Change-Id: Id6b3a7845ae194a18cc01b0f9d2fa6b178867b76

Set some default mock icd values for vkGetPhysicalDeviceProperties2
query of VkPhysicalDeviceDescriptorIndexingPropertiesEXT props.

Update layer_data struct to track safe_VkPhysicalDeviceFeatures2 struct
instead of VkPhysicalDeviceFeatures. Currently just doing previous
checks using the VkPhysicalDeviceFeatures member of GPDF2 struct. Will
follow-on with checking against extension structs.

In PHYSICAL_DEVICE_STATE track safe_VkPhysicalDeviceFeatures2 instead
of the old VkPhysicalDeviceFeatures. Initially just passing the
VkPhysicalDeviceFeatures of VkPhysicalDeviceFeatures2 wherever this was
used. This is laying groundwork to start using/comparing extended
features in core validation.

Make sure we generate vk_typemap_helper.h for mock icd.

Checks for TRANSFER_SRC / TRANSFER_DST formats were added to:
- vkCmdClearColorImage
- vkCmdCopyImage
- vkCmdResolveImage
- vkCmdCopyImageToBuffer
- vkCmdCopyBufferToImage
These checks require VK_KHR_maintenance1 or Vulkan 1.1.

Existing checks were adapted to use the new helper function:
- vkCmdClearDepthStencilImage
- vkCmdBlitImage

Fix an oversight when recording a 3D image subresource layout transition
that is set by way of a 2D or 2D_ARRAY image view.

Change-Id: Ifbf7645fb2705fef6f82030f4e3e1da0e9da2f2c

Added output for a MacOSSurface in vulkaninfo
and modified the CMake files to include the
new files needed for this change:

metal_view.m Objective-C code to hook into Metal
metal_view.h metal_view.m's header file

Change-Id: Ib7f9185a86703b2a300b2ce3be46f130c90d1b6b

Update mock ICD generator so that if we have an extension that's been
promoted to core, the core version will call any existing custom
implementation of the extension function.

Update Features query to detect the two existing extension structs
(VkPhysicalDeviceDescriptorIndexingFeaturesEXT and
VkPhysicalDeviceBlendOperationAdvancedFeaturesEXT) and enable all of
those features when those queries are present.

Change-Id: Ie09bfcd61c57024f4c7a0ff884001989fec2a565

Change-Id: Id8f0301c4f208d185abe2cc3df71021cb4aa29b1

Change-Id: I3b672a9d54e1509b550407f1b4ffae58b48beb85

Before loading a manifest file, its full path is strcmp() with already
loaded ones in order not to load the same file twice. While being simple
and efficent, this mechanism is not able to detect subtle duplicates
like these:

/usr/share/vulkan/icd.d/intel_icd.x86_64.json
/usr/share//vulkan/icd.d/intel_icd.x86_64.json

This patch ensure that searched paths do not contains such duplicated
directory delimiting characters in order to avoid this kind of problem.

Fixes #2331
Fixes #2629

This test as written was guaranteed to be skipped. It would skip the
test if the first graphics queue family found had a
minImageTransferGranularity with a width, height, or depth of less
than 4. The spec guarantees that any graphics queue family's min
transfer granularity will be 1 for the width, height, and depth. The
rewritten test is now only skipped if no queue supporting copy
operations is found with a large enough granularity.

Also the usage flags on the src and dst image needed to be fixed.
Additionally, the extent of the copies dst and src images were changed
to always be multiples of the min transfer granularity.

Finally, I added a device profile that has a queue with a large enough
min transfer granularity to not skip this test.

Simplified this test by using VkImageObj in place of direct vulkan
calls. Additionally added a fake multiplane image format to the geforce
940m device profile so that this repo would have a device profile that
doesn't skip this test.

Previously buffer_validation covered part of this, but not everything.

V2: - Fix repeated ATTACHMENT_RESOLVE where I meant ATTACHMENT_PRESERVE
    - Fix camelcase noise in a bunch of places
    - Rename VkAttachmentReference& variables to attachment_ref to avoid
      attachment.attachment etc.
    - clang-format tidy-up
    - Move positive case to VkPositiveLayerTest

Previously this would contribute to skip /after/ the call, which
accomplishes nothing.

Simplified this test by using VkImageObj in place of direct Vulkan
calls.

Change-Id: Ie74c686af656e346789b30ec1aba41457170c6d2

Change-Id: I43886920e5ce1feb2e293ed46a3e8c23a217ca69

Change-Id: If11059dca9716a90134d9d362351634fa07a8ecd

Piers found a couple of typos in the device extension handling.

Change-Id: I0068f07bdbfca9715cac319feb56a3fc2b6402fb

Simplified this test by using test helpers in place of direct Vulkan
calls. Also added a TODO to switch this test from checking for an error
string to checking for an error VUID when the specific error is given a
VUID.

Add VkDisplayKHR and VkDisplayModeKHR tracking to
VK_LAYER_LUNARG_object_tracker.

Change-Id: I63004cfe734793593bea12700ea9cd1bdefeab0d

- Use appropriate VUIDs for all checks
- Update tests to check for VUIDs instead of message substrings
- Image granularity tests should be also be run on compressed images.
  They're *mostly* redundant with existing compressed-format-only checks,
  but I can't convince myself that they're *always* redundant, and the
  spec doesn't say to skip them.
- Removed comments that the checks should be moved elsewhere; IMO they're
  fine where they are, but I welcome second opinions.
- Compressed-format alignment checks should also apply to _422 single-plane
  formats.

Modified dispatch table helper to replace null device extension
function pointers with an appropriate stub function.

Change-Id: If5e22333f084d3f41c0eefdf3e82c9060b872e8c

The ActualTimeLate function in cube.c is supposed to check if a
VkPastPresentationTimingGOOGLE structure indicates that the
presentation happened within a refresh-cycle duration of the
desired presentation time.

However currently ActualTimeLate will always return false, because
it is calculating the presentation deadline as `actual + rdur`.
This change calculated the presentation deadline as
`desired + rdur`.

Reviewed-by: Andy Ritger <aritger@nvidia.com>

New tests to exercise new checks for indexed draw count and offset.

Add VkRenderFramework::DestroyRenderTarget() and call it from
VkTriangleTest so that VkTriangleTest cleans up after itself,
allowing it to be called multiple times within a single framework
instantiation.

Introduce VkBufferObj as a typedef for vk_testing::Buffer.

Fixes #2448

Change-Id: Iab513c7f50ef390c210ea76e7ef3f67d7e3d40f8

Check that the first index and index count are within the bounds
of an index buffer for DrawIndexed, taking into account any
offset specified in the binding of the index buffer.

Fixes #2448

Change-Id: I47a1acbf0bb0e586692c75380d8d43a1c72450fd

Simplified this test by removing an uneccessary creation of a VkImage.
Instead, the test now just directly allocates unmappable memory and
attempts to map it.

Implementation was incomplete, and had been reduced to producing
non-specified warnings.

Change-Id: I34a4bbf1f572f337b5f9f9316e9717f52645d8dc

Change debug log message to use correct VkObjectType to
VkDebugReportObjectTypeEXT method, avoiding the index out-of-bounds
lookup that caused a SEGV.

Change-Id: Ibb6e485eccb36ff0e9b49f0722d8f2d0bf660982

To avoid Destroy/Create and Free/Alloc race conditions (when a driver
returns a destroyed(freed) handle between the PreCall and PostCall
phases of a validation intercept routine, PostCallRecord functionality
has been moved to PreCall.  This affects the following calls

	FreeMemory
	DestroyFence
	DestroySemaphore
	DestroyEvent
	DestroyQueryPool
	DestroyBuffer
	DestroyBufferView
	DestroyImage
	DestroyImageView
	DestroyPipeline
	DestroySampler
	DestroyDescriptorSetLayout
	DestroyFramebuffer
	DestroyRenderPass
	FreeDescriptorSets

Did not alter the destroy functions for Instance, Device, or the Debug
or layer extensions.

Change-Id: I5133982eaa1a0c94283fa922d9c3bcc90b5a6223

Extend check for load operation vs first time use image layout to
new read-only depth stencil image layouts.

The new layouts were added in VK_KHR_maintenance2 / Vulkan 1.1
and follow similar restrictions to VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL.

Spec:

"
For any member of pAttachments with a loadOp equal to VK_ATTACHMENT_LOAD_OP_CLEAR,
 the first use of that attachment must not specify a layout equal to
VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL.

For any member of pAttachments with a stencilLoadOp equal to VK_ATTACHMENT_LOAD_OP_CLEAR,
the first use of that attachment must not specify a layout equal to
VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL.
"

Fixes #2614

Fixed the following issues with the previous checks:

- Only alpha blend factors were checked
- The checks did not use the individual VUIDs for
  each blend factor field
- The check was being run at pipeline bind time,
  not pipeline creation time

The layers files fit are not really configuration files and are rarely
edited by users. So, let's move the install location into the
data root directory.

Signed-off-by: Nick Sarnie <sarnex@gentoo.org>