backport: re PR tree-optimization/81388 (Incorrect code generation with -O1)

Backport from 2017-07-20 trunk r250384.

	PR tree-optimization/81388
	Revert r238585:
	2016-07-21  Bin Cheng  <bin.cheng@arm.com>

	* tree-ssa-loop-niter.c (number_of_iterations_lt_to_ne): Clean up
	by removing computation of may_be_zero.

	gcc/testsuite
	PR tree-optimization/81388
	* gcc.dg/tree-ssa/pr81388-1.c: New test.
	* gcc.dg/tree-ssa/pr81388-2.c: New test.

From-SVN: r250497
This commit is contained in:
Bin Cheng 2017-07-25 08:56:26 +00:00 committed by Bin Cheng
parent cc30cfd553
commit 16ac9369b0
5 changed files with 124 additions and 28 deletions

View File

@ -1,3 +1,14 @@
2017-07-25 Bin Cheng <bin.cheng@arm.com>
Backport from 2017-07-20 trunk r250384.
PR tree-optimization/81388
Revert r238585:
2016-07-21 Bin Cheng <bin.cheng@arm.com>
* tree-ssa-loop-niter.c (number_of_iterations_lt_to_ne): Clean up
by removing computation of may_be_zero.
2017-07-23 Uros Bizjak <ubizjak@gmail.com> 2017-07-23 Uros Bizjak <ubizjak@gmail.com>
PR target/80569 PR target/80569

View File

@ -1,3 +1,11 @@
2017-07-25 Bin Cheng <bin.cheng@arm.com>
Backport from 2017-07-20 trunk r250384.
PR tree-optimization/81388
* gcc.dg/tree-ssa/pr81388-1.c: New test.
* gcc.dg/tree-ssa/pr81388-2.c: New test.
2017-07-23 Uros Bizjak <ubizjak@gmail.com> 2017-07-23 Uros Bizjak <ubizjak@gmail.com>
PR target/80569 PR target/80569

View File

@ -0,0 +1,14 @@
/* { dg-do compile } */
/* { dg-options "-O2 -fno-strict-overflow -fdump-tree-ivcanon-details" } */
void bar();
void foo(char *dst)
{
char *const end = dst;
do {
bar();
dst += 2;
} while (dst < end);
}
/* { dg-final { scan-tree-dump " zero if " "ivcanon" } } */

View File

@ -0,0 +1,14 @@
/* { dg-do compile } */
/* { dg-options "-O2 -fdump-tree-ivcanon-details" } */
void bar();
void foo(unsigned dst)
{
unsigned end = dst;
do {
bar();
dst += 2;
} while (dst < end);
}
/* { dg-final { scan-tree-dump " zero if " "ivcanon" } } */

View File

@ -1142,8 +1142,12 @@ number_of_iterations_lt_to_ne (tree type, affine_iv *iv0, affine_iv *iv1,
tree niter_type = TREE_TYPE (step); tree niter_type = TREE_TYPE (step);
tree mod = fold_build2 (FLOOR_MOD_EXPR, niter_type, *delta, step); tree mod = fold_build2 (FLOOR_MOD_EXPR, niter_type, *delta, step);
tree tmod; tree tmod;
tree assumption = boolean_true_node, bound; mpz_t mmod;
tree type1 = (POINTER_TYPE_P (type)) ? sizetype : type; tree assumption = boolean_true_node, bound, noloop;
bool ret = false, fv_comp_no_overflow;
tree type1 = type;
if (POINTER_TYPE_P (type))
type1 = sizetype;
if (TREE_CODE (mod) != INTEGER_CST) if (TREE_CODE (mod) != INTEGER_CST)
return false; return false;
@ -1151,51 +1155,96 @@ number_of_iterations_lt_to_ne (tree type, affine_iv *iv0, affine_iv *iv1,
mod = fold_build2 (MINUS_EXPR, niter_type, step, mod); mod = fold_build2 (MINUS_EXPR, niter_type, step, mod);
tmod = fold_convert (type1, mod); tmod = fold_convert (type1, mod);
mpz_init (mmod);
wi::to_mpz (mod, mmod, UNSIGNED);
mpz_neg (mmod, mmod);
/* If the induction variable does not overflow and the exit is taken, /* If the induction variable does not overflow and the exit is taken,
then the computation of the final value does not overflow. There then the computation of the final value does not overflow. This is
are three cases: also obviously the case if the new final value is equal to the
1) The case if the new final value is equal to the current one. current one. Finally, we postulate this for pointer type variables,
2) Induction varaible has pointer type, as the code cannot rely as the code cannot rely on the object to that the pointer points being
on the object to that the pointer points being placed at the placed at the end of the address space (and more pragmatically,
end of the address space (and more pragmatically, TYPE_{MIN,MAX}_VALUE is not defined for pointers). */
TYPE_{MIN,MAX}_VALUE is not defined for pointers). if (integer_zerop (mod) || POINTER_TYPE_P (type))
3) EXIT_MUST_BE_TAKEN is true, note it implies that the induction fv_comp_no_overflow = true;
variable does not overflow. */ else if (!exit_must_be_taken)
if (!integer_zerop (mod) && !POINTER_TYPE_P (type) && !exit_must_be_taken) fv_comp_no_overflow = false;
else
fv_comp_no_overflow =
(iv0->no_overflow && integer_nonzerop (iv0->step))
|| (iv1->no_overflow && integer_nonzerop (iv1->step));
if (integer_nonzerop (iv0->step))
{ {
if (integer_nonzerop (iv0->step)) /* The final value of the iv is iv1->base + MOD, assuming that this
computation does not overflow, and that
iv0->base <= iv1->base + MOD. */
if (!fv_comp_no_overflow)
{ {
/* The final value of the iv is iv1->base + MOD, assuming
that this computation does not overflow, and that
iv0->base <= iv1->base + MOD. */
bound = fold_build2 (MINUS_EXPR, type1, bound = fold_build2 (MINUS_EXPR, type1,
TYPE_MAX_VALUE (type1), tmod); TYPE_MAX_VALUE (type1), tmod);
assumption = fold_build2 (LE_EXPR, boolean_type_node, assumption = fold_build2 (LE_EXPR, boolean_type_node,
iv1->base, bound); iv1->base, bound);
if (integer_zerop (assumption))
goto end;
} }
if (mpz_cmp (mmod, bnds->below) < 0)
noloop = boolean_false_node;
else if (POINTER_TYPE_P (type))
noloop = fold_build2 (GT_EXPR, boolean_type_node,
iv0->base,
fold_build_pointer_plus (iv1->base, tmod));
else else
noloop = fold_build2 (GT_EXPR, boolean_type_node,
iv0->base,
fold_build2 (PLUS_EXPR, type1,
iv1->base, tmod));
}
else
{
/* The final value of the iv is iv0->base - MOD, assuming that this
computation does not overflow, and that
iv0->base - MOD <= iv1->base. */
if (!fv_comp_no_overflow)
{ {
/* The final value of the iv is iv0->base - MOD, assuming
that this computation does not overflow, and that
iv0->base - MOD <= iv1->base. */
bound = fold_build2 (PLUS_EXPR, type1, bound = fold_build2 (PLUS_EXPR, type1,
TYPE_MIN_VALUE (type1), tmod); TYPE_MIN_VALUE (type1), tmod);
assumption = fold_build2 (GE_EXPR, boolean_type_node, assumption = fold_build2 (GE_EXPR, boolean_type_node,
iv0->base, bound); iv0->base, bound);
if (integer_zerop (assumption))
goto end;
} }
if (integer_zerop (assumption)) if (mpz_cmp (mmod, bnds->below) < 0)
return false; noloop = boolean_false_node;
else if (!integer_nonzerop (assumption)) else if (POINTER_TYPE_P (type))
niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node, noloop = fold_build2 (GT_EXPR, boolean_type_node,
niter->assumptions, assumption); fold_build_pointer_plus (iv0->base,
fold_build1 (NEGATE_EXPR,
type1, tmod)),
iv1->base);
else
noloop = fold_build2 (GT_EXPR, boolean_type_node,
fold_build2 (MINUS_EXPR, type1,
iv0->base, tmod),
iv1->base);
} }
/* Since we are transforming LT to NE and DELTA is constant, there if (!integer_nonzerop (assumption))
is no need to compute may_be_zero because this loop must roll. */ niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
niter->assumptions,
assumption);
if (!integer_zerop (noloop))
niter->may_be_zero = fold_build2 (TRUTH_OR_EXPR, boolean_type_node,
niter->may_be_zero,
noloop);
bounds_add (bnds, wi::to_widest (mod), type); bounds_add (bnds, wi::to_widest (mod), type);
*delta = fold_build2 (PLUS_EXPR, niter_type, *delta, mod); *delta = fold_build2 (PLUS_EXPR, niter_type, *delta, mod);
return true;
ret = true;
end:
mpz_clear (mmod);
return ret;
} }
/* Add assertions to NITER that ensure that the control variable of the loop /* Add assertions to NITER that ensure that the control variable of the loop