/* Optimize by combining instructions for GNU compiler.
Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
- 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
+ 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
Free Software Foundation, Inc.
This file is part of GCC.
#include "expr.h"
#include "insn-attr.h"
#include "recog.h"
-#include "real.h"
+#include "diagnostic-core.h"
#include "toplev.h"
#include "target.h"
#include "optabs.h"
static int label_tick;
-/* Reset to label_tick for each label. */
+/* Reset to label_tick for each extended basic block in scanning order. */
static int label_tick_ebb_start;
static void setup_incoming_promotions (rtx);
static void set_nonzero_bits_and_sign_copies (rtx, const_rtx, void *);
static int cant_combine_insn_p (rtx);
-static int can_combine_p (rtx, rtx, rtx, rtx, rtx *, rtx *);
-static int combinable_i3pat (rtx, rtx *, rtx, rtx, int, rtx *);
+static int can_combine_p (rtx, rtx, rtx, rtx, rtx, rtx, rtx *, rtx *);
+static int combinable_i3pat (rtx, rtx *, rtx, rtx, rtx, int, int, rtx *);
static int contains_muldiv (rtx);
-static rtx try_combine (rtx, rtx, rtx, int *);
+static rtx try_combine (rtx, rtx, rtx, rtx, int *);
static void undo_all (void);
static void undo_commit (void);
-static rtx *find_split_point (rtx *, rtx);
+static rtx *find_split_point (rtx *, rtx, bool);
static rtx subst (rtx, rtx, rtx, int, int);
static rtx combine_simplify_rtx (rtx, enum machine_mode, int);
static rtx simplify_if_then_else (rtx);
static int reg_dead_at_p (rtx, rtx);
static void move_deaths (rtx, rtx, int, rtx, rtx *);
static int reg_bitfield_target_p (rtx, rtx);
-static void distribute_notes (rtx, rtx, rtx, rtx, rtx, rtx);
+static void distribute_notes (rtx, rtx, rtx, rtx, rtx, rtx, rtx);
static void distribute_links (rtx);
static void mark_used_regs_combine (rtx);
static void record_promoted_value (rtx, rtx);
\f
/* Subroutine of try_combine. Determine whether the combine replacement
patterns NEWPAT, NEWI2PAT and NEWOTHERPAT are cheaper according to
- insn_rtx_cost that the original instruction sequence I1, I2, I3 and
- undobuf.other_insn. Note that I1 and/or NEWI2PAT may be NULL_RTX.
+ insn_rtx_cost that the original instruction sequence I0, I1, I2, I3 and
+ undobuf.other_insn. Note that I1 and/or NEWI2PAT may be NULL_RTX.
NEWOTHERPAT and undobuf.other_insn may also both be NULL_RTX. This
function returns false, if the costs of all instructions can be
estimated, and the replacements are more expensive than the original
sequence. */
static bool
-combine_validate_cost (rtx i1, rtx i2, rtx i3, rtx newpat, rtx newi2pat,
- rtx newotherpat)
+combine_validate_cost (rtx i0, rtx i1, rtx i2, rtx i3, rtx newpat,
+ rtx newi2pat, rtx newotherpat)
{
- int i1_cost, i2_cost, i3_cost;
+ int i0_cost, i1_cost, i2_cost, i3_cost;
int new_i2_cost, new_i3_cost;
int old_cost, new_cost;
if (i1)
{
i1_cost = INSN_COST (i1);
- old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0)
- ? i1_cost + i2_cost + i3_cost : 0;
+ if (i0)
+ {
+ i0_cost = INSN_COST (i0);
+ old_cost = (i0_cost > 0 && i1_cost > 0 && i2_cost > 0 && i3_cost > 0
+ ? i0_cost + i1_cost + i2_cost + i3_cost : 0);
+ }
+ else
+ {
+ old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0
+ ? i1_cost + i2_cost + i3_cost : 0);
+ i0_cost = 0;
+ }
}
else
{
old_cost = (i2_cost > 0 && i3_cost > 0) ? i2_cost + i3_cost : 0;
- i1_cost = 0;
+ i1_cost = i0_cost = 0;
}
/* Calculate the replacement insn_rtx_costs. */
{
if (dump_file)
{
- if (i1)
+ if (i0)
+ {
+ fprintf (dump_file,
+ "rejecting combination of insns %d, %d, %d and %d\n",
+ INSN_UID (i0), INSN_UID (i1), INSN_UID (i2),
+ INSN_UID (i3));
+ fprintf (dump_file, "original costs %d + %d + %d + %d = %d\n",
+ i0_cost, i1_cost, i2_cost, i3_cost, old_cost);
+ }
+ else if (i1)
{
fprintf (dump_file,
"rejecting combination of insns %d, %d and %d\n",
register and establishing log links when def is encountered.
Note that we do not clear next_use array in order to save time,
so we have to test whether the use is in the same basic block as def.
-
+
There are a few cases below when we do not consider the definition or
usage -- these are taken from original flow.c did. Don't ask me why it is
done this way; I don't know and if it works, I don't want to know. */
free_INSN_LIST_list (&LOG_LINKS (insn));
}
+/* Walk the LOG_LINKS of insn B to see if we find a reference to A. Return
+ true if we found a LOG_LINK that proves that A feeds B. This only works
+ if there are no instructions between A and B which could have a link
+ depending on A, since in that case we would not record a link for B. */
-
+static bool
+insn_a_feeds_b (rtx a, rtx b)
+{
+ rtx links;
+ for (links = LOG_LINKS (b); links; links = XEXP (links, 1))
+ if (XEXP (links, 0) == a)
+ return true;
+ return false;
+}
\f
/* Main entry point for combiner. F is the first insn of the function.
NREGS is the first unused pseudo-reg number.
#endif
rtx links, nextlinks;
rtx first;
+ basic_block last_bb;
int new_direct_jump_p = 0;
problems when, for example, we have j <<= 1 in a loop. */
nonzero_sign_valid = 0;
+ label_tick = label_tick_ebb_start = 1;
/* Scan all SETs and see if we can deduce anything about what
bits are known to be zero for some registers and how many copies
for what bits are known to be set. */
setup_incoming_promotions (first);
+ /* Allow the entry block and the first block to fall into the same EBB.
+ Conceptually the incoming promotions are assigned to the entry block. */
+ last_bb = ENTRY_BLOCK_PTR;
create_log_links ();
- label_tick_ebb_start = ENTRY_BLOCK_PTR->index;
FOR_EACH_BB (this_basic_block)
{
optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
last_call_luid = 0;
mem_last_set = -1;
- label_tick = this_basic_block->index;
+
+ label_tick++;
if (!single_pred_p (this_basic_block)
- || single_pred (this_basic_block)->index != label_tick - 1)
+ || single_pred (this_basic_block) != last_bb)
label_tick_ebb_start = label_tick;
+ last_bb = this_basic_block;
+
FOR_BB_INSNS (this_basic_block, insn)
if (INSN_P (insn) && BLOCK_FOR_INSN (insn))
{
nonzero_sign_valid = 1;
/* Now scan all the insns in forward order. */
-
- label_tick_ebb_start = ENTRY_BLOCK_PTR->index;
+ label_tick = label_tick_ebb_start = 1;
init_reg_last ();
setup_incoming_promotions (first);
+ last_bb = ENTRY_BLOCK_PTR;
FOR_EACH_BB (this_basic_block)
{
optimize_this_for_speed_p = optimize_bb_for_speed_p (this_basic_block);
last_call_luid = 0;
mem_last_set = -1;
- label_tick = this_basic_block->index;
+
+ label_tick++;
if (!single_pred_p (this_basic_block)
- || single_pred (this_basic_block)->index != label_tick - 1)
+ || single_pred (this_basic_block) != last_bb)
label_tick_ebb_start = label_tick;
+ last_bb = this_basic_block;
+
rtl_profile_for_bb (this_basic_block);
for (insn = BB_HEAD (this_basic_block);
insn != NEXT_INSN (BB_END (this_basic_block));
/* Try this insn with each insn it links back to. */
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
- if ((next = try_combine (insn, XEXP (links, 0),
+ if ((next = try_combine (insn, XEXP (links, 0), NULL_RTX,
NULL_RTX, &new_direct_jump_p)) != 0)
goto retry;
for (nextlinks = LOG_LINKS (link);
nextlinks;
nextlinks = XEXP (nextlinks, 1))
- if ((next = try_combine (insn, link,
- XEXP (nextlinks, 0),
+ if ((next = try_combine (insn, link, XEXP (nextlinks, 0),
+ NULL_RTX,
&new_direct_jump_p)) != 0)
goto retry;
}
&& NONJUMP_INSN_P (prev)
&& sets_cc0_p (PATTERN (prev)))
{
- if ((next = try_combine (insn, prev,
- NULL_RTX, &new_direct_jump_p)) != 0)
+ if ((next = try_combine (insn, prev, NULL_RTX, NULL_RTX,
+ &new_direct_jump_p)) != 0)
goto retry;
for (nextlinks = LOG_LINKS (prev); nextlinks;
nextlinks = XEXP (nextlinks, 1))
- if ((next = try_combine (insn, prev,
- XEXP (nextlinks, 0),
+ if ((next = try_combine (insn, prev, XEXP (nextlinks, 0),
+ NULL_RTX,
&new_direct_jump_p)) != 0)
goto retry;
}
&& GET_CODE (PATTERN (insn)) == SET
&& reg_mentioned_p (cc0_rtx, SET_SRC (PATTERN (insn))))
{
- if ((next = try_combine (insn, prev,
- NULL_RTX, &new_direct_jump_p)) != 0)
+ if ((next = try_combine (insn, prev, NULL_RTX, NULL_RTX,
+ &new_direct_jump_p)) != 0)
goto retry;
for (nextlinks = LOG_LINKS (prev); nextlinks;
nextlinks = XEXP (nextlinks, 1))
- if ((next = try_combine (insn, prev,
- XEXP (nextlinks, 0),
+ if ((next = try_combine (insn, prev, XEXP (nextlinks, 0),
+ NULL_RTX,
&new_direct_jump_p)) != 0)
goto retry;
}
&& NONJUMP_INSN_P (prev)
&& sets_cc0_p (PATTERN (prev))
&& (next = try_combine (insn, XEXP (links, 0),
- prev, &new_direct_jump_p)) != 0)
+ prev, NULL_RTX,
+ &new_direct_jump_p)) != 0)
goto retry;
#endif
for (nextlinks = XEXP (links, 1); nextlinks;
nextlinks = XEXP (nextlinks, 1))
if ((next = try_combine (insn, XEXP (links, 0),
- XEXP (nextlinks, 0),
+ XEXP (nextlinks, 0), NULL_RTX,
&new_direct_jump_p)) != 0)
goto retry;
+ /* Try four-instruction combinations. */
+ for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
+ {
+ rtx next1;
+ rtx link = XEXP (links, 0);
+
+ /* If the linked insn has been replaced by a note, then there
+ is no point in pursuing this chain any further. */
+ if (NOTE_P (link))
+ continue;
+
+ for (next1 = LOG_LINKS (link); next1; next1 = XEXP (next1, 1))
+ {
+ rtx link1 = XEXP (next1, 0);
+ if (NOTE_P (link1))
+ continue;
+ /* I0 -> I1 -> I2 -> I3. */
+ for (nextlinks = LOG_LINKS (link1); nextlinks;
+ nextlinks = XEXP (nextlinks, 1))
+ if ((next = try_combine (insn, link, link1,
+ XEXP (nextlinks, 0),
+ &new_direct_jump_p)) != 0)
+ goto retry;
+ /* I0, I1 -> I2, I2 -> I3. */
+ for (nextlinks = XEXP (next1, 1); nextlinks;
+ nextlinks = XEXP (nextlinks, 1))
+ if ((next = try_combine (insn, link, link1,
+ XEXP (nextlinks, 0),
+ &new_direct_jump_p)) != 0)
+ goto retry;
+ }
+
+ for (next1 = XEXP (links, 1); next1; next1 = XEXP (next1, 1))
+ {
+ rtx link1 = XEXP (next1, 0);
+ if (NOTE_P (link1))
+ continue;
+ /* I0 -> I2; I1, I2 -> I3. */
+ for (nextlinks = LOG_LINKS (link); nextlinks;
+ nextlinks = XEXP (nextlinks, 1))
+ if ((next = try_combine (insn, link, link1,
+ XEXP (nextlinks, 0),
+ &new_direct_jump_p)) != 0)
+ goto retry;
+ /* I0 -> I1; I1, I2 -> I3. */
+ for (nextlinks = LOG_LINKS (link1); nextlinks;
+ nextlinks = XEXP (nextlinks, 1))
+ if ((next = try_combine (insn, link, link1,
+ XEXP (nextlinks, 0),
+ &new_direct_jump_p)) != 0)
+ goto retry;
+ }
+ }
+
/* Try this insn with each REG_EQUAL note it links back to. */
for (links = LOG_LINKS (insn); links; links = XEXP (links, 1))
{
i2mod = temp;
i2mod_old_rhs = copy_rtx (orig);
i2mod_new_rhs = copy_rtx (note);
- next = try_combine (insn, i2mod, NULL_RTX,
+ next = try_combine (insn, i2mod, NULL_RTX, NULL_RTX,
&new_direct_jump_p);
i2mod = NULL_RTX;
if (next)
unsigned int i;
reg_stat_type *p;
- for (i = 0; VEC_iterate (reg_stat_type, reg_stat, i, p); ++i)
+ FOR_EACH_VEC_ELT (reg_stat_type, reg_stat, i, p)
memset (p, 0, offsetof (reg_stat_type, sign_bit_copies));
}
\f
bool strictly_local = false;
for (arg = DECL_ARGUMENTS (current_function_decl); arg;
- arg = TREE_CHAIN (arg))
+ arg = DECL_CHAIN (arg))
{
- rtx reg = DECL_INCOMING_RTL (arg);
+ rtx x, reg = DECL_INCOMING_RTL (arg);
int uns1, uns3;
enum machine_mode mode1, mode2, mode3, mode4;
mode2 = TYPE_MODE (DECL_ARG_TYPE (arg));
uns3 = TYPE_UNSIGNED (DECL_ARG_TYPE (arg));
- /* The mode and signedness of the argument as it is actually passed,
+ /* The mode and signedness of the argument as it is actually passed,
after any TARGET_PROMOTE_FUNCTION_ARGS-driven ABI promotions. */
mode3 = promote_function_mode (DECL_ARG_TYPE (arg), mode2, &uns3,
TREE_TYPE (cfun->decl), 0);
/* The mode of the register in which the argument is being passed. */
mode4 = GET_MODE (reg);
- /* Eliminate sign extensions in the callee when possible. Only
- do this when:
- (a) a mode promotion has occurred;
- (b) the mode of the register is the same as the mode of
- the argument as it is passed; and
- (c) the signedness does not change across any of the promotions; and
- (d) when no language-level promotions (which we cannot guarantee
- will have been done by an external caller) are necessary,
- unless we know that this function is only ever called from
- the current compilation unit -- all of whose call sites will
- do the mode1 --> mode2 promotion. */
- if (mode1 != mode3
- && mode3 == mode4
- && uns1 == uns3
- && (mode1 == mode2 || strictly_local))
- {
- /* Record that the value was promoted from mode1 to mode3,
- so that any sign extension at the head of the current
- function may be eliminated. */
- rtx x;
- x = gen_rtx_CLOBBER (mode1, const0_rtx);
- x = gen_rtx_fmt_e ((uns3 ? ZERO_EXTEND : SIGN_EXTEND), mode3, x);
- record_value_for_reg (reg, first, x);
- }
+ /* Eliminate sign extensions in the callee when:
+ (a) A mode promotion has occurred; */
+ if (mode1 == mode3)
+ continue;
+ /* (b) The mode of the register is the same as the mode of
+ the argument as it is passed; */
+ if (mode3 != mode4)
+ continue;
+ /* (c) There's no language level extension; */
+ if (mode1 == mode2)
+ ;
+ /* (c.1) All callers are from the current compilation unit. If that's
+ the case we don't have to rely on an ABI, we only have to know
+ what we're generating right now, and we know that we will do the
+ mode1 to mode2 promotion with the given sign. */
+ else if (!strictly_local)
+ continue;
+ /* (c.2) The combination of the two promotions is useful. This is
+ true when the signs match, or if the first promotion is unsigned.
+ In the later case, (sign_extend (zero_extend x)) is the same as
+ (zero_extend (zero_extend x)), so make sure to force UNS3 true. */
+ else if (uns1)
+ uns3 = true;
+ else if (uns3)
+ continue;
+
+ /* Record that the value was promoted from mode1 to mode3,
+ so that any sign extension at the head of the current
+ function may be eliminated. */
+ x = gen_rtx_CLOBBER (mode1, const0_rtx);
+ x = gen_rtx_fmt_e ((uns3 ? ZERO_EXTEND : SIGN_EXTEND), mode3, x);
+ record_value_for_reg (reg, first, x);
}
}
}
}
\f
-/* See if INSN can be combined into I3. PRED and SUCC are optionally
- insns that were previously combined into I3 or that will be combined
- into the merger of INSN and I3.
+/* See if INSN can be combined into I3. PRED, PRED2, SUCC and SUCC2 are
+ optionally insns that were previously combined into I3 or that will be
+ combined into the merger of INSN and I3. The order is PRED, PRED2,
+ INSN, SUCC, SUCC2, I3.
Return 0 if the combination is not allowed for any reason.
will return 1. */
static int
-can_combine_p (rtx insn, rtx i3, rtx pred ATTRIBUTE_UNUSED, rtx succ,
+can_combine_p (rtx insn, rtx i3, rtx pred ATTRIBUTE_UNUSED,
+ rtx pred2 ATTRIBUTE_UNUSED, rtx succ, rtx succ2,
rtx *pdest, rtx *psrc)
{
int i;
#ifdef AUTO_INC_DEC
rtx link;
#endif
- int all_adjacent = (succ ? (next_active_insn (insn) == succ
- && next_active_insn (succ) == i3)
- : next_active_insn (insn) == i3);
+ bool all_adjacent = true;
+ if (succ)
+ {
+ if (succ2)
+ {
+ if (next_active_insn (succ2) != i3)
+ all_adjacent = false;
+ if (next_active_insn (succ) != succ2)
+ all_adjacent = false;
+ }
+ else if (next_active_insn (succ) != i3)
+ all_adjacent = false;
+ if (next_active_insn (insn) != succ)
+ all_adjacent = false;
+ }
+ else if (next_active_insn (insn) != i3)
+ all_adjacent = false;
+
/* Can combine only if previous insn is a SET of a REG, a SUBREG or CC0.
or a PARALLEL consisting of such a SET and CLOBBERs.
/* Don't substitute into an incremented register. */
|| FIND_REG_INC_NOTE (i3, dest)
|| (succ && FIND_REG_INC_NOTE (succ, dest))
+ || (succ2 && FIND_REG_INC_NOTE (succ2, dest))
/* Don't substitute into a non-local goto, this confuses CFG. */
|| (JUMP_P (i3) && find_reg_note (i3, REG_NON_LOCAL_GOTO, NULL_RTX))
/* Make sure that DEST is not used after SUCC but before I3. */
- || (succ && ! all_adjacent
- && reg_used_between_p (dest, succ, i3))
+ || (!all_adjacent
+ && ((succ2
+ && (reg_used_between_p (dest, succ2, i3)
+ || reg_used_between_p (dest, succ, succ2)))
+ || (!succ2 && succ && reg_used_between_p (dest, succ, i3))))
/* Make sure that the value that is to be substituted for the register
does not use any registers whose values alter in between. However,
If the insns are adjacent, a use can't cross a set even though we
if (GET_CODE (src) == ASM_OPERANDS || volatile_refs_p (src))
{
- /* Make sure succ doesn't contain a volatile reference. */
+ /* Make sure neither succ nor succ2 contains a volatile reference. */
+ if (succ2 != 0 && volatile_refs_p (PATTERN (succ2)))
+ return 0;
if (succ != 0 && volatile_refs_p (PATTERN (succ)))
return 0;
-
- for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
- if (INSN_P (p) && p != succ && volatile_refs_p (PATTERN (p)))
- return 0;
+ /* We'll check insns between INSN and I3 below. */
}
/* If INSN is an asm, and DEST is a hard register, reject, since it has
they might affect machine state. */
for (p = NEXT_INSN (insn); p != i3; p = NEXT_INSN (p))
- if (INSN_P (p) && p != succ && volatile_insn_p (PATTERN (p)))
+ if (INSN_P (p) && p != succ && p != succ2 && volatile_insn_p (PATTERN (p)))
return 0;
/* If INSN contains an autoincrement or autodecrement, make sure that
|| reg_used_between_p (XEXP (link, 0), insn, i3)
|| (pred != NULL_RTX
&& reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred)))
+ || (pred2 != NULL_RTX
+ && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (pred2)))
|| (succ != NULL_RTX
&& reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ)))
+ || (succ2 != NULL_RTX
+ && reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (succ2)))
|| reg_overlap_mentioned_p (XEXP (link, 0), PATTERN (i3))))
return 0;
#endif
of a PARALLEL of the pattern. We validate that it is valid for combining.
One problem is if I3 modifies its output, as opposed to replacing it
- entirely, we can't allow the output to contain I2DEST or I1DEST as doing
- so would produce an insn that is not equivalent to the original insns.
+ entirely, we can't allow the output to contain I2DEST, I1DEST or I0DEST as
+ doing so would produce an insn that is not equivalent to the original insns.
Consider:
must reject the combination. This case occurs when I2 and I1 both
feed into I3, rather than when I1 feeds into I2, which feeds into I3.
If I1_NOT_IN_SRC is nonzero, it means that finding I1 in the source
- of a SET must prevent combination from occurring.
+ of a SET must prevent combination from occurring. The same situation
+ can occur for I0, in which case I0_NOT_IN_SRC is set.
Before doing the above check, we first try to expand a field assignment
into a set of logical operations.
Return 1 if the combination is valid, zero otherwise. */
static int
-combinable_i3pat (rtx i3, rtx *loc, rtx i2dest, rtx i1dest,
- int i1_not_in_src, rtx *pi3dest_killed)
+combinable_i3pat (rtx i3, rtx *loc, rtx i2dest, rtx i1dest, rtx i0dest,
+ int i1_not_in_src, int i0_not_in_src, rtx *pi3dest_killed)
{
rtx x = *loc;
if ((inner_dest != dest &&
(!MEM_P (inner_dest)
|| rtx_equal_p (i2dest, inner_dest)
- || (i1dest && rtx_equal_p (i1dest, inner_dest)))
+ || (i1dest && rtx_equal_p (i1dest, inner_dest))
+ || (i0dest && rtx_equal_p (i0dest, inner_dest)))
&& (reg_overlap_mentioned_p (i2dest, inner_dest)
- || (i1dest && reg_overlap_mentioned_p (i1dest, inner_dest))))
+ || (i1dest && reg_overlap_mentioned_p (i1dest, inner_dest))
+ || (i0dest && reg_overlap_mentioned_p (i0dest, inner_dest))))
/* This is the same test done in can_combine_p except we can't test
all_adjacent; we don't have to, since this instruction will stay
&& REGNO (inner_dest) < FIRST_PSEUDO_REGISTER
&& (! HARD_REGNO_MODE_OK (REGNO (inner_dest),
GET_MODE (inner_dest))))
- || (i1_not_in_src && reg_overlap_mentioned_p (i1dest, src)))
+ || (i1_not_in_src && reg_overlap_mentioned_p (i1dest, src))
+ || (i0_not_in_src && reg_overlap_mentioned_p (i0dest, src)))
return 0;
/* If DEST is used in I3, it is being killed in this insn, so
int i;
for (i = 0; i < XVECLEN (x, 0); i++)
- if (! combinable_i3pat (i3, &XVECEXP (x, 0, i), i2dest, i1dest,
- i1_not_in_src, pi3dest_killed))
+ if (! combinable_i3pat (i3, &XVECEXP (x, 0, i), i2dest, i1dest, i0dest,
+ i1_not_in_src, i0_not_in_src, pi3dest_killed))
return 0;
}
if (GET_CODE (dest) == SUBREG)
dest = SUBREG_REG (dest);
if (REG_P (src) && REG_P (dest)
- && ((REGNO (src) < FIRST_PSEUDO_REGISTER
- && ! fixed_regs[REGNO (src)]
- && CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (src))))
- || (REGNO (dest) < FIRST_PSEUDO_REGISTER
- && ! fixed_regs[REGNO (dest)]
- && CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (dest))))))
+ && ((HARD_REGISTER_P (src)
+ && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (src))
+ && targetm.class_likely_spilled_p (REGNO_REG_CLASS (REGNO (src))))
+ || (HARD_REGISTER_P (dest)
+ && ! TEST_HARD_REG_BIT (fixed_reg_set, REGNO (dest))
+ && targetm.class_likely_spilled_p (REGNO_REG_CLASS (REGNO (dest))))))
return 1;
return 0;
unsigned regno, nregs;
/* We assume here that no machine mode needs more than
32 hard registers when the value overlaps with a register
- for which FUNCTION_VALUE_REGNO_P is true. */
+ for which TARGET_FUNCTION_VALUE_REGNO_P is true. */
unsigned mask;
struct likely_spilled_retval_info info;
if (!NONJUMP_INSN_P (use) || GET_CODE (PATTERN (use)) != USE || insn == use)
return 0;
reg = XEXP (PATTERN (use), 0);
- if (!REG_P (reg) || !FUNCTION_VALUE_REGNO_P (REGNO (reg)))
+ if (!REG_P (reg) || !targetm.calls.function_value_regno_p (REGNO (reg)))
return 0;
regno = REGNO (reg);
nregs = hard_regno_nregs[regno][GET_MODE (reg)];
do
{
if ((mask & 1 << nregs)
- && CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (regno + nregs)))
+ && targetm.class_likely_spilled_p (REGNO_REG_CLASS (regno + nregs)))
return 1;
} while (nregs--);
return 0;
struct rtx_subst_pair
{
- rtx from, to;
- bool changed;
-#ifdef AUTO_INC_DEC
+ rtx to;
bool adjusted;
bool after;
-#endif
};
-/* Clean up any auto-updates in PAIR->to the first time it is called
- for a PAIR. PAIR->adjusted is used to tell whether we've cleaned
- up before. */
+/* DATA points to an rtx_subst_pair. Return the value that should be
+ substituted. */
-static void
-auto_adjust_pair (struct rtx_subst_pair *pair ATTRIBUTE_UNUSED)
+static rtx
+propagate_for_debug_subst (rtx from, const_rtx old_rtx, void *data)
{
-#ifdef AUTO_INC_DEC
+ struct rtx_subst_pair *pair = (struct rtx_subst_pair *)data;
+
+ if (!rtx_equal_p (from, old_rtx))
+ return NULL_RTX;
if (!pair->adjusted)
{
pair->adjusted = true;
+#ifdef AUTO_INC_DEC
pair->to = cleanup_auto_inc_dec (pair->to, pair->after, VOIDmode);
- }
+#else
+ pair->to = copy_rtx (pair->to);
#endif
-}
-
-/* If *LOC is the same as FROM in the struct rtx_subst_pair passed as
- DATA, replace it with a copy of TO. Handle SUBREGs of *LOC as
- well. */
-
-static int
-propagate_for_debug_subst (rtx *loc, void *data)
-{
- struct rtx_subst_pair *pair = (struct rtx_subst_pair *)data;
- rtx from = pair->from, to = pair->to;
- rtx x = *loc, s = x;
-
- if (rtx_equal_p (x, from)
- || (GET_CODE (x) == SUBREG && rtx_equal_p ((s = SUBREG_REG (x)), from)))
- {
- auto_adjust_pair (pair);
- if (pair->to != to)
- to = pair->to;
- else
- to = copy_rtx (to);
- if (s != x)
- {
- gcc_assert (GET_CODE (x) == SUBREG && SUBREG_REG (x) == s);
- to = simplify_gen_subreg (GET_MODE (x), to,
- GET_MODE (from), SUBREG_BYTE (x));
- }
- *loc = wrap_constant (GET_MODE (x), to);
- pair->changed = true;
- return -1;
+ pair->to = make_compound_operation (pair->to, SET);
+ return pair->to;
}
-
- return 0;
+ return copy_rtx (pair->to);
}
/* Replace occurrences of DEST with SRC in DEBUG_INSNs between INSN
static void
propagate_for_debug (rtx insn, rtx last, rtx dest, rtx src, bool move)
{
- struct rtx_subst_pair p;
- rtx next, move_pos = move ? last : NULL_RTX;
+ rtx next, move_pos = move ? last : NULL_RTX, loc;
- p.from = dest;
+ struct rtx_subst_pair p;
p.to = src;
- p.changed = false;
-
-#ifdef AUTO_INC_DEC
p.adjusted = false;
p.after = move;
-#endif
next = NEXT_INSN (insn);
while (next != last)
next = NEXT_INSN (insn);
if (DEBUG_INSN_P (insn))
{
- for_each_rtx (&INSN_VAR_LOCATION_LOC (insn),
- propagate_for_debug_subst, &p);
- if (!p.changed)
+ loc = simplify_replace_fn_rtx (INSN_VAR_LOCATION_LOC (insn),
+ dest, propagate_for_debug_subst, &p);
+ if (loc == INSN_VAR_LOCATION_LOC (insn))
continue;
- p.changed = false;
+ INSN_VAR_LOCATION_LOC (insn) = loc;
if (move_pos)
{
remove_insn (insn);
}
}
-/* Delete the conditional jump INSN and adjust the CFG correspondingly.
+/* Delete the unconditional jump INSN and adjust the CFG correspondingly.
Note that the INSN should be deleted *after* removing dead edges, so
that the kept edge is the fallthrough edge for a (set (pc) (pc))
but not for a (set (pc) (label_ref FOO)). */
update_cfg_for_uncondjump (rtx insn)
{
basic_block bb = BLOCK_FOR_INSN (insn);
+ bool at_end = (BB_END (bb) == insn);
- if (BB_END (bb) == insn)
+ if (at_end)
purge_dead_edges (bb);
delete_insn (insn);
- if (EDGE_COUNT (bb->succs) == 1)
+ if (at_end && EDGE_COUNT (bb->succs) == 1)
single_succ_edge (bb)->flags |= EDGE_FALLTHRU;
}
+/* Try to combine the insns I0, I1 and I2 into I3.
+ Here I0, I1 and I2 appear earlier than I3.
+ I0 and I1 can be zero; then we combine just I2 into I3, or I1 and I2 into
+ I3.
-/* Try to combine the insns I1 and I2 into I3.
- Here I1 and I2 appear earlier than I3.
- I1 can be zero; then we combine just I2 into I3.
-
- If we are combining three insns and the resulting insn is not recognized,
- try splitting it into two insns. If that happens, I2 and I3 are retained
- and I1 is pseudo-deleted by turning it into a NOTE. Otherwise, I1 and I2
- are pseudo-deleted.
+ If we are combining more than two insns and the resulting insn is not
+ recognized, try splitting it into two insns. If that happens, I2 and I3
+ are retained and I1/I0 are pseudo-deleted by turning them into a NOTE.
+ Otherwise, I0, I1 and I2 are pseudo-deleted.
Return 0 if the combination does not work. Then nothing is changed.
If we did the combination, return the insn at which combine should
new direct jump instruction. */
static rtx
-try_combine (rtx i3, rtx i2, rtx i1, int *new_direct_jump_p)
+try_combine (rtx i3, rtx i2, rtx i1, rtx i0, int *new_direct_jump_p)
{
/* New patterns for I3 and I2, respectively. */
rtx newpat, newi2pat = 0;
rtvec newpat_vec_with_clobbers = 0;
- int substed_i2 = 0, substed_i1 = 0;
- /* Indicates need to preserve SET in I1 or I2 in I3 if it is not dead. */
- int added_sets_1, added_sets_2;
+ int substed_i2 = 0, substed_i1 = 0, substed_i0 = 0;
+ /* Indicates need to preserve SET in I0, I1 or I2 in I3 if it is not
+ dead. */
+ int added_sets_0, added_sets_1, added_sets_2;
/* Total number of SETs to put into I3. */
int total_sets;
- /* Nonzero if I2's body now appears in I3. */
- int i2_is_used;
+ /* Nonzero if I2's or I1's body now appears in I3. */
+ int i2_is_used = 0, i1_is_used = 0;
/* INSN_CODEs for new I3, new I2, and user of condition code. */
int insn_code_number, i2_code_number = 0, other_code_number = 0;
/* Contains I3 if the destination of I3 is used in its source, which means
that the old life of I3 is being killed. If that usage is placed into
I2 and not in I3, a REG_DEAD note must be made. */
rtx i3dest_killed = 0;
- /* SET_DEST and SET_SRC of I2 and I1. */
- rtx i2dest = 0, i2src = 0, i1dest = 0, i1src = 0;
+ /* SET_DEST and SET_SRC of I2, I1 and I0. */
+ rtx i2dest = 0, i2src = 0, i1dest = 0, i1src = 0, i0dest = 0, i0src = 0;
/* Set if I2DEST was reused as a scratch register. */
bool i2scratch = false;
- /* PATTERN (I1) and PATTERN (I2), or a copy of it in certain cases. */
- rtx i1pat = 0, i2pat = 0;
+ /* The PATTERNs of I0, I1, and I2, or a copy of them in certain cases. */
+ rtx i0pat = 0, i1pat = 0, i2pat = 0;
/* Indicates if I2DEST or I1DEST is in I2SRC or I1_SRC. */
int i2dest_in_i2src = 0, i1dest_in_i1src = 0, i2dest_in_i1src = 0;
- int i2dest_killed = 0, i1dest_killed = 0;
- int i1_feeds_i3 = 0;
+ int i0dest_in_i0src = 0, i1dest_in_i0src = 0, i2dest_in_i0src = 0;
+ int i2dest_killed = 0, i1dest_killed = 0, i0dest_killed = 0;
+ int i1_feeds_i2_n = 0, i0_feeds_i2_n = 0, i0_feeds_i1_n = 0;
/* Notes that must be added to REG_NOTES in I3 and I2. */
rtx new_i3_notes, new_i2_notes;
/* Notes that we substituted I3 into I2 instead of the normal case. */
rtx new_other_notes;
int i;
+ /* Only try four-insn combinations when there's high likelihood of
+ success. Look for simple insns, such as loads of constants or
+ binary operations involving a constant. */
+ if (i0)
+ {
+ int i;
+ int ngood = 0;
+ int nshift = 0;
+
+ if (!flag_expensive_optimizations)
+ return 0;
+
+ for (i = 0; i < 4; i++)
+ {
+ rtx insn = i == 0 ? i0 : i == 1 ? i1 : i == 2 ? i2 : i3;
+ rtx set = single_set (insn);
+ rtx src;
+ if (!set)
+ continue;
+ src = SET_SRC (set);
+ if (CONSTANT_P (src))
+ {
+ ngood += 2;
+ break;
+ }
+ else if (BINARY_P (src) && CONSTANT_P (XEXP (src, 1)))
+ ngood++;
+ else if (GET_CODE (src) == ASHIFT || GET_CODE (src) == ASHIFTRT
+ || GET_CODE (src) == LSHIFTRT)
+ nshift++;
+ }
+ if (ngood < 2 && nshift < 2)
+ return 0;
+ }
+
/* Exit early if one of the insns involved can't be used for
combinations. */
if (cant_combine_insn_p (i3)
|| cant_combine_insn_p (i2)
|| (i1 && cant_combine_insn_p (i1))
+ || (i0 && cant_combine_insn_p (i0))
|| likely_spilled_retval_p (i3))
return 0;
if (dump_file && (dump_flags & TDF_DETAILS))
{
- if (i1)
+ if (i0)
+ fprintf (dump_file, "\nTrying %d, %d, %d -> %d:\n",
+ INSN_UID (i0), INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
+ else if (i1)
fprintf (dump_file, "\nTrying %d, %d -> %d:\n",
INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
else
INSN_UID (i2), INSN_UID (i3));
}
- /* If I1 and I2 both feed I3, they can be in any order. To simplify the
- code below, set I1 to be the earlier of the two insns. */
+ /* If multiple insns feed into one of I2 or I3, they can be in any
+ order. To simplify the code below, reorder them in sequence. */
+ if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i2))
+ temp = i2, i2 = i0, i0 = temp;
+ if (i0 && DF_INSN_LUID (i0) > DF_INSN_LUID (i1))
+ temp = i1, i1 = i0, i0 = temp;
if (i1 && DF_INSN_LUID (i1) > DF_INSN_LUID (i2))
temp = i1, i1 = i2, i2 = temp;
subst_insn = i3;
subst_low_luid = DF_INSN_LUID (i2);
- added_sets_2 = added_sets_1 = 0;
+ added_sets_2 = added_sets_1 = added_sets_0 = 0;
i2src = SET_DEST (PATTERN (i3));
i2dest = SET_SRC (PATTERN (i3));
i2dest_killed = dead_or_set_p (i2, i2dest);
if (offset >= 0
&& (GET_MODE_BITSIZE (GET_MODE (SET_DEST (temp)))
- <= HOST_BITS_PER_WIDE_INT * 2))
+ <= HOST_BITS_PER_DOUBLE_INT))
{
- HOST_WIDE_INT mhi, ohi, ihi;
- HOST_WIDE_INT mlo, olo, ilo;
+ double_int m, o, i;
rtx inner = SET_SRC (PATTERN (i3));
rtx outer = SET_SRC (temp);
- if (CONST_INT_P (outer))
- {
- olo = INTVAL (outer);
- ohi = olo < 0 ? -1 : 0;
- }
- else
- {
- olo = CONST_DOUBLE_LOW (outer);
- ohi = CONST_DOUBLE_HIGH (outer);
- }
-
- if (CONST_INT_P (inner))
- {
- ilo = INTVAL (inner);
- ihi = ilo < 0 ? -1 : 0;
- }
- else
- {
- ilo = CONST_DOUBLE_LOW (inner);
- ihi = CONST_DOUBLE_HIGH (inner);
- }
-
- if (width < HOST_BITS_PER_WIDE_INT)
- {
- mlo = ((unsigned HOST_WIDE_INT) 1 << width) - 1;
- mhi = 0;
- }
- else if (width < HOST_BITS_PER_WIDE_INT * 2)
- {
- mhi = ((unsigned HOST_WIDE_INT) 1
- << (width - HOST_BITS_PER_WIDE_INT)) - 1;
- mlo = -1;
- }
- else
- {
- mlo = -1;
- mhi = -1;
- }
+ o = rtx_to_double_int (outer);
+ i = rtx_to_double_int (inner);
- ilo &= mlo;
- ihi &= mhi;
-
- if (offset >= HOST_BITS_PER_WIDE_INT)
- {
- mhi = mlo << (offset - HOST_BITS_PER_WIDE_INT);
- mlo = 0;
- ihi = ilo << (offset - HOST_BITS_PER_WIDE_INT);
- ilo = 0;
- }
- else if (offset > 0)
- {
- mhi = (mhi << offset) | ((unsigned HOST_WIDE_INT) mlo
- >> (HOST_BITS_PER_WIDE_INT - offset));
- mlo = mlo << offset;
- ihi = (ihi << offset) | ((unsigned HOST_WIDE_INT) ilo
- >> (HOST_BITS_PER_WIDE_INT - offset));
- ilo = ilo << offset;
- }
-
- olo = (olo & ~mlo) | ilo;
- ohi = (ohi & ~mhi) | ihi;
+ m = double_int_mask (width);
+ i = double_int_and (i, m);
+ m = double_int_lshift (m, offset, HOST_BITS_PER_DOUBLE_INT, false);
+ i = double_int_lshift (i, offset, HOST_BITS_PER_DOUBLE_INT, false);
+ o = double_int_ior (double_int_and_not (o, m), i);
combine_merges++;
subst_insn = i3;
subst_low_luid = DF_INSN_LUID (i2);
- added_sets_2 = added_sets_1 = 0;
+ added_sets_2 = added_sets_1 = added_sets_0 = 0;
i2dest = SET_DEST (temp);
i2dest_killed = dead_or_set_p (i2, i2dest);
+ /* Replace the source in I2 with the new constant and make the
+ resulting insn the new pattern for I3. Then skip to where we
+ validate the pattern. Everything was set up above. */
SUBST (SET_SRC (temp),
- immed_double_const (olo, ohi, GET_MODE (SET_DEST (temp))));
+ immed_double_int_const (o, GET_MODE (SET_DEST (temp))));
newpat = PATTERN (i2);
+
+ /* The dest of I3 has been replaced with the dest of I2. */
+ changed_i3_dest = 1;
goto validate_replacement;
}
}
as I2 will not cause a problem. */
i1 = gen_rtx_INSN (VOIDmode, INSN_UID (i2), NULL_RTX, i2,
- BLOCK_FOR_INSN (i2), INSN_LOCATOR (i2),
- XVECEXP (PATTERN (i2), 0, 1), -1, NULL_RTX);
+ BLOCK_FOR_INSN (i2), XVECEXP (PATTERN (i2), 0, 1),
+ INSN_LOCATOR (i2), -1, NULL_RTX);
SUBST (PATTERN (i2), XVECEXP (PATTERN (i2), 0, 0));
SUBST (XEXP (SET_SRC (PATTERN (i2)), 0),
#endif
/* Verify that I2 and I1 are valid for combining. */
- if (! can_combine_p (i2, i3, i1, NULL_RTX, &i2dest, &i2src)
- || (i1 && ! can_combine_p (i1, i3, NULL_RTX, i2, &i1dest, &i1src)))
+ if (! can_combine_p (i2, i3, i0, i1, NULL_RTX, NULL_RTX, &i2dest, &i2src)
+ || (i1 && ! can_combine_p (i1, i3, i0, NULL_RTX, i2, NULL_RTX,
+ &i1dest, &i1src))
+ || (i0 && ! can_combine_p (i0, i3, NULL_RTX, NULL_RTX, i1, i2,
+ &i0dest, &i0src)))
{
undo_all ();
return 0;
i2dest_in_i2src = reg_overlap_mentioned_p (i2dest, i2src);
i1dest_in_i1src = i1 && reg_overlap_mentioned_p (i1dest, i1src);
i2dest_in_i1src = i1 && reg_overlap_mentioned_p (i2dest, i1src);
+ i0dest_in_i0src = i0 && reg_overlap_mentioned_p (i0dest, i0src);
+ i1dest_in_i0src = i0 && reg_overlap_mentioned_p (i1dest, i0src);
+ i2dest_in_i0src = i0 && reg_overlap_mentioned_p (i2dest, i0src);
i2dest_killed = dead_or_set_p (i2, i2dest);
i1dest_killed = i1 && dead_or_set_p (i1, i1dest);
+ i0dest_killed = i0 && dead_or_set_p (i0, i0dest);
- /* See if I1 directly feeds into I3. It does if I1DEST is not used
- in I2SRC. */
- i1_feeds_i3 = i1 && ! reg_overlap_mentioned_p (i1dest, i2src);
+ /* For the earlier insns, determine which of the subsequent ones they
+ feed. */
+ i1_feeds_i2_n = i1 && insn_a_feeds_b (i1, i2);
+ i0_feeds_i1_n = i0 && insn_a_feeds_b (i0, i1);
+ i0_feeds_i2_n = (i0 && (!i0_feeds_i1_n ? insn_a_feeds_b (i0, i2)
+ : (!reg_overlap_mentioned_p (i1dest, i0dest)
+ && reg_overlap_mentioned_p (i0dest, i2src))));
/* Ensure that I3's pattern can be the destination of combines. */
- if (! combinable_i3pat (i3, &PATTERN (i3), i2dest, i1dest,
- i1 && i2dest_in_i1src && i1_feeds_i3,
+ if (! combinable_i3pat (i3, &PATTERN (i3), i2dest, i1dest, i0dest,
+ i1 && i2dest_in_i1src && !i1_feeds_i2_n,
+ i0 && ((i2dest_in_i0src && !i0_feeds_i2_n)
+ || (i1dest_in_i0src && !i0_feeds_i1_n)),
&i3dest_killed))
{
undo_all ();
here. */
if (GET_CODE (i2src) == MULT
|| (i1 != 0 && GET_CODE (i1src) == MULT)
+ || (i0 != 0 && GET_CODE (i0src) == MULT)
|| (GET_CODE (PATTERN (i3)) == SET
&& GET_CODE (SET_SRC (PATTERN (i3))) == MULT))
have_mult = 1;
feed into I3, the set in I1 needs to be kept around if I1DEST dies
or is set in I3. Otherwise (if I1 feeds I2 which feeds I3), the set
in I1 needs to be kept around unless I1DEST dies or is set in either
- I2 or I3. We can distinguish these cases by seeing if I2SRC mentions
- I1DEST. If so, we know I1 feeds into I2. */
+ I2 or I3. The same consideration applies to I0. */
- added_sets_2 = ! dead_or_set_p (i3, i2dest);
+ added_sets_2 = !dead_or_set_p (i3, i2dest);
+
+ if (i1)
+ added_sets_1 = !(dead_or_set_p (i3, i1dest)
+ || (i1_feeds_i2_n && dead_or_set_p (i2, i1dest)));
+ else
+ added_sets_1 = 0;
- added_sets_1
- = i1 && ! (i1_feeds_i3 ? dead_or_set_p (i3, i1dest)
- : (dead_or_set_p (i3, i1dest) || dead_or_set_p (i2, i1dest)));
+ if (i0)
+ added_sets_0 = !(dead_or_set_p (i3, i0dest)
+ || (i0_feeds_i2_n && dead_or_set_p (i2, i0dest))
+ || (i0_feeds_i1_n && dead_or_set_p (i1, i0dest)));
+ else
+ added_sets_0 = 0;
/* If the set in I2 needs to be kept around, we must make a copy of
PATTERN (I2), so that when we substitute I1SRC for I1DEST in
i1pat = copy_rtx (PATTERN (i1));
}
+ if (added_sets_0)
+ {
+ if (GET_CODE (PATTERN (i0)) == PARALLEL)
+ i0pat = gen_rtx_SET (VOIDmode, i0dest, copy_rtx (i0src));
+ else
+ i0pat = copy_rtx (PATTERN (i0));
+ }
+
combine_merges++;
/* Substitute in the latest insn for the regs set by the earlier ones. */
i2src, const0_rtx))
!= GET_MODE (SET_DEST (newpat))))
{
- if (can_change_dest_mode(SET_DEST (newpat), added_sets_2,
- compare_mode))
+ if (can_change_dest_mode (SET_DEST (newpat), added_sets_2,
+ compare_mode))
{
unsigned int regno = REGNO (SET_DEST (newpat));
rtx new_dest;
n_occurrences = 0; /* `subst' counts here */
- /* If I1 feeds into I2 (not into I3) and I1DEST is in I1SRC, we
- need to make a unique copy of I2SRC each time we substitute it
- to avoid self-referential rtl. */
+ /* If I1 feeds into I2 and I1DEST is in I1SRC, we need to make a
+ unique copy of I2SRC each time we substitute it to avoid
+ self-referential rtl. */
subst_low_luid = DF_INSN_LUID (i2);
newpat = subst (PATTERN (i3), i2dest, i2src, 0,
- ! i1_feeds_i3 && i1dest_in_i1src);
+ ((i1_feeds_i2_n && i1dest_in_i1src)
+ || (i0_feeds_i2_n && i0dest_in_i0src)));
substed_i2 = 1;
/* Record whether i2's body now appears within i3's body. */
This happens if I1DEST is mentioned in I2 and dies there, and
has disappeared from the new pattern. */
if ((FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
- && !i1_feeds_i3
+ && i1_feeds_i2_n
&& dead_or_set_p (i2, i1dest)
&& !reg_overlap_mentioned_p (i1dest, newpat))
/* Before we can do this substitution, we must redo the test done
above (see detailed comments there) that ensures that I1DEST
isn't mentioned in any SETs in NEWPAT that are field assignments. */
- || !combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX, 0, 0))
+ || !combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX, NULL_RTX,
+ 0, 0, 0))
{
undo_all ();
return 0;
n_occurrences = 0;
subst_low_luid = DF_INSN_LUID (i1);
- newpat = subst (newpat, i1dest, i1src, 0, 0);
+ newpat = subst (newpat, i1dest, i1src, 0,
+ i0_feeds_i1_n && i0dest_in_i0src);
substed_i1 = 1;
+ i1_is_used = n_occurrences;
+ }
+ if (i0 && GET_CODE (newpat) != CLOBBER)
+ {
+ if ((FIND_REG_INC_NOTE (i0, NULL_RTX) != 0
+ && ((i0_feeds_i2_n && dead_or_set_p (i2, i0dest))
+ || (i0_feeds_i1_n && dead_or_set_p (i1, i0dest)))
+ && !reg_overlap_mentioned_p (i0dest, newpat))
+ || !combinable_i3pat (NULL_RTX, &newpat, i0dest, NULL_RTX, NULL_RTX,
+ 0, 0, 0))
+ {
+ undo_all ();
+ return 0;
+ }
+
+ n_occurrences = 0;
+ subst_low_luid = DF_INSN_LUID (i0);
+ newpat = subst (newpat, i0dest, i0src, 0,
+ i0_feeds_i1_n && i0dest_in_i0src);
+ substed_i0 = 1;
}
/* Fail if an autoincrement side-effect has been duplicated. Be careful
if ((FIND_REG_INC_NOTE (i2, NULL_RTX) != 0
&& i2_is_used + added_sets_2 > 1)
|| (i1 != 0 && FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
- && (n_occurrences + added_sets_1 + (added_sets_2 && ! i1_feeds_i3)
+ && (i1_is_used + added_sets_1 + (added_sets_2 && i1_feeds_i2_n)
+ > 1))
+ || (i0 != 0 && FIND_REG_INC_NOTE (i0, NULL_RTX) != 0
+ && (n_occurrences + added_sets_0
+ + (added_sets_1 && i0_feeds_i1_n)
+ + (added_sets_2 && i0_feeds_i2_n)
> 1))
/* Fail if we tried to make a new register. */
|| max_reg_num () != maxreg
we must make a new PARALLEL for the latest insn
to hold additional the SETs. */
- if (added_sets_1 || added_sets_2)
+ if (added_sets_0 || added_sets_1 || added_sets_2)
{
+ int extra_sets = added_sets_0 + added_sets_1 + added_sets_2;
combine_extras++;
if (GET_CODE (newpat) == PARALLEL)
{
rtvec old = XVEC (newpat, 0);
- total_sets = XVECLEN (newpat, 0) + added_sets_1 + added_sets_2;
+ total_sets = XVECLEN (newpat, 0) + extra_sets;
newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
memcpy (XVEC (newpat, 0)->elem, &old->elem[0],
sizeof (old->elem[0]) * old->num_elem);
else
{
rtx old = newpat;
- total_sets = 1 + added_sets_1 + added_sets_2;
+ total_sets = 1 + extra_sets;
newpat = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (total_sets));
XVECEXP (newpat, 0, 0) = old;
}
+ if (added_sets_0)
+ XVECEXP (newpat, 0, --total_sets) = i0pat;
+
if (added_sets_1)
- XVECEXP (newpat, 0, --total_sets) = i1pat;
+ {
+ rtx t = i1pat;
+ if (i0_feeds_i1_n)
+ t = subst (t, i0dest, i0src, 0, 0);
+ XVECEXP (newpat, 0, --total_sets) = t;
+ }
if (added_sets_2)
{
- /* If there is no I1, use I2's body as is. We used to also not do
- the subst call below if I2 was substituted into I3,
- but that could lose a simplification. */
- if (i1 == 0)
- XVECEXP (newpat, 0, --total_sets) = i2pat;
- else
- /* See comment where i2pat is assigned. */
- XVECEXP (newpat, 0, --total_sets)
- = subst (i2pat, i1dest, i1src, 0, 0);
+ rtx t = i2pat;
+ if (i0_feeds_i2_n)
+ t = subst (t, i0dest, i0src, 0, 0);
+ if (i1_feeds_i2_n)
+ t = subst (t, i1dest, i1src, 0, 0);
+
+ XVECEXP (newpat, 0, --total_sets) = t;
}
}
- /* We come here when we are replacing a destination in I2 with the
- destination of I3. */
validate_replacement:
/* Note which hard regs this insn has as inputs. */
/* If we can split it and use I2DEST, go ahead and see if that
helps things be recognized. Verify that none of the registers
are set between I2 and I3. */
- if (insn_code_number < 0 && (split = find_split_point (&newpat, i3)) != 0
+ if (insn_code_number < 0
+ && (split = find_split_point (&newpat, i3, false)) != 0
#ifdef HAVE_cc0
&& REG_P (i2dest)
#endif
i2scratch = true;
+ /* *SPLIT may be part of I2SRC, so make sure we have the
+ original expression around for later debug processing.
+ We should not need I2SRC any more in other cases. */
+ if (MAY_HAVE_DEBUG_INSNS)
+ i2src = copy_rtx (i2src);
+ else
+ i2src = NULL;
+
/* Get NEWDEST as a register in the proper mode. We have already
validated that we can do this. */
if (GET_MODE (i2dest) != split_mode && split_mode != VOIDmode)
i2_code_number = recog_for_combine (&newi2pat, i2, &new_i2_notes);
if (i2_code_number >= 0)
- insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
+ {
+ /* recog_for_combine might have added CLOBBERs to newi2pat.
+ Make sure NEWPAT does not depend on the clobbered regs. */
+ if (GET_CODE (newi2pat) == PARALLEL)
+ {
+ for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
+ if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
+ {
+ rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
+ if (reg_overlap_mentioned_p (reg, newpat))
+ break;
+ }
+
+ if (i >= 0)
+ {
+ /* CLOBBERs on newi2pat prevent it going first.
+ Try the other order of the insns if possible. */
+ temp = newpat;
+ newpat = XVECEXP (newi2pat, 0, 0);
+ newi2pat = temp;
+#ifdef HAVE_cc0
+ if (reg_referenced_p (cc0_rtx, newpat))
+ {
+ undo_all ();
+ return 0;
+ }
+#endif
+
+ i2_code_number = recog_for_combine (&newi2pat, i2,
+ &new_i2_notes);
+ if (i2_code_number < 0)
+ {
+ undo_all ();
+ return 0;
+ }
+
+ if (GET_CODE (newi2pat) == PARALLEL)
+ for (i = XVECLEN (newi2pat, 0) - 1; i >= 0; i--)
+ if (GET_CODE (XVECEXP (newi2pat, 0, i)) == CLOBBER)
+ {
+ rtx reg = XEXP (XVECEXP (newi2pat, 0, i), 0);
+ if (reg_overlap_mentioned_p (reg, newpat))
+ {
+ undo_all ();
+ return 0;
+ }
+ }
+ }
+ }
+
+ insn_code_number = recog_for_combine (&newpat, i3, &new_i3_notes);
+ }
}
/* If it still isn't recognized, fail and change things back the way they
/* Only allow this combination if insn_rtx_costs reports that the
replacement instructions are cheaper than the originals. */
- if (!combine_validate_cost (i1, i2, i3, newpat, newi2pat, other_pat))
+ if (!combine_validate_cost (i0, i1, i2, i3, newpat, newi2pat, other_pat))
{
undo_all ();
return 0;
}
distribute_notes (new_other_notes, undobuf.other_insn,
- undobuf.other_insn, NULL_RTX, NULL_RTX, NULL_RTX);
+ undobuf.other_insn, NULL_RTX, NULL_RTX, NULL_RTX,
+ NULL_RTX);
}
if (swap_i2i3)
}
{
- rtx i3notes, i2notes, i1notes = 0;
- rtx i3links, i2links, i1links = 0;
+ rtx i3notes, i2notes, i1notes = 0, i0notes = 0;
+ rtx i3links, i2links, i1links = 0, i0links = 0;
rtx midnotes = 0;
+ int from_luid;
unsigned int regno;
/* Compute which registers we expect to eliminate. newi2pat may be setting
either i3dest or i2dest, so we must check it. Also, i1dest may be the
same as i3dest, in which case newi2pat may be setting i1dest. */
rtx elim_i2 = ((newi2pat && reg_set_p (i2dest, newi2pat))
- || i2dest_in_i2src || i2dest_in_i1src
+ || i2dest_in_i2src || i2dest_in_i1src || i2dest_in_i0src
|| !i2dest_killed
? 0 : i2dest);
- rtx elim_i1 = (i1 == 0 || i1dest_in_i1src
+ rtx elim_i1 = (i1 == 0 || i1dest_in_i1src || i1dest_in_i0src
|| (newi2pat && reg_set_p (i1dest, newi2pat))
|| !i1dest_killed
? 0 : i1dest);
+ rtx elim_i0 = (i0 == 0 || i0dest_in_i0src
+ || (newi2pat && reg_set_p (i0dest, newi2pat))
+ || !i0dest_killed
+ ? 0 : i0dest);
/* Get the old REG_NOTES and LOG_LINKS from all our insns and
clear them. */
i2notes = REG_NOTES (i2), i2links = LOG_LINKS (i2);
if (i1)
i1notes = REG_NOTES (i1), i1links = LOG_LINKS (i1);
+ if (i0)
+ i0notes = REG_NOTES (i0), i0links = LOG_LINKS (i0);
/* Ensure that we do not have something that should not be shared but
occurs multiple times in the new insns. Check this by first
reset_used_flags (i3notes);
reset_used_flags (i2notes);
reset_used_flags (i1notes);
+ reset_used_flags (i0notes);
reset_used_flags (newpat);
reset_used_flags (newi2pat);
if (undobuf.other_insn)
i3notes = copy_rtx_if_shared (i3notes);
i2notes = copy_rtx_if_shared (i2notes);
i1notes = copy_rtx_if_shared (i1notes);
+ i0notes = copy_rtx_if_shared (i0notes);
newpat = copy_rtx_if_shared (newpat);
newi2pat = copy_rtx_if_shared (newi2pat);
if (undobuf.other_insn)
call_usage = copy_rtx (call_usage);
if (substed_i2)
- replace_rtx (call_usage, i2dest, i2src);
+ {
+ /* I2SRC must still be meaningful at this point. Some splitting
+ operations can invalidate I2SRC, but those operations do not
+ apply to calls. */
+ gcc_assert (i2src);
+ replace_rtx (call_usage, i2dest, i2src);
+ }
if (substed_i1)
replace_rtx (call_usage, i1dest, i1src);
+ if (substed_i0)
+ replace_rtx (call_usage, i0dest, i0src);
CALL_INSN_FUNCTION_USAGE (i3) = call_usage;
}
SET_INSN_DELETED (i1);
}
+ if (i0)
+ {
+ LOG_LINKS (i0) = 0;
+ REG_NOTES (i0) = 0;
+ if (MAY_HAVE_DEBUG_INSNS)
+ propagate_for_debug (i0, i3, i0dest, i0src, false);
+ SET_INSN_DELETED (i0);
+ }
+
/* Get death notes for everything that is now used in either I3 or
I2 and used to die in a previous insn. If we built two new
patterns, move from I1 to I2 then I2 to I3 so that we get the
proper movement on registers that I2 modifies. */
- if (newi2pat)
- {
- move_deaths (newi2pat, NULL_RTX, DF_INSN_LUID (i1), i2, &midnotes);
- move_deaths (newpat, newi2pat, DF_INSN_LUID (i1), i3, &midnotes);
- }
+ if (i0)
+ from_luid = DF_INSN_LUID (i0);
+ else if (i1)
+ from_luid = DF_INSN_LUID (i1);
else
- move_deaths (newpat, NULL_RTX, i1 ? DF_INSN_LUID (i1) : DF_INSN_LUID (i2),
- i3, &midnotes);
+ from_luid = DF_INSN_LUID (i2);
+ if (newi2pat)
+ move_deaths (newi2pat, NULL_RTX, from_luid, i2, &midnotes);
+ move_deaths (newpat, newi2pat, from_luid, i3, &midnotes);
/* Distribute all the LOG_LINKS and REG_NOTES from I1, I2, and I3. */
if (i3notes)
distribute_notes (i3notes, i3, i3, newi2pat ? i2 : NULL_RTX,
- elim_i2, elim_i1);
+ elim_i2, elim_i1, elim_i0);
if (i2notes)
distribute_notes (i2notes, i2, i3, newi2pat ? i2 : NULL_RTX,
- elim_i2, elim_i1);
+ elim_i2, elim_i1, elim_i0);
if (i1notes)
distribute_notes (i1notes, i1, i3, newi2pat ? i2 : NULL_RTX,
- elim_i2, elim_i1);
+ elim_i2, elim_i1, elim_i0);
+ if (i0notes)
+ distribute_notes (i0notes, i0, i3, newi2pat ? i2 : NULL_RTX,
+ elim_i2, elim_i1, elim_i0);
if (midnotes)
distribute_notes (midnotes, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
- elim_i2, elim_i1);
+ elim_i2, elim_i1, elim_i0);
/* Distribute any notes added to I2 or I3 by recog_for_combine. We
know these are REG_UNUSED and want them to go to the desired insn,
so we always pass it as i3. */
if (newi2pat && new_i2_notes)
- distribute_notes (new_i2_notes, i2, i2, NULL_RTX, NULL_RTX, NULL_RTX);
-
+ distribute_notes (new_i2_notes, i2, i2, NULL_RTX, NULL_RTX, NULL_RTX,
+ NULL_RTX);
+
if (new_i3_notes)
- distribute_notes (new_i3_notes, i3, i3, NULL_RTX, NULL_RTX, NULL_RTX);
+ distribute_notes (new_i3_notes, i3, i3, NULL_RTX, NULL_RTX, NULL_RTX,
+ NULL_RTX);
/* If I3DEST was used in I3SRC, it really died in I3. We may need to
put a REG_DEAD note for it somewhere. If NEWI2PAT exists and sets
if (newi2pat && reg_set_p (i3dest_killed, newi2pat))
distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
NULL_RTX),
- NULL_RTX, i2, NULL_RTX, elim_i2, elim_i1);
+ NULL_RTX, i2, NULL_RTX, elim_i2, elim_i1, elim_i0);
else
distribute_notes (alloc_reg_note (REG_DEAD, i3dest_killed,
NULL_RTX),
NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
- elim_i2, elim_i1);
+ elim_i2, elim_i1, elim_i0);
}
if (i2dest_in_i2src)
{
+ rtx new_note = alloc_reg_note (REG_DEAD, i2dest, NULL_RTX);
if (newi2pat && reg_set_p (i2dest, newi2pat))
- distribute_notes (alloc_reg_note (REG_DEAD, i2dest, NULL_RTX),
- NULL_RTX, i2, NULL_RTX, NULL_RTX, NULL_RTX);
- else
- distribute_notes (alloc_reg_note (REG_DEAD, i2dest, NULL_RTX),
- NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
+ distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
NULL_RTX, NULL_RTX);
+ else
+ distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
+ NULL_RTX, NULL_RTX, NULL_RTX);
}
if (i1dest_in_i1src)
{
+ rtx new_note = alloc_reg_note (REG_DEAD, i1dest, NULL_RTX);
if (newi2pat && reg_set_p (i1dest, newi2pat))
- distribute_notes (alloc_reg_note (REG_DEAD, i1dest, NULL_RTX),
- NULL_RTX, i2, NULL_RTX, NULL_RTX, NULL_RTX);
+ distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
+ NULL_RTX, NULL_RTX);
else
- distribute_notes (alloc_reg_note (REG_DEAD, i1dest, NULL_RTX),
- NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
+ distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
+ NULL_RTX, NULL_RTX, NULL_RTX);
+ }
+
+ if (i0dest_in_i0src)
+ {
+ rtx new_note = alloc_reg_note (REG_DEAD, i0dest, NULL_RTX);
+ if (newi2pat && reg_set_p (i0dest, newi2pat))
+ distribute_notes (new_note, NULL_RTX, i2, NULL_RTX, NULL_RTX,
NULL_RTX, NULL_RTX);
+ else
+ distribute_notes (new_note, NULL_RTX, i3, newi2pat ? i2 : NULL_RTX,
+ NULL_RTX, NULL_RTX, NULL_RTX);
}
distribute_links (i3links);
distribute_links (i2links);
distribute_links (i1links);
+ distribute_links (i0links);
if (REG_P (i2dest))
{
INC_REG_N_SETS (regno, -1);
}
+ if (i0 && REG_P (i0dest))
+ {
+ rtx link;
+ rtx i0_insn = 0, i0_val = 0, set;
+
+ for (link = LOG_LINKS (i3); link; link = XEXP (link, 1))
+ if ((set = single_set (XEXP (link, 0))) != 0
+ && rtx_equal_p (i0dest, SET_DEST (set)))
+ i0_insn = XEXP (link, 0), i0_val = SET_SRC (set);
+
+ record_value_for_reg (i0dest, i0_insn, i0_val);
+
+ regno = REGNO (i0dest);
+ if (! added_sets_0 && ! i0dest_in_i0src)
+ INC_REG_N_SETS (regno, -1);
+ }
+
/* Update reg_stat[].nonzero_bits et al for any changes that may have
been made to this insn. The order of
set_nonzero_bits_and_sign_copies() is important. Because newi2pat
df_insn_rescan (undobuf.other_insn);
}
+ if (i0 && !(NOTE_P(i0) && (NOTE_KIND (i0) == NOTE_INSN_DELETED)))
+ {
+ if (dump_file)
+ {
+ fprintf (dump_file, "modifying insn i1 ");
+ dump_insn_slim (dump_file, i0);
+ }
+ df_insn_rescan (i0);
+ }
+
if (i1 && !(NOTE_P(i1) && (NOTE_KIND (i1) == NOTE_INSN_DELETED)))
{
if (dump_file)
}
df_insn_rescan (i3);
}
-
+
/* Set new_direct_jump_p if a new return or simple jump instruction
has been created. Adjust the CFG accordingly. */
*new_direct_jump_p = 1;
update_cfg_for_uncondjump (i3);
}
-
+
combine_successes++;
undo_commit ();
two insns. */
static rtx *
-find_split_point (rtx *loc, rtx insn)
+find_split_point (rtx *loc, rtx insn, bool set_src)
{
rtx x = *loc;
enum rtx_code code = GET_CODE (x);
if (MEM_P (SUBREG_REG (x)))
return loc;
#endif
- return find_split_point (&SUBREG_REG (x), insn);
+ return find_split_point (&SUBREG_REG (x), insn, false);
case MEM:
#ifdef HAVE_lo_sum
if (GET_CODE (XEXP (x, 0)) == CONST
|| GET_CODE (XEXP (x, 0)) == SYMBOL_REF)
{
+ enum machine_mode address_mode
+ = targetm.addr_space.address_mode (MEM_ADDR_SPACE (x));
+
SUBST (XEXP (x, 0),
- gen_rtx_LO_SUM (Pmode,
- gen_rtx_HIGH (Pmode, XEXP (x, 0)),
+ gen_rtx_LO_SUM (address_mode,
+ gen_rtx_HIGH (address_mode, XEXP (x, 0)),
XEXP (x, 0)));
return &XEXP (XEXP (x, 0), 0);
}
it will not remain in the result. */
if (GET_CODE (XEXP (x, 0)) == PLUS
&& CONST_INT_P (XEXP (XEXP (x, 0), 1))
- && ! memory_address_p (GET_MODE (x), XEXP (x, 0)))
+ && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
+ MEM_ADDR_SPACE (x)))
{
rtx reg = regno_reg_rtx[FIRST_PSEUDO_REGISTER];
rtx seq = combine_split_insns (gen_rtx_SET (VOIDmode, reg,
&& NONJUMP_INSN_P (NEXT_INSN (seq))
&& GET_CODE (PATTERN (NEXT_INSN (seq))) == SET
&& SET_DEST (PATTERN (NEXT_INSN (seq))) == reg
- && memory_address_p (GET_MODE (x),
- SET_SRC (PATTERN (NEXT_INSN (seq)))))
+ && memory_address_addr_space_p
+ (GET_MODE (x), SET_SRC (PATTERN (NEXT_INSN (seq))),
+ MEM_ADDR_SPACE (x)))
{
rtx src1 = SET_SRC (PATTERN (seq));
rtx src2 = SET_SRC (PATTERN (NEXT_INSN (seq)));
/* If we have a PLUS whose first operand is complex, try computing it
separately by making a split there. */
if (GET_CODE (XEXP (x, 0)) == PLUS
- && ! memory_address_p (GET_MODE (x), XEXP (x, 0))
+ && ! memory_address_addr_space_p (GET_MODE (x), XEXP (x, 0),
+ MEM_ADDR_SPACE (x))
&& ! OBJECT_P (XEXP (XEXP (x, 0), 0))
&& ! (GET_CODE (XEXP (XEXP (x, 0), 0)) == SUBREG
&& OBJECT_P (SUBREG_REG (XEXP (XEXP (x, 0), 0)))))
#endif
/* See if we can split SET_SRC as it stands. */
- split = find_split_point (&SET_SRC (x), insn);
+ split = find_split_point (&SET_SRC (x), insn, true);
if (split && split != &SET_SRC (x))
return split;
/* See if we can split SET_DEST as it stands. */
- split = find_split_point (&SET_DEST (x), insn);
+ split = find_split_point (&SET_DEST (x), insn, false);
if (split && split != &SET_DEST (x))
return split;
SUBST (SET_DEST (x), dest);
- split = find_split_point (&SET_SRC (x), insn);
+ split = find_split_point (&SET_SRC (x), insn, true);
if (split && split != &SET_SRC (x))
return split;
}
if (extraction != 0)
{
SUBST (SET_SRC (x), extraction);
- return find_split_point (loc, insn);
+ return find_split_point (loc, insn, false);
}
}
break;
XEXP (SET_SRC (x), 0),
GEN_INT (pos))));
- split = find_split_point (&SET_SRC (x), insn);
+ split = find_split_point (&SET_SRC (x), insn, true);
if (split && split != &SET_SRC (x))
return split;
}
GEN_INT (pos)),
GEN_INT (((HOST_WIDE_INT) 1 << len) - 1)));
- split = find_split_point (&SET_SRC (x), insn);
+ split = find_split_point (&SET_SRC (x), insn, true);
if (split && split != &SET_SRC (x))
return split;
}
- len - pos)),
GEN_INT (GET_MODE_BITSIZE (mode) - len)));
- split = find_split_point (&SET_SRC (x), insn);
+ split = find_split_point (&SET_SRC (x), insn, true);
if (split && split != &SET_SRC (x))
return split;
}
GET_MODE (x),
XEXP (XEXP (x, 0), 0),
XEXP (XEXP (x, 1), 0))));
- return find_split_point (loc, insn);
+ return find_split_point (loc, insn, set_src);
}
/* Many RISC machines have a large set of logical insns. If the
}
break;
+ case PLUS:
+ case MINUS:
+ /* Canonicalization can produce (minus A (mult B C)), where C is a
+ constant. It may be better to try splitting (plus (mult B -C) A)
+ instead if this isn't a multiply by a power of two. */
+ if (set_src && code == MINUS && GET_CODE (XEXP (x, 1)) == MULT
+ && GET_CODE (XEXP (XEXP (x, 1), 1)) == CONST_INT
+ && exact_log2 (INTVAL (XEXP (XEXP (x, 1), 1))) < 0)
+ {
+ enum machine_mode mode = GET_MODE (x);
+ unsigned HOST_WIDE_INT this_int = INTVAL (XEXP (XEXP (x, 1), 1));
+ HOST_WIDE_INT other_int = trunc_int_for_mode (-this_int, mode);
+ SUBST (*loc, gen_rtx_PLUS (mode, gen_rtx_MULT (mode,
+ XEXP (XEXP (x, 1), 0),
+ GEN_INT (other_int)),
+ XEXP (x, 0)));
+ return find_split_point (loc, insn, set_src);
+ }
+
+ /* Split at a multiply-accumulate instruction. However if this is
+ the SET_SRC, we likely do not have such an instruction and it's
+ worthless to try this split. */
+ if (!set_src && GET_CODE (XEXP (x, 0)) == MULT)
+ return loc;
+
default:
break;
}
{
case RTX_BITFIELD_OPS: /* This is ZERO_EXTRACT and SIGN_EXTRACT. */
case RTX_TERNARY:
- split = find_split_point (&XEXP (x, 2), insn);
+ split = find_split_point (&XEXP (x, 2), insn, false);
if (split)
return split;
/* ... fall through ... */
case RTX_COMM_ARITH:
case RTX_COMPARE:
case RTX_COMM_COMPARE:
- split = find_split_point (&XEXP (x, 1), insn);
+ split = find_split_point (&XEXP (x, 1), insn, false);
if (split)
return split;
/* ... fall through ... */
if (GET_CODE (x) != AND && GET_CODE (XEXP (x, 0)) == AND)
return &XEXP (x, 0);
- split = find_split_point (&XEXP (x, 0), insn);
+ split = find_split_point (&XEXP (x, 0), insn, false);
if (split)
return split;
return loc;
force_to_mode (XEXP (x, 0), GET_MODE (XEXP (x, 0)),
GET_MODE_MASK (mode), 0));
+ /* We can truncate a constant value and return it. */
+ if (CONST_INT_P (XEXP (x, 0)))
+ return gen_int_mode (INTVAL (XEXP (x, 0)), mode);
+
/* Similarly to what we do in simplify-rtx.c, a truncate of a register
whose value is a comparison can be replaced with a subreg if
STORE_FLAG_VALUE permits. */
}
/* Try simplify a*(b/c) as (a*b)/c. */
- if (FLOAT_MODE_P (mode) && flag_associative_math
+ if (FLOAT_MODE_P (mode) && flag_associative_math
&& GET_CODE (XEXP (x, 0)) == DIV)
{
rtx tem = simplify_binary_operation (MULT, mode,
if (mode == tmode)
return new_rtx;
- if (CONST_INT_P (new_rtx))
- return gen_int_mode (INTVAL (new_rtx), mode);
+ if (CONST_INT_P (new_rtx)
+ || GET_CODE (new_rtx) == CONST_DOUBLE)
+ return simplify_unary_operation (unsignedp ? ZERO_EXTEND : SIGN_EXTEND,
+ mode, new_rtx, tmode);
/* If we know that no extraneous bits are set, and that the high
bit is not set, convert the extraction to the cheaper of
address, we stay there. If we have a comparison, set to COMPARE,
but once inside, go back to our default of SET. */
- next_code = (code == MEM || code == PLUS || code == MINUS ? MEM
+ next_code = (code == MEM ? MEM
+ : ((code == PLUS || code == MINUS)
+ && SCALAR_INT_MODE_P (mode)) ? MEM
: ((code == COMPARE || COMPARISON_P (x))
&& XEXP (x, 1) == const0_rtx) ? COMPARE
: in_code == COMPARE ? SET : in_code);
&& INTVAL (XEXP (x, 1)) < HOST_BITS_PER_WIDE_INT
&& INTVAL (XEXP (x, 1)) >= 0)
{
+ HOST_WIDE_INT count = INTVAL (XEXP (x, 1));
+ HOST_WIDE_INT multval = (HOST_WIDE_INT) 1 << count;
+
new_rtx = make_compound_operation (XEXP (x, 0), next_code);
- new_rtx = gen_rtx_MULT (mode, new_rtx,
- GEN_INT ((HOST_WIDE_INT) 1
- << INTVAL (XEXP (x, 1))));
+ if (GET_CODE (new_rtx) == NEG)
+ {
+ new_rtx = XEXP (new_rtx, 0);
+ multval = -multval;
+ }
+ multval = trunc_int_for_mode (multval, mode);
+ new_rtx = gen_rtx_MULT (mode, new_rtx, GEN_INT (multval));
}
break;
+ case PLUS:
+ lhs = XEXP (x, 0);
+ rhs = XEXP (x, 1);
+ lhs = make_compound_operation (lhs, next_code);
+ rhs = make_compound_operation (rhs, next_code);
+ if (GET_CODE (lhs) == MULT && GET_CODE (XEXP (lhs, 0)) == NEG
+ && SCALAR_INT_MODE_P (mode))
+ {
+ tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (lhs, 0), 0),
+ XEXP (lhs, 1));
+ new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
+ }
+ else if (GET_CODE (lhs) == MULT
+ && (CONST_INT_P (XEXP (lhs, 1)) && INTVAL (XEXP (lhs, 1)) < 0))
+ {
+ tem = simplify_gen_binary (MULT, mode, XEXP (lhs, 0),
+ simplify_gen_unary (NEG, mode,
+ XEXP (lhs, 1),
+ mode));
+ new_rtx = simplify_gen_binary (MINUS, mode, rhs, tem);
+ }
+ else
+ {
+ SUBST (XEXP (x, 0), lhs);
+ SUBST (XEXP (x, 1), rhs);
+ goto maybe_swap;
+ }
+ x = gen_lowpart (mode, new_rtx);
+ goto maybe_swap;
+
+ case MINUS:
+ lhs = XEXP (x, 0);
+ rhs = XEXP (x, 1);
+ lhs = make_compound_operation (lhs, next_code);
+ rhs = make_compound_operation (rhs, next_code);
+ if (GET_CODE (rhs) == MULT && GET_CODE (XEXP (rhs, 0)) == NEG
+ && SCALAR_INT_MODE_P (mode))
+ {
+ tem = simplify_gen_binary (MULT, mode, XEXP (XEXP (rhs, 0), 0),
+ XEXP (rhs, 1));
+ new_rtx = simplify_gen_binary (PLUS, mode, tem, lhs);
+ }
+ else if (GET_CODE (rhs) == MULT
+ && (CONST_INT_P (XEXP (rhs, 1)) && INTVAL (XEXP (rhs, 1)) < 0))
+ {
+ tem = simplify_gen_binary (MULT, mode, XEXP (rhs, 0),
+ simplify_gen_unary (NEG, mode,
+ XEXP (rhs, 1),
+ mode));
+ new_rtx = simplify_gen_binary (PLUS, mode, tem, lhs);
+ }
+ else
+ {
+ SUBST (XEXP (x, 0), lhs);
+ SUBST (XEXP (x, 1), rhs);
+ return x;
+ }
+ return gen_lowpart (mode, new_rtx);
+
case AND:
/* If the second operand is not a constant, we can't do anything
with it. */
/* Call ourselves recursively on the inner expression. If we are
narrowing the object and it has a different RTL code from
what it originally did, do this SUBREG as a force_to_mode. */
-
- tem = make_compound_operation (SUBREG_REG (x), in_code);
-
{
- rtx simplified;
- simplified = simplify_subreg (GET_MODE (x), tem, GET_MODE (tem),
- SUBREG_BYTE (x));
+ rtx inner = SUBREG_REG (x), simplified;
+
+ tem = make_compound_operation (inner, in_code);
+ simplified
+ = simplify_subreg (mode, tem, GET_MODE (inner), SUBREG_BYTE (x));
if (simplified)
tem = simplified;
- if (GET_CODE (tem) != GET_CODE (SUBREG_REG (x))
- && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (tem))
+ if (GET_CODE (tem) != GET_CODE (inner)
+ && GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (inner))
&& subreg_lowpart_p (x))
{
- rtx newer = force_to_mode (tem, mode, ~(HOST_WIDE_INT) 0,
- 0);
+ rtx newer = force_to_mode (tem, mode, ~(HOST_WIDE_INT) 0, 0);
/* If we have something other than a SUBREG, we might have
done an expansion, so rerun ourselves. */
newer = make_compound_operation (newer, in_code);
/* force_to_mode can expand compounds. If it just re-expanded the
- compound use gen_lowpart instead to convert to the desired
- mode. */
- if (rtx_equal_p (newer, x))
+ compound, use gen_lowpart to convert to the desired mode. */
+ if (rtx_equal_p (newer, x)
+ /* Likewise if it re-expanded the compound only partially.
+ This happens for SUBREG of ZERO_EXTRACT if they extract
+ the same number of bits. */
+ || (GET_CODE (newer) == SUBREG
+ && (GET_CODE (SUBREG_REG (newer)) == LSHIFTRT
+ || GET_CODE (SUBREG_REG (newer)) == ASHIFTRT)
+ && GET_CODE (inner) == AND
+ && rtx_equal_p (SUBREG_REG (newer), XEXP (inner, 0))))
return gen_lowpart (GET_MODE (x), tem);
return newer;
SUBST (XVECEXP (x, i, j), new_rtx);
}
+ maybe_swap:
/* If this is a commutative operation, the changes to the operands
may have made it noncanonical. */
if (COMMUTATIVE_ARITH_P (x)
get_pos_from_mask (unsigned HOST_WIDE_INT m, unsigned HOST_WIDE_INT *plen)
{
/* Get the bit number of the first 1 bit from the right, -1 if none. */
- int pos = exact_log2 (m & -m);
+ int pos = m ? ctz_hwi (m) : -1;
int len = 0;
if (pos >= 0)
enum rtx_code outer_code, inner_code;
rtx decomposed, distributed, inner_op0, inner_op1, new_op0, new_op1, tmp;
+ /* Distributivity is not true for floating point as it can change the
+ value. So we don't do it unless -funsafe-math-optimizations. */
+ if (FLOAT_MODE_P (GET_MODE (x))
+ && ! flag_unsafe_math_optimizations)
+ return NULL_RTX;
+
decomposed = XEXP (x, n);
if (!ARITHMETIC_P (decomposed))
return NULL_RTX;
> GET_MODE_SIZE (GET_MODE (varop)))
&& (unsigned int) ((GET_MODE_SIZE (GET_MODE (SUBREG_REG (varop)))
+ (UNITS_PER_WORD - 1)) / UNITS_PER_WORD)
- == mode_words)
+ == mode_words
+ && GET_MODE_CLASS (GET_MODE (varop)) == MODE_INT
+ && GET_MODE_CLASS (GET_MODE (SUBREG_REG (varop))) == MODE_INT)
{
varop = SUBREG_REG (varop);
if (GET_MODE_SIZE (GET_MODE (varop)) > GET_MODE_SIZE (mode))
if (REG_P (XEXP (XVECEXP (newpat, 0, i), 0))
&& ! reg_dead_at_p (XEXP (XVECEXP (newpat, 0, i), 0), insn))
return -1;
- if (GET_CODE (XEXP (XVECEXP (newpat, 0, i), 0)) != SCRATCH)
+ if (GET_CODE (XEXP (XVECEXP (newpat, 0, i), 0)) != SCRATCH)
{
gcc_assert (REG_P (XEXP (XVECEXP (newpat, 0, i), 0)));
notes = alloc_reg_note (REG_UNUSED,
{
int zero_extended;
+ /* If this is a test for negative, we can make an explicit
+ test of the sign bit. Test this first so we can use
+ a paradoxical subreg to extend OP0. */
+
+ if (op1 == const0_rtx && (code == LT || code == GE)
+ && GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
+ {
+ op0 = simplify_gen_binary (AND, tmode,
+ gen_lowpart (tmode, op0),
+ GEN_INT ((HOST_WIDE_INT) 1
+ << (GET_MODE_BITSIZE (mode)
+ - 1)));
+ code = (code == LT) ? NE : EQ;
+ break;
+ }
+
/* If the only nonzero bits in OP0 and OP1 are those in the
narrower mode and this is an equality or unsigned comparison,
we can use the wider mode. Similarly for sign-extended
XEXP (op0, 0)),
gen_lowpart (tmode,
XEXP (op0, 1)));
-
- op0 = gen_lowpart (tmode, op0);
- if (zero_extended && CONST_INT_P (op1))
- op1 = GEN_INT (INTVAL (op1) & GET_MODE_MASK (mode));
- op1 = gen_lowpart (tmode, op1);
- break;
- }
-
- /* If this is a test for negative, we can make an explicit
- test of the sign bit. */
-
- if (op1 == const0_rtx && (code == LT || code == GE)
- && GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
- {
- op0 = simplify_gen_binary (AND, tmode,
- gen_lowpart (tmode, op0),
- GEN_INT ((HOST_WIDE_INT) 1
- << (GET_MODE_BITSIZE (mode)
- - 1)));
- code = (code == LT) ? NE : EQ;
- break;
+ else
+ {
+ if (zero_extended)
+ {
+ op0 = simplify_gen_unary (ZERO_EXTEND, tmode, op0, mode);
+ op1 = simplify_gen_unary (ZERO_EXTEND, tmode, op1, mode);
+ }
+ else
+ {
+ op0 = simplify_gen_unary (SIGN_EXTEND, tmode, op0, mode);
+ op1 = simplify_gen_unary (SIGN_EXTEND, tmode, op1, mode);
+ }
+ break;
+ }
}
}
case, we must replace it with (clobber (const_int 0)) to prevent
infinite loops. */
rsp = VEC_index (reg_stat_type, reg_stat, regno);
- if (value && ! get_last_value_validate (&value, insn,
- rsp->last_set_label, 0))
+ if (value && !get_last_value_validate (&value, insn, label_tick, 0))
{
value = copy_rtx (value);
- if (! get_last_value_validate (&value, insn,
- rsp->last_set_label, 1))
+ if (!get_last_value_validate (&value, insn, label_tick, 1))
value = 0;
}
}
}
\f
-/* Utility routine for the following function. Verify that all the registers
- mentioned in *LOC are valid when *LOC was part of a value set when
- label_tick == TICK. Return 0 if some are not.
-
- If REPLACE is nonzero, replace the invalid reference with
- (clobber (const_int 0)) and return 1. This replacement is useful because
- we often can get useful information about the form of a value (e.g., if
- it was produced by a shift that always produces -1 or 0) even though
- we don't know exactly what registers it was produced from. */
+/* Verify that all the registers and memory references mentioned in *LOC are
+ still valid. *LOC was part of a value set in INSN when label_tick was
+ equal to TICK. Return 0 if some are not. If REPLACE is nonzero, replace
+ the invalid references with (clobber (const_int 0)) and return 1. This
+ replacement is useful because we often can get useful information about
+ the form of a value (e.g., if it was produced by a shift that always
+ produces -1 or 0) even though we don't know exactly what registers it
+ was produced from. */
static int
get_last_value_validate (rtx *loc, rtx insn, int tick, int replace)
return 1;
}
- /* If this is a memory reference, make sure that there were
- no stores after it that might have clobbered the value. We don't
- have alias info, so we assume any store invalidates it. */
+ /* If this is a memory reference, make sure that there were no stores after
+ it that might have clobbered the value. We don't have alias info, so we
+ assume any store invalidates it. Moreover, we only have local UIDs, so
+ we also assume that there were stores in the intervening basic blocks. */
else if (MEM_P (x) && !MEM_READONLY_P (x)
- && DF_INSN_LUID (insn) <= mem_last_set)
+ && (tick != label_tick || DF_INSN_LUID (insn) <= mem_last_set))
{
if (replace)
*loc = gen_rtx_CLOBBER (GET_MODE (x), const0_rtx);
return 0;
/* If the value has all its registers valid, return it. */
- if (get_last_value_validate (&value, rsp->last_set,
- rsp->last_set_label, 0))
+ if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 0))
return value;
/* Otherwise, make a copy and replace any invalid register with
(clobber (const_int 0)). If that fails for some reason, return 0. */
value = copy_rtx (value);
- if (get_last_value_validate (&value, rsp->last_set,
- rsp->last_set_label, 1))
+ if (get_last_value_validate (&value, rsp->last_set, rsp->last_set_label, 1))
return value;
return 0;
return 0;
}
-
-/* Return the next insn after INSN that is neither a NOTE nor a
- DEBUG_INSN. This routine does not look inside SEQUENCEs. */
-
-static rtx
-next_nonnote_nondebug_insn (rtx insn)
-{
- while (insn)
- {
- insn = NEXT_INSN (insn);
- if (insn == 0)
- break;
- if (NOTE_P (insn))
- continue;
- if (DEBUG_INSN_P (insn))
- continue;
- break;
- }
-
- return insn;
-}
-
-
\f
/* Given a chain of REG_NOTES originally from FROM_INSN, try to place them
as appropriate. I3 and I2 are the insns resulting from the combination
static void
distribute_notes (rtx notes, rtx from_insn, rtx i3, rtx i2, rtx elim_i2,
- rtx elim_i1)
+ rtx elim_i1, rtx elim_i0)
{
rtx note, next_note;
rtx tem;
place = i2;
else
{
- gcc_assert (flag_non_call_exceptions);
+ gcc_assert (cfun->can_throw_non_call_exceptions);
if (may_trap_p (i3))
place = i3;
else if (i2 && may_trap_p (i2))
&& !(i2mod
&& reg_overlap_mentioned_p (XEXP (note, 0),
i2mod_old_rhs)))
- || rtx_equal_p (XEXP (note, 0), elim_i1))
+ || rtx_equal_p (XEXP (note, 0), elim_i1)
+ || rtx_equal_p (XEXP (note, 0), elim_i0))
break;
tem = i3;
}
REG_NOTES (tem) = NULL;
distribute_notes (old_notes, tem, tem, NULL_RTX,
- NULL_RTX, NULL_RTX);
+ NULL_RTX, NULL_RTX, NULL_RTX);
distribute_links (LOG_LINKS (tem));
SET_INSN_DELETED (tem);
distribute_notes (old_notes, cc0_setter,
cc0_setter, NULL_RTX,
- NULL_RTX, NULL_RTX);
+ NULL_RTX, NULL_RTX, NULL_RTX);
distribute_links (LOG_LINKS (cc0_setter));
SET_INSN_DELETED (cc0_setter);
NULL_RTX);
distribute_notes (new_note, place, place,
- NULL_RTX, NULL_RTX, NULL_RTX);
+ NULL_RTX, NULL_RTX, NULL_RTX,
+ NULL_RTX);
}
else if (! refers_to_regno_p (i, i + 1,
PATTERN (place), 0)
OSDN Git Service
