[Bug c/70039] New: Data placed into rodata that could be encoded as immediates

["andrew.burgess at embecosm dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70039

            Bug ID: 70039
           Summary: Data placed into rodata that could be encoded as
                    immediates
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: andrew.burgess at embecosm dot com
  Target Milestone: ---

Created attachment 37840
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37840&action=edit
An example of this behaviour

Compiling the attached file with gcc from SVN 1-Mar-2016 on x86-64 GNU/Linux:

  gcc -O2 -Wall -Wextra -o fill.s -S fill.i

Here's some of the source:

  void
  fill_from_array (char * d)
  {
    static const char a[] =
      {
        'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j',
        'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't',
        'u', 'v', 'w', 'x', 'y', 'z',
      };
    memcpy (d, a, 20);
  }

  void
  fill_from_string (char * d)
  {
    static const char a[] = "abcdefghijklmnopqrstuvwxyz";
    memcpy (d, a, 20);
  }

in the generated assembler of fill_from_string the data is written to the
destination pointer (d) as constants, like this:

  movabsq $7523094288207667809, %rax
  movl    $1953722993, 16(%rdi)
  movq    %rax, (%rdi)
  movabsq $8101815670912281193, %rax
  movq    %rax, 8(%rdi)

However, in the fill_from_array case the array data is held in the .rodata
section, loaded, and then stored to the destination.

My expectation would have been that the code generated in each case would have
been identical.