blob: e26b2c832cda4f62bd7549a6435ef4cb37cf5cdb [file] [log] [blame]
/* If-conversion support.
Copyright (C) 2000-2015 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/>. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "rtl.h"
#include "regs.h"
#include "hashtab.h"
#include "hash-set.h"
#include "vec.h"
#include "machmode.h"
#include "hard-reg-set.h"
#include "input.h"
#include "function.h"
#include "flags.h"
#include "insn-config.h"
#include "recog.h"
#include "except.h"
#include "predict.h"
#include "dominance.h"
#include "cfg.h"
#include "cfgrtl.h"
#include "cfganal.h"
#include "cfgcleanup.h"
#include "basic-block.h"
#include "symtab.h"
#include "statistics.h"
#include "double-int.h"
#include "real.h"
#include "fixed-value.h"
#include "alias.h"
#include "wide-int.h"
#include "inchash.h"
#include "tree.h"
#include "expmed.h"
#include "dojump.h"
#include "explow.h"
#include "calls.h"
#include "emit-rtl.h"
#include "varasm.h"
#include "stmt.h"
#include "expr.h"
#include "output.h"
#include "insn-codes.h"
#include "optabs.h"
#include "diagnostic-core.h"
#include "tm_p.h"
#include "cfgloop.h"
#include "target.h"
#include "tree-pass.h"
#include "df.h"
#include "dbgcnt.h"
#include "shrink-wrap.h"
#include "ifcvt.h"
#ifndef HAVE_conditional_move
#define HAVE_conditional_move 0
#endif
#ifndef HAVE_incscc
#define HAVE_incscc 0
#endif
#ifndef HAVE_decscc
#define HAVE_decscc 0
#endif
#ifndef HAVE_trap
#define HAVE_trap 0
#endif
#ifndef MAX_CONDITIONAL_EXECUTE
#define MAX_CONDITIONAL_EXECUTE \
(BRANCH_COST (optimize_function_for_speed_p (cfun), false) \
+ 1)
#endif
#ifndef HAVE_cbranchcc4
#define HAVE_cbranchcc4 0
#endif
#define IFCVT_MULTIPLE_DUMPS 1
#define NULL_BLOCK ((basic_block) NULL)
/* True if after combine pass. */
static bool ifcvt_after_combine;
/* # of IF-THEN or IF-THEN-ELSE blocks we looked at */
static int num_possible_if_blocks;
/* # of IF-THEN or IF-THEN-ELSE blocks were converted to conditional
execution. */
static int num_updated_if_blocks;
/* # of changes made. */
static int num_true_changes;
/* Whether conditional execution changes were made. */
static int cond_exec_changed_p;
/* Forward references. */
static int count_bb_insns (const_basic_block);
static bool cheap_bb_rtx_cost_p (const_basic_block, int, int);
static rtx_insn *first_active_insn (basic_block);
static rtx_insn *last_active_insn (basic_block, int);
static rtx_insn *find_active_insn_before (basic_block, rtx_insn *);
static rtx_insn *find_active_insn_after (basic_block, rtx_insn *);
static basic_block block_fallthru (basic_block);
static int cond_exec_process_insns (ce_if_block *, rtx_insn *, rtx, rtx, int,
int);
static rtx cond_exec_get_condition (rtx_insn *);
static rtx noce_get_condition (rtx_insn *, rtx_insn **, bool);
static int noce_operand_ok (const_rtx);
static void merge_if_block (ce_if_block *);
static int find_cond_trap (basic_block, edge, edge);
static basic_block find_if_header (basic_block, int);
static int block_jumps_and_fallthru_p (basic_block, basic_block);
static int noce_find_if_block (basic_block, edge, edge, int);
static int cond_exec_find_if_block (ce_if_block *);
static int find_if_case_1 (basic_block, edge, edge);
static int find_if_case_2 (basic_block, edge, edge);
static int dead_or_predicable (basic_block, basic_block, basic_block,
edge, int);
static void noce_emit_move_insn (rtx, rtx);
static rtx_insn *block_has_only_trap (basic_block);
/* Count the number of non-jump active insns in BB. */
static int
count_bb_insns (const_basic_block bb)
{
int count = 0;
rtx_insn *insn = BB_HEAD (bb);
while (1)
{
if (active_insn_p (insn) && !JUMP_P (insn))
count++;
if (insn == BB_END (bb))
break;
insn = NEXT_INSN (insn);
}
return count;
}
/* Determine whether the total insn_rtx_cost on non-jump insns in
basic block BB is less than MAX_COST. This function returns
false if the cost of any instruction could not be estimated.
The cost of the non-jump insns in BB is scaled by REG_BR_PROB_BASE
as those insns are being speculated. MAX_COST is scaled with SCALE
plus a small fudge factor. */
static bool
cheap_bb_rtx_cost_p (const_basic_block bb, int scale, int max_cost)
{
int count = 0;
rtx_insn *insn = BB_HEAD (bb);
bool speed = optimize_bb_for_speed_p (bb);
/* Set scale to REG_BR_PROB_BASE to void the identical scaling
applied to insn_rtx_cost when optimizing for size. Only do
this after combine because if-conversion might interfere with
passes before combine.
Use optimize_function_for_speed_p instead of the pre-defined
variable speed to make sure it is set to same value for all
basic blocks in one if-conversion transformation. */
if (!optimize_function_for_speed_p (cfun) && ifcvt_after_combine)
scale = REG_BR_PROB_BASE;
/* Our branch probability/scaling factors are just estimates and don't
account for cases where we can get speculation for free and other
secondary benefits. So we fudge the scale factor to make speculating
appear a little more profitable when optimizing for performance. */
else
scale += REG_BR_PROB_BASE / 8;
max_cost *= scale;
while (1)
{
if (NONJUMP_INSN_P (insn))
{
int cost = insn_rtx_cost (PATTERN (insn), speed) * REG_BR_PROB_BASE;
if (cost == 0)
return false;
/* If this instruction is the load or set of a "stack" register,
such as a floating point register on x87, then the cost of
speculatively executing this insn may need to include
the additional cost of popping its result off of the
register stack. Unfortunately, correctly recognizing and
accounting for this additional overhead is tricky, so for
now we simply prohibit such speculative execution. */
#ifdef STACK_REGS
{
rtx set = single_set (insn);
if (set && STACK_REG_P (SET_DEST (set)))
return false;
}
#endif
count += cost;
if (count >= max_cost)
return false;
}
else if (CALL_P (insn))
return false;
if (insn == BB_END (bb))
break;
insn = NEXT_INSN (insn);
}
return true;
}
/* Return the first non-jump active insn in the basic block. */
static rtx_insn *
first_active_insn (basic_block bb)
{
rtx_insn *insn = BB_HEAD (bb);
if (LABEL_P (insn))
{
if (insn == BB_END (bb))
return NULL;
insn = NEXT_INSN (insn);
}
while (NOTE_P (insn) || DEBUG_INSN_P (insn))
{
if (insn == BB_END (bb))
return NULL;
insn = NEXT_INSN (insn);
}
if (JUMP_P (insn))
return NULL;
return insn;
}
/* Return the last non-jump active (non-jump) insn in the basic block. */
static rtx_insn *
last_active_insn (basic_block bb, int skip_use_p)
{
rtx_insn *insn = BB_END (bb);
rtx_insn *head = BB_HEAD (bb);
while (NOTE_P (insn)
|| JUMP_P (insn)
|| DEBUG_INSN_P (insn)
|| (skip_use_p
&& NONJUMP_INSN_P (insn)
&& GET_CODE (PATTERN (insn)) == USE))
{
if (insn == head)
return NULL;
insn = PREV_INSN (insn);
}
if (LABEL_P (insn))
return NULL;
return insn;
}
/* Return the active insn before INSN inside basic block CURR_BB. */
static rtx_insn *
find_active_insn_before (basic_block curr_bb, rtx_insn *insn)
{
if (!insn || insn == BB_HEAD (curr_bb))
return NULL;
while ((insn = PREV_INSN (insn)) != NULL_RTX)
{
if (NONJUMP_INSN_P (insn) || JUMP_P (insn) || CALL_P (insn))
break;
/* No other active insn all the way to the start of the basic block. */
if (insn == BB_HEAD (curr_bb))
return NULL;
}
return insn;
}
/* Return the active insn after INSN inside basic block CURR_BB. */
static rtx_insn *
find_active_insn_after (basic_block curr_bb, rtx_insn *insn)
{
if (!insn || insn == BB_END (curr_bb))
return NULL;
while ((insn = NEXT_INSN (insn)) != NULL_RTX)
{
if (NONJUMP_INSN_P (insn) || JUMP_P (insn) || CALL_P (insn))
break;
/* No other active insn all the way to the end of the basic block. */
if (insn == BB_END (curr_bb))
return NULL;
}
return insn;
}
/* Return the basic block reached by falling though the basic block BB. */
static basic_block
block_fallthru (basic_block bb)
{
edge e = find_fallthru_edge (bb->succs);
return (e) ? e->dest : NULL_BLOCK;
}
/* Return true if RTXs A and B can be safely interchanged. */
static bool
rtx_interchangeable_p (const_rtx a, const_rtx b)
{
if (!rtx_equal_p (a, b))
return false;
if (GET_CODE (a) != MEM)
return true;
/* A dead type-unsafe memory reference is legal, but a live type-unsafe memory
reference is not. Interchanging a dead type-unsafe memory reference with
a live type-safe one creates a live type-unsafe memory reference, in other
words, it makes the program illegal.
We check here conservatively whether the two memory references have equal
memory attributes. */
return mem_attrs_eq_p (get_mem_attrs (a), get_mem_attrs (b));
}
/* Go through a bunch of insns, converting them to conditional
execution format if possible. Return TRUE if all of the non-note
insns were processed. */
static int
cond_exec_process_insns (ce_if_block *ce_info ATTRIBUTE_UNUSED,
/* if block information */rtx_insn *start,
/* first insn to look at */rtx end,
/* last insn to look at */rtx test,
/* conditional execution test */int prob_val,
/* probability of branch taken. */int mod_ok)
{
int must_be_last = FALSE;
rtx_insn *insn;
rtx xtest;
rtx pattern;
if (!start || !end)
return FALSE;
for (insn = start; ; insn = NEXT_INSN (insn))
{
/* dwarf2out can't cope with conditional prologues. */
if (NOTE_P (insn) && NOTE_KIND (insn) == NOTE_INSN_PROLOGUE_END)
return FALSE;
if (NOTE_P (insn) || DEBUG_INSN_P (insn))
goto insn_done;
gcc_assert (NONJUMP_INSN_P (insn) || CALL_P (insn));
/* dwarf2out can't cope with conditional unwind info. */
if (RTX_FRAME_RELATED_P (insn))
return FALSE;
/* Remove USE insns that get in the way. */
if (reload_completed && GET_CODE (PATTERN (insn)) == USE)
{
/* ??? Ug. Actually unlinking the thing is problematic,
given what we'd have to coordinate with our callers. */
SET_INSN_DELETED (insn);
goto insn_done;
}
/* Last insn wasn't last? */
if (must_be_last)
return FALSE;
if (modified_in_p (test, insn))
{
if (!mod_ok)
return FALSE;
must_be_last = TRUE;
}
/* Now build the conditional form of the instruction. */
pattern = PATTERN (insn);
xtest = copy_rtx (test);
/* If this is already a COND_EXEC, rewrite the test to be an AND of the
two conditions. */
if (GET_CODE (pattern) == COND_EXEC)
{
if (GET_MODE (xtest) != GET_MODE (COND_EXEC_TEST (pattern)))
return FALSE;
xtest = gen_rtx_AND (GET_MODE (xtest), xtest,
COND_EXEC_TEST (pattern));
pattern = COND_EXEC_CODE (pattern);
}
pattern = gen_rtx_COND_EXEC (VOIDmode, xtest, pattern);
/* If the machine needs to modify the insn being conditionally executed,
say for example to force a constant integer operand into a temp
register, do so here. */
#ifdef IFCVT_MODIFY_INSN
IFCVT_MODIFY_INSN (ce_info, pattern, insn);
if (! pattern)
return FALSE;
#endif
validate_change (insn, &PATTERN (insn), pattern, 1);
if (CALL_P (insn) && prob_val >= 0)
validate_change (insn, &REG_NOTES (insn),
gen_rtx_INT_LIST ((machine_mode) REG_BR_PROB,
prob_val, REG_NOTES (insn)), 1);
insn_done:
if (insn == end)
break;
}
return TRUE;
}
/* Return the condition for a jump. Do not do any special processing. */
static rtx
cond_exec_get_condition (rtx_insn *jump)
{
rtx test_if, cond;
if (any_condjump_p (jump))
test_if = SET_SRC (pc_set (jump));
else
return NULL_RTX;
cond = XEXP (test_if, 0);
/* If this branches to JUMP_LABEL when the condition is false,
reverse the condition. */
if (GET_CODE (XEXP (test_if, 2)) == LABEL_REF
&& LABEL_REF_LABEL (XEXP (test_if, 2)) == JUMP_LABEL (jump))
{
enum rtx_code rev = reversed_comparison_code (cond, jump);
if (rev == UNKNOWN)
return NULL_RTX;
cond = gen_rtx_fmt_ee (rev, GET_MODE (cond), XEXP (cond, 0),
XEXP (cond, 1));
}
return cond;
}
/* Given a simple IF-THEN or IF-THEN-ELSE block, attempt to convert it
to conditional execution. Return TRUE if we were successful at
converting the block. */
static int
cond_exec_process_if_block (ce_if_block * ce_info,
/* if block information */int do_multiple_p)
{
basic_block test_bb = ce_info->test_bb; /* last test block */
basic_block then_bb = ce_info->then_bb; /* THEN */
basic_block else_bb = ce_info->else_bb; /* ELSE or NULL */
rtx test_expr; /* expression in IF_THEN_ELSE that is tested */
rtx_insn *then_start; /* first insn in THEN block */
rtx_insn *then_end; /* last insn + 1 in THEN block */
rtx_insn *else_start = NULL; /* first insn in ELSE block or NULL */
rtx_insn *else_end = NULL; /* last insn + 1 in ELSE block */
int max; /* max # of insns to convert. */
int then_mod_ok; /* whether conditional mods are ok in THEN */
rtx true_expr; /* test for else block insns */
rtx false_expr; /* test for then block insns */
int true_prob_val; /* probability of else block */
int false_prob_val; /* probability of then block */
rtx_insn *then_last_head = NULL; /* Last match at the head of THEN */
rtx_insn *else_last_head = NULL; /* Last match at the head of ELSE */
rtx_insn *then_first_tail = NULL; /* First match at the tail of THEN */
rtx_insn *else_first_tail = NULL; /* First match at the tail of ELSE */
int then_n_insns, else_n_insns, n_insns;
enum rtx_code false_code;
rtx note;
/* If test is comprised of && or || elements, and we've failed at handling
all of them together, just use the last test if it is the special case of
&& elements without an ELSE block. */
if (!do_multiple_p && ce_info->num_multiple_test_blocks)
{
if (else_bb || ! ce_info->and_and_p)
return FALSE;
ce_info->test_bb = test_bb = ce_info->last_test_bb;
ce_info->num_multiple_test_blocks = 0;
ce_info->num_and_and_blocks = 0;
ce_info->num_or_or_blocks = 0;
}
/* Find the conditional jump to the ELSE or JOIN part, and isolate
the test. */
test_expr = cond_exec_get_condition (BB_END (test_bb));
if (! test_expr)
return FALSE;
/* If the conditional jump is more than just a conditional jump,
then we can not do conditional execution conversion on this block. */
if (! onlyjump_p (BB_END (test_bb)))
return FALSE;
/* Collect the bounds of where we're to search, skipping any labels, jumps
and notes at the beginning and end of the block. Then count the total
number of insns and see if it is small enough to convert. */
then_start = first_active_insn (then_bb);
then_end = last_active_insn (then_bb, TRUE);
then_n_insns = ce_info->num_then_insns = count_bb_insns (then_bb);
n_insns = then_n_insns;
max = MAX_CONDITIONAL_EXECUTE;
if (else_bb)
{
int n_matching;
max *= 2;
else_start = first_active_insn (else_bb);
else_end = last_active_insn (else_bb, TRUE);
else_n_insns = ce_info->num_else_insns = count_bb_insns (else_bb);
n_insns += else_n_insns;
/* Look for matching sequences at the head and tail of the two blocks,
and limit the range of insns to be converted if possible. */
n_matching = flow_find_cross_jump (then_bb, else_bb,
&then_first_tail, &else_first_tail,
NULL);
if (then_first_tail == BB_HEAD (then_bb))
then_start = then_end = NULL;
if (else_first_tail == BB_HEAD (else_bb))
else_start = else_end = NULL;
if (n_matching > 0)
{
if (then_end)
then_end = find_active_insn_before (then_bb, then_first_tail);
if (else_end)
else_end = find_active_insn_before (else_bb, else_first_tail);
n_insns -= 2 * n_matching;
}
if (then_start
&& else_start
&& then_n_insns > n_matching
&& else_n_insns > n_matching)
{
int longest_match = MIN (then_n_insns - n_matching,
else_n_insns - n_matching);
n_matching
= flow_find_head_matching_sequence (then_bb, else_bb,
&then_last_head,
&else_last_head,
longest_match);
if (n_matching > 0)
{
rtx_insn *insn;
/* We won't pass the insns in the head sequence to
cond_exec_process_insns, so we need to test them here
to make sure that they don't clobber the condition. */
for (insn = BB_HEAD (then_bb);
insn != NEXT_INSN (then_last_head);
insn = NEXT_INSN (insn))
if (!LABEL_P (insn) && !NOTE_P (insn)
&& !DEBUG_INSN_P (insn)
&& modified_in_p (test_expr, insn))
return FALSE;
}
if (then_last_head == then_end)
then_start = then_end = NULL;
if (else_last_head == else_end)
else_start = else_end = NULL;
if (n_matching > 0)
{
if (then_start)
then_start = find_active_insn_after (then_bb, then_last_head);
if (else_start)
else_start = find_active_insn_after (else_bb, else_last_head);
n_insns -= 2 * n_matching;
}
}
}
if (n_insns > max)
return FALSE;
/* Map test_expr/test_jump into the appropriate MD tests to use on
the conditionally executed code. */
true_expr = test_expr;
false_code = reversed_comparison_code (true_expr, BB_END (test_bb));
if (false_code != UNKNOWN)
false_expr = gen_rtx_fmt_ee (false_code, GET_MODE (true_expr),
XEXP (true_expr, 0), XEXP (true_expr, 1));
else
false_expr = NULL_RTX;
#ifdef IFCVT_MODIFY_TESTS
/* If the machine description needs to modify the tests, such as setting a
conditional execution register from a comparison, it can do so here. */
IFCVT_MODIFY_TESTS (ce_info, true_expr, false_expr);
/* See if the conversion failed. */
if (!true_expr || !false_expr)
goto fail;
#endif
note = find_reg_note (BB_END (test_bb), REG_BR_PROB, NULL_RTX);
if (note)
{
true_prob_val = XINT (note, 0);
false_prob_val = REG_BR_PROB_BASE - true_prob_val;
}
else
{
true_prob_val = -1;
false_prob_val = -1;
}
/* If we have && or || tests, do them here. These tests are in the adjacent
blocks after the first block containing the test. */
if (ce_info->num_multiple_test_blocks > 0)
{
basic_block bb = test_bb;
basic_block last_test_bb = ce_info->last_test_bb;
if (! false_expr)
goto fail;
do
{
rtx_insn *start, *end;
rtx t, f;
enum rtx_code f_code;
bb = block_fallthru (bb);
start = first_active_insn (bb);
end = last_active_insn (bb, TRUE);
if (start
&& ! cond_exec_process_insns (ce_info, start, end, false_expr,
false_prob_val, FALSE))
goto fail;
/* If the conditional jump is more than just a conditional jump, then
we can not do conditional execution conversion on this block. */
if (! onlyjump_p (BB_END (bb)))
goto fail;
/* Find the conditional jump and isolate the test. */
t = cond_exec_get_condition (BB_END (bb));
if (! t)
goto fail;
f_code = reversed_comparison_code (t, BB_END (bb));
if (f_code == UNKNOWN)
goto fail;
f = gen_rtx_fmt_ee (f_code, GET_MODE (t), XEXP (t, 0), XEXP (t, 1));
if (ce_info->and_and_p)
{
t = gen_rtx_AND (GET_MODE (t), true_expr, t);
f = gen_rtx_IOR (GET_MODE (t), false_expr, f);
}
else
{
t = gen_rtx_IOR (GET_MODE (t), true_expr, t);
f = gen_rtx_AND (GET_MODE (t), false_expr, f);
}
/* If the machine description needs to modify the tests, such as
setting a conditional execution register from a comparison, it can
do so here. */
#ifdef IFCVT_MODIFY_MULTIPLE_TESTS
IFCVT_MODIFY_MULTIPLE_TESTS (ce_info, bb, t, f);
/* See if the conversion failed. */
if (!t || !f)
goto fail;
#endif
true_expr = t;
false_expr = f;
}
while (bb != last_test_bb);
}
/* For IF-THEN-ELSE blocks, we don't allow modifications of the test
on then THEN block. */
then_mod_ok = (else_bb == NULL_BLOCK);
/* Go through the THEN and ELSE blocks converting the insns if possible
to conditional execution. */
if (then_end
&& (! false_expr
|| ! cond_exec_process_insns (ce_info, then_start, then_end,
false_expr, false_prob_val,
then_mod_ok)))
goto fail;
if (else_bb && else_end
&& ! cond_exec_process_insns (ce_info, else_start, else_end,
true_expr, true_prob_val, TRUE))
goto fail;
/* If we cannot apply the changes, fail. Do not go through the normal fail
processing, since apply_change_group will call cancel_changes. */
if (! apply_change_group ())
{
#ifdef IFCVT_MODIFY_CANCEL
/* Cancel any machine dependent changes. */
IFCVT_MODIFY_CANCEL (ce_info);
#endif
return FALSE;
}
#ifdef IFCVT_MODIFY_FINAL
/* Do any machine dependent final modifications. */
IFCVT_MODIFY_FINAL (ce_info);
#endif
/* Conversion succeeded. */
if (dump_file)
fprintf (dump_file, "%d insn%s converted to conditional execution.\n",
n_insns, (n_insns == 1) ? " was" : "s were");
/* Merge the blocks! If we had matching sequences, make sure to delete one
copy at the appropriate location first: delete the copy in the THEN branch
for a tail sequence so that the remaining one is executed last for both
branches, and delete the copy in the ELSE branch for a head sequence so
that the remaining one is executed first for both branches. */
if (then_first_tail)
{
rtx_insn *from = then_first_tail;
if (!INSN_P (from))
from = find_active_insn_after (then_bb, from);
delete_insn_chain (from, BB_END (then_bb), false);
}
if (else_last_head)
delete_insn_chain (first_active_insn (else_bb), else_last_head, false);
merge_if_block (ce_info);
cond_exec_changed_p = TRUE;
return TRUE;
fail:
#ifdef IFCVT_MODIFY_CANCEL
/* Cancel any machine dependent changes. */
IFCVT_MODIFY_CANCEL (ce_info);
#endif
cancel_changes (0);
return FALSE;
}
/* Used by noce_process_if_block to communicate with its subroutines.
The subroutines know that A and B may be evaluated freely. They
know that X is a register. They should insert new instructions
before cond_earliest. */
struct noce_if_info
{
/* The basic blocks that make up the IF-THEN-{ELSE-,}JOIN block. */
basic_block test_bb, then_bb, else_bb, join_bb;
/* The jump that ends TEST_BB. */
rtx_insn *jump;
/* The jump condition. */
rtx cond;
/* New insns should be inserted before this one. */
rtx_insn *cond_earliest;
/* Insns in the THEN and ELSE block. There is always just this
one insns in those blocks. The insns are single_set insns.
If there was no ELSE block, INSN_B is the last insn before
COND_EARLIEST, or NULL_RTX. In the former case, the insn
operands are still valid, as if INSN_B was moved down below
the jump. */
rtx_insn *insn_a, *insn_b;
/* The SET_SRC of INSN_A and INSN_B. */
rtx a, b;
/* The SET_DEST of INSN_A. */
rtx x;
/* True if this if block is not canonical. In the canonical form of
if blocks, the THEN_BB is the block reached via the fallthru edge
from TEST_BB. For the noce transformations, we allow the symmetric
form as well. */
bool then_else_reversed;
/* Estimated cost of the particular branch instruction. */
int branch_cost;
};
static rtx noce_emit_store_flag (struct noce_if_info *, rtx, int, int);
static int noce_try_move (struct noce_if_info *);
static int noce_try_store_flag (struct noce_if_info *);
static int noce_try_addcc (struct noce_if_info *);
static int noce_try_store_flag_constants (struct noce_if_info *);
static int noce_try_store_flag_mask (struct noce_if_info *);
static rtx noce_emit_cmove (struct noce_if_info *, rtx, enum rtx_code, rtx,
rtx, rtx, rtx);
static int noce_try_cmove (struct noce_if_info *);
static int noce_try_cmove_arith (struct noce_if_info *);
static rtx noce_get_alt_condition (struct noce_if_info *, rtx, rtx_insn **);
static int noce_try_minmax (struct noce_if_info *);
static int noce_try_abs (struct noce_if_info *);
static int noce_try_sign_mask (struct noce_if_info *);
/* Helper function for noce_try_store_flag*. */
static rtx
noce_emit_store_flag (struct noce_if_info *if_info, rtx x, int reversep,
int normalize)
{
rtx cond = if_info->cond;
int cond_complex;
enum rtx_code code;
cond_complex = (! general_operand (XEXP (cond, 0), VOIDmode)
|| ! general_operand (XEXP (cond, 1), VOIDmode));
/* If earliest == jump, or when the condition is complex, try to
build the store_flag insn directly. */
if (cond_complex)
{
rtx set = pc_set (if_info->jump);
cond = XEXP (SET_SRC (set), 0);
if (GET_CODE (XEXP (SET_SRC (set), 2)) == LABEL_REF
&& LABEL_REF_LABEL (XEXP (SET_SRC (set), 2)) == JUMP_LABEL (if_info->jump))
reversep = !reversep;
if (if_info->then_else_reversed)
reversep = !reversep;
}
if (reversep)
code = reversed_comparison_code (cond, if_info->jump);
else
code = GET_CODE (cond);
if ((if_info->cond_earliest == if_info->jump || cond_complex)
&& (normalize == 0 || STORE_FLAG_VALUE == normalize))
{
rtx src = gen_rtx_fmt_ee (code, GET_MODE (x), XEXP (cond, 0),
XEXP (cond, 1));
rtx set = gen_rtx_SET (VOIDmode, x, src);
start_sequence ();
rtx_insn *insn = emit_insn (set);
if (recog_memoized (insn) >= 0)
{
rtx_insn *seq = get_insns ();
end_sequence ();
emit_insn (seq);
if_info->cond_earliest = if_info->jump;
return x;
}
end_sequence ();
}
/* Don't even try if the comparison operands or the mode of X are weird. */
if (cond_complex || !SCALAR_INT_MODE_P (GET_MODE (x)))
return NULL_RTX;
return emit_store_flag (x, code, XEXP (cond, 0),
XEXP (cond, 1), VOIDmode,
(code == LTU || code == LEU
|| code == GEU || code == GTU), normalize);
}
/* Emit instruction to move an rtx, possibly into STRICT_LOW_PART.
X is the destination/target and Y is the value to copy. */
static void
noce_emit_move_insn (rtx x, rtx y)
{
machine_mode outmode;
rtx outer, inner;
int bitpos;
if (GET_CODE (x) != STRICT_LOW_PART)
{
rtx_insn *seq, *insn;
rtx target;
optab ot;
start_sequence ();
/* Check that the SET_SRC is reasonable before calling emit_move_insn,
otherwise construct a suitable SET pattern ourselves. */
insn = (OBJECT_P (y) || CONSTANT_P (y) || GET_CODE (y) == SUBREG)
? emit_move_insn (x, y)
: emit_insn (gen_rtx_SET (VOIDmode, x, y));
seq = get_insns ();
end_sequence ();
if (recog_memoized (insn) <= 0)
{
if (GET_CODE (x) == ZERO_EXTRACT)
{
rtx op = XEXP (x, 0);
unsigned HOST_WIDE_INT size = INTVAL (XEXP (x, 1));
unsigned HOST_WIDE_INT start = INTVAL (XEXP (x, 2));
/* store_bit_field expects START to be relative to
BYTES_BIG_ENDIAN and adjusts this value for machines with
BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN. In order to be able to
invoke store_bit_field again it is necessary to have the START
value from the first call. */
if (BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN)
{
if (MEM_P (op))
start = BITS_PER_UNIT - start - size;
else
{
gcc_assert (REG_P (op));
start = BITS_PER_WORD - start - size;
}
}
gcc_assert (start < (MEM_P (op) ? BITS_PER_UNIT : BITS_PER_WORD));
store_bit_field (op, size, start, 0, 0, GET_MODE (x), y);
return;
}
switch (GET_RTX_CLASS (GET_CODE (y)))
{
case RTX_UNARY:
ot = code_to_optab (GET_CODE (y));
if (ot)
{
start_sequence ();
target = expand_unop (GET_MODE (y), ot, XEXP (y, 0), x, 0);
if (target != NULL_RTX)
{
if (target != x)
emit_move_insn (x, target);
seq = get_insns ();
}
end_sequence ();
}
break;
case RTX_BIN_ARITH:
case RTX_COMM_ARITH:
ot = code_to_optab (GET_CODE (y));
if (ot)
{
start_sequence ();
target = expand_binop (GET_MODE (y), ot,
XEXP (y, 0), XEXP (y, 1),
x, 0, OPTAB_DIRECT);
if (target != NULL_RTX)
{
if (target != x)
emit_move_insn (x, target);
seq = get_insns ();
}
end_sequence ();
}
break;
default:
break;
}
}
emit_insn (seq);
return;
}
outer = XEXP (x, 0);
inner = XEXP (outer, 0);
outmode = GET_MODE (outer);
bitpos = SUBREG_BYTE (outer) * BITS_PER_UNIT;
store_bit_field (inner, GET_MODE_BITSIZE (outmode), bitpos,
0, 0, outmode, y);
}
/* Return the CC reg if it is used in COND. */
static rtx
cc_in_cond (rtx cond)
{
if (HAVE_cbranchcc4 && cond
&& GET_MODE_CLASS (GET_MODE (XEXP (cond, 0))) == MODE_CC)
return XEXP (cond, 0);
return NULL_RTX;
}
/* Return sequence of instructions generated by if conversion. This
function calls end_sequence() to end the current stream, ensures
that are instructions are unshared, recognizable non-jump insns.
On failure, this function returns a NULL_RTX. */
static rtx_insn *
end_ifcvt_sequence (struct noce_if_info *if_info)
{
rtx_insn *insn;
rtx_insn *seq = get_insns ();
rtx cc = cc_in_cond (if_info->cond);
set_used_flags (if_info->x);
set_used_flags (if_info->cond);
set_used_flags (if_info->a);
set_used_flags (if_info->b);
unshare_all_rtl_in_chain (seq);
end_sequence ();
/* Make sure that all of the instructions emitted are recognizable,
and that we haven't introduced a new jump instruction.
As an exercise for the reader, build a general mechanism that
allows proper placement of required clobbers. */
for (insn = seq; insn; insn = NEXT_INSN (insn))
if (JUMP_P (insn)
|| recog_memoized (insn) == -1
/* Make sure new generated code does not clobber CC. */
|| (cc && set_of (cc, insn)))
return NULL;
return seq;
}
/* Convert "if (a != b) x = a; else x = b" into "x = a" and
"if (a == b) x = a; else x = b" into "x = b". */
static int
noce_try_move (struct noce_if_info *if_info)
{
rtx cond = if_info->cond;
enum rtx_code code = GET_CODE (cond);
rtx y;
rtx_insn *seq;
if (code != NE && code != EQ)
return FALSE;
/* This optimization isn't valid if either A or B could be a NaN
or a signed zero. */
if (HONOR_NANS (if_info->x)
|| HONOR_SIGNED_ZEROS (if_info->x))
return FALSE;
/* Check whether the operands of the comparison are A and in
either order. */
if ((rtx_equal_p (if_info->a, XEXP (cond, 0))
&& rtx_equal_p (if_info->b, XEXP (cond, 1)))
|| (rtx_equal_p (if_info->a, XEXP (cond, 1))
&& rtx_equal_p (if_info->b, XEXP (cond, 0))))
{
if (!rtx_interchangeable_p (if_info->a, if_info->b))
return FALSE;
y = (code == EQ) ? if_info->a : if_info->b;
/* Avoid generating the move if the source is the destination. */
if (! rtx_equal_p (if_info->x, y))
{
start_sequence ();
noce_emit_move_insn (if_info->x, y);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
}
return TRUE;
}
return FALSE;
}
/* Convert "if (test) x = 1; else x = 0".
Only try 0 and STORE_FLAG_VALUE here. Other combinations will be
tried in noce_try_store_flag_constants after noce_try_cmove has had
a go at the conversion. */
static int
noce_try_store_flag (struct noce_if_info *if_info)
{
int reversep;
rtx target;
rtx_insn *seq;
if (CONST_INT_P (if_info->b)
&& INTVAL (if_info->b) == STORE_FLAG_VALUE
&& if_info->a == const0_rtx)
reversep = 0;
else if (if_info->b == const0_rtx
&& CONST_INT_P (if_info->a)
&& INTVAL (if_info->a) == STORE_FLAG_VALUE
&& (reversed_comparison_code (if_info->cond, if_info->jump)
!= UNKNOWN))
reversep = 1;
else
return FALSE;
start_sequence ();
target = noce_emit_store_flag (if_info, if_info->x, reversep, 0);
if (target)
{
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (! seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
else
{
end_sequence ();
return FALSE;
}
}
/* Convert "if (test) x = a; else x = b", for A and B constant. */
static int
noce_try_store_flag_constants (struct noce_if_info *if_info)
{
rtx target;
rtx_insn *seq;
int reversep;
HOST_WIDE_INT itrue, ifalse, diff, tmp;
int normalize, can_reverse;
machine_mode mode;
if (CONST_INT_P (if_info->a)
&& CONST_INT_P (if_info->b))
{
mode = GET_MODE (if_info->x);
ifalse = INTVAL (if_info->a);
itrue = INTVAL (if_info->b);
diff = (unsigned HOST_WIDE_INT) itrue - ifalse;
/* Make sure we can represent the difference between the two values. */
if ((diff > 0)
!= ((ifalse < 0) != (itrue < 0) ? ifalse < 0 : ifalse < itrue))
return FALSE;
diff = trunc_int_for_mode (diff, mode);
can_reverse = (reversed_comparison_code (if_info->cond, if_info->jump)
!= UNKNOWN);
reversep = 0;
if (diff == STORE_FLAG_VALUE || diff == -STORE_FLAG_VALUE)
normalize = 0;
else if (ifalse == 0 && exact_log2 (itrue) >= 0
&& (STORE_FLAG_VALUE == 1
|| if_info->branch_cost >= 2))
normalize = 1;
else if (itrue == 0 && exact_log2 (ifalse) >= 0 && can_reverse
&& (STORE_FLAG_VALUE == 1 || if_info->branch_cost >= 2))
normalize = 1, reversep = 1;
else if (itrue == -1
&& (STORE_FLAG_VALUE == -1
|| if_info->branch_cost >= 2))
normalize = -1;
else if (ifalse == -1 && can_reverse
&& (STORE_FLAG_VALUE == -1 || if_info->branch_cost >= 2))
normalize = -1, reversep = 1;
else if ((if_info->branch_cost >= 2 && STORE_FLAG_VALUE == -1)
|| if_info->branch_cost >= 3)
normalize = -1;
else
return FALSE;
if (reversep)
{
tmp = itrue; itrue = ifalse; ifalse = tmp;
diff = trunc_int_for_mode (-(unsigned HOST_WIDE_INT) diff, mode);
}
start_sequence ();
target = noce_emit_store_flag (if_info, if_info->x, reversep, normalize);
if (! target)
{
end_sequence ();
return FALSE;
}
/* if (test) x = 3; else x = 4;
=> x = 3 + (test == 0); */
if (diff == STORE_FLAG_VALUE || diff == -STORE_FLAG_VALUE)
{
target = expand_simple_binop (mode,
(diff == STORE_FLAG_VALUE
? PLUS : MINUS),
gen_int_mode (ifalse, mode), target,
if_info->x, 0, OPTAB_WIDEN);
}
/* if (test) x = 8; else x = 0;
=> x = (test != 0) << 3; */
else if (ifalse == 0 && (tmp = exact_log2 (itrue)) >= 0)
{
target = expand_simple_binop (mode, ASHIFT,
target, GEN_INT (tmp), if_info->x, 0,
OPTAB_WIDEN);
}
/* if (test) x = -1; else x = b;
=> x = -(test != 0) | b; */
else if (itrue == -1)
{
target = expand_simple_binop (mode, IOR,
target, gen_int_mode (ifalse, mode),
if_info->x, 0, OPTAB_WIDEN);
}
/* if (test) x = a; else x = b;
=> x = (-(test != 0) & (b - a)) + a; */
else
{
target = expand_simple_binop (mode, AND,
target, gen_int_mode (diff, mode),
if_info->x, 0, OPTAB_WIDEN);
if (target)
target = expand_simple_binop (mode, PLUS,
target, gen_int_mode (ifalse, mode),
if_info->x, 0, OPTAB_WIDEN);
}
if (! target)
{
end_sequence ();
return FALSE;
}
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
return FALSE;
}
/* Convert "if (test) foo++" into "foo += (test != 0)", and
similarly for "foo--". */
static int
noce_try_addcc (struct noce_if_info *if_info)
{
rtx target;
rtx_insn *seq;
int subtract, normalize;
if (GET_CODE (if_info->a) == PLUS
&& rtx_equal_p (XEXP (if_info->a, 0), if_info->b)
&& (reversed_comparison_code (if_info->cond, if_info->jump)
!= UNKNOWN))
{
rtx cond = if_info->cond;
enum rtx_code code = reversed_comparison_code (cond, if_info->jump);
/* First try to use addcc pattern. */
if (general_operand (XEXP (cond, 0), VOIDmode)
&& general_operand (XEXP (cond, 1), VOIDmode))
{
start_sequence ();
target = emit_conditional_add (if_info->x, code,
XEXP (cond, 0),
XEXP (cond, 1),
VOIDmode,
if_info->b,
XEXP (if_info->a, 1),
GET_MODE (if_info->x),
(code == LTU || code == GEU
|| code == LEU || code == GTU));
if (target)
{
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
end_sequence ();
}
/* If that fails, construct conditional increment or decrement using
setcc. */
if (if_info->branch_cost >= 2
&& (XEXP (if_info->a, 1) == const1_rtx
|| XEXP (if_info->a, 1) == constm1_rtx))
{
start_sequence ();
if (STORE_FLAG_VALUE == INTVAL (XEXP (if_info->a, 1)))
subtract = 0, normalize = 0;
else if (-STORE_FLAG_VALUE == INTVAL (XEXP (if_info->a, 1)))
subtract = 1, normalize = 0;
else
subtract = 0, normalize = INTVAL (XEXP (if_info->a, 1));
target = noce_emit_store_flag (if_info,
gen_reg_rtx (GET_MODE (if_info->x)),
1, normalize);
if (target)
target = expand_simple_binop (GET_MODE (if_info->x),
subtract ? MINUS : PLUS,
if_info->b, target, if_info->x,
0, OPTAB_WIDEN);
if (target)
{
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
end_sequence ();
}
}
return FALSE;
}
/* Convert "if (test) x = 0;" to "x &= -(test == 0);" */
static int
noce_try_store_flag_mask (struct noce_if_info *if_info)
{
rtx target;
rtx_insn *seq;
int reversep;
reversep = 0;
if ((if_info->branch_cost >= 2
|| STORE_FLAG_VALUE == -1)
&& ((if_info->a == const0_rtx
&& rtx_equal_p (if_info->b, if_info->x))
|| ((reversep = (reversed_comparison_code (if_info->cond,
if_info->jump)
!= UNKNOWN))
&& if_info->b == const0_rtx
&& rtx_equal_p (if_info->a, if_info->x))))
{
start_sequence ();
target = noce_emit_store_flag (if_info,
gen_reg_rtx (GET_MODE (if_info->x)),
reversep, -1);
if (target)
target = expand_simple_binop (GET_MODE (if_info->x), AND,
if_info->x,
target, if_info->x, 0,
OPTAB_WIDEN);
if (target)
{
int old_cost, new_cost, insn_cost;
int speed_p;
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
speed_p = optimize_bb_for_speed_p (BLOCK_FOR_INSN (if_info->insn_a));
insn_cost = insn_rtx_cost (PATTERN (if_info->insn_a), speed_p);
old_cost = COSTS_N_INSNS (if_info->branch_cost) + insn_cost;
new_cost = seq_cost (seq, speed_p);
if (new_cost > old_cost)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
end_sequence ();
}
return FALSE;
}
/* Helper function for noce_try_cmove and noce_try_cmove_arith. */
static rtx
noce_emit_cmove (struct noce_if_info *if_info, rtx x, enum rtx_code code,
rtx cmp_a, rtx cmp_b, rtx vfalse, rtx vtrue)
{
rtx target ATTRIBUTE_UNUSED;
int unsignedp ATTRIBUTE_UNUSED;
/* If earliest == jump, try to build the cmove insn directly.
This is helpful when combine has created some complex condition
(like for alpha's cmovlbs) that we can't hope to regenerate
through the normal interface. */
if (if_info->cond_earliest == if_info->jump)
{
rtx cond = gen_rtx_fmt_ee (code, GET_MODE (if_info->cond), cmp_a, cmp_b);
rtx if_then_else = gen_rtx_IF_THEN_ELSE (GET_MODE (x),
cond, vtrue, vfalse);
rtx set = gen_rtx_SET (VOIDmode, x, if_then_else);
start_sequence ();
rtx_insn *insn = emit_insn (set);
if (recog_memoized (insn) >= 0)
{
rtx_insn *seq = get_insns ();
end_sequence ();
emit_insn (seq);
return x;
}
end_sequence ();
}
/* Don't even try if the comparison operands are weird
except that the target supports cbranchcc4. */
if (! general_operand (cmp_a, GET_MODE (cmp_a))
|| ! general_operand (cmp_b, GET_MODE (cmp_b)))
{
if (!(HAVE_cbranchcc4)
|| GET_MODE_CLASS (GET_MODE (cmp_a)) != MODE_CC
|| cmp_b != const0_rtx)
return NULL_RTX;
}
#if HAVE_conditional_move
unsignedp = (code == LTU || code == GEU
|| code == LEU || code == GTU);
target = emit_conditional_move (x, code, cmp_a, cmp_b, VOIDmode,
vtrue, vfalse, GET_MODE (x),
unsignedp);
if (target)
return target;
/* We might be faced with a situation like:
x = (reg:M TARGET)
vtrue = (subreg:M (reg:N VTRUE) BYTE)
vfalse = (subreg:M (reg:N VFALSE) BYTE)
We can't do a conditional move in mode M, but it's possible that we
could do a conditional move in mode N instead and take a subreg of
the result.
If we can't create new pseudos, though, don't bother. */
if (reload_completed)
return NULL_RTX;
if (GET_CODE (vtrue) == SUBREG && GET_CODE (vfalse) == SUBREG)
{
rtx reg_vtrue = SUBREG_REG (vtrue);
rtx reg_vfalse = SUBREG_REG (vfalse);
unsigned int byte_vtrue = SUBREG_BYTE (vtrue);
unsigned int byte_vfalse = SUBREG_BYTE (vfalse);
rtx promoted_target;
if (GET_MODE (reg_vtrue) != GET_MODE (reg_vfalse)
|| byte_vtrue != byte_vfalse
|| (SUBREG_PROMOTED_VAR_P (vtrue)
!= SUBREG_PROMOTED_VAR_P (vfalse))
|| (SUBREG_PROMOTED_GET (vtrue)
!= SUBREG_PROMOTED_GET (vfalse)))
return NULL_RTX;
promoted_target = gen_reg_rtx (GET_MODE (reg_vtrue));
target = emit_conditional_move (promoted_target, code, cmp_a, cmp_b,
VOIDmode, reg_vtrue, reg_vfalse,
GET_MODE (reg_vtrue), unsignedp);
/* Nope, couldn't do it in that mode either. */
if (!target)
return NULL_RTX;
target = gen_rtx_SUBREG (GET_MODE (vtrue), promoted_target, byte_vtrue);
SUBREG_PROMOTED_VAR_P (target) = SUBREG_PROMOTED_VAR_P (vtrue);
SUBREG_PROMOTED_SET (target, SUBREG_PROMOTED_GET (vtrue));
emit_move_insn (x, target);
return x;
}
else
return NULL_RTX;
#else
/* We'll never get here, as noce_process_if_block doesn't call the
functions involved. Ifdef code, however, should be discouraged
because it leads to typos in the code not selected. However,
emit_conditional_move won't exist either. */
return NULL_RTX;
#endif
}
/* Try only simple constants and registers here. More complex cases
are handled in noce_try_cmove_arith after noce_try_store_flag_arith
has had a go at it. */
static int
noce_try_cmove (struct noce_if_info *if_info)
{
enum rtx_code code;
rtx target;
rtx_insn *seq;
if ((CONSTANT_P (if_info->a) || register_operand (if_info->a, VOIDmode))
&& (CONSTANT_P (if_info->b) || register_operand (if_info->b, VOIDmode)))
{
start_sequence ();
code = GET_CODE (if_info->cond);
target = noce_emit_cmove (if_info, if_info->x, code,
XEXP (if_info->cond, 0),
XEXP (if_info->cond, 1),
if_info->a, if_info->b);
if (target)
{
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
}
else
{
end_sequence ();
return FALSE;
}
}
return FALSE;
}
/* Try more complex cases involving conditional_move. */
static int
noce_try_cmove_arith (struct noce_if_info *if_info)
{
rtx a = if_info->a;
rtx b = if_info->b;
rtx x = if_info->x;
rtx orig_a, orig_b;
rtx_insn *insn_a, *insn_b;
rtx target;
int is_mem = 0;
int insn_cost;
enum rtx_code code;
rtx_insn *ifcvt_seq;
/* A conditional move from two memory sources is equivalent to a
conditional on their addresses followed by a load. Don't do this
early because it'll screw alias analysis. Note that we've
already checked for no side effects. */
/* ??? FIXME: Magic number 5. */
if (cse_not_expected
&& MEM_P (a) && MEM_P (b)
&& MEM_ADDR_SPACE (a) == MEM_ADDR_SPACE (b)
&& if_info->branch_cost >= 5)
{
machine_mode address_mode = get_address_mode (a);
a = XEXP (a, 0);
b = XEXP (b, 0);
x = gen_reg_rtx (address_mode);
is_mem = 1;
}
/* ??? We could handle this if we knew that a load from A or B could
not trap or fault. This is also true if we've already loaded
from the address along the path from ENTRY. */
else if (may_trap_or_fault_p (a) || may_trap_or_fault_p (b))
return FALSE;
/* if (test) x = a + b; else x = c - d;
=> y = a + b;
x = c - d;
if (test)
x = y;
*/
code = GET_CODE (if_info->cond);
insn_a = if_info->insn_a;
insn_b = if_info->insn_b;
/* Total insn_rtx_cost should be smaller than branch cost. Exit
if insn_rtx_cost can't be estimated. */
if (insn_a)
{
insn_cost
= insn_rtx_cost (PATTERN (insn_a),
optimize_bb_for_speed_p (BLOCK_FOR_INSN (insn_a)));
if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (if_info->branch_cost))
return FALSE;
}
else
insn_cost = 0;
if (insn_b)
{
insn_cost
+= insn_rtx_cost (PATTERN (insn_b),
optimize_bb_for_speed_p (BLOCK_FOR_INSN (insn_b)));
if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (if_info->branch_cost))
return FALSE;
}
/* Possibly rearrange operands to make things come out more natural. */
if (reversed_comparison_code (if_info->cond, if_info->jump) != UNKNOWN)
{
int reversep = 0;
if (rtx_equal_p (b, x))
reversep = 1;
else if (general_operand (b, GET_MODE (b)))
reversep = 1;
if (reversep)
{
rtx tmp;
rtx_insn *tmp_insn;
code = reversed_comparison_code (if_info->cond, if_info->jump);
tmp = a, a = b, b = tmp;
tmp_insn = insn_a, insn_a = insn_b, insn_b = tmp_insn;
}
}
start_sequence ();
orig_a = a;
orig_b = b;
/* If either operand is complex, load it into a register first.
The best way to do this is to copy the original insn. In this
way we preserve any clobbers etc that the insn may have had.
This is of course not possible in the IS_MEM case. */
if (! general_operand (a, GET_MODE (a)))
{
rtx_insn *insn;
if (is_mem)
{
rtx reg = gen_reg_rtx (GET_MODE (a));
insn = emit_insn (gen_rtx_SET (VOIDmode, reg, a));
}
else if (! insn_a)
goto end_seq_and_fail;
else
{
a = gen_reg_rtx (GET_MODE (a));
rtx_insn *copy_of_a = as_a <rtx_insn *> (copy_rtx (insn_a));
rtx set = single_set (copy_of_a);
SET_DEST (set) = a;
insn = emit_insn (PATTERN (copy_of_a));
}
if (recog_memoized (insn) < 0)
goto end_seq_and_fail;
}
if (! general_operand (b, GET_MODE (b)))
{
rtx pat;
rtx_insn *last;
rtx_insn *new_insn;
if (is_mem)
{
rtx reg = gen_reg_rtx (GET_MODE (b));
pat = gen_rtx_SET (VOIDmode, reg, b);
}
else if (! insn_b)
goto end_seq_and_fail;
else
{
b = gen_reg_rtx (GET_MODE (b));
rtx_insn *copy_of_insn_b = as_a <rtx_insn *> (copy_rtx (insn_b));
rtx set = single_set (copy_of_insn_b);
SET_DEST (set) = b;
pat = PATTERN (copy_of_insn_b);
}
/* If insn to set up A clobbers any registers B depends on, try to
swap insn that sets up A with the one that sets up B. If even
that doesn't help, punt. */
last = get_last_insn ();
if (last && modified_in_p (orig_b, last))
{
new_insn = emit_insn_before (pat, get_insns ());
if (modified_in_p (orig_a, new_insn))
goto end_seq_and_fail;
}
else
new_insn = emit_insn (pat);
if (recog_memoized (new_insn) < 0)
goto end_seq_and_fail;
}
target = noce_emit_cmove (if_info, x, code, XEXP (if_info->cond, 0),
XEXP (if_info->cond, 1), a, b);
if (! target)
goto end_seq_and_fail;
/* If we're handling a memory for above, emit the load now. */
if (is_mem)
{
rtx mem = gen_rtx_MEM (GET_MODE (if_info->x), target);
/* Copy over flags as appropriate. */
if (MEM_VOLATILE_P (if_info->a) || MEM_VOLATILE_P (if_info->b))
MEM_VOLATILE_P (mem) = 1;
if (MEM_ALIAS_SET (if_info->a) == MEM_ALIAS_SET (if_info->b))
set_mem_alias_set (mem, MEM_ALIAS_SET (if_info->a));
set_mem_align (mem,
MIN (MEM_ALIGN (if_info->a), MEM_ALIGN (if_info->b)));
gcc_assert (MEM_ADDR_SPACE (if_info->a) == MEM_ADDR_SPACE (if_info->b));
set_mem_addr_space (mem, MEM_ADDR_SPACE (if_info->a));
noce_emit_move_insn (if_info->x, mem);
}
else if (target != x)
noce_emit_move_insn (x, target);
ifcvt_seq = end_ifcvt_sequence (if_info);
if (!ifcvt_seq)
return FALSE;
emit_insn_before_setloc (ifcvt_seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
return TRUE;
end_seq_and_fail:
end_sequence ();
return FALSE;
}
/* For most cases, the simplified condition we found is the best
choice, but this is not the case for the min/max/abs transforms.
For these we wish to know that it is A or B in the condition. */
static rtx
noce_get_alt_condition (struct noce_if_info *if_info, rtx target,
rtx_insn **earliest)
{
rtx cond, set;
rtx_insn *insn;
int reverse;
/* If target is already mentioned in the known condition, return it. */
if (reg_mentioned_p (target, if_info->cond))
{
*earliest = if_info->cond_earliest;
return if_info->cond;
}
set = pc_set (if_info->jump);
cond = XEXP (SET_SRC (set), 0);
reverse
= GET_CODE (XEXP (SET_SRC (set), 2)) == LABEL_REF
&& LABEL_REF_LABEL (XEXP (SET_SRC (set), 2)) == JUMP_LABEL (if_info->jump);
if (if_info->then_else_reversed)
reverse = !reverse;
/* If we're looking for a constant, try to make the conditional
have that constant in it. There are two reasons why it may
not have the constant we want:
1. GCC may have needed to put the constant in a register, because
the target can't compare directly against that constant. For
this case, we look for a SET immediately before the comparison
that puts a constant in that register.
2. GCC may have canonicalized the conditional, for example
replacing "if x < 4" with "if x <= 3". We can undo that (or
make equivalent types of changes) to get the constants we need
if they're off by one in the right direction. */
if (CONST_INT_P (target))
{
enum rtx_code code = GET_CODE (if_info->cond);
rtx op_a = XEXP (if_info->cond, 0);
rtx op_b = XEXP (if_info->cond, 1);
rtx prev_insn;
/* First, look to see if we put a constant in a register. */
prev_insn = prev_nonnote_insn (if_info->cond_earliest);
if (prev_insn
&& BLOCK_FOR_INSN (prev_insn)
== BLOCK_FOR_INSN (if_info->cond_earliest)
&& INSN_P (prev_insn)
&& GET_CODE (PATTERN (prev_insn)) == SET)
{
rtx src = find_reg_equal_equiv_note (prev_insn);
if (!src)
src = SET_SRC (PATTERN (prev_insn));
if (CONST_INT_P (src))
{
if (rtx_equal_p (op_a, SET_DEST (PATTERN (prev_insn))))
op_a = src;
else if (rtx_equal_p (op_b, SET_DEST (PATTERN (prev_insn))))
op_b = src;
if (CONST_INT_P (op_a))
{
rtx tmp = op_a;
op_a = op_b;
op_b = tmp;
code = swap_condition (code);
}
}
}
/* Now, look to see if we can get the right constant by
adjusting the conditional. */
if (CONST_INT_P (op_b))
{
HOST_WIDE_INT desired_val = INTVAL (target);
HOST_WIDE_INT actual_val = INTVAL (op_b);
switch (code)
{
case LT:
if (actual_val == desired_val + 1)
{
code = LE;
op_b = GEN_INT (desired_val);
}
break;
case LE:
if (actual_val == desired_val - 1)
{
code = LT;
op_b = GEN_INT (desired_val);
}
break;
case GT:
if (actual_val == desired_val - 1)
{
code = GE;
op_b = GEN_INT (desired_val);
}
break;
case GE:
if (actual_val == desired_val + 1)
{
code = GT;
op_b = GEN_INT (desired_val);
}
break;
default:
break;
}
}
/* If we made any changes, generate a new conditional that is
equivalent to what we started with, but has the right
constants in it. */
if (code != GET_CODE (if_info->cond)
|| op_a != XEXP (if_info->cond, 0)
|| op_b != XEXP (if_info->cond, 1))
{
cond = gen_rtx_fmt_ee (code, GET_MODE (cond), op_a, op_b);
*earliest = if_info->cond_earliest;
return cond;
}
}
cond = canonicalize_condition (if_info->jump, cond, reverse,
earliest, target, HAVE_cbranchcc4, true);
if (! cond || ! reg_mentioned_p (target, cond))
return NULL;
/* We almost certainly searched back to a different place.
Need to re-verify correct lifetimes. */
/* X may not be mentioned in the range (cond_earliest, jump]. */
for (insn = if_info->jump; insn != *earliest; insn = PREV_INSN (insn))
if (INSN_P (insn) && reg_overlap_mentioned_p (if_info->x, PATTERN (insn)))
return NULL;
/* A and B may not be modified in the range [cond_earliest, jump). */
for (insn = *earliest; insn != if_info->jump; insn = NEXT_INSN (insn))
if (INSN_P (insn)
&& (modified_in_p (if_info->a, insn)
|| modified_in_p (if_info->b, insn)))
return NULL;
return cond;
}
/* Convert "if (a < b) x = a; else x = b;" to "x = min(a, b);", etc. */
static int
noce_try_minmax (struct noce_if_info *if_info)
{
rtx cond, target;
rtx_insn *earliest, *seq;
enum rtx_code code, op;
int unsignedp;
/* ??? Reject modes with NaNs or signed zeros since we don't know how
they will be resolved with an SMIN/SMAX. It wouldn't be too hard
to get the target to tell us... */
if (HONOR_SIGNED_ZEROS (if_info->x)
|| HONOR_NANS (if_info->x))
return FALSE;
cond = noce_get_alt_condition (if_info, if_info->a, &earliest);
if (!cond)
return FALSE;
/* Verify the condition is of the form we expect, and canonicalize
the comparison code. */
code = GET_CODE (cond);
if (rtx_equal_p (XEXP (cond, 0), if_info->a))
{
if (! rtx_equal_p (XEXP (cond, 1), if_info->b))
return FALSE;
}
else if (rtx_equal_p (XEXP (cond, 1), if_info->a))
{
if (! rtx_equal_p (XEXP (cond, 0), if_info->b))
return FALSE;
code = swap_condition (code);
}
else
return FALSE;
/* Determine what sort of operation this is. Note that the code is for
a taken branch, so the code->operation mapping appears backwards. */
switch (code)
{
case LT:
case LE:
case UNLT:
case UNLE:
op = SMAX;
unsignedp = 0;
break;
case GT:
case GE:
case UNGT:
case UNGE:
op = SMIN;
unsignedp = 0;
break;
case LTU:
case LEU:
op = UMAX;
unsignedp = 1;
break;
case GTU:
case GEU:
op = UMIN;
unsignedp = 1;
break;
default:
return FALSE;
}
start_sequence ();
target = expand_simple_binop (GET_MODE (if_info->x), op,
if_info->a, if_info->b,
if_info->x, unsignedp, OPTAB_WIDEN);
if (! target)
{
end_sequence ();
return FALSE;
}
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump, INSN_LOCATION (if_info->insn_a));
if_info->cond = cond;
if_info->cond_earliest = earliest;
return TRUE;
}
/* Convert "if (a < 0) x = -a; else x = a;" to "x = abs(a);",
"if (a < 0) x = ~a; else x = a;" to "x = one_cmpl_abs(a);",
etc. */
static int
noce_try_abs (struct noce_if_info *if_info)
{
rtx cond, target, a, b, c;
rtx_insn *earliest, *seq;
int negate;
bool one_cmpl = false;
/* Reject modes with signed zeros. */
if (HONOR_SIGNED_ZEROS (if_info->x))
return FALSE;
/* Recognize A and B as constituting an ABS or NABS. The canonical
form is a branch around the negation, taken when the object is the
first operand of a comparison against 0 that evaluates to true. */
a = if_info->a;
b = if_info->b;
if (GET_CODE (a) == NEG && rtx_equal_p (XEXP (a, 0), b))
negate = 0;
else if (GET_CODE (b) == NEG && rtx_equal_p (XEXP (b, 0), a))
{
c = a; a = b; b = c;
negate = 1;
}
else if (GET_CODE (a) == NOT && rtx_equal_p (XEXP (a, 0), b))
{
negate = 0;
one_cmpl = true;
}
else if (GET_CODE (b) == NOT && rtx_equal_p (XEXP (b, 0), a))
{
c = a; a = b; b = c;
negate = 1;
one_cmpl = true;
}
else
return FALSE;
cond = noce_get_alt_condition (if_info, b, &earliest);
if (!cond)
return FALSE;
/* Verify the condition is of the form we expect. */
if (rtx_equal_p (XEXP (cond, 0), b))
c = XEXP (cond, 1);
else if (rtx_equal_p (XEXP (cond, 1), b))
{
c = XEXP (cond, 0);
negate = !negate;
}
else
return FALSE;
/* Verify that C is zero. Search one step backward for a
REG_EQUAL note or a simple source if necessary. */
if (REG_P (c))
{
rtx set;
rtx_insn *insn = prev_nonnote_insn (earliest);
if (insn
&& BLOCK_FOR_INSN (insn) == BLOCK_FOR_INSN (earliest)
&& (set = single_set (insn))
&& rtx_equal_p (SET_DEST (set), c))
{
rtx note = find_reg_equal_equiv_note (insn);
if (note)
c = XEXP (note, 0);
else
c = SET_SRC (set);
}
else
return FALSE;
}
if (MEM_P (c)
&& GET_CODE (XEXP (c, 0)) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (XEXP (c, 0)))
c = get_pool_constant (XEXP (c, 0));
/* Work around funny ideas get_condition has wrt canonicalization.
Note that these rtx constants are known to be CONST_INT, and
therefore imply integer comparisons.
The one_cmpl case is more complicated, as we want to handle
only x < 0 ? ~x : x or x >= 0 ? x : ~x to one_cmpl_abs (x)
and x < 0 ? x : ~x or x >= 0 ? ~x : x to ~one_cmpl_abs (x),
but not other cases (x > -1 is equivalent of x >= 0). */
if (c == constm1_rtx && GET_CODE (cond) == GT)
;
else if (c == const1_rtx && GET_CODE (cond) == LT)
{
if (one_cmpl)
return FALSE;
}
else if (c == CONST0_RTX (GET_MODE (b)))
{
if (one_cmpl
&& GET_CODE (cond) != GE
&& GET_CODE (cond) != LT)
return FALSE;
}
else
return FALSE;
/* Determine what sort of operation this is. */
switch (GET_CODE (cond))
{
case LT:
case LE:
case UNLT:
case UNLE:
negate = !negate;
break;
case GT:
case GE:
case UNGT:
case UNGE:
break;
default:
return FALSE;
}
start_sequence ();
if (one_cmpl)
target = expand_one_cmpl_abs_nojump (GET_MODE (if_info->x), b,
if_info->x);
else
target = expand_abs_nojump (GET_MODE (if_info->x), b, if_info->x, 1);
/* ??? It's a quandary whether cmove would be better here, especially
for integers. Perhaps combine will clean things up. */
if (target && negate)
{
if (one_cmpl)
target = expand_simple_unop (GET_MODE (target), NOT, target,
if_info->x, 0);
else
target = expand_simple_unop (GET_MODE (target), NEG, target,
if_info->x, 0);
}
if (! target)
{
end_sequence ();
return FALSE;
}
if (target != if_info->x)
noce_emit_move_insn (if_info->x, target);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump, INSN_LOCATION (if_info->insn_a));
if_info->cond = cond;
if_info->cond_earliest = earliest;
return TRUE;
}
/* Convert "if (m < 0) x = b; else x = 0;" to "x = (m >> C) & b;". */
static int
noce_try_sign_mask (struct noce_if_info *if_info)
{
rtx cond, t, m, c;
rtx_insn *seq;
machine_mode mode;
enum rtx_code code;
bool t_unconditional;
cond = if_info->cond;
code = GET_CODE (cond);
m = XEXP (cond, 0);
c = XEXP (cond, 1);
t = NULL_RTX;
if (if_info->a == const0_rtx)
{
if ((code == LT && c == const0_rtx)
|| (code == LE && c == constm1_rtx))
t = if_info->b;
}
else if (if_info->b == const0_rtx)
{
if ((code == GE && c == const0_rtx)
|| (code == GT && c == constm1_rtx))
t = if_info->a;
}
if (! t || side_effects_p (t))
return FALSE;
/* We currently don't handle different modes. */
mode = GET_MODE (t);
if (GET_MODE (m) != mode)
return FALSE;
/* This is only profitable if T is unconditionally executed/evaluated in the
original insn sequence or T is cheap. The former happens if B is the
non-zero (T) value and if INSN_B was taken from TEST_BB, or there was no
INSN_B which can happen for e.g. conditional stores to memory. For the
cost computation use the block TEST_BB where the evaluation will end up
after the transformation. */
t_unconditional =
(t == if_info->b
&& (if_info->insn_b == NULL_RTX
|| BLOCK_FOR_INSN (if_info->insn_b) == if_info->test_bb));
if (!(t_unconditional
|| (set_src_cost (t, optimize_bb_for_speed_p (if_info->test_bb))
< COSTS_N_INSNS (2))))
return FALSE;
start_sequence ();
/* Use emit_store_flag to generate "m < 0 ? -1 : 0" instead of expanding
"(signed) m >> 31" directly. This benefits targets with specialized
insns to obtain the signmask, but still uses ashr_optab otherwise. */
m = emit_store_flag (gen_reg_rtx (mode), LT, m, const0_rtx, mode, 0, -1);
t = m ? expand_binop (mode, and_optab, m, t, NULL_RTX, 0, OPTAB_DIRECT)
: NULL_RTX;
if (!t)
{
end_sequence ();
return FALSE;
}
noce_emit_move_insn (if_info->x, t);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump, INSN_LOCATION (if_info->insn_a));
return TRUE;
}
/* Optimize away "if (x & C) x |= C" and similar bit manipulation
transformations. */
static int
noce_try_bitop (struct noce_if_info *if_info)
{
rtx cond, x, a, result;
rtx_insn *seq;
machine_mode mode;
enum rtx_code code;
int bitnum;
x = if_info->x;
cond = if_info->cond;
code = GET_CODE (cond);
/* Check for no else condition. */
if (! rtx_equal_p (x, if_info->b))
return FALSE;
/* Check for a suitable condition. */
if (code != NE && code != EQ)
return FALSE;
if (XEXP (cond, 1) != const0_rtx)
return FALSE;
cond = XEXP (cond, 0);
/* ??? We could also handle AND here. */
if (GET_CODE (cond) == ZERO_EXTRACT)
{
if (XEXP (cond, 1) != const1_rtx
|| !CONST_INT_P (XEXP (cond, 2))
|| ! rtx_equal_p (x, XEXP (cond, 0)))
return FALSE;
bitnum = INTVAL (XEXP (cond, 2));
mode = GET_MODE (x);
if (BITS_BIG_ENDIAN)
bitnum = GET_MODE_BITSIZE (mode) - 1 - bitnum;
if (bitnum < 0 || bitnum >= HOST_BITS_PER_WIDE_INT)
return FALSE;
}
else
return FALSE;
a = if_info->a;
if (GET_CODE (a) == IOR || GET_CODE (a) == XOR)
{
/* Check for "if (X & C) x = x op C". */
if (! rtx_equal_p (x, XEXP (a, 0))
|| !CONST_INT_P (XEXP (a, 1))
|| (INTVAL (XEXP (a, 1)) & GET_MODE_MASK (mode))
!= (unsigned HOST_WIDE_INT) 1 << bitnum)
return FALSE;
/* if ((x & C) == 0) x |= C; is transformed to x |= C. */
/* if ((x & C) != 0) x |= C; is transformed to nothing. */
if (GET_CODE (a) == IOR)
result = (code == NE) ? a : NULL_RTX;
else if (code == NE)
{
/* if ((x & C) == 0) x ^= C; is transformed to x |= C. */
result = gen_int_mode ((HOST_WIDE_INT) 1 << bitnum, mode);
result = simplify_gen_binary (IOR, mode, x, result);
}
else
{
/* if ((x & C) != 0) x ^= C; is transformed to x &= ~C. */
result = gen_int_mode (~((HOST_WIDE_INT) 1 << bitnum), mode);
result = simplify_gen_binary (AND, mode, x, result);
}
}
else if (GET_CODE (a) == AND)
{
/* Check for "if (X & C) x &= ~C". */
if (! rtx_equal_p (x, XEXP (a, 0))
|| !CONST_INT_P (XEXP (a, 1))
|| (INTVAL (XEXP (a, 1)) & GET_MODE_MASK (mode))
!= (~((HOST_WIDE_INT) 1 << bitnum) & GET_MODE_MASK (mode)))
return FALSE;
/* if ((x & C) == 0) x &= ~C; is transformed to nothing. */
/* if ((x & C) != 0) x &= ~C; is transformed to x &= ~C. */
result = (code == EQ) ? a : NULL_RTX;
}
else
return FALSE;
if (result)
{
start_sequence ();
noce_emit_move_insn (x, result);
seq = end_ifcvt_sequence (if_info);
if (!seq)
return FALSE;
emit_insn_before_setloc (seq, if_info->jump,
INSN_LOCATION (if_info->insn_a));
}
return TRUE;
}
/* Similar to get_condition, only the resulting condition must be
valid at JUMP, instead of at EARLIEST.
If THEN_ELSE_REVERSED is true, the fallthrough does not go to the
THEN block of the caller, and we have to reverse the condition. */
static rtx
noce_get_condition (rtx_insn *jump, rtx_insn **earliest, bool then_else_reversed)
{
rtx cond, set, tmp;
bool reverse;
if (! any_condjump_p (jump))
return NULL_RTX;
set = pc_set (jump);
/* If this branches to JUMP_LABEL when the condition is false,
reverse the condition. */
reverse = (GET_CODE (XEXP (SET_SRC (set), 2)) == LABEL_REF
&& LABEL_REF_LABEL (XEXP (SET_SRC (set), 2)) == JUMP_LABEL (jump));
/* We may have to reverse because the caller's if block is not canonical,
i.e. the THEN block isn't the fallthrough block for the TEST block
(see find_if_header). */
if (then_else_reversed)
reverse = !reverse;
/* If the condition variable is a register and is MODE_INT, accept it. */
cond = XEXP (SET_SRC (set), 0);
tmp = XEXP (cond, 0);
if (REG_P (tmp) && GET_MODE_CLASS (GET_MODE (tmp)) == MODE_INT
&& (GET_MODE (tmp) != BImode
|| !targetm.small_register_classes_for_mode_p (BImode)))
{
*earliest = jump;
if (reverse)
cond = gen_rtx_fmt_ee (reverse_condition (GET_CODE (cond)),
GET_MODE (cond), tmp, XEXP (cond, 1));
return cond;
}
/* Otherwise, fall back on canonicalize_condition to do the dirty
work of manipulating MODE_CC values and COMPARE rtx codes. */
tmp = canonicalize_condition (jump, cond, reverse, earliest,
NULL_RTX, HAVE_cbranchcc4, true);
/* We don't handle side-effects in the condition, like handling
REG_INC notes and making sure no duplicate conditions are emitted. */
if (tmp != NULL_RTX && side_effects_p (tmp))
return NULL_RTX;
return tmp;
}
/* Return true if OP is ok for if-then-else processing. */
static int
noce_operand_ok (const_rtx op)
{
if (side_effects_p (op))
return FALSE;
/* We special-case memories, so handle any of them with
no address side effects. */
if (MEM_P (op))
return ! side_effects_p (XEXP (op, 0));
return ! may_trap_p (op);
}
/* Return true if a write into MEM may trap or fault. */
static bool
noce_mem_write_may_trap_or_fault_p (const_rtx mem)
{
rtx addr;
if (MEM_READONLY_P (mem))
return true;
if (may_trap_or_fault_p (mem))
return true;
addr = XEXP (mem, 0);
/* Call target hook to avoid the effects of -fpic etc.... */
addr = targetm.delegitimize_address (addr);
while (addr)
switch (GET_CODE (addr))
{
case CONST:
case PRE_DEC:
case PRE_INC:
case POST_DEC:
case POST_INC:
case POST_MODIFY:
addr = XEXP (addr, 0);
break;
case LO_SUM:
case PRE_MODIFY:
addr = XEXP (addr, 1);
break;
case PLUS:
if (CONST_INT_P (XEXP (addr, 1)))
addr = XEXP (addr, 0);
else
return false;
break;
case LABEL_REF:
return true;
case SYMBOL_REF:
if (SYMBOL_REF_DECL (addr)
&& decl_readonly_section (SYMBOL_REF_DECL (addr), 0))
return true;
return false;
default:
return false;
}
return false;
}
/* Return whether we can use store speculation for MEM. TOP_BB is the
basic block above the conditional block where we are considering
doing the speculative store. We look for whether MEM is set
unconditionally later in the function. */
static bool
noce_can_store_speculate_p (basic_block top_bb, const_rtx mem)
{
basic_block dominator;
for (dominator = get_immediate_dominator (CDI_POST_DOMINATORS, top_bb);
dominator != NULL;
dominator = get_immediate_dominator (CDI_POST_DOMINATORS, dominator))
{
rtx_insn *insn;
FOR_BB_INSNS (dominator, insn)
{
/* If we see something that might be a memory barrier, we
have to stop looking. Even if the MEM is set later in
the function, we still don't want to set it
unconditionally before the barrier. */
if (INSN_P (insn)
&& (volatile_insn_p (PATTERN (insn))
|| (CALL_P (insn) && (!RTL_CONST_CALL_P (insn)))))
return false;
if (memory_must_be_modified_in_insn_p (mem, insn))
return true;
if (modified_in_p (XEXP (mem, 0), insn))
return false;
}
}
return false;
}
/* Given a simple IF-THEN-JOIN or IF-THEN-ELSE-JOIN block, attempt to convert
it without using conditional execution. Return TRUE if we were successful
at converting the block. */
static int
noce_process_if_block (struct noce_if_info *if_info)
{
basic_block test_bb = if_info->test_bb; /* test block */
basic_block then_bb = if_info->then_bb; /* THEN */
basic_block else_bb = if_info->else_bb; /* ELSE or NULL */
basic_block join_bb = if_info->join_bb; /* JOIN */
rtx_insn *jump = if_info->jump;
rtx cond = if_info->cond;
rtx_insn *insn_a, *insn_b;
rtx set_a, set_b;
rtx orig_x, x, a, b;
rtx cc;
/* We're looking for patterns of the form
(1) if (...) x = a; else x = b;
(2) x = b; if (...) x = a;
(3) if (...) x = a; // as if with an initial x = x.
The later patterns require jumps to be more expensive.
??? For future expansion, look for multiple X in such patterns. */
/* Look for one of the potential sets. */
insn_a = first_active_insn (then_bb);
if (! insn_a
|| insn_a != last_active_insn (then_bb, FALSE)
|| (set_a = single_set (insn_a)) == NULL_RTX)
return FALSE;
x = SET_DEST (set_a);
a = SET_SRC (set_a);
/* Look for the other potential set. Make sure we've got equivalent
destinations. */
/* ??? This is overconservative. Storing to two different mems is
as easy as conditionally computing the address. Storing to a
single mem merely requires a scratch memory to use as one of the
destination addresses; often the memory immediately below the
stack pointer is available for this. */
set_b = NULL_RTX;
if (else_bb)
{
insn_b = first_active_insn (else_bb);
if (! insn_b
|| insn_b != last_active_insn (else_bb, FALSE)
|| (set_b = single_set (insn_b)) == NULL_RTX
|| ! rtx_interchangeable_p (x, SET_DEST (set_b)))
return FALSE;
}
else
{
insn_b = prev_nonnote_nondebug_insn (if_info->cond_earliest);
/* We're going to be moving the evaluation of B down from above
COND_EARLIEST to JUMP. Make sure the relevant data is still
intact. */
if (! insn_b
|| BLOCK_FOR_INSN (insn_b) != BLOCK_FOR_INSN (if_info->cond_earliest)
|| !NONJUMP_INSN_P (insn_b)
|| (set_b = single_set (insn_b)) == NULL_RTX
|| ! rtx_interchangeable_p (x, SET_DEST (set_b))
|| ! noce_operand_ok (SET_SRC (set_b))
|| reg_overlap_mentioned_p (x, SET_SRC (set_b))
|| modified_between_p (SET_SRC (set_b), insn_b, jump)
/* Avoid extending the lifetime of hard registers on small
register class machines. */
|| (REG_P (SET_SRC (set_b))
&& HARD_REGISTER_P (SET_SRC (set_b))
&& targetm.small_register_classes_for_mode_p
(GET_MODE (SET_SRC (set_b))))
/* Likewise with X. In particular this can happen when
noce_get_condition looks farther back in the instruction
stream than one might expect. */
|| reg_overlap_mentioned_p (x, cond)
|| reg_overlap_mentioned_p (x, a)
|| modified_between_p (x, insn_b, jump))
{
insn_b = NULL;
set_b = NULL_RTX;
}
}
/* If x has side effects then only the if-then-else form is safe to
convert. But even in that case we would need to restore any notes
(such as REG_INC) at then end. That can be tricky if
noce_emit_move_insn expands to more than one insn, so disable the
optimization entirely for now if there are side effects. */
if (side_effects_p (x))
return FALSE;
b = (set_b ? SET_SRC (set_b) : x);
/* Only operate on register destinations, and even then avoid extending
the lifetime of hard registers on small register class machines. */
orig_x = x;
if (!REG_P (x)
|| (HARD_REGISTER_P (x)
&& targetm.small_register_classes_for_mode_p (GET_MODE (x))))
{
if (GET_MODE (x) == BLKmode)
return FALSE;
if (GET_CODE (x) == ZERO_EXTRACT
&& (!CONST_INT_P (XEXP (x, 1))
|| !CONST_INT_P (XEXP (x, 2))))
return FALSE;
x = gen_reg_rtx (GET_MODE (GET_CODE (x) == STRICT_LOW_PART
? XEXP (x, 0) : x));
}
/* Don't operate on sources that may trap or are volatile. */
if (! noce_operand_ok (a) || ! noce_operand_ok (b))
return FALSE;
retry:
/* Set up the info block for our subroutines. */
if_info->insn_a = insn_a;
if_info->insn_b = insn_b;
if_info->x = x;
if_info->a = a;
if_info->b = b;
/* Skip it if the instruction to be moved might clobber CC. */
cc = cc_in_cond (cond);
if (cc
&& (set_of (cc, insn_a)
|| (insn_b && set_of (cc, insn_b))))
return FALSE;
/* Try optimizations in some approximation of a useful order. */
/* ??? Should first look to see if X is live incoming at all. If it
isn't, we don't need anything but an unconditional set. */
/* Look and see if A and B are really the same. Avoid creating silly
cmove constructs that no one will fix up later. */
if (rtx_interchangeable_p (a, b))
{
/* If we have an INSN_B, we don't have to create any new rtl. Just
move the instruction that we already have. If we don't have an
INSN_B, that means that A == X, and we've got a noop move. In
that case don't do anything and let the code below delete INSN_A. */
if (insn_b && else_bb)
{
rtx note;
if (else_bb && insn_b == BB_END (else_bb))
BB_END (else_bb) = PREV_INSN (insn_b);
reorder_insns (insn_b, insn_b, PREV_INSN (jump));
/* If there was a REG_EQUAL note, delete it since it may have been
true due to this insn being after a jump. */
if ((note = find_reg_note (insn_b, REG_EQUAL, NULL_RTX)) != 0)
remove_note (insn_b, note);
insn_b = NULL;
}
/* If we have "x = b; if (...) x = a;", and x has side-effects, then
x must be executed twice. */
else if (insn_b && side_effects_p (orig_x))
return FALSE;
x = orig_x;
goto success;
}
if (!set_b && MEM_P (orig_x))
{
/* Disallow the "if (...) x = a;" form (implicit "else x = x;")
for optimizations if writing to x may trap or fault,
i.e. it's a memory other than a static var or a stack slot,
is misaligned on strict aligned machines or is read-only. If
x is a read-only memory, then the program is valid only if we
avoid the store into it. If there are stores on both the
THEN and ELSE arms, then we can go ahead with the conversion;
either the program is broken, or the condition is always
false such that the other memory is selected. */
if (noce_mem_write_may_trap_or_fault_p (orig_x))
return FALSE;
/* Avoid store speculation: given "if (...) x = a" where x is a
MEM, we only want to do the store if x is always set
somewhere in the function. This avoids cases like
if (pthread_mutex_trylock(mutex))
++global_variable;
where we only want global_variable to be changed if the mutex
is held. FIXME: This should ideally be expressed directly in
RTL somehow. */
if (!noce_can_store_speculate_p (test_bb, orig_x))
return FALSE;
}
if (noce_try_move (if_info))
goto success;
if (noce_try_store_flag (if_info))
goto success;
if (noce_try_bitop (if_info))
goto success;
if (noce_try_minmax (if_info))
goto success;
if (noce_try_abs (if_info))
goto success;
if (HAVE_conditional_move
&& noce_try_cmove (if_info))
goto success;
if (! targetm.have_conditional_execution ())
{
if (noce_try_store_flag_constants (if_info))
goto success;
if (noce_try_addcc (if_info))
goto success;
if (noce_try_store_flag_mask (if_info))
goto success;
if (HAVE_conditional_move
&& noce_try_cmove_arith (if_info))
goto success;
if (noce_try_sign_mask (if_info))
goto success;
}
if (!else_bb && set_b)
{
insn_b = NULL;
set_b = NULL_RTX;
b = orig_x;
goto retry;
}
return FALSE;
success:
/* If we used a temporary, fix it up now. */
if (orig_x != x)
{
rtx_insn *seq;
start_sequence ();
noce_emit_move_insn (orig_x, x);
seq = get_insns ();
set_used_flags (orig_x);
unshare_all_rtl_in_chain (seq);
end_sequence ();
emit_insn_before_setloc (seq, BB_END (test_bb), INSN_LOCATION (insn_a));
}
/* The original THEN and ELSE blocks may now be removed. The test block
must now jump to the join block. If the test block and the join block
can be merged, do so. */
if (else_bb)
{
delete_basic_block (else_bb);
num_true_changes++;
}
else
remove_edge (find_edge (test_bb, join_bb));
remove_edge (find_edge (then_bb, join_bb));
redirect_edge_and_branch_force (single_succ_edge (test_bb), join_bb);
delete_basic_block (then_bb);
num_true_changes++;
if (can_merge_blocks_p (test_bb, join_bb))
{
merge_blocks (test_bb, join_bb);
num_true_changes++;
}
num_updated_if_blocks++;
return TRUE;
}
/* Check whether a block is suitable for conditional move conversion.
Every insn must be a simple set of a register to a constant or a
register. For each assignment, store the value in the pointer map
VALS, keyed indexed by register pointer, then store the register
pointer in REGS. COND is the condition we will test. */
static int
check_cond_move_block (basic_block bb,
hash_map<rtx, rtx> *vals,
vec<rtx> *regs,
rtx cond)
{
rtx_insn *insn;
rtx cc = cc_in_cond (cond);
/* We can only handle simple jumps at the end of the basic block.
It is almost impossible to update the CFG otherwise. */
insn = BB_END (bb);
if (JUMP_P (insn) && !onlyjump_p (insn))
return FALSE;
FOR_BB_INSNS (bb, insn)
{
rtx set, dest, src;
if (!NONDEBUG_INSN_P (insn) || JUMP_P (insn))
continue;
set = single_set (insn);
if (!set)
return FALSE;
dest = SET_DEST (set);
src = SET_SRC (set);
if (!REG_P (dest)
|| (HARD_REGISTER_P (dest)
&& targetm.small_register_classes_for_mode_p (GET_MODE (dest))))
return FALSE;
if (!CONSTANT_P (src) && !register_operand (src, VOIDmode))
return FALSE;
if (side_effects_p (src) || side_effects_p (dest))
return FALSE;
if (may_trap_p (src) || may_trap_p (dest))
return FALSE;
/* Don't try to handle this if the source register was
modified earlier in the block. */
if ((REG_P (src)
&& vals->get (src))
|| (GET_CODE (src) == SUBREG && REG_P (SUBREG_REG (src))
&& vals->get (SUBREG_REG (src))))
return FALSE;
/* Don't try to handle this if the destination register was
modified earlier in the block. */
if (vals->get (dest))
return FALSE;
/* Don't try to handle this if the condition uses the
destination register. */
if (reg_overlap_mentioned_p (dest, cond))
return FALSE;
/* Don't try to handle this if the source register is modified
later in the block. */
if (!CONSTANT_P (src)
&& modified_between_p (src, insn, NEXT_INSN (BB_END (bb))))
return FALSE;
/* Skip it if the instruction to be moved might clobber CC. */
if (cc && set_of (cc, insn))
return FALSE;
vals->put (dest, src);
regs->safe_push (dest);
}
return TRUE;
}
/* Given a basic block BB suitable for conditional move conversion,
a condition COND, and pointer maps THEN_VALS and ELSE_VALS containing
the register values depending on COND, emit the insns in the block as
conditional moves. If ELSE_BLOCK is true, THEN_BB was already
processed. The caller has started a sequence for the conversion.
Return true if successful, false if something goes wrong. */
static bool
cond_move_convert_if_block (struct noce_if_info *if_infop,
basic_block bb, rtx cond,
hash_map<rtx, rtx> *then_vals,
hash_map<rtx, rtx> *else_vals,
bool else_block_p)
{
enum rtx_code code;
rtx_insn *insn;
rtx cond_arg0, cond_arg1;
code = GET_CODE (cond);
cond_arg0 = XEXP (cond, 0);
cond_arg1 = XEXP (cond, 1);
FOR_BB_INSNS (bb, insn)
{
rtx set, target, dest, t, e;
/* ??? Maybe emit conditional debug insn? */
if (!NONDEBUG_INSN_P (insn) || JUMP_P (insn))
continue;
set = single_set (insn);
gcc_assert (set && REG_P (SET_DEST (set)));
dest = SET_DEST (set);
rtx *then_slot = then_vals->get (dest);
rtx *else_slot = else_vals->get (dest);
t = then_slot ? *then_slot : NULL_RTX;
e = else_slot ? *else_slot : NULL_RTX;
if (else_block_p)
{
/* If this register was set in the then block, we already
handled this case there. */
if (t)
continue;
t = dest;
gcc_assert (e);
}
else
{
gcc_assert (t);
if (!e)
e = dest;
}
target = noce_emit_cmove (if_infop, dest, code, cond_arg0, cond_arg1,
t, e);
if (!target)
return false;
if (target != dest)
noce_emit_move_insn (dest, target);
}
return true;
}
/* Given a simple IF-THEN-JOIN or IF-THEN-ELSE-JOIN block, attempt to convert
it using only conditional moves. Return TRUE if we were successful at
converting the block. */
static int
cond_move_process_if_block (struct noce_if_info *if_info)
{
basic_block test_bb = if_info->test_bb;
basic_block then_bb = if_info->then_bb;
basic_block else_bb = if_info->else_bb;
basic_block join_bb = if_info->join_bb;
rtx_insn *jump = if_info->jump;
rtx cond = if_info->cond;
rtx_insn *seq, *loc_insn;
rtx reg;
int c;
vec<rtx> then_regs = vNULL;
vec<rtx> else_regs = vNULL;
unsigned int i;
int success_p = FALSE;
/* Build a mapping for each block to the value used for each
register. */
hash_map<rtx, rtx> then_vals;
hash_map<rtx, rtx> else_vals;
/* Make sure the blocks are suitable. */
if (!check_cond_move_block (then_bb, &then_vals, &then_regs, cond)
|| (else_bb
&& !check_cond_move_block (else_bb, &else_vals, &else_regs, cond)))
goto done;
/* Make sure the blocks can be used together. If the same register
is set in both blocks, and is not set to a constant in both
cases, then both blocks must set it to the same register. We
have already verified that if it is set to a register, that the
source register does not change after the assignment. Also count
the number of registers set in only one of the blocks. */
c = 0;
FOR_EACH_VEC_ELT (then_regs, i, reg)
{
rtx *then_slot = then_vals.get (reg);
rtx *else_slot = else_vals.get (reg);
gcc_checking_assert (then_slot);
if (!else_slot)
++c;
else
{
rtx then_val = *then_slot;
rtx else_val = *else_slot;
if (!CONSTANT_P (then_val) && !CONSTANT_P (else_val)
&& !rtx_equal_p (then_val, else_val))
goto done;
}
}
/* Finish off c for MAX_CONDITIONAL_EXECUTE. */
FOR_EACH_VEC_ELT (else_regs, i, reg)
{
gcc_checking_assert (else_vals.get (reg));
if (!then_vals.get (reg))
++c;
}
/* Make sure it is reasonable to convert this block. What matters
is the number of assignments currently made in only one of the
branches, since if we convert we are going to always execute
them. */
if (c > MAX_CONDITIONAL_EXECUTE)
goto done;
/* Try to emit the conditional moves. First do the then block,
then do anything left in the else blocks. */
start_sequence ();
if (!cond_move_convert_if_block (if_info, then_bb, cond,
&then_vals, &else_vals, false)
|| (else_bb
&& !cond_move_convert_if_block (if_info, else_bb, cond,
&then_vals, &else_vals, true)))
{
end_sequence ();
goto done;
}
seq = end_ifcvt_sequence (if_info);
if (!seq)
goto done;
loc_insn = first_active_insn (then_bb);
if (!loc_insn)
{
loc_insn = first_active_insn (else_bb);
gcc_assert (loc_insn);
}
emit_insn_before_setloc (seq, jump, INSN_LOCATION (loc_insn));
if (else_bb)
{
delete_basic_block (else_bb);
num_true_changes++;
}
else
remove_edge (find_edge (test_bb, join_bb));
remove_edge (find_edge (then_bb, join_bb));
redirect_edge_and_branch_force (single_succ_edge (test_bb), join_bb);
delete_basic_block (then_bb);
num_true_changes++;
if (can_merge_blocks_p (test_bb, join_bb))
{
merge_blocks (test_bb, join_bb);
num_true_changes++;
}
num_updated_if_blocks++;
success_p = TRUE;
done:
then_regs.release ();
else_regs.release ();
return success_p;
}
/* Determine if a given basic block heads a simple IF-THEN-JOIN or an
IF-THEN-ELSE-JOIN block.
If so, we'll try to convert the insns to not require the branch,
using only transformations that do not require conditional execution.
Return TRUE if we were successful at converting the block. */
static int
noce_find_if_block (basic_block test_bb, edge then_edge, edge else_edge,
int pass)
{
basic_block then_bb, else_bb, join_bb;
bool then_else_reversed = false;
rtx_insn *jump;
rtx cond;
rtx_insn *cond_earliest;
struct noce_if_info if_info;
/* We only ever should get here before reload. */
gcc_assert (!reload_completed);
/* Recognize an IF-THEN-ELSE-JOIN block. */
if (single_pred_p (then_edge->dest)
&& single_succ_p (then_edge->dest)
&& single_pred_p (else_edge->dest)
&& single_succ_p (else_edge->dest)
&& single_succ (then_edge->dest) == single_succ (else_edge->dest))
{
then_bb = then_edge->dest;
else_bb = else_edge->dest;
join_bb = single_succ (then_bb);
}
/* Recognize an IF-THEN-JOIN block. */
else if (single_pred_p (then_edge->dest)
&& single_succ_p (then_edge->dest)
&& single_succ (then_edge->dest) == else_edge->dest)
{
then_bb = then_edge->dest;
else_bb = NULL_BLOCK;
join_bb = else_edge->dest;
}
/* Recognize an IF-ELSE-JOIN block. We can have those because the order
of basic blocks in cfglayout mode does not matter, so the fallthrough
edge can go to any basic block (and not just to bb->next_bb, like in
cfgrtl mode). */
else if (single_pred_p (else_edge->dest)
&& single_succ_p (else_edge->dest)
&& single_succ (else_edge->dest<