# Vulkan Confused with Vulkan subpass dependency

## Recommended Posts

I am looking at the SaschaWillems subpass example for getting some insight into subpass depdendencies but its hard to understand whats going on without any comments. Also there is not a lot of documentation on subpass dependencies overall.

Looking at the code, I can see that user specifies the src subpass, dst subpass and src state, dst state. But there is no mention of which resource the dependency is on. Is a subpass dependency like a pipeline barrier. If yes, how does it issue the barrier? Is the pipeline barrier issued on all attachments in the subpass with the input src and dst access flags? Any explanation will really clear a lot of doubts on subpass dependencies.

Thank you

Edited by mark_braga

##### Share on other sites

I recently wrote an abstraction for this mechanism so my graphics API would not be D3D12 specific.  Given that, I can only really describe this from the point of view of writing the code but since things seem to be working, I believe the details I figured out are pretty close to accurate.

First off, you need to look at the three related info structures again since they most certainly do tell you exactly which images are being referenced, it is just a bit indirect. Basically there is an array of all images used in the overall pass found in the render pass info structure, sub passes reference these images via 0 based indexing.

As to the behavior, at the start and end of each subpass the API issues an image transition barrier if needed to put the attachment in the requested format.  So, for instance, if you were doing a post processing blur, you might end up with the following chain of events:

NextSubPass
Transition attachment 0 to writable
NextSubPass
Transition attachment 1 to writable
.. Draw post processing quad to run vertical blur with input attachment 0 and output attachment 1
NextSubPass
Transition attachment 0 to writable
.. Draw post processing quad to run horizontal blur with input attachment 1 and output attachment 0

So the attachments involved are ping ponging from readable to writable as required for the post processing to occur.

Hopefully this makes sense and helps you out.  I had to look at those structures quite a few times till I figured out the details.  The structures themselves are pretty simple, it's just the relationships that are hard to see until you try and fail a couple times to get the correct behavior.

##### Share on other sites

Thanks for the explanation.

1 hour ago, Hiwas said:

NextSubPass
Transition attachment 0 to writable
NextSubPass
Transition attachment 1 to writable
.. Draw post processing quad to run vertical blur with input attachment 0 and output attachment 1

Here are you talking about the attachment in the subpass or the renderpass? (Is attachment0 relative to the pColorAttachments in the subpass or pAttachments in the renderpass)

##### Share on other sites

In the subpass descriptions you have arrays of VkAttachmentReference which is a uint and layout.  The uint is the 0 based index into the VkRenderPassCreateInfo structure's pAttachment array where you listed all of the attachments for the render pass.  So, effectively, what I'm saying with those is:

// assume you have pRenderPass and pSubPass pointers to the respective Vk structures.
theImageWeWantToMessWith = pRenderPass->pAttachments[ pSubPass->pInputAttachments.attachment ]

That is effectively what is going on behind the scenes to figure out which image to call memory barriers on.
So, when I said attachment 0 and 1, I was talking about the index into the VkRenderPassCreateInfo structure's pAttachments array.  Note that render pass info does not separate inputs/outputs etc, it just takes one big list, only subpasses care about usage.

Hope that clarifies things.

##### Share on other sites

So how is the image barrier issued. Is the logic something like this:

for (uint32_t i = 0; i < dependencyCount; ++i)
{
if (pDependencies[i].srcSubpass == currentSubpass)
{
for (uint32_t att = 0; att < pRenderPass->attachmentCount; ++att)
{
if (pRenderPass->pAttachments[att]->srcAccessFlag == pDepdendencies[i].srcAccessFlag)
{
// transition the attachment to pDependencies.dstAccess?
}
}
}
}

##### Share on other sites

In a general way, that is fairly close to a very simplistic solution.  Unfortunately at this level it is really all about how clever the drivers get when they solve the path through the dag generated by the subpasses.  They could do the very simplistic solution of just issuing a vkCmdPipelineBarrier with top and bottom of pipe flags set between subpasses with dependencies or they could look at the subpass attachments in detail and figure out a more refined approach.  Since this is all just a state transition chain, building a simple DAG allows for a much more optimized approach to issuing a mix of pipeline and memory barriers.

I can't find the article I remember that describes some of this but this one may be of interest: https://gpuopen.com/vulkan-barriers-explained/ as it is related.

## Create an account

Register a new account

• ### Similar Content

• I'm writing a small 3D Vulkan game engine using C++. I'm working in a team, and the other members really don't know almost anything about C++. About three years ago i found this new programming language called D wich seems very interesting, as it's very similar to C++. My idea was to implement core systems like rendering, math, serialization and so on using C++ and then wrapping all with a D framework, easier to use and less complicated. Is it worth it or I should stick only to C++ ? Does it have less performance compared to a pure c++ application ?

• Cannot get rid of z-fighting (severity varies between: no errors at all - ~40% fail).
* up-to-date validation layer has nothing to say.
* pipelines are nearly identical (differences: color attachments, descriptor sets for textures, depth write, depth compare op - LESS for prepass and EQUAL later).
* did not notice anything funny when comparing the draw commands via NSight either - except, see end of this post.
* "invariant gl_Position" for all participating vertex shaders makes no difference ('invariant' does not show up in decompile, but is present in SPIR-V).
* gl_Position calculations are identical for all (also using identical source data: push constants + vertex attribs)
However, when decompiling SPIR-V back to GLSL via NSight i noticed something rather strange:
Depth prepass has "gl_Position.z = 2.0 * gl_Position.z - gl_Position.w;" added to it. What is this!? "gl_Position.y = -gl_Position.y;", which is always added to everything, i can understand - vulcans NDC is vertically flipped by default in comparison to OpenGL. That is fine. What is the muckery with z there for? And why is it only selectively added?
Looking at my perspective projection code (the usual matrix multiplication, just simplified):
vec4 projection(vec3 v) { return vec4(v.xy * par.proj.xy, v.z * par.proj.z + par.proj.w, -v.z); }
All it ends up doing is doubling w-part of 'proj' in z (proj = vec4(1.0, 1.33.., -1.0, 0.2)). How does anything show at all given that i draw with compare op EQUAL. Decompile bug?
I am out of ideas.

• I have a rather specific question. I'm trying to learn about linked multi GPU in Vulkan 1.1; the only real source I can find (other than the spec itself) is the following video:

Anyway, each node in the linked configuration gets its own internal heap pointer. You can swizzle the node mask to your liking to make one node pull from another's memory. However, the only way to perform the "swizzling" is to rebind a new VkImage / VkBuffer instance to the same VkDeviceMemory handle (but with a different node configuration). This is effectively aliasing the memory between two instances with identical properties.
I'm curious whether this configuration requires special barriers. How do image barriers work in this case? Does a layout transition on one alias automatically affect the other. I'm coming from DX12 land where placed resources require custom aliasing barriers, and each placed resource has its own independent state. It seems like Vulkan functions a bit differently.
Thanks.

• bs::framework is a newly released, free and open-source C++ game development framework. It aims to provide a modern C++14 API & codebase, focus on high-end technologies comparable to commercial engine offerings and a highly optimized core capable of running demanding projects. Additionally it aims to offer a clean, simple architecture with lightweight implementations that allow the framework to be easily enhanced with new features and therefore be ready for future growth.
Some of the currently available features include a physically based renderer based on Vulkan, DirectX and OpenGL, unified shading language, systems for animation, audio, GUI, physics, scripting, heavily multi-threaded core, full API documentation + user manuals, support for Windows, Linux and macOS and more.
The next few updates are focusing on adding support for scripting languages like C#, Python and Lua, further enhancing the rendering fidelity and adding sub-systems for particle and terrain rendering.
A complete editor based on the framework is also in development, currently available in pre-alpha stage.

View full story

• bs::framework is a newly released, free and open-source C++ game development framework. It aims to provide a modern C++14 API & codebase, focus on high-end technologies comparable to commercial engine offerings and a highly optimized core capable of running demanding projects. Additionally it aims to offer a clean, simple architecture with lightweight implementations that allow the framework to be easily enhanced with new features and therefore be ready for future growth.
Some of the currently available features include a physically based renderer based on Vulkan, DirectX and OpenGL, unified shading language, systems for animation, audio, GUI, physics, scripting, heavily multi-threaded core, full API documentation + user manuals, support for Windows, Linux and macOS and more.
The next few updates are focusing on adding support for scripting languages like C#, Python and Lua, further enhancing the rendering fidelity and adding sub-systems for particle and terrain rendering.
A complete editor based on the framework is also in development, currently available in pre-alpha stage.

• Trying to figure out why input attachment reads as black with NSight VS plugin - and failing.
This is what i can see at the invocation point of the shader:
* attachment is filled with correct data (just a clear to bright red in previous renderpass) and used by the fragment shader:
// SPIR-V decompiled to GLSL #version 450 layout(binding = 0) uniform sampler2D accum; // originally: layout(input_attachment_index=0, set=0, binding=0) uniform subpassInput accum; layout(location = 0) out vec4 fbFinal; void main(){ fbFinal = vec4(texelFetch(accum, ivec2(gl_FragCoord.xy), 0).xyz + vec3(0.0, 0.0, 1.0), 1.0); // originally: fbFinal = vec4(subpassLoad(accum).rgb + vec3(0.0, 0.0, 1.0), 1.0); } * the resulting image is bright blue - instead of the expected bright purple (red+blue)
How can this happen?
'fbFinal' format is B8G8R8A8_UNORM and 'accum' format is R16G16B16A16_UNORM - ie. nothing weird.

• Hi, running Vulkan with the latest SDK, validation layers enabled I just got the following warning:
That is really strange, because in DX11 we can have 15 constant buffers per shader stage. And my device (Nvidia GTX 1050 is DX11 compatible of course) Did anyone else run into the same issue? How is it usually handled? I would prefer not enforcing less amount of CBs for the Vulkan device and be as closely compliant to DX11 as possible. Any idea what could be the reason behind this limitation?

• Hello everyone!
For my engine, I want to be able to automatically generate pipeline layouts based on shader resources. That works perfectly well in D3D12 as shader resources are not required to specify descriptor tables, so I use reflection system and map different shader registers to tables as I need. In Vulkan, however, looks like descriptor sets must be specified in both SPIRV bytecode and when creating pipeline layout (why is that?). So it looks like I will have to mess around with the bytecode to tweak bindings and descriptor sets. I looked at SPIRV-cross but it seems like it can't emit SPIRV (funny enough). I also use glslang to compile GLSL to SPIRV and for some reason, binding decoration is only present for these resources that I explicitly defined.
Does anybody know if there is a tool to change bindings in SPIRV bytecode?

• Hi, I am having problems with all of my compute shaders in Vulkan. They are not writing to resources, even though there are no problems in the debug layer, every descriptor seem correctly bound in the graphics debugger, and the shaders definitely take time to execute. I understand that this is probably a bug in my implementation which is a bit complex, trying to emulate a DX11 style rendering API, but maybe I'm missing something trivial in my logic here? Currently I am doing these:
Set descriptors, such as VK_DESCRIPTOR_TYPE_STORAGE_BUFFER for a read-write structured buffer (which is non formatted buffer) Bind descriptor table / validate correctness by debug layer Dispatch on graphics/compute queue, the same one that is feeding graphics rendering commands.  Insert memory barrier with both stagemasks as VK_PIPELINE_STAGE_ALL_COMMANDS_BIT and srcAccessMask VK_ACCESS_SHADER_WRITE_BIT to dstAccessMask VK_ACCESS_SHADER_READ_BIT Also insert buffer memory barrier just for the storage buffer I wanted to write Both my application behaves like the buffers are empty, and Nsight debugger also shows empty buffers (ssems like everything initialized to 0). Also, I tried the most trivial shader, writing value of 1 to the first element of uint buffer. Am I missing something trivial here? What could be an other way to debug this further?

• By khawk
LunarG has released new Vulkan SDKs for Windows, Linux, and macOS based on the 1.1.73 header. The new SDK includes:

View full story

• 35
• 12
• 10
• 9
• 9
• ### Forum Statistics

• Total Topics
631358
• Total Posts
2999531
×