/* Loop unrolling and peeling.
- Copyright (C) 2002, 2003, 2004 Free Software Foundation, Inc.
+ Copyright (C) 2002, 2003, 2004, 2005, 2007, 2008, 2010
+ Free Software Foundation, Inc.
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
-Software Foundation; either version 2, or (at your option) any later
+Software Foundation; either version 3, or (at your option) any later
version.
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
for more details.
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. */
+along with GCC; see the file COPYING3. If not see
+<http://www.gnu.org/licenses/>. */
#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 "expr.h"
#include "hashtab.h"
#include "recog.h"
+#include "target.h"
/* This pass performs loop unrolling and peeling. We only perform these
optimizations on innermost loops (with single exception) because
rtx base_var; /* The variable on that the values in the further
iterations are based. */
rtx step; /* Step of the induction variable. */
+ struct iv_to_split *next; /* Next entry in walking order. */
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]). */
+ XEXP (XEXP (single_set, loc[0]), loc[1]). */
};
-struct split_ivs_info
+/* 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. */
+ struct var_to_expand *next; /* Next entry in walking order. */
+ 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]. */
+ unsigned accum_pos; /* The position in which the accumulator is placed in
+ the insn src. For example in x = x + something
+ accum_pos is 0 while in x = something + x accum_pos
+ is 1. */
+};
+
+/* Information about optimization applied in
+ the unrolled loop. */
+
+struct opt_info
{
- htab_t insns_to_split; /* A hashtable of insns to split. */
- unsigned first_new_block; /* The first basic block that was
- duplicated. */
+ htab_t insns_to_split; /* A hashtable of insns to split. */
+ struct iv_to_split *iv_to_split_head; /* The first iv to split. */
+ struct iv_to_split **iv_to_split_tail; /* Pointer to the tail of the list. */
+ htab_t insns_with_var_to_expand; /* A hashtable of insns with accumulators
+ to expand. */
+ struct var_to_expand *var_to_expand_head; /* The first var to expand. */
+ struct var_to_expand **var_to_expand_tail; /* Pointer to the tail of the list. */
+ 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_unrolling_and_peeling (int);
+static void peel_loops_completely (int);
static void decide_peel_simple (struct loop *, int);
static void decide_peel_once_rolling (struct loop *, int);
static void decide_peel_completely (struct loop *, int);
static void decide_unroll_stupid (struct loop *, int);
static void decide_unroll_constant_iterations (struct loop *, int);
static void decide_unroll_runtime_iterations (struct loop *, int);
-static void peel_loop_simple (struct loops *, struct loop *);
-static void peel_loop_completely (struct loops *, struct loop *);
-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 split_ivs_info *analyze_ivs_to_split (struct loop *);
-static void si_info_start_duplication (struct split_ivs_info *);
-static void split_ivs_in_copies (struct split_ivs_info *, unsigned, bool, bool);
-static void free_si_info (struct split_ivs_info *);
+static void peel_loop_simple (struct loop *);
+static void peel_loop_completely (struct loop *);
+static void unroll_loop_stupid (struct loop *);
+static void unroll_loop_constant_iterations (struct loop *);
+static void unroll_loop_runtime_iterations (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, int *);
+static struct iv_to_split *analyze_iv_to_split_insn (rtx);
+static void expand_var_during_unrolling (struct var_to_expand *, rtx);
+static void insert_var_expansion_initialization (struct var_to_expand *,
+ basic_block);
+static void combine_var_copies_in_loop_exit (struct var_to_expand *,
+ basic_block);
+static rtx get_expansion (struct var_to_expand *);
/* Unroll and/or peel (depending on FLAGS) LOOPS. */
void
-unroll_and_peel_loops (struct loops *loops, int flags)
+unroll_and_peel_loops (int flags)
{
- struct loop *loop, *next;
+ struct loop *loop;
bool check;
+ loop_iterator li;
/* First perform complete loop peeling (it is almost surely a win,
and affects parameters for further decision a lot). */
- peel_loops_completely (loops, flags);
+ peel_loops_completely (flags);
/* Now decide rest of unrolling and peeling. */
- decide_unrolling_and_peeling (loops, flags);
-
- loop = loops->tree_root;
- while (loop->inner)
- loop = loop->inner;
+ decide_unrolling_and_peeling (flags);
/* Scan the loops, inner ones first. */
- while (loop != loops->tree_root)
+ FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST)
{
- if (loop->next)
- {
- next = loop->next;
- while (next->inner)
- next = next->inner;
- }
- else
- next = loop->outer;
-
check = true;
/* And perform the appropriate transformations. */
switch (loop->lpt_decision.decision)
/* Already done. */
gcc_unreachable ();
case LPT_PEEL_SIMPLE:
- peel_loop_simple (loops, loop);
+ peel_loop_simple (loop);
break;
case LPT_UNROLL_CONSTANT:
- unroll_loop_constant_iterations (loops, loop);
+ unroll_loop_constant_iterations (loop);
break;
case LPT_UNROLL_RUNTIME:
- unroll_loop_runtime_iterations (loops, loop);
+ unroll_loop_runtime_iterations (loop);
break;
case LPT_UNROLL_STUPID:
- unroll_loop_stupid (loops, loop);
+ unroll_loop_stupid (loop);
break;
case LPT_NONE:
check = false;
{
#ifdef ENABLE_CHECKING
verify_dominators (CDI_DOMINATORS);
- verify_loop_structure (loops);
+ verify_loop_structure ();
#endif
}
- loop = next;
}
iv_analysis_done ();
return true;
}
-/* Check whether to peel LOOPS (depending on FLAGS) completely and do so. */
+/* Depending on FLAGS, check whether to peel loops completely and do so. */
static void
-peel_loops_completely (struct loops *loops, int flags)
+peel_loops_completely (int flags)
{
- struct loop *loop, *next;
+ struct loop *loop;
+ loop_iterator li;
- loop = loops->tree_root;
- while (loop->inner)
- loop = loop->inner;
-
- while (loop != loops->tree_root)
+ /* Scan the loops, the inner ones first. */
+ FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST)
{
- if (loop->next)
- {
- next = loop->next;
- while (next->inner)
- next = next->inner;
- }
- else
- next = loop->outer;
-
loop->lpt_decision.decision = LPT_NONE;
if (dump_file)
if (loop->lpt_decision.decision == LPT_PEEL_COMPLETELY)
{
- peel_loop_completely (loops, loop);
+ peel_loop_completely (loop);
#ifdef ENABLE_CHECKING
verify_dominators (CDI_DOMINATORS);
- verify_loop_structure (loops);
+ verify_loop_structure ();
#endif
}
- loop = next;
}
}
-/* Decide whether unroll or peel LOOPS (depending on FLAGS) and how much. */
+/* Decide whether unroll or peel loops (depending on FLAGS) and how much. */
static void
-decide_unrolling_and_peeling (struct loops *loops, int flags)
+decide_unrolling_and_peeling (int flags)
{
- struct loop *loop = loops->tree_root, *next;
-
- while (loop->inner)
- loop = loop->inner;
+ struct loop *loop;
+ loop_iterator li;
/* Scan the loops, inner ones first. */
- while (loop != loops->tree_root)
+ FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST)
{
- if (loop->next)
- {
- next = loop->next;
- while (next->inner)
- next = next->inner;
- }
- else
- next = loop->outer;
-
loop->lpt_decision.decision = LPT_NONE;
if (dump_file)
fprintf (dump_file, "\n;; *** Considering loop %d ***\n", loop->num);
/* Do not peel cold areas. */
- if (!maybe_hot_bb_p (loop->header))
+ if (optimize_loop_for_size_p (loop))
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, cold area\n");
- loop = next;
continue;
}
if (dump_file)
fprintf (dump_file,
";; Not considering loop, cannot duplicate\n");
- loop = next;
continue;
}
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
- loop = next;
continue;
}
decide_unroll_stupid (loop, flags);
if (loop->lpt_decision.decision == LPT_NONE)
decide_peel_simple (loop, flags);
-
- loop = next;
}
}
/* Check number of iterations. */
if (!desc->simple_p
|| desc->assumptions
+ || desc->infinite
|| !desc->const_iter
|| desc->niter != 0)
{
}
/* Do not peel cold areas. */
- if (!maybe_hot_bb_p (loop->header))
+ if (optimize_loop_for_size_p (loop))
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, cold area\n");
/* Check number of iterations. */
if (!desc->simple_p
|| desc->assumptions
- || !desc->const_iter)
+ || !desc->const_iter
+ || desc->infinite)
{
if (dump_file)
fprintf (dump_file,
body; i++;
*/
static void
-peel_loop_completely (struct loops *loops, struct loop *loop)
+peel_loop_completely (struct loop *loop)
{
sbitmap wont_exit;
unsigned HOST_WIDE_INT npeel;
- unsigned n_remove_edges, i;
- edge *remove_edges, ei;
+ unsigned i;
+ VEC (edge, heap) *remove_edges;
+ edge ein;
struct niter_desc *desc = get_simple_loop_desc (loop);
- struct split_ivs_info *si_info = NULL;
+ struct opt_info *opt_info = NULL;
npeel = desc->niter;
if (npeel)
{
+ 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));
- n_remove_edges = 0;
+ remove_edges = NULL;
if (flag_split_ivs_in_unroller)
- si_info = analyze_ivs_to_split (loop);
-
- si_info_start_duplication (si_info);
- if (!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))
- abort ();
+ opt_info = analyze_insns_in_loop (loop);
+
+ opt_info_start_duplication (opt_info);
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ npeel,
+ wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ
+ | DLTHE_FLAG_COMPLETTE_PEEL
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER : 0));
+ gcc_assert (ok);
free (wont_exit);
- if (si_info)
- {
- split_ivs_in_copies (si_info, npeel, false, true);
- free_si_info (si_info);
- }
+ 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++)
- remove_path (loops, remove_edges[i]);
- free (remove_edges);
+ FOR_EACH_VEC_ELT (edge, remove_edges, i, ein)
+ remove_path (ein);
+ VEC_free (edge, heap, 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 (ein);
if (dump_file)
fprintf (dump_file, ";; Peeled loop completely, %d times\n", (int) npeel);
loop->lpt_decision.decision = LPT_UNROLL_CONSTANT;
loop->lpt_decision.times = best_unroll;
-
+
if (dump_file)
fprintf (dump_file,
";; Decided to unroll the constant times rolling loop, %d times.\n",
}
*/
static void
-unroll_loop_constant_iterations (struct loops *loops, struct loop *loop)
+unroll_loop_constant_iterations (struct loop *loop)
{
unsigned HOST_WIDE_INT niter;
unsigned exit_mod;
sbitmap wont_exit;
- unsigned n_remove_edges, i;
- edge *remove_edges;
+ unsigned i;
+ VEC (edge, heap) *remove_edges;
+ edge e;
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);
- struct split_ivs_info *si_info = NULL;
+ struct opt_info *opt_info = NULL;
+ bool ok;
niter = desc->niter;
wont_exit = sbitmap_alloc (max_unroll + 1);
sbitmap_ones (wont_exit);
- remove_edges = xcalloc (max_unroll + exit_mod + 1, sizeof (edge));
- n_remove_edges = 0;
-
- if (flag_split_ivs_in_unroller)
- si_info = analyze_ivs_to_split (loop);
+ remove_edges = NULL;
+ if (flag_split_ivs_in_unroller
+ || flag_variable_expansion_in_unroller)
+ opt_info = analyze_insns_in_loop (loop);
if (!exit_at_end)
{
if (exit_mod)
{
- si_info_start_duplication (si_info);
- if (!duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
- loops, exit_mod,
+ opt_info_start_duplication (opt_info);
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ exit_mod,
wont_exit, desc->out_edge,
- remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ))
- abort ();
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info && exit_mod > 1
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
- if (si_info && exit_mod > 1)
- split_ivs_in_copies (si_info, exit_mod, false, false);
+ 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;
if (desc->noloop_assumptions)
RESET_BIT (wont_exit, 1);
- si_info_start_duplication (si_info);
- if (!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))
- abort ();
+ opt_info_start_duplication (opt_info);
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ exit_mod + 1,
+ wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info && exit_mod > 0
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
- if (si_info && exit_mod > 0)
- split_ivs_in_copies (si_info, exit_mod + 1, false, false);
+ 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. */
- si_info_start_duplication (si_info);
- if (!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))
- abort ();
- if (si_info)
+ opt_info_start_duplication (opt_info);
+ ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
+ max_unroll,
+ wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
+
+ if (opt_info)
{
- split_ivs_in_copies (si_info, max_unroll, true, true);
- free_si_info (si_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);
}
}
desc->niter_expr = GEN_INT (desc->niter);
/* Remove the edges. */
- for (i = 0; i < n_remove_edges; i++)
- remove_path (loops, remove_edges[i]);
- free (remove_edges);
+ FOR_EACH_VEC_ELT (edge, remove_edges, i, e)
+ remove_path (e);
+ VEC_free (edge, heap, remove_edges);
if (dump_file)
fprintf (dump_file,
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
+ if (targetm.loop_unroll_adjust)
+ nunroll = targetm.loop_unroll_adjust (nunroll, loop);
+
/* Skip big loops. */
if (nunroll <= 1)
{
loop->lpt_decision.decision = LPT_UNROLL_RUNTIME;
loop->lpt_decision.times = i - 1;
-
+
if (dump_file)
fprintf (dump_file,
";; Decided to unroll the runtime computable "
loop->lpt_decision.times);
}
+/* Splits edge E and inserts the sequence of instructions INSNS on it, and
+ returns the newly created block. If INSNS is NULL_RTX, nothing is changed
+ and NULL is returned instead. */
+
+basic_block
+split_edge_and_insert (edge e, rtx insns)
+{
+ basic_block bb;
+
+ if (!insns)
+ return NULL;
+ bb = split_edge (e);
+ emit_insn_after (insns, BB_END (bb));
+
+ /* ??? We used to assume that INSNS can contain control flow insns, and
+ that we had to try to find sub basic blocks in BB to maintain a valid
+ CFG. For this purpose we used to set the BB_SUPERBLOCK flag on BB
+ and call break_superblocks when going out of cfglayout mode. But it
+ turns out that this never happens; and that if it does ever happen,
+ the verify_flow_info call in loop_optimizer_finalize would fail.
+
+ There are two reasons why we expected we could have control flow insns
+ in INSNS. The first is when a comparison has to be done in parts, and
+ the second is when the number of iterations is computed for loops with
+ the number of iterations known at runtime. In both cases, test cases
+ to get control flow in INSNS appear to be impossible to construct:
+
+ * If do_compare_rtx_and_jump needs several branches to do comparison
+ in a mode that needs comparison by parts, we cannot analyze the
+ number of iterations of the loop, and we never get to unrolling it.
+
+ * The code in expand_divmod that was suspected to cause creation of
+ branching code seems to be only accessed for signed division. The
+ divisions used by # of iterations analysis are always unsigned.
+ Problems might arise on architectures that emits branching code
+ for some operations that may appear in the unroller (especially
+ for division), but we have no such architectures.
+
+ Considering all this, it was decided that we should for now assume
+ that INSNS can in theory contain control flow insns, but in practice
+ it never does. So we don't handle the theoretical case, and should
+ a real failure ever show up, we have a pretty good clue for how to
+ fix it. */
+
+ return bb;
+}
+
/* Unroll LOOP for that we are able to count number of iterations in runtime
LOOP->LPT_DECISION.TIMES + 1 times. The transformation does this (with some
extra care for case n < 0):
}
*/
static void
-unroll_loop_runtime_iterations (struct loops *loops, struct loop *loop)
+unroll_loop_runtime_iterations (struct loop *loop)
{
rtx old_niter, niter, init_code, branch_code, tmp;
unsigned i, j, p;
- basic_block preheader, *body, *dom_bbs, swtch, ezc_swtch;
- unsigned n_dom_bbs;
+ basic_block preheader, *body, swtch, ezc_swtch;
+ VEC (basic_block, heap) *dom_bbs;
sbitmap wont_exit;
int may_exit_copy;
- unsigned n_peel, n_remove_edges;
- edge *remove_edges, e;
+ unsigned n_peel;
+ VEC (edge, heap) *remove_edges;
+ edge e;
bool extra_zero_check, last_may_exit;
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);
- struct split_ivs_info *si_info = NULL;
+ struct opt_info *opt_info = NULL;
+ bool ok;
- if (flag_split_ivs_in_unroller)
- si_info = analyze_ivs_to_split (loop);
+ 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));
- n_dom_bbs = 0;
+ dom_bbs = NULL;
body = get_loop_body (loop);
for (i = 0; i < loop->num_nodes; i++)
{
- unsigned nldom;
- basic_block *ldom;
+ VEC (basic_block, heap) *ldom;
+ basic_block bb;
- nldom = get_dominated_by (CDI_DOMINATORS, body[i], &ldom);
- for (j = 0; j < nldom; j++)
- if (!flow_bb_inside_loop_p (loop, ldom[j]))
- dom_bbs[n_dom_bbs++] = ldom[j];
+ ldom = get_dominated_by (CDI_DOMINATORS, body[i]);
+ FOR_EACH_VEC_ELT (basic_block, ldom, j, bb)
+ if (!flow_bb_inside_loop_p (loop, bb))
+ VEC_safe_push (basic_block, heap, dom_bbs, bb);
- free (ldom);
+ VEC_free (basic_block, heap, ldom);
}
free (body);
init_code = get_insns ();
end_sequence ();
+ unshare_all_rtl_in_chain (init_code);
/* Precondition the loop. */
- loop_split_edge_with (loop_preheader_edge (loop), init_code);
+ split_edge_and_insert (loop_preheader_edge (loop), init_code);
- remove_edges = xcalloc (max_unroll + n_peel + 1, sizeof (edge));
- n_remove_edges = 0;
+ remove_edges = NULL;
wont_exit = sbitmap_alloc (max_unroll + 2);
&& !desc->noloop_assumptions)
SET_BIT (wont_exit, 1);
ezc_swtch = loop_preheader_edge (loop)->src;
- if (!duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
- loops, 1,
- wont_exit, desc->out_edge, remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ))
- abort ();
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ 1, wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ);
+ gcc_assert (ok);
/* Record the place where switch will be built for preconditioning. */
- swtch = loop_split_edge_with (loop_preheader_edge (loop),
- NULL_RTX);
+ swtch = split_edge (loop_preheader_edge (loop));
for (i = 0; i < n_peel; i++)
{
sbitmap_zero (wont_exit);
if (i != n_peel - 1 || !last_may_exit)
SET_BIT (wont_exit, 1);
- if (!duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
- loops, 1,
- wont_exit, desc->out_edge, remove_edges, &n_remove_edges,
- DLTHE_FLAG_UPDATE_FREQ))
- abort ();
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ 1, wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ);
+ gcc_assert (ok);
/* Create item for switch. */
j = n_peel - i - (extra_zero_check ? 0 : 1);
p = REG_BR_PROB_BASE / (i + 2);
- preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
+ preheader = split_edge (loop_preheader_edge (loop));
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);
+ /* We rely on the fact that the compare and jump cannot be optimized out,
+ and hence the cfg we create is correct. */
+ gcc_assert (branch_code != NULL_RTX);
+
+ swtch = split_edge_and_insert (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;
}
/* Add branch for zero iterations. */
p = REG_BR_PROB_BASE / (max_unroll + 1);
swtch = ezc_swtch;
- preheader = loop_split_edge_with (loop_preheader_edge (loop), NULL_RTX);
+ preheader = split_edge (loop_preheader_edge (loop));
branch_code = compare_and_jump_seq (copy_rtx (niter), const0_rtx, EQ,
- block_label (preheader), p, NULL_RTX);
+ block_label (preheader), p,
+ NULL_RTX);
+ gcc_assert (branch_code != NULL_RTX);
- swtch = loop_split_edge_with (swtch->succ, branch_code);
+ swtch = split_edge_and_insert (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;
}
/* Recount dominators for outer blocks. */
- iterate_fix_dominators (CDI_DOMINATORS, dom_bbs, n_dom_bbs);
+ iterate_fix_dominators (CDI_DOMINATORS, dom_bbs, false);
/* And unroll loop. */
sbitmap_ones (wont_exit);
RESET_BIT (wont_exit, may_exit_copy);
-
- si_info_start_duplication (si_info);
- if (!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))
- abort ();
-
- if (si_info)
+ opt_info_start_duplication (opt_info);
+
+ ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
+ max_unroll,
+ wont_exit, desc->out_edge,
+ &remove_edges,
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
+
+ if (opt_info)
{
- split_ivs_in_copies (si_info, max_unroll, true, true);
- free_si_info (si_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. */
-
- if (exit_block->succ->dest == desc->out_edge->dest)
+ 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 (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);
}
}
/* Remove the edges. */
- for (i = 0; i < n_remove_edges; i++)
- remove_path (loops, remove_edges[i]);
- free (remove_edges);
+ FOR_EACH_VEC_ELT (edge, remove_edges, i, e)
+ remove_path (e);
+ VEC_free (edge, heap, remove_edges);
/* We must be careful when updating the number of iterations due to
preconditioning and the fact that the value must be valid at entry
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));
+
+ VEC_free (basic_block, heap, dom_bbs);
}
/* Decide whether to simply peel LOOP and how much. */
/* Success. */
loop->lpt_decision.decision = LPT_PEEL_SIMPLE;
loop->lpt_decision.times = npeel;
-
+
if (dump_file)
fprintf (dump_file, ";; Decided to simply peel the loop, %d times.\n",
loop->lpt_decision.times);
end: ;
*/
static void
-peel_loop_simple (struct loops *loops, struct loop *loop)
+peel_loop_simple (struct loop *loop)
{
sbitmap wont_exit;
unsigned npeel = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
- struct split_ivs_info *si_info = NULL;
+ struct opt_info *opt_info = NULL;
+ bool ok;
if (flag_split_ivs_in_unroller && npeel > 1)
- si_info = analyze_ivs_to_split (loop);
+ opt_info = analyze_insns_in_loop (loop);
wont_exit = sbitmap_alloc (npeel + 1);
sbitmap_zero (wont_exit);
- si_info_start_duplication (si_info);
- if (!duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
- loops, npeel, wont_exit, NULL, NULL, NULL,
- DLTHE_FLAG_UPDATE_FREQ))
- abort ();
+ opt_info_start_duplication (opt_info);
+
+ ok = duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
+ npeel, wont_exit, NULL,
+ NULL, DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
free (wont_exit);
- if (si_info)
+ if (opt_info)
{
- split_ivs_in_copies (si_info, npeel, false, false);
- free_si_info (si_info);
+ apply_opt_in_copies (opt_info, npeel, false, false);
+ free_opt_info (opt_info);
}
if (desc->simple_p)
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
+ if (targetm.loop_unroll_adjust)
+ nunroll = targetm.loop_unroll_adjust (nunroll, loop);
+
/* Skip big loops. */
if (nunroll <= 1)
{
loop->lpt_decision.decision = LPT_UNROLL_STUPID;
loop->lpt_decision.times = i - 1;
-
+
if (dump_file)
fprintf (dump_file,
";; Decided to unroll the loop stupidly, %d times.\n",
}
*/
static void
-unroll_loop_stupid (struct loops *loops, struct loop *loop)
+unroll_loop_stupid (struct loop *loop)
{
sbitmap wont_exit;
unsigned nunroll = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
- struct split_ivs_info *si_info = NULL;
+ 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);
- if (flag_split_ivs_in_unroller)
- si_info = analyze_ivs_to_split (loop);
wont_exit = sbitmap_alloc (nunroll + 1);
sbitmap_zero (wont_exit);
-
- si_info_start_duplication (si_info);
- if (!duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
- loops, nunroll, wont_exit, NULL, NULL, NULL,
- DLTHE_FLAG_UPDATE_FREQ))
- abort ();
-
- if (si_info)
+ opt_info_start_duplication (opt_info);
+
+ ok = duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
+ nunroll, wont_exit,
+ NULL, NULL,
+ DLTHE_FLAG_UPDATE_FREQ
+ | (opt_info
+ ? DLTHE_RECORD_COPY_NUMBER
+ : 0));
+ gcc_assert (ok);
+
+ if (opt_info)
{
- split_ivs_in_copies (si_info, nunroll, true, true);
- free_si_info (si_info);
+ apply_opt_in_copies (opt_info, nunroll, true, true);
+ free_opt_info (opt_info);
}
free (wont_exit);
static hashval_t
si_info_hash (const void *ivts)
{
- return htab_hash_pointer (((struct iv_to_split *) ivts)->insn);
+ return (hashval_t) INSN_UID (((const 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;
+ const struct iv_to_split *const i1 = (const struct iv_to_split *) ivts1;
+ const struct iv_to_split *const i2 = (const struct iv_to_split *) 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 (hashval_t) INSN_UID (((const 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 *const i1 = (const struct var_to_expand *) ivts1;
+ const struct var_to_expand *const i2 = (const struct var_to_expand *) ivts2;
return i1->insn == i2->insn;
}
+/* Returns true if REG is referenced in one nondebug insn in LOOP.
+ Set *DEBUG_USES to the number of debug insns that reference the
+ variable. */
+
+bool
+referenced_in_one_insn_in_loop_p (struct loop *loop, rtx reg,
+ int *debug_uses)
+{
+ 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))
+ continue;
+ else if (DEBUG_INSN_P (insn))
+ ++*debug_uses;
+ else if (++count_ref > 1)
+ break;
+ }
+ free (body);
+ return (count_ref == 1);
+}
+
+/* Reset the DEBUG_USES debug insns in LOOP that reference REG. */
+
+static void
+reset_debug_uses_in_loop (struct loop *loop, rtx reg, int debug_uses)
+{
+ basic_block *body, bb;
+ unsigned i;
+ rtx insn;
+
+ body = get_loop_body (loop);
+ for (i = 0; debug_uses && i < loop->num_nodes; i++)
+ {
+ bb = body[i];
+
+ FOR_BB_INSNS (bb, insn)
+ if (!DEBUG_INSN_P (insn) || !rtx_referenced_p (reg, insn))
+ continue;
+ else
+ {
+ validate_change (insn, &INSN_VAR_LOCATION_LOC (insn),
+ gen_rtx_UNKNOWN_VAR_LOC (), 0);
+ if (!--debug_uses)
+ break;
+ }
+ }
+ free (body);
+}
+
+/* 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;
+ struct var_to_expand *ves;
+ unsigned accum_pos;
+ enum rtx_code code;
+ int debug_uses = 0;
+
+ set = single_set (insn);
+ if (!set)
+ return NULL;
+
+ dest = SET_DEST (set);
+ src = SET_SRC (set);
+ code = GET_CODE (src);
+
+ if (code != PLUS && code != MINUS && code != MULT && code != FMA)
+ return NULL;
+
+ if (FLOAT_MODE_P (GET_MODE (dest)))
+ {
+ if (!flag_associative_math)
+ return NULL;
+ /* In the case of FMA, we're also changing the rounding. */
+ if (code == FMA && !flag_unsafe_math_optimizations)
+ 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 (code, GET_MODE (src)))
+ return NULL;
+
+ if (!REG_P (dest)
+ && !(GET_CODE (dest) == SUBREG
+ && REG_P (SUBREG_REG (dest))))
+ return NULL;
+
+ /* Find the accumulator use within the operation. */
+ if (code == FMA)
+ {
+ /* We only support accumulation via FMA in the ADD position. */
+ if (!rtx_equal_p (dest, XEXP (src, 2)))
+ return NULL;
+ accum_pos = 2;
+ }
+ else if (rtx_equal_p (dest, XEXP (src, 0)))
+ accum_pos = 0;
+ else if (rtx_equal_p (dest, XEXP (src, 1)))
+ {
+ /* The method of expansion that we are using; which includes the
+ initialization of the expansions with zero and the summation of
+ the expansions at the end of the computation will yield wrong
+ results for (x = something - x) thus avoid using it in that case. */
+ if (code == MINUS)
+ return NULL;
+ accum_pos = 1;
+ }
+ else
+ return NULL;
+
+ /* It must not otherwise be used. */
+ if (code == FMA)
+ {
+ if (rtx_referenced_p (dest, XEXP (src, 0))
+ || rtx_referenced_p (dest, XEXP (src, 1)))
+ return NULL;
+ }
+ else if (rtx_referenced_p (dest, XEXP (src, 1 - accum_pos)))
+ return NULL;
+
+ /* It must be used in exactly one insn. */
+ if (!referenced_in_one_insn_in_loop_p (loop, dest, &debug_uses))
+ return NULL;
+
+ if (dump_file)
+ {
+ fprintf (dump_file, "\n;; Expanding Accumulator ");
+ print_rtl (dump_file, dest);
+ fprintf (dump_file, "\n");
+ }
+
+ if (debug_uses)
+ /* Instead of resetting the debug insns, we could replace each
+ debug use in the loop with the sum or product of all expanded
+ accummulators. Since we'll only know of all expansions at the
+ end, we'd have to keep track of which vars_to_expand a debug
+ insn in the loop references, take note of each copy of the
+ debug insn during unrolling, and when it's all done, compute
+ the sum or product of each variable and adjust the original
+ debug insn and each copy thereof. What a pain! */
+ reset_debug_uses_in_loop (loop, dest, debug_uses);
+
+ /* Record the accumulator to expand. */
+ ves = XNEW (struct var_to_expand);
+ ves->insn = insn;
+ ves->reg = copy_rtx (dest);
+ ves->var_expansions = VEC_alloc (rtx, heap, 1);
+ ves->next = NULL;
+ ves->op = GET_CODE (src);
+ ves->expansion_count = 0;
+ ves->reuse_expansion = 0;
+ ves->accum_pos = accum_pos;
+ return ves;
+}
+
/* Determine whether there is an induction variable in INSN that
- we would like to split during unrolling. Return NULL if INSN
- contains no interesting IVs. Otherwise, allocate an IV_TO_SPLIT
- structure, fill it with the relevant information and return a
+ 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 *
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. */
if (!biv_p (insn, dest))
return NULL;
- if (!iv_analyze (insn, dest, &iv))
- abort ();
+ 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 = xmalloc (sizeof (struct iv_to_split));
+ ivts = XNEW (struct iv_to_split);
ivts->insn = insn;
ivts->base_var = NULL_RTX;
ivts->step = iv.step;
+ ivts->next = NULL;
ivts->n_loc = 1;
ivts->loc[0] = 1;
-
+
return ivts;
}
-/* Determines which of induction variables in LOOP to split.
- Return a SPLIT_IVS_INFO struct with the hash table filled
- with all insns to split IVs in. The FIRST_NEW_BLOCK field
+/* 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 split_ivs_info *
-analyze_ivs_to_split (struct loop *loop)
+static struct opt_info *
+analyze_insns_in_loop (struct loop *loop)
{
basic_block *body, bb;
unsigned i;
- struct split_ivs_info *si_info = xcalloc (1, sizeof (struct split_ivs_info));
+ struct opt_info *opt_info = XCNEW (struct opt_info);
rtx insn;
- struct iv_to_split *ivts;
- PTR *slot;
-
- si_info->insns_to_split = htab_create (5 * loop->num_nodes,
- si_info_hash, si_info_eq, free);
+ struct iv_to_split *ivts = NULL;
+ struct var_to_expand *ves = NULL;
+ PTR *slot1;
+ PTR *slot2;
+ VEC (edge, heap) *edges = get_loop_exit_edges (loop);
+ edge exit;
+ 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);
+ opt_info->iv_to_split_head = NULL;
+ opt_info->iv_to_split_tail = &opt_info->iv_to_split_head;
+ }
+
+ /* Record the loop exit bb and loop preheader before the unrolling. */
+ opt_info->loop_preheader = loop_preheader_edge (loop)->src;
+
+ if (VEC_length (edge, edges) == 1)
+ {
+ exit = VEC_index (edge, edges, 0);
+ if (!(exit->flags & EDGE_COMPLEX))
+ {
+ opt_info->loop_exit = split_edge (exit);
+ 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);
+ opt_info->var_to_expand_head = NULL;
+ opt_info->var_to_expand_tail = &opt_info->var_to_expand_head;
+ }
+
for (i = 0; i < loop->num_nodes; i++)
{
bb = body[i];
continue;
FOR_BB_INSNS (bb, insn)
- {
- if (!INSN_P (insn))
- continue;
-
- ivts = analyze_iv_to_split_insn (insn);
-
- if (!ivts)
- continue;
-
- slot = htab_find_slot (si_info->insns_to_split, ivts, INSERT);
- *slot = ivts;
- }
+ {
+ 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);
+ gcc_assert (*slot1 == NULL);
+ *slot1 = ivts;
+ *opt_info->iv_to_split_tail = ivts;
+ opt_info->iv_to_split_tail = &ivts->next;
+ 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);
+ gcc_assert (*slot2 == NULL);
+ *slot2 = ves;
+ *opt_info->var_to_expand_tail = ves;
+ opt_info->var_to_expand_tail = &ves->next;
+ }
+ }
}
+ VEC_free (edge, heap, edges);
free (body);
-
- return si_info;
+ return opt_info;
}
/* Called just before loop duplication. Records start of duplicated area
- to SI_INFO. */
+ to OPT_INFO. */
-static void
-si_info_start_duplication (struct split_ivs_info *si_info)
+static void
+opt_info_start_duplication (struct opt_info *opt_info)
{
- if (si_info)
- si_info->first_new_block = last_basic_block;
+ if (opt_info)
+ opt_info->first_new_block = last_basic_block;
}
/* Determine the number of iterations between initialization of the base
return ret;
}
-/* Allocate basic variable for the induction variable chain. Callback for
- htab_traverse. */
+/* Allocate basic variable for the induction variable chain. */
-static int
-allocate_basic_variable (void **slot, void *data ATTRIBUTE_UNUSED)
+static void
+allocate_basic_variable (struct iv_to_split *ivts)
{
- 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
seq = get_insns ();
end_sequence ();
emit_insn_before (seq, insn);
-
+
if (validate_change (insn, loc, var, 0))
return;
emit_move_insn (dest, src);
seq = get_insns ();
end_sequence ();
-
+
emit_insn_before (seq, insn);
delete_insn (insn);
}
-/* Splits induction variables (that are marked in SI_INFO) in copies of loop.
- I.e. replace
- i = i + 1;
- ...
- i = i + 1;
- ...
- i = i + 1;
- ...
+/* Return one expansion of the accumulator recorded in struct VE. */
- type chains by
+static rtx
+get_expansion (struct var_to_expand *ve)
+{
+ rtx reg;
- i0 = i + 1
- ...
- i = i0 + 1
- ...
- i = i0 + 2
- ...
+ 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), ve->accum_pos), 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. PLACE is the
+ loop-preheader basic block where the initialization of the
+ expansions should take place. The expansions are initialized with
+ (-0) when the operation is plus or minus to honor sign zero. This
+ way we can prevent cases where the sign of the final result is
+ effected by the sign of the expansion. Here is an example to
+ demonstrate this:
+
+ for (i = 0 ; i < n; i++)
+ sum += something;
+
+ ==>
+
+ sum += something
+ ....
+ i = i+1;
+ sum1 += something
+ ....
+ i = i+1
+ sum2 += something;
+ ....
+
+ When SUM is initialized with -zero and SOMETHING is also -zero; the
+ final result of sum should be -zero thus the expansions sum1 and sum2
+ should be initialized with -zero as well (otherwise we will get +zero
+ as the final result). */
+
+static void
+insert_var_expansion_initialization (struct var_to_expand *ve,
+ basic_block place)
+{
+ rtx seq, var, zero_init, insn;
+ unsigned i;
+ enum machine_mode mode = GET_MODE (ve->reg);
+ bool honor_signed_zero_p = HONOR_SIGNED_ZEROS (mode);
+
+ if (VEC_length (rtx, ve->var_expansions) == 0)
+ return;
+
+ start_sequence ();
+ switch (ve->op)
+ {
+ case FMA:
+ /* Note that we only accumulate FMA via the ADD operand. */
+ case PLUS:
+ case MINUS:
+ FOR_EACH_VEC_ELT (rtx, ve->var_expansions, i, var)
+ {
+ if (honor_signed_zero_p)
+ zero_init = simplify_gen_unary (NEG, mode, CONST0_RTX (mode), mode);
+ else
+ zero_init = CONST0_RTX (mode);
+ emit_move_insn (var, zero_init);
+ }
+ break;
+
+ case MULT:
+ FOR_EACH_VEC_ELT (rtx, ve->var_expansions, i, var)
+ {
+ zero_init = CONST1_RTX (GET_MODE (var));
+ emit_move_insn (var, zero_init);
+ }
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+
+ 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);
+}
+
+/* Combine the variable expansions at the loop exit. PLACE is the
+ loop exit basic block where the summation of the expansions should
+ take place. */
+
+static void
+combine_var_copies_in_loop_exit (struct var_to_expand *ve, basic_block place)
+{
+ rtx sum = ve->reg;
+ rtx expr, seq, var, insn;
+ unsigned i;
+
+ if (VEC_length (rtx, ve->var_expansions) == 0)
+ return;
+
+ start_sequence ();
+ switch (ve->op)
+ {
+ case FMA:
+ /* Note that we only accumulate FMA via the ADD operand. */
+ case PLUS:
+ case MINUS:
+ FOR_EACH_VEC_ELT (rtx, ve->var_expansions, i, var)
+ sum = simplify_gen_binary (PLUS, GET_MODE (ve->reg), var, sum);
+ break;
+
+ case MULT:
+ FOR_EACH_VEC_ELT (rtx, ve->var_expansions, i, var)
+ sum = simplify_gen_binary (MULT, GET_MODE (ve->reg), var, sum);
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+
+ 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);
+}
+
+/* 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
peeling of the loop). */
static void
-split_ivs_in_copies (struct split_ivs_info *si_info, unsigned n_copies,
- bool unrolling, bool rewrite_original_loop)
+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). */
- htab_traverse (si_info->insns_to_split, allocate_basic_variable, NULL);
+ if (opt_info->insns_to_split)
+ for (ivts = opt_info->iv_to_split_head; ivts; ivts = ivts->next)
+ allocate_basic_variable (ivts);
- for (i = si_info->first_new_block; i < (unsigned) last_basic_block; i++)
+ for (i = opt_info->first_new_block; i < (unsigned) last_basic_block; i++)
{
bb = BASIC_BLOCK (i);
- orig_bb = bb->rbi->original;
+ orig_bb = get_bb_original (bb);
- delta = determine_split_iv_delta (bb->rbi->copy_number, n_copies,
+ /* 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;
- ivts = htab_find (si_info->insns_to_split, &ivts_templ);
- if (ivts)
- {
-
-#ifdef ENABLE_CHECKING
- if (!rtx_equal_p (PATTERN (insn), PATTERN (orig_insn)))
- abort ();
-#endif
-
- if (!delta)
- insert_base_initialization (ivts, insn);
- split_iv (ivts, insn, delta);
- }
- orig_insn = NEXT_INSN (orig_insn);
- }
+ {
+ 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 = (struct iv_to_split *)
+ 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 = (struct var_to_expand *)
+ 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)
+ {
+ for (ves = opt_info->var_to_expand_head; ves; ves = ves->next)
+ insert_var_expansion_initialization (ves, opt_info->loop_preheader);
+ for (ves = opt_info->var_to_expand_head; ves; ves = ves->next)
+ combine_var_copies_in_loop_exit (ves, 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
- bb->rbi->original->copy == bb. */
- for (i = si_info->first_new_block; i < (unsigned) last_basic_block; i++)
+ 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 = bb->rbi->original;
- if (orig_bb->rbi->copy != bb)
+ 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;
- ivts = htab_find (si_info->insns_to_split, &ivts_templ);
- if (!ivts)
- continue;
-
- if (!delta)
- insert_base_initialization (ivts, orig_insn);
- split_iv (ivts, orig_insn, delta);
- }
+ 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 = (struct iv_to_split *)
+ 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 SI_INFO. */
+/* Release OPT_INFO. */
static void
-free_si_info (struct split_ivs_info *si_info)
+free_opt_info (struct opt_info *opt_info)
{
- htab_delete (si_info->insns_to_split);
- free (si_info);
+ if (opt_info->insns_to_split)
+ htab_delete (opt_info->insns_to_split);
+ if (opt_info->insns_with_var_to_expand)
+ {
+ struct var_to_expand *ves;
+
+ for (ves = opt_info->var_to_expand_head; ves; ves = ves->next)
+ VEC_free (rtx, heap, ves->var_expansions);
+ htab_delete (opt_info->insns_with_var_to_expand);
+ }
+ free (opt_info);
}
OSDN Git Service
