[PATCH] match.pd: Simplify (t * u) / (t * v) [PR112994]

[PATCH] match.pd: Simplify (t * u) / (t * v) [PR112994]

[Jakub Jelinek]

Hi!

On top of the previously posted patch, this simplifies say (x * 16) / (x * 4)
into 4.  Unlike the previous pattern, this is something we didn't fold
previously on GENERIC, so I think it shouldn't be all wrapped with #if
GIMPLE.  The question whether there should be fold_overflow_warning for the
TYPE_OVERFLOW_UNDEFINED case remains.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk
(with or without fold_overflow_warning)?

2023-12-13  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/112994
	* match.pd ((t * u) / (t * v) -> (u / v)): New simplification.

	* gcc.dg/tree-ssa/pr112994-2.c: New test.

--- gcc/match.pd.jj	2023-12-13 18:43:51.277839661 +0100
+++ gcc/match.pd	2023-12-13 18:43:45.891913683 +0100
@@ -962,6 +960,23 @@ (define_operator_list SYNC_FETCH_AND_AND
 	  && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr1))
       (mult @0 (div! @1 @2))))
 #endif
+   )))
+ /* Simplify (t * u) / (t * v) -> (u / v) if u is multiple of v.  */
+ (simplify
+  (div (mult @0 INTEGER_CST@1) (mult @0 INTEGER_CST@2))
+  (if (INTEGRAL_TYPE_P (type)
+       && wi::multiple_of_p (wi::to_widest (@1), wi::to_widest (@2), SIGNED))
+   (if (TYPE_OVERFLOW_UNDEFINED (type) && !TYPE_OVERFLOW_SANITIZED (type))
+    (div @1 @2)
+#if GIMPLE
+    (with {value_range vr0, vr1, vr2;}
+     (if (get_range_query (cfun)->range_of_expr (vr0, @0)
+	  && get_range_query (cfun)->range_of_expr (vr1, @1)
+	  && get_range_query (cfun)->range_of_expr (vr2, @2)
+	  && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr1)
+	  && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr2))
+      (div @1 @2)))
+#endif
    ))))
 
 #if GIMPLE
--- gcc/testsuite/gcc.dg/tree-ssa/pr112994-2.c.jj	2023-12-13 19:07:20.882475735 +0100
+++ gcc/testsuite/gcc.dg/tree-ssa/pr112994-2.c	2023-12-13 19:07:02.597726855 +0100
@@ -0,0 +1,15 @@
+/* PR tree-optimization/112994 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+/* { dg-final { scan-tree-dump-times "return 2;" 3 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "return 7;" 3 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "return -7;" 2 "optimized" } } */
+
+int f1 (int x) { return (x * 4) / (x * 2); }
+int f2 (int x) { return (x * 56) / (x * 8); }
+int f3 (int x) { return (x * 56) / (x * -8); }
+int f4 (int x) { int y = x * 4; return y / (x * 2); }
+int f5 (int x) { int y = x * 56; return y / (x * 8); }
+int f6 (int x) { int y = x * 56; return y / (x * -8); }
+unsigned f7 (unsigned x) { if (x > ~0U / 4) __builtin_unreachable (); unsigned y = x * 4; return y / (x * 2); }
+unsigned f8 (unsigned x) { if (x > ~0U / 56) __builtin_unreachable (); unsigned y = x * 56; return y / (x * 8); }

	Jakub

Re: [PATCH] match.pd: Simplify (t * u) / (t * v) [PR112994]

[Richard Biener]

> Am 14.12.2023 um 08:37 schrieb Jakub Jelinek <jakub@redhat.com>:
> 
> Hi!
> 
> On top of the previously posted patch, this simplifies say (x * 16) / (x * 4)
> into 4.  Unlike the previous pattern, this is something we didn't fold
> previously on GENERIC, so I think it shouldn't be all wrapped with #if
> GIMPLE.  The question whether there should be fold_overflow_warning for the
> TYPE_OVERFLOW_UNDEFINED case remains.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk
> (with or without fold_overflow_warning)?

Ok without the warning.

Richard 

> 2023-12-13  Jakub Jelinek  <jakub@redhat.com>
> 
>    PR tree-optimization/112994
>    * match.pd ((t * u) / (t * v) -> (u / v)): New simplification.
> 
>    * gcc.dg/tree-ssa/pr112994-2.c: New test.
> 
> --- gcc/match.pd.jj    2023-12-13 18:43:51.277839661 +0100
> +++ gcc/match.pd    2023-12-13 18:43:45.891913683 +0100
> @@ -962,6 +960,23 @@ (define_operator_list SYNC_FETCH_AND_AND
>      && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr1))
>       (mult @0 (div! @1 @2))))
> #endif
> +   )))
> + /* Simplify (t * u) / (t * v) -> (u / v) if u is multiple of v.  */
> + (simplify
> +  (div (mult @0 INTEGER_CST@1) (mult @0 INTEGER_CST@2))
> +  (if (INTEGRAL_TYPE_P (type)
> +       && wi::multiple_of_p (wi::to_widest (@1), wi::to_widest (@2), SIGNED))
> +   (if (TYPE_OVERFLOW_UNDEFINED (type) && !TYPE_OVERFLOW_SANITIZED (type))
> +    (div @1 @2)
> +#if GIMPLE
> +    (with {value_range vr0, vr1, vr2;}
> +     (if (get_range_query (cfun)->range_of_expr (vr0, @0)
> +      && get_range_query (cfun)->range_of_expr (vr1, @1)
> +      && get_range_query (cfun)->range_of_expr (vr2, @2)
> +      && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr1)
> +      && range_op_handler (MULT_EXPR).overflow_free_p (vr0, vr2))
> +      (div @1 @2)))
> +#endif
>    ))))
> 
> #if GIMPLE
> --- gcc/testsuite/gcc.dg/tree-ssa/pr112994-2.c.jj    2023-12-13 19:07:20.882475735 +0100
> +++ gcc/testsuite/gcc.dg/tree-ssa/pr112994-2.c    2023-12-13 19:07:02.597726855 +0100
> @@ -0,0 +1,15 @@
> +/* PR tree-optimization/112994 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fdump-tree-optimized" } */
> +/* { dg-final { scan-tree-dump-times "return 2;" 3 "optimized" } } */
> +/* { dg-final { scan-tree-dump-times "return 7;" 3 "optimized" } } */
> +/* { dg-final { scan-tree-dump-times "return -7;" 2 "optimized" } } */
> +
> +int f1 (int x) { return (x * 4) / (x * 2); }
> +int f2 (int x) { return (x * 56) / (x * 8); }
> +int f3 (int x) { return (x * 56) / (x * -8); }
> +int f4 (int x) { int y = x * 4; return y / (x * 2); }
> +int f5 (int x) { int y = x * 56; return y / (x * 8); }
> +int f6 (int x) { int y = x * 56; return y / (x * -8); }
> +unsigned f7 (unsigned x) { if (x > ~0U / 4) __builtin_unreachable (); unsigned y = x * 4; return y / (x * 2); }
> +unsigned f8 (unsigned x) { if (x > ~0U / 56) __builtin_unreachable (); unsigned y = x * 56; return y / (x * 8); }
> 
>    Jakub
>