[PATCH 07/20] glamor: Add simple upload/download functions in glamor_transfer

Keith Packard keithp at keithp.com
Thu Mar 20 14:44:59 PDT 2014


Markus Wick <markus at selfnet.de> writes:

> So in the end, I suggest to use PBO all the time. It's likely an 
> overhead compared to the internal staging buffer of the driver, but imo 
> this shouldn't matter compared with the memcpy time.

As you'll note, the fallback API does use PBOs through this same
interface. The other main user of these APIs are GetImage and PutImage,
both of which transfer data between the GPU and network buffers. As we
don't (currently) control allocation of those buffers, we're stuck doing
at least one copy from network buffer to GPU. I suggest that we leave
the precise details of implementing that efficiently to the GL driver
and not attempt to optimize it further in the X code, which has very
little good information about the best way to do that.

> How should we deal with plattforms without ARB_pbo? I think glamor 
> should work, but performance doesn't matter that hard. I don't want to 
> fork this function to work fast with and without PBO.

Agreed -- the fallback code is dramatically faster now that I've
switched from malloc to PBOs, but switching back for drivers without
PBOs would be quite easy, if that seems necessary.

-- 
keith.packard at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 810 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg-devel/attachments/20140320/45333b43/attachment.sig>


More information about the xorg-devel mailing list