Fix NegativeRayTracing.AccelerationStructuresOverlappingMemory4

[spencer-lunarg]

When ran on intel mesa I get

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x440300000, 0x440300180).

on Lavapipe I get

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x790c1d987000, 0x790c1d987098).

(or sometimes lavapipe gets Did not receive expected error 'VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03704')

[arno-lunarg]

can you give me the full error log printed with --print-vu?

On lavapipe the fact that you don't always get the same errors tells me you need to blacklist this test. Due to optimizations I had to make, I may not emit errors for all memory overlaps, results depend on what drivers do. If there is one though it is always detected, that's all I care about. Going through the hassle of adding yet another "m_errorMonitor->detect_at_least_one_error_and_if_there_are_more_thats_actually_fine()" is not worth it IMO

[spencer-lunarg]

can you give me the full error log printed with --print-vu?

the errors in the summary are what I get

... as long as one machine on CI runs this, I am happy to add whatever GTEST_SKIP you want

[arno-lunarg]

Ah you only get those? No others?

[spencer-lunarg]

yes, just those

[arno-lunarg]

But so you get a bunch of did not receive expected error?
Can you get me the full test output please, I'm confused

[spencer-lunarg]

.... I run AccelerationStructuresOverlappingMemory no issues

---

I run AccelerationStructuresOverlappingMemory3 no issues

---

I run AccelerationStructuresOverlappingMemory4, I get (on Intel Mesa, lets start there)

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x440300000, 0x440300180).

[arno-lunarg]

So you do get other (expected) errors x)
what's your --print-vu log for Mesa?

[spencer-lunarg]

Intel Mesa

---

Lavapipe

[arno-lunarg]

Is that with --print-vu? I need to look at errors to better understand what gets detected

[spencer-lunarg]

no, that is as normal, here is a --print-vu

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[0].scratchData overlaps with srcAccelerationStructure of pInfos[0], which is in update mode, on device address range [0x440300000, 0x440300180).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the srcAccelerationStructure member of any element of pInfos with a mode equal to VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR (including the same element), which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705)
Objects: 2
    [0] VkCommandBuffer 0x57e8a867cda0
    [1] VkAccelerationStructureKHR 0x140000000014

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[1].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x440300000, 0x440300180).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the srcAccelerationStructure member of any element of pInfos with a mode equal to VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR (including the same element), which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705)
Objects: 2
    [0] VkCommandBuffer 0x57e8a867cda0
    [1] VkAccelerationStructureKHR 0x1f000000001f

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[1].scratchData overlaps with srcAccelerationStructure of pInfos[0], which is in update mode, on device address range [0x440300000, 0x440300180).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the srcAccelerationStructure member of any element of pInfos with a mode equal to VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR (including the same element), which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705)
Objects: 2
    [0] VkCommandBuffer 0x57e8a867cda0
    [1] VkAccelerationStructureKHR 0x140000000014

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x440300000, 0x440300180).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the srcAccelerationStructure member of any element of pInfos with a mode equal to VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR (including the same element), which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705)
Objects: 2
    [0] VkCommandBuffer 0x57e8a867cda0
    [1] VkAccelerationStructureKHR 0x2a000000002a

/home/fricke/github/Vulkan-ValidationLayers/tests/framework/error_monitor.cpp:309: Failure
Failed
Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705 ] | MessageID = 0x30b2f3fd
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].srcAccelerationStructure overlaps with scratch buffer of pInfos[0] on device address range [0x440300000, 0x440300180).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the srcAccelerationStructure member of any element of pInfos with a mode equal to VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR (including the same element), which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03705)
Objects: 2
    [0] VkCommandBuffer 0x57e8a867cda0
    [1] VkAccelerationStructureKHR 0x2a000000002a

Validation Error: [ VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03704 ] | MessageID = 0x331537d3
vkCmdBuildAccelerationStructuresKHR(): pInfos[2].scratchData overlaps with scratch buffer of pInfos[1] on device address range [0x440300000, 0x4403100f4).
The Vulkan spec states: The range of memory backing the scratchData member of any element of pInfos that is accessed by this command must not overlap the memory backing the scratchData member of any other element of pInfos, which is accessed by this command (https://docs.vulkan.org/spec/latest/chapters/accelstructures.html#VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03704)
Objects: 1
    [0] VkCommandBuffer 0x57e8a867cda0

/home/fricke/github/Vulkan-ValidationLayers/tests/framework/error_monitor.cpp:349: Failure
Failed
Did not receive expected error 'VUID-vkCmdBuildAccelerationStructuresKHR-scratchData-03704'

[arno-lunarg]

Hummm yeah I think you are going to have to skip those tests ... Memory overlaps detection depends on the result of a map insertion, and this is kind of unpredictable (depends on the map implementation, the values inserted)... It was already tedious to get green CI for those, so I don't want to spend more time on this unless I reeeeeally need to (〜￣▽￣)〜

[spencer-lunarg]

Hummm yeah I think you are going to have to skip those tests

If the test is broken on both drivers on my machine, I rather just remove it then, sounds like a broken test and just a matter of time before it creeps into CI failures