Panagiotis Apostolou [Mon, 26 Mar 2018 11:21:26 +0000 (13:21 +0200)]
Missing pipeline barriers from fragment variants of dEQP-VK.subgroups.*
This commit adds a missing image pipeline barrier in the fragment variant
of the subgroups test group. After the end of the renderpass and before the
vkCmdCopyImageToBuffer, a pipeline barrier is required to delay the copy
command so that all the writes to the color attachment are complete.
Tests affected:
dEQP-VK.subgroups.*
Components: Vulkan
VK-GL-CTS issue: 1135
Change-Id: I82a5179f97cf4435f96dd378f123f37bc26bd082
Chris Forbes [Thu, 12 Apr 2018 15:54:00 +0000 (08:54 -0700)]
Request GPDP2 extension if required in shared presentable image tests
Change-Id: Ifa4b19c6046d4f0b95a5fc65443bcd1df051b6a5
Components: Vulkan
Affects: dEQP-VK.wsi.*.shared_presentable_image.*
VK-GL-CTS: 1078
Peter Quayle [Wed, 28 Mar 2018 17:18:28 +0000 (18:18 +0100)]
Fix improper use of vkCmdWaitEvents
dEQP-VK.synchronization.smoke.events was calling vkCmdWaitEvents inside
a render-pass, which is not allowed if the event is later triggerd using
vkSetEvent. Additionally, the test was only waiting 1ns to determine if
a queue had made forward progress, which is too short to give the
implementation a chance to do any work.
Affects:
dEQP-VK.synchronization.smoke.events
Components: Vulkan
VK-GL-CTS issue: 1089
Change-Id: I6d9c9a635d82840f49fd3d9ed3736f91c4b6f88d
Dae Kim [Tue, 27 Mar 2018 09:54:23 +0000 (10:54 +0100)]
Fixed rounding issue in spirv_assembly tests
The test assumes that the host system and the device use the same
rounding mode. When this is not the case, the derived rounding errors
make the test fail unexpectedly. By rounding the input values down we
avoid differences in the test output caused by the rounding modes in
use.
Affects:
dEQP-VK.spirv_assembly.instruction.compute.loop_control.dependency_infinite
Components: Vulkan
VK-GL-CTS issue: 1097
Change-Id: I3678e38cbe5c6d57cecef91e7398df0d04763851
Caio Marcelo de Oliveira Filho [Mon, 12 Mar 2018 17:25:58 +0000 (10:25 -0700)]
Fetch new tags when fetching new Git sources
Change-Id: I1a906ebf2698eac33c3acbf1da74df9642b775a2
Component: Framework
(cherry picked from commit
204d597504c94fa52ed0f5a2c21895482236d5cc)
Alexander Galazin [Wed, 7 Mar 2018 15:45:21 +0000 (16:45 +0100)]
Update glslang, SPIR-V commits
Components: Framework
Affects: dEQP-VK.*
Change-Id: If5830108278597a28f8043dd947bed3c20c9688d
Alexander Galazin [Fri, 2 Mar 2018 11:17:24 +0000 (12:17 +0100)]
Add a note to the README
Explain that the verification scrpt will report a warning
for a correctly generated git-log
Components: Vulkan
Change-Id: Ie8e0ab090f50c776150f9a7f4ce7d69583186953
Marcin Rogucki [Thu, 18 Jan 2018 09:55:07 +0000 (10:55 +0100)]
Validation errors fix in copying 3d images
A minor chages to fix validation errors.
Two tests are not passing currently!
Affected tests:
dEQP-VK.api.copy_and_blit.dedicated_allocation.image_to_image.3d_images.*
Components: Vulkan
VK-GL-CTS issue: 208
Change-Id: Id3c5ec869d52db6b1229f45bbeacf24e35e75e48
(cherry picked from commit
c23ba94d7978ba84f9ffd2e75b6afbf8e3395633)
Arkadiusz Sarwa [Tue, 27 Feb 2018 11:38:31 +0000 (12:38 +0100)]
Fix: Memory alignment
Affects:
dEQP-VK.memory.mapping*
Components: Vulkan
VK-GL-CTS issue: 1039
Change-Id: Ia5093148fa145c9eecc03fe3b76102cdf77ad472
Panagiotis Apostolou [Thu, 22 Feb 2018 15:28:54 +0000 (16:28 +0100)]
Spec relaxation for protected YCbCr Implicit chroma reconstruction
Incorporates spec relaxation in the protected memory YCbCr tests which
describes that when doing implicit reconstruction with nearest chroma
filtering, an implementation may behave as if both xChromaOffset and
yChromaOffset were VK_CHROMA_LOCATION_MIDPOINT.
Tests affected:
dEQP-VK.protected_memory.interaction.ycbcr.*
Components: Vulkan
VK-GL-CTS issue: 1032
Change-Id: I25ef589ed05b1b167475eb5b6d782b261b1b3532
Slawomir Cygan [Wed, 28 Feb 2018 18:11:21 +0000 (19:11 +0100)]
Set usable instance version in external memory tests
These tests were using vkEnumerateInstanceVersion() to get
instance version. This failed, when Vulkan 1.1 Loader is used with
Vulkan 1.0 ICD: the function was returning version 1.1,
so test was not requesting any needed instance extensions, as these
were assumed to be already in core.
This is out of sync with dispatch code dispatching to
1.0 + KHR extension functions, when calling entrypoints like
vkGetPhysicalDeviceExternalBufferProperties(). As no KHR extensions
were requested during instance creation, the behavior is undefined.
The dispatch code is using context.getUsedApiVersion() utility
to guess Vulkan version, so same utility should be used for instance
creation, avoiding raw calls to vkEnumerateInstanceVersion().
Components: Vulkan
VK-GL-CTS Issue: 1047, 854
Affects:
dEQP-VK.api.external.memory.*
dEQP-VK.api.external.fence.*
dEQP-VK.api.external.semaphore.*
Change-Id: Iecf9bb086e04c6d88118fee9c4867ac5647d1ab7
Panagiotis Apostolou [Thu, 22 Feb 2018 15:04:24 +0000 (16:04 +0100)]
Remove vertex stage variants from dEQP-VK.protected_memory.interaction.ycbcr.*
"Except for framebuffer-space pipeline stages, compute shader stage, and
transfer stage, any read from or write to protected memory during protected
queue operations results in undefined behavior"
The existing dEQP-VK.protected_memory.interaction.ycbcr.* group include
tests that sample the protected texture from inside the vertex shader. This
commit removes these tests.
Tests affected:
dEQP-VK.protected_memory.interaction.ycbcr.*
Components: Vulkan
VK-GL-CTS issue: 1031
Change-Id: I459da0f3028e418c90285e55fb17ac1fda8d8467
Alexander Galazin [Thu, 1 Mar 2018 13:24:31 +0000 (14:24 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I8c2317b97955bbee922af624b91287071fbc7661
Dae Kim [Fri, 23 Feb 2018 08:39:12 +0000 (08:39 +0000)]
Added missing gl_PointSizes when drawing points
Some tests using point list primitive topology were not specifying point
sizes with the gl_PointSize built-in.
Affects:
dEQP-VK.subgroups.*
dEQP-VK.synchronization.*
Components: Vulkan
VK-GL-CTS issue: 1035
Change-Id: I8588e26ee47c58f6ec4e9ee5fdce70546ec28903
Slawomir Cygan [Thu, 22 Feb 2018 17:55:01 +0000 (18:55 +0100)]
Do not access temporary C++ object members by reference
Fix undefined C++ behavior (resulting in errors when doing runtime memory checks),
where a temporary object returned by getInputAspects() method is used by a reference
to a member after the temporary is destructed.
It seems getInputAspects should just return reference.
Affects: dEQP-VK.renderpass.*.input*
Component: Vulkan
VK-GL-CTS Issue: 1033
Change-Id: I71edabed71ea63f46cdbc45c966d57cd020565ab
Peter Quayle [Wed, 21 Feb 2018 15:34:55 +0000 (15:34 +0000)]
Ensure sync_fd tests use valid file descriptors
The external.*.sync_fd.* tests fail on some implementations
because vkGet*FdKHR may return -1 (meaning pre-signalled) instead
of a valid file descriptor.
Affects:
dEQP-VK.api.external.semaphore.sync_fd*
dEQP-VK.api.external.fence.sync_fd*
Components: Vulkan
VK-GL-CTS issue: 957
Change-Id: I91087db1595a19605956dad460caa344e850491e
Dae Kim [Tue, 20 Feb 2018 13:37:39 +0000 (13:37 +0000)]
Fix verification in robust buffer access tests
Some buffer accesses to addresses less than 16 bytes away from the out
of bounds address where not considered potentially out of bounds, even
when there were out of bounds accesses in the same SPIR-V block.
Affects:
dEQP-VK.robustness.buffer_access.*:
Components: Vulkan
VK-GL-CTS issue: 1025
Change-Id: Ibcb61fb84cecc360d753ba5d3acc1f9d48169651
Alexander Galazin [Thu, 22 Feb 2018 11:45:50 +0000 (12:45 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I115b3e16d7282ef5f9abeb5bd382f744d891b5e3
Tom Cooper [Mon, 12 Feb 2018 16:20:58 +0000 (16:20 +0000)]
Fix low memory in dedicated_alloc buffer tests
Observe platform memory limits to avoid low memory conditions.
Components: Vulkan
VK-GL-CTS issue: 1008
Affects: dEQP-VK.api.buffer.dedicated_alloc.create_buffer_*
Change-Id: Ib184b648126e8cb6e1acff533447a86be9a1a612
Alexander Galazin [Wed, 14 Feb 2018 09:17:22 +0000 (10:17 +0100)]
Update glslang/spir-v headers tags
Components: Framework
Affects: dEQP-VK.*
Change-Id: I1282fe5924abfb86fba7d32d3c67b94993a252fe
Juan A. Suarez Romero [Fri, 10 Nov 2017 11:20:36 +0000 (11:20 +0000)]
Do not use depthClampEnable unconditionally.
Depth clamping is not needed by these tests, so
do not rely on the optional feature when creating
the pipeline object.
Affects:
dEQP-VK.spirv_assembly.*
Components: Vulkan
VK-GL-CTS issue: 846
Change-Id: I248759e6ac209ccbd7661068c3ef5f1dfb8ca501
(cherry picked from commit
d33f62c935d51b4ecba0b6b97ea1baafe67ea3f4)
Lionel Landwerlin [Fri, 9 Feb 2018 15:20:25 +0000 (15:20 +0000)]
YCbCr: go through the correct location ranges when sampling
Affects: dEQP-VK.ycbcr.conversion.*
Components: Vulkan
VK-GL-CTS issue: 1005
Change-Id: Ifac1026e82e1bb48195fc7ab5429951a15324042
Alexander Galazin [Mon, 5 Feb 2018 12:13:44 +0000 (13:13 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I36a59b8e50449e1788779380d6dc1dc4668b78fc
Vamsidhar Gaddam [Wed, 31 Jan 2018 10:23:48 +0000 (11:23 +0100)]
Pass in proper pWaitDstStageMask in vktWsiSharedPresentableImageTests
One of the submitInfos for a queue submit had
a semaphore passed in without passing in any stage mask. This is a
spec violation.
Affects:
dEQP-VK.wsi.android.shared_presentable_image.*
Components: Vulkan
VK-GL-CTS issue: 974
Change-Id: Iee2ea838e971a3610f44b8c1f765869a8a21fedb
Panagiotis Apostolou [Wed, 17 Jan 2018 09:01:58 +0000 (10:01 +0100)]
Spec relaxation for YCbCr Implicit chroma reconstruction
Incorporates spec relaxation in the YCbCr tests which describes that when
doing implicit reconstruction with nearest chroma filtering, an
implementation may behave as if both xChromaOffset and yChromaOffset were
VK_CHROMA_LOCATION_MIDPOINT.
Tests affected:
dEQP-VK.ycbcr.conversion.*
Components: Vulkan
VK-GL-CTS issue: 979
Change-Id: I10e83d2f5845c3e74673aa9e42fc4f0950f9e6cd
Noah Fredriks [Mon, 15 Jan 2018 21:09:55 +0000 (16:09 -0500)]
Fix illegal memory access for robustness test
Don't subtract 3 from buffer pointer for VK_FORMAT_A2B10G10R10_UNORM_PACK32
alpha channel because this format is a deUint32 instead of Vec4.
Affects: dEQP-VK.robustness.vertex_access.a2b10g10r10_unorm_pack32.*
Components: Vulkan
VK-GL-CTS issue: 946
Change-Id: I2a5241191e01252ce071658b8dbd79d3e424aa41
Alexander Galazin [Fri, 19 Jan 2018 09:08:43 +0000 (10:08 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I8db9696a5f7b9bafdf20e76afc8d1d42b96cfe1c
Yanjun Zhang [Wed, 10 Jan 2018 00:12:47 +0000 (16:12 -0800)]
Check GS support before testing adjacent primitives
Affects: dEQP-VK.draw.basic_draw.draw.*
Components: Vulkan
VK-GL-CTS issue: 926
Change-Id: I02569b1944eed52f641966f9206d44983d1ff165
Chris Forbes [Wed, 10 Jan 2018 02:13:15 +0000 (18:13 -0800)]
Fix buffer usage flags for index and indirect buffers
Previously, createAndUploadBuffer would provide
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT for all buffers. This is no good for
index buffers and indirect buffers.
VK-GL-CTS: 930
Change-Id: I774beef7c7226f4fbd5670591eb66475300ee208
Components: Vulkan
Affects: dEQP-VK.draw.instanced.*
Alexander Galazin [Thu, 18 Jan 2018 09:51:31 +0000 (10:51 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I79e2e20649232a7ebd78088981f06ade6c5e6323
Piers Daniell [Fri, 5 Jan 2018 21:04:29 +0000 (14:04 -0700)]
Add VK_ACCESS_SHADER_READ_BIT barrier access bit
The dEQP-VK.query_pool.statistics_query.compute_shader_invocations.secondary*
tests were missing a VK_ACCESS_SHADER_READ_BIT bit in the
pipeline barrier between compute dispatch calls. Each dispatch
was doing a read-modify-write of the same SSBO locations, but
the test only used the VK_ACCESS_SHADER_WRITE_BIT access bit.
This may cause the compute shader to read stale cache data
since the dependency between each dispatch was not expressed
correctly.
This fix simple adds the missing VK_ACCESS_SHADER_READ_BIT
access bit.
Affects:
dEQP-VK.query_pool.statistics_query.compute_shader_invocations.secondary*
Components: Vulkan
VK-GL-CTS issue: 927
Change-Id: I946d6113c897a03116db7a93a182e27497adfe6f
Yanjun Zhang [Thu, 11 Jan 2018 19:44:54 +0000 (11:44 -0800)]
Set maxDescriptorSet* limit table based on shader stages
Re-submit the patch based on WG review and feedback.
Affects: dEQP-VK.api.info.device.properties
Components: Vulkan
VK-GL-CTS issue: 923
Change-Id: Ie627b52cfae099d87f5b6183118d42bcccd313ca
Alexander Galazin [Fri, 12 Jan 2018 14:03:32 +0000 (15:03 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I02ba5bcb0273934d93d91e2375f9d272c6c69904
Yanjun Zhang [Wed, 10 Jan 2018 00:26:50 +0000 (16:26 -0800)]
Check sampleRateShading for sample_qualifier_distinct_values.* tests
Affects: dEQP-VK.pipeline.*.sample_qualifier_distinct_values.*
Components: Vulkan
VK-GL-CTS issue: 925
Change-Id: Ife1166e8ed88699c67ee6acde79857f34c013398
Chris Forbes [Tue, 9 Jan 2018 23:26:34 +0000 (15:26 -0800)]
Use unique indices for each binding in out_of_pool_memory test
Previously, all the layout bindings would use binding=0, which is
invalid. Give them their own binding indices.
VK-GL-CTS: 929
Change-Id: I28e62db6c257ed0843dd9b77b2d8cb01d02e3eba
Components: Vulkan
Affects: dEQP-VK.api.descriptor_pool.out_of_pool_memory
Alexander Galazin [Thu, 11 Jan 2018 12:40:39 +0000 (13:40 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I918575ed59c764d988fb98b718745605c5285b06
Chris Forbes [Wed, 10 Jan 2018 19:42:41 +0000 (11:42 -0800)]
Build prebuilts for HLSL shaders
Previously these prebuilts would be missing, causing all the HLSL tests
to fail when run via the Android CTS integration which relies on them.
VK-GL-CTS Issue: 939
Change-Id: Ib69f884818ea24692222cc74d8b69b0e0bcb465d
Components: Vulkan
Affects: dEQP-VK.*hlsl*
Piers Daniell [Thu, 30 Nov 2017 21:40:00 +0000 (14:40 -0700)]
Skip allocation subtests that allocate too much VA
For 32-bit builds of CTS we can't allocate more host visible
memory than there is virtual address space for the process
since the allocation will fail and cause a false negative
result.
This CL skips over host visible allocations that would add
up to more than 1.5GB of virtual address space. This can
happen on platforms that expose very large heaps that
support memory types that include host visible memory.
The implementation can't cap the available memory size
since the heap can support allocations of memory other
than host visible and it doesn't want to restrict those
that don't consume CPU virtual address space. Its up
to CTS to limit how much virtual address space it
chooses to allocate.
Affects:
dEQP-VK.memory.allocation.basic.percent_1.*.count_12
Components: Vulkan
VK-GL-CTS issue: 859
Change-Id: I321f5cad567245571722f53397df26e23a4cee97
Marcin Rogucki [Wed, 13 Dec 2017 14:07:32 +0000 (15:07 +0100)]
Fix for #895 - physical_device_groups testing
For 1.1 version instances we do not require physical devices groups extension to be supported. Instead on 1.1 instances we just assume it's now a part of core implementation.
Components: Vulkan
VK-GL-CTS issue: 895
Affects: All tests using createInstanceWithExtension function
Change-Id: Ib11999498c28156e1e5d0c5e9025cb6ca4ba75a0
Noah Fredriks [Fri, 3 Nov 2017 18:20:06 +0000 (14:20 -0400)]
Round up vertex buffer size to stride multiple to fix OOB attribute reads
Partial stride multiple vertex buffer elements are now considered OOB
to accomodate implementations that fetch stride-sized elements
Affects: dEQP-VK.texture.*
Components: Vulkan
VK-GL-CTS issue: 812
Change-Id: I72a2143146b4ea0d893d70d41b7ca2c84c34542a
(cherry picked from commit
50c0544ec8c2e99158bf8a375eacfc9baf364a91)
Alexander Galazin [Thu, 7 Dec 2017 11:27:21 +0000 (12:27 +0100)]
Vulkan test list without waivers
Adds new test list - vk-default-no-waivers.txt which is the same as
the regular mustpass (vk-default.txt) but without exclusions for
waivers.
Component: Vulkan
Change-Id: I7127255bf975ee02ede284a6217b82baa3af2414
Oivind Boge [Tue, 19 Dec 2017 10:06:16 +0000 (11:06 +0100)]
Fixed tests issue where vertex shader was misssing gl_PointSize
Affects:
dEQP-VK.protected_memory.ssbo.*
Components: Vulkan
VK-GL-CTS issue:: 915
Change-Id: Ia7cc2b9a6b1593ca15f585a9dc052cb2528f9ee3
Graeme Leese [Thu, 14 Dec 2017 13:53:45 +0000 (13:53 +0000)]
Also set gl_PointSize in subgroup quad tests
Affects: dEQP-VK.subgroups.quad.*vertex_framebuffer
Components: Vulkan
VK-GL-CTS issue: 872
Change-Id: Ie73fb744737e079d42792f6c974534e57ae006ef
Alex Walters [Wed, 13 Dec 2017 15:47:23 +0000 (15:47 +0000)]
Invalid handle in create_display_plane_surface
This test attempts to create a display plane surface but incorrectly
fills in the VkDisplaySurfaceCreateInfo with an index into dEQP's array
of display modes instead of the VkDisplayModeKHR handle itself.
The app is required to specify a valid handle.
An invalid handle results in undefined behaviour.
Affects:
dEQP-VK.wsi.display.create_display_plane_surface
Components: Vulkan
VK-GL-CTS issue: 906
Change-Id: I9fbbd40926e26bf82ed11dc1b5e807922cbc4e73
Noah Fredriks [Thu, 14 Dec 2017 19:48:48 +0000 (14:48 -0500)]
Fix Invalid VkImageSubresourceRange
Pipeline image barrier was using the arraySize of the wrong image.
Affects: dEQP-VK.api.copy_and_blit.core.resolve_image.*
Components: Vulkan
VK-GL-CTS Issue: 910
Change-Id: I021bdd0ba5eff9e894f21186b3d85bb2f2c99f16
Alexander Galazin [Thu, 14 Dec 2017 10:52:40 +0000 (11:52 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I3b358114c4158d2130e68632d4e02d7246c6f9af
Marcin Rogucki [Mon, 20 Nov 2017 15:44:51 +0000 (16:44 +0100)]
Fix for getPhysicalDeviceFeatures2 loader issue
For all instance methods taking VkPhysicalDevice
as a first parameter that were moved to core in
1.1 there is now a dynamic dispatch based on
physical device api version. If version is 1.1 or
above a core method is used. If version is lower
than that a counterpart extension method is used.
Additionally a pre-instance is created to enumerate
all physical devices and select lowest version to
be able to enable proper extensions set for final
instance.
Components: Vulkan
VK-GL-CTS issue: 854
Affects: All tests using getPhysicalDeviceFeatures2
Change-Id: Ia68a5cac23ae35566fcf99bc56c93c9cb77f33e0
Jeff Leger [Sat, 2 Dec 2017 02:16:28 +0000 (21:16 -0500)]
Modify spirv_assembly tests adding dependency on 16bit_storage extension
Several tests were using 16bit_storage, but were not checking the
extension was supported nor declaring the required opCapability.
Affects:
dEQP-VK.spirv_assembly.instruction.compute.sconvert.*
dEQP-VK.spirv_assembly.instruction.compute.uconvert.*
Components: Vulkan
VK-GL-CTS issue: 875
Change-Id: I07fda13c69cdef09967d4aab5067edff07c47fa1
Noah Fredriks [Wed, 29 Nov 2017 23:27:52 +0000 (18:27 -0500)]
Avoid creating vkInstances for not supported cases
Defer vkInstance creation for cross instance tests until not supported
checks complete to shorten the runtime of these 10k total test cases.
Affects: dEQP-VK.synchronization.cross_instance.*
Components: Vulkan
VK-GL-CTS Issue: 885
Change-Id: I481de90d8c0363c23a287aaf457ac101c327c558
Igor Ostrowski [Wed, 15 Nov 2017 17:33:51 +0000 (18:33 +0100)]
Fix mediump zero division in matrix inverse tests
To avoid zero division all input matrix determinants should be same signed.
Affects:
dEQP-VK.glsl.matrix.inverse.*
Components: Vulkan
VK-GL-CTS issue: 881
Change-Id: Ib702f81c4eb7f25ba9081db6ead67da02f484ca9
Maciej Jesionowski [Wed, 29 Nov 2017 13:07:42 +0000 (14:07 +0100)]
Remove max expected query result in pipeline statistics tests
Query result is implementation-defined and some internal rendering
operations may count towards the results of queries.
Affects: dEQP-VK.query_pool.statistics_query.*
Components: Vulkan
VK-GL-CTS issue: 876
Change-Id: I0c86e1101c1ae00049623bd49dfe0c4de5798832
Stu Smith [Thu, 30 Nov 2017 13:30:20 +0000 (13:30 +0000)]
Add missing gl_PointSize in subgroup test shaders when drawing points
Affects: dEQP-VK.subgroups.*vertex_framebuffer
Components: Vulkan
VK-GL-CTS issue: 872
Change-Id: Iddf8362024d3dd750e6f0b9d4c71e1adcdfa63ca
Jeff Leger [Wed, 22 Nov 2017 17:12:58 +0000 (12:12 -0500)]
Reduce GPU workload of pipeline_barrier tests
Tests with large buffer sizes will now clamp the
GPU workload to avoid triggering execution timeout
on implementations.
Affects: dEQP-VK-memory.pipeline_barrier.*
Components: Vulkan
VK_GL_CTS issue: 813
Change-Id: I479f6ad04ef2ec6f9d87b65a39a2bc13bcbcc6a2
Alexander Galazin [Wed, 22 Nov 2017 11:36:02 +0000 (12:36 +0100)]
Update glslang tag
Update glslang tag to vulkan-1.1-rc6
Affects: dEQP-VK.*
Change-Id: I5bfbf717e1f0be51802b33f2a7b658d395c72019
Slawomir Cygan [Thu, 23 Nov 2017 10:51:23 +0000 (11:51 +0100)]
Fix regression from "Add contextSupports functions"
Fixes contextSupports condition for missing VK1.1.
This regression was introduced in
39a3dba4.
Affects:
dEQP-VK.spirv_assembly.instruction.compute.loop_control.dependency_infinite
Components: Vulkan
VK-GL-CTS issue: 860
Change-Id: I866c815b918af2860d888775179635011f757535
(cherry picked from commit
4f5fa8a6a62071624d9a2a703b9219abc301c72a)
Alexander Galazin [Thu, 30 Nov 2017 13:39:20 +0000 (14:39 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I9fbb4d01a6050ae79808e61649783ef698a1481d
Piers Daniell [Wed, 22 Nov 2017 20:23:12 +0000 (13:23 -0700)]
Fix some CTS bugs that cause validation failures
Most of these are misuse of the dstStageMask parameter to
vkCmdPipelineBarrier. The dstStageMask has to be compatible
with the destination access type of the buffer or image.
The other large source of validation errors is a bug
in the CTS framework when calling vkCreateDevice. The
wrong pNext member was used for setting the enabled
features. The effect is that all features were disabled.
Affects:
dEQP-VK.pipeline.push_constant.graphics_pipeline.*
dEQP-VK.pipeline.multisample_shader_builtin.sample_mask.pattern.*
dEQP-VK.*
Components: Vulkan
VK-GL-CTS issue: 862
Change-Id: I58ca4e033d7d66b52c6726a635808b9b747546d8
Noah Fredriks [Thu, 9 Nov 2017 21:00:29 +0000 (16:00 -0500)]
Enforce OOB requirements for binding with multiple attributes
The spec states that, if any vertex input attribute using a specific
vertex input binding is out of bounds, then all vertex input attributes
using that vertex input binding for that vertex shader invocation are
considered out of bounds.
Affects: dEQP-VK.robustness.vertex_access.*.vertex_incomplete
Components: Vulkan
VK-GL-CTS issue: 848
Change-Id: I6a9222a0b2a53368af361419105793a5301a6bf6
Tom Cooper [Wed, 22 Nov 2017 15:57:37 +0000 (15:57 +0000)]
Pipeline must be bound before calling vkCmdSetViewport/Scissor etc
Affects:
dEQP-VK.dynamic_state.*
Components: Vulkan
VK-GL-CTS issue: 861
Change-Id: I0f5312c83096cebebfa4ab01987c2ab9ede7306b
Jari Komppa [Tue, 14 Nov 2017 12:21:06 +0000 (14:21 +0200)]
Added constraints for random access flags
The pipeline barrier tests randomize access flags and stages, sometimes
resulting in access flags being enabled for stages that do not support
said accesses. This fix removes the illegal access flags while leaving
the random flag generation intact.
Affects:
dEQP-VK.memory.pipeline_barrier.*
Components: Vulkan
VK-GL-CTS issue: 814
Change-Id: I135ea8a9e41b5d8428ef77fe6ecbdf2380d80c7d
Slawomir Cygan [Fri, 17 Nov 2017 12:59:14 +0000 (13:59 +0100)]
Fix regression from "Add contextSupports functions"
Fixes contextSupports condition for missing VK1.1.
This regression was introduced in
39a3dba450.
Affects:
dEQP-VK.spirv_assembly.instruction.compute.loop_control.dependency_infinite
dEQP-VK.spirv_assembly.instruction.compute.loop_control.dependency_length
Components: Vulkan
VK-GL-CTS issue: 727
Change-Id: I64c0b5fbd40f67ae6cfe314ac66ddc9d7f201dbb
Arkadiusz Sarwa [Thu, 19 Oct 2017 11:20:24 +0000 (13:20 +0200)]
get_physical_device_properties2: add information print
Affects:
dEQP-VK.api.info.get_physical_device_properties2*
Components: Vulkan
VK-GL-CTS issue:: 777
Change-Id: Ib7462d63de521574f7fab018d7aed2d449a58dd3
(cherry picked from commit
45b164c54682c8e1d3de14d9ef425860adae4ec1)
Noah Fredriks [Fri, 3 Nov 2017 18:20:06 +0000 (14:20 -0400)]
Round up vertex buffer size to stride multiple to fix OOB attribute reads
Partial stride multiple vertex buffer elements are now considered OOB
to accomodate implementations that fetch stride-sized elements
Affects: dEQP-VK.texture.*
Components: Vulkan
VK-GL-CTS issue: 812
Change-Id: I72a2143146b4ea0d893d70d41b7ca2c84c34542a
Arkadiusz Sarwa [Mon, 9 Oct 2017 15:13:48 +0000 (17:13 +0200)]
Add contextSupports functions
Components: Vulkan
VK-GL-CTS issue: 727
Change-Id: Ic2912106268402eb61c2a7aa4d487a2c3c7755e3
Alexander Galazin [Fri, 10 Nov 2017 10:24:18 +0000 (11:24 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I38a420841bb08c1d2d993eee39c16046c7680c3f
Jorg Wagner [Mon, 6 Nov 2017 13:56:49 +0000 (14:56 +0100)]
Do not use depthClampEnable unconditionally.
Depth clamping is not needed by this test, so
do not rely on the optional feature when creating
the pipeline object.
Affects:
dEQP-VK.api.smoke.*
Components: Vulkan
VK-GL-CTS issue: 817
Change-Id: I6794f7f5d17d7610ff43dcb4080a6a30e12ed95b
Ari Suonpaa [Fri, 13 Oct 2017 11:59:43 +0000 (14:59 +0300)]
Fixed double type declarations and invalid OpUConvert usage.
Fixed access chain indexing test by removing double declarations
of the same integer type. Also replaced OpUConvert instruction
with OpCopyObject when the conversion would result into the
same bit width as the original.
Affects: dEQP-VK.spirv_assembly.instruction.*.indexing.*
VK-GL-CTS issue: 702
Change-Id: I91a9de22976b109ae988391d5cb7e4388bda4adf
Vikram Kushwaha [Wed, 1 Nov 2017 23:02:57 +0000 (16:02 -0700)]
Fix image layout transitions for peer Image
When SFR/AFR copies are performed by peer device into
the primary device's render target, the image layout transitions are
also being performed by peer device.
This is only correct if GENERIC_SRC\GENERIC_DST feature flags
are supported.
VK-GL-CTS issue: 807
Change-Id: Ie3ffe6a68976ad644de12bd2c923cd0a34defe5f
Affects: dEQP-VK.device_group.*
Components: Vulkan
Graeme Leese [Thu, 2 Nov 2017 13:11:03 +0000 (13:11 +0000)]
Fix subgroup tests error message
Basic is a feature bit, not a shader stage, so use the correct string
getter.
Components: Vulkan
Affects: dEQP-VK.subgroups.basic.*
Change-Id: Ibb9b88cf1efcf4f6459b277cb4e4beaadeda01c9
Alexander Galazin [Tue, 24 Oct 2017 08:06:06 +0000 (10:06 +0200)]
Update glslang tag
Update glslang tag to vulkan-1.1-rc4
Affects: dEQP-VK.*
Change-Id: Idd20ba8a7c3d58124062da119af5186acc08a4d2
Chris Forbes [Mon, 23 Oct 2017 22:13:38 +0000 (15:13 -0700)]
KHR_swapchain: Trigger the fence off vkQueueSubmit
Another instance of
438ea0888e8ccae.
VK-GL-CTS issue: 756
Component: Vulkan
Affects:
dEQP-VK.wsi.*.swapchain.*
Change-Id: I87da6800e0c8402c2c50ca88eb3360b31c3a515d
Chris Forbes [Mon, 23 Oct 2017 22:09:56 +0000 (15:09 -0700)]
GOOGLE_display_timing: Trigger the fence off vkQueueSubmit
Another instance of
438ea0888e8ccae.
VK-GL-CTS issue: 756
Component: Vulkan
Affects:
dEQP-VK.wsi.*.display_timing.*
Change-Id: If94644402579ba1f46347d155313f6ef52d125a0
Alexander Galazin [Tue, 31 Oct 2017 11:42:28 +0000 (12:42 +0100)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: If71ec6d93e19e3dd2744f9c1375ec08553dec392
Maciej Jesionowski [Mon, 23 Oct 2017 10:14:36 +0000 (12:14 +0200)]
Add missing barrier to texel_view_compatible tests
There was no GPU to Host barrier. GPU writes are not automatically
visible to the Host.
Affected tests: dEQP-VK.image.texel_view_compatible.compute.*
VK-GL-CTS issue: 786
Components: Vulkan
Change-Id: I47797add1c5e48dfa1a7c7be8897e822091c2044
Jason Ekstrand [Thu, 19 Oct 2017 16:07:52 +0000 (09:07 -0700)]
VK_KHR_variable_pointers: Don't use OpPtrAccessChain on a bare SSBO
You can't use OpPtrAccessChain on the direct result of an OpVariable
because it's not the element of any array. Instead, we need to do our
OpPtrAccessChain on a pointer to an actual array element and that
pointer needs to be properly decorated with a stride.
Component: Vulkan
VK-GL-CTS issue: https://gitlab.khronos.org/Tracker/vk-gl-cts/issues/686
Affects:
dEQP-VK.spirv_assembly.instruction.graphics.variable_pointers.multi_buffer_*
Change-Id: I9c45fb1320099e404672ee169a6a8e5c41eb13f7
Jason Ekstrand [Thu, 19 Oct 2017 15:43:36 +0000 (08:43 -0700)]
VK_KHR_variable_pointers: Delete an unused and invalid variable
The %float_var variable was not being used in any way by the test. It
was also invalid because it had the StorageBuffer storage class (so it's
in external memory) but also had a constant initializer. The real
variable being tested in these tests is %f32_ptr_var which has the
Function storage class so a constant initializer is fine.
Component: Vulkan
VK-GL-CTS issue: https://gitlab.khronos.org/Tracker/vk-gl-cts/issues/686
Affects:
dEQP-VK.spirv_assembly.instruction.graphics.variable_pointers.nullptr_graphics.opvariable_initialized_null_*
dEQP-VK.spirv_assembly.instruction.graphics.variable_pointers.nullptr_graphics.opselect_null_or_valid_ptr_*
Change-Id: I9e2f118b1c774a84c0b6900ef0db081cb5256c2f
Tom Cooper [Wed, 25 Oct 2017 06:26:39 +0000 (08:26 +0200)]
YCbCr image views need a VkSamplerYcbcrConversionInfo
Ensure the image views are created using the
VkSamplerYcbcrConversionInfo object via the pNext
chaining, as this is mandated by the Spec.
Affects:
dEQP-VK.ycbcr.query.*
Components: Vulkan
VK-GL-CTS issue: 788
Change-Id: If641b08432c26ca5b32c084c4c0b9c7f935bcb94
Marcin Rogucki [Wed, 25 Oct 2017 11:33:44 +0000 (13:33 +0200)]
gl_PrimitiveIDIn tests have sketchy loop bounds.
Components: Vulkan
VK-GL-CTS issue: 778
Updates:
dEQP-VK.geometry.basic.primitive_id_in
dEQP-VK.geometry.basic.primitive_id_in_restarted
Change-Id: Icd2be56814de18067ce058a5f2a1da80338f80d6
Piers Daniell [Tue, 17 Oct 2017 20:11:55 +0000 (14:11 -0600)]
Update README for Vulkan multi-device platforms
Added some language to the Vulkan CTS README describing the
valid CTS log criteria for products that are multi-device
capable.
Change-Id: Ia5d39d5ab2c14b73bcad09531065b7dc5251c14b
Components: Vulkan
Piers Daniell [Wed, 18 Oct 2017 17:01:13 +0000 (11:01 -0600)]
Avoid hitting fp16 NaN values
Modify the test references values to avoid situations where
they hit NaN values when viewed as fp16. This can happen
when uploading {0.1, 0.5, 0.9, 0.2} to r8g8b8a8_unorm which,
depending on the rounding of the 0.5, could end up as a 0x7F
value which when viewed as the MSB of a fp16 value is a NaN.
Affects:
dEQP-VK.image.mutable.*
Components: Vulkan
VK-GL-CTS Issue: 784
Change-Id: I2adbd27aa7d94254c9f19d788eacace2ea95d10c
Arkadiusz Sarwa [Thu, 19 Oct 2017 12:08:13 +0000 (14:08 +0200)]
Fix subgroups basic test
Affects:
dEQP-VK.subgroups.basic.*_fragment_framebuffer
Components: Vulkan
VK-GL-CTS issue: 782
Change-Id: I54a125af75863e7dd56bf1c3785ee5e51be3d7b4
Chris Forbes [Mon, 16 Oct 2017 19:42:35 +0000 (12:42 -0700)]
Avoid watchdog timeout in Vulkan builtin precision tests
Calculating the reference intervals can take a substantial amount of
time. On some very slow targets, this can cause us to trip the watchdog
in the verification loop.
Same approach and limit as in modules/glshared/glsBuiltinPrecisionTests.cpp
Affected tests:
dEQP-VK.glsl.builtin.precision.*
Component: Vulkan
Bug:
67854695
Test: run on google devices
Change-Id: I492b656e7e2a9aa724bb3785bdf1416a4c22ec04
Mais Alnasser [Sun, 15 Oct 2017 19:10:59 +0000 (15:10 -0400)]
Use env.apiVersion when creating instance for object management test group
Affects: dEQP-VK.api.object_management.single.device_group
Components: Vulkan
VK-GL-CTS issue: 773
Change-Id: Ie6673133bc4d0de352b054fa797e2455433423d2
Mais Alnasser [Sun, 15 Oct 2017 15:16:15 +0000 (11:16 -0400)]
Fix compile error caused by usage of unit32_t instead of deUint32
Affects: Compilation
Components: Vulkan
VK-GL-CTS issue: 774
Change-Id: Ibc647d3714a818b61f99d1fc67903a24f59f8d6a
Vikram Kushwaha [Wed, 18 Oct 2017 19:55:03 +0000 (12:55 -0700)]
Do not use SFR bindings when device group has 1 device
As per the spec, when device group has 1 device:
sfrRects must not be specified
image create flag should not contain the BIND_SFR_BIT
Also use peer memory reads if VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
is supported. This saves an additionaly copy in the test.
VK-GL-CTS issue: 779
Change-Id: Ic316fc8858886f99156dc1b9c0a0810b3f48e714
Affects: dEQP-VK.device_group.sfr*
Components: Vulkan
Piers Daniell [Thu, 12 Oct 2017 18:09:00 +0000 (12:09 -0600)]
Ignore the patch number when comparing apiVersions
Modify the getSpirvVersionForAsm() and getSpirvVersionForGlsl()
functions to ignore the patch version number when checking
for Vulkan 1.0 vs Vulkan 1.1. It's often the case that the loader
or implementation will have a non-zero patch number to indicate
what version of the spec and vulkan header they built with.
Affects: startup
Components: Vulkan
VK-GL-CTS issue: 767
Change-Id: I947cb43db5e1f93e5b0b9cb93d205863b8e3c4b1
Daniel Koch [Thu, 12 Oct 2017 05:25:39 +0000 (01:25 -0400)]
Detect Core VK 1.1 device_group_creation support
Check if the core version of VK_KHR_device_group_creation
is supported in addition to checking for the extension string.
VK-GL-CTS Issue: 766
Component: Vulkan
Affects: dEQP-VK.api.object_management.*device_group
Change-Id: I5d3ee7733293d27e9ae94cdc085b85ea144be445
Andrey Tuganov [Thu, 28 Sep 2017 21:18:51 +0000 (17:18 -0400)]
Remove OpUConvert to signed integer tests
OpUConvert only convers to unsigned integers.
Affects: dEQP-VK.spirv_assembly.instruction.
compute.uconvert.*
Components: Vulkan
VK-GL-CTS issue: 720
Change-Id: I68b13c8584d64359e52910091347de03b933ba42
Slawomir Cygan [Wed, 11 Oct 2017 12:17:25 +0000 (14:17 +0200)]
Fix VK version selection in version_check.entry_points test
The test was validating only VK1.0 entrypoints due to wrong
version check.
Components: Vulkan
VK-GL-CTS Issue: 764
Affects: dEQP-VK.api.version_check.entry_points
Change-Id: Id555746f79d40c06cc8a95013608f9a1928847e2
Marcin Rogucki [Wed, 11 Oct 2017 09:26:33 +0000 (11:26 +0200)]
out_of_pool_memory test reports wrong error code
While creating default device a device version was used to select
enabled extensions however this should be based on used API version
(a minimum of Vulkan version and device version). It is fixed now.
Component: Vulkan
VK-GL-CTS issue: 763
Affects: dEQP-VK.api.descriptor_pool.out_of_pool_memory
Change-Id: I879d296ab56c1e2118dca84845cce6abc0e75f92
Marcin Rogucki [Mon, 9 Oct 2017 13:56:44 +0000 (15:56 +0200)]
dEQP-VK.ycbcr.* tests throw NotSupported incorrectly
Generic way of checking samplerYcbcrConversion support
added to DefaultDevice and Context.
Used in dEQP-VK.ycbcr.* tests group in checkImageSupport() function.
Component: Vulkan
VK-GL-CTS issue: 755
Affects: dEQP-VK.ycbcr.*
Change-Id: I0774a476f7ed94351c00387bc2e75d9c09492970
Marcin Rogucki [Tue, 10 Oct 2017 09:04:05 +0000 (11:04 +0200)]
Dispatch base test does not enable pipeline flag
dEQP-VK.compute.device_group.dispatch_base does not enable pipeline flag which violates valid usage rule: If any of baseGroupX, baseGroupY, or baseGroupZ are not zero, then the currently bound
compute pipeline must have been created with the VK_PIPELINE_CREATE_DISPATCH_BASE flag.
An overloaded method vktComputeTestsUtil.cpp::makeComputePipeline was added that allows to pass additional parameters:
- flags for pipeline creation
- flags for compute shader creation
Old method redirected to call a new one with flags set to zero.
A method vktComputeTestsUtil.cpp::makePipelineLayout no longer takes useDeviceGroups parameter and always sets pipeline layout flags to 0.
New method used in test with proper flags passed in.
Component: Vulkan
VK-GL-CTS issue: 760
Affects: dEQP-VK.compute.device_group.dispatch_base
Change-Id: I2c7400a55809b28bb7096dcdfeba2095c1541e1b
Slawomir Cygan [Tue, 10 Oct 2017 16:13:51 +0000 (18:13 +0200)]
Adjust the result verification in loop_control tests
Change epsilon value to 0.001f
Component: Vulkan
VK-GL-CTS issue: 742
Affects: dEQP-VK.spirv_assembly.instruction.compute.loop_control.*
Change-Id: I4601e026e53a2b4256f77b690c6938ac53d5991b
Alexander Galazin [Tue, 10 Oct 2017 07:36:18 +0000 (09:36 +0200)]
Update Vulkan README
Update references to the mustpass location,
parameterize package name.
Components: Vulkan
VK-GL-CTS issue: 759
Change-Id: I4f15a1aa5eb54ad4b11e3b7a3a5a391fe81b9dac
Maciej Jesionowski [Tue, 10 Oct 2017 09:29:26 +0000 (11:29 +0200)]
Exclude invalid spirv_assembly indexing cases
32 bit int cases were trying to convert to the same bit width, which is
forbidden by the spec. We can safely remove these cases as all other
tests in this group use 32 bit access, so no coverage is lost.
Excluded tests:
* dEQP-VK.spirv_assembly.instruction.*.indexing.*u32*
* dEQP-VK.spirv_assembly.instruction.*.indexing.*s32*
VK-GL-CTS issue: 702
Components: Vulkan
Change-Id: I049ab07824a92f757829c4db810ab099a3fb59e6
Alexander Galazin [Tue, 10 Oct 2017 07:11:48 +0000 (09:11 +0200)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I1c26d0567159f88871148d48d2509e469a3c8abf
Jason Ekstrand [Fri, 6 Oct 2017 23:10:26 +0000 (16:10 -0700)]
VK_KHR_incremental_present: Trigger the fence off vkQueueSubmit
Previously, the tests were using AcquireNextImage to trigger the fence
and using it to guard the vkFreeCommandBuffers call. However, the
acquire fence tells you when the PE is done with an image and when you
can start rendering to it. This meant that the test was waiting for the
start of the previous frame's rendering not the end before deleting the
previous frame's command buffer. This meant that the previous frame's
rendering could still be active on the GPU when the command buffer was
freed. By triggering the fence off of the vkQueueSubmit, we wait for
the end of the previous frame's rendering before freeing its command
buffer.
VK-GL-CTS issue: 756
Component: Vulkan
Affects:
dEQP-VK.wsi.*.incremental_present.*
Change-Id: I87b25eb2a25e4ae48c8c527d5d63c45d89a31a30
(cherry picked from commit
db916576a76a17c02c14390fc5d5b3db1488d1d2)
Igor Ostrowski [Mon, 9 Oct 2017 14:41:31 +0000 (16:41 +0200)]
DrawUtil: Fix depthBounds check
Affects:
dEQP-VK.glsl.builtin_var.fragdepth*
Components: Vulkan
VK-GL-CTS issue: 758
Change-Id: I4da79b236921e64f11566fe386b144497249123b
Alexander Galazin [Mon, 9 Oct 2017 08:32:25 +0000 (10:32 +0200)]
Merge vk-gl-cts/vulkan-cts-1.0.2 into vk-gl-cts/vulkan-cts-1.1.0
Change-Id: I3cef9c5da6238a99dcb3f96cdbe356963c79d88d
Jason Ekstrand [Fri, 6 Oct 2017 23:10:26 +0000 (16:10 -0700)]
VK_KHR_incremental_present: Trigger the fence off vkQueueSubmit
Previously, the tests were using AcquireNextImage to trigger the fence
and using it to guard the vkFreeCommandBuffers call. However, the
acquire fence tells you when the PE is done with an image and when you
can start rendering to it. This meant that the test was waiting for the
start of the previous frame's rendering not the end before deleting the
previous frame's command buffer. This meant that the previous frame's
rendering could still be active on the GPU when the command buffer was
freed. By triggering the fence off of the vkQueueSubmit, we wait for
the end of the previous frame's rendering before freeing its command
buffer.
VK-GL-CTS issue: 756
Component: Vulkan
Affects:
dEQP-VK.wsi.*.incremental_present.*
Change-Id: I87b25eb2a25e4ae48c8c527d5d63c45d89a31a30