Vkguide 2 Beta Comments

[utterances-bot]

- Vulkan Guide

Practical guide to vulkan graphics programming

https://www.vkguide.dev/docs/new_vkguide/chapter_1/vulkan_mainloop/

[vblanco20-1]

This is the feedback comment channel for Vk guide 2 Early Beta. The comments here will be removed once full release is done. If you find issues with the current version or typos, post here to report them.

[khushmmehta]

Do you have any idea how i could set up this project for CLion?

[MasterDrake]

The hyperlink "Initializing Vulkan" on Chapter0: Code Walkthrough leads to the old version.

[MasterDrake]

1) I don't know if I did something wrong using the starting point or I missed some configuration that had to be done, but my engine project wasn't linked to "vkboostrap" so I coulnd't include the header or link with the .lib. I had to do it manually and the "project dependencies" thing didn't help either.

2) Something I never understood is this: "By doing that = {} thing, we are letting the compiler initialize the entire struct to zero. This is critical, as in general Vulkan structs will have their defaults set in a way that 0 is relatively safe. By doing that, we make sure we don’t leave uninitialized data in the struct." If so, then that doesn't also mean that .pNext is already set to either 0 or nullptr, so it's pointless to set it?

3)Typos:

• "Now that we have the VkCommandPool created, and stored in the _commandPool member, we can allocate our command buffer from it.``" Extra " "
• Command buffers can be Primary of Secondary level. Should be "or"
• Some enum aren't highlighted like "VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT" unlike "VK_COMMAND_BUFFER_LEVEL_PRIMARY"

4) Initialization steps lack the hyperlink to the next step

[MasterDrake]

• Mainloop Code 1) Add something to the frame data struct that tells to add that code to your previous. Silly I know, but I got confused for a moment. 2) "(VkFenceCreateInfo)[https://registry.khronos.org/vulkan/specs/1.3-extensions/html/chap7.html#VkFenceCreateInfo], (VkSemaphoreCreateInfo)[https://registry.khronos.org/vulkan/specs/1.3-extensions/html/chap7.html#VkSemaphoreCreateInfo]" are not properly formatted. 3) Typos:
• "when a command buffer is started", when should be capitalized.
• "this is to shorten all other references to it", this should be capitalized.
• "transitioning a image has loads of possible options", should be "Transitioning an image
• "we are doing ALL_COMMANDs. This is inneficient, as it will stall the GPU pipeline a bit.", 's' of command should be capitalized as well and "inneficient" -> inefficient
• "Other than that, it only needs the semaphore handle. device index parameter is used for multi-gpu semaphore usage, but we wont do any of it. and value is used for timeline semaphores, which are a special type of semaphore where they work through a counter intead of a binary state. We will also not be using them, so we can default it to 1" needs some minor rework.

[MasterDrake]

• Improving the render loop 1) Suggest how to include VmaAllocator -> "#include "vk_mem_alloc.h" 2) "We have an AllocatedImage structure that already holds format, size, imageview, and the image itself". We don't as far as I know by the time we reach that chapter. VkTypes hasn't been touched yet(it's also the place where vk_mem_alloc is included). 3) Typoes:
• "new draw loop" should be capitalized.
• "Vulkan has 2 main ways of copying one image to another. you can use VkCmdCopyImage or VkCmdBlitImage CopyImage is faster" needs a dot after VkCmdBlitImage.
• "//transtion the draw image and the swapchain image into their correct transfer layouts" in the last code piece -> "transition"
• "The main diference we have in the render loop is that we no longer do the clear on the swapchain image" -> "difference" 4) I got linking error due to Vma, I guess #define VMA_IMPLEMENTATION is needed. I added it in the main. 5) You say "In our case, we want TransferSRC because we will have it as a copy source to the swapchain, Storage because thats the “compute shader can write to it” layout, and Color Attachment so that we can use graphics pipelines to draw geometry into it." but in the code snippet "VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT" is missing. 6) So far everything works as intended(flashing blue on screen) but I got a validation error that has to do with how drawImage was created but I'm not sure how to fix it: "Validation Error: [ VUID-vkCmdClearColorImage-image-00002 ] Object 0: handle = 0x1a047dde250, type = VK_OBJECT_TYPE_COMMAND_BUFFER; Object 1: handle = 0xcfef35000000000a, type = VK_OBJECT_TYPE_IMAGE; | MessageID = 0x636f8691 | vkCmdClearColorImage(): image (VkImage 0xcfef35000000000a[]) was created with usage VK_IMAGE_USAGE_TRANSFER_SRC_BIT|VK_IMAGE_USAGE_STORAGE_BIT|VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT (missing VK_IMAGE_USAGE_TRANSFER_DST_BIT). The Vulkan spec states: image must have been created with VK_IMAGE_USAGE_TRANSFER_DST_BIT usage flag (https://vulkan.lunarg.com/doc/view/1.3.268.0/windows/1.3-extensions/vkspec.html#VUID-vkCmdClearColorImage-image-00002)"

I think it's a me problem :/

[MasterDrake]

• Vulkan Shaders - Code 1) "We are specifyinf" ->" specifying". 2) "initializing the layout and descriptors" should be capitalized. 3) "we will be storing one of those descriptor allocators in our engine as the global allocator." should be capitalized. 4)
  • The second half of the init_descriptors code is missing
• "With this done, we now have a descriptor set we can use to bind our draw image, and the layou we needed. We can finally proceed to creating the compute pipeline" needs an ending dot and "layout". 5)
  • Minor nitpick but it would be useful to state the includes for vk_pipelines .h/cpp, "#include " and "#include " respectively.
• It would be nice, IMO, to drop the *outShader + return bool and use directly a std::optional, since VkShaderModule should be trivially copyable(I guess) and it's very c++-ish. 6) "On this shader we have, we wont need those so we can skip them, leaving only the DescriptorSetLayout.": I guess "we have ," is a typo. 7) Add "#include

  include " as suggestion in the second half of "init_pipelines" code.

  Or even better, "#include "fmt/printf.h"" and use "fmt::print("Error when building the compute shader\n");" instead of std::cout 8) "We are now ready to draw with it" needs a dot. 9) "Go back to the draw() function, we will replace the vkCmdClear with a compute shader invocation.": I guess you mean "draw_main()". 10) I got a linking error because "allocate" from DescriptorAllocator wasn't defined, I guess it's missing from 4). 11) I got an error when I tried to load a shader, because apparently it wasn't compiled. Even when I added "Shaders" to the project dependencies of "engine", it still didn't build. You have to go to "Configuration Manager" when you click "Debug/Release" and then tick "build" on the Shader project. Could be useful for other VS users. 12) I got a validation error about push constants because I was loading the wrong shader. " if (!vkutil::load_shader_module("../../shaders/gradient_color.comp.spv", _device, &computeDrawShader)) ": it's "gradient.comp.spv" not "gradient_color.comp.spv" 13) The validation error in the previous chapter about VkClear and draw Image is gone, but so far in this chapter the descriptor set/pool/set_layout are not deleted properly.

[MasterDrake]

• Setting up ImGui 1) Minor nitpick but indicate where to put the new code relative to the old code, to avoid confusion about order and what not.
• init_command()
• init_sync_structures() 2) I needed to add imgui as dependency to "engine" to link it but manually add the header directory :/ Again, feels like I fucked up somewhere somehow. I also needed to link manually "imgui.lib" :/ 3) "Imgui will draw using actual gpu draws with meshes and shaders, it will not do a compute draw like we are doing at te moment. ": "the" moment. 4) "With the attachment info done, we can make the VkRenderingInfo." You re-pasted the "VkRenderingAttachmentInfo vkinit::attachment_info" code instead of VkRenderingInfo.

[MasterDrake]

• Push Constants and new shaders 1) "we will be using PushConstants" should be capitalized. 2) "lets now use them from the render loop"-> "Let's now use them from the render loop:" 3) Minor trivialities: const ComputeEffect& effect = backgroundEffects[currentBackgroundEffect]; and "ImGui::InputFloat4("data1", reinterpret_cast<float>(&selected.data.data1)); ImGui::InputFloat4("data2", reinterpret_cast<float>(&selected.data.data2)); ImGui::InputFloat4("data3", reinterpret_cast<float>(&selected.data.data3)); ImGui::InputFloat4("data4", reinterpret_cast<float>(&selected.data.data4));"

maybe?

[MasterDrake]

• Push Constants and new shaders: Two minor nitpicks: 1) In order to demostrate the awesomeness of imgui and the newest shader, I suggest this instead: "if (ImGui::Begin("Background")) { ComputeEffect& selected = backgroundEffects[currentBackgroundEffect];

  ImGui::Text("Selected effect: ", selected.name);

  ImGui::SliderInt("Effect Index", &currentBackgroundEffect, 0, backgroundEffects.size() - 1); ImGui::ColorEdit4("Data1", reinterpret_cast<float>(&selected.data.data1)); ImGui::ColorEdit4("Data2", reinterpret_cast<float>(&selected.data.data2));

  ImGui::End(); }"

2) I have a doubt, VS reports me that I'm using 275 MB whether I'm in release or debug mode. Is it normal? I'm not even loading meshes or textures :/

[vblanco20-1]

@MasterDrake, thanks for the reports, ill begin fixing all of those.

[MasterDrake]

@MasterDrake, thanks for the reports, ill begin fixing all of those.

Have fun 😊 I'll continue tomorrow and see how it goes Thanks you so much for this, btw!!

[vblanco20-1]

@MasterDrake Ive tested it, my debug ram cost is also at 200+ MBs. seems like the cost just comes by initializing vulkan and its validation layers.

[MasterDrake]

• Mesh Buffers 1) Put AllocatedBuffer in "vk_types.h" 2) "Now we need a function to create those buffers and fill them on the gpu.": Missing function decleration and header (love the use of span). 3) "On the index buffer we use VK_BUFFER_USAGE_INDEX_BUFFER_BIT to signal that we are going to be using that buffer for indexed draws." and VK_BUFFER_USAGE_TRANSFER_DST_BIT as well. 4) init_triangle_pipeline() pipeline builder has reference to depth testing yet to be touched. 5) std::cout -> fmt::printf 6) Very very very annoying problem: I had #define VMA_IMPLEMENTATION in the main and it worked with no problem so far. But now with the upload mesh I kept getting "error C2027: use of undefined type 'VmaAllocation_T'". I ended up declaring "#define VMA_IMPLEMENTATION #include "vk_mem_alloc.h"" in the vk_engine.cpp file. Again, I don't know if it's me or what, but shit's annoying.

[MasterDrake]

• Mesh Buffers I got into two problems when I run the program: 1) I got a validation error about an extension not being added. Apparently you skipped a portion where you added: " VkPhysicalDeviceVulkan12Features features12{}; features12.bufferDeviceAddress = true; features12.descriptorIndexing = true; " + .set_required_features_12(features12) in init_vulkan(); 2) Speaking of init_vulkan(), the second validation error was about vma allocator, it was missing a flag for device addressing: "allocatorInfo.flags = VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT;"

[MasterDrake]

• Mesh Loading 1) I guess "#include " is needed in vk_loaders.h. You also forward declared class VulkanEngine;

2) "fmt::print("Loading GLTF: {}.", filePath);"

3) "std::optional<std::vector<std::shared_ptr>> loadGltfMeshes(VulkanEngine* engine, std::string_view filePath);" -> std::filesystem::path filePath also #include .

4) "#include <fastgltf/parser.hpp>

include <fastgltf/tools.hpp>

include <fastgltf/glm_element_traits.hpp>"

Also "//needed for the fastgltf variants template struct overloaded : Ts... { using Ts::operator()...; };" Never used this library so a lot of this stuff was new to me. I also was wondering if there was a way to query indices and vertices size from fastgltf so you can resize the vectors accordingly. It would be a good idea to const the accessors.

5) Since we're using c++20, I'm not asking to use ranges for stuff like: " constexpr bool OverrideColors = false; if (OverrideColors) std::ranges::for_each(vertices, [](Vertex& vtx) { vtx.color = glm::vec4(vtx.normal, 1.f); });" But it would be cool to use if stataments with initializer like this: "if (auto normals = p.findAttribute("NORMAL"); normals != p.attributes.end()) if (auto uv = p.findAttribute("TEXCOORD_0"); uv != p.attributes.end()) if (auto colors = p.findAttribute("COLOR_0"); colors != p.attributes.end())" Reducing the scope isn't vital here, but it's a c++17-ism that's pretty neat IMO.

6) #include <glm/gtx/transform.hpp> when fixing the camera Y axis.

7) "Now we need to change the render pass begin info to use this depth attachment and clear it correctly.": specify that is in the draw_geometry() function. 8) "We made the depth option when the pipelinebuilder was made, buit left it dsiabled": "dsiabled" -> "disabled" 9) " std::vector testMeshes;" -> "std::vector<std::shared_ptr>" 10) Again, I had to manually link fastgltf and fastgltf_simdjson. :( 11) I got validation errors screaming at me because a pipeline builder depth image format was create with VK_FORMAT_UNDEFINED instead of VK_FORMAT_D32_SFLOAT so apparently I had to change this line in init_triangle_pipeline(): "pipelineBuilder.set_depth_format(VK_FORMAT_UNDEFINED);" to " pipelineBuilder.set_depth_format(_depthImage.imageFormat);" even if the depth testing was disabled.

I wish you can name vulkan objects so the validation errors would be useful or should I say precise -.-

[MasterDrake]

• Descriptor Abstractions 1) In DescriptorAllocatorGrowable::create_pool, if we know we're gonna add poolRatios. size() elements to the vector, we should resize it before hand. No biggie, I know. poolSizes.resize(poolRatios.size()); 2)" std::deque is guaranteed to keep pointers to elements valid, so we can take advante": "advante" -> "advantage" 3) "We will hold view materia and projection matrix separated": "materia" -> "matrix" 4) OK, something's off. I don't have any:"_frameDescriptors" in my FrameData. I also never changed DescriptorAllocator to DescriptorAllocatorGrowable in the init_descriptors neither I added the uniform descriptor type neither I initialized the said DAG in the FrameData :O Nor I changed "globalDescriptorAllocator " to a DescriptorAllocatorGrowable. Ok, I see it's in the next chapter but still :/

[MasterDrake]

• Textures 1) "We can now update the init_mesh_pipeline() function with this. We will be modifying the start part, changing the fragment shader and connecting the descriptor set layout to the pipelinelayout creation.": You copy-pasted the wrong snippet there. 2) So far my code doesn't work properly. It just displays a red rectangle. I hope next chapters refactor will help trim down the error or remove it altogether :(

3) Mea culpa, I previously suggested to "poolSizes.resize(poolRatios.size());" in the DescriptorAllocatorGrowable::create_pool, but I obviously meant ".reserve" -.- I keep making this mistake even when I'm aware of this pitfall, grr

[MasterDrake]

• Engine Architecture 1) "(we will improve tis later)" -> this unless it was on purpose and the next c++ snippet code isn't syntactically highlighted. 2) "Next: Meshes and Camera": should point to "setting up materials".

[MasterDrake]

• Setting up materials 1) Btw, state that:
• "RenderObject " should go in "vk_engine.h"
• "Material-Data/Pipeline/Pass" should go in "vk_types."
• "GLTFMetallic_Roughness" should go in "vk_engine.h" 2) Indicate that " metalRoughMaterial.build_pipelines(this);" just has to be added in init_pipelines, it doesn't replace its entire content. 3) "^nextlink" instead of "meshes and camera" link

[MasterDrake]

• Meshes and camera

1) Scene node classes go into "vk_types.h".

2) Not sure whether you're supposed to use "auto& c : children" or "auto c : children" when using shared_ptrs, but you're using both approaches and it's confusing.

3) "MaterialInstance material;": "MaterialData has become MaterialInstance*.

4) "void MeshNode::Draw(const glm::mat4& topMatrix, DrawContext& ctx)" goes into "vk_engine.cpp"

5) So far, GeoSurface doesn't have a material, so "def.material = &s.material->data;" is an error.

6) "We are binding the data every draw which is inneficient but we will fix that later.": "inneficient"->"inefficient"

7) "std::unordered_map<std::string, shared_ptr<Node*>> loadedNodes;": should be std::shared_ptr

8) "loadedNodes[m.name] = std::move(newNode);": should be m->name.

9) "newNode->mesh = std::make_shared(m);": this becomes "newNode->mesh = m;" in the source code, not sure the difference. Can't even compile without changing it.

10) For some messy reason, I didn't have "Cube" loaded so I crashed. I used a mesh I knew I had in my unordered_map(ehi, can we change it to martinus one and see how much is better) and it's huge so I'm not sure I'm doing ok :D Hopefully with the free camera... .Mantaining a project like this with so many refactors can become a pain in the ass. IMO, some of the deprecated stuff should be resolved before jumping to the next stage.

[MasterDrake]

• Interactive Camera 1) "#include "SDL_events.h" #include " should be added in camera.h 2) "Improving it is an excersise to the reader.": "excersise" -> "exercise". 3) "For mouse motion, we acumulate horizontal and vertical movement of the mouse into the pitch and yaw properties.": "acumulate" -> "accumulate". 4) getXMatrix() can be made const, as well the actual objects created in the functions.

[MasterDrake]

• GLTF Scene nodes 1) "LoadedGLTF" goes into "vk_loader.h" 2) Change iostream print to fmt and change to ""Failed to load binary glTF:" when loading a GLB to differentiate the kind of error message depending on the type. 3) Broken indentation "```

Before we begin loading everything, We are going to create some arrays to hold the structures. In GLTF files, everything works through indices, so we need a way to handle that . For example a mesh node will give the mesh index, not name or anything similar.

4) "    loadedScenes["structure"]->Draw(glm::mat4{ 1.f }, drawCommands);": "drawCommands" should be  "mainDrawContext".
5) I already said that but it needs a clean up since test meshes are still loaded and rendered.
6) It's needed "#include "vk_descriptors.h" in "vk_loaders.h" because of DescriptorAllocatorGrowable we just added.
7) "if (node->parent.get() == nullptr)": gives me an error that I can't understand and the source code is this: "if (node->parent.lock() == nullptr)"
8) Not gonna lie, this is a pretty big function we've got here and it's not over yet

[MasterDrake]

• GLTF Textures 1) load_image is a vk_loader.cpp function. 2) "#include "stb_image.h" at the top of "vk_loader.cpp" file. 3) "fastgltf::Image& image" in load_image can be made const, same when we're looping "for (fastgltf::Image& image : asset->images) " or meshes as well. 4) "// we failed to load, so lets give the slot a default white texture to not completely break loading images.push_back(engine->_errorCheckerboardImage);": you say white texture but then you set the checkerboard one. 5) I don't have a "destroy_image" function vk_engine.cpp 6) For some reasons, it doesn't load every texture and my checkerboard texture is not black-magenta. I guess it has to do with shaders but I just copied them from you because I couldn't be bothered to reconfigure the projects every time I wanted to compile the shaders. But again, that's a me-problem. The fact the only some are loaded is funny to me. (I should investigate but now I want to finish, your source code projects works fine so it's not my gpu) I'll do a file-diff when I'm done. 7) If you're aware of some fancy imgui widget for display camera info and maybe modifying it, let me know. 8) " for (auto& r : drawCommands.OpaqueSurfaces) ": it's mainDrawContext. 9) "Make sure to reset the array of transparent draws too at the end of the function": I guess you mean at the beginning of "update_scene" function.

[MasterDrake]

• Faster Draw 1) " MaterialInstance* lastMaterial = nullptr;" should be MaterialData. 2) "drawCommands" should be mainDrawContext. 3) That was awesome. It flows much better and it's even easier to grasp. Bugs and performance aside, I really like this. I'll see what can I do to fix my mess and I'll post my final considerations and suggestions/wishlist.

Thank you so much!!

[senorsenorjunior]

"We set queueFamilyIndex to the _graphicsQueueFamily that we grabbed before. This means that the command pool will create commands that are compatible with any queue of that “graphics” family."

Both the code in the lesson and the function and the function in vk_initializers.cpp don't actually set the queueFamilyIndex field.

    VkCommandPoolCreateFlags flags /*= 0*/)
{
    VkCommandPoolCreateInfo info = {};
    info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
    info.pNext = nullptr;

    info.queueFamilyIndex = queueFamilyIndex;
    info.flags = flags;
    return info;
}```

By the way, thanks a ton for this updated guide! :)

[MasterDrake]

• Refactoring I don't know if you care but now I'm trying to fix my mess, maybe it can help with what's missing from the guide :)

1) My "LoadedGLTF::clearAll" is different from the one in your source code. Not to mention the pointless "samplersToDestroy" variable after the samplers are destroyed. Unless I'm missing something. 2) Apparently I don't have a "GPUGLTFMaterial" for the uniform buffer :/ 3) I don't have an "AllocatedBuffer cameraBuffer" in FrameData 4) I don't have "EngineStats" as well, but that's not surprising. 5) I don't have " // the format for the draw image VkFormat _drawFormat;" 6) you added "init_renderables" and got rid of "init_triangle_pipeline" and "init_mesh_pipeline" 7) there's not render_nodes() for me. 8) I don't have "std::vector _framebuffers;" nor "VkDescriptorSet _defaultGLTFdescriptor;" 9)"globalDescriptorAllocator" is not Growable, it's fine? 10) I don't have " VkDescriptorPool _descriptorPool;" either. 11) I don't have " AllocatedBuffer _defaultGLTFMaterialData;"

12) In the end I had to got rid of: " VkPipelineLayout _trianglePipelineLayout; VkPipeline _trianglePipeline; VkDescriptorSetLayout _singleImageDescriptorLayout;

MaterialData defaultData;

VkPipelineLayout _meshPipelineLayout; VkPipeline _meshPipeline;

std::vector<std::shared_ptr> testMeshes;"

I'm not ready for vk_engine.cpp :(

[vblanco20-1]

@MasterDrake most of those you have mentioned are never used, and its just dead code i havent fixed yet. you dont have to worry. Chapter 5 in particular is prone to that dead code because its the "original" code for the tutorial chapters, chapters 2 to 4 were made after it.

[MasterDrake]

• The night before the game awards is long so it's better if I do something in the meanwhile 1)"// we want to immediately abort when there is an error. In normal engines this // would give an error message to the user, or perform a dump of state. using namespace std;": not sure where this message is supposed to be about and I'm not sure that is the right place either 2) Now init_default_data() initializes rect_vertices uv coordinates. 3) everything about default_data in init_default_data() is gone. 4) loadedEngine is not nulled in cleanup(). 5) Now draw_main draws the background and the geometry, but we should remove "draw the triangle" comment. 6) Ah, render_nodes() is empty :( So i guess this is the leftover you're talking about. 7) How about " if (VkResult e = vkAcquireNextImageKHR(_device, _swapchain, 1000000000, get_current_frame()._swapchainSemaphore, nullptr, &swapchainImageIndex); e == VK_ERROR_OUT_OF_DATE_KHR) return;" ? :D 8) this "VK_CHECK(vkResetCommandBuffer(get_current_frame()._mainCommandBuffer, 0));" should come after "VkCommandBuffer cmd = get_current_frame()._mainCommandBuffer;" for brevity, as you mentioned. 9)" VkResult presentResult = vkQueuePresentKHR(_graphicsQueue, &presentInfo);" is never used. 10) Don't know about you but I think it's RenderObject job to provide the < operator overload. If you don't want to taint the struct, at least as global/free function. Or not, it's used once :D 10) Why not "if(ImGui::Begin("Stats")) like you did with background? 11) "mainDrawContext.OpaqueSurfaces.clear(); mainDrawContext.TransparentSurfaces.clear();" are in a different location. 12) " 1700.f / 900.f": we should use windows extents. 13) No " _mainDeletionQueue.push_function([&]() { vmaDestroyAllocator(_allocator); });" in your code in init_vulkan() :/ 14) use_default_format_selection is disabled and you're using this instead:" .set_desired_format(VkSurfaceFormatKHR{ .format = VK_FORMAT_B8G8R8A8_UNORM, .colorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR })" 15) Shouldn't the container for sizes in init_descriptors be an std::array since we know it's size at compile time? I know I know, there are three elements and it's run once, but still. Good practices! 16) Added " _frames[i].cameraBuffer = create_buffer(sizeof(GPUSceneData), VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VMA_MEMORY_USAGE_CPU_TO_GPU);" in the init_descriptors 17) "void GLTFMetallic_Roughness::clear_resources(VkDevice device)" is still empty. 18) Added code for transparency in MeshNode::Draw 19) Removed init_triangle_pipeline and init_mesh_pipeline 20) After everything I've done, I'm getting validation errors about me destroying destroyed pipelines :/

Yeah, I know nothing too fancy, but at least I'm done :D

[MasterDrake]

I figure out why I coulnd't load the textures

It was missing this and apparently my vk_loader code is broken but I still missed the bug while file-diffing :( "#define STB_IMAGE_IMPLEMENTATION

include "stb_image.h""

in "vk_images.cpp" I hate these headers implementation

Finally!

[senorsenorjunior]

Completely stuck at the Textures lesson, as soon as I set the _singleImageDescriptorSet to the _meshPipelineLayout.pSetLayouts field, the validation layer vomits this error and nothing renders

Validation Error: [ VUID-vkCmdDrawIndexed-None-02697 ] Object 0: handle = 0x3fbcd60000000028, type = VK_OBJECT_TYPE_PIPELINE; Object 1: VK_NULL_HANDLE, type = VK_OBJECT_TYPE_PIPELINE_LAYOUT; | MessageID = 0x9888fef3 | vkCmdDrawIndexed(): The VkPipeline 0x3fbcd60000000028[] (created with VkPipelineLayout 0xb097c90000000027[]) statically uses descriptor set (index #0) which is not compatible with the currently bound descriptor set's pipeline layout (VkPipelineLayout 0x0[]) The Vulkan spec states: For each set n that is statically used by the VkPipeline bound to the pipeline bind point used by this command, a descriptor set must have been bound to n at the same pipeline bind point, with a VkPipelineLayout that is compatible for set n, with the VkPipelineLayout used to create the current VkPipeline, as described in Pipeline Layout Compatibility (https://vulkan.lunarg.com/doc/view/1.3.239.0/windows/1.3-extensions/vkspec.html#VUID-vkCmdDrawIndexed-None-02697)

[vblanco20-1]

@mattwhitson The texture chapter had an error on the code snippets where the init_mesh_pipeline() is updated with the new descriptor set layout and new shaders. Check that chapter again as i just fixed it

[aer-i]

If someone else also have problems with buffer device address (in my case vertices have wrong positions) try using alignas() for the vertex struct

[vblanco20-1]

@aer-i Did you modify the vertex struct? With the exact code shown in the guide, it should align exactly to what the GPU wants in the shader with std430 rules. If you didnt, what gpu did you use?

[aer-i]

@vblanco20-1 Sorry, I should have mentioned that I have different vertex struct. Everything is fine, with alignment in the tutorial. I just tried moving to device address vertex buffer from normal vertex buffer, but then vertices broke in my game engine. I was a bit confused so I wanted to inform people with similar problem.

[aer-i]

• Vulkan Shaders - Code

"void VulkanEngine::init_descriptors() { // other code

^code init_desc_2 chapter-2/vk_engine.cpp }"

[vblanco20-1]

Error on init_descriptors() fixed

[YanTree]

follow "2 Drawing with Compute / Setting up IMGUI" of Vkguide2, the visual studio catch one error below, any advice can fix it? VulkanEngine initialized Assertion failed: GImGui != 0 && "No current context. Did you call ImGui::CreateContext() and ImGui::SetCurrentContext() ?", file D:\vulkan-guide\third_party\imgui\imgui.cpp, line 4239

D:\vulkan-guide\bin\Debug\engine.exe (process 40164) exited with code 3.

[vblanco20-1]

@YanTree make sure you call init_imgui() from the main init() function of the engine, you likely missed it.

[YanTree]

@vblanco20-1 oh! Right here, i missed call it, thx your advice.

[C1oud555]

For a Mac user, Vulkan is supported by the MoltenVK, which only support api_version 1.2;

Although the required extension dynamic rendering and synchronization is supported, but as KHR extensions.

In order to do that, use latest Volk is not that hard

1. use cmake to load volk

   add_subdirectory(volk) # complete repo of Volk, not existing one under third_party

2. define corresponding features so Volk will load related functions

   
   #define VK_KHR_synchronization2
   #define VK_NO_PROTOTYPES

include

3. add KHR suffix in all synchronization2 functions, structures.... eg:

VkImageMemoryBarrier2 imageBarrier{};

-------> VkImageMemoryBarrier2KHR imageBarrier{};

Emmmmm, could be a advice or a note for Mac users.

[wesleywong1995]

I followed the tutorial in the IMGUI chapter and found an error indicating that the setLayout was not closed after closing the program

[mrriddleman]

Found a few issues so far in Section 2:

1. We don't seem to be destroying the descriptor pool. I added these lines at the end of init_descriptors():

_mainDeletionQueue.push_function([&]() { vkDestroyDescriptorSetLayout(_device, _drawImageDescriptorLayout, nullptr); globalDescriptorAllocator.destroy_pool(_device); });

2. I'm not sure we're destroying the effect pipelines correctly in the PushConstants sections. I was receiving validation errors about destroying the pipelines.

I got it to not complain with this code at the end of init_background_pipelines():

_mainDeletionQueue.push_function([&]() { vkDestroyPipelineLayout(_device, _gradientPipelineLayout, nullptr);

    for (auto&& effect : backgroundEffects)
    {
        vkDestroyPipeline(_device, effect.pipeline, nullptr);
    }

});

3. For the PushConstants lecture - There's no changes to the sky.comp shader if we change the push constant data in the UI. This is obviously because push constants weren't added to the sky.comp shader.

I modified the sky.comp compute shader by adding the constants:

version 450

layout (local_size_x = 16, local_size_y = 16) in; layout(rgba8,set = 0, binding = 0) uniform image2D image;

//push constants block layout( push_constant ) uniform constants { vec4 data1; vec4 data2; vec4 data3; vec4 data4; } PushConstants;

Then I changed the mainImage a little to make the sky change color + change the star density

void mainImage( out vec4 fragColor, in vec2 fragCoord ) { vec4 pushedData = PushConstants.data1; //new

vec2 iResolution = imageSize(image);
// Sky Background Color
vec3 vColor = pushedData.xyz * fragCoord.y / iResolution.y; //use the push constants

// Note: Choose fThreshhold in the range [0.99, 0.9999].
// Higher values (i.e., closer to one) yield a sparser starfield.
float StarFieldThreshhold = pushedData.w; //use the push constants

// Stars with a slow crawl.
float xRate = 0.2;
float yRate = -0.06;
vec2 vSamplePos = fragCoord.xy + vec2( xRate * float( 1 ), yRate * float( 1 ) );
float StarVal = StableStarField( vSamplePos, StarFieldThreshhold );
vColor += vec3( StarVal );

fragColor = vec4(vColor, 1.0);

}

Enjoying the tutorial so far!

[OptimisticMonkey]

To avoid crash when closing app (after Mesh Buffers section):

     destroy_buffer(rectangle.indexBuffer);
     destroy_buffer(rectangle.vertexBuffer);
     });```

[OptimisticMonkey]

To avoid crash when closing app (after Mesh Buffers section):

_mainDeletionQueue.push_function([&]() {
     destroy_buffer(rectangle.indexBuffer);
     destroy_buffer(rectangle.vertexBuffer);
     });```

[OptimisticMonkey]

Likewise, need to add delete stuff after Mesh Loading section:

_mainDeletionQueue.push_function([&]() {
    for( auto&& meshAsset : testMeshes)
        {
        destroy_buffer(meshAsset->meshBuffers.indexBuffer);
        destroy_buffer(meshAsset->meshBuffers.vertexBuffer);
        }
    });

[OptimisticMonkey]

One more minor cleanup tip - when exiting there is a high chance the current in flight frame will not have the _deletionQueue called. Added to cleanup():

//Make sure to flush all possible in flight frames
for (FrameData framedata : _frames)
{
    framedata._deletionQueue.flush();
}

[hartmutbehrens]

If you're trying to follow along on macOS then you've likely run into linker issues. Inserting this after the filrst call to target_link_libraries(vkbootstrap,...) helped me:

target_link_libraries(vkbootstrap PUBLIC "iconv -framework AudioToolbox -framework Carbon -framework Cocoa -framework CoreFoundation -framework CoreGraphics -framework CoreHaptics -framework CoreAudio -framework CoreVideo -framework ForceFeedback -framework GameController -framework IOKit -framework Metal")

[SNAKFRIEN]

If I understand correctly, some vma memory usage flags, such as VMA_MEMORY_USAGE_GPU_ONLY, are deprecated. This link contains more info: https://gpuopen-librariesandsdks.github.io/VulkanMemoryAllocator/html/deprecated.html

[Fletterio]

I just finished chapter 2

• Some math for the shaders wasn't adding up and I wanted to check the thing from RenderDoc. However, launching the executable from RenderDoc crashes on startup due to an assert from vkboostrap

I don't know what causes it or if it is reproducible for anyone else. I noticed this in my own version of the code (starting from the starting point project an adding stuff as I go) and then I built the chapter-2 code in the all-chapters project and the same thing happens

If anyone has the time can you check if you get the same issue from RenderDoc? I've spent the whole day trying to debug it and didn't get anywhere