[PATCH] match.pd: Fix up __builtin_mul_overflow_p signed type optimization [PR105984]

[PATCH] match.pd: Fix up __builtin_mul_overflow_p signed type optimization [PR105984]

[Jakub Jelinek]

Hi!

Earlier in the simplification pattern, we require that @0 has compatible
type to the type of IMAGPART_EXPR, but for @1 which is a non-zero constant
all we require is that it the constant fits into that type.
Later the code checks if the constant is negative, because when min / max
values are divided by negative divisor, lo will be higher than hi.
In the following testcase, @1 has unsigned char type, while @0 has
int type, so @1 which is 254 is wi::neg_p and we were swapping lo and hi,
even when @1 cast to int isn't negative.

We could use tree_int_cst_sgn (@1) < 0 as the check instead and it would
work both for narrower types of @1 and even same or wider ones, but
I've noticed we probably don't want to call fold_convert (TREE_TYPE (@0), @1)
twice and when we save that result in a temporary, we can just use wi::neg_p
on that temporary.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2022-06-16  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/105984
	* match.pd (__builtin_mul_overflow_p (x, cst, (stype) 0) ->
	x > stype_max / cst || x < stype_min / cst): fold_convert @1
	to TREE_TYPE (@0) just once and test for negative divisor
	also on that folded constant instead of on @1.

	* gcc.c-torture/execute/pr105984.c: New test.

--- gcc/match.pd.jj	2022-06-15 10:43:46.000000000 +0200
+++ gcc/match.pd	2022-06-15 12:52:04.640981511 +0200
@@ -5995,16 +5995,15 @@ (define_operator_list SYNC_FETCH_AND_AND
       (convert (eq @0 { TYPE_MIN_VALUE (TREE_TYPE (@0)); }))
       (with
        {
+	 tree div = fold_convert (TREE_TYPE (@0), @1);
 	 tree lo = int_const_binop (TRUNC_DIV_EXPR,
-				    TYPE_MIN_VALUE (TREE_TYPE (@0)),
-				    fold_convert (TREE_TYPE (@0), @1));
+				    TYPE_MIN_VALUE (TREE_TYPE (@0)), div);
 	 tree hi = int_const_binop (TRUNC_DIV_EXPR,
-				    TYPE_MAX_VALUE (TREE_TYPE (@0)),
-				    fold_convert (TREE_TYPE (@0), @1));
+				    TYPE_MAX_VALUE (TREE_TYPE (@0)), div);
 	 tree etype = range_check_type (TREE_TYPE (@0));
 	 if (etype)
 	   {
-	     if (wi::neg_p (wi::to_wide (@1)))
+	     if (wi::neg_p (wi::to_wide (div)))
 	       std::swap (lo, hi);
 	     lo = fold_convert (etype, lo);
 	     hi = fold_convert (etype, hi);
--- gcc/testsuite/gcc.c-torture/execute/pr105984.c.jj	2022-06-15 13:11:30.695091900 +0200
+++ gcc/testsuite/gcc.c-torture/execute/pr105984.c	2022-06-15 13:10:56.267443900 +0200
@@ -0,0 +1,19 @@
+/* PR tree-optimization/105984 */
+
+unsigned long long g;
+
+static inline unsigned long long
+foo (unsigned char c)
+{
+  g -= __builtin_mul_overflow_p (4, (unsigned char) ~c, 0);
+  return g;
+}
+
+int
+main ()
+{
+  unsigned long long x = foo (1);
+  if (x != 0)
+    __builtin_abort ();
+  return 0;
+}

	Jakub

Re: [PATCH] match.pd: Fix up __builtin_mul_overflow_p signed type optimization [PR105984]

[Richard Biener]

> Am 16.06.2022 um 11:10 schrieb Jakub Jelinek <jakub@redhat.com>:
> 
> Hi!
> 
> Earlier in the simplification pattern, we require that @0 has compatible
> type to the type of IMAGPART_EXPR, but for @1 which is a non-zero constant
> all we require is that it the constant fits into that type.
> Later the code checks if the constant is negative, because when min / max
> values are divided by negative divisor, lo will be higher than hi.
> In the following testcase, @1 has unsigned char type, while @0 has
> int type, so @1 which is 254 is wi::neg_p and we were swapping lo and hi,
> even when @1 cast to int isn't negative.
> 
> We could use tree_int_cst_sgn (@1) < 0 as the check instead and it would
> work both for narrower types of @1 and even same or wider ones, but
> I've noticed we probably don't want to call fold_convert (TREE_TYPE (@0), @1)
> twice and when we save that result in a temporary, we can just use wi::neg_p
> on that temporary.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 

Ok.

Thanks,
Richard 
> 2022-06-16  Jakub Jelinek  <jakub@redhat.com>
> 
>    PR tree-optimization/105984
>    * match.pd (__builtin_mul_overflow_p (x, cst, (stype) 0) ->
>    x > stype_max / cst || x < stype_min / cst): fold_convert @1
>    to TREE_TYPE (@0) just once and test for negative divisor
>    also on that folded constant instead of on @1.
> 
>    * gcc.c-torture/execute/pr105984.c: New test.
> 
> --- gcc/match.pd.jj    2022-06-15 10:43:46.000000000 +0200
> +++ gcc/match.pd    2022-06-15 12:52:04.640981511 +0200
> @@ -5995,16 +5995,15 @@ (define_operator_list SYNC_FETCH_AND_AND
>       (convert (eq @0 { TYPE_MIN_VALUE (TREE_TYPE (@0)); }))
>       (with
>        {
> +     tree div = fold_convert (TREE_TYPE (@0), @1);
>     tree lo = int_const_binop (TRUNC_DIV_EXPR,
> -                    TYPE_MIN_VALUE (TREE_TYPE (@0)),
> -                    fold_convert (TREE_TYPE (@0), @1));
> +                    TYPE_MIN_VALUE (TREE_TYPE (@0)), div);
>     tree hi = int_const_binop (TRUNC_DIV_EXPR,
> -                    TYPE_MAX_VALUE (TREE_TYPE (@0)),
> -                    fold_convert (TREE_TYPE (@0), @1));
> +                    TYPE_MAX_VALUE (TREE_TYPE (@0)), div);
>     tree etype = range_check_type (TREE_TYPE (@0));
>     if (etype)
>       {
> -         if (wi::neg_p (wi::to_wide (@1)))
> +         if (wi::neg_p (wi::to_wide (div)))
>           std::swap (lo, hi);
>         lo = fold_convert (etype, lo);
>         hi = fold_convert (etype, hi);
> --- gcc/testsuite/gcc.c-torture/execute/pr105984.c.jj    2022-06-15 13:11:30.695091900 +0200
> +++ gcc/testsuite/gcc.c-torture/execute/pr105984.c    2022-06-15 13:10:56.267443900 +0200
> @@ -0,0 +1,19 @@
> +/* PR tree-optimization/105984 */
> +
> +unsigned long long g;
> +
> +static inline unsigned long long
> +foo (unsigned char c)
> +{
> +  g -= __builtin_mul_overflow_p (4, (unsigned char) ~c, 0);
> +  return g;
> +}
> +
> +int
> +main ()
> +{
> +  unsigned long long x = foo (1);
> +  if (x != 0)
> +    __builtin_abort ();
> +  return 0;
> +}
> 
>    Jakub
>