Hi! The following patch implements an optimization suggested in the PR, copysign(x,-x) can be optimized into -x (even without -ffast-math, should work fine even for signed zeros and infinities or nans).
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2020-08-24 Jakub Jelinek <ja...@redhat.com> PR tree-optimization/96715 * match.pd (copysign(x,-x) -> -x): New simplification. * gcc.dg/tree-ssa/copy-sign-3.c: New test. --- gcc/match.pd.jj 2020-08-24 10:00:01.000000000 +0200 +++ gcc/match.pd 2020-08-24 15:53:56.001381109 +0200 @@ -5293,6 +5293,11 @@ (define_operator_list COND_TERNARY (COPYSIGN_ALL @0 @0) @0) +(simplify + /* copysign(x,-x) -> -x. */ + (COPYSIGN_ALL @0 (negate@1 @0)) + @1) + (simplify /* copysign(x,y) -> fabs(x) if y is nonnegative. */ (COPYSIGN_ALL @0 tree_expr_nonnegative_p@1) --- gcc/testsuite/gcc.dg/tree-ssa/copy-sign-3.c.jj 2020-08-24 16:09:11.543453434 +0200 +++ gcc/testsuite/gcc.dg/tree-ssa/copy-sign-3.c 2020-08-24 16:08:31.360020186 +0200 @@ -0,0 +1,23 @@ +/* PR tree-optimization/96715 */ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-optimized" } */ +/* { dg-final { scan-tree-dump-not "= __builtin_copysign" "optimized" } } */ +/* { dg-final { scan-tree-dump-times " = -x_\[0-9]*\\(D\\)" 3 "optimized" } } */ + +float +foo (float x) +{ + return __builtin_copysignf (x, -x); +} + +double +bar (double x) +{ + return __builtin_copysign (x, -x); +} + +long double +baz (long double x) +{ + return __builtin_copysignl (x, -x); +} Jakub