[dpdk-stable] AVX512 bug on SkyLake

Thomas Monjalon thomas at monjalon.net
Mon Nov 12 10:21:56 CET 2018


12/11/2018 10:09, Christian Ehrhardt:
> > -    vmovdqu8 xmm0,XMMWORD PTR [rax*8+0x6]
> > +    vmovdqu xmm0,XMMWORD PTR [rax*8+0x60]
> >      vinserti128 ymm0,ymm0,XMMWORD PTR [rax*8+0x70],0x1
> >      vmovups XMMWORD PTR [rsi+0x60],xmm0
> >      vextracti128 XMMWORD PTR [rsi+0x70],ymm0,0x1
> >
> > > Not sure what causing compiler behaves that way.
> > > BTW, looking though testpmd objdump output - it seems that only mlx5 driver
> > > exhibits such problem (I didn't enable mlx4 actually, probably same problem here).
> > > Which looks a bit weird to me.
> >
> > Yes it's weird. I don't see how the mlx5 code could influence
> > the compiler to generate this bad code in AVX512 mode.
> 
> Thomas you have all this set up, do you have any chance to test this
> on the GCC's in Ubuntu 18.10 and 19.04
> If easy I'd love to see results wit hgcc-7 & gcc-8 as in Ubuntu 19.04
> (current -dev).
> If the above is too hard, at least could you try the gcc-8 in Bionic
> is 8.2.0-1ubuntu2~18.04 that is rather close.

I already tested updated GCC 7 and 8 (it is the same result):
	https://bugs.dpdk.org/show_bug.cgi?id=97#c18

Versions are:
	gcc-7 (Ubuntu 7.3.0-27ubuntu1~18.04) 7.3.0
	gcc-8 (Ubuntu 8.2.0-1ubuntu2~18.04) 8.2.0

> If you could share the simplified build options that you need to
> reproduce that would help (at least me) - thanks in advance

You just need to compile mlx5, nothing special.




More information about the stable mailing list