[Bug target/77276] New: strlen expanded inline with -Os, emits larger code than with -O2

["msebor at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77276

            Bug ID: 77276
           Summary: strlen expanded inline with -Os, emits larger code
                    than with -O2
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: msebor at gcc dot gnu.org
  Target Milestone: ---

On x86_64 (but not on powerpc64le, for example), a call to __builtin_strlen
with a non-constant argument is expanded inline at -Os (as well as -Og and
-O1), but not at -O2 even though the size of the code emitted at -O2 is less
than that at -Os.

$ (set -x && cat xyz.c && for o in s 2; do /build/gcc-trunk-svn/gcc/xgcc -B
/build/gcc-trunk-svn/gcc -O$o -Wall -Wextra -c xyz.c && objdump -d xyz.o; done)
+ cat xyz.c
static const char* volatile s = "123";

unsigned f (void)
{
  return __builtin_strlen (s);
}
+ for o in s 2
+ /build/gcc-trunk-svn/gcc/xgcc -B /build/gcc-trunk-svn/gcc -Os -Wall -Wextra
-c xyz.c
+ objdump -d xyz.o

xyz.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <f>:
   0:   48 8b 3d 00 00 00 00    mov    0x0(%rip),%rdi        # 7 <f+0x7>
   7:   31 c0                   xor    %eax,%eax
   9:   48 83 c9 ff             or     $0xffffffffffffffff,%rcx
   d:   f2 ae                   repnz scas %es:(%rdi),%al
   f:   48 89 c8                mov    %rcx,%rax
  12:   48 f7 d0                not    %rax
  15:   48 ff c8                dec    %rax
  18:   c3                      retq   
+ for o in s 2
+ /build/gcc-trunk-svn/gcc/xgcc -B /build/gcc-trunk-svn/gcc -O2 -Wall -Wextra
-c xyz.c
+ objdump -d xyz.o

xyz.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <f>:
   0:   48 83 ec 08             sub    $0x8,%rsp
   4:   48 8b 3d 00 00 00 00    mov    0x0(%rip),%rdi        # b <f+0xb>
   b:   e8 00 00 00 00          callq  10 <f+0x10>
  10:   48 83 c4 08             add    $0x8,%rsp
  14:   c3                      retq