This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug target/81085] New: inefficient union with long double argument on 32-bit x86


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81085

            Bug ID: 81085
           Summary: inefficient union with long double argument on 32-bit
                    x86
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jsm28 at gcc dot gnu.org
  Target Milestone: ---
            Target: i?86-*-*

Consider the following code accessing part of the representation of a long
double function argument (which is similar to what various code in glibc libm
does).

union u { long double d; unsigned int u[4]; };
unsigned int f (long double x) { union u u = { .d = x }; return u.u[0]; }

On x86_64-linux-gnu, building with -m64 -O2, I get the expected code:

        movl    8(%rsp), %eax
        ret

But with -m32 -O2 I get:

        subl    $28, %esp
        .cfi_def_cfa_offset 32
        fldt    32(%esp)
        fstpt   (%esp)
        movl    (%esp), %eax
        addl    $28, %esp
        .cfi_def_cfa_offset 4
        ret

The argument is already on the stack.  There is no need to load and store a
copy of it in order to access part of it as an integer.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]