[Bug middle-end/84309] [8 Regression] Wrong-code with -ffast-math

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84309

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2018-02-09
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 43387
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43387&action=edit
gcc8-pr84309.patch

That said, if the target has c99 runtime and the constant is a power of 2, I
can't imagine why exp2 (log2 (C) * x) wouldn't be faster and more precise than
exp (log (C) * x).