| Age | Commit message (Collapse) | Author |
|---|
|
DeviceLimits
|
|
|
|
Refactored DeviceLimits to store maps in the layer_data struct in order to reduce map look-ups.
Added new checks and documentation related to Features checking. If specific features are requested at vkCreateDevice() time by the app, then they are checked against the supported features of the physical device. Requesting unsupported features cause validation errors. If the user did not query features and is asking for any unsupported features, then a warning message recommending that they query features is also printed along with the error(s) for unsupported features.
|
|
|
|
|
|
Don't use object as dispatchable after destroy.
|
|
The purpose of this layer is to validate the application's usage of WSI, not to
validate the driver's conformance with WSI.
|
|
|
|
|
|
Windows compiler won't allow static on specialization.
|
|
Image layer had been calling imageMap.reserve to recover from memset of imageMap.
That is no longer needed.
|
|
get_my_data_ptr was zeroing data after using new.
It should leave the layer data as its constructor set it.
|
|
|
|
|
|
|
|
Windows doesn't define snprintf so we need to #define
it to what Windows does use in our SDK platform header.
|
|
|
|
Integrate changes from merge request #5.
|
|
|
|
|
|
|
|
|
|
Validated that these are not called in an active RP, added validation tests
for these and some other renderpass-dependent APIs.
|
|
Instance and device specific objects.
Refcounts of duplicate non-dispatchable objects.
|
|
ImageView
|
|
|
|
Moved checks related to ImageView creation from the driver into the Image layer. Added image.h header and ENUMs for various Image error types, along with documenation for each type.
|
|
This patch contains fixes to structure layout for draws.
These structures are accessed by HW and must be in a specific order.
This change also includes reordering of the parameters for
vkCmdDraw and vkCmdDrawIndexed.
|
|
The header only had one include in it.
|
|
The generator was getting confused by the argument list of
vkCmdDrawIndexed(cmdBuffer, indexCount, instanceCount, firstIndex, vertexOffset, firstInstance)
It saw instanceCount and then wanted to expand firstInstance.
I had to terminate it match search after one miss to stop it recognizing firstInstance as an array of Instance items.
|
|
Cannot reach these points with an invalid instance value.
|
|
Release mutex when last instance is destroyed.
|
|
Don't complain about lingering objects more than once.
Once their device or instance is gone don't keep their information around.
This should change dramatically to have independent maps for each instance/device.
Don't complain about other devices remaining in vkDestroyDevice.
|
|
|
|
|
|
These checks are covered by ObjectTracker.
|
|
This commit covers phase 2 of the removal of dynamic
state objects. Now, an application can include an array
of VK_DYNAMIC_STATE_* values that tell the driver
which PSO dynamic state elements to use.
I.e. if VK_DYNAMIC_STATE_LINE_WIDTH was specified in the
pDynamicState array then the ICD should use the lineWidth
defined at PSO create time and ignore any set using
vkCmdSetLineWidth.
To accomplish that the driver will make a copy of the
dynamic state specified in the PSO as well as a bitmask
of the affected state. When vkCmdSet* is called, the
driver will check if a PSO override is current and ignore
the call if so.
At PSO bind time the command buffer's dynamic state
will be updated and the PSO override bitmask set so that
any future vkCmdSet*'s will be appropriately ignored.
TODO: Validation layer should probably indicate a warning
if app tries to do vkCmdSet on state defined by the PSO.
|
|
|
|
This is phase 1 of the replacement of dynamic state objects.
In this commit we remove the create and destroy functions
and rename the CmdBind... with CmdSet...
|
|
A layer can use will_log_msg to determine if any output
at the designated notification level(s) will be sent
somewhere. This will allow a validation layer to avoid
high compute validation tests if the results are going
to be looked at.
|
|
|
|
|
|
Moved validation into device_limits layer, created layer utils
files for format utilities, added validation tests, updated validation
doc, updated test framework to call GetPhysicalDeviceProperties.
|
|
|
|
cJSON not thread safe. Add a lock around cJSON calls so instance
extension and layer queries don't need the big lock. This allows layers
to do queries from CreateInstance.
|
|
vkCreateDevice()
|
|
Also enabling DeviceLimits and Image layers when --validate option selected.
|
|
Primarily, this deals with new binding field in metadata rather than having
location field do double duty.
|
|
A structure's sType was missed in the grand renaming that took place a while
ago, and was fixed with the latest WSI header changes.
|
|
|
