/* Loop unrolling and peeling.
- Copyright (C) 2002, 2003, 2004 Free Software Foundation, Inc.
+ Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
This file is part of GCC.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING. If not, write to the Free
-Software Foundation, 59 Temple Place - Suite 330, Boston, MA
-02111-1307, USA. */
+Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
+02110-1301, USA. */
#include "config.h"
#include "system.h"
#include "tm.h"
#include "rtl.h"
#include "hard-reg-set.h"
+#include "obstack.h"
#include "basic-block.h"
#include "cfgloop.h"
#include "cfglayout.h"
#include "params.h"
#include "output.h"
#include "expr.h"
+#include "hashtab.h"
+#include "recog.h"
/* This pass performs loop unrolling and peeling. We only perform these
optimizations on innermost loops (with single exception) because
showed that this choice may affect performance in order of several %.
*/
+/* Information about induction variables to split. */
+
+struct iv_to_split
+{
+ rtx insn; /* The insn in that the induction variable occurs. */
+ rtx base_var; /* The variable on that the values in the further
+ iterations are based. */
+ rtx step; /* Step of the induction variable. */
+ unsigned n_loc;
+ unsigned loc[3]; /* Location where the definition of the induction
+ variable occurs in the insn. For example if
+ N_LOC is 2, the expression is located at
+ XEXP (XEXP (single_set, loc[0]), loc[1]). */
+};
+
+/* Information about accumulators to expand. */
+
+struct var_to_expand
+{
+ rtx insn; /* The insn in that the variable expansion occurs. */
+ rtx reg; /* The accumulator which is expanded. */
+ VEC(rtx,heap) *var_expansions; /* The copies of the accumulator which is expanded. */
+ enum rtx_code op; /* The type of the accumulation - addition, subtraction
+ or multiplication. */
+ int expansion_count; /* Count the number of expansions generated so far. */
+ int reuse_expansion; /* The expansion we intend to reuse to expand
+ the accumulator. If REUSE_EXPANSION is 0 reuse
+ the original accumulator. Else use
+ var_expansions[REUSE_EXPANSION - 1]. */
+};
+
+/* Information about optimization applied in
+ the unrolled loop. */
+
+struct opt_info
+{
+ htab_t insns_to_split; /* A hashtable of insns to split. */
+ htab_t insns_with_var_to_expand; /* A hashtable of insns with accumulators
+ to expand. */
+ unsigned first_new_block; /* The first basic block that was
+ duplicated. */
+ basic_block loop_exit; /* The loop exit basic block. */
+ basic_block loop_preheader; /* The loop preheader basic block. */
+};
+
static void decide_unrolling_and_peeling (struct loops *, int);
static void peel_loops_completely (struct loops *, int);
static void decide_peel_simple (struct loop *, int);
static void unroll_loop_stupid (struct loops *, struct loop *);
static void unroll_loop_constant_iterations (struct loops *, struct loop *);
static void unroll_loop_runtime_iterations (struct loops *, struct loop *);
+static struct opt_info *analyze_insns_in_loop (struct loop *);
+static void opt_info_start_duplication (struct opt_info *);
+static void apply_opt_in_copies (struct opt_info *, unsigned, bool, bool);
+static void free_opt_info (struct opt_info *);
+static struct var_to_expand *analyze_insn_to_expand_var (struct loop*, rtx);
+static bool referenced_in_one_insn_in_loop_p (struct loop *, rtx);
+static struct iv_to_split *analyze_iv_to_split_insn (rtx);
+static void expand_var_during_unrolling (struct var_to_expand *, rtx);
+static int insert_var_expansion_initialization (void **, void *);
+static int combine_var_copies_in_loop_exit (void **, void *);
+static int release_var_copies (void **, void *);
+static rtx get_expansion (struct var_to_expand *);
/* Unroll and/or peel (depending on FLAGS) LOOPS. */
void
/* Check number of iterations. */
if (!desc->simple_p
|| desc->assumptions
+ || desc->infinite
|| !desc->const_iter
|| desc->niter != 0)
{
/* Check number of iterations. */
if (!desc->simple_p
|| desc->assumptions
- || !desc->const_iter)
+ || !desc->const_iter
+ || desc->infinite)
{
if (dump_file)
fprintf (dump_file,
sbitmap wont_exit;
unsigned HOST_WIDE_INT npeel;
unsigned n_remove_edges, i;
- edge *remove_edges, ei;
+ edge *remove_edges, ein;
struct niter_desc *desc = get_simple_loop_desc (loop);
-
+ struct opt_info *opt_info = NULL;
+
npeel = desc->niter;
if (npeel)
{
- int ok;
-
+ bool ok;
+
wont_exit = sbitmap_alloc (npeel + 1);
sbitmap_ones (wont_exit);
RESET_BIT (wont_exit, 0);
if (desc->noloop_assumptions)
RESET_BIT (wont_exit, 1);
- remove_edges = xcalloc (npeel, sizeof (edge));
+ remove_edges = XCNEWVEC (edge, npeel);
n_remove_edges = 0;
+ if (flag_split_ivs_in_unroller)
+ opt_info = analyze_insns_in_loop (loop);
+
+ opt_info_start_duplication (opt_info);
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
loops, npeel,
wont_exit, desc->out_edge,
remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | DLTHE_FLAG_COMPLETTE_PEEL
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER : 0));
gcc_assert (ok);
free (wont_exit);
+
+ if (opt_info)
+ {
+ apply_opt_in_copies (opt_info, npeel, false, true);
+ free_opt_info (opt_info);
+ }
/* Remove the exit edges. */
for (i = 0; i < n_remove_edges; i++)
free (remove_edges);
}
- ei = desc->in_edge;
+ ein = desc->in_edge;
free_simple_loop_desc (loop);
/* Now remove the unreachable part of the last iteration and cancel
the loop. */
- remove_path (loops, ei);
+ remove_path (loops, ein);
if (dump_file)
fprintf (dump_file, ";; Peeled loop completely, %d times\n", (int) npeel);
unsigned max_unroll = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
bool exit_at_end = loop_exit_at_end_p (loop);
- int ok;
-
+ struct opt_info *opt_info = NULL;
+ bool ok;
+
niter = desc->niter;
- /* Should not assert out here (such loop should be peeled instead). */
+ /* Should not get here (such loop should be peeled instead). */
gcc_assert (niter > max_unroll + 1);
exit_mod = niter % (max_unroll + 1);
wont_exit = sbitmap_alloc (max_unroll + 1);
sbitmap_ones (wont_exit);
- remove_edges = xcalloc (max_unroll + exit_mod + 1, sizeof (edge));
+ remove_edges = XCNEWVEC (edge, max_unroll + exit_mod + 1);
n_remove_edges = 0;
-
+ if (flag_split_ivs_in_unroller
+ || flag_variable_expansion_in_unroller)
+ opt_info = analyze_insns_in_loop (loop);
+
if (!exit_at_end)
{
/* The exit is not at the end of the loop; leave exit test
if (exit_mod)
{
- int ok;
-
- ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ opt_info_start_duplication (opt_info);
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
loops, exit_mod,
wont_exit, desc->out_edge,
remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info && exit_mod > 1
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
+ if (opt_info && exit_mod > 1)
+ apply_opt_in_copies (opt_info, exit_mod, false, false);
+
desc->noloop_assumptions = NULL_RTX;
desc->niter -= exit_mod;
desc->niter_max -= exit_mod;
if (exit_mod != max_unroll
|| desc->noloop_assumptions)
{
- int ok;
-
RESET_BIT (wont_exit, 0);
if (desc->noloop_assumptions)
RESET_BIT (wont_exit, 1);
-
+
+ opt_info_start_duplication (opt_info);
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
loops, exit_mod + 1,
wont_exit, desc->out_edge,
remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info && exit_mod > 0
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
+
+ if (opt_info && exit_mod > 0)
+ apply_opt_in_copies (opt_info, exit_mod + 1, false, false);
desc->niter -= exit_mod + 1;
desc->niter_max -= exit_mod + 1;
}
/* Now unroll the loop. */
+
+ opt_info_start_duplication (opt_info);
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
loops, max_unroll,
wont_exit, desc->out_edge,
remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
+ if (opt_info)
+ {
+ apply_opt_in_copies (opt_info, max_unroll, true, true);
+ free_opt_info (opt_info);
+ }
+
free (wont_exit);
if (exit_at_end)
{
- basic_block exit_block = desc->in_edge->src->rbi->copy;
+ basic_block exit_block = get_bb_copy (desc->in_edge->src);
/* Find a new in and out edge; they are in the last copy we have made. */
- if (exit_block->succ->dest == desc->out_edge->dest)
+ if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
{
- desc->out_edge = exit_block->succ;
- desc->in_edge = exit_block->succ->succ_next;
+ desc->out_edge = EDGE_SUCC (exit_block, 0);
+ desc->in_edge = EDGE_SUCC (exit_block, 1);
}
else
{
- desc->out_edge = exit_block->succ->succ_next;
- desc->in_edge = exit_block->succ;
+ desc->out_edge = EDGE_SUCC (exit_block, 1);
+ desc->in_edge = EDGE_SUCC (exit_block, 0);
}
}
unsigned max_unroll = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
bool exit_at_end = loop_exit_at_end_p (loop);
- int ok;
-
+ struct opt_info *opt_info = NULL;
+ bool ok;
+
+ if (flag_split_ivs_in_unroller
+ || flag_variable_expansion_in_unroller)
+ opt_info = analyze_insns_in_loop (loop);
+
/* Remember blocks whose dominators will have to be updated. */
- dom_bbs = xcalloc (n_basic_blocks, sizeof (basic_block));
+ dom_bbs = XCNEWVEC (basic_block, n_basic_blocks);
n_dom_bbs = 0;
body = get_loop_body (loop);
/* Precondition the loop. */
loop_split_edge_with (loop_preheader_edge (loop), init_code);
- remove_edges = xcalloc (max_unroll + n_peel + 1, sizeof (edge));
+ remove_edges = XCNEWVEC (edge, max_unroll + n_peel + 1);
n_remove_edges = 0;
wont_exit = sbitmap_alloc (max_unroll + 2);
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
branch_code = compare_and_jump_seq (copy_rtx (niter), GEN_INT (j), EQ,
- block_label (preheader), p, NULL_RTX);
+ block_label (preheader), p,
+ NULL_RTX);
- swtch = loop_split_edge_with (swtch->pred, branch_code);
+ swtch = loop_split_edge_with (single_pred_edge (swtch), branch_code);
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
- swtch->succ->probability = REG_BR_PROB_BASE - p;
+ single_pred_edge (swtch)->probability = REG_BR_PROB_BASE - p;
e = make_edge (swtch, preheader,
- swtch->succ->flags & EDGE_IRREDUCIBLE_LOOP);
+ single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
e->probability = p;
}
swtch = ezc_swtch;
preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
branch_code = compare_and_jump_seq (copy_rtx (niter), const0_rtx, EQ,
- block_label (preheader), p, NULL_RTX);
+ block_label (preheader), p,
+ NULL_RTX);
- swtch = loop_split_edge_with (swtch->succ, branch_code);
+ swtch = loop_split_edge_with (single_succ_edge (swtch), branch_code);
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
- swtch->succ->probability = REG_BR_PROB_BASE - p;
+ single_succ_edge (swtch)->probability = REG_BR_PROB_BASE - p;
e = make_edge (swtch, preheader,
- swtch->succ->flags & EDGE_IRREDUCIBLE_LOOP);
+ single_succ_edge (swtch)->flags & EDGE_IRREDUCIBLE_LOOP);
e->probability = p;
}
sbitmap_ones (wont_exit);
RESET_BIT (wont_exit, may_exit_copy);
-
+ opt_info_start_duplication (opt_info);
+
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
loops, max_unroll,
wont_exit, desc->out_edge,
remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
+
+ if (opt_info)
+ {
+ apply_opt_in_copies (opt_info, max_unroll, true, true);
+ free_opt_info (opt_info);
+ }
free (wont_exit);
if (exit_at_end)
{
- basic_block exit_block = desc->in_edge->src->rbi->copy;
- /* Find a new in and out edge; they are in the last copy we have made. */
+ basic_block exit_block = get_bb_copy (desc->in_edge->src);
+ /* Find a new in and out edge; they are in the last copy we have
+ made. */
- if (exit_block->succ->dest == desc->out_edge->dest)
+ if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
{
- desc->out_edge = exit_block->succ;
- desc->in_edge = exit_block->succ->succ_next;
+ desc->out_edge = EDGE_SUCC (exit_block, 0);
+ desc->in_edge = EDGE_SUCC (exit_block, 1);
}
else
{
- desc->out_edge = exit_block->succ->succ_next;
- desc->in_edge = exit_block->succ;
+ desc->out_edge = EDGE_SUCC (exit_block, 1);
+ desc->in_edge = EDGE_SUCC (exit_block, 0);
}
}
the correct new number of iterations is this: */
gcc_assert (!desc->const_iter);
desc->niter_expr =
- simplify_gen_binary (UDIV, desc->mode, old_niter, GEN_INT (max_unroll + 1));
+ simplify_gen_binary (UDIV, desc->mode, old_niter,
+ GEN_INT (max_unroll + 1));
desc->niter_max /= max_unroll + 1;
if (exit_at_end)
{
";; Unrolled loop %d times, counting # of iterations "
"in runtime, %i insns\n",
max_unroll, num_loop_insns (loop));
+
+ if (dom_bbs)
+ free (dom_bbs);
}
/* Decide whether to simply peel LOOP and how much. */
sbitmap wont_exit;
unsigned npeel = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
- int ok;
-
+ struct opt_info *opt_info = NULL;
+ bool ok;
+
+ if (flag_split_ivs_in_unroller && npeel > 1)
+ opt_info = analyze_insns_in_loop (loop);
+
wont_exit = sbitmap_alloc (npeel + 1);
sbitmap_zero (wont_exit);
-
+
+ opt_info_start_duplication (opt_info);
+
ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
loops, npeel, wont_exit,
- NULL, NULL, NULL,
- DLTHE_FLAG_UPDATE_FREQ);
+ NULL, NULL,
+ NULL, DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
free (wont_exit);
+
+ if (opt_info)
+ {
+ apply_opt_in_copies (opt_info, npeel, false, false);
+ free_opt_info (opt_info);
+ }
if (desc->simple_p)
{
sbitmap wont_exit;
unsigned nunroll = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
- int ok;
-
+ struct opt_info *opt_info = NULL;
+ bool ok;
+
+ if (flag_split_ivs_in_unroller
+ || flag_variable_expansion_in_unroller)
+ opt_info = analyze_insns_in_loop (loop);
+
+
wont_exit = sbitmap_alloc (nunroll + 1);
sbitmap_zero (wont_exit);
-
+ opt_info_start_duplication (opt_info);
+
ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
loops, nunroll, wont_exit,
NULL, NULL, NULL,
- DLTHE_FLAG_UPDATE_FREQ);
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
gcc_assert (ok);
+
+ if (opt_info)
+ {
+ apply_opt_in_copies (opt_info, nunroll, true, true);
+ free_opt_info (opt_info);
+ }
free (wont_exit);
fprintf (dump_file, ";; Unrolled loop %d times, %i insns\n",
nunroll, num_loop_insns (loop));
}
+
+/* A hash function for information about insns to split. */
+
+static hashval_t
+si_info_hash (const void *ivts)
+{
+ return htab_hash_pointer (((struct iv_to_split *) ivts)->insn);
+}
+
+/* An equality functions for information about insns to split. */
+
+static int
+si_info_eq (const void *ivts1, const void *ivts2)
+{
+ const struct iv_to_split *i1 = ivts1;
+ const struct iv_to_split *i2 = ivts2;
+
+ return i1->insn == i2->insn;
+}
+
+/* Return a hash for VES, which is really a "var_to_expand *". */
+
+static hashval_t
+ve_info_hash (const void *ves)
+{
+ return htab_hash_pointer (((struct var_to_expand *) ves)->insn);
+}
+
+/* Return true if IVTS1 and IVTS2 (which are really both of type
+ "var_to_expand *") refer to the same instruction. */
+
+static int
+ve_info_eq (const void *ivts1, const void *ivts2)
+{
+ const struct var_to_expand *i1 = ivts1;
+ const struct var_to_expand *i2 = ivts2;
+
+ return i1->insn == i2->insn;
+}
+
+/* Returns true if REG is referenced in one insn in LOOP. */
+
+bool
+referenced_in_one_insn_in_loop_p (struct loop *loop, rtx reg)
+{
+ basic_block *body, bb;
+ unsigned i;
+ int count_ref = 0;
+ rtx insn;
+
+ body = get_loop_body (loop);
+ for (i = 0; i < loop->num_nodes; i++)
+ {
+ bb = body[i];
+
+ FOR_BB_INSNS (bb, insn)
+ {
+ if (rtx_referenced_p (reg, insn))
+ count_ref++;
+ }
+ }
+ return (count_ref == 1);
+}
+
+/* Determine whether INSN contains an accumulator
+ which can be expanded into separate copies,
+ one for each copy of the LOOP body.
+
+ for (i = 0 ; i < n; i++)
+ sum += a[i];
+
+ ==>
+
+ sum += a[i]
+ ....
+ i = i+1;
+ sum1 += a[i]
+ ....
+ i = i+1
+ sum2 += a[i];
+ ....
+
+ Return NULL if INSN contains no opportunity for expansion of accumulator.
+ Otherwise, allocate a VAR_TO_EXPAND structure, fill it with the relevant
+ information and return a pointer to it.
+*/
+
+static struct var_to_expand *
+analyze_insn_to_expand_var (struct loop *loop, rtx insn)
+{
+ rtx set, dest, src, op1;
+ struct var_to_expand *ves;
+ enum machine_mode mode1, mode2;
+
+ set = single_set (insn);
+ if (!set)
+ return NULL;
+
+ dest = SET_DEST (set);
+ src = SET_SRC (set);
+
+ if (GET_CODE (src) != PLUS
+ && GET_CODE (src) != MINUS
+ && GET_CODE (src) != MULT)
+ return NULL;
+
+ /* Hmm, this is a bit paradoxical. We know that INSN is a valid insn
+ in MD. But if there is no optab to generate the insn, we can not
+ perform the variable expansion. This can happen if an MD provides
+ an insn but not a named pattern to generate it, for example to avoid
+ producing code that needs additional mode switches like for x87/mmx.
+
+ So we check have_insn_for which looks for an optab for the operation
+ in SRC. If it doesn't exist, we can't perform the expansion even
+ though INSN is valid. */
+ if (!have_insn_for (GET_CODE (src), GET_MODE (src)))
+ return NULL;
+
+ if (!XEXP (src, 0))
+ return NULL;
+
+ op1 = XEXP (src, 0);
+
+ if (!REG_P (dest)
+ && !(GET_CODE (dest) == SUBREG
+ && REG_P (SUBREG_REG (dest))))
+ return NULL;
+
+ if (!rtx_equal_p (dest, op1))
+ return NULL;
+
+ if (!referenced_in_one_insn_in_loop_p (loop, dest))
+ return NULL;
+
+ if (rtx_referenced_p (dest, XEXP (src, 1)))
+ return NULL;
+
+ mode1 = GET_MODE (dest);
+ mode2 = GET_MODE (XEXP (src, 1));
+ if ((FLOAT_MODE_P (mode1)
+ || FLOAT_MODE_P (mode2))
+ && !flag_unsafe_math_optimizations)
+ return NULL;
+
+ /* Record the accumulator to expand. */
+ ves = XNEW (struct var_to_expand);
+ ves->insn = insn;
+ ves->var_expansions = VEC_alloc (rtx, heap, 1);
+ ves->reg = copy_rtx (dest);
+ ves->op = GET_CODE (src);
+ ves->expansion_count = 0;
+ ves->reuse_expansion = 0;
+ return ves;
+}
+
+/* Determine whether there is an induction variable in INSN that
+ we would like to split during unrolling.
+
+ I.e. replace
+
+ i = i + 1;
+ ...
+ i = i + 1;
+ ...
+ i = i + 1;
+ ...
+
+ type chains by
+
+ i0 = i + 1
+ ...
+ i = i0 + 1
+ ...
+ i = i0 + 2
+ ...
+
+ Return NULL if INSN contains no interesting IVs. Otherwise, allocate
+ an IV_TO_SPLIT structure, fill it with the relevant information and return a
+ pointer to it. */
+
+static struct iv_to_split *
+analyze_iv_to_split_insn (rtx insn)
+{
+ rtx set, dest;
+ struct rtx_iv iv;
+ struct iv_to_split *ivts;
+ bool ok;
+
+ /* For now we just split the basic induction variables. Later this may be
+ extended for example by selecting also addresses of memory references. */
+ set = single_set (insn);
+ if (!set)
+ return NULL;
+
+ dest = SET_DEST (set);
+ if (!REG_P (dest))
+ return NULL;
+
+ if (!biv_p (insn, dest))
+ return NULL;
+
+ ok = iv_analyze_result (insn, dest, &iv);
+
+ /* This used to be an assert under the assumption that if biv_p returns
+ true that iv_analyze_result must also return true. However, that
+ assumption is not strictly correct as evidenced by pr25569.
+
+ Returning NULL when iv_analyze_result returns false is safe and
+ avoids the problems in pr25569 until the iv_analyze_* routines
+ can be fixed, which is apparently hard and time consuming
+ according to their author. */
+ if (! ok)
+ return NULL;
+
+ if (iv.step == const0_rtx
+ || iv.mode != iv.extend_mode)
+ return NULL;
+
+ /* Record the insn to split. */
+ ivts = XNEW (struct iv_to_split);
+ ivts->insn = insn;
+ ivts->base_var = NULL_RTX;
+ ivts->step = iv.step;
+ ivts->n_loc = 1;
+ ivts->loc[0] = 1;
+
+ return ivts;
+}
+
+/* Determines which of insns in LOOP can be optimized.
+ Return a OPT_INFO struct with the relevant hash tables filled
+ with all insns to be optimized. The FIRST_NEW_BLOCK field
+ is undefined for the return value. */
+
+static struct opt_info *
+analyze_insns_in_loop (struct loop *loop)
+{
+ basic_block *body, bb;
+ unsigned i, num_edges = 0;
+ struct opt_info *opt_info = XCNEW (struct opt_info);
+ rtx insn;
+ struct iv_to_split *ivts = NULL;
+ struct var_to_expand *ves = NULL;
+ PTR *slot1;
+ PTR *slot2;
+ edge *edges = get_loop_exit_edges (loop, &num_edges);
+ bool can_apply = false;
+
+ iv_analysis_loop_init (loop);
+
+ body = get_loop_body (loop);
+
+ if (flag_split_ivs_in_unroller)
+ opt_info->insns_to_split = htab_create (5 * loop->num_nodes,
+ si_info_hash, si_info_eq, free);
+
+ /* Record the loop exit bb and loop preheader before the unrolling. */
+ if (!loop_preheader_edge (loop)->src)
+ {
+ loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
+ opt_info->loop_preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
+ }
+ else
+ opt_info->loop_preheader = loop_preheader_edge (loop)->src;
+
+ if (num_edges == 1
+ && !(edges[0]->flags & EDGE_COMPLEX))
+ {
+ opt_info->loop_exit = loop_split_edge_with (edges[0], NULL_RTX);
+ can_apply = true;
+ }
+
+ if (flag_variable_expansion_in_unroller
+ && can_apply)
+ opt_info->insns_with_var_to_expand = htab_create (5 * loop->num_nodes,
+ ve_info_hash, ve_info_eq, free);
+
+ for (i = 0; i < loop->num_nodes; i++)
+ {
+ bb = body[i];
+ if (!dominated_by_p (CDI_DOMINATORS, loop->latch, bb))
+ continue;
+
+ FOR_BB_INSNS (bb, insn)
+ {
+ if (!INSN_P (insn))
+ continue;
+
+ if (opt_info->insns_to_split)
+ ivts = analyze_iv_to_split_insn (insn);
+
+ if (ivts)
+ {
+ slot1 = htab_find_slot (opt_info->insns_to_split, ivts, INSERT);
+ *slot1 = ivts;
+ continue;
+ }
+
+ if (opt_info->insns_with_var_to_expand)
+ ves = analyze_insn_to_expand_var (loop, insn);
+
+ if (ves)
+ {
+ slot2 = htab_find_slot (opt_info->insns_with_var_to_expand, ves, INSERT);
+ *slot2 = ves;
+ }
+ }
+ }
+
+ free (edges);
+ free (body);
+ return opt_info;
+}
+
+/* Called just before loop duplication. Records start of duplicated area
+ to OPT_INFO. */
+
+static void
+opt_info_start_duplication (struct opt_info *opt_info)
+{
+ if (opt_info)
+ opt_info->first_new_block = last_basic_block;
+}
+
+/* Determine the number of iterations between initialization of the base
+ variable and the current copy (N_COPY). N_COPIES is the total number
+ of newly created copies. UNROLLING is true if we are unrolling
+ (not peeling) the loop. */
+
+static unsigned
+determine_split_iv_delta (unsigned n_copy, unsigned n_copies, bool unrolling)
+{
+ if (unrolling)
+ {
+ /* If we are unrolling, initialization is done in the original loop
+ body (number 0). */
+ return n_copy;
+ }
+ else
+ {
+ /* If we are peeling, the copy in that the initialization occurs has
+ number 1. The original loop (number 0) is the last. */
+ if (n_copy)
+ return n_copy - 1;
+ else
+ return n_copies;
+ }
+}
+
+/* Locate in EXPR the expression corresponding to the location recorded
+ in IVTS, and return a pointer to the RTX for this location. */
+
+static rtx *
+get_ivts_expr (rtx expr, struct iv_to_split *ivts)
+{
+ unsigned i;
+ rtx *ret = &expr;
+
+ for (i = 0; i < ivts->n_loc; i++)
+ ret = &XEXP (*ret, ivts->loc[i]);
+
+ return ret;
+}
+
+/* Allocate basic variable for the induction variable chain. Callback for
+ htab_traverse. */
+
+static int
+allocate_basic_variable (void **slot, void *data ATTRIBUTE_UNUSED)
+{
+ struct iv_to_split *ivts = *slot;
+ rtx expr = *get_ivts_expr (single_set (ivts->insn), ivts);
+
+ ivts->base_var = gen_reg_rtx (GET_MODE (expr));
+
+ return 1;
+}
+
+/* Insert initialization of basic variable of IVTS before INSN, taking
+ the initial value from INSN. */
+
+static void
+insert_base_initialization (struct iv_to_split *ivts, rtx insn)
+{
+ rtx expr = copy_rtx (*get_ivts_expr (single_set (insn), ivts));
+ rtx seq;
+
+ start_sequence ();
+ expr = force_operand (expr, ivts->base_var);
+ if (expr != ivts->base_var)
+ emit_move_insn (ivts->base_var, expr);
+ seq = get_insns ();
+ end_sequence ();
+
+ emit_insn_before (seq, insn);
+}
+
+/* Replace the use of induction variable described in IVTS in INSN
+ by base variable + DELTA * step. */
+
+static void
+split_iv (struct iv_to_split *ivts, rtx insn, unsigned delta)
+{
+ rtx expr, *loc, seq, incr, var;
+ enum machine_mode mode = GET_MODE (ivts->base_var);
+ rtx src, dest, set;
+
+ /* Construct base + DELTA * step. */
+ if (!delta)
+ expr = ivts->base_var;
+ else
+ {
+ incr = simplify_gen_binary (MULT, mode,
+ ivts->step, gen_int_mode (delta, mode));
+ expr = simplify_gen_binary (PLUS, GET_MODE (ivts->base_var),
+ ivts->base_var, incr);
+ }
+
+ /* Figure out where to do the replacement. */
+ loc = get_ivts_expr (single_set (insn), ivts);
+
+ /* If we can make the replacement right away, we're done. */
+ if (validate_change (insn, loc, expr, 0))
+ return;
+
+ /* Otherwise, force EXPR into a register and try again. */
+ start_sequence ();
+ var = gen_reg_rtx (mode);
+ expr = force_operand (expr, var);
+ if (expr != var)
+ emit_move_insn (var, expr);
+ seq = get_insns ();
+ end_sequence ();
+ emit_insn_before (seq, insn);
+
+ if (validate_change (insn, loc, var, 0))
+ return;
+
+ /* The last chance. Try recreating the assignment in insn
+ completely from scratch. */
+ set = single_set (insn);
+ gcc_assert (set);
+
+ start_sequence ();
+ *loc = var;
+ src = copy_rtx (SET_SRC (set));
+ dest = copy_rtx (SET_DEST (set));
+ src = force_operand (src, dest);
+ if (src != dest)
+ emit_move_insn (dest, src);
+ seq = get_insns ();
+ end_sequence ();
+
+ emit_insn_before (seq, insn);
+ delete_insn (insn);
+}
+
+
+/* Return one expansion of the accumulator recorded in struct VE. */
+
+static rtx
+get_expansion (struct var_to_expand *ve)
+{
+ rtx reg;
+
+ if (ve->reuse_expansion == 0)
+ reg = ve->reg;
+ else
+ reg = VEC_index (rtx, ve->var_expansions, ve->reuse_expansion - 1);
+
+ if (VEC_length (rtx, ve->var_expansions) == (unsigned) ve->reuse_expansion)
+ ve->reuse_expansion = 0;
+ else
+ ve->reuse_expansion++;
+
+ return reg;
+}
+
+
+/* Given INSN replace the uses of the accumulator recorded in VE
+ with a new register. */
+
+static void
+expand_var_during_unrolling (struct var_to_expand *ve, rtx insn)
+{
+ rtx new_reg, set;
+ bool really_new_expansion = false;
+
+ set = single_set (insn);
+ gcc_assert (set);
+
+ /* Generate a new register only if the expansion limit has not been
+ reached. Else reuse an already existing expansion. */
+ if (PARAM_VALUE (PARAM_MAX_VARIABLE_EXPANSIONS) > ve->expansion_count)
+ {
+ really_new_expansion = true;
+ new_reg = gen_reg_rtx (GET_MODE (ve->reg));
+ }
+ else
+ new_reg = get_expansion (ve);
+
+ validate_change (insn, &SET_DEST (set), new_reg, 1);
+ validate_change (insn, &XEXP (SET_SRC (set), 0), new_reg, 1);
+
+ if (apply_change_group ())
+ if (really_new_expansion)
+ {
+ VEC_safe_push (rtx, heap, ve->var_expansions, new_reg);
+ ve->expansion_count++;
+ }
+}
+
+/* Initialize the variable expansions in loop preheader.
+ Callbacks for htab_traverse. PLACE_P is the loop-preheader
+ basic block where the initialization of the expansions
+ should take place. */
+
+static int
+insert_var_expansion_initialization (void **slot, void *place_p)
+{
+ struct var_to_expand *ve = *slot;
+ basic_block place = (basic_block)place_p;
+ rtx seq, var, zero_init, insn;
+ unsigned i;
+
+ if (VEC_length (rtx, ve->var_expansions) == 0)
+ return 1;
+
+ start_sequence ();
+ if (ve->op == PLUS || ve->op == MINUS)
+ for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
+ {
+ zero_init = CONST0_RTX (GET_MODE (var));
+ emit_move_insn (var, zero_init);
+ }
+ else if (ve->op == MULT)
+ for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
+ {
+ zero_init = CONST1_RTX (GET_MODE (var));
+ emit_move_insn (var, zero_init);
+ }
+
+ seq = get_insns ();
+ end_sequence ();
+
+ insn = BB_HEAD (place);
+ while (!NOTE_INSN_BASIC_BLOCK_P (insn))
+ insn = NEXT_INSN (insn);
+
+ emit_insn_after (seq, insn);
+ /* Continue traversing the hash table. */
+ return 1;
+}
+
+/* Combine the variable expansions at the loop exit.
+ Callbacks for htab_traverse. PLACE_P is the loop exit
+ basic block where the summation of the expansions should
+ take place. */
+
+static int
+combine_var_copies_in_loop_exit (void **slot, void *place_p)
+{
+ struct var_to_expand *ve = *slot;
+ basic_block place = (basic_block)place_p;
+ rtx sum = ve->reg;
+ rtx expr, seq, var, insn;
+ unsigned i;
+
+ if (VEC_length (rtx, ve->var_expansions) == 0)
+ return 1;
+
+ start_sequence ();
+ if (ve->op == PLUS || ve->op == MINUS)
+ for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
+ {
+ sum = simplify_gen_binary (PLUS, GET_MODE (ve->reg),
+ var, sum);
+ }
+ else if (ve->op == MULT)
+ for (i = 0; VEC_iterate (rtx, ve->var_expansions, i, var); i++)
+ {
+ sum = simplify_gen_binary (MULT, GET_MODE (ve->reg),
+ var, sum);
+ }
+
+ expr = force_operand (sum, ve->reg);
+ if (expr != ve->reg)
+ emit_move_insn (ve->reg, expr);
+ seq = get_insns ();
+ end_sequence ();
+
+ insn = BB_HEAD (place);
+ while (!NOTE_INSN_BASIC_BLOCK_P (insn))
+ insn = NEXT_INSN (insn);
+
+ emit_insn_after (seq, insn);
+
+ /* Continue traversing the hash table. */
+ return 1;
+}
+
+/* Apply loop optimizations in loop copies using the
+ data which gathered during the unrolling. Structure
+ OPT_INFO record that data.
+
+ UNROLLING is true if we unrolled (not peeled) the loop.
+ REWRITE_ORIGINAL_BODY is true if we should also rewrite the original body of
+ the loop (as it should happen in complete unrolling, but not in ordinary
+ peeling of the loop). */
+
+static void
+apply_opt_in_copies (struct opt_info *opt_info,
+ unsigned n_copies, bool unrolling,
+ bool rewrite_original_loop)
+{
+ unsigned i, delta;
+ basic_block bb, orig_bb;
+ rtx insn, orig_insn, next;
+ struct iv_to_split ivts_templ, *ivts;
+ struct var_to_expand ve_templ, *ves;
+
+ /* Sanity check -- we need to put initialization in the original loop
+ body. */
+ gcc_assert (!unrolling || rewrite_original_loop);
+
+ /* Allocate the basic variables (i0). */
+ if (opt_info->insns_to_split)
+ htab_traverse (opt_info->insns_to_split, allocate_basic_variable, NULL);
+
+ for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
+ {
+ bb = BASIC_BLOCK (i);
+ orig_bb = get_bb_original (bb);
+
+ /* bb->aux holds position in copy sequence initialized by
+ duplicate_loop_to_header_edge. */
+ delta = determine_split_iv_delta ((size_t)bb->aux, n_copies,
+ unrolling);
+ bb->aux = 0;
+ orig_insn = BB_HEAD (orig_bb);
+ for (insn = BB_HEAD (bb); insn != NEXT_INSN (BB_END (bb)); insn = next)
+ {
+ next = NEXT_INSN (insn);
+ if (!INSN_P (insn))
+ continue;
+
+ while (!INSN_P (orig_insn))
+ orig_insn = NEXT_INSN (orig_insn);
+
+ ivts_templ.insn = orig_insn;
+ ve_templ.insn = orig_insn;
+
+ /* Apply splitting iv optimization. */
+ if (opt_info->insns_to_split)
+ {
+ ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
+
+ if (ivts)
+ {
+ gcc_assert (GET_CODE (PATTERN (insn))
+ == GET_CODE (PATTERN (orig_insn)));
+
+ if (!delta)
+ insert_base_initialization (ivts, insn);
+ split_iv (ivts, insn, delta);
+ }
+ }
+ /* Apply variable expansion optimization. */
+ if (unrolling && opt_info->insns_with_var_to_expand)
+ {
+ ves = htab_find (opt_info->insns_with_var_to_expand, &ve_templ);
+ if (ves)
+ {
+ gcc_assert (GET_CODE (PATTERN (insn))
+ == GET_CODE (PATTERN (orig_insn)));
+ expand_var_during_unrolling (ves, insn);
+ }
+ }
+ orig_insn = NEXT_INSN (orig_insn);
+ }
+ }
+
+ if (!rewrite_original_loop)
+ return;
+
+ /* Initialize the variable expansions in the loop preheader
+ and take care of combining them at the loop exit. */
+ if (opt_info->insns_with_var_to_expand)
+ {
+ htab_traverse (opt_info->insns_with_var_to_expand,
+ insert_var_expansion_initialization,
+ opt_info->loop_preheader);
+ htab_traverse (opt_info->insns_with_var_to_expand,
+ combine_var_copies_in_loop_exit,
+ opt_info->loop_exit);
+ }
+
+ /* Rewrite also the original loop body. Find them as originals of the blocks
+ in the last copied iteration, i.e. those that have
+ get_bb_copy (get_bb_original (bb)) == bb. */
+ for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
+ {
+ bb = BASIC_BLOCK (i);
+ orig_bb = get_bb_original (bb);
+ if (get_bb_copy (orig_bb) != bb)
+ continue;
+
+ delta = determine_split_iv_delta (0, n_copies, unrolling);
+ for (orig_insn = BB_HEAD (orig_bb);
+ orig_insn != NEXT_INSN (BB_END (bb));
+ orig_insn = next)
+ {
+ next = NEXT_INSN (orig_insn);
+
+ if (!INSN_P (orig_insn))
+ continue;
+
+ ivts_templ.insn = orig_insn;
+ if (opt_info->insns_to_split)
+ {
+ ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
+ if (ivts)
+ {
+ if (!delta)
+ insert_base_initialization (ivts, orig_insn);
+ split_iv (ivts, orig_insn, delta);
+ continue;
+ }
+ }
+
+ }
+ }
+}
+
+/* Release the data structures used for the variable expansion
+ optimization. Callbacks for htab_traverse. */
+
+static int
+release_var_copies (void **slot, void *data ATTRIBUTE_UNUSED)
+{
+ struct var_to_expand *ve = *slot;
+
+ VEC_free (rtx, heap, ve->var_expansions);
+
+ /* Continue traversing the hash table. */
+ return 1;
+}
+
+/* Release OPT_INFO. */
+
+static void
+free_opt_info (struct opt_info *opt_info)
+{
+ if (opt_info->insns_to_split)
+ htab_delete (opt_info->insns_to_split);
+ if (opt_info->insns_with_var_to_expand)
+ {
+ htab_traverse (opt_info->insns_with_var_to_expand,
+ release_var_copies, NULL);
+ htab_delete (opt_info->insns_with_var_to_expand);
+ }
+ free (opt_info);
+}
OSDN Git Service
