[FFmpeg-devel] Amazing intrinsics improvments in gcc 4

Michael Niedermayer michaelni
Wed Mar 19 19:07:03 CET 2008


2004-03-12 i submitted bug
14552 (compiled trivial vector intrinsic code is inefficient) to gcc

the source was:
 typedef short mmxw  __attribute__ ((mode(V4HI)));
 typedef int   mmxdw __attribute__ ((mode(V2SI)));

mmxdw dw;
mmxw w;

void test(){
    dw= (mmxdw)w;

what gcc-3.4 -O3 -mtune=pentium3 -march=pentium3 -S  generated was:
        movq    w, %mm1 
        pushl   %ebp 
        movl    %esp, %ebp 
        popl    %ebp 
        psllw   $1, %mm1 
        movq    %mm1, w 
        movq    w, %mm0 
        movq    %mm0, dw 

what a human would generate:
        movq w, %mm1 
        paddw %mm1,%mm1 
        movq %mm1, w 
        movq %mm1,dw 

Now finally after 4 years the bug has been closed (with wontfix of course)
and gcc-4.3 -O3 -mtune=pentium3 -march=pentium3 -S now generates:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $16, %esp
        movl    w, %eax
        movl    w+4, %edx
        movl    %ebx, -8(%ebp)
        movl    %esi, -4(%ebp)
        movl    %eax, -16(%ebp)
        movl    %edx, -12(%ebp)
        movswl  -16(%ebp),%ecx
        movswl  -14(%ebp),%ebx
        movswl  -12(%ebp),%esi
        movswl  -10(%ebp),%eax
        addl    %ecx, %ecx
        addl    %ebx, %ebx
        addl    %esi, %esi
        addl    %eax, %eax
        movw    %bx, w+2
        movl    -8(%ebp), %ebx
        movw    %si, w+4
        movl    -4(%ebp), %esi
        movw    %ax, w+6
        movw    %cx, w
        movl    w+4, %edx
        movl    w, %eax
        movl    %edx, dw+4
        movl    %eax, dw
        movl    %ebp, %esp
        popl    %ebp

I thought some people here would be interrested as there were various claims
on gccs abilities and improvments posted here lately ...

Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I am the wisest man alive, for I know one thing, and that is that I know
nothing. -- Socrates
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080319/7f4a2121/attachment.pgp>

More information about the ffmpeg-devel mailing list