| Age | Commit message (Collapse) | Author |
|
Change-Id: Ic1d0f67ec2f8ca438c4bd546e6beb4d565daf89a
|
|
|
|
1) Ninja build and QTCreator IDE support
2) Out of tree builds that aren't based on using update_external_sources and build_windows_target scripts
3) Allowing the developer to use non-default locations for glslang and SPIRV-Tools binaries and sources
4) Fix linux build when developer chooses not to rename SPIRV-Tools to spirv-tools
Change-Id: Ib6118c47dc780e6721ec0538aae1a6ee444eed78
|
|
Mirroring commit d288cb9a, stop using directories outside
of our project location to host external files. Small
changes to update_external_sources_android.sh/bat to find
glslang, spirv-tool, shaderc in "external" instead of ".."
|
|
Also add info about need for layer work.
|
|
|
|
Previously we'd assumed these were dense. It is however valid to provide
VK_ATTACHMENT_UNUSED for a color attachment reference.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
|
|
Signed-off-by: Chris Forbes <chrisforbes@google.com>
|
|
Save VkInstance in layer_data so that we query vkCreateDevice with a valid
instance.
|
|
Handle device commands as well. Move handling of interface functions to
v0's vkGetInstanceProcAddr.
|
|
s/vkEnumerateInstanceDeviceProperties/vkEnumerateDeviceExtensionProperties/.
|
|
Split command intercepting out to intercept_core_instance_command.
|
|
They are defined but unused.
|
|
It returns the function pointers for all intercepted core device commands.
Call intercept_core_device_command from GetDeviceProcAddr. Assert that
device is valid in GetDeviceProcAddr.
|
|
Call down the next layer unless pLayerName is device_limits.
|
|
Replace dl_global_layers with global_layer. There is no plan to define
more than one layer in the same namespace.
|
|
Drop VK_LAYER_EXPORT as well.
|
|
Put all layer fucntions/data into device_limits namespace. I had to add
some wrappers to make everything work.
|
|
Move them toward the end of the file.
|
|
Save VkInstance in layer_data so that we query vkCreateDevice with a valid
instance.
|
|
Handle device commands as well. Move handling of interface functions to
v0's vkGetInstanceProcAddr.
|
|
Split command intercepting out to intercept_core_instance_command.
|
|
It returns the function pointers for all intercepted core device commands.
Call intercept_core_device_command from GetDeviceProcAddr. Assert that
device is valid in GetDeviceProcAddr.
|
|
Call down the next layer unless pLayerName is image.
|
|
Replace pc_global_layers with global_layer. There is no plan to define
more than one layer in the same namespace.
|
|
Drop VK_LAYER_EXPORT as well.
|
|
Put all layer fucntions/data into image namespace. I had to add some
wrappers to make everything work.
|
|
Move them toward the end of the file.
|
|
Change-Id: I889a00132af4c6f8db944ed21f03290ee8f14063
|
|
DescriptorSetCompatibility test hits all of the PIPELINE_LAYOUTS_INCOMPATIBLE
validation errors so just needed to update documentation.
|
|
A number of test names in vk_validation_layer_details.md were incorrect
so updated names as appropriate and removed other names for tests that
were not in layer_validation_tests.cpp.
|
|
This adds a set of checks to the document validator that will iterate through
all of the checks documented in vk_validation_layer_details.md and compare the
names of the test(s) for each check against the names of the actual tests in
layer_validation_tests.cpp.
Currently, all checks not ending in "_NONE" that don't have a valid test
recorded in the table will report a warning with a final count of the warnings
given. There are initially 131 such warnings, but I suspect at least a handful
of them are implemented but just need to be correctly documented.
Once all the test holes are filled in we can promote the warnings to errors
to make sure that any new checks are both documented and have a valid test
implemented.
Rough support of wild-carded testnames such as "CreatePipeline*NotConsumed" is
included.
|
|
Change-Id: I77f771943aac77d9bafc9d33336adfb009d1b616
|
|
Change-Id: I9cca32a98a1b984b44eca42e4805461c220ae04f
|
|
All dynamic state binding cases had tests, but were not correctly documented.
This changes merges them all into a single test, DynamicStatesNotBound, to speed
test time and documentation was updated to record that test for the 9 related tests.
|
|
At the time of a Draw or Dispatch, a pipeline must be bound to the
command buffer. This adds a check to verify that.
Added tests for both draw and compute cases to verify the check.
|
|
Have update_external_sources create a sub-directory "external" to
fetch these components into. This way the update_external_sources
does not potentially overwrite files outside the directory tree.
Change-Id: Ia03d1fb9e4dad9d9db26a46e596fb1bf24fcb033
|
|
Change-Id: I6fb9d1d0ffd8b51988bc0768afedc0e9fb510cc5
|
|
Adds an AMD rasterization order extension
Change-Id: I62d17b0437e190da4d44f65a59b8cc246f05b567
|
|
Because the WSI functions are statically exported (i.e. no need to call gpa()
functions), the swapchain layer is supposed to issue an error if any WSI
function is called without its corresponding extension being enabled. For
example, the VK_KHR_swapchain extension must be enabled before the
vkCreateSwapchainKHR() function is called.
|
|
|
|
Change-Id: Ic45e21bd3137dc0474c59f0f4cf9331f070dac20
|
|
|
|
|
|
After lh and rh layout descriptorCounts were verified to be the same, the
next check was testing the descriptor count for each binding in the lh binding
against the entire descriptor count for the rh layout. The check should be made
just against the descriptor count for the specific binding we're checking in
the rh layout.
|
|
Change-Id: Icd9c2b63f8e59b39ad0349bc5a4456cbd88a5803
|
|
Change-Id: Iaa1afa361d29022388005c173054f714c64fb7d7
|
|
Move the get_dispatch_key(instance/device) call before calling the
driver's vkDestroyDevice/vkDestroyInstance functions in
unique_objects. This addresses an access violation error reported
by Application Verifier on Windows due to the previously destroyed
instance/device handle being dereferenced.
Change-Id: I151d42fb8a897d9d3b3c988813108b0b6b5b6b0d
|
|
|
|
|