Chris Forbes [Mon, 26 Sep 2016 02:18:57 +0000 (15:18 +1300)]
layers: Tweak naming of enums moved from SwapChain to CV
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 06:34:20 +0000 (18:34 +1200)]
layers: Move image-not-acquired etc from swapchain to core validation
Now that Core Validation tracks the acquisition state of every swapchain
image, we can do this.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 05:11:06 +0000 (17:11 +1200)]
layers: Add tracking of acquired state to Core Validation
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 04:49:47 +0000 (16:49 +1200)]
layers: Add IMAGE_NODE::acquired flag
This is necessary for moving the image ownership stuff across from
swapchain
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 04:40:27 +0000 (16:40 +1200)]
layers: Move ACQUIRE_NO_SYNC check from Swapchain to Core Validation
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Mark Young [Mon, 26 Sep 2016 17:38:46 +0000 (11:38 -0600)]
loader: gh971 EnumPhysDev return VK_INCOMPLETE
vkEnumeratePhysicalDevices should return VK_INCOMPLETE if the
following conditions are met:
a) pPhysicalDeviceCount < number of physical devices available
b) pPhysicalDevices != NULL
The code was actually in place, but was not getting hit.
Change-Id: I725fec5e57239f990590d9e447bb057ebc3502a1
Mike Stroyan [Wed, 21 Sep 2016 17:19:07 +0000 (11:19 -0600)]
build: Enable parallel windows build
Mike Stroyan [Wed, 21 Sep 2016 20:05:35 +0000 (14:05 -0600)]
build: Fix parallel build of layers tests
Make json copy targets depend on layer targets to first create destination directory.
Make header generation go through a custom target.
Mike Stroyan [Thu, 22 Sep 2016 20:55:16 +0000 (14:55 -0600)]
build: Make windows build of demos parallel build safe
Use direct output to SPIRV file names to prevent collisions.
Mike Stroyan [Fri, 16 Sep 2016 15:45:14 +0000 (09:45 -0600)]
build: Fix parallel build of layers
Change layer dependencies to prevent collisions of scripts creating headers.
Use dependency on one custom target for each group of headers built by a custom command.
The layers json targets need to depend on the layers targets.
They expect to copy into a directory created by those targets.
Make dependency of vk_safe_struct.cpp indirect through generate_vk_layer_helpers.
That will wait for a single build of the generated files.
Mike Stroyan [Tue, 20 Sep 2016 15:08:18 +0000 (09:08 -0600)]
build: Fix ImageMagick_Magick++ search
One cmake test was looking at ImageMagick_MagickWand_INCLUDE_DIR
instead of ImageMagick_Magick++_INCLUDE_DIR
Mike Stroyan [Tue, 20 Sep 2016 15:05:28 +0000 (09:05 -0600)]
build: UNSET cache variables for FIND_IMAGEMAGICK_API
The way that the ImageMagick component search calls find_path
multiple times with the same variable but different names
does not expect old cache values in the variable.
Chris Forbes [Thu, 22 Sep 2016 02:55:13 +0000 (14:55 +1200)]
layers: Remove spurious using BASE_NODE::*.
Inheritance is all public (except for FRAMEBUFFER_NODE, which ought to
have been), and there's no template inheritance involved, so these
didn't do anything interesting.
Fix FRAMEBUFFER_NODE to publicly inherit BASE_NODE, and drop the noise
everywhere.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 06:53:59 +0000 (18:53 +1200)]
layers: Get rid of some junk in RENDER_PASS_NODE::ctor
Many moons ago, ShaderChecker used this state. Now, it digs directly in
the CreateInfo, so there are no clients. Nuke it.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 22 Sep 2016 06:51:58 +0000 (18:51 +1200)]
layers: Explicitly delete IMAGE_NODE, BUFFER_NODE copy constructors
Previously we defined copy constructors that would almost certainly lead
to bugs if these objects were copied (they copied the atomic values,
etc). The layer doesn't copy these, just disallow it.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Alex Smith [Fri, 23 Sep 2016 11:18:51 +0000 (12:18 +0100)]
layers: Fix VerifyCopyUpdateContents on combined image sampler descriptors
A missing break at the end of the ImageSampler case results in falling
through to the Image case, leading to spurious validation errors such as:
vkUpdateDescriptorsSets() failed copy update from Descriptor Set 0x108
to Descriptor Set 0x137 with error: Attempted copy update to image
descriptor failed due to: Invalid VkImageView: 0x100000000
Józef Kucia [Fri, 23 Sep 2016 10:45:04 +0000 (12:45 +0200)]
layers: Prevent reading uninitialized "count" when validating array parameters
Cody Northrop [Thu, 22 Sep 2016 20:39:16 +0000 (14:39 -0600)]
android: Add Android and C++ support to a couple files
These are needed downstream in VulkanTools for Android support.
Cody Northrop [Thu, 22 Sep 2016 19:55:49 +0000 (13:55 -0600)]
demos: Enable arm64 build of Smoke
Cody Northrop [Thu, 22 Sep 2016 19:42:22 +0000 (13:42 -0600)]
demos: Update Smoke permissions for trace/replay
Cody Northrop [Thu, 22 Sep 2016 17:08:37 +0000 (11:08 -0600)]
demos: Fix Smoke build for Android
It doesn't appear that this ever worked.
Cody Northrop [Thu, 22 Sep 2016 15:42:23 +0000 (09:42 -0600)]
demos: Update gradle properties for smoke
Cody Northrop [Thu, 22 Sep 2016 14:18:52 +0000 (08:18 -0600)]
demos: Update smoke build script to be automatable
Cody Northrop [Thu, 22 Sep 2016 19:32:04 +0000 (13:32 -0600)]
toolchain: Use multiple processes to build glslang
Tobin Ehlis [Thu, 22 Sep 2016 16:52:00 +0000 (10:52 -0600)]
layers: Distinguish never-bound from un-bound memory cases
Fixes #964
Added special MEMORY_UNBOUND handle (0xF..FE) that indicates that memory
bound to an object has been freed. When attempting to bind memory or
checking for bound memory, distinguish the never-bound case from the
memory un-bound case.
For sparse binding case allow for memory to be re-bound.
Update tests to account for new error messages.
There's a sliver of exposure here if an actual memory handle is MEMORY_UNBOUND.
We could remove that exposure by never having unique_objects return MEMORY_UNBOUND
as a handle. I believe the exposure is small enough that we don't need to do that,
but am open to other opinions.
Tony Barbour [Wed, 21 Sep 2016 19:10:56 +0000 (13:10 -0600)]
demos: Change cube to use FIFO present mode
And throttle using fences
Change-Id: I13cd622f9a401fdb44bf7b74343589b70e340b77
Cody Northrop [Thu, 22 Sep 2016 13:41:08 +0000 (07:41 -0600)]
demos: Update Android documentation
Cody Northrop [Thu, 22 Sep 2016 13:32:22 +0000 (07:32 -0600)]
demos: Give Cube and Tri distinct app names
Cody Northrop [Thu, 22 Sep 2016 13:22:28 +0000 (07:22 -0600)]
demos: Clean up APK creation for cube and tri
Instead of moving manifest files, run the following commands:
android update project -s -p . -t "android-23"
ant -buildfile cube debug
ant -buildfile tri debug
And the APKs can coexist:
./cube/bin/NativeActivity-debug.apk
./tri/bin/NativeActivity-debug.apk
Tobin Ehlis [Thu, 22 Sep 2016 00:32:11 +0000 (18:32 -0600)]
tests: Fix unexpected errors in InvalidMemoryAliasing
Memory bindings are immutable so create a separate image2 to bind
to the second memory allocation instead of re-binding same image.
Tobin Ehlis [Wed, 21 Sep 2016 23:32:49 +0000 (17:32 -0600)]
tests: Add DescriptorImageUpdateNoMemoryBound test
Attempt to update an image descriptor that doesn't have memory bound.
Tobin Ehlis [Wed, 21 Sep 2016 23:32:26 +0000 (17:32 -0600)]
tests: Update error monitor to handle multiple expected errors
Tobin Ehlis [Wed, 21 Sep 2016 21:09:45 +0000 (15:09 -0600)]
layers: Clean up no memory bound checks
When memory is freed make sure to clear bindings in associated objects.
Also clean up reporting of no memory bound errors. Old assumption was that
no memory had every been bound, but it's also possible to hit such errors
if the bound memory was freed prior to operation of interest.
In descriptor error cases where no memory is bound, add a final error string
as clarification since descriptor errors are built up of multiple strings
with details from each function call in the tree.
Tobin Ehlis [Wed, 21 Sep 2016 15:19:15 +0000 (09:19 -0600)]
layers: Verify memory is bound to images used to update descriptors
Fixes #663
For vkUpdateDescriptorSets() calls that include an image, verify that memory is bound to the image.
Tobin Ehlis [Thu, 22 Sep 2016 13:30:05 +0000 (07:30 -0600)]
layers: A few minor clean-ups from code review
Better use of auto and killing some unneeded code.
Tobin Ehlis [Wed, 21 Sep 2016 20:33:41 +0000 (14:33 -0600)]
tests: Add FramebufferImageInUseDestroyedSignaled test
Create a framebuffer, bind it to cmd buffer as part of renderpass and then
destroy the child image while in-flight to verify that correct error is
triggered.
Tobin Ehlis [Wed, 21 Sep 2016 20:31:52 +0000 (14:31 -0600)]
tests: Make initial renderpass color attachment image layout UNDEFINED
This is more flexible for handling various initial image layouts and makes it
easier to add tests that avoid unexpected errors.
Tobin Ehlis [Wed, 21 Sep 2016 20:28:42 +0000 (14:28 -0600)]
layers: Add in-use check at DestroyImage() time
Flag error if an in-use image is destroyed.
Also clean up a couple minor issues related to binding objects to command buffer.
Tobin Ehlis [Mon, 19 Sep 2016 22:23:01 +0000 (16:23 -0600)]
tests: Fix incorrect descriptor type
SamplerInUseDestroyedSignaled test was incorrectly using VK_DESCRIPTOR_TYPE_SAMPLER
instead of VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER
Tobin Ehlis [Mon, 19 Sep 2016 22:15:45 +0000 (16:15 -0600)]
tests: Add ImageViewInUseDestroyedSignaled test
Bind combined image/sampler descriptor to cmd buffer, then destroy
the imageView while cmd buffer in-flight and verify in-use error.
Tobin Ehlis [Mon, 19 Sep 2016 20:20:37 +0000 (14:20 -0600)]
layers: Add in-use check for renderPass
At DestroyRenderPass() time verify that renderPass is not in-use.
Tobin Ehlis [Mon, 19 Sep 2016 20:02:58 +0000 (14:02 -0600)]
layer: Add in-use validation for imageView
Bind imageView from a descriptor to the command buffer at CmdBindDescriptorSets
time and check for an in-use imageView at DestroyImageView time.
Also use new AddCommandBufferBindingImageView() helper to bind children of
framebuffer in AddFrameBufferBinding().
Tobin Ehlis [Mon, 19 Sep 2016 19:10:37 +0000 (13:10 -0600)]
layers: Add in-use check for imageView
Add DestroyImageView time, make sure imageView is not in use.
Tobin Ehlis [Wed, 14 Sep 2016 21:26:09 +0000 (15:26 -0600)]
layers: Add binding for framebuffer children
When BeginCommandBuffer and CmdBeginRenderPass are called, need to bind
not only the framebuffer, but all of the framebuffer's children objects
to the command buffer.
This includes the imageViews that are in the attachments as well as their
children and the renderpass.
Added this binding and included IMAGE_VIEW_STATE* in framebuffer
attachment struct for simplicity when creating binding.
Tobin Ehlis [Wed, 14 Sep 2016 17:21:55 +0000 (11:21 -0600)]
layers: Add support code to handle in_use for missing objects
Add case block to in-use Increment/Decrement functions to handle the missing
object types that may be bound to a command buffer. The actual bindings for
these missing types will be added in a future commit.
Here are the types that were added:
VkBufferView
VkImageView
VkDescriptorPool
VkCommandPool
VkFramebuffer
VkRenderPass
VkDeviceMemory
Also added new error enums for all of these type except RENDERPASS which already
had an appropriate error enum.
Unify object BASE_NODE look-up for removeCommandBufferBinding() and
DecrementBoundResources() to use common GetStateStructPtrFromObject() function.
Finally, had to update the state-wrapping class for DescriptorPool, DeviceMemory
and RenderPass to derive from BASE_NODE so that in_use can be tracked.
Tobin Ehlis [Wed, 14 Sep 2016 14:12:08 +0000 (08:12 -0600)]
layers: Update BufferView state tracking
Create BUFFER_VIEW_STATE class to hold VkBufferViewCreateInfo and update related
state tracking. This class derives from BASE_NODE in preparation for tracking
in_use for BufferViews consistent to other object in_use tracking.
Tobin Ehlis [Wed, 14 Sep 2016 14:02:49 +0000 (08:02 -0600)]
layers: Update ImageView state tracking
Create IMAGE_VIEW_STATE class to hold VkImageViewCreateInfo and update related
state tracking. This class derives from BASE_NODE in preparation for tracking
in_use for ImageViews consistent to other object in_use tracking.
Tobin Ehlis [Wed, 14 Sep 2016 13:01:11 +0000 (07:01 -0600)]
layers: Clean up in-use decrement for bound objects
Decrement in-use in a single location via ptr to BASE_NODE for each bound object.
Tobin Ehlis [Wed, 14 Sep 2016 12:55:51 +0000 (06:55 -0600)]
layers: Clean up in-use increment for bound objects
Simplify code with a single error call and in_use increment based on common
base class.
Chris Forbes [Wed, 21 Sep 2016 01:36:19 +0000 (13:36 +1200)]
layers: Fix tracking of WSI fence retirement
Previously we'd note that the fence wasn't submitted to a queue, and so
never call RetireWorkOnQueue, which is the only thing that marked the
fence as retired. In the WSI fence case, we still aren't doing any
tracking of completion of the WSI operation (we should! image ownership
is easily fouled up!) but we can at least retire the fence.
Fixes #954.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 20 Sep 2016 23:18:30 +0000 (11:18 +1200)]
layers: Extract RetireFence helper from WaitForFences/GetFenceStatus.
This does the same thing in both places.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Tony Barbour [Wed, 7 Sep 2016 22:07:56 +0000 (16:07 -0600)]
demos: Use ownership barriers for separate queues in cube
Change-Id: Ia3456405add6ea35b9cc774b2247e6f7d9a80768
Tony Barbour [Tue, 6 Sep 2016 17:40:12 +0000 (11:40 -0600)]
demos: Use queue search from spec example
To favor finding a queue that supports both graphics and present
Change-Id: I0223b0973b4244c006ade31c6515f149f9696ff8
Mark Young [Fri, 16 Sep 2016 16:18:42 +0000 (10:18 -0600)]
loader: More changes for gh181
With Piers' fix, another path was broke because we still needed
to intercept the call before it went to the ICD. Now, with his
change and this change, all paths work.
Also, clean up some code based on comments from Ian and Courtney
reviews.
Finally, update docs to indicate behavior of new loader 3
interface with ICDs.
Change-Id: I2d3f962baffb21c1edeb93b132ffad40f298c8e7
Piers Daniell [Wed, 14 Sep 2016 17:24:36 +0000 (11:24 -0600)]
loader: Fixes for ICD vk{Create|Destroy}SurfaceKHR
Need to keep VkIcdSurface private to the loader.
The definition may conflict with some redefinitions of
VK_DEFINE_NON_DISPATCHABLE_HANDLE and it's not necessary to make public
anyway.
Change-Id: I30e166f0ecce2fd3ea36c745dc073b705973d75f
Mark Young [Tue, 28 Jun 2016 16:52:43 +0000 (10:52 -0600)]
loader: gh181 use ICD for SurfaceKHR
Use the ICD to create and destroy VkSurfaceKHR objects instead
of just performing the work in the ICD. This only occurs if the ICD
exports the appropriate entry-points, and exposes version 3 of the
loader/icd interface.
Change-Id: I5e7bf9506318823c57ad75cf19d3f53fdfa6451e
Mark Young [Mon, 19 Sep 2016 20:25:33 +0000 (14:25 -0600)]
unique_object: Remove unneccessary code.
There is some duplication in querying the WSI extensions in the
unique_object layer. Remove the duplication.
Change-Id: I2f240e8f801bd3d154fce8b1cbc486b62b5a50ee
Gregory Mitrano [Mon, 19 Sep 2016 03:48:29 +0000 (23:48 -0400)]
GH-951 Fixed consistency issue with mip size calculations
Gregory Mitrano [Mon, 19 Sep 2016 01:54:49 +0000 (21:54 -0400)]
GH-951 - Improve validation for minImageTransferGranularity
Added missing BC1_RGBA formats to vk_format_is_compressed function
Added function in vk_layer_utils to get block sizes for compressed formats
Chris Forbes [Fri, 16 Sep 2016 05:58:00 +0000 (17:58 +1200)]
tests: Fix VkLayerTest.InvalidPipeline
- Remove repeated BeginCommandBuffer.
- Ensure enough state is set up for implicit BeginRenderPass for
graphics test.
- Ensure renderpass is finished for compute test.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 05:48:53 +0000 (17:48 +1200)]
tests: Partially fix VkLayerTest.InvalidCmdBufferPipelineDestroyed
- If there is no dynamic state, MUST NOT provide the struct
(dynamicStateCount==0 is not allowed).
- Set a real line width. This was being left at 0, which is outside
device limits.
- Disable rasterization since we don't need it, to avoid needing to
populate even more structures.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 05:42:42 +0000 (17:42 +1200)]
tests: Fix VkLayerTest.BufferMemoryNotBound
Wrong usage flags.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 05:37:17 +0000 (17:37 +1200)]
tests: Use CmdFillBuffer to slim down InvalidCmdBufferBufferDestroyed
- Fix the usage flags (in neither version did we want
UNIFORM_BUFFER_BIT).
- Get rid of the image distraction, which had broken layout, /and/ would
count as another referenced object to be complained about.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 05:11:50 +0000 (17:11 +1200)]
layers: Get rid of old CBSTATUS-based dynamic viewport and scissor
These were superceded by the precise tracking of dynamic scissors and
viewports.
Fix up the tests to look for the precise errors, and get rid of the
broken clearing of static scissors & viewports, which triggered other
errors.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 04:45:16 +0000 (16:45 +1200)]
tests: Fix VKLayerTest.StencilLoadOp
Leaving this as 0 isn't valid.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 02:36:39 +0000 (14:36 +1200)]
tests: Capture unexpected messages before first desired message is set
Otherwise, tests that specify what they want late are vulnerable to
accidental errors in their setup code.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Fri, 16 Sep 2016 02:13:53 +0000 (14:13 +1200)]
tests: Fix VKTriangleTest to set the usage flags correctly on buffer.
By default, you get transfer_src|transfer_dst.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Mark Young [Fri, 16 Sep 2016 16:21:31 +0000 (10:21 -0600)]
build: FindImageMagick file needed updating.
Incorrectly updated the one in VulkanTools insteda of the one
in LoaderAndValidationLayers.
Thanks to Arda for the catch (even though I originally thought
he messed up).
Change-Id: Ic60d68d9c17437c4c711a8b359a8ad9246537e30
Karl Schultz [Tue, 13 Sep 2016 20:23:19 +0000 (14:23 -0600)]
layers: Cleanup unexpected push_constants msgs
Fix to core validation CreatePipelineLayout to change
a message from warning to error.
Fixes to PushConstants layer tests to fix unintended
validation errors and to avoid generating multiple
intentional errors in a single test.
Change-Id: I618d6e5aff55fa918da5bd81c47092174b266da1
Chris Forbes [Tue, 13 Sep 2016 06:19:21 +0000 (18:19 +1200)]
tests: Fix VkLayerTest.EnableWsiBeforeUse
This was producing a spurious error for completion of the ANI call not
being observable without either a fence or a semaphore.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 13 Sep 2016 06:13:58 +0000 (18:13 +1200)]
tests: Fix VkLayerTest.CreateBufferViewNoMemoryBoundToBuffer
This was using the wrong usage bit.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 13 Sep 2016 06:10:34 +0000 (18:10 +1200)]
tests: Fix VkLayerTest.FramebufferIncompatible
This was trying to execute a secondary command buffer within a subpass
whose contents were marked as inline.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 13 Sep 2016 05:39:18 +0000 (17:39 +1200)]
tests: Silence spurious error in MapMemWithoutHostVisibleBit
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Tony Barbour [Tue, 13 Sep 2016 17:45:37 +0000 (11:45 -0600)]
layers: Only save p-devices if call was successful
Change-Id: Ife211b1b2dc97c5bab0d6e7b177ead6b1a6105d3
Ian Elliott [Thu, 1 Sep 2016 17:09:21 +0000 (11:09 -0600)]
demos: Give Android versions permissions for internet & read/write files
This is in support of RenderDoc (and other tracing libraries). On Android,
these "native" applications must have permission to read/write files on the
Android filesystem so that a trace-capture library can write a trace file.
After building and installing the application, the following commands should be
run from a Linux shell, in order to truly grant those permissions (not normally
required for an Android Java-based app, with a GUI; but is required for these
shell-launched, native apps):
adb shell pm grant com.example.Cube android.permission.READ_EXTERNAL_STORAGE
adb shell pm grant com.example.Cube android.permission.WRITE_EXTERNAL_STORAGE
Tony Barbour [Mon, 12 Sep 2016 20:02:43 +0000 (14:02 -0600)]
demos: Fix issue of making Vulkan calls from callback
Change-Id: I6fcb41cbcd13e8ce5ace6f857d7dd54bad48aa55
Chris Forbes [Tue, 13 Sep 2016 04:49:57 +0000 (16:49 +1200)]
tests: Complain about unexpected errors
This won't give us test failures yet, but does make the spurious errors
visible.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 13 Sep 2016 04:32:57 +0000 (16:32 +1200)]
tests: Add test for depth/stencil state required in pipeline
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Tue, 13 Sep 2016 03:52:00 +0000 (15:52 +1200)]
tests: Hold PipelineObj::m_ds_state by ptr.
Makes it possible to pass nullptr for this structure. In fact, pretty
much all the time, we can pass nullptr here, as no depth buffer!
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 05:55:54 +0000 (17:55 +1200)]
layers: Require VkGraphicsPipelineCreateInfo::pDepthStencilState
If the subpass uses a depth/stencil attachment and rasterization is not
disabled, then this structure must be present.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Mon, 12 Sep 2016 23:22:22 +0000 (11:22 +1200)]
layers: tweak comments from review feedback
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 23:26:20 +0000 (11:26 +1200)]
layers: Don't do cross-stage validation if individual stages failed
There are only crashes and nonsense down this path.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 23:23:39 +0000 (11:23 +1200)]
layers: Don't bother continuing after missing entrypoint
There's very little we can do here.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 22:50:24 +0000 (10:50 +1200)]
tests: Reintroduce test for missing entrypoint
This used to exist, but got nuked at some point...
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Mark Young [Mon, 12 Sep 2016 17:55:10 +0000 (11:55 -0600)]
loader: Fix compiler warning
I had an unused variable left after some recent changes.
Change-Id: I934ea4a7410c6be9bbafe250bc09f38feb3d6f7c
Karl Schultz [Mon, 12 Sep 2016 20:09:03 +0000 (14:09 -0600)]
build: Fix CMake coding error
Need STREQUAL when compating strings.
Change-Id: Id65055d18af2c0bbd3e4959f21587e2ba9afa607
Karl Schultz [Mon, 12 Sep 2016 19:28:09 +0000 (13:28 -0600)]
demos: fix vulkaninfo warnings in display-only build
Change-Id: Iabc8edd7cd90f848d4957d005eb14c1094e94790
Norbert Nopper [Thu, 8 Sep 2016 07:31:36 +0000 (09:31 +0200)]
Fix, that display only build under Linux works
Derrick Owens [Fri, 9 Sep 2016 19:49:07 +0000 (15:49 -0400)]
loader: Fix another possible memory leak
Added another cleanup case, just in case...
Change-Id: Ia6e09ce12c29314a1cce6b1cdcd72bdd7749f655
Derrick Owens [Fri, 9 Sep 2016 19:45:13 +0000 (15:45 -0400)]
loader: Fixed conformance memory leak.
This fixes a memory leak found whilst running Khronos conformance with
more than 1 ICD installed.
Change-Id: I34ed978b2929c33d9e2c49cf633f09ada2788a92
Lenny Komow [Fri, 9 Sep 2016 15:43:17 +0000 (09:43 -0600)]
winsdk: Fix ConfigureRT.exe not being deleted
Mark Young [Fri, 9 Sep 2016 13:26:48 +0000 (07:26 -0600)]
loader: Add back func calls
Previously removed function calls in extension_create_instance
because I believed they were no longer needed. I was incorrect.
Change-Id: Ibc63009f0038a413b7e1bc03225818b046a62352
Chris Forbes [Fri, 9 Sep 2016 02:36:33 +0000 (14:36 +1200)]
layers: make CmdClearAttachments messages errors again for now.
There's some spec contradictions, captured in !488. For now, let's say
that these are errors. This commit can be reverted if !488's resolution
retains the noop behavior.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 03:35:36 +0000 (15:35 +1200)]
layers: Make d/s CmdClearAttachments without d/s attachment warning
This is pointless, but well-defined. Rework warning message to not
reference the VkClearAttachment::colorAttachment, as it's only
meaningful for color clears.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 03:30:59 +0000 (15:30 +1200)]
layers: Add VK_ATTACHMENT_UNUSED warning case for CmdClearAttachments
This is a pointless clear, but well-defined.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Chris Forbes [Thu, 8 Sep 2016 03:25:24 +0000 (15:25 +1200)]
layers: CmdClearAttachments colorAttachment is index into subpass
We were previously treating this as being an index into the renderpass
attachment descriptions, but that's nonsense. It's also well-defined
what happens if an out-of-range index is provided, so downgrade to a
warning.
Signed-off-by: Chris Forbes <chrisforbes@google.com>
Mark Young [Fri, 9 Sep 2016 00:36:32 +0000 (18:36 -0600)]
loader: Remove extension enable checks
Apparently, the intent of Vulkan is to have the loader do no
run-time checking in each of the functions as to whether or not
the extension is enabled. This should only be done in the
validation layers.
This resolves the crash Jeff was seeing in the Nvidia driver when
using the new extension.
Change-Id: Ic90db0bde062d6b2511954abb8677f9f20bb4285
Mark Young [Thu, 8 Sep 2016 18:28:38 +0000 (12:28 -0600)]
loader: Add checks for usage of wsi extensions
The loader really should validate that the WSI extensions are
enabled before being called. Additionally, I needed to add
more checks for the KHR_display_swapchain extension in the
parameter_validation and object_tracker layers.
Change-Id: I3d07d46baf551be6f5f07e5374d6c683e3f52e7e
Mark Mueller [Thu, 8 Sep 2016 17:11:25 +0000 (11:11 -0600)]
tests: GH428 removal of offending line
The ill constructed draw is not needed and causes issues on some platforms.
Change-Id: I695a386e3a70a2d004b047044ee534e204a6499a
Ian Elliott [Thu, 8 Sep 2016 17:34:52 +0000 (11:34 -0600)]
layers: Remove old/unnecessary FIXME from swapchain layer
This FIXME comment does not apply. It may have applied, if the line being
referred to once copied a pointer to user data, but at this point, the actual
data is being copied into a struct owned by the swapchain layer, which is what
the comment was about.