On Saturday, March 5, 2016 4:02:17 PM PST Jordan Justen wrote:
> On 2016-03-04 16:07:08, Matt Turner wrote:
> > Alternatively, if overwriting FFTID is bad, instead of emitting
> >
> > shr(1) sr0.1:UD g0.0:UD 16UD
> >
> > you could write only the low word:
> >
> > mov(1) sr0.2:UW g0.1:UW
> >
> > or maybe use an AND to ensure you're writing zeros to the MBZ bits of
> > the status register.
>
> Ok, with Matt's suggestion, it is pretty easy to remove the FFTID
> concern. I made this change:
>
> - abld.SHR(suboffset(brw_sr0_reg(), 1), g0, brw_imm_d(16));
> + abld.MOV(retype(suboffset(brw_sr0_reg(), 1), BRW_REGISTER_TYPE_UW),
> + suboffset(retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW),
1));
>
> Which produces:
>
> mov(1) sr0.2<1>UW g0.1<0,1,0>UW { align1
WE_all };
>
> And, therefore shouldn't touch the FFTID bits in the upper 16-bits of
> sr0.1.
>
> Matt, Ken: Does that look okay?
>
> -Jordan
> Sounds good to me. I trust Matt's review, so I'm not going to bother; feel free to have an Ack if you want one.
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
