#include "toplev.h"
#include "target.h"
#include "rtlhooks-def.h"
+/* Include output.h for dump_file. */
+#include "output.h"
/* Number of attempts to combine instructions in this function. */
those blocks as starting points. */
static sbitmap refresh_blocks;
\f
+/* The following array records the combine_insn_cost for every insn
+ in the instruction stream. */
+
+static int *uid_insn_cost;
+
+/* Length of the currently allocated uid_insn_cost array. */
+
+static int last_insn_cost;
+
/* Incremented for each label. */
static int label_tick;
#define SUBST_INT(INTO, NEWVAL) do_SUBST_INT(&(INTO), (NEWVAL))
\f
+/* Calculate the rtx_cost of a single instruction. A return value of zero
+ indicates an instruction without a known cost. */
+
+static int
+combine_insn_cost (rtx pat)
+{
+ int i, cost;
+ rtx set;
+
+ /* Extract the single set rtx from the instruction pattern.
+ We can't use single_set since we only have the pattern. */
+ if (GET_CODE (pat) == SET)
+ set = pat;
+ else if (GET_CODE (pat) == PARALLEL)
+ {
+ set = NULL_RTX;
+ for (i = 0; i < XVECLEN (pat, 0); i++)
+ {
+ rtx x = XVECEXP (pat, 0, i);
+ if (GET_CODE (x) == SET)
+ {
+ if (set)
+ return 0;
+ set = x;
+ }
+ }
+ if (!set)
+ return 0;
+ }
+ else
+ return 0;
+
+ cost = rtx_cost (SET_SRC (set), SET);
+ return cost > 0 ? cost : COSTS_N_INSNS (1);
+}
+
+/* Subroutine of try_combine. Determine whether the combine replacement
+ patterns NEWPAT and NEWI2PAT are cheaper according to combine_insn_cost
+ that the original instruction sequence I1, I2 and I3. Note that I1
+ and/or NEWI2PAT may 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)
+{
+ int i1_cost, i2_cost, i3_cost;
+ int new_i2_cost, new_i3_cost;
+ int old_cost, new_cost;
+
+ /* Lookup the original combine_insn_costs. */
+ i2_cost = INSN_UID (i2) <= last_insn_cost
+ ? uid_insn_cost[INSN_UID (i2)] : 0;
+ i3_cost = INSN_UID (i3) <= last_insn_cost
+ ? uid_insn_cost[INSN_UID (i3)] : 0;
+
+ if (i1)
+ {
+ i1_cost = INSN_UID (i1) <= last_insn_cost
+ ? uid_insn_cost[INSN_UID (i1)] : 0;
+ old_cost = (i1_cost > 0 && i2_cost > 0 && i3_cost > 0)
+ ? i1_cost + i2_cost + i3_cost : 0;
+ }
+ else
+ {
+ old_cost = (i2_cost > 0 && i3_cost > 0) ? i2_cost + i3_cost : 0;
+ i1_cost = 0;
+ }
+
+ /* Calculate the replacement combine_insn_costs. */
+ new_i3_cost = combine_insn_cost (newpat);
+ if (newi2pat)
+ {
+ new_i2_cost = combine_insn_cost (newi2pat);
+ new_cost = (new_i2_cost > 0 && new_i3_cost > 0)
+ ? new_i2_cost + new_i3_cost : 0;
+ }
+ else
+ {
+ new_cost = new_i3_cost;
+ new_i2_cost = 0;
+ }
+
+ /* Disallow this recombination if both new_cost and old_cost are
+ greater than zero, and new_cost is greater than old cost. */
+ if (!undobuf.other_insn
+ && old_cost > 0
+ && new_cost > old_cost)
+ {
+ if (dump_file)
+ {
+ if (i1)
+ {
+ fprintf (dump_file,
+ "rejecting combination of insns %d, %d and %d\n",
+ INSN_UID (i1), INSN_UID (i2), INSN_UID (i3));
+ fprintf (dump_file, "original costs %d + %d + %d = %d\n",
+ i1_cost, i2_cost, i3_cost, old_cost);
+ }
+ else
+ {
+ fprintf (dump_file,
+ "rejecting combination of insns %d and %d\n",
+ INSN_UID (i2), INSN_UID (i3));
+ fprintf (dump_file, "original costs %d + %d = %d\n",
+ i2_cost, i3_cost, old_cost);
+ }
+
+ if (newi2pat)
+ {
+ fprintf (dump_file, "replacement costs %d + %d = %d\n",
+ new_i2_cost, new_i3_cost, new_cost);
+ }
+ else
+ fprintf (dump_file, "replacement cost %d\n", new_cost);
+ }
+
+ return false;
+ }
+
+ /* Update the uid_insn_cost array with the replacement costs. */
+ uid_insn_cost[INSN_UID (i2)] = new_i2_cost;
+ uid_insn_cost[INSN_UID (i3)] = new_i3_cost;
+ if (i1)
+ uid_insn_cost[INSN_UID (i1)] = 0;
+
+ return true;
+}
+\f
/* Main entry point for combiner. F is the first insn of the function.
NREGS is the first unused pseudo-reg number.
refresh_blocks = sbitmap_alloc (last_basic_block);
sbitmap_zero (refresh_blocks);
+ /* Allocate array of current combine_insn_costs. */
+ uid_insn_cost = xcalloc (max_uid_cuid + 1, sizeof (int));
+ last_insn_cost = max_uid_cuid;
+
for (insn = f, i = 0; insn; insn = NEXT_INSN (insn))
{
uid_cuid[INSN_UID (insn)] = ++i;
set_nonzero_bits_and_sign_copies (XEXP (links, 0), NULL_RTX,
NULL);
#endif
+
+ /* Record the current combine_insn_cost of this instruction. */
+ uid_insn_cost[INSN_UID (insn)] = combine_insn_cost (PATTERN (insn));
+ if (dump_file)
+ fprintf(dump_file, "insn_cost %d: %d\n",
+ INSN_UID (insn), uid_insn_cost[INSN_UID (insn)]);
}
if (GET_CODE (insn) == CODE_LABEL)
/* Clean up. */
sbitmap_free (refresh_blocks);
+ free (uid_insn_cost);
free (reg_stat);
free (uid_cuid);
are intervening stores. Also, don't move a volatile asm or
UNSPEC_VOLATILE across any other insns. */
|| (! all_adjacent
- && (((GET_CODE (src) != MEM
+ && (((!MEM_P (src)
|| ! find_reg_note (insn, REG_EQUIV, src))
&& use_crosses_set_p (src, INSN_CUID (insn)))
|| (GET_CODE (src) == ASM_OPERANDS && MEM_VOLATILE_P (src))
into the address of a MEM, so only prevent the combination if
i1 or i2 set the same MEM. */
if ((inner_dest != dest &&
- (GET_CODE (inner_dest) != MEM
+ (!MEM_P (inner_dest)
|| rtx_equal_p (i2dest, inner_dest)
|| (i1dest && rtx_equal_p (i1dest, inner_dest)))
&& (reg_overlap_mentioned_p (i2dest, inner_dest)
#if 0
if (!(GET_CODE (PATTERN (i3)) == SET
&& REG_P (SET_SRC (PATTERN (i3)))
- && GET_CODE (SET_DEST (PATTERN (i3))) == MEM
+ && MEM_P (SET_DEST (PATTERN (i3)))
&& (GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_INC
|| GET_CODE (XEXP (SET_DEST (PATTERN (i3)), 0)) == POST_DEC)))
/* It's not the exception. */
#ifdef INSN_SCHEDULING
/* If *SPLIT is a paradoxical SUBREG, when we split it, it should
be written as a ZERO_EXTEND. */
- if (split_code == SUBREG && GET_CODE (SUBREG_REG (*split)) == MEM)
+ if (split_code == SUBREG && MEM_P (SUBREG_REG (*split)))
{
#ifdef LOAD_EXTEND_OP
/* Or as a SIGN_EXTEND if LOAD_EXTEND_OP says that that's
}
#endif
+ /* Only allow this combination if combine_insn_costs reports that the
+ replacement instructions are cheaper than the originals. */
+ if (!combine_validate_cost (i1, i2, i3, newpat, newi2pat))
+ {
+ undo_all ();
+ return 0;
+ }
+
/* We now know that we can do this combination. Merge the insns and
update the status of registers and LOG_LINKS. */
PATTERN (i2) = newi2pat;
}
else
- {
- PUT_CODE (i2, NOTE);
- NOTE_LINE_NUMBER (i2) = NOTE_INSN_DELETED;
- NOTE_SOURCE_FILE (i2) = 0;
- }
+ SET_INSN_DELETED (i2);
if (i1)
{
LOG_LINKS (i1) = 0;
REG_NOTES (i1) = 0;
- PUT_CODE (i1, NOTE);
- NOTE_LINE_NUMBER (i1) = NOTE_INSN_DELETED;
- NOTE_SOURCE_FILE (i1) = 0;
+ SET_INSN_DELETED (i1);
}
/* Get death notes for everything that is now used in either I3 or
#ifdef INSN_SCHEDULING
/* If we are making a paradoxical SUBREG invalid, it becomes a split
point. */
- if (GET_CODE (SUBREG_REG (x)) == MEM)
+ if (MEM_P (SUBREG_REG (x)))
return loc;
#endif
return find_split_point (&SUBREG_REG (x), insn);
/* Don't change the mode of the MEM if that would change the meaning
of the address. */
- if (GET_CODE (SUBREG_REG (x)) == MEM
+ if (MEM_P (SUBREG_REG (x))
&& (MEM_VOLATILE_P (SUBREG_REG (x))
|| mode_dependent_address_p (XEXP (SUBREG_REG (x), 0))))
return gen_rtx_CLOBBER (mode, const0_rtx);
&& SUBREG_BYTE (src) == 0
&& (GET_MODE_SIZE (GET_MODE (src))
> GET_MODE_SIZE (GET_MODE (SUBREG_REG (src))))
- && GET_CODE (SUBREG_REG (src)) == MEM)
+ && MEM_P (SUBREG_REG (src)))
{
SUBST (SET_SRC (x),
gen_rtx_fmt_e (LOAD_EXTEND_OP (GET_MODE (SUBREG_REG (src))),
The subreg adds or removes high bits; its mode is
irrelevant to the meaning of this extraction,
since POS and LEN count from the lsb. */
- if (GET_CODE (SUBREG_REG (inner)) == MEM)
+ if (MEM_P (SUBREG_REG (inner)))
is_mode = GET_MODE (SUBREG_REG (inner));
inner = SUBREG_REG (inner);
}
if (tmode != BLKmode
&& ! (spans_byte && inner_mode != tmode)
&& ((pos_rtx == 0 && (pos % BITS_PER_WORD) == 0
- && GET_CODE (inner) != MEM
+ && !MEM_P (inner)
&& (! in_dest
|| (REG_P (inner)
&& have_insn_for (STRICT_LOW_PART, tmode))))
- || (GET_CODE (inner) == MEM && pos_rtx == 0
+ || (MEM_P (inner) && pos_rtx == 0
&& (pos
% (STRICT_ALIGNMENT ? GET_MODE_ALIGNMENT (tmode)
: BITS_PER_UNIT)) == 0
If INNER is not a MEM, get a piece consisting of just the field
of interest (in this case POS % BITS_PER_WORD must be 0). */
- if (GET_CODE (inner) == MEM)
+ if (MEM_P (inner))
{
HOST_WIDE_INT offset;
make a STRICT_LOW_PART unless we made a MEM. */
if (in_dest)
- return (GET_CODE (new) == MEM ? new
+ return (MEM_P (new) ? new
: (GET_CODE (new) != SUBREG
? gen_rtx_CLOBBER (tmode, const0_rtx)
: gen_rtx_STRICT_LOW_PART (VOIDmode, new)));
length is not 1. In all other cases, we would only be going outside
our object in cases when an original shift would have been
undefined. */
- if (! spans_byte && GET_CODE (inner) == MEM
+ if (! spans_byte && MEM_P (inner)
&& ((pos_rtx == 0 && pos + len > GET_MODE_BITSIZE (is_mode))
|| (pos_rtx != 0 && len != 1)))
return 0;
/* If this is not from memory, the desired mode is wanted_inner_reg_mode;
if we have to change the mode of memory and cannot, the desired mode is
EXTRACTION_MODE. */
- if (GET_CODE (inner) != MEM)
+ if (!MEM_P (inner))
wanted_inner_mode = wanted_inner_reg_mode;
else if (inner_mode != wanted_inner_mode
&& (mode_dependent_address_p (XEXP (inner, 0))
If it's a MEM we need to recompute POS relative to that.
However, if we're extracting from (or inserting into) a register,
we want to recompute POS relative to wanted_inner_mode. */
- int width = (GET_CODE (inner) == MEM
+ int width = (MEM_P (inner)
? GET_MODE_BITSIZE (is_mode)
: GET_MODE_BITSIZE (wanted_inner_mode));
pos_rtx
= gen_rtx_MINUS (GET_MODE (pos_rtx), GEN_INT (width - len), pos_rtx);
/* POS may be less than 0 now, but we check for that below.
- Note that it can only be less than 0 if GET_CODE (inner) != MEM. */
+ Note that it can only be less than 0 if !MEM_P (inner). */
}
/* If INNER has a wider mode, make it smaller. If this is a constant
the value. */
if (wanted_inner_mode != VOIDmode
&& GET_MODE_SIZE (wanted_inner_mode) < GET_MODE_SIZE (is_mode)
- && ((GET_CODE (inner) == MEM
+ && ((MEM_P (inner)
&& (inner_mode == wanted_inner_mode
|| (! mode_dependent_address_p (XEXP (inner, 0))
&& ! MEM_VOLATILE_P (inner))))))
/* If INNER is not memory, we can always get it into the proper mode. If we
are changing its mode, POS must be a constant and smaller than the size
of the new mode. */
- else if (GET_CODE (inner) != MEM)
+ else if (!MEM_P (inner))
{
if (GET_MODE (inner) != wanted_inner_mode
&& (pos_rtx != 0
/* Check for a paradoxical SUBREG of a MEM compared with the MEM.
Note that all SUBREGs of MEM are paradoxical; otherwise they
would have been rewritten. */
- if (GET_CODE (x) == MEM && GET_CODE (y) == SUBREG
- && GET_CODE (SUBREG_REG (y)) == MEM
+ if (MEM_P (x) && GET_CODE (y) == SUBREG
+ && MEM_P (SUBREG_REG (y))
&& rtx_equal_p (SUBREG_REG (y),
gen_lowpart (GET_MODE (SUBREG_REG (y)), x)))
return 1;
- if (GET_CODE (y) == MEM && GET_CODE (x) == SUBREG
- && GET_CODE (SUBREG_REG (x)) == MEM
+ if (MEM_P (y) && GET_CODE (x) == SUBREG
+ && MEM_P (SUBREG_REG (x))
&& rtx_equal_p (SUBREG_REG (x),
gen_lowpart (GET_MODE (SUBREG_REG (x)), y)))
return 1;
/* X might be a paradoxical (subreg (mem)). In that case, gen_lowpart
won't know what to do. So we will strip off the SUBREG here and
process normally. */
- if (GET_CODE (x) == SUBREG && GET_CODE (SUBREG_REG (x)) == MEM)
+ if (GET_CODE (x) == SUBREG && MEM_P (SUBREG_REG (x)))
{
x = SUBREG_REG (x);
if (GET_MODE (x) == mode)
if (result)
return result;
- if (GET_CODE (x) == MEM)
+ if (MEM_P (x))
{
int offset = 0;
break;
case GEU:
- /* >= C is equivalent to < (C - 1). */
+ /* >= C is equivalent to > (C - 1). */
if (const_op > 1)
{
const_op -= 1;
else
record_value_for_reg (dest, record_dead_insn, NULL_RTX);
}
- else if (GET_CODE (dest) == MEM
+ else if (MEM_P (dest)
/* Ignore pushes, they clobber nothing. */
&& ! push_operand (dest, GET_MODE (dest)))
mem_last_set = INSN_CUID (record_dead_insn);
/* 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. */
- else if (GET_CODE (x) == MEM && ! RTX_UNCHANGING_P (x)
+ else if (MEM_P (x) && ! RTX_UNCHANGING_P (x)
&& INSN_CUID (insn) <= mem_last_set)
{
if (replace)
case CLOBBER:
/* If we are clobbering a MEM, mark any hard registers inside the
address as used. */
- if (GET_CODE (XEXP (x, 0)) == MEM)
+ if (MEM_P (XEXP (x, 0)))
mark_used_regs_combine (XEXP (XEXP (x, 0), 0));
return;
|| GET_CODE (testreg) == STRICT_LOW_PART)
testreg = XEXP (testreg, 0);
- if (GET_CODE (testreg) == MEM)
+ if (MEM_P (testreg))
mark_used_regs_combine (XEXP (testreg, 0));
mark_used_regs_combine (SET_SRC (x));
For a REG (the only other possibility), the entire value is
being replaced so the old value is not used in this insn. */
- if (GET_CODE (dest) == MEM)
+ if (MEM_P (dest))
move_deaths (XEXP (dest, 0), maybe_kill_insn, from_cuid,
to_insn, pnotes);
return;
distribute_notes (old_notes, tem, tem, NULL_RTX);
distribute_links (LOG_LINKS (tem));
- PUT_CODE (tem, NOTE);
- NOTE_LINE_NUMBER (tem) = NOTE_INSN_DELETED;
- NOTE_SOURCE_FILE (tem) = 0;
+ SET_INSN_DELETED (tem);
#ifdef HAVE_cc0
/* Delete the setter too. */
cc0_setter, NULL_RTX);
distribute_links (LOG_LINKS (cc0_setter));
- PUT_CODE (cc0_setter, NOTE);
- NOTE_LINE_NUMBER (cc0_setter)
- = NOTE_INSN_DELETED;
- NOTE_SOURCE_FILE (cc0_setter) = 0;
+ SET_INSN_DELETED (cc0_setter);
}
#endif
}
rtx x = *loc;
if (x != NULL_RTX
- && (REG_P (x) || GET_CODE (x) == MEM)
+ && (REG_P (x) || MEM_P (x))
&& ! reg_mentioned_p (x, (rtx) expr))
return 1;
return 0;