#include "config.h"
#include "system.h"
#include "coretypes.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"
#include "varray.h"
struct iv_to_split
{
rtx insn;
rtx base_var;
rtx step;
unsigned n_loc;
unsigned loc[3];
};
struct var_to_expand
{
rtx insn;
rtx reg;
varray_type var_expansions;
enum rtx_code op;
int expansion_count;
int reuse_expansion;
};
struct opt_info
{
htab_t insns_to_split;
htab_t insns_with_var_to_expand;
unsigned first_new_block;
basic_block loop_exit;
basic_block loop_preheader;
};
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 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 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 *);
void
unroll_and_peel_loops (struct loops *loops, int flags)
{
struct loop *loop, *next;
bool check;
peel_loops_completely (loops, flags);
decide_unrolling_and_peeling (loops, flags);
loop = loops->tree_root;
while (loop->inner)
loop = loop->inner;
while (loop != loops->tree_root)
{
if (loop->next)
{
next = loop->next;
while (next->inner)
next = next->inner;
}
else
next = loop->outer;
check = true;
switch (loop->lpt_decision.decision)
{
case LPT_PEEL_COMPLETELY:
gcc_unreachable ();
case LPT_PEEL_SIMPLE:
peel_loop_simple (loops, loop);
break;
case LPT_UNROLL_CONSTANT:
unroll_loop_constant_iterations (loops, loop);
break;
case LPT_UNROLL_RUNTIME:
unroll_loop_runtime_iterations (loops, loop);
break;
case LPT_UNROLL_STUPID:
unroll_loop_stupid (loops, loop);
break;
case LPT_NONE:
check = false;
break;
default:
gcc_unreachable ();
}
if (check)
{
#ifdef ENABLE_CHECKING
verify_dominators (CDI_DOMINATORS);
verify_loop_structure (loops);
#endif
}
loop = next;
}
iv_analysis_done ();
}
static bool
loop_exit_at_end_p (struct loop *loop)
{
struct niter_desc *desc = get_simple_loop_desc (loop);
rtx insn;
if (desc->in_edge->dest != loop->latch)
return false;
FOR_BB_INSNS (loop->latch, insn)
{
if (INSN_P (insn))
return false;
}
return true;
}
static void
peel_loops_completely (struct loops *loops, int flags)
{
struct loop *loop, *next;
loop = loops->tree_root;
while (loop->inner)
loop = loop->inner;
while (loop != loops->tree_root)
{
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 for complete peeling ***\n",
loop->num);
loop->ninsns = num_loop_insns (loop);
decide_peel_once_rolling (loop, flags);
if (loop->lpt_decision.decision == LPT_NONE)
decide_peel_completely (loop, flags);
if (loop->lpt_decision.decision == LPT_PEEL_COMPLETELY)
{
peel_loop_completely (loops, loop);
#ifdef ENABLE_CHECKING
verify_dominators (CDI_DOMINATORS);
verify_loop_structure (loops);
#endif
}
loop = next;
}
}
static void
decide_unrolling_and_peeling (struct loops *loops, int flags)
{
struct loop *loop = loops->tree_root, *next;
while (loop->inner)
loop = loop->inner;
while (loop != loops->tree_root)
{
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);
if (!maybe_hot_bb_p (loop->header))
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, cold area\n");
loop = next;
continue;
}
if (!can_duplicate_loop_p (loop))
{
if (dump_file)
fprintf (dump_file,
";; Not considering loop, cannot duplicate\n");
loop = next;
continue;
}
if (loop->inner)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
loop = next;
continue;
}
loop->ninsns = num_loop_insns (loop);
loop->av_ninsns = average_num_loop_insns (loop);
decide_unroll_constant_iterations (loop, flags);
if (loop->lpt_decision.decision == LPT_NONE)
decide_unroll_runtime_iterations (loop, flags);
if (loop->lpt_decision.decision == LPT_NONE)
decide_unroll_stupid (loop, flags);
if (loop->lpt_decision.decision == LPT_NONE)
decide_peel_simple (loop, flags);
loop = next;
}
}
static void
decide_peel_once_rolling (struct loop *loop, int flags ATTRIBUTE_UNUSED)
{
struct niter_desc *desc;
if (dump_file)
fprintf (dump_file, "\n;; Considering peeling once rolling loop\n");
if ((unsigned) PARAM_VALUE (PARAM_MAX_ONCE_PEELED_INSNS) < loop->ninsns)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (!desc->simple_p
|| desc->assumptions
|| !desc->const_iter
|| desc->niter != 0)
{
if (dump_file)
fprintf (dump_file,
";; Unable to prove that the loop rolls exactly once\n");
return;
}
if (dump_file)
fprintf (dump_file, ";; Decided to peel exactly once rolling loop\n");
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
}
static void
decide_peel_completely (struct loop *loop, int flags ATTRIBUTE_UNUSED)
{
unsigned npeel;
struct niter_desc *desc;
if (dump_file)
fprintf (dump_file, "\n;; Considering peeling completely\n");
if (loop->inner)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is not innermost\n");
return;
}
if (!maybe_hot_bb_p (loop->header))
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, cold area\n");
return;
}
if (!can_duplicate_loop_p (loop))
{
if (dump_file)
fprintf (dump_file,
";; Not considering loop, cannot duplicate\n");
return;
}
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS) / loop->ninsns;
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES))
npeel = PARAM_VALUE (PARAM_MAX_COMPLETELY_PEEL_TIMES);
if (!npeel)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (!desc->simple_p
|| desc->assumptions
|| !desc->const_iter)
{
if (dump_file)
fprintf (dump_file,
";; Unable to prove that the loop iterates constant times\n");
return;
}
if (desc->niter > npeel - 1)
{
if (dump_file)
{
fprintf (dump_file,
";; Not peeling loop completely, rolls too much (");
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC, desc->niter);
fprintf (dump_file, " iterations > %d [maximum peelings])\n", npeel);
}
return;
}
if (dump_file)
fprintf (dump_file, ";; Decided to peel loop completely\n");
loop->lpt_decision.decision = LPT_PEEL_COMPLETELY;
}
static void
peel_loop_completely (struct loops *loops, struct loop *loop)
{
sbitmap wont_exit;
unsigned HOST_WIDE_INT npeel;
unsigned n_remove_edges, i;
edge *remove_edges, ein;
struct niter_desc *desc = get_simple_loop_desc (loop);
struct opt_info *opt_info = NULL;
npeel = desc->niter;
if (npeel)
{
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;
if (flag_split_ivs_in_unroller)
opt_info = analyze_insns_in_loop (loop);
opt_info_start_duplication (opt_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 ();
free (wont_exit);
if (opt_info)
{
apply_opt_in_copies (opt_info, npeel, false, true);
free_opt_info (opt_info);
}
for (i = 0; i < n_remove_edges; i++)
remove_path (loops, remove_edges[i]);
free (remove_edges);
}
ein = desc->in_edge;
free_simple_loop_desc (loop);
remove_path (loops, ein);
if (dump_file)
fprintf (dump_file, ";; Peeled loop completely, %d times\n", (int) npeel);
}
static void
decide_unroll_constant_iterations (struct loop *loop, int flags)
{
unsigned nunroll, nunroll_by_av, best_copies, best_unroll = 0, n_copies, i;
struct niter_desc *desc;
if (!(flags & UAP_UNROLL))
{
return;
}
if (dump_file)
fprintf (dump_file,
"\n;; Considering unrolling loop with constant "
"number of iterations\n");
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
nunroll_by_av
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
if (nunroll > nunroll_by_av)
nunroll = nunroll_by_av;
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
if (nunroll <= 1)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (!desc->simple_p || !desc->const_iter || desc->assumptions)
{
if (dump_file)
fprintf (dump_file,
";; Unable to prove that the loop iterates constant times\n");
return;
}
if (desc->niter < 2 * nunroll)
{
if (dump_file)
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
return;
}
best_copies = 2 * nunroll + 10;
i = 2 * nunroll + 2;
if (i - 1 >= desc->niter)
i = desc->niter - 2;
for (; i >= nunroll - 1; i--)
{
unsigned exit_mod = desc->niter % (i + 1);
if (!loop_exit_at_end_p (loop))
n_copies = exit_mod + i + 1;
else if (exit_mod != (unsigned) i
|| desc->noloop_assumptions != NULL_RTX)
n_copies = exit_mod + i + 2;
else
n_copies = i + 1;
if (n_copies < best_copies)
{
best_copies = n_copies;
best_unroll = i;
}
}
if (dump_file)
fprintf (dump_file, ";; max_unroll %d (%d copies, initial %d).\n",
best_unroll + 1, best_copies, nunroll);
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",
loop->lpt_decision.times);
}
static void
unroll_loop_constant_iterations (struct loops *loops, struct loop *loop)
{
unsigned HOST_WIDE_INT niter;
unsigned exit_mod;
sbitmap wont_exit;
unsigned n_remove_edges, i;
edge *remove_edges;
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 opt_info *opt_info = NULL;
niter = desc->niter;
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));
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)
{
if (dump_file)
fprintf (dump_file, ";; Condition on beginning of loop.\n");
RESET_BIT (wont_exit, 0);
if (desc->noloop_assumptions)
RESET_BIT (wont_exit, 1);
if (exit_mod)
{
opt_info_start_duplication (opt_info);
if (!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))
abort ();
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;
}
SET_BIT (wont_exit, 1);
}
else
{
if (dump_file)
fprintf (dump_file, ";; Condition on end of loop.\n");
if (exit_mod != max_unroll
|| desc->noloop_assumptions)
{
RESET_BIT (wont_exit, 0);
if (desc->noloop_assumptions)
RESET_BIT (wont_exit, 1);
opt_info_start_duplication (opt_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 ();
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;
desc->noloop_assumptions = NULL_RTX;
SET_BIT (wont_exit, 0);
SET_BIT (wont_exit, 1);
}
RESET_BIT (wont_exit, max_unroll);
}
opt_info_start_duplication (opt_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 (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;
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
{
desc->out_edge = EDGE_SUCC (exit_block, 0);
desc->in_edge = EDGE_SUCC (exit_block, 1);
}
else
{
desc->out_edge = EDGE_SUCC (exit_block, 1);
desc->in_edge = EDGE_SUCC (exit_block, 0);
}
}
desc->niter /= max_unroll + 1;
desc->niter_max /= max_unroll + 1;
desc->niter_expr = GEN_INT (desc->niter);
for (i = 0; i < n_remove_edges; i++)
remove_path (loops, remove_edges[i]);
free (remove_edges);
if (dump_file)
fprintf (dump_file,
";; Unrolled loop %d times, constant # of iterations %i insns\n",
max_unroll, num_loop_insns (loop));
}
static void
decide_unroll_runtime_iterations (struct loop *loop, int flags)
{
unsigned nunroll, nunroll_by_av, i;
struct niter_desc *desc;
if (!(flags & UAP_UNROLL))
{
return;
}
if (dump_file)
fprintf (dump_file,
"\n;; Considering unrolling loop with runtime "
"computable number of iterations\n");
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
nunroll_by_av = PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
if (nunroll > nunroll_by_av)
nunroll = nunroll_by_av;
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
if (nunroll <= 1)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (!desc->simple_p || desc->assumptions)
{
if (dump_file)
fprintf (dump_file,
";; Unable to prove that the number of iterations "
"can be counted in runtime\n");
return;
}
if (desc->const_iter)
{
if (dump_file)
fprintf (dump_file, ";; Loop iterates constant times\n");
return;
}
if (loop->header->count && expected_loop_iterations (loop) < 2 * nunroll)
{
if (dump_file)
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
return;
}
for (i = 1; 2 * i <= nunroll; i *= 2)
continue;
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 "
"times rolling loop, %d times.\n",
loop->lpt_decision.times);
}
static void
unroll_loop_runtime_iterations (struct loops *loops, 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;
sbitmap wont_exit;
int may_exit_copy;
unsigned n_peel, n_remove_edges;
edge *remove_edges, 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 opt_info *opt_info = NULL;
if (flag_split_ivs_in_unroller
|| flag_variable_expansion_in_unroller)
opt_info = analyze_insns_in_loop (loop);
dom_bbs = xcalloc (n_basic_blocks, sizeof (basic_block));
n_dom_bbs = 0;
body = get_loop_body (loop);
for (i = 0; i < loop->num_nodes; i++)
{
unsigned nldom;
basic_block *ldom;
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];
free (ldom);
}
free (body);
if (!exit_at_end)
{
may_exit_copy = 0;
n_peel = max_unroll - 1;
extra_zero_check = true;
last_may_exit = false;
}
else
{
may_exit_copy = max_unroll;
n_peel = max_unroll;
extra_zero_check = false;
last_may_exit = true;
}
start_sequence ();
old_niter = niter = gen_reg_rtx (desc->mode);
tmp = force_operand (copy_rtx (desc->niter_expr), niter);
if (tmp != niter)
emit_move_insn (niter, tmp);
niter = expand_simple_binop (desc->mode, AND,
niter,
GEN_INT (max_unroll),
NULL_RTX, 0, OPTAB_LIB_WIDEN);
init_code = get_insns ();
end_sequence ();
loop_split_edge_with (loop_preheader_edge (loop), init_code);
remove_edges = xcalloc (max_unroll + n_peel + 1, sizeof (edge));
n_remove_edges = 0;
wont_exit = sbitmap_alloc (max_unroll + 2);
sbitmap_zero (wont_exit);
if (extra_zero_check
&& !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 ();
swtch = loop_split_edge_with (loop_preheader_edge (loop),
NULL_RTX);
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 ();
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);
branch_code = compare_and_jump_seq (copy_rtx (niter), GEN_INT (j), EQ,
block_label (preheader), p, NULL_RTX);
swtch = loop_split_edge_with (EDGE_PRED (swtch, 0), branch_code);
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
EDGE_SUCC (swtch, 0)->probability = REG_BR_PROB_BASE - p;
e = make_edge (swtch, preheader,
EDGE_SUCC (swtch, 0)->flags & EDGE_IRREDUCIBLE_LOOP);
e->probability = p;
}
if (extra_zero_check)
{
p = REG_BR_PROB_BASE / (max_unroll + 1);
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);
swtch = loop_split_edge_with (EDGE_SUCC (swtch, 0), branch_code);
set_immediate_dominator (CDI_DOMINATORS, preheader, swtch);
EDGE_SUCC (swtch, 0)->probability = REG_BR_PROB_BASE - p;
e = make_edge (swtch, preheader,
EDGE_SUCC (swtch, 0)->flags & EDGE_IRREDUCIBLE_LOOP);
e->probability = p;
}
iterate_fix_dominators (CDI_DOMINATORS, dom_bbs, n_dom_bbs);
sbitmap_ones (wont_exit);
RESET_BIT (wont_exit, may_exit_copy);
opt_info_start_duplication (opt_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 (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;
if (EDGE_SUCC (exit_block, 0)->dest == desc->out_edge->dest)
{
desc->out_edge = EDGE_SUCC (exit_block, 0);
desc->in_edge = EDGE_SUCC (exit_block, 1);
}
else
{
desc->out_edge = EDGE_SUCC (exit_block, 1);
desc->in_edge = EDGE_SUCC (exit_block, 0);
}
}
for (i = 0; i < n_remove_edges; i++)
remove_path (loops, remove_edges[i]);
free (remove_edges);
gcc_assert (!desc->const_iter);
desc->niter_expr =
simplify_gen_binary (UDIV, desc->mode, old_niter, GEN_INT (max_unroll + 1));
desc->niter_max /= max_unroll + 1;
if (exit_at_end)
{
desc->niter_expr =
simplify_gen_binary (MINUS, desc->mode, desc->niter_expr, const1_rtx);
desc->noloop_assumptions = NULL_RTX;
desc->niter_max--;
}
if (dump_file)
fprintf (dump_file,
";; Unrolled loop %d times, counting # of iterations "
"in runtime, %i insns\n",
max_unroll, num_loop_insns (loop));
}
static void
decide_peel_simple (struct loop *loop, int flags)
{
unsigned npeel;
struct niter_desc *desc;
if (!(flags & UAP_PEEL))
{
return;
}
if (dump_file)
fprintf (dump_file, "\n;; Considering simply peeling loop\n");
npeel = PARAM_VALUE (PARAM_MAX_PEELED_INSNS) / loop->ninsns;
if (npeel > (unsigned) PARAM_VALUE (PARAM_MAX_PEEL_TIMES))
npeel = PARAM_VALUE (PARAM_MAX_PEEL_TIMES);
if (!npeel)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (desc->simple_p && !desc->assumptions && desc->const_iter)
{
if (dump_file)
fprintf (dump_file, ";; Loop iterates constant times\n");
return;
}
if (num_loop_branches (loop) > 1)
{
if (dump_file)
fprintf (dump_file, ";; Not peeling, contains branches\n");
return;
}
if (loop->header->count)
{
unsigned niter = expected_loop_iterations (loop);
if (niter + 1 > npeel)
{
if (dump_file)
{
fprintf (dump_file, ";; Not peeling loop, rolls too much (");
fprintf (dump_file, HOST_WIDEST_INT_PRINT_DEC,
(HOST_WIDEST_INT) (niter + 1));
fprintf (dump_file, " iterations > %d [maximum peelings])\n",
npeel);
}
return;
}
npeel = niter + 1;
}
else
{
if (dump_file)
fprintf (dump_file,
";; Not peeling loop, no evidence it will be profitable\n");
return;
}
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);
}
static void
peel_loop_simple (struct loops *loops, struct loop *loop)
{
sbitmap wont_exit;
unsigned npeel = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
struct opt_info *opt_info = NULL;
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);
if (!duplicate_loop_to_header_edge (loop, loop_preheader_edge (loop),
loops, npeel, wont_exit, NULL, NULL, NULL,
DLTHE_FLAG_UPDATE_FREQ))
abort ();
free (wont_exit);
if (opt_info)
{
apply_opt_in_copies (opt_info, npeel, false, false);
free_opt_info (opt_info);
}
if (desc->simple_p)
{
if (desc->const_iter)
{
desc->niter -= npeel;
desc->niter_expr = GEN_INT (desc->niter);
desc->noloop_assumptions = NULL_RTX;
}
else
{
free_simple_loop_desc (loop);
}
}
if (dump_file)
fprintf (dump_file, ";; Peeling loop %d times\n", npeel);
}
static void
decide_unroll_stupid (struct loop *loop, int flags)
{
unsigned nunroll, nunroll_by_av, i;
struct niter_desc *desc;
if (!(flags & UAP_UNROLL_ALL))
{
return;
}
if (dump_file)
fprintf (dump_file, "\n;; Considering unrolling loop stupidly\n");
nunroll = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / loop->ninsns;
nunroll_by_av
= PARAM_VALUE (PARAM_MAX_AVERAGE_UNROLLED_INSNS) / loop->av_ninsns;
if (nunroll > nunroll_by_av)
nunroll = nunroll_by_av;
if (nunroll > (unsigned) PARAM_VALUE (PARAM_MAX_UNROLL_TIMES))
nunroll = PARAM_VALUE (PARAM_MAX_UNROLL_TIMES);
if (nunroll <= 1)
{
if (dump_file)
fprintf (dump_file, ";; Not considering loop, is too big\n");
return;
}
desc = get_simple_loop_desc (loop);
if (desc->simple_p && !desc->assumptions)
{
if (dump_file)
fprintf (dump_file, ";; The loop is simple\n");
return;
}
if (num_loop_branches (loop) > 1)
{
if (dump_file)
fprintf (dump_file, ";; Not unrolling, contains branches\n");
return;
}
if (loop->header->count
&& expected_loop_iterations (loop) < 2 * nunroll)
{
if (dump_file)
fprintf (dump_file, ";; Not unrolling loop, doesn't roll\n");
return;
}
for (i = 1; 2 * i <= nunroll; i *= 2)
continue;
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",
loop->lpt_decision.times);
}
static void
unroll_loop_stupid (struct loops *loops, struct loop *loop)
{
sbitmap wont_exit;
unsigned nunroll = loop->lpt_decision.times;
struct niter_desc *desc = get_simple_loop_desc (loop);
struct opt_info *opt_info = NULL;
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);
if (!duplicate_loop_to_header_edge (loop, loop_latch_edge (loop),
loops, nunroll, wont_exit, NULL, NULL, NULL,
DLTHE_FLAG_UPDATE_FREQ))
abort ();
if (opt_info)
{
apply_opt_in_copies (opt_info, nunroll, true, true);
free_opt_info (opt_info);
}
free (wont_exit);
if (desc->simple_p)
{
desc->simple_p = false;
}
if (dump_file)
fprintf (dump_file, ";; Unrolled loop %d times, %i insns\n",
nunroll, num_loop_insns (loop));
}
static hashval_t
si_info_hash (const void *ivts)
{
return htab_hash_pointer (((struct iv_to_split *) ivts)->insn);
}
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;
}
static hashval_t
ve_info_hash (const void *ves)
{
return htab_hash_pointer (((struct var_to_expand *) ves)->insn);
}
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;
}
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);
}
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;
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;
ves = xmalloc (sizeof (struct var_to_expand));
ves->insn = insn;
VARRAY_RTX_INIT (ves->var_expansions, 1, "var_expansions");
ves->reg = copy_rtx (dest);
ves->op = GET_CODE (src);
ves->expansion_count = 0;
ves->reuse_expansion = 0;
return ves;
}
static struct iv_to_split *
analyze_iv_to_split_insn (rtx insn)
{
rtx set, dest;
struct rtx_iv iv;
struct iv_to_split *ivts;
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;
if (!iv_analyze (insn, dest, &iv))
abort ();
if (iv.step == const0_rtx
|| iv.mode != iv.extend_mode)
return NULL;
ivts = xmalloc (sizeof (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;
}
static struct opt_info *
analyze_insns_in_loop (struct loop *loop)
{
basic_block *body, bb;
unsigned i, n_edges = 0;
struct opt_info *opt_info = xcalloc (1, sizeof (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, &n_edges);
basic_block preheader;
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);
if (!loop_preheader_edge (loop)->src)
{
preheader = 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 (n_edges == 1
&& !(edges[0]->flags & EDGE_COMPLEX)
&& (edges[0]->flags & EDGE_LOOP_EXIT))
{
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;
}
static void
opt_info_start_duplication (struct opt_info *opt_info)
{
if (opt_info)
opt_info->first_new_block = last_basic_block;
}
static unsigned
determine_split_iv_delta (unsigned n_copy, unsigned n_copies, bool unrolling)
{
if (unrolling)
{
return n_copy;
}
else
{
if (n_copy)
return n_copy - 1;
else
return n_copies;
}
}
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;
}
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;
}
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);
}
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;
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);
}
loc = get_ivts_expr (single_set (insn), ivts);
if (validate_change (insn, loc, expr, 0))
return;
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;
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);
}
static rtx
get_expansion (struct var_to_expand *ve)
{
rtx reg;
if (ve->reuse_expansion == 0)
reg = ve->reg;
else
reg = VARRAY_RTX (ve->var_expansions, ve->reuse_expansion - 1);
if (VARRAY_ACTIVE_SIZE (ve->var_expansions) == (unsigned) ve->reuse_expansion)
ve->reuse_expansion = 0;
else
ve->reuse_expansion++;
return reg;
}
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);
if (!set)
abort ();
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)
{
VARRAY_PUSH_RTX (ve->var_expansions, new_reg);
ve->expansion_count++;
}
}
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 (VARRAY_ACTIVE_SIZE (ve->var_expansions) == 0)
return 1;
start_sequence ();
if (ve->op == PLUS || ve->op == MINUS)
for (i = 0; i < VARRAY_ACTIVE_SIZE (ve->var_expansions); i++)
{
var = VARRAY_RTX (ve->var_expansions, i);
zero_init = CONST0_RTX (GET_MODE (var));
emit_move_insn (var, zero_init);
}
else if (ve->op == MULT)
for (i = 0; i < VARRAY_ACTIVE_SIZE (ve->var_expansions); i++)
{
var = VARRAY_RTX (ve->var_expansions, 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);
return 1;
}
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 (VARRAY_ACTIVE_SIZE (ve->var_expansions) == 0)
return 1;
start_sequence ();
if (ve->op == PLUS || ve->op == MINUS)
for (i = 0; i < VARRAY_ACTIVE_SIZE (ve->var_expansions); i++)
{
var = VARRAY_RTX (ve->var_expansions, i);
sum = simplify_gen_binary (PLUS, GET_MODE (ve->reg),
var, sum);
}
else if (ve->op == MULT)
for (i = 0; i < VARRAY_ACTIVE_SIZE (ve->var_expansions); i++)
{
var = VARRAY_RTX (ve->var_expansions, 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);
return 1;
}
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;
gcc_assert (!unrolling || rewrite_original_loop);
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 = bb->rbi->original;
delta = determine_split_iv_delta (bb->rbi->copy_number, n_copies,
unrolling);
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;
if (opt_info->insns_to_split)
{
ivts = htab_find (opt_info->insns_to_split, &ivts_templ);
if (ivts)
{
#ifdef ENABLE_CHECKING
gcc_assert (rtx_equal_p (PATTERN (insn), PATTERN (orig_insn)));
#endif
if (!delta)
insert_base_initialization (ivts, insn);
split_iv (ivts, insn, delta);
}
}
if (unrolling && opt_info->insns_with_var_to_expand)
{
ves = htab_find (opt_info->insns_with_var_to_expand, &ve_templ);
if (ves)
{
#ifdef ENABLE_CHECKING
gcc_assert (rtx_equal_p (PATTERN (insn), PATTERN (orig_insn)));
#endif
expand_var_during_unrolling (ves, insn);
}
}
orig_insn = NEXT_INSN (orig_insn);
}
}
if (!rewrite_original_loop)
return;
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);
}
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)
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;
}
}
}
}
}
static int
release_var_copies (void **slot, void *data ATTRIBUTE_UNUSED)
{
struct var_to_expand *ve = *slot;
VARRAY_CLEAR (ve->var_expansions);
return 1;
}
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);
}