gcc/gcc/tree-ssa-loop-prefetch.c

1061 lines
30 KiB
C
Raw Normal View History

2006-02-14 14:51:51 +01:00
/* Array prefetching.
Copyright (C) 2005 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 version.
GCC is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
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. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "tree.h"
#include "rtl.h"
#include "tm_p.h"
#include "hard-reg-set.h"
#include "basic-block.h"
#include "output.h"
#include "diagnostic.h"
#include "tree-flow.h"
#include "tree-dump.h"
#include "timevar.h"
#include "cfgloop.h"
#include "varray.h"
#include "expr.h"
#include "tree-pass.h"
#include "ggc.h"
#include "insn-config.h"
#include "recog.h"
#include "hashtab.h"
#include "tree-chrec.h"
#include "tree-scalar-evolution.h"
#include "toplev.h"
#include "params.h"
#include "langhooks.h"
/* This pass inserts prefetch instructions to optimize cache usage during
accesses to arrays in loops. It processes loops sequentially and:
1) Gathers all memory references in the single loop.
2) For each of the references it decides when it is profitable to prefetch
it. To do it, we evaluate the reuse among the accesses, and determines
two values: PREFETCH_BEFORE (meaning that it only makes sense to do
prefetching in the first PREFETCH_BEFORE iterations of the loop) and
PREFETCH_MOD (meaning that it only makes sense to prefetch in the
iterations of the loop that are zero modulo PREFETCH_MOD). For example
(assuming cache line size is 64 bytes, char has size 1 byte and there
is no hardware sequential prefetch):
char *a;
for (i = 0; i < max; i++)
{
a[255] = ...; (0)
a[i] = ...; (1)
a[i + 64] = ...; (2)
a[16*i] = ...; (3)
a[187*i] = ...; (4)
a[187*i + 50] = ...; (5)
}
(0) obviously has PREFETCH_BEFORE 1
(1) has PREFETCH_BEFORE 64, since (2) accesses the same memory
location 64 iterations before it, and PREFETCH_MOD 64 (since
it hits the same cache line otherwise).
(2) has PREFETCH_MOD 64
(3) has PREFETCH_MOD 4
(4) has PREFETCH_MOD 1. We do not set PREFETCH_BEFORE here, since
the cache line accessed by (4) is the same with probability only
7/32.
(5) has PREFETCH_MOD 1 as well.
3) We determine how much ahead we need to prefetch. The number of
iterations needed is time to fetch / time spent in one iteration of
the loop. The problem is that we do not know either of these values,
so we just make a heuristic guess based on a magic (possibly)
target-specific constant and size of the loop.
4) Determine which of the references we prefetch. We take into account
that there is a maximum number of simultaneous prefetches (provided
by machine description). We prefetch as many prefetches as possible
while still within this bound (starting with those with lowest
prefetch_mod, since they are responsible for most of the cache
misses).
5) We unroll and peel loops so that we are able to satisfy PREFETCH_MOD
and PREFETCH_BEFORE requirements (within some bounds), and to avoid
prefetching nonaccessed memory.
TODO -- actually implement peeling.
6) We actually emit the prefetch instructions. ??? Perhaps emit the
prefetch instructions with guards in cases where 5) was not sufficient
to satisfy the constraints?
Some other TODO:
-- write and use more general reuse analysis (that could be also used
in other cache aimed loop optimizations)
-- make it behave sanely together with the prefetches given by user
(now we just ignore them; at the very least we should avoid
optimizing loops in that user put his own prefetches)
-- we assume cache line size alignment of arrays; this could be
improved. */
/* Magic constants follow. These should be replaced by machine specific
numbers. */
/* True if write can be prefetched by a read prefetch. */
#ifndef WRITE_CAN_USE_READ_PREFETCH
#define WRITE_CAN_USE_READ_PREFETCH 1
#endif
/* True if read can be prefetched by a write prefetch. */
#ifndef READ_CAN_USE_WRITE_PREFETCH
#define READ_CAN_USE_WRITE_PREFETCH 0
#endif
2006-11-12 20:17:02 +01:00
/* The size of the block loaded by a single prefetch. Usually, this is
the same as cache line size (at the moment, we only consider one level
of cache hierarchy). */
2006-02-14 14:51:51 +01:00
#ifndef PREFETCH_BLOCK
2006-11-12 20:17:02 +01:00
#define PREFETCH_BLOCK L1_CACHE_LINE_SIZE
2006-02-14 14:51:51 +01:00
#endif
/* Do we have a forward hardware sequential prefetching? */
#ifndef HAVE_FORWARD_PREFETCH
#define HAVE_FORWARD_PREFETCH 0
#endif
/* Do we have a backward hardware sequential prefetching? */
#ifndef HAVE_BACKWARD_PREFETCH
#define HAVE_BACKWARD_PREFETCH 0
#endif
/* In some cases we are only able to determine that there is a certain
probability that the two accesses hit the same cache line. In this
case, we issue the prefetches for both of them if this probability
is less then (1000 - ACCEPTABLE_MISS_RATE) promile. */
#ifndef ACCEPTABLE_MISS_RATE
#define ACCEPTABLE_MISS_RATE 50
#endif
#ifndef HAVE_prefetch
#define HAVE_prefetch 0
#endif
/* The group of references between that reuse may occur. */
struct mem_ref_group
{
tree base; /* Base of the reference. */
HOST_WIDE_INT step; /* Step of the reference. */
struct mem_ref *refs; /* References in the group. */
struct mem_ref_group *next; /* Next group of references. */
};
/* Assigned to PREFETCH_BEFORE when all iterations are to be prefetched. */
#define PREFETCH_ALL (~(unsigned HOST_WIDE_INT) 0)
/* The memory reference. */
struct mem_ref
{
tree stmt; /* Statement in that the reference appears. */
tree mem; /* The reference. */
HOST_WIDE_INT delta; /* Constant offset of the reference. */
bool write_p; /* Is it a write? */
struct mem_ref_group *group; /* The group of references it belongs to. */
unsigned HOST_WIDE_INT prefetch_mod;
/* Prefetch only each PREFETCH_MOD-th
iteration. */
unsigned HOST_WIDE_INT prefetch_before;
/* Prefetch only first PREFETCH_BEFORE
iterations. */
bool issue_prefetch_p; /* Should we really issue the prefetch? */
struct mem_ref *next; /* The next reference in the group. */
};
/* Dumps information about reference REF to FILE. */
2006-02-14 14:51:51 +01:00
static void
dump_mem_ref (FILE *file, struct mem_ref *ref)
{
fprintf (file, "Reference %p:\n", (void *) ref);
fprintf (file, " group %p (base ", (void *) ref->group);
print_generic_expr (file, ref->group->base, TDF_SLIM);
fprintf (file, ", step ");
fprintf (file, HOST_WIDE_INT_PRINT_DEC, ref->group->step);
fprintf (file, ")\n");
fprintf (file, " delta ");
2006-02-14 14:51:51 +01:00
fprintf (file, HOST_WIDE_INT_PRINT_DEC, ref->delta);
fprintf (file, "\n");
fprintf (file, " %s\n", ref->write_p ? "write" : "read");
fprintf (file, "\n");
}
/* Finds a group with BASE and STEP in GROUPS, or creates one if it does not
exist. */
static struct mem_ref_group *
find_or_create_group (struct mem_ref_group **groups, tree base,
HOST_WIDE_INT step)
{
struct mem_ref_group *group;
for (; *groups; groups = &(*groups)->next)
{
if ((*groups)->step == step
&& operand_equal_p ((*groups)->base, base, 0))
return *groups;
/* Keep the list of groups sorted by decreasing step. */
if ((*groups)->step < step)
break;
}
group = xcalloc (1, sizeof (struct mem_ref_group));
group->base = base;
group->step = step;
group->refs = NULL;
group->next = *groups;
*groups = group;
return group;
}
/* Records a memory reference MEM in GROUP with offset DELTA and write status
WRITE_P. The reference occurs in statement STMT. */
static void
record_ref (struct mem_ref_group *group, tree stmt, tree mem,
HOST_WIDE_INT delta, bool write_p)
{
struct mem_ref **aref;
/* Do not record the same address twice. */
for (aref = &group->refs; *aref; aref = &(*aref)->next)
{
/* It does not have to be possible for write reference to reuse the read
prefetch, or vice versa. */
if (!WRITE_CAN_USE_READ_PREFETCH
&& write_p
&& !(*aref)->write_p)
continue;
if (!READ_CAN_USE_WRITE_PREFETCH
&& !write_p
&& (*aref)->write_p)
continue;
if ((*aref)->delta == delta)
return;
}
(*aref) = xcalloc (1, sizeof (struct mem_ref));
(*aref)->stmt = stmt;
(*aref)->mem = mem;
(*aref)->delta = delta;
(*aref)->write_p = write_p;
(*aref)->prefetch_before = PREFETCH_ALL;
(*aref)->prefetch_mod = 1;
(*aref)->issue_prefetch_p = false;
(*aref)->group = group;
(*aref)->next = NULL;
if (dump_file && (dump_flags & TDF_DETAILS))
dump_mem_ref (dump_file, *aref);
}
/* Release memory references in GROUPS. */
static void
release_mem_refs (struct mem_ref_group *groups)
{
struct mem_ref_group *next_g;
struct mem_ref *ref, *next_r;
for (; groups; groups = next_g)
{
next_g = groups->next;
for (ref = groups->refs; ref; ref = next_r)
{
next_r = ref->next;
free (ref);
}
free (groups);
}
}
/* A structure used to pass arguments to idx_analyze_ref. */
struct ar_data
{
struct loop *loop; /* Loop of the reference. */
tree stmt; /* Statement of the reference. */
HOST_WIDE_INT *step; /* Step of the memory reference. */
HOST_WIDE_INT *delta; /* Offset of the memory reference. */
};
/* Analyzes a single INDEX of a memory reference to obtain information
described at analyze_ref. Callback for for_each_index. */
static bool
idx_analyze_ref (tree base, tree *index, void *data)
{
struct ar_data *ar_data = data;
tree ibase, step, stepsize;
HOST_WIDE_INT istep, idelta = 0, imult = 1;
affine_iv iv;
if (TREE_CODE (base) == MISALIGNED_INDIRECT_REF
|| TREE_CODE (base) == ALIGN_INDIRECT_REF)
return false;
if (!simple_iv (ar_data->loop, ar_data->stmt, *index, &iv, false))
return false;
ibase = iv.base;
step = iv.step;
2006-12-24 00:05:52 +01:00
if (!cst_and_fits_in_hwi (step))
return false;
istep = int_cst_value (step);
2006-02-14 14:51:51 +01:00
if (TREE_CODE (ibase) == PLUS_EXPR
&& cst_and_fits_in_hwi (TREE_OPERAND (ibase, 1)))
{
idelta = int_cst_value (TREE_OPERAND (ibase, 1));
ibase = TREE_OPERAND (ibase, 0);
}
if (cst_and_fits_in_hwi (ibase))
{
idelta += int_cst_value (ibase);
ibase = build_int_cst (TREE_TYPE (ibase), 0);
2006-02-14 14:51:51 +01:00
}
if (TREE_CODE (base) == ARRAY_REF)
{
stepsize = array_ref_element_size (base);
if (!cst_and_fits_in_hwi (stepsize))
return false;
imult = int_cst_value (stepsize);
istep *= imult;
idelta *= imult;
}
*ar_data->step += istep;
*ar_data->delta += idelta;
*index = ibase;
return true;
}
/* Tries to express REF_P in shape &BASE + STEP * iter + DELTA, where DELTA and
2006-02-14 14:51:51 +01:00
STEP are integer constants and iter is number of iterations of LOOP. The
reference occurs in statement STMT. Strips nonaddressable component
references from REF_P. */
2006-02-14 14:51:51 +01:00
static bool
analyze_ref (struct loop *loop, tree *ref_p, tree *base,
2006-02-14 14:51:51 +01:00
HOST_WIDE_INT *step, HOST_WIDE_INT *delta,
tree stmt)
{
struct ar_data ar_data;
tree off;
HOST_WIDE_INT bit_offset;
tree ref = *ref_p;
2006-02-14 14:51:51 +01:00
*step = 0;
*delta = 0;
/* First strip off the component references. Ignore bitfields. */
if (TREE_CODE (ref) == COMPONENT_REF
&& DECL_NONADDRESSABLE_P (TREE_OPERAND (ref, 1)))
ref = TREE_OPERAND (ref, 0);
*ref_p = ref;
2006-02-14 14:51:51 +01:00
for (; TREE_CODE (ref) == COMPONENT_REF; ref = TREE_OPERAND (ref, 0))
{
off = DECL_FIELD_BIT_OFFSET (TREE_OPERAND (ref, 1));
bit_offset = TREE_INT_CST_LOW (off);
gcc_assert (bit_offset % BITS_PER_UNIT == 0);
*delta += bit_offset / BITS_PER_UNIT;
}
*base = unshare_expr (ref);
ar_data.loop = loop;
ar_data.stmt = stmt;
ar_data.step = step;
ar_data.delta = delta;
return for_each_index (base, idx_analyze_ref, &ar_data);
}
/* Record a memory reference REF to the list REFS. The reference occurs in
LOOP in statement STMT and it is write if WRITE_P. */
static void
gather_memory_references_ref (struct loop *loop, struct mem_ref_group **refs,
tree ref, bool write_p, tree stmt)
{
tree base;
HOST_WIDE_INT step, delta;
struct mem_ref_group *agrp;
if (!analyze_ref (loop, &ref, &base, &step, &delta, stmt))
2006-02-14 14:51:51 +01:00
return;
/* Now we know that REF = &BASE + STEP * iter + DELTA, where DELTA and STEP
are integer constants. */
agrp = find_or_create_group (refs, base, step);
record_ref (agrp, stmt, ref, delta, write_p);
}
/* Record the suitable memory references in LOOP. */
static struct mem_ref_group *
gather_memory_references (struct loop *loop)
{
basic_block *body = get_loop_body_in_dom_order (loop);
basic_block bb;
unsigned i;
block_stmt_iterator bsi;
tree stmt, lhs, rhs;
struct mem_ref_group *refs = NULL;
/* Scan the loop body in order, so that the former references precede the
later ones. */
for (i = 0; i < loop->num_nodes; i++)
{
bb = body[i];
if (bb->loop_father != loop)
continue;
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
{
stmt = bsi_stmt (bsi);
if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
2006-02-14 14:51:51 +01:00
continue;
lhs = GIMPLE_STMT_OPERAND (stmt, 0);
rhs = GIMPLE_STMT_OPERAND (stmt, 1);
2006-02-14 14:51:51 +01:00
if (REFERENCE_CLASS_P (rhs))
gather_memory_references_ref (loop, &refs, rhs, false, stmt);
if (REFERENCE_CLASS_P (lhs))
gather_memory_references_ref (loop, &refs, lhs, true, stmt);
}
}
free (body);
return refs;
}
/* Prune the prefetch candidate REF using the self-reuse. */
static void
prune_ref_by_self_reuse (struct mem_ref *ref)
{
HOST_WIDE_INT step = ref->group->step;
bool backward = step < 0;
if (step == 0)
{
/* Prefetch references to invariant address just once. */
ref->prefetch_before = 1;
return;
}
if (backward)
step = -step;
if (step > PREFETCH_BLOCK)
return;
if ((backward && HAVE_BACKWARD_PREFETCH)
|| (!backward && HAVE_FORWARD_PREFETCH))
{
ref->prefetch_before = 1;
return;
}
ref->prefetch_mod = PREFETCH_BLOCK / step;
}
/* Divides X by BY, rounding down. */
static HOST_WIDE_INT
ddown (HOST_WIDE_INT x, unsigned HOST_WIDE_INT by)
{
gcc_assert (by > 0);
if (x >= 0)
return x / by;
else
return (x + by - 1) / by;
}
/* Prune the prefetch candidate REF using the reuse with BY.
If BY_IS_BEFORE is true, BY is before REF in the loop. */
static void
prune_ref_by_group_reuse (struct mem_ref *ref, struct mem_ref *by,
bool by_is_before)
{
HOST_WIDE_INT step = ref->group->step;
bool backward = step < 0;
HOST_WIDE_INT delta_r = ref->delta, delta_b = by->delta;
HOST_WIDE_INT delta = delta_b - delta_r;
HOST_WIDE_INT hit_from;
unsigned HOST_WIDE_INT prefetch_before, prefetch_block;
if (delta == 0)
{
/* If the references has the same address, only prefetch the
former. */
if (by_is_before)
ref->prefetch_before = 0;
return;
}
if (!step)
{
/* If the reference addresses are invariant and fall into the
same cache line, prefetch just the first one. */
if (!by_is_before)
return;
if (ddown (ref->delta, PREFETCH_BLOCK)
!= ddown (by->delta, PREFETCH_BLOCK))
return;
ref->prefetch_before = 0;
return;
}
/* Only prune the reference that is behind in the array. */
if (backward)
{
if (delta > 0)
return;
/* Transform the data so that we may assume that the accesses
are forward. */
delta = - delta;
step = -step;
delta_r = PREFETCH_BLOCK - 1 - delta_r;
delta_b = PREFETCH_BLOCK - 1 - delta_b;
}
else
{
if (delta < 0)
return;
}
/* Check whether the two references are likely to hit the same cache
line, and how distant the iterations in that it occurs are from
each other. */
if (step <= PREFETCH_BLOCK)
{
/* The accesses are sure to meet. Let us check when. */
hit_from = ddown (delta_b, PREFETCH_BLOCK) * PREFETCH_BLOCK;
prefetch_before = (hit_from - delta_r + step - 1) / step;
if (prefetch_before < ref->prefetch_before)
ref->prefetch_before = prefetch_before;
return;
}
/* A more complicated case. First let us ensure that size of cache line
and step are coprime (here we assume that PREFETCH_BLOCK is a power
of two. */
prefetch_block = PREFETCH_BLOCK;
while ((step & 1) == 0
&& prefetch_block > 1)
{
step >>= 1;
prefetch_block >>= 1;
delta >>= 1;
}
/* Now step > prefetch_block, and step and prefetch_block are coprime.
Determine the probability that the accesses hit the same cache line. */
prefetch_before = delta / step;
delta %= step;
if ((unsigned HOST_WIDE_INT) delta
<= (prefetch_block * ACCEPTABLE_MISS_RATE / 1000))
{
if (prefetch_before < ref->prefetch_before)
ref->prefetch_before = prefetch_before;
return;
}
/* Try also the following iteration. */
prefetch_before++;
delta = step - delta;
if ((unsigned HOST_WIDE_INT) delta
<= (prefetch_block * ACCEPTABLE_MISS_RATE / 1000))
{
if (prefetch_before < ref->prefetch_before)
ref->prefetch_before = prefetch_before;
return;
}
/* The ref probably does not reuse by. */
return;
}
/* Prune the prefetch candidate REF using the reuses with other references
in REFS. */
static void
prune_ref_by_reuse (struct mem_ref *ref, struct mem_ref *refs)
{
struct mem_ref *prune_by;
bool before = true;
prune_ref_by_self_reuse (ref);
for (prune_by = refs; prune_by; prune_by = prune_by->next)
{
if (prune_by == ref)
{
before = false;
continue;
}
if (!WRITE_CAN_USE_READ_PREFETCH
&& ref->write_p
&& !prune_by->write_p)
continue;
if (!READ_CAN_USE_WRITE_PREFETCH
&& !ref->write_p
&& prune_by->write_p)
continue;
prune_ref_by_group_reuse (ref, prune_by, before);
}
}
/* Prune the prefetch candidates in GROUP using the reuse analysis. */
static void
prune_group_by_reuse (struct mem_ref_group *group)
{
struct mem_ref *ref_pruned;
for (ref_pruned = group->refs; ref_pruned; ref_pruned = ref_pruned->next)
{
prune_ref_by_reuse (ref_pruned, group->refs);
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "Reference %p:", (void *) ref_pruned);
if (ref_pruned->prefetch_before == PREFETCH_ALL
&& ref_pruned->prefetch_mod == 1)
fprintf (dump_file, " no restrictions");
else if (ref_pruned->prefetch_before == 0)
fprintf (dump_file, " do not prefetch");
else if (ref_pruned->prefetch_before <= ref_pruned->prefetch_mod)
fprintf (dump_file, " prefetch once");
else
{
if (ref_pruned->prefetch_before != PREFETCH_ALL)
{
fprintf (dump_file, " prefetch before ");
fprintf (dump_file, HOST_WIDE_INT_PRINT_DEC,
ref_pruned->prefetch_before);
}
if (ref_pruned->prefetch_mod != 1)
{
fprintf (dump_file, " prefetch mod ");
fprintf (dump_file, HOST_WIDE_INT_PRINT_DEC,
ref_pruned->prefetch_mod);
}
}
fprintf (dump_file, "\n");
}
}
}
/* Prune the list of prefetch candidates GROUPS using the reuse analysis. */
static void
prune_by_reuse (struct mem_ref_group *groups)
{
for (; groups; groups = groups->next)
prune_group_by_reuse (groups);
}
/* Returns true if we should issue prefetch for REF. */
static bool
should_issue_prefetch_p (struct mem_ref *ref)
{
/* For now do not issue prefetches for only first few of the
iterations. */
if (ref->prefetch_before != PREFETCH_ALL)
return false;
return true;
}
/* Decide which of the prefetch candidates in GROUPS to prefetch.
AHEAD is the number of iterations to prefetch ahead (which corresponds
to the number of simultaneous instances of one prefetch running at a
time). UNROLL_FACTOR is the factor by that the loop is going to be
unrolled. Returns true if there is anything to prefetch. */
static bool
schedule_prefetches (struct mem_ref_group *groups, unsigned unroll_factor,
unsigned ahead)
{
unsigned remaining_prefetch_slots, n_prefetches, prefetch_slots;
unsigned slots_per_prefetch;
2006-02-14 14:51:51 +01:00
struct mem_ref *ref;
bool any = false;
/* At most SIMULTANEOUS_PREFETCHES should be running at the same time. */
remaining_prefetch_slots = SIMULTANEOUS_PREFETCHES;
2006-02-14 14:51:51 +01:00
/* The prefetch will run for AHEAD iterations of the original loop, i.e.,
AHEAD / UNROLL_FACTOR iterations of the unrolled loop. In each iteration,
it will need a prefetch slot. */
slots_per_prefetch = (ahead + unroll_factor / 2) / unroll_factor;
2006-02-14 14:51:51 +01:00
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Each prefetch instruction takes %u prefetch slots.\n",
slots_per_prefetch);
2006-02-14 14:51:51 +01:00
/* For now we just take memory references one by one and issue
prefetches for as many as possible. The groups are sorted
starting with the largest step, since the references with
large step are more likely to cause many cache misses. */
2006-02-14 14:51:51 +01:00
for (; groups; groups = groups->next)
for (ref = groups->refs; ref; ref = ref->next)
{
if (!should_issue_prefetch_p (ref))
continue;
/* If we need to prefetch the reference each PREFETCH_MOD iterations,
and we unroll the loop UNROLL_FACTOR times, we need to insert
ceil (UNROLL_FACTOR / PREFETCH_MOD) instructions in each
iteration. */
2006-02-14 14:51:51 +01:00
n_prefetches = ((unroll_factor + ref->prefetch_mod - 1)
/ ref->prefetch_mod);
prefetch_slots = n_prefetches * slots_per_prefetch;
/* If more than half of the prefetches would be lost anyway, do not
issue the prefetch. */
if (2 * remaining_prefetch_slots < prefetch_slots)
continue;
ref->issue_prefetch_p = true;
2006-02-14 14:51:51 +01:00
if (remaining_prefetch_slots <= prefetch_slots)
return true;
remaining_prefetch_slots -= prefetch_slots;
2006-02-14 14:51:51 +01:00
any = true;
}
return any;
}
/* Determine whether there is any reference suitable for prefetching
in GROUPS. */
static bool
anything_to_prefetch_p (struct mem_ref_group *groups)
{
struct mem_ref *ref;
for (; groups; groups = groups->next)
for (ref = groups->refs; ref; ref = ref->next)
if (should_issue_prefetch_p (ref))
return true;
return false;
}
/* Issue prefetches for the reference REF into loop as decided before.
HEAD is the number of iterations to prefetch ahead. UNROLL_FACTOR
is the factor by which LOOP was unrolled. */
2006-02-14 14:51:51 +01:00
static void
issue_prefetch_ref (struct mem_ref *ref, unsigned unroll_factor, unsigned ahead)
{
HOST_WIDE_INT delta;
tree addr, addr_base, prefetch, params, write_p;
block_stmt_iterator bsi;
unsigned n_prefetches, ap;
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Issued prefetch for %p.\n", (void *) ref);
bsi = bsi_for_stmt (ref->stmt);
n_prefetches = ((unroll_factor + ref->prefetch_mod - 1)
/ ref->prefetch_mod);
addr_base = build_fold_addr_expr_with_type (ref->mem, ptr_type_node);
addr_base = force_gimple_operand_bsi (&bsi, unshare_expr (addr_base), true, NULL);
write_p = ref->write_p ? integer_one_node : integer_zero_node;
2006-02-14 14:51:51 +01:00
for (ap = 0; ap < n_prefetches; ap++)
{
/* Determine the address to prefetch. */
delta = (ahead + ap * ref->prefetch_mod) * ref->group->step;
addr = fold_build2 (PLUS_EXPR, ptr_type_node,
addr_base, build_int_cst (ptr_type_node, delta));
addr = force_gimple_operand_bsi (&bsi, unshare_expr (addr), true, NULL);
/* Create the prefetch instruction. */
params = tree_cons (NULL_TREE, addr,
tree_cons (NULL_TREE, write_p, NULL_TREE));
2006-02-14 14:51:51 +01:00
prefetch = build_function_call_expr (built_in_decls[BUILT_IN_PREFETCH],
params);
bsi_insert_before (&bsi, prefetch, BSI_SAME_STMT);
}
}
/* Issue prefetches for the references in GROUPS into loop as decided before.
HEAD is the number of iterations to prefetch ahead. UNROLL_FACTOR is the
factor by that LOOP was unrolled. */
static void
issue_prefetches (struct mem_ref_group *groups,
unsigned unroll_factor, unsigned ahead)
{
struct mem_ref *ref;
for (; groups; groups = groups->next)
for (ref = groups->refs; ref; ref = ref->next)
if (ref->issue_prefetch_p)
issue_prefetch_ref (ref, unroll_factor, ahead);
}
/* Determines whether we can profitably unroll LOOP FACTOR times, and if
this is the case, fill in DESC by the description of number of
iterations. */
static bool
should_unroll_loop_p (struct loop *loop, struct tree_niter_desc *desc,
unsigned factor)
{
if (!can_unroll_loop_p (loop, factor, desc))
return false;
/* We only consider loops without control flow for unrolling. This is not
a hard restriction -- tree_unroll_loop works with arbitrary loops
as well; but the unrolling/prefetching is usually more profitable for
loops consisting of a single basic block, and we want to limit the
code growth. */
if (loop->num_nodes > 2)
return false;
return true;
}
/* Determine the coefficient by that unroll LOOP, from the information
contained in the list of memory references REFS. Description of
umber of iterations of LOOP is stored to DESC. AHEAD is the number
of iterations ahead that we need to prefetch. NINSNS is number of
insns of the LOOP. */
static unsigned
determine_unroll_factor (struct loop *loop, struct mem_ref_group *refs,
unsigned ninsns, struct tree_niter_desc *desc)
2006-02-14 14:51:51 +01:00
{
unsigned upper_bound;
unsigned nfactor, factor, mod_constraint;
2006-02-14 14:51:51 +01:00
struct mem_ref_group *agp;
struct mem_ref *ref;
/* First check whether the loop is not too large to unroll. We ignore
PARAM_MAX_UNROLL_TIMES, because for small loops, it prevented us
from unrolling them enough to make exactly one cache line covered by each
iteration. Also, the goal of PARAM_MAX_UNROLL_TIMES is to prevent
us from unrolling the loops too many times in cases where we only expect
gains from better scheduling and decreasing loop overhead, which is not
the case here. */
upper_bound = PARAM_VALUE (PARAM_MAX_UNROLLED_INSNS) / ninsns;
if (upper_bound <= 1)
2006-02-14 14:51:51 +01:00
return 1;
/* Choose the factor so that we may prefetch each cache just once,
but bound the unrolling by UPPER_BOUND. */
factor = 1;
2006-02-14 14:51:51 +01:00
for (agp = refs; agp; agp = agp->next)
for (ref = agp->refs; ref; ref = ref->next)
if (should_issue_prefetch_p (ref))
{
mod_constraint = ref->prefetch_mod;
nfactor = least_common_multiple (mod_constraint, factor);
if (nfactor <= upper_bound)
factor = nfactor;
}
2006-02-14 14:51:51 +01:00
if (!should_unroll_loop_p (loop, desc, factor))
return 1;
return factor;
}
/* Issue prefetch instructions for array references in LOOP. Returns
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
true if the LOOP was unrolled. */
2006-02-14 14:51:51 +01:00
static bool
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
loop_prefetch_arrays (struct loop *loop)
2006-02-14 14:51:51 +01:00
{
struct mem_ref_group *refs;
unsigned ahead, ninsns, unroll_factor;
struct tree_niter_desc desc;
bool unrolled = false;
/* Step 1: gather the memory references. */
refs = gather_memory_references (loop);
/* Step 2: estimate the reuse effects. */
prune_by_reuse (refs);
if (!anything_to_prefetch_p (refs))
goto fail;
/* Step 3: determine the ahead and unroll factor. */
/* FIXME: We should use not size of the loop, but the average number of
instructions executed per iteration of the loop. */
ninsns = tree_num_loop_insns (loop);
ahead = (PREFETCH_LATENCY + ninsns - 1) / ninsns;
unroll_factor = determine_unroll_factor (loop, refs, ninsns, &desc);
2006-02-14 14:51:51 +01:00
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Ahead %d, unroll factor %d\n", ahead, unroll_factor);
/* If the loop rolls less than the required unroll factor, prefetching
is useless. */
if (unroll_factor > 1
&& cst_and_fits_in_hwi (desc.niter)
&& (unsigned HOST_WIDE_INT) int_cst_value (desc.niter) < unroll_factor)
goto fail;
/* Step 4: what to prefetch? */
if (!schedule_prefetches (refs, unroll_factor, ahead))
goto fail;
/* Step 5: unroll the loop. TODO -- peeling of first and last few
iterations so that we do not issue superfluous prefetches. */
if (unroll_factor != 1)
{
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
tree_unroll_loop (loop, unroll_factor,
2006-02-14 14:51:51 +01:00
single_dom_exit (loop), &desc);
unrolled = true;
}
/* Step 6: issue the prefetches. */
issue_prefetches (refs, unroll_factor, ahead);
fail:
release_mem_refs (refs);
return unrolled;
}
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
/* Issue prefetch instructions for array references in loops. */
2006-02-14 14:51:51 +01:00
unsigned int
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
tree_ssa_prefetch_arrays (void)
2006-02-14 14:51:51 +01:00
{
tree-vrp.c (adjust_range_with_scev): Use get_chrec_loop. * tree-vrp.c (adjust_range_with_scev): Use get_chrec_loop. * loop-unswitch.c (unswitch_loops): Use FOR_EACH_LOOP. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree_ssa_unswitch_loops (tree_ssa_unswitch_loops): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-loop-ivcanon.c (canonicalize_induction_variables, tree_unroll_loops_completely): Ditto. * predict.c (predict_loops): Ditto. * tree-if-conv.c (main_tree_if_conversion): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling): Ditto. * cfgloopmanip.c (unloop): Use delete_loop. (place_new_loop): Access larray vector instead of parray. (create_preheaders, force_single_succ_latches, fix_loop_structure): Use FOR_EACH_LOOP and delete_loop.. * loop-doloop.c (doloop_optimize_loops): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-cfg.c (replace_uses_by): Ditto. * tree-ssa-loop-prefetch.c (tree_ssa_prefetch_arrays): Ditto. * tree-chrec.h (CHREC_VAR, CHREC_LEFT, CHREC_RIGHT, CHREC_VARIABLE): Moved to ... * tree.h (CHREC_VAR, CHREC_LEFT, CHREC_RIGHT, CHREC_VARIABLE): ... here. * tree-scalar-evolution.c (chrec_contains_symbols_defined_in_loop, compute_overall_effect_of_inner_loop, chrec_is_positive): Use get_loop and get_chrec_loop. (number_of_iterations_for_all_loops): Use number_of_loops. (scev_initialize, scev_reset, scev_const_prop): Use FOR_EACH_LOOP. * tree-scalar-evolution.h (get_chrec_loop): New inline function. * cfgloopanal.c (mark_irreducible_loops): Use number_of_loops, and FOR_EACH_LOOP. * tree-chrec.c (evolution_function_is_invariant_rec_p, chrec_convert_1): Use get_loop and get_chrec_loop. * loop-init.c (loop_optimizer_init): Use number_of_loops. (loop_optimizer_init): Use FOR_EACH_LOOP. * tree-vectorizer.c (vect_loops_num): Removed. (vectorize_loops): Store number of loops locally. Use FOR_EACH_LOOP and get_loop. * tree-vectorizer.h (vect_loops_num): Removed. * tree-data-ref.c (get_number_of_iters_for_loop): Use get_loop. (find_data_references_in_loop): Do not set parallel_p. * tree-data-ref.h: Do not declare VEC(loop_p). * cfgloop.c (flow_loops_dump, mark_single_exit_loops, verify_loop_structure): Use FOR_EACH_LOOP. (flow_loops_free): Use FOR_EACH_LOOP, free larray vector. (initialize_loops_parallel_p): Removed. (flow_loops_find): Push the loops into a vector. (delete_loop): New function. (cancel_loop): Use delete_loop. * cfgloop.h: Declare VEC(loop_p). (struct loop): Remove parallel_p field. (struct loops): Replace num and parray field by larray vector. Remove shared_headers field. (delete_loop): Declare. (get_loop, get_loops, number_of_loops, fel_next, fel_init, FOR_EACH_LOOP): New. * doc/loop.tex: Document new accessor functions. From-SVN: r119713
2006-12-10 23:17:15 +01:00
loop_iterator li;
2006-02-14 14:51:51 +01:00
struct loop *loop;
bool unrolled = false;
int todo_flags = 0;
2006-02-14 14:51:51 +01:00
if (!HAVE_prefetch
/* It is possible to ask compiler for say -mtune=i486 -march=pentium4.
-mtune=i486 causes us having PREFETCH_BLOCK 0, since this is part
of processor costs and i486 does not have prefetch, but
-march=pentium4 causes HAVE_prefetch to be true. Ugh. */
|| PREFETCH_BLOCK == 0)
return 0;
2006-02-14 14:51:51 +01:00
2006-11-12 20:17:02 +01:00
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "Prefetching parameters:\n");
fprintf (dump_file, " simultaneous prefetches: %d\n",
SIMULTANEOUS_PREFETCHES);
fprintf (dump_file, " prefetch latency: %d\n", PREFETCH_LATENCY);
fprintf (dump_file, " L1 cache size: %d (%d bytes)\n",
L1_CACHE_SIZE, L1_CACHE_SIZE * L1_CACHE_LINE_SIZE);
fprintf (dump_file, " L1 cache line size: %d\n", L1_CACHE_LINE_SIZE);
fprintf (dump_file, " prefetch block size: %d\n", PREFETCH_BLOCK);
fprintf (dump_file, "\n");
}
2006-02-14 14:51:51 +01:00
initialize_original_copy_tables ();
if (!built_in_decls[BUILT_IN_PREFETCH])
{
tree type = build_function_type (void_type_node,
tree_cons (NULL_TREE,
const_ptr_type_node,
NULL_TREE));
decl.c: Include langhooks.h. * gcc/java/decl.c: Include langhooks.h. (builtin_function): Remove. (java_init_decl_processing): Replace calls to builtin_function with add_builtin_function. * gcc/java/Make-lang.in (jc1$(exeext)): Depend on and link with attribs.o. (java/decl.o): Depend on langhooks.h. * gcc/java/java-tree.h (builtin_function): Remove. * gcc/tree.c (local_define_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/objc/objc-act.c (synth_module_prologue): Replace calls to builtin_function with add_builtin_function. (build_next_objc_exception_stuff): Replace calls to builtin_function with add_builtin_function. (build_objc_exception_stuff): Replace calls to builtin_function with add_builtin_function. * gcc/objcp/objcp-decl.h (objcp_builtin_function): Remove. * gcc/cp/decl.c (builtin_function_1): Move common code to add_builtin_function. (builtin_function): Rename to cxx_builtin_function. Change the signature. * gcc/cp/call.c: Include langhooks.h. (build_java_interface_fn_ref): Replace calls to builtin_function with add_builtin_function. * gcc/cp/Make-lang.in (cp/call.o): Depend on langhooks.h. * gcc/cp/cp-objcp-common.h (LANG_HOOKS_BUILTIN_FUNCTION): Define as cxx_builtin_function. * gcc/cp/cp-tree.h (builtin_function): Rename to cxx_builtin_function. Change the signature. * gcc/c-objc-common.h (LANG_HOOKS_BUILTIN_FUNCTION): Define as c_builtin_function. * gcc/c-tree.h (builtin_function): Rename to c_builtin_function. Change the signature. * gcc/ada/utils.c (builtin_function): Rename to gnat_builtin_function. Move common code to add_builtin_function. * gcc/ada/misc.c (LANG_HOOKS_BUILTIN_FUNCTION): Define as gnat_builtin_function. * gcc/ada/gigi.h (builtin_function): Rename to gnat_builtin_function. Change the signature. * gcc/c-decl.c (builtin_function): Rename to c_builtin_function. Move common code to add_builtin_function. * gcc/fortran/Make-lang.in (f951$(exeext)): Depend on and link with attribs.o. * gcc/fortran/trans.h (builtin_function): Rename to gfc_builtin_function. Change the signature. * gcc/fortran/f95-lang.c (LANG_HOOKS_BUILTIN_FUNCTION): Define as gfc_builtin_function. (builtin_function): Rename to gfc_builtin_function. Move common code to builtin_function. (gfc_define_builtin): Replace calls to builtin_function with gfc_define_builtin. * gcc/langhooks.c (add_builtin_function): New function. (lhd_builtin_function): New function. * gcc/langhooks.h (struct lang_hooks): Change the signature of builtin_function. (add_builtin_function): New function. * gcc/treelang/treetree.c (builtin_function): Remove. * gcc/tree-ssa-loop-prefetch.c (tree_ssa_prefetch_arrays): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/c-common.c (decl_builtin_1): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/alpha/alpha.c (alpha_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/frv/frv.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/s390/s390.c (s390_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/sparc/sparc.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/i386/i386.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/sh/sh.c (sh_media_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/c4x/c4x.c (c4x_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/stormy16/stormy16.c (xstormy16_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/iq2000/iq2000.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/ia64/ia64.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/rs6000/rs6000.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. (altivec_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/arm/arm.c (def_mbuiltin): Replace calls to lang_hooks.builtin_function with add_builtin_function. (arm_init_tls_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/mips/mips.c (mips_init_builtins): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/config/bfin/bfin.c (def_builtin): Replace calls to lang_hooks.builtin_function with add_builtin_function. * gcc/langhooks-def.h (lhd_builtin_function): New function. (LANG_HOOKS_BUILTIN_FUNCTION): Define as lhd_builtin_function. From-SVN: r117981
2006-10-23 21:47:58 +02:00
tree decl = add_builtin_function ("__builtin_prefetch", type,
BUILT_IN_PREFETCH, BUILT_IN_NORMAL,
NULL, NULL_TREE);
2006-02-14 14:51:51 +01:00
DECL_IS_NOVOPS (decl) = true;
built_in_decls[BUILT_IN_PREFETCH] = decl;
}
/* We assume that size of cache line is a power of two, so verify this
here. */
gcc_assert ((PREFETCH_BLOCK & (PREFETCH_BLOCK - 1)) == 0);
tree-vrp.c (adjust_range_with_scev): Use get_chrec_loop. * tree-vrp.c (adjust_range_with_scev): Use get_chrec_loop. * loop-unswitch.c (unswitch_loops): Use FOR_EACH_LOOP. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree_ssa_unswitch_loops (tree_ssa_unswitch_loops): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-loop-ivcanon.c (canonicalize_induction_variables, tree_unroll_loops_completely): Ditto. * predict.c (predict_loops): Ditto. * tree-if-conv.c (main_tree_if_conversion): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling): Ditto. * cfgloopmanip.c (unloop): Use delete_loop. (place_new_loop): Access larray vector instead of parray. (create_preheaders, force_single_succ_latches, fix_loop_structure): Use FOR_EACH_LOOP and delete_loop.. * loop-doloop.c (doloop_optimize_loops): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-cfg.c (replace_uses_by): Ditto. * tree-ssa-loop-prefetch.c (tree_ssa_prefetch_arrays): Ditto. * tree-chrec.h (CHREC_VAR, CHREC_LEFT, CHREC_RIGHT, CHREC_VARIABLE): Moved to ... * tree.h (CHREC_VAR, CHREC_LEFT, CHREC_RIGHT, CHREC_VARIABLE): ... here. * tree-scalar-evolution.c (chrec_contains_symbols_defined_in_loop, compute_overall_effect_of_inner_loop, chrec_is_positive): Use get_loop and get_chrec_loop. (number_of_iterations_for_all_loops): Use number_of_loops. (scev_initialize, scev_reset, scev_const_prop): Use FOR_EACH_LOOP. * tree-scalar-evolution.h (get_chrec_loop): New inline function. * cfgloopanal.c (mark_irreducible_loops): Use number_of_loops, and FOR_EACH_LOOP. * tree-chrec.c (evolution_function_is_invariant_rec_p, chrec_convert_1): Use get_loop and get_chrec_loop. * loop-init.c (loop_optimizer_init): Use number_of_loops. (loop_optimizer_init): Use FOR_EACH_LOOP. * tree-vectorizer.c (vect_loops_num): Removed. (vectorize_loops): Store number of loops locally. Use FOR_EACH_LOOP and get_loop. * tree-vectorizer.h (vect_loops_num): Removed. * tree-data-ref.c (get_number_of_iters_for_loop): Use get_loop. (find_data_references_in_loop): Do not set parallel_p. * tree-data-ref.h: Do not declare VEC(loop_p). * cfgloop.c (flow_loops_dump, mark_single_exit_loops, verify_loop_structure): Use FOR_EACH_LOOP. (flow_loops_free): Use FOR_EACH_LOOP, free larray vector. (initialize_loops_parallel_p): Removed. (flow_loops_find): Push the loops into a vector. (delete_loop): New function. (cancel_loop): Use delete_loop. * cfgloop.h: Declare VEC(loop_p). (struct loop): Remove parallel_p field. (struct loops): Replace num and parray field by larray vector. Remove shared_headers field. (delete_loop): Declare. (get_loop, get_loops, number_of_loops, fel_next, fel_init, FOR_EACH_LOOP): New. * doc/loop.tex: Document new accessor functions. From-SVN: r119713
2006-12-10 23:17:15 +01:00
FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST)
2006-02-14 14:51:51 +01:00
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Processing loop %d:\n", loop->num);
tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * tree-vrp.c (execute_vrp): Do not pass loops structure through arguments. * loop-unswitch.c (unswitch_loops, unswitch_single_loop, unswitch_loop): Ditto. * tree-loop-linear.c (linear_transform_loops): Ditto. * tree-ssa-loop-im.c (determine_lsm, tree_ssa_lim_initialize, tree_ssa_lim): Ditto. * tree-ssa-loop-niter.c (estimate_numbers_of_iterations, free_numbers_of_iterations_estimates): Ditto. * tree-ssa-loop-unswitch.c (tree_ssa_unswitch_loops, tree_unswitch_single_loop, tree_unswitch_loop): Ditto. * cfgloopmanip.c (fix_bb_placement, fix_bb_placements, remove_path, add_loop, loopify, unloop, fix_loop_placements, place_new_loop, duplicate_loop, duplicate_subloops, update_single_exit_for_duplicated_loops, copy_loops_to, duplicate_loop_to_header_edge, create_preheaders, force_single_succ_latches, loop_version, fix_loop_structure): Ditto. * tree-ssa-loop-manip.c (tree_duplicate_loop_to_header_edge, tree_unroll_loop): Ditto. * tree-ssa-loop-ch.c (copy_loop_headers): Ditto. * tree-scalar-evolution.c (select_loops_exit_conditions, scev_initialize, scev_analysis): Ditto. * tree-scalar-evolution.h (scev_initialize): Ditto. * cfghooks.c (cfg_hook_duplicate_loop_to_header_edge): Ditto. * cfgloopanal.c (mark_irreducible_loops, mark_loop_exit_edges): Ditto. * tree-ssa-loop-ivopts.c (tree_ssa_iv_optimize): Ditto. * modulo-sched.c (sms_schedule): Ditto. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Ditto. * loop-init.c (loop_optimizer_init, rtl_move_loop_invariants, rtl_unswitch, rtl_unroll_and_peel_loops, rtl_doloop): Ditto. * ifcvt.c (if_convert): Ditto. * tree-ssa-loop-ivcanon.c (try_unroll_loop_completely, canonicalize_loop_induction_variables, canonicalize_induction_variables, tree_unroll_loops_completely, remove_empty_loops): Ditto. * tree-ssa-loop.c (tree_ssa_loop_init, tree_ssa_loop_im, tree_ssa_loop_unswitch, tree_vectorize, tree_linear_transform, tree_ssa_loop_ivcanon, tree_ssa_empty_loop, tree_ssa_loop_bounds, tree_complete_unroll, tree_ssa_loop_prefetch, tree_ssa_loop_ivopts, tree_ssa_loop_done): Ditto. * predict.c (predict_loops, tree_estimate_probability, estimate_loops, estimate_bb_frequencies): Ditto. * tree-vectorizer.c (slpeel_tree_duplicate_loop_to_edge_cfg, slpeel_tree_peel_loop_to_edge, vectorize_loops): Ditto. * loop-unroll.c (unroll_and_peel_loops, peel_loops_completely, decide_unrolling_and_peeling, peel_loop_completely, unroll_loop_constant_iterations, unroll_loop_runtime_iterations, peel_loop_simple, unroll_loop_stupid): Ditto. * loop-doloop.c (doloop_optimize_loops): Ditto. * tree-cfgcleanup.c (cleanup_tree_cfg_loop): Ditto. * loop-invariant.c (move_loop_invariants): Ditto. * tree-ssa-dce.c (tree_ssa_dce_loop): Ditto. * tree-ssa-loop-prefetch.c (loop_prefetch_arrays, tree_ssa_prefetch_arrays): Ditto. * lambda-code.c (gcc_loopnest_to_lambda_loopnest, perfect_nestify): Ditto. * tree-vect-transform.c (vect_do_peeling_for_loop_bound, vect_do_peeling_for_alignment, vect_transform_loop): Ditto. * cfgloop.c (flow_loops_cfg_dump, flow_loops_dump, mark_single_exit_loops, cancel_loop, cancel_loop_tree, verify_loop_structure): Ditto. * tree-flow.h (vectorize_loops, tree_ssa_lim, tree_ssa_unswitch_loops, canonicalize_induction_variables, tree_unroll_loops_completely, tree_ssa_prefetch_arrays, remove_empty_loops, tree_ssa_iv_optimize, estimate_numbers_of_iterations, free_numbers_of_iterations_estimates, tree_duplicate_loop_to_header_edge, tree_ssa_loop_version, tree_unroll_loop, linear_transform_loops): Declaration changed. * basic-block.h: Remove declaration of struct loops. * cfghooks.h (struct cfg_hooks): Change type of cfg_hook_duplicate_loop_to_header_edge. (cfg_hook_duplicate_loop_to_header_edge): Declaration changed. * tree-vectorizer.h (slpeel_tree_peel_loop_to_edge, vect_transform_loop): Declaration changed. * lambda.h (gcc_loopnest_to_lambda_loopnest): Declaration changed. * cfgloop.h (flow_loops_dump, fix_loop_structure, mark_irreducible_loops, mark_single_exit_loops, mark_loop_exit_edges, cancel_loop_tree, create_preheaders, force_single_succ_latches, verify_loop_structure, duplicate_loop, duplicate_loop_to_header_edge, loopify, loop_version, remove_path, unswitch_loops, unroll_and_peel_loops, doloop_optimize_loops, move_loop_invariants): Declaration changed. From-SVN: r119189
2006-11-25 11:34:13 +01:00
unrolled |= loop_prefetch_arrays (loop);
2006-02-14 14:51:51 +01:00
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "\n\n");
}
if (unrolled)
{
scev_reset ();
todo_flags |= TODO_cleanup_cfg;
2006-02-14 14:51:51 +01:00
}
free_original_copy_tables ();
return todo_flags;
2006-02-14 14:51:51 +01:00
}