Extremely low FPS when using Qt6 on Windows

[dov]

With the goal of running my application on Windows with Qt6, I have tested the following scenarios, with varying success:

• vsgQt with Qt5 - Works fine and I get the expected 60 fps
• vsgQt with Qt6 - Extremely slow, about 2fps
• https://github.com/nolankramer/vsgQt6 with Qt6 - Extremely slow, about 1-2fps

I have recompiled Qt6 on my own on Windows, using vcpkg with the "vulkan" feature enabled.

When running the vsgQt6 vsgqtviewer cube10.vsgb with validation layers I get the following warnings:

vsgqtviewer.exe -d c:\Temp\cube10.vsgb
VUID-VkShaderModuleCreateInfo-pCode-08737(ERROR / SPEC): msgNum: -1520283006 - Validation Error: [ VUID-VkShaderModuleCreateInfo-pCode-08737 ] | MessageID = 0xa5625282 | vkCreateShaderModule(): pCreateInfo->pCode (spirv-val produced an error):
Expected Image to have the same type as Result Type Image
  %222 = OpSampledImage %221 %215 %219
. The Vulkan spec states: If pCode is a pointer to SPIR-V code, pCode must adhere to the validation rules described by the Validation Rules within a Module section of the SPIR-V Environment appendix (https://vulkan.lunarg.com/doc/view/1.3.290.0/windows/1.3-extensions/vkspec.html#VUID-VkShaderModuleCreateInfo-pCode-08737)
    Objects: 0
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849(ERROR / SPEC): msgNum: 1132206547 - Validation Error: [ VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 ] | MessageID = 0x437c19d3 | vkCreateGraphicsPipelines(): pCreateInfos[0].pStages[1] After specialization was applied, VkShaderModule 0x27d60e0000000019[] produces a spirv-val error (stage VK_SHADER_STAGE_FRAGMENT_BIT):
Expected Image to have the same type as Result Type Image
  %222 = OpSampledImage %221 %215 %219
. The Vulkan spec states: If a shader module identifier is not specified, the shader code used by the pipeline must be valid as described by the Khronos SPIR-V Specification after applying the specializations provided in pSpecializationInfo, if any, and then converting all specialization constants into fixed constants (https://vulkan.lunarg.com/doc/view/1.3.290.0/windows/1.3-extensions/vkspec.html#VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849)
    Objects: 0
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849(ERROR / SPEC): msgNum: 1132206547 - Validation Error: [ VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 ] | MessageID = 0x437c19d3 | vkCreateGraphicsPipelines(): pCreateInfos[0].pStages[1] After specialization was applied, VkShaderModule 0x27d60e0000000019[] produces a spirv-val error (stage VK_SHADER_STAGE_FRAGMENT_BIT):
Expected Image to have the same type as Result Type Image
  %222 = OpSampledImage %221 %215 %219
. The Vulkan spec states: If a shader module identifier is not specified, the shader code used by the pipeline must be valid as described by the Khronos SPIR-V Specification after applying the specializations provided in pSpecializationInfo, if any, and then converting all specialization constants into fixed constants (https://vulkan.lunarg.com/doc/view/1.3.290.0/windows/1.3-extensions/vkspec.html#VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849)
    Objects: 0

I do get some warnings with the vsgviewer without qt as well, but much less, and then the FPS is the expected 60 fps.

c:\...\vsgExamples\build\bin>vsgviewer -d c:\Temp\cube10.vsgb
VUID-VkShaderModuleCreateInfo-pCode-08737(ERROR / SPEC): msgNum: -1520283006 - Validation Error: [ VUID-VkShaderModuleCreateInfo-pCode-08737 ] | MessageID = 0xa5625282 | vkCreateShaderModule(): pCreateInfo->pCode (spirv-val produced an error):
Expected Image to have the same type as Result Type Image
  %222 = OpSampledImage %221 %215 %219
. The Vulkan spec states: If pCode is a pointer to SPIR-V code, pCode must adhere to the validation rules described by the Validation Rules within a Module section of the SPIR-V Environment appendix (https://vulkan.lunarg.com/doc/view/1.3.290.0/windows/1.3-extensions/vkspec.html#VUID-VkShaderModuleCreateInfo-pCode-08737)
    Objects: 0
VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849(ERROR / SPEC): msgNum: 1132206547 - Validation Error: [ VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849 ] | MessageID = 0x437c19d3 | vkCreateGraphicsPipelines(): pCreateInfos[0].pStages[1] After specialization was applied, VkShaderModule 0x27d60e0000000019[] produces a spirv-val error (stage VK_SHADER_STAGE_FRAGMENT_BIT):
Expected Image to have the same type as Result Type Image
  %222 = OpSampledImage %221 %215 %219
. The Vulkan spec states: If a shader module identifier is not specified, the shader code used by the pipeline must be valid as described by the Khronos SPIR-V Specification after applying the specializations provided in pSpecializationInfo, if any, and then converting all specialization constants into fixed constants (https://vulkan.lunarg.com/doc/view/1.3.290.0/windows/1.3-extensions/vkspec.html#VUID-VkPipelineShaderStageCreateInfo-pSpecializationInfo-06849)
    Objects: 0

The file cube10.vsgb is the output of vsgconv cube10.stl cube10.vsgb, i.e. a simple no frills cube.

Does anyone have any idea what is going on, and how can I debug this?

[reedev]

Hi,

I am on Qt6 for a long time with a few apps of mine, on Linux and Windows, and fwiw I don't see such differences. One difference though: I use the 'standard' vsgQt project and not the one from Nolan. Just specify -DQT_PACKAGE_NAME:STRING=Qt6 when running cmake does it for me.

On some platforms, I use Qt5 and that's working ok too.

Can you try vsgQt?

[dov]

Thanks. I got the same result with standard vsgQt as well, which is why I tried Nolan's version.

Did you compile vsgQt on your own?

I should really check that the non-vsg Vulkan stuff works as expected on Qt6.

[reedev]

aha, yes, I build VSG and the projects myself.

If you can make the issue small, I can try building and running here too.

[dov]

Sorry, I actually meant to ask if you build Qt6 on your own.

[reedev]

oh yes, I build Qt myself too (on each platform)

[dov]

The problem occurs with vsgqtviewer of the vsgQt project, so there is really nothing to simplify in terms of the test.

I'll try run a non-vsg Qt6 program with Vulkan later on today to check if it runs ok.

[reedev]

ok I will try to try it out too. At least to sync with you...

[robertosfield]

The validation error is due to an issue within the Vulkan Validation Layer that is enforcing a rule that has now been removed. Future versions of the Validaition Layer should fix this false positive.

This reported "error" is likely just a red herring for the slow down you are seeing. A big slow downs like this I'd suspect a synchronization issue somewhere or that CPU vulkan driver is being used rather than GPU driver.

[robertosfield]

The vsgdeviceselection example shows how to query the available Vulkan drivers/devices.

[dov]

Ok, thanks. I'll try to integrate that code into a vsgQt6 example and will see what I get.

[dov]

Found it! :-) It was a question of wrong Qt environment.

The story and the solution were as follows:

1. For compilation of Qt6, I used vcpkg with a vcpkg.json file that referenced Qt6 (with the vulkan feature turned on).
2. After compilation I used windeployqt.exe to setup the dependencies so that I can run the executable in the compilation directory.
3. However, for whatever reason, there are two different version of this executable in the the vcpkg tree. One in ...\installed\x64-windows\tools\Qt6\bin and one in ...\packages\qtbase_x64-windows\tools\Qt6\bin. I had initially used the "installed" version, which triggered the bad behavior. (I was confused by the fact that I needed to use the --force flag of windeployqt.exe in order to get a running executable.)
4. Once I switched to the "packages" version of windeployqt6 the slow FPS disappeared!

Thanks for the help! Just the confirmation that vsg worked fine under Qt6 gave me confidence to continue looking for the solution.

[reedev]

Nice! Glad I could give you the confidence.

Interesting to hear about the windeploy tool, I collect the dlls manually.