[Mesa-dev] [PATCH 15/18] docs: Import extension spec for GLX_MESA_query_renderer
idr at freedesktop.org
Thu Nov 7 16:23:15 PST 2013
On 11/07/2013 01:22 PM, Dave Airlie wrote:
> On Sat, Oct 12, 2013 at 8:10 AM, Ian Romanick <idr at freedesktop.org> wrote:
>> From: Ian Romanick <ian.d.romanick at intel.com>
>> The enumerated values are currently allocated from Intel's range.
> Some highlevel comments below,
>> + GLX renderer attribute number description
>> + of values
>> + ---------------------- --------- -----------
>> + GLX_RENDERER_VENDOR_ID_MESA 1 PCI ID of the device vendor
>> + GLX_RENDERER_DEVICE_ID_MESA 1 PCI ID of the device
> Okay PCI IDs might seem useful but we have a lot of situations where
> they aren't available,
> a) ARM devices generally don't have them
> b) nvidia generally espouse using PCI IDs in favour of a chip family
> ID they bake into the GPUs,
> nouveau doesn't even know the actual PCI IDs in userspace, we could
> find them but it does take a bit more work, like new kernel interface.
Multiple ISVs, including Wine, asked specifically for PCI IDs. If a PCI
ID doesn't exist or can't be provided... zero for the
GLX_RENDERER_DEVICE_ID_MESA? It should always be possible to provide a
reasonable value for GLX_RENDERER_VENDOR_ID_MESA. I can add an issue
for that and adjust the documentation of those queries.
Does that sound reasonable?
>> + GLX renderer attribute description
>> + ---------------------- -----------
>> + GLX_RENDERER_VENDOR_ID_MESA Name of the renderer provider. This may
>> + differ from the vendor name of the
>> + underlying hardware.
>> + GLX_RENDERER_DEVICE_ID_MESA Name of the renderer. This may differ from
>> + the name of the underlying hardware (e.g.,
>> + for a software renderer).
>> + If <attribute> is not a recognized value, NULL is returned, but no GLX
>> + error is generated.
>> + The string returned for GLX_RENDERER_VENDOR_ID_MESA will have the same
>> + format as the string that would be returned by glGetString of GL_VENDOR.
>> + It may, however, have a different value.
> Any reason for the value to be different here? Like a use case?
Vendor and device are the only ones that make any sense to be strings.
Is there any utility in getting the memory size as a string? :)
>> + 14) Why not make the queries from issue #14 GL functions (instead of GLX)?
> from issue 14? this is issue 14
Right... that should say "issue #13".
More information about the mesa-dev