[PATCH v2 15/27] gpu: host1x: Add support for Tegra114

[Thierry Reding]

On Mon, Oct 14, 2013 at 08:30:28AM +0300, Terje Bergström wrote:
> On 12.10.2013 14:24, Thierry Reding wrote:
> > Yeah, I don't like very much how this is currently done. I mean about
> > half of this is actually duplicate code because of the static inline
> > functions used for register defines. As discussed elsewhere this was
> > originally meant to be used for coverage testing, but nobody's done
> > anything like that as far as I know. I'm also not convinced that these
> > would be very useful in coverage testing, but adding Terje on Cc and
> > unless he or anyone else has any (strong) objections I'll go and remove
> > the duplicate definitions and while at it see if I can come up with a
> > way to share more code/definitions between versions of host1x.
> > 
> > Do you see this as a blocker for 3.13 or can I do the cleanup after
> > that? As far as I know Tegra124 has a more heavily modified version of
> > host1x so implementing Tegra124 support might be a good opportunity to
> > clean this up anyway.
> 
> For Tegra114 the register level changes are small (one channel added,
> wait bases, expanded fifo, etc) but for Tegra124 they're bigger. So
> let's keep the long term in mind.

I agree with Stephen here. We should really be working together with HW
to prevent as many incompatibilities as we can.

> I'm ok with removing the inlines. Just the energy spent in discussing it
> over and over again is more than the benefits that a proper compiler
> gives over the preprocessor (type checking, real syntax checking,
> coverage, gdb breakpoints, etc). For some reason (which escapes me)
> preprocessor magic seems to be the norm in Linux kernel.
> 
> Parameterized code gets easily really ugly, difficult to debug and
> difficult to port to new chips. There's much less work and opportunity
> of mistake in maintaining data in a couple of files than subtle
> algorithms all over the driver. If we don't have inlines in hw headers,
> the headers become tiny and any benefits of parameterized code versus
> per-hw version headers vanish.

I'll keep that in mind when I revisit this.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/dri-devel/attachments/20131015/e3546c93/attachment-0001.pgp>