[Bug middle-end/96175] New: Failure to optimize memory stores of double literals properly

[Bug middle-end/96175] New: Failure to optimize memory stores of double literals properly

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96175

            Bug ID: 96175
           Summary: Failure to optimize memory stores of double literals
                    properly
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

void f(double *x)
{
    *x = 1.0;
}

With -O3, LLVM generates this :

f(double*): # @f(double*)
  movabs rax, 4607182418800017408
  mov qword ptr [rdi], rax
  ret

GCC generates this :

f(double*):
  mov rax, QWORD PTR .LC0[rip]
  mov QWORD PTR [rdi], rax
  ret
.LC0:
  .long 0
  .long 1072693248

This also occurs on every other architecture I tested (arm and powerpc64le,
though only with GCC 9). GCC is weirdly insistent on having a literal stored in
const data instead of just having it in the code, which seems inefficient.

[Bug target/96175] Failure to optimize memory stores of double literals properly

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96175

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Each target decides how liberals are handled.  So this is a target issue.  Iirc
aarch64 was fixed in gcc 10 for an example.

[Bug target/96175] Failure to optimize memory stores of double literals properly

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96175

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-07-13
     Ever confirmed|0                           |1
             Target|                            |x86_64-*-* i?86-*-*
             Status|UNCONFIRMED                 |NEW

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think one issue is we don't know how many times a constant literal is used
and thus for example for -Os whether it's more beneficial to inline it or
not.  The instructions with large immediate are also not very friendly
to modern architecture yop caches and that up more space there.

But confirmed.