iv);
gcc_assert (ok);
+ /* When the condition is a comparison of the candidate IV against
+ zero, prefer this IV.
+
+ TODO: The constant that we're substracting from the cost should
+ be target-dependent. This information should be added to the
+ target costs for each backend. */
+ if (!infinite_cost_p (elim_cost) /* Do not try to decrease infinite! */
+ && integer_zerop (*bound_cst)
+ && (operand_equal_p (*control_var, cand->var_after, 0)
+ || operand_equal_p (*control_var, cand->var_before, 0)))
+ elim_cost.cost -= 1;
+
express_cost = get_computation_cost (data, use, cand, false,
&depends_on_express, NULL);
fd_ivopts_data = data;
if (cand->pos != IP_ORIGINAL
|| DECL_ARTIFICIAL (SSA_NAME_VAR (cand->var_before)))
cost++;
-
+
/* Prefer not to insert statements into latch unless there are some
already (so that we do not create unnecessary jumps). */
if (cand->pos == IP_END
etc.). For now the reserve is a constant 3.
Let I be the number of induction variables.
-
+
-- if U + I + R <= A, the cost is I * SMALL_COST (just not to encourage
make a lot of ivs without a reason).
-- if A - R < U + I <= A, the cost is I * PRES_COST
if (!iv_ca_has_deps (ivs, new_cp))
continue;
-
+
if (!cheaper_cost_pair (new_cp, old_cp))
continue;
continue;
if (!iv_ca_has_deps (ivs, cp))
continue;
-
+
if (!cheaper_cost_pair (cp, new_cp))
continue;
continue;
if (!iv_ca_has_deps (ivs, cp))
continue;
-
+
if (!cheaper_cost_pair (cp, new_cp))
continue;
/* Already tried this. */
if (cand->important && cand->iv->base_object == NULL_TREE)
continue;
-
+
if (iv_ca_cand_used_p (ivs, cand))
continue;
for (i = 0; i < n_iv_cands (data); i++)
{
cand = iv_cand (data, i);
-
+
if (iv_ca_cand_used_p (ivs, cand))
continue;
/* Rewrite the increment so that it uses var_before directly. */
find_interesting_uses_op (data, cand->var_after)->selected = cand;
-
+
return;
}
-
+
gimple_add_tmp_var (cand->var_before);
add_referenced_var (cand->var_before);
if (TREE_CODE (old_ref) == TARGET_MEM_REF)
copy_mem_ref_info (new_ref, old_ref);
else
- TMR_ORIGINAL (new_ref) = unshare_and_remove_ssa_names (old_ref);
+ {
+ TMR_ORIGINAL (new_ref) = unshare_and_remove_ssa_names (old_ref);
+ TREE_SIDE_EFFECTS (new_ref) = TREE_SIDE_EFFECTS (old_ref);
+ TREE_THIS_VOLATILE (new_ref) = TREE_THIS_VOLATILE (old_ref);
+ }
}
/* Rewrites USE (address that is an iv) using candidate CAND. */
{
aff_tree aff;
gimple_stmt_iterator bsi = gsi_for_stmt (use->stmt);
+ tree base_hint = NULL_TREE;
tree ref;
bool ok;
gcc_assert (ok);
unshare_aff_combination (&aff);
- ref = create_mem_ref (&bsi, TREE_TYPE (*use->op_p), &aff, data->speed);
+ /* To avoid undefined overflow problems, all IV candidates use unsigned
+ integer types. The drawback is that this makes it impossible for
+ create_mem_ref to distinguish an IV that is based on a memory object
+ from one that represents simply an offset.
+
+ To work around this problem, we pass a hint to create_mem_ref that
+ indicates which variable (if any) in aff is an IV based on a memory
+ object. Note that we only consider the candidate. If this is not
+ based on an object, the base of the reference is in some subexpression
+ of the use -- but these will use pointer types, so they are recognized
+ by the create_mem_ref heuristics anyway. */
+ if (cand->iv->base_object)
+ base_hint = var_at_stmt (data->current_loop, cand, use->stmt);
+
+ ref = create_mem_ref (&bsi, TREE_TYPE (*use->op_p), &aff, base_hint,
+ data->speed);
copy_ref_info (ref, *use->op_p);
*use->op_p = ref;
}
default:
gcc_unreachable ();
}
-
+
update_stmt (use->stmt);
}
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "Processing loop %d\n", loop->num);
-
+
exit = single_dom_exit (loop);
if (exit)
{
/* Create the new induction variables (item 4, part 1). */
create_new_ivs (data, iv_ca);
iv_ca_free (&iv_ca);
-
+
/* Rewrite the uses (item 4, part 2). */
rewrite_uses (data);