| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Updated MoltenVK to latest known good
|
|
|
|
|
|
The previous implementation did not differentiate when a single queue
supported a specific surface. It will now print 'true' if all surfaces
support presentation with a queue, 'false' if none, and otherwise prints
each surface type's support if they aren't uniform in value.
This makes the info regarding which queue supports presentation on a
surface much easier to validate the correctness of.
|
|
Previous implementation would always run the function 3 times no matter what.
|
|
|
|
Instead of unconditionally querying properties twice,
only call the query again if the given output struct has array members.
Also fix the case where if there are no output structs with array properties (VulkanSC),
then don't end up with unresolved external symbols.
|
|
|
|
Needed custom implementation of vkCreatePipelineBinariesKHR in mock_icd
|
|
This option restores behavior that prints out structures which have
been promoted to core vulkan and their members combined into the
VkPhysicalDeviceVulkanXXX structs.
For example, VkPhysicalDeviceVulkan12Features contains all of the features
that VkPhysicalDeviceVulkanMemoryModelFeatures does, so only the former
needs to be output. However, some users will want both structs present
regardless and the --show-promoted-structs option allows that behavior.
This commit also removes redundant checks when iterating the pNext chains
while printing, as those same checks occur when the chain is being created,
removing duplicate code-gen and runtime checks.
|
|
wl_display_dispatch() may not process all events.
|
|
|
|
These surface extensions shouldn't be enabled if the metal
surface extension is present. This fixes a deprecation notice
the validation layers emit - as well as not provide redundant
information (since the metal extension is a direct replacement
for the mvk/ios extensions).
|
|
ALooper_pollAll is removed in NDK r27, switch to using ALooper_pollOnce
instead. Since it was already being called in a loop, all we have to do
is change the function name.
Fixes #1015
|
|
|
|
Cube is using a deprecated API for r27
https://github.com/KhronosGroup/Vulkan-Tools/issues/1015
|
|
Rollback MoltenVK to 1.2.9 from 1.2.10 due to stability issues
|
|
Updated known_good for latest MoltenVK
|
|
|
|
|
|
|
|
There are now multiple structs in pNext chains which require calling once to
get the size, allocating memory, then calling again to get the data. The
codegen now handles such functions, although not very cleanly.
|
|
|
|
Closes #959
|
|
The Vulkan-Headers and volk dependencies are always required, while the Vulkan-Loader
dependency is only necessary on APPLE hardware or when testing is enabled.
|
|
|
|
|
|
|
|
|
|
|
|
Enable the Linux/BSD paths that are not for Wayland also for GNU/Hurd.
|
|
It is valid for most of the Unix platforms, including GNU/Hurd.
|
|
Since volk is used, this include is no more needed explicitly.
|
|
See also:
Cannot build DEMOS.sln
https://gitlab.khronos.org/vulkan/Vulkan-SDK-Packaging/-/issues/1417
Volk requires that VK_NO_PROTOTYPES be defined before vulkan.h or vulkan.hpp is
included. Currently, the various flavors of vkcube hide this definition in the
cube/CMakeLists.txt file, which can confuse users who may copy the source for their
own use, and may require investigation to figure out why it doesn't "just work".
This change makes the #define explicit in the cube.c and cube.cpp source files,
which should both be clearer and be more similar to how most applications use Volk.
cube/CMakeLists.txt:
- remove
- vulkaninfo already had a #VK_NO_PROTOTYPES somewhere
- macOS DemoViewController includes of mvk_vulkan.h are confusing Volk and don't seem to be necessary,
since cube.c and cube.cpp #include vulkan.h and vulkan.hpp, respectively
|
|
Updated with latest MoltenVK tag
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Android builds of vkcube have been using the same shaders as the desktop
build since the NDK build was removed.
|
|
|
|
|
