Re: [PATCH 2/3] lib/rendercopy: Convert rendercopy_gen9 to use genxml pack headers

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Fri, Apr 24, 2026 at 05:37:37PM +0200, Zbigniew Kempczyński wrote:
> On Fri, Apr 24, 2026 at 01:46:38PM +0300, Ville Syrjälä wrote:
> > On Fri, Apr 24, 2026 at 12:21:37PM +0200, Zbigniew Kempczyński wrote:
> > > On Tue, Apr 07, 2026 at 06:00:49PM +0300, Ville Syrjälä wrote:
> > > 
> > > <cut>
> > > 
> > > > > +	/* MOCS encoding: genxml has a single 7-bit MOCS field (bits 30:24).
> > > > > +	 * The old struct had mocs_index:6 at bits 30:25 and pxp:1 at bit 24.
> > > > > +	 * Reproduce the same bit layout. */
> > > > > +	mocs = (buf->mocs_index << 1) | (intel_buf_pxp(buf) ? 1 : 0);
> > > > 
> > > > This annoying mocs_index stuff should be nuked throughout igt,
> > > > and replaced with the full mocs field. I've already gotten
> > > > confused by this multiple times when it looked like the 
> > > > relevant macros were off by one bit when compared to the spec.
> > > 
> > > IMO we can't stop using mocs index. BLOCK_COPY for Xe2+ has mocs index
> > > on bits[27:24] and encrypt is on bit[21]. Bits[23-22] are reserved.
> > 
> > If that's the exception then it would be better to handle it there.
> > Everything else just has a single MOCS field, and the encryption bit
> > (if it exists) is just part of it. The current thing makes zero sense
> > for platforms that don't have the encrypt bit, and for the platforms
> > where the encrypt bit is part of MOCS the whole thing is just
> > intentionally confusing.
> 
> I've found 4 exceptions in xe2.xml where MOCS doesn't use previous
> [7] or [6+1] bit arrangements.
> 
> Interesting is EXECUTE_INDIRECT_DISPATCH contains MOCS index only. We're
> not using this command in IGT, but this shows index also can appear
> standalone. I think IGT should provide both interfaces - for getting
> MOCS index standalone and for getting the full MOCS field and on
> usecase depends which interface should be used.

IIRC Mesa uses the full MOCS field. We should do the same, especially
if we now plan to start using genxml.

-- 
Ville Syrjälä
Intel