On Sat, Oct 13, 2018 at 6:12 PM Jason Ekstrand <jason@xxxxxxxxxxxxxx> wrote:
>
> On Sat, Oct 13, 2018 at 10:58 AM Bas Nieuwenhuizen <bas@xxxxxxxxxxxxxxxxxxx> wrote:
>>
>> On Fri, Oct 12, 2018 at 10:38 PM Keith Packard <keithp@xxxxxxxxxx> wrote:
>> >
>> > According to the Vulkan spec:
>> >
>> > "Vulkan 1.0 initially required all new physical-device-level extension
>> > functionality to be structured within an instance extension. In order
>> > to avoid using an instance extension, which often requires loader
>> > support, physical-device-level extension functionality may be
>> > implemented within device extensions"
>> >
>> > The code that checks for enabled extension APIs currently only passes
>> > functions with VkDevice or VkCommandBuffer as their first
>> > argument. This patch extends that to also allow functions with
>> > VkPhysicalDevice parameters, in support of the above quote from the
>> > Vulkan spec.
>> >
>>
>> Also "To obtain a function pointer for a physical-device-level command
>> from a device extension, an application can use vkGetInstanceProcAddr.
>> "
>>
>> As far as I can tell the device_command member is only to make sure we
>> return NULL from vkGetDeviceProcAddr, and per the spec (3.1 table 2)
>> we still have to return NULL there for functions which take
>> VkPhysicalDevice? So the old behavior seems correct to me.
>
>
> I think anv is ignoring that line in the table which is why it works for us. If only someone wrote tests for these things...
>
> I think the correct interpretation would be that any physical device functions that are part of a core version or instance extension should yield NULL but any physical device functions from a device extension should return a valid function pointer. Sadly, that behavior is kind-of a pain to implement. :-(
How would you read that into the spec? As quoted above it explicitly
tells you to use vkGetInstanceProcAddr for VkPhysicalDevice functions,
even if they are based on a device extension. (And you cannot really
use vkGetDeviceProcAddr anyway as the typical usecase is before you've
created a device).
>
>>
>> > Signed-off-by: Keith Packard <keithp@xxxxxxxxxx>
>> > ---
>> > src/amd/vulkan/radv_entrypoints_gen.py | 2 +-
>> > 1 file changed, 1 insertion(+), 1 deletion(-)
>> >
>> > diff --git a/src/amd/vulkan/radv_entrypoints_gen.py b/src/amd/vulkan/radv_entrypoints_gen.py
>> > index 377b544c2aa..69e6fc3e0eb 100644
>> > --- a/src/amd/vulkan/radv_entrypoints_gen.py
>> > +++ b/src/amd/vulkan/radv_entrypoints_gen.py
>> > @@ -352,7 +352,7 @@ class Entrypoint(EntrypointBase):
>> > self.return_type = return_type
>> > self.params = params
>> > self.guard = guard
>> > - self.device_command = len(params) > 0 and (params[0].type == 'VkDevice' or params[0].type == 'VkQueue' or params[0].type == 'VkCommandBuffer')
>> > + self.device_command = len(params) > 0 and (params[0].type == 'VkPhysicalDevice' or params[0].type == 'VkDevice' or params[0].type == 'VkQueue' or params[0].type == 'VkCommandBuffer')
>> >
>> > def prefixed_name(self, prefix):
>> > assert self.name.startswith('vk')
>> > --
>> > 2.19.1
>> >
>> > _______________________________________________
>> > mesa-dev mailing list
>> > mesa-dev@xxxxxxxxxxxxxxxxxxxxx
>> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev@xxxxxxxxxxxxxxxxxxxxx
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
