re PR tree-optimization/18754 (unrolling happens too late/SRA does not happen late enough)

2008-04-27  Richard Guenther  <rguenther@suse.de>

	PR tree-optimization/18754
	PR tree-optimization/34223
	* tree-pass.h (pass_complete_unrolli): Declare.
	* tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Print
	loop size before and after unconditionally of UL_NO_GROWTH in effect.
	Rewrite loop into loop closed SSA form if it is not already.
	(tree_unroll_loops_completely): Re-structure to iterate over
	innermost loops with intermediate CFG cleanups.
	Unroll outermost loops only if requested or the code does not grow
	doing so.
	* tree-ssa-loop.c (gate_tree_vectorize): Don't shortcut if no
	loops are available.
	(tree_vectorize): Instead do so here.
	(tree_complete_unroll): Also unroll outermost loops.
	(tree_complete_unroll_inner): New function.
	(gate_tree_complete_unroll_inner): Likewise.
	(pass_complete_unrolli): New pass.
	* tree-ssa-loop-manip.c (find_uses_to_rename_use): Only record
	uses outside of the loop.
	(tree_duplicate_loop_to_header_edge): Only verify loop-closed SSA
	form if it is available.  
	* tree-flow.h (tree_unroll_loops_completely): Add extra parameter.
	* passes.c (init_optimization_passes): Schedule complete inner
	loop unrolling pass before the first CCP pass after final inlining.

	* gcc.dg/tree-ssa/loop-36.c: New testcase.
	* gcc.dg/tree-ssa/loop-37.c: Likewise.
	* gcc.dg/vect/vect-118.c: Likewise.
	* gcc.dg/Wunreachable-8.c: XFAIL bogus warning.
	* gcc.dg/vect/vect-66.c: Increase loop trip count.
	* gcc.dg/vect/no-section-anchors-vect-66.c: Likewise.
	* gcc.dg/vect/no-section-anchors-vect-69.c: Likewise.
	* gcc.dg/vect/vect-76.c: Likewise.
	* gcc.dg/vect/vect-outer-6.c: Likewise.
	* gcc.dg/vect/vect-outer-1.c: Likewise.
	* gcc.dg/vect/vect-outer-1a.c: Likewise.
	* gcc.dg/vect/vect-11a.c: Likewise.
	* gcc.dg/vect/vect-shift-1.c: Likewise.
	* gcc.target/i386/vectorize1.c: Likewise.

From-SVN: r134730
This commit is contained in:
Richard Guenther 2008-04-27 16:27:08 +00:00 committed by Richard Biener
parent feb60f0328
commit d6e840eed2
22 changed files with 307 additions and 87 deletions

View file

@ -1,3 +1,30 @@
2008-04-27 Richard Guenther <rguenther@suse.de>
PR tree-optimization/18754
PR tree-optimization/34223
* tree-pass.h (pass_complete_unrolli): Declare.
* tree-ssa-loop-ivcanon.c (try_unroll_loop_completely): Print
loop size before and after unconditionally of UL_NO_GROWTH in effect.
Rewrite loop into loop closed SSA form if it is not already.
(tree_unroll_loops_completely): Re-structure to iterate over
innermost loops with intermediate CFG cleanups.
Unroll outermost loops only if requested or the code does not grow
doing so.
* tree-ssa-loop.c (gate_tree_vectorize): Don't shortcut if no
loops are available.
(tree_vectorize): Instead do so here.
(tree_complete_unroll): Also unroll outermost loops.
(tree_complete_unroll_inner): New function.
(gate_tree_complete_unroll_inner): Likewise.
(pass_complete_unrolli): New pass.
* tree-ssa-loop-manip.c (find_uses_to_rename_use): Only record
uses outside of the loop.
(tree_duplicate_loop_to_header_edge): Only verify loop-closed SSA
form if it is available.
* tree-flow.h (tree_unroll_loops_completely): Add extra parameter.
* passes.c (init_optimization_passes): Schedule complete inner
loop unrolling pass before the first CCP pass after final inlining.
2008-04-27 Nathan Sidwell <nathan@codesourcery.com>
* targhooks.h (default_emutls_var_fields,

View file

@ -567,6 +567,7 @@ init_optimization_passes (void)
NEXT_PASS (pass_rename_ssa_copies);
/* Initial scalar cleanups. */
NEXT_PASS (pass_complete_unrolli);
NEXT_PASS (pass_ccp);
NEXT_PASS (pass_phiprop);
NEXT_PASS (pass_fre);

View file

@ -1,3 +1,22 @@
2008-04-27 Richard Guenther <rguenther@suse.de>
PR tree-optimization/18754
PR tree-optimization/34223
* gcc.dg/tree-ssa/loop-36.c: New testcase.
* gcc.dg/tree-ssa/loop-37.c: Likewise.
* gcc.dg/vect/vect-118.c: Likewise.
* gcc.dg/Wunreachable-8.c: XFAIL bogus warning.
* gcc.dg/vect/vect-66.c: Increase loop trip count.
* gcc.dg/vect/no-section-anchors-vect-66.c: Likewise.
* gcc.dg/vect/no-section-anchors-vect-69.c: Likewise.
* gcc.dg/vect/vect-76.c: Likewise.
* gcc.dg/vect/vect-outer-6.c: Likewise.
* gcc.dg/vect/vect-outer-1.c: Likewise.
* gcc.dg/vect/vect-outer-1a.c: Likewise.
* gcc.dg/vect/vect-11a.c: Likewise.
* gcc.dg/vect/vect-shift-1.c: Likewise.
* gcc.target/i386/vectorize1.c: Likewise.
2008-04-27 Nathan Sidwell <nathan@codesourcery.com>
* gcc.dg/tls/section-2.c: New.

View file

@ -6,7 +6,7 @@ float Factorial(float X)
int k,j;
for (k=1; k < 5; k++)
{
val += 1.0;
val += 1.0; /* { dg-bogus "will never be executed" "" { xfail *-*-* } } */
}
return (val); /* { dg-bogus "will never be executed" } */
}

View file

@ -0,0 +1,21 @@
/* { dg-do compile } */
/* { dg-options "-O2 -fdump-tree-dce2" } */
struct X { float array[4]; };
struct X a,b;
float foobar () {
float s = 0;
unsigned int d;
struct X c;
for (d=0; d<4; ++d)
c.array[d] = a.array[d] * b.array[d];
for (d=0; d<4; ++d)
s+=c.array[d];
return s;
}
/* The temporary structure should have been promoted to registers
by FRE after the loops have been unrolled by the early unrolling pass. */
/* { dg-final { scan-tree-dump-not "c\.array" "dce2" } } */

View file

@ -0,0 +1,27 @@
/* { dg-do link } */
/* { dg-options "-O2 -fdump-tree-optimized" } */
extern void link_error (void);
static const int my_array [3] = { 4, 5, 6 };
void f0 (void)
{
int j, sum = 0;
for (j = 0; j < 3; j ++)
sum += my_array [j];
if (15 != sum)
link_error ();
}
int f1 (int a [])
{
int j, sum = 0;
for (j = 0; j < 3; j ++)
sum += a [j] + my_array [j];
return sum;
}
int main() { }
/* { dg-final { scan-tree-dump-not "my_array" "optimized" } } */
/* { dg-final { cleanup-tree-dump "optimized" } } */

View file

@ -3,11 +3,10 @@
#include <stdarg.h>
#include "tree-vect.h"
#define N 16
#define N 8
int ib[6] = {0,3,6,9,12,15};
int ia[8][5][6];
int ic[16][16][5][6];
int ia[8][5][N+2];
int ic[16][16][5][N+2];
__attribute__ ((noinline))
int main1 ()
@ -17,7 +16,7 @@ int main1 ()
/* Multidimensional array. Aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
ia[2][6][j] = 5;
}
@ -26,7 +25,7 @@ int main1 ()
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
if (ia[2][6][j] != 5)
abort();
@ -35,14 +34,14 @@ int main1 ()
/* Multidimensional array. Aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
ia[3][6][j+2] = 5;
}
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 2; j < 6; j++)
for (j = 2; j < N+2; j++)
{
if (ia[3][6][j] != 5)
abort();
@ -52,7 +51,7 @@ int main1 ()
/* Multidimensional array. Not aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
ic[2][1][6][j+1] = 5;
}
@ -61,7 +60,7 @@ int main1 ()
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
if (ic[2][1][6][j+1] != 5)
abort();

View file

@ -3,7 +3,7 @@
#include <stdarg.h>
#include "tree-vect.h"
#define N 8
#define N 12
struct s{
int m;

View file

@ -0,0 +1,34 @@
/* { dg-require-effective-target vect_int } */
/* { dg-options "-O3 -fdump-tree-vect-details" } */
#include "tree-vect.h"
#define M 10
#define N 3
void __attribute__((noinline))
foo (int n, int *ub, int *uc)
{
int i, j, tmp1;
for (i = 0; i < n; i++)
{
tmp1 = 0;
for (j = 0; j < M; j++)
{
tmp1 += uc[i] * ((int)(j << N) / M);
}
ub[i] = tmp1;
}
}
int main()
{
int uc[16], ub[16];
check_vect ();
foo (16, uc, ub);
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { cleanup-tree-dump "vect" } } */

View file

@ -9,15 +9,18 @@ extern void abort (void);
__attribute__ ((noinline))
void u ()
{
unsigned int A[4] = {0x08000000,0xffffffff,0xff0000ff,0xf0000001};
unsigned int B[4] = {0x08000000,0x08000001,0xff0000ff,0xf0000001};
unsigned int Answer[4] = {0,0xf7ffffff,0x0200fe01,0xe0000001};
unsigned int C[4];
unsigned int A[8] = {0x08000000,0xffffffff,0xff0000ff,0xf0000001,
0x08000000,0xffffffff,0xff0000ff,0xf0000001};
unsigned int B[8] = {0x08000000,0x08000001,0xff0000ff,0xf0000001,
0x08000000,0x08000001,0xff0000ff,0xf0000001};
unsigned int Answer[8] = {0,0xf7ffffff,0x0200fe01,0xe0000001,
0,0xf7ffffff,0x0200fe01,0xe0000001};
unsigned int C[8];
int i, j;
for (i=0; i<4; i++)
for (i=0; i<8; i++)
C[i] = A[i] * B[i];
for (i=0; i<4; i++)
for (i=0; i<8; i++)
if (C[i] != Answer[i])
abort ();
}
@ -25,15 +28,18 @@ void u ()
__attribute__ ((noinline))
void s()
{
signed int A[4] = {0x08000000,0xffffffff,0xff0000ff,0xf0000001};
signed int B[4] = {0x08000000,0x08000001,0xff0000ff,0xf0000001};
signed int Answer[4] = {0,0xf7ffffff,0x0200fe01, 0xe0000001};
signed int C[4];
signed int A[8] = {0x08000000,0xffffffff,0xff0000ff,0xf0000001,
0x08000000,0xffffffff,0xff0000ff,0xf0000001};
signed int B[8] = {0x08000000,0x08000001,0xff0000ff,0xf0000001,
0x08000000,0x08000001,0xff0000ff,0xf0000001};
signed int Answer[8] = {0,0xf7ffffff,0x0200fe01, 0xe0000001,
0,0xf7ffffff,0x0200fe01, 0xe0000001};
signed int C[8];
int i, j;
for (i=0; i<4; i++)
for (i=0; i<8; i++)
C[i] = A[i] * B[i];
for (i=0; i<4; i++)
for (i=0; i<8; i++)
if (C[i] != Answer[i])
abort ();
}

View file

@ -3,20 +3,18 @@
#include <stdarg.h>
#include "tree-vect.h"
#define N 16
#define N 8
__attribute__ ((noinline))
int main1 ()
void main1 ()
{
int i, j;
int ib[6] = {0,3,6,9,12,15};
int ia[8][5][6];
int ic[16][16][5][6];
int ia[8][5][N+2];
/* Multidimensional array. Aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
ia[2][6][j] = 5;
}
@ -25,33 +23,48 @@ int main1 ()
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
if (ia[2][6][j] != 5)
abort();
}
}
}
__attribute__ ((noinline))
void main2 ()
{
int i, j;
int ia[8][5][N+2];
/* Multidimensional array. Aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
ia[3][6][j+2] = 5;
}
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 2; j < 6; j++)
for (j = 2; j < N+2; j++)
{
if (ia[3][6][j] != 5)
abort();
}
}
}
__attribute__ ((noinline))
void main3 ()
{
int i, j;
int ic[16][16][5][N+2];
/* Multidimensional array. Not aligned. */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
ic[2][1][6][j+1] = 5;
}
@ -60,22 +73,24 @@ int main1 ()
/* check results: */
for (i = 0; i < 16; i++)
{
for (j = 0; j < 4; j++)
for (j = 0; j < N; j++)
{
if (ic[2][1][6][j+1] != 5)
abort();
}
}
return 0;
}
int main (void)
{
check_vect ();
return main1 ();
main1 ();
main2 ();
main3 ();
return 0;
}
/* { dg-final { scan-tree-dump-times "vectorized 3 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 3 "vect" } } */
/* { dg-final { cleanup-tree-dump "vect" } } */

View file

@ -3,7 +3,7 @@
#include <stdarg.h>
#include "tree-vect.h"
#define N 8
#define N 12
#define OFF 4
/* Check handling of accesses for which the "initial condition" -

View file

@ -1,6 +1,6 @@
/* { dg-do compile } */
#define N 40
#define N 64
signed short image[N][N] __attribute__ ((__aligned__(16)));
signed short block[N][N] __attribute__ ((__aligned__(16)));
signed short out[N] __attribute__ ((__aligned__(16)));

View file

@ -1,6 +1,6 @@
/* { dg-do compile } */
#define N 40
#define N 64
signed short image[N][N] __attribute__ ((__aligned__(16)));
signed short block[N][N] __attribute__ ((__aligned__(16)));

View file

@ -20,10 +20,10 @@ int main1 ()
int i, j;
for (i = 0; i < 8; i++)
for (i = 0; i < N; i++)
{
s = 0;
for (j=0; j<8; j+=4)
for (j = 0; j < N; j += 4)
s += C[j];
A[i] = s;
}
@ -49,10 +49,10 @@ int main ()
main1();
/* check results: */
for (i = 0; i < 8; i++)
for (i = 0; i < N; i++)
{
s = 0;
for (j=0; j<8; j+=4)
for (j = 0; j < N; j += 4)
s += C[j];
if (A[i] != s)
abort ();

View file

@ -4,12 +4,14 @@
#include <stdarg.h>
#include "tree-vect.h"
#define N 4
#define N 8
int main ()
{
unsigned int A[N] = { 0x08000000, 0x08000001, 0x0ff0000ff, 0xf0000001 };
unsigned int B[N] = { 0x01000000, 0x01000000, 0x01fe0001f, 0x1e000000 };
unsigned int A[N] = { 0x08000000, 0x08000001, 0x0ff0000ff, 0xf0000001,
0x08000000, 0x08000001, 0x0ff0000ff, 0xf0000001 };
unsigned int B[N] = { 0x01000000, 0x01000000, 0x01fe0001f, 0x1e000000,
0x01000000, 0x01000000, 0x01fe0001f, 0x1e000000 };
int i;
check_vect ();

View file

@ -4,14 +4,14 @@
extern char lanip[3][40];
typedef struct
{
char *t[4];
char *t[8];
}tx_typ;
int set_names (void)
{
static tx_typ tt1;
int ln;
for (ln = 0; ln < 4; ln++)
for (ln = 0; ln < 8; ln++)
tt1.t[ln] = lanip[1];
}

View file

@ -1018,7 +1018,7 @@ basic_block *blocks_in_phiopt_order (void);
void tree_ssa_lim (void);
unsigned int tree_ssa_unswitch_loops (void);
unsigned int canonicalize_induction_variables (void);
unsigned int tree_unroll_loops_completely (bool);
unsigned int tree_unroll_loops_completely (bool, bool);
unsigned int tree_ssa_prefetch_arrays (void);
unsigned int remove_empty_loops (void);
void tree_ssa_iv_optimize (void);

View file

@ -290,6 +290,7 @@ extern struct gimple_opt_pass pass_if_conversion;
extern struct gimple_opt_pass pass_loop_distribution;
extern struct gimple_opt_pass pass_vectorize;
extern struct gimple_opt_pass pass_complete_unroll;
extern struct gimple_opt_pass pass_complete_unrolli;
extern struct gimple_opt_pass pass_parallelize_loops;
extern struct gimple_opt_pass pass_loop_prefetch;
extern struct gimple_opt_pass pass_iv_optimize;

View file

@ -187,23 +187,20 @@ try_unroll_loop_completely (struct loop *loop,
> (unsigned) PARAM_VALUE (PARAM_MAX_COMPLETELY_PEELED_INSNS))
return false;
if (ul == UL_NO_GROWTH)
unr_insns = estimated_unrolled_size (ninsns, n_unroll);
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " Loop size: %d\n", (int) ninsns);
fprintf (dump_file, " Estimated size after unrolling: %d\n",
(int) unr_insns);
}
if (ul == UL_NO_GROWTH
&& unr_insns > ninsns)
{
unr_insns = estimated_unrolled_size (ninsns, n_unroll);
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " Loop size: %d\n", (int) ninsns);
fprintf (dump_file, " Estimated size after unrolling: %d\n",
(int) unr_insns);
}
if (unr_insns > ninsns)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, "Not unrolling loop %d:\n", loop->num);
return false;
}
fprintf (dump_file, "Not unrolling loop %d.\n", loop->num);
return false;
}
}
@ -339,30 +336,45 @@ canonicalize_induction_variables (void)
size of the code does not increase. */
unsigned int
tree_unroll_loops_completely (bool may_increase_size)
tree_unroll_loops_completely (bool may_increase_size, bool unroll_outer)
{
loop_iterator li;
struct loop *loop;
bool changed = false;
bool changed;
enum unroll_level ul;
FOR_EACH_LOOP (li, loop, 0)
do
{
if (may_increase_size && maybe_hot_bb_p (loop->header))
ul = UL_ALL;
else
ul = UL_NO_GROWTH;
changed |= canonicalize_loop_induction_variables (loop,
false, ul,
!flag_tree_loop_ivcanon);
changed = false;
FOR_EACH_LOOP (li, loop, LI_ONLY_INNERMOST)
{
if (may_increase_size && maybe_hot_bb_p (loop->header)
/* Unroll outermost loops only if asked to do so or they do
not cause code growth. */
&& (unroll_outer
|| loop_outer (loop_outer (loop))))
ul = UL_ALL;
else
ul = UL_NO_GROWTH;
changed |= canonicalize_loop_induction_variables
(loop, false, ul, !flag_tree_loop_ivcanon);
}
if (changed)
{
/* This will take care of removing completely unrolled loops
from the loop structures so we can continue unrolling now
innermost loops. */
cleanup_tree_cfg ();
/* Clean up the information about numbers of iterations, since
complete unrolling might have invalidated it. */
scev_reset ();
}
}
while (changed);
/* Clean up the information about numbers of iterations, since complete
unrolling might have invalidated it. */
scev_reset ();
if (changed)
return TODO_cleanup_cfg;
return 0;
}

View file

@ -248,10 +248,15 @@ find_uses_to_rename_use (basic_block bb, tree use, bitmap *use_blocks,
return;
def_loop = def_bb->loop_father;
/* If the definition is not inside loop, it is not interesting. */
/* If the definition is not inside a loop, it is not interesting. */
if (!loop_outer (def_loop))
return;
/* If the use is not outside of the loop it is defined in, it is not
interesting. */
if (flow_bb_inside_loop_p (def_loop, bb))
return;
if (!use_blocks[ver])
use_blocks[ver] = BITMAP_ALLOC (NULL);
bitmap_set_bit (use_blocks[ver], bb->index);
@ -592,7 +597,8 @@ tree_duplicate_loop_to_header_edge (struct loop *loop, edge e,
return false;
#ifdef ENABLE_CHECKING
verify_loop_closed_ssa ();
if (loops_state_satisfies_p (LOOP_CLOSED_SSA))
verify_loop_closed_ssa ();
#endif
first_new_block = last_basic_block;

View file

@ -223,13 +223,16 @@ struct gimple_opt_pass pass_predcom =
static unsigned int
tree_vectorize (void)
{
if (number_of_loops () <= 1)
return 0;
return vectorize_loops ();
}
static bool
gate_tree_vectorize (void)
{
return flag_tree_vectorize && number_of_loops () > 1;
return flag_tree_vectorize;
}
struct gimple_opt_pass pass_vectorize =
@ -466,7 +469,7 @@ tree_complete_unroll (void)
return tree_unroll_loops_completely (flag_unroll_loops
|| flag_peel_loops
|| optimize >= 3);
|| optimize >= 3, true);
}
static bool
@ -495,6 +498,53 @@ struct gimple_opt_pass pass_complete_unroll =
}
};
/* Complete unrolling of inner loops. */
static unsigned int
tree_complete_unroll_inner (void)
{
unsigned ret = 0;
loop_optimizer_init (LOOPS_NORMAL
| LOOPS_HAVE_RECORDED_EXITS);
if (number_of_loops () > 1)
{
scev_initialize ();
ret = tree_unroll_loops_completely (optimize >= 3, false);
free_numbers_of_iterations_estimates ();
scev_finalize ();
}
loop_optimizer_finalize ();
return ret;
}
static bool
gate_tree_complete_unroll_inner (void)
{
return optimize >= 2;
}
struct gimple_opt_pass pass_complete_unrolli =
{
{
GIMPLE_PASS,
"cunrolli", /* name */
gate_tree_complete_unroll_inner, /* gate */
tree_complete_unroll_inner, /* execute */
NULL, /* sub */
NULL, /* next */
0, /* static_pass_number */
TV_COMPLETE_UNROLL, /* tv_id */
PROP_cfg | PROP_ssa, /* properties_required */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
TODO_dump_func | TODO_verify_loops
| TODO_ggc_collect /* todo_flags_finish */
}
};
/* Parallelization. */
static bool