* [PATCH] Propagate out predicate inversions in forwprop
@ 2011-09-07 11:55 Richard Guenther
2011-09-07 19:51 ` Andrew Pinski
0 siblings, 1 reply; 2+ messages in thread
From: Richard Guenther @ 2011-09-07 11:55 UTC (permalink / raw)
To: gcc-patches
This makes sure that we propagate comparisons that we cannot invert
into inverted conditions by swapping edges or the conditional ops.
For the testcase it transforms
<bb 2>:
D.2724_4 = xx_2(D) < xy_3(D);
p_5 = (int) D.2724_4;
D.2725_6 = p_5 == 0;
np_7 = (int) D.2725_6;
if (np_7 != 0)
to
<bb 2>:
if (xx_2(D) < xy_3(D))
while without the patch we would be left with
<bb 2>:
D.2724_4 = xx_2(D) < xy_3(D);
if (D.2724_4 == 0)
Bootstrap and regtest ongoing on x86_64-unknown-linux-gnu.
Richard.
2011-09-07 Richard Guenther <rguenther@suse.de>
* tree-ssa-forwprop.c (forward_propagate_into_gimple_cond):
Canonicalize negated predicates by swapping edges.
(forward_propagate_into_cond): Likewise.
* gcc.dg/tree-ssa/forwprop-16.c: New testcase.
Index: gcc/tree-ssa-forwprop.c
===================================================================
*** gcc/tree-ssa-forwprop.c (revision 178633)
--- gcc/tree-ssa-forwprop.c (working copy)
*************** forward_propagate_into_gimple_cond (gimp
*** 534,539 ****
--- 534,556 ----
return (cfg_changed || is_gimple_min_invariant (tmp)) ? 2 : 1;
}
+ /* Canonicalize _Bool == 0 and _Bool != 1 to _Bool != 0 by swapping edges. */
+ if ((TREE_CODE (TREE_TYPE (rhs1)) == BOOLEAN_TYPE
+ || (INTEGRAL_TYPE_P (TREE_TYPE (rhs1))
+ && TYPE_PRECISION (TREE_TYPE (rhs1)) == 1))
+ && ((code == EQ_EXPR
+ && integer_zerop (rhs2))
+ || (code == NE_EXPR
+ && integer_onep (rhs2))))
+ {
+ basic_block bb = gimple_bb (stmt);
+ gimple_cond_set_code (stmt, NE_EXPR);
+ gimple_cond_set_rhs (stmt, build_zero_cst (TREE_TYPE (rhs1)));
+ EDGE_SUCC (bb, 0)->flags ^= (EDGE_TRUE_VALUE|EDGE_FALSE_VALUE);
+ EDGE_SUCC (bb, 1)->flags ^= (EDGE_TRUE_VALUE|EDGE_FALSE_VALUE);
+ return 1;
+ }
+
return 0;
}
*************** forward_propagate_into_cond (gimple_stmt
*** 548,553 ****
--- 565,571 ----
gimple stmt = gsi_stmt (*gsi_p);
tree tmp = NULL_TREE;
tree cond = gimple_assign_rhs1 (stmt);
+ bool swap = false;
/* We can do tree combining on SSA_NAME and comparison expressions. */
if (COMPARISON_CLASS_P (cond))
*************** forward_propagate_into_cond (gimple_stmt
*** 557,573 ****
TREE_OPERAND (cond, 1));
else if (TREE_CODE (cond) == SSA_NAME)
{
tree name = cond;
gimple def_stmt = get_prop_source_stmt (name, true, NULL);
if (!def_stmt || !can_propagate_from (def_stmt))
return 0;
! if (TREE_CODE_CLASS (gimple_assign_rhs_code (def_stmt)) == tcc_comparison)
tmp = fold_build2_loc (gimple_location (def_stmt),
! gimple_assign_rhs_code (def_stmt),
boolean_type_node,
gimple_assign_rhs1 (def_stmt),
gimple_assign_rhs2 (def_stmt));
}
if (tmp)
--- 575,601 ----
TREE_OPERAND (cond, 1));
else if (TREE_CODE (cond) == SSA_NAME)
{
+ enum tree_code code;
tree name = cond;
gimple def_stmt = get_prop_source_stmt (name, true, NULL);
if (!def_stmt || !can_propagate_from (def_stmt))
return 0;
! code = gimple_assign_rhs_code (def_stmt);
! if (TREE_CODE_CLASS (code) == tcc_comparison)
tmp = fold_build2_loc (gimple_location (def_stmt),
! code,
boolean_type_node,
gimple_assign_rhs1 (def_stmt),
gimple_assign_rhs2 (def_stmt));
+ else if ((code == BIT_NOT_EXPR
+ && TYPE_PRECISION (TREE_TYPE (cond)) == 1)
+ || (code == BIT_XOR_EXPR
+ && integer_onep (gimple_assign_rhs2 (def_stmt))))
+ {
+ tmp = gimple_assign_rhs1 (def_stmt);
+ swap = true;
+ }
}
if (tmp)
*************** forward_propagate_into_cond (gimple_stmt
*** 586,592 ****
else if (integer_zerop (tmp))
gimple_assign_set_rhs_from_tree (gsi_p, gimple_assign_rhs3 (stmt));
else
! gimple_assign_set_rhs1 (stmt, unshare_expr (tmp));
stmt = gsi_stmt (*gsi_p);
update_stmt (stmt);
--- 614,628 ----
else if (integer_zerop (tmp))
gimple_assign_set_rhs_from_tree (gsi_p, gimple_assign_rhs3 (stmt));
else
! {
! gimple_assign_set_rhs1 (stmt, unshare_expr (tmp));
! if (swap)
! {
! tree t = gimple_assign_rhs2 (stmt);
! gimple_assign_set_rhs2 (stmt, gimple_assign_rhs3 (stmt));
! gimple_assign_set_rhs3 (stmt, t);
! }
! }
stmt = gsi_stmt (*gsi_p);
update_stmt (stmt);
Index: gcc/testsuite/gcc.dg/tree-ssa/forwprop-16.c
===================================================================
*** gcc/testsuite/gcc.dg/tree-ssa/forwprop-16.c (revision 0)
--- gcc/testsuite/gcc.dg/tree-ssa/forwprop-16.c (revision 0)
***************
*** 0 ****
--- 1,14 ----
+ /* { dg-do compile } */
+ /* { dg-options "-O -fdump-tree-forwprop1" } */
+
+ int foo (double xx, double xy)
+ {
+ int p = xx < xy;
+ int np = !p;
+ if (np)
+ return 5;
+ return 2;
+ }
+
+ /* { dg-final { scan-tree-dump "if \\\(x" "forwprop1" } } */
+ /* { dg-final { cleanup-tree-dump "forwprop1" } } */
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Propagate out predicate inversions in forwprop
2011-09-07 11:55 [PATCH] Propagate out predicate inversions in forwprop Richard Guenther
@ 2011-09-07 19:51 ` Andrew Pinski
0 siblings, 0 replies; 2+ messages in thread
From: Andrew Pinski @ 2011-09-07 19:51 UTC (permalink / raw)
To: Richard Guenther; +Cc: gcc-patches
On Wed, Sep 7, 2011 at 4:46 AM, Richard Guenther <rguenther@suse.de> wrote:
>
> This makes sure that we propagate comparisons that we cannot invert
> into inverted conditions by swapping edges or the conditional ops.
> For the testcase it transforms
>
> <bb 2>:
> D.2724_4 = xx_2(D) < xy_3(D);
> p_5 = (int) D.2724_4;
> D.2725_6 = p_5 == 0;
> np_7 = (int) D.2725_6;
> if (np_7 != 0)
>
> to
>
> <bb 2>:
> if (xx_2(D) < xy_3(D))
>
> while without the patch we would be left with
>
> <bb 2>:
> D.2724_4 = xx_2(D) < xy_3(D);
> if (D.2724_4 == 0)
This sounds like PR 15558.
Thanks,
Andrew Pinski
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2011-09-07 19:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-09-07 11:55 [PATCH] Propagate out predicate inversions in forwprop Richard Guenther
2011-09-07 19:51 ` Andrew Pinski
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).