[Mesa-dev] [PATCH v3 22/42] intel/compiler: don't propagate HF immediates to 3-src instructions

[Iago Toral]

On Tue, 2019-01-22 at 16:18 -0800, Matt Turner wrote:
> On Tue, Jan 15, 2019 at 5:54 AM Iago Toral Quiroga <itoral at igalia.com
> > wrote:
> > 
> > 3-src instructions don't support immediates, but since
> > 36bc5f06dd22,
> > we allow them on MAD and LRP relying on the combine constants pass
> > to
> > fix it up later. However, that pass is specialized for 32-bit float
> > immediates and can't handle HF constants at present, so this patch
> > ensures that copy-propagation only does this for 32-bit constants.
> 
> There's a patch later in the series that adds HF support to constant
> combining (and presumably removes this code). Maybe it's the best
> thing to add and remove the code in the same series, but it's good to
> at least mention that in the commit message so that reviewers
> understand what the plan is.

Yes, makes sense.

> I see from a later thread that you're going to try to handle more
> than
> just F/HF types in constant combining, and I guess you'll resend this
> series. If that's the case, I'd just leave this patch out if it's
> possible to reorder things.

Yes, I'll do that.

> Oh, another thing: Gen10+ can take *1* immediate HF argument in 3-src
> instructions. We haven't added that support yet, though all the
> low-level brw_inst_* functions exist. Not asking you to do that
> without hardware, but just thought you'd be interested to know :)

Sure, that's good to know, thanks for the info! :)