[PATCH 1/4] winex11.drv: Only return vulkan functions if the host function is available.

[Georg Lehmann]

On 03.07.21 20:45, Liam Middlebrook wrote:
> I like how simple this solution is.
> 
> At first I was a bit worried that this would make the outstanding 
> patchset from Derek Lesho more complex, but I think the same strategy 
> you're applying here for a map of Win32 function to Unix platform 
> function could be applied. That said, it's still not the cleanest feeling.
>  >
> When I filed https://bugs.winehq.org/show_bug.cgi?id=51360 I was 
> thinking a solution for this would look like the follow steps (roughly):
> 
> 1. Add logic in make_vulkan to correspond what extensions must be 
> supported by the host implementation for a command to be enabled.
> 
> 2. At Get*ProcAddr time check that function name against the list of 
> extensions, and ensure that said extensions are supported by the 
> VkInstance/VkDevice.
> 
> 
> You're solution is a lot simpler, and I like it. But I can't help but 
> worry that with Derek's pending work (or any potential future work in 
> the same vein) this will cause issues.
>


Yes, for platform specific extensions we have to manually replace the 
name of the win32 function. For Derek's (CC'ed) patches this should be 
relatively simple as far as I can tell, vkGetMemoryWin32HandleKHR maps 
to vkGetMemoryFdKHR and vkGetMemoryWin32HandlePropertiesKHR to 
vkGetMemoryFdPropertiesKHR.
It's the same story for the external semaphore/fence extensions, if we 
ever get around to support those.

I share your general skepticism but given that there aren't many 
platform specific extensions and I see none which could cause issues, I 
think this is good enough. If we have to in the future, we can still go 
for something similar to your proposed solution. But I don't really want 
to make make_vulkan more complex without a good reason.

Thanks,

Georg Lehmann

> 
> I'm not conceptually opposed to this, but I want to make sure the 
> implications of this simpler solution are discussed before I give my 
> sign-off on it.
> 
> 
> Thanks,
> 
> Liam Middlebrook
> 
> On 7/3/21 9:42 AM, Georg Lehmann wrote:
>> Signed-off-by: Georg Lehmann <dadschoorse at gmail.com>
>> ---
>>   dlls/winex11.drv/vulkan.c | 16 ++++++++++++++++
>>   1 file changed, 16 insertions(+)
>>
>> diff --git a/dlls/winex11.drv/vulkan.c b/dlls/winex11.drv/vulkan.c
>> index bdc287afeea..1bbdba2ce1d 100644
>> --- a/dlls/winex11.drv/vulkan.c
>> +++ b/dlls/winex11.drv/vulkan.c
>> @@ -435,12 +435,25 @@ static VkResult 
>> X11DRV_vkGetDeviceGroupSurfacePresentModesKHR(VkDevice device,
>>       return pvkGetDeviceGroupSurfacePresentModesKHR(device, 
>> x11_surface->surface, flags);
>>   }
>> +static const char *wine_vk_native_fn_name(const char *name)
>> +{
>> +    if (!strcmp(name, "vkCreateWin32SurfaceKHR"))
>> +        return "vkCreateXlibSurfaceKHR";
>> +    if (!strcmp(name, "vkGetPhysicalDeviceWin32PresentationSupportKHR"))
>> +        return "vkGetPhysicalDeviceXlibPresentationSupportKHR";
>> +
>> +    return name;
>> +}
>> +
>>   static void *X11DRV_vkGetDeviceProcAddr(VkDevice device, const char 
>> *name)
>>   {
>>       void *proc_addr;
>>       TRACE("%p, %s\n", device, debugstr_a(name));
>> +    if (!pvkGetDeviceProcAddr(device, wine_vk_native_fn_name(name)))
>> +        return NULL;
>> +
>>       if ((proc_addr = X11DRV_get_vk_device_proc_addr(name)))
>>           return proc_addr;
>> @@ -453,6 +466,9 @@ static void 
>> *X11DRV_vkGetInstanceProcAddr(VkInstance instance, const char *name)
>>       TRACE("%p, %s\n", instance, debugstr_a(name));
>> +    if (!pvkGetInstanceProcAddr(instance, wine_vk_native_fn_name(name)))
>> +        return NULL;
>> +
>>       if ((proc_addr = X11DRV_get_vk_instance_proc_addr(instance, name)))
>>           return proc_addr;
>>
>