[Bug rtl-optimization/45670] New: Less efficient x86 addressing mode selection on 4.6, causes -Os size regression from 4.5

[sgunderson at bigfoot dot com]

Hi,

Given the following test C++ file:

class Class
{
public:
        void func();

        float *buf;
        int size;
};

void Class::func()
{
        for (int i = 0; i < size; ++i) {
                buf[i] = 0;
        }
}

4.6 (see below for exact version) will generate larger code (36 vs. 30 bytes)
than 4.5.1 (Debian 4.5.1-6) given -Os. The output is

00000000 <Class::func()>:
   0:   55                      push   %ebp
   1:   31 c0                   xor    %eax,%eax
   3:   89 e5                   mov    %esp,%ebp
   5:   8b 4d 08                mov    0x8(%ebp),%ecx
   8:   53                      push   %ebx
   9:   8b 59 04                mov    0x4(%ecx),%ebx
   c:   eb 10                   jmp    1e <Class::func()+0x1e>
   e:   8d 14 85 00 00 00 00    lea    0x0(,%eax,4),%edx
  15:   40                      inc    %eax
  16:   03 11                   add    (%ecx),%edx
  18:   c7 02 00 00 00 00       movl   $0x0,(%edx)
  1e:   39 d8                   cmp    %ebx,%eax
  20:   7c ec                   jl     e <Class::func()+0xe>
  22:   5b                      pop    %ebx
  23:   5d                      pop    %ebp
  24:   c3                      ret    

Basically the problem is that the lea is large (due to the zero immediate
taking up 32 bits); 4.5 uses a variation where the address calculation takes
both a base and an index register, which has a shorter form not requiring to
store the zero. (The joys of x86; lea edx, [eax*4 + ecx] takes less space then
lea edx, [eax*4]...)

===

Configured with: ../src/configure -v --with-pkgversion='Debian 20100828-1'
--with-bugurl=file:///usr/share/doc/gcc-snapshot/README.Bugs
--enable-languages=c,ada,c++,fortran,objc,obj-c++
--prefix=/usr/lib/gcc-snapshot --enable-shared --enable-multiarch
--enable-linker-build-id --with-system-zlib --disable-nls --enable-clocale=gnu
--enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-plugin
--enable-gold --with-plugin-ld=ld.gold --enable-objc-gc --enable-targets=all
--with-arch-32=i586 --with-tune=generic --disable-werror --enable-checking=yes
--build=i486-linux-gnu --host=i486-linux-gnu --target=i486-linux-gnu
Thread model: posix
gcc version 4.6.0 20100828 (experimental) [trunk revision 163616] (Debian
20100828-1)


-- 
           Summary: Less efficient x86 addressing mode selection on 4.6,
                    causes -Os size regression from 4.5
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: sgunderson at bigfoot dot com
 GCC build triplet: i486-linux-gnu
  GCC host triplet: i486-linux-gnu
GCC target triplet: i486-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45670