Ping: [PATCH] x86: don't use AVX512BW vmovdqu variants without -mavx512bw

Jan Hubicka hubicka@ucw.cz
Tue Jan 2 13:30:00 GMT 2018


> >>> On 02.01.18 at 11:05, <hubicka@ucw.cz> wrote:
> >> >>> On 13.12.17 at 10:33,  wrote:
> >> > Simply mirror the MODE_XI logic of handling unaligned operands in
> >> > mov<mode>_internal into MODE_TI / MODE_OI handling.
> >> > 
> >> > gcc/
> >> > 2017-12-13  Jan Beulich  <jbeulich@suse.com>
> >> > 
> >> > 	* sse.md (mov<mode>_internal): Tighten condition for when to use
> >> > 	vmovdqu<ssescalarsize> for TI and OI modes.
> >> > 
> >> > gcc/testsuite/
> >> > 2017-12-13  Jan Beulich  <jbeulich@suse.com>
> >> > 
> >> > 	* gcc.target/i386/avx512vl-no-vmovdqu8.c,
> >> > 	gcc.target/i386/avx512vl-no-vmovdqu16.c: New.
> > 
> > Looks OK. We do not need to update instruction attribute becuase we make no
> > difference between those instructions and both ends up being ssemov?
> 
> I think so, yes, but my opinion here is certainly not meaning much.

Seems to be that case. So patch is OK.
(we do not model avx512 on core or avx256 on Zen for scheduling very well. I guess
we should eventually)

Honza
> 
> Jan



More information about the Gcc-patches mailing list