[Bug rtl-optimization/17236] inefficient code for long long multiply on x86

[bonzini at gnu dot org]

------- Comment #19 from bonzini at gnu dot org  2007-12-19 12:13 -------
Created an attachment (id=14792)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14792&action=view)
patch to almost fix the bug

With this patch:
1) local-alloc first tries to allocate registers that go into small classes
2) regclass tries to use the same class for an output and an input that dies
3) i386 modes_tiable_p allows tying SImode with DImode (but not vice versa)

I get now this:
        pushl   %esi
        pushl   %ebx
        movl    12(%esp), %ecx
        movl    20(%esp), %eax
        movl    24(%esp), %ebx
        imull   %ecx, %ebx
        movl    16(%esp), %esi
        imull   %eax, %esi
        addl    %esi, %ebx
        mull    %ecx
        leal    (%ebx,%edx), %edx
        popl    %ebx
        popl    %esi

which has no moves.  It still spills two registers because it is not able to
use %edx (I'll look into it later).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17236