diff mbox series

match.pd, v2: Optimize a * !a to 0 [PR114009]

Message ID ZehsBg/ts9JvUXDl@tucnak
State New
Headers show
Series match.pd, v2: Optimize a * !a to 0 [PR114009] | expand

Commit Message

Jakub Jelinek March 6, 2024, 1:13 p.m. UTC
On Wed, Mar 06, 2024 at 11:45:42AM +0100, Richard Biener wrote:
> OK, though feel free to add ARG_UNUSED to 'captures' as well.

Ok, done below.

> I think the INTEGRAL_TYPE_P should be redundant - the pattern
> should work for vectors and complex as well (with integer components
> which integer_zerop constrains).

Guess you're right.  I think it is correct as is for both complex and
int vector, while float etc. shouldn't appear there because of the
integer_zerop and pointer because we don't do multiplication on pointer
types.

That said, it won't really actually match the vector int case actually
emitted, where we have a VEC_COND_EXPR, added those as new patterns.
And it doesn't match what is emitted for _Complex, though I think it
isn't that important to simplify that.
_Complex int
freddy (_Complex int x)
{
  return x * !x;
}

_Complex int
garply (_Complex int x)
{
  return x * (x == 0);
}

results in original in:
  return COMPLEX_EXPR <REALPART_EXPR <SAVE_EXPR <x>> == 0 && IMAGPART_EXPR <SAVE_EXPR <x>> == 0, 0> * x;
and
  return (x == __complex__ (0, 0) ? __complex__ (1, 0) : __complex__ (0, 0)) * x;
but after gimplification it is already something that is really
too hard to optimize.

So, is the following still ok if it passes bootstrap/regtest?

2024-03-06  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/114009
	* genmatch.cc (decision_tree::gen): Emit ARG_UNUSED for captures
	argument even for GENERIC, not just for GIMPLE.
	* match.pd (a * !a -> 0): New simplifications.

	* gcc.dg/tree-ssa/pr114009.c: New test.



	Jakub

Comments

Richard Biener March 6, 2024, 1:32 p.m. UTC | #1
On Wed, 6 Mar 2024, Jakub Jelinek wrote:

> On Wed, Mar 06, 2024 at 11:45:42AM +0100, Richard Biener wrote:
> > OK, though feel free to add ARG_UNUSED to 'captures' as well.
> 
> Ok, done below.
> 
> > I think the INTEGRAL_TYPE_P should be redundant - the pattern
> > should work for vectors and complex as well (with integer components
> > which integer_zerop constrains).
> 
> Guess you're right.  I think it is correct as is for both complex and
> int vector, while float etc. shouldn't appear there because of the
> integer_zerop and pointer because we don't do multiplication on pointer
> types.
> 
> That said, it won't really actually match the vector int case actually
> emitted, where we have a VEC_COND_EXPR, added those as new patterns.
> And it doesn't match what is emitted for _Complex, though I think it
> isn't that important to simplify that.
> _Complex int
> freddy (_Complex int x)
> {
>   return x * !x;
> }
> 
> _Complex int
> garply (_Complex int x)
> {
>   return x * (x == 0);
> }
> 
> results in original in:
>   return COMPLEX_EXPR <REALPART_EXPR <SAVE_EXPR <x>> == 0 && IMAGPART_EXPR <SAVE_EXPR <x>> == 0, 0> * x;
> and
>   return (x == __complex__ (0, 0) ? __complex__ (1, 0) : __complex__ (0, 0)) * x;
> but after gimplification it is already something that is really
> too hard to optimize.
> 
> So, is the following still ok if it passes bootstrap/regtest?

OK.

Thanks,
Richard.

> 2024-03-06  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR tree-optimization/114009
> 	* genmatch.cc (decision_tree::gen): Emit ARG_UNUSED for captures
> 	argument even for GENERIC, not just for GIMPLE.
> 	* match.pd (a * !a -> 0): New simplifications.
> 
> 	* gcc.dg/tree-ssa/pr114009.c: New test.
> 
> --- gcc/genmatch.cc.jj	2024-02-26 10:09:16.594613314 +0100
> +++ gcc/genmatch.cc	2024-03-06 13:47:35.001458638 +0100
> @@ -4071,7 +4071,7 @@ decision_tree::gen (vec <FILE *> &files,
>  	  for (unsigned i = 0;
>  	       i < as_a <expr *>(s->s->s->match)->ops.length (); ++i)
>  	    fp_decl (f, " tree ARG_UNUSED (_p%d),", i);
> -	  fp_decl (f, " tree *captures");
> +	  fp_decl (f, " tree *ARG_UNUSED (captures)");
>  	}
>        for (unsigned i = 0; i < s->s->s->for_subst_vec.length (); ++i)
>  	{
> --- gcc/match.pd.jj	2024-03-06 09:34:36.608271525 +0100
> +++ gcc/match.pd	2024-03-06 13:58:27.400588289 +0100
> @@ -1219,6 +1219,17 @@ (define_operator_list SYNC_FETCH_AND_AND
>         && tree_nop_conversion_p (type, TREE_TYPE (@1)))
>     (lshift @0 @2)))
>  
> +/* Fold a * !a into 0.  */
> +(simplify
> + (mult:c @0 (convert? (eq @0 integer_zerop)))
> +  { build_zero_cst (type); })
> +(simplify
> + (mult:c @0 (vec_cond (eq @0 integer_zerop) @1 integer_zerop))
> +  { build_zero_cst (type); })
> +(simplify
> + (mult:c @0 (vec_cond (ne @0 integer_zerop) integer_zerop @1))
> +  { build_zero_cst (type); })
> +
>  /* Shifts by precision or greater result in zero.  */
>  (for shift (lshift rshift)
>   (simplify
> --- gcc/testsuite/gcc.dg/tree-ssa/pr114009.c.jj	2024-03-06 13:44:33.481924326 +0100
> +++ gcc/testsuite/gcc.dg/tree-ssa/pr114009.c	2024-03-06 14:04:09.724918390 +0100
> @@ -0,0 +1,33 @@
> +/* PR tree-optimization/114009 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -Wno-psabi -fdump-tree-forwprop1" } */
> +/* { dg-final { scan-tree-dump-times "  return 0;" 3 "forwprop1" } } */
> +/* { dg-final { scan-tree-dump-times "  (?:return|<retval> =) { 0, 0, 0, 0 };" 1 "forwprop1" } } */
> +
> +int
> +foo (int x)
> +{
> +  x = (x / 2) * 2;
> +  return (!x) * x;
> +}
> +
> +int
> +bar (int x, int y)
> +{
> +  (void) x;
> +  return y * !y;
> +}
> +
> +unsigned long long
> +baz (unsigned long long x)
> +{
> +  return (!x) * x;
> +}
> +
> +typedef int V __attribute__((vector_size (4 * sizeof (int))));
> +
> +V
> +qux (V x)
> +{
> +  return x * (x == 0);
> +}
> 
> 
> 	Jakub
> 
>
diff mbox series

Patch

--- gcc/genmatch.cc.jj	2024-02-26 10:09:16.594613314 +0100
+++ gcc/genmatch.cc	2024-03-06 13:47:35.001458638 +0100
@@ -4071,7 +4071,7 @@  decision_tree::gen (vec <FILE *> &files,
 	  for (unsigned i = 0;
 	       i < as_a <expr *>(s->s->s->match)->ops.length (); ++i)
 	    fp_decl (f, " tree ARG_UNUSED (_p%d),", i);
-	  fp_decl (f, " tree *captures");
+	  fp_decl (f, " tree *ARG_UNUSED (captures)");
 	}
       for (unsigned i = 0; i < s->s->s->for_subst_vec.length (); ++i)
 	{
--- gcc/match.pd.jj	2024-03-06 09:34:36.608271525 +0100
+++ gcc/match.pd	2024-03-06 13:58:27.400588289 +0100
@@ -1219,6 +1219,17 @@  (define_operator_list SYNC_FETCH_AND_AND
        && tree_nop_conversion_p (type, TREE_TYPE (@1)))
    (lshift @0 @2)))
 
+/* Fold a * !a into 0.  */
+(simplify
+ (mult:c @0 (convert? (eq @0 integer_zerop)))
+  { build_zero_cst (type); })
+(simplify
+ (mult:c @0 (vec_cond (eq @0 integer_zerop) @1 integer_zerop))
+  { build_zero_cst (type); })
+(simplify
+ (mult:c @0 (vec_cond (ne @0 integer_zerop) integer_zerop @1))
+  { build_zero_cst (type); })
+
 /* Shifts by precision or greater result in zero.  */
 (for shift (lshift rshift)
  (simplify
--- gcc/testsuite/gcc.dg/tree-ssa/pr114009.c.jj	2024-03-06 13:44:33.481924326 +0100
+++ gcc/testsuite/gcc.dg/tree-ssa/pr114009.c	2024-03-06 14:04:09.724918390 +0100
@@ -0,0 +1,33 @@ 
+/* PR tree-optimization/114009 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -Wno-psabi -fdump-tree-forwprop1" } */
+/* { dg-final { scan-tree-dump-times "  return 0;" 3 "forwprop1" } } */
+/* { dg-final { scan-tree-dump-times "  (?:return|<retval> =) { 0, 0, 0, 0 };" 1 "forwprop1" } } */
+
+int
+foo (int x)
+{
+  x = (x / 2) * 2;
+  return (!x) * x;
+}
+
+int
+bar (int x, int y)
+{
+  (void) x;
+  return y * !y;
+}
+
+unsigned long long
+baz (unsigned long long x)
+{
+  return (!x) * x;
+}
+
+typedef int V __attribute__((vector_size (4 * sizeof (int))));
+
+V
+qux (V x)
+{
+  return x * (x == 0);
+}