Allows the tests to set the exact values vk-bootstrap will receieve
rather than require a real vulkan driver to be present on the system.
This is done by creating a `vulkan-1.dll` on windows that implements
all of the API functions, and on linux/macOS intercepts `dlsym` to
make it return the mock's functions.
VulkanSC was added to vk.xml and broke the generation script. This was an
easy fix, just needed to specify which API the version information should
use. Still, took the time to cleanup the code by running pylint and fixing
anything it warned about.
Previously, the code tried to make it possible to use vk-bootstrap with
Vulkan 1.0 headers. This makes the code complicated and left several paths
completly untested. Making Vulkan 1.1 headers allows for simpler code.
Note: This does NOT make a Vulkan 1.1 capable driver required to use
vk-bootstrap. This only requires that the headers used to build
vk-bootstrap are 1.1 or above.
Cleaned up the logic for handling when to use the core, the KHR, and when
not to use vkGetPhysicalDeviceFeatures2. This should make the logic easier
to follow with better names, and handle a few cases that were wrong, like
not caring if the extension is enabled by vulkan 1.1 is not.
If an application uses add_required_extension_features but also adds
a VkPhysicalDeviceFeatures2 to the pNext chain of VkDeviceCreateInfo,
this will now result in an error. The reason is that Vk-Bootstrap is
perfectly capable of adding VkPhysicalDeviceFeatures to the pNext
chain itself but only if the user didn't already add their own.
The XML added requirements which had multiple extensions be required.
This broke the existing autogen because it assumed there would only
be one extension required at a time. The fix is easy, split the
requirement string into a list then add it to the list of requirements.
- format feature flags should not be used, as set_image_usage_flags is a better alternative
- split format search in two phases (desired / best)
- explain usage of is_unextended_present_mode for basic validation
This allows the image views to be created with the same usage flags as
the swapchain. This necessitated passing in version information down
the object creation chain because to add the usage flags requires 1.1
to be enabled.
Some alias types are not pure promotional types, and need to be excluded from the
code which replaces types with their alias.
This may be a stopgap solution for a larger underlying problem of not being able to
determine if an alias is promoted or not.
Makes it easy to use vulkan on macOS since vk-bootstrap will auto-enable it if
the extension is available.
This can be disabled by calling `disable_portability_subset`.
This lets applications see all the devices which meet a set of requirements
then pick whichever of these devices is most appropriate for them. The intent
is for applications that want to find all the suitable devices and let the
user of the application pick the physical device from a list.
The version macro selector was looking for the constant VK_API_VERSION instead of the correct macro VK_MAKE_VERSION. Replaced the preprocessors accordingly.
Allows apps to find all the extensions available. If they do not enable the layer the
extension comes from, the instance will fail creation due to missing the layer that
implements the extension.
The autogen accidentally took a reference instead of a copy, so all the aliases
were using the same data as the base type, leading to incorrect macro guards.
In addition, any aliased function would use the base funciton's types, leading
to incompatibilities when compiling with older headers. The chosen solution is
to always use the alias type if it exists.
Lastly, NVX was added to a blacklist, no functions from NVX extensions will be
generated, since they are experimental in nature.
Previously it was difficult to know which extensions were actually enabled.
If an app wanted to optionally enable certain extensions it was hard to know if
they got enabled.
A recent loader change results in NULL being returned for pre-instance functions
if the instance handle passed into vkGetInstanceProcAddr isn't NULL. This was
reverted but going forward this change will require updating applications.
- Replaced GenericFeaturesPNextNode's set function with a template constructor.
- Moved validation asserts for extension feature structs into source.
- Dispatch table now generates in UTF-8.
Previously, vk-bootstrap didn't support calling vkGetPhyicalDeviceFeaturesKHR
on devices which support the extension. This makes vk-bootstrap more amenable
to 1.0 only hardware, like rasberry pi's.
- Internal tables are now populated through internal vkGetDeviceProcAddr rather than through passed dispatch table
- Device now has its own internal table as well.
- Cleaned up get_proc_addr functions and removed internal functions that are now held in a types internal tables.
