On Tue, Nov 9, 2021 at 5:12 AM Navid Rahimi via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Hi GCC community,
>
> This patch will add the missed pattern described in bug 102232 [1] to the
> match.pd. The testcase will test whether the multiplication and division has
> been removed from the code or not. The correctness proof for this pattern is
> here [2] in case anyone is curious.
>
> PR tree-optimization/102232
> * match.pd (x * (1 + y / x) - y) -> (x - y % x): New optimization.
+/* x * (1 + y / x) - y -> x - y % x */
+(simplify
+ (minus (mult:cs @0 (plus:cs integer_onep (trunc_div:s @1 @0))) @1)
the canonical order of the plus is (plus:s (trunc_div ...) integer_onep) as
constants come last - you can then remove the 'c'
+ (if (INTEGRAL_TYPE_P (TREE_TYPE (@0))
you can use INTEGRAL_TYPE_P (type).
+ && types_match (@0, @1))
this test is not necessary
+ (minus @0 (trunc_mod @1 @0))))
But should we also optimize x * (2 + y/x) - y -> 2*x - y % x? So
it looks like a conflict with the x * (1 + b) <-> x + x * b transform
(fold_plusminus_mult)? That said, the special case of one
definitely makes the result cheaper (one less operation).
Please move the pattern next to the most relatest which I think is
/* X - (X / Y) * Y is the same as X % Y. */
(simplify
(minus (convert1? @0) (convert2? (mult:c (trunc_div @@0 @@1) @1)))
(if (INTEGRAL_TYPE_P (type) || VECTOR_INTEGER_TYPE_P (type))
(convert (trunc_mod @0 @1))))
+int
+main (void)
+{
+ // few randomly generated test cases
+ if (foo (71856034, 238) != 212)
+ {
+ return 1;
the return value of 1 is an unreliable way to fail, please instead call
__builtin_abort ();
+/* { dg-options "-O3 -fdump-tree-optimized" } */
do we really need -O3 for this? I think that -O should be enough here?
Thanks,
Richard.
> * gcc.dg/tree-ssa/pr102232.c: testcase for this optimization.
>
>
> 1) https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102232
> 2) https://alive2.llvm.org/ce/z/2VScjD
>
> Best wishes,
> Navid.
