Add a new sample - subgroups operations by Patryk-Jastrzebski-Mobica · Pull Request #715 · KhronosGroup/Vulkan-Samples

Patryk-Jastrzebski-Mobica · 2023-06-12T08:42:44Z

Description

Please include a summary of the change, new sample or fixed issue. Please also include relevant motivation and context.
Please read the contribution guidelines

Fixes #

General Checklist:

Please ensure the following points are checked:

Sample Checklist

If your PR contains a new or modified sample, these further checks must be carried out in addition to the General Checklist:

I have tested the sample on at least one compliant Vulkan implementation
If the sample is vendor-specific, I have tagged it appropriately
I have stated on what implementation the sample has been tested so that others can test on different implementations and platforms
Any dependent assets have been merged and published in downstream modules
For new samples, I have added a paragraph with a summary to the appropriate chapter in the samples readme

Szymon-Rajczakowski · 2023-06-12T20:52:03Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	VK_CHECK(vkAllocateDescriptorSets(get_device().get_handle(), &alloc_info, &compute.descriptor_set));
+}
+
+void SubgroupsOperations::preapre_compute_pipeline_layout()


typo. Should be "prepare_compute_pipeline_layout"

Szymon-Rajczakowski · 2023-06-12T21:18:52Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	            VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+	            1u,
+	            &image_descriptor)};
+	vkUpdateDescriptorSets(get_device().get_handle(), static_cast<uint32_t>(write_descriptor_sets.size()), write_descriptor_sets.data(), 0u, NULL);


How about using nullptr instead of NULL? The last argument of vkUpdateDescriptorSets is a pointer to an array of VkCopyDescriptorSet so nullptr better suggest that is the pointer

out to date

Szymon-Rajczakowski · 2023-06-12T21:34:15Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+		VkViewport viewport = vkb::initializers::viewport(static_cast<float>(width), static_cast<float>(height), 0.0f, 1.0f);
+		vkCmdSetViewport(draw_cmd_buffers[i], 0u, 1u, &viewport);
+
+		VkRect2D scissor = vkb::initializers::rect2D(width, height, 0, 0);


Perhaps it can be explicitly cast by static_cast<int32_t>(width) and static_cast<int32_t>(height). This is not necessary, just a small suggestion

Szymon-Rajczakowski · 2023-06-12T21:41:16Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+
+		// draw texture
+		{
+			vkCmdBindDescriptorSets(draw_cmd_buffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, texture_object.pipeline.pipeline_layout, 0, 1, &texture_object.descriptor_set, 0, nullptr);


For code style consistency I suggest to add "u" to numbers
vkCmdBindDescriptorSets(draw_cmd_buffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, texture_object.pipeline.pipeline_layout, 0u, 1u, &texture_object.descriptor_set, 0u, nullptr);

out to date

Szymon-Rajczakowski · 2023-06-12T21:44:08Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+			vkCmdBindPipeline(draw_cmd_buffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, texture_object.pipeline.pipeline);
+
+			VkDeviceSize offset[] = {0};
+			vkCmdBindVertexBuffers(draw_cmd_buffers[i], 0, 1, texture_object.buffers.vertex->get(), offset);


For code style consistency I suggest to add "u" to unsigned int arguments

out to date

Szymon-Rajczakowski · 2023-06-12T21:48:16Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+			vkCmdBindVertexBuffers(draw_cmd_buffers[i], 0, 1, texture_object.buffers.vertex->get(), offset);
+			vkCmdBindIndexBuffer(draw_cmd_buffers[i], texture_object.buffers.index->get_handle(), 0, VK_INDEX_TYPE_UINT32);
+
+			vkCmdDrawIndexed(draw_cmd_buffers[i], texture_object.buffers.index_count, 1u, 0u, 0u, 0u);


vertexOffset argument is int32_t so it should be
vkCmdDrawIndexed(draw_cmd_buffers[i], texture_object.buffers.index_count, 1u, 0u, 0, 0u);

out to date

Szymon-Rajczakowski · 2023-06-12T21:52:51Z

samples/extensions/subgroups_operations/subgroups_operations.h

+
+	struct GuiSettings
+	{
+		int32_t                         selected_filtr = 0;


typo. Should be selected_filter

out to date

Szymon-Rajczakowski · 2023-06-12T21:53:12Z

samples/extensions/subgroups_operations/subgroups_operations.h

+		int32_t                         selected_filtr = 0;
+		static std::vector<std::string> init_filters_name()
+		{
+			std::vector<std::string> filtrs = {


typo. Should be std::vectorstd::string filters

out to date

asuessenbach · 2023-06-29T12:13:18Z

That's a nice little sample!
Is it still work in progress, as the title says?
Would it be easy to extend that sample by adding a non-subgroups path for comparison?

Patryk-Jastrzebski-Mobica · 2023-06-30T06:01:20Z

Hi @asuessenbach,
Yes, I'm still working on this sample. I'm currently implementing the FFT algorithm. Next week, I plan to push another commit/change with the vertex animation ready.
And as for the last question, I cannot answer you yet.

SaschaWillems · 2023-08-03T16:00:14Z

Any update on this? Having a subgroups sample would be very much appreciated :)

SaschaWillems · 2023-11-19T18:19:26Z

Any updates on this sample? Having a subgroup sample would be a great addition :)

Patryk-Jastrzebski-Mobica · 2023-11-20T09:13:38Z

Hi @SaschaWillems, I apologise for not updating for such a long time. Unfortunately the implementation of the sample is not yet finished. I still have some work left to do:

implement BRDF reflections,
fixing problems with pipeline barriers - on the Intel Iris Xe the sample runs stable, but on the RTX 2060 it crash after a few seconds,
adding a description to the README file.

SaschaWillems · 2023-11-20T15:44:35Z

No need to apologize. Your work is very much appreciated and I'm happy to hear that this sample is coming along :)

SaschaWillems · 2024-02-18T15:31:57Z

Any update on this?

CLAassistant · 2024-03-20T08:33:59Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
0 out of 5 committers have signed the CLA.

❌ Patryk-Jastrzebski-Mobica
❌ Krzysztof-Dmitruk-Mobica
❌ kdmitruk
❌ Seweryn-Zielas-Mobica
❌ Piotr-Plebanski-Mobica
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

marty-johnson59 · 2024-07-03T14:05:01Z

Hi @Patryk-Jastrzebski-Mobica, we noticed there's an issue with the CLA on this one - looks like the test got confused with the login alias (kdmitruk). Can you look into this? Not quite sure what the fix is - perhaps sign the CLA again using the alias? Thanks

Patryk-Jastrzebski-Mobica · 2024-07-04T06:11:49Z

Hi @marty-johnson59, I will to do something about it this week. I hope to find some time in the near future to finish the sample :).

marty-johnson59 · 2024-10-09T16:00:25Z

Hi @Patryk-Jastrzebski-Mobica, quick reminder on this. We briefly discussed this on the Samples call this week and there's definitely interest in seeing this sample released. Let us know if/how we can help. Thanks!

Patryk-Jastrzebski-Mobica · 2024-10-10T06:09:00Z

Hi @marty-johnson59, I am very sorry for such a long delay. I got back to work on this example this week. For now, I am pushing the fixes onto my own branch https://github.com/Mobica/Vulkan-Samples/tree/subgroups_operations_WIP3 . As soon as I have fixed the reflections on the model, I will immediately start to rebase this branch and deliver the code.

marty-johnson59 · 2024-10-10T14:57:32Z

Awesome, thank you! (and NP on the delay - just following up and letting you know there is still strong interest in this sample :) )

…en fixed

… remove useless shaders

…ilde_h0 shader and removed commented lines

Patryk-Jastrzebski-Mobica · 2024-10-16T12:09:00Z

Hi,
I have added the option to disable/enable subgroups. After adding shaders without subgroups enabled something like this started to happen to the mesh? @marty-johnson59 or @SaschaWillems do you have any idea where this problem is coming from? I'm not able to figure out myself if this is a problem with SPIR-V, the graphics driver or my implementation.
I need help with this issue.

asuessenbach · 2024-10-23T16:33:47Z

I don't know why, but this little patch makes the subgroups enabled case look correct again.
0001-Make-subgroups-enabled-work.patch

cforfang · 2024-10-23T16:49:39Z

Hey, I wanted to check -- does this sample work for all possible legal subgroup sizes? I didn't notice any checks for the exposed subgroupSize, nor any logic to handle different ones. It wasn't clear to me how/if the algorithm is such that it 'just works' whatever size is exposed or not.

EDIT:
Semi-related, it would probably be good if the readme explained how subgroups are used by the sample, and why this is useful/good compared to alternative approaches.

SaschaWillems · 2024-10-23T16:55:29Z

@cforfang : The sample is still work-in-progress

cforfang · 2024-10-24T13:26:29Z

@cforfang : The sample is still work-in-progress

Understood, just to highlight already now in case incompatibility might come up as a problem later for example :)

marty-johnson59 · 2024-11-19T19:36:24Z

Hi @Patryk-Jastrzebski-Mobica, just checking status on this one. Is it still in flight? Any help you need? Thanks

marty-johnson59 · 2025-04-24T18:34:23Z

Re-assigning to Steve for now.

gpx1000 · 2025-12-31T11:47:32Z

HI @Patryk-Jastrzebski-Mobica, @Krzysztof-Dmitruk-Mobica, @Piotr-Plebanski-Mobica, @Seweryn-Zielas-Mobica, @kdmitruk The CLA still hasn't been signed. Could I ask you guys to take a look and ensure that is handled?

I have updated this PR locally and might have a few fixes for the bugs observed above. I'll push what I have once I have it completely working.

gpx1000 · 2025-12-31T12:50:14Z

Mobica#9 <-- I think I've fixed everything in this PR including merging with master. All pipelines work and there are no more VVL errors or warnings. Please give it a look Mobica guys when you get a chance and let's hopefully get this excellent PR merged.

asuessenbach · 2026-01-05T15:15:10Z

samples/CMakeLists.txt

+	    "hpp_texture_compression_comparison"
+
+        #General Samples
+        "mobile_nerf")


Seems, you've removed a couple of samples?

Mobica#9 <-- hopefully they are not removed after the merge.

asuessenbach · 2026-01-05T15:16:49Z

samples/extensions/subgroups_operations/README.adoc

+* `#extension GL_KHR_shader_subgroup_quad`
+
+
+This sample focuse on `GL_KHR_shader_subgroup_basic` extension.


Typo: focuse -> focuses

asuessenbach · 2026-01-05T15:17:56Z

samples/extensions/subgroups_operations/README.adoc

+
+In order to use subgroups operations, the required extensions must be enabled, an instance of the Vulkan API must be created with a minimum of version 1.1 and SPIR-V 1.4 must be used.
+
+VkDevice must be created with this `VK_EXT_subgroup_size_control` extension.


Typo this -> the?

asuessenbach · 2026-01-05T15:22:20Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	prepare_uniform_buffers();
+	prepare_compute();
+
+	// prepare grpahics pipeline


Typo: grpahics -> graphics

asuessenbach

I did not try to get all the math used here... maybe worth to extend the README.adoc a bit?

asuessenbach · 2026-01-05T16:38:06Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	auto  rndVal = []() -> float {
+        std::random_device                    rndDevice;
+        std::mt19937                          mt(rndDevice());
+        std::uniform_real_distribution<float> dis(0.0f, 1.0f);


You could make those three variables static, to prevent their construction on every random number generation.

asuessenbach · 2026-01-05T16:41:29Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	do
+	{
+		x1 = 2.0f * rndVal() - 1.0f;
+		x2 = 2.0f * rndVal() - 1.0f;


Instead of re-mapping a random value from [0.0, 1.0] to [-1.0, 1.0], you could just generate a value in that range by using
std::uniform_real_distribution<float> dis(-1.0f, 1.0f);

asuessenbach · 2026-01-05T16:43:45Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+
+void SubgroupsOperations::generate_plane()
+{
+	uint32_t              dim_gird     = grid_size;


Typo: dim_gird -> dim_grid?

And... why are you introducing that variable at all? You could just directly use grid_size.

asuessenbach · 2026-01-05T16:56:18Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+void SubgroupsOperations::generate_plane()
+{
+	uint32_t              dim_gird     = grid_size;
+	uint32_t              vertex_count = dim_gird + 1u;


I think, the name vertex_count might be a bit misleading. This is not the number of vertices (in total), but the number of vertices per row (or per column).

asuessenbach · 2026-01-05T17:01:07Z

samples/extensions/subgroups_operations/subgroups_operations.h

+	struct GridBuffers
+	{
+		std::unique_ptr<vkb::core::BufferC> vertex      = {VK_NULL_HANDLE};
+		std::unique_ptr<vkb::core::BufferC> index       = {VK_NULL_HANDLE};


Wouldn't vertices and indices be better names here?

asuessenbach · 2026-01-05T17:03:58Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	    vkb::initializers::descriptor_pool_size(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 20u),
+	    vkb::initializers::descriptor_pool_size(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 20u)};
+	VkDescriptorPoolCreateInfo descriptor_pool_create_info =
+	    vkb::initializers::descriptor_pool_create_info(static_cast<uint32_t>(pool_sizes.size()), pool_sizes.data(), 15u);


Some magic values 20u and 15u here? Where do they come from?

asuessenbach · 2026-01-05T17:32:44Z

samples/extensions/subgroups_operations/subgroups_operations.cpp

+	vkUpdateDescriptorSets(get_device().get_handle(), static_cast<uint32_t>(write_descriptor_sets.size()), write_descriptor_sets.data(), 0u, nullptr);
+}
+
+VkDescriptorImageInfo SubgroupsOperations::create_ia_descriptor(ImageAttachment &attachment)


What does ia in create_ia_descriptor stand for?

asuessenbach · 2026-01-05T17:48:10Z

When I start this sample, I get this VVL warning:
[warning] -140549851 - WARNING-CoreValidation-AllocateDescriptorSets-WrongType: vkAllocateDescriptorSets(): pAllocateInfo->pSetLayouts[0] binding 4 was created with VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER but VkDescriptorPool 0x3b000000003b was not created with any VkDescriptorPoolSize::type with VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (Some implementations will not return VK_ERROR_OUT_OF_POOL_MEMORY as they should with VK_KHR_maintenance1).

gpx1000 · 2026-01-05T18:04:38Z

When I start this sample, I get this VVL warning: [warning] -140549851 - WARNING-CoreValidation-AllocateDescriptorSets-WrongType: vkAllocateDescriptorSets(): pAllocateInfo->pSetLayouts[0] binding 4 was created with VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER but VkDescriptorPool 0x3b000000003b was not created with any VkDescriptorPoolSize::type with VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER (Some implementations will not return VK_ERROR_OUT_OF_POOL_MEMORY as they should with VK_KHR_maintenance1).

Mobica#9 <-- VVL error is fixed here.

Patryk-Jastrzebski-Mobica changed the title ~~WIP: Add a new sample - subgroups operation~~ WIP: Add a new sample - subgroups operations Jun 12, 2023

Szymon-Rajczakowski reviewed Jun 12, 2023

View reviewed changes

Patryk-Jastrzebski-Mobica force-pushed the subgroups_operations branch from 6a2eef8 to a67ff3e Compare November 20, 2023 09:07

Patryk-Jastrzebski-Mobica and others added 14 commits October 11, 2024 13:04

Add a new sample - subgroups operation

e275c44

Add ocean fft

00fe99c

Fix grid generation algorithm

fe0326f

Add compute pipeline and compute queue

0cd32b8

Add ability to resize the grid in the UI and a few small changes

44aa375

Animated grid using the sine and cosine functions

3ea91d5

Add a new compute pipeline

fc9009f

Fixed a descriptor set, the function generating random numbers has be…

8026831

…en fixed

Transfer of philips spectrum calculation from GPU to CPU.

b62b93f

Refactoring and changes in graphic pipeline

db0b8b2

Add the FBAttachment type

b3dce47

Add butterfly precomp shader and texture

063784f

Add tilde textures and pipeline

b2b74ac

Refactor and add butterfly and tild texture generation

d5f02b8

Patryk-Jastrzebski-Mobica added 11 commits October 14, 2024 14:16

Add invert fft and clean-up

7501909

Some fixes, add tessellation

64f366d

add skybox

b7b7798

Convert a README.md file to a README.adoc file

ec7b26c

Add pipeline barriers, update README file

204bfb4

Add a new sample - subgroups operation

c0065a5

[WIP3] subgroups -Rebase and compilation fix

4d94890

Add a new sample - subgroups operation

213059b

A rebase fix

5fa5527

[WIP] reflections experiments

acbbaac

Add a new sample - subgroups operation

f1936a7

Patryk-Jastrzebski-Mobica force-pushed the subgroups_operations branch from a67ff3e to f1936a7 Compare October 14, 2024 14:18

Patryk-Jastrzebski-Mobica added 3 commits October 15, 2024 11:22

[WIP] Remove skybox to make sample simpler. Update copyright data and…

dae01a0

… remove useless shaders

[WIP] Add missing description, enable subgroup basic feature in fft_t…

8340fc9

…ilde_h0 shader and removed commented lines

Update README and add option to disable/enable subgroups

1f26700

Patryk-Jastrzebski-Mobica changed the title ~~WIP: Add a new sample - subgroups operations~~ Add a new sample - subgroups operations Oct 16, 2024

marty-johnson59 assigned gpx1000 Apr 24, 2025

asuessenbach reviewed Jan 5, 2026

View reviewed changes

		* `#extension GL_KHR_shader_subgroup_quad`


		This sample focuse on `GL_KHR_shader_subgroup_basic` extension.


		In order to use subgroups operations, the required extensions must be enabled, an instance of the Vulkan API must be created with a minimum of version 1.1 and SPIR-V 1.4 must be used.

		VkDevice must be created with this `VK_EXT_subgroup_size_control` extension.

Conversation

Patryk-Jastrzebski-Mobica commented Jun 12, 2023

Description

General Checklist:

Sample Checklist

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

asuessenbach commented Jun 29, 2023

Uh oh!

Patryk-Jastrzebski-Mobica commented Jun 30, 2023

Uh oh!

SaschaWillems commented Aug 3, 2023

Uh oh!

SaschaWillems commented Nov 19, 2023

Uh oh!

Patryk-Jastrzebski-Mobica commented Nov 20, 2023

Uh oh!

SaschaWillems commented Nov 20, 2023

Uh oh!

SaschaWillems commented Feb 18, 2024

Uh oh!

CLAassistant commented Mar 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

marty-johnson59 commented Jul 3, 2024

Uh oh!

Patryk-Jastrzebski-Mobica commented Jul 4, 2024

Uh oh!

marty-johnson59 commented Oct 9, 2024

Uh oh!

Patryk-Jastrzebski-Mobica commented Oct 10, 2024

Uh oh!

marty-johnson59 commented Oct 10, 2024

Uh oh!

Patryk-Jastrzebski-Mobica commented Oct 16, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

asuessenbach commented Oct 23, 2024

Uh oh!

cforfang commented Oct 23, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SaschaWillems commented Oct 23, 2024

Uh oh!

cforfang commented Oct 24, 2024

Uh oh!

marty-johnson59 commented Nov 19, 2024

Uh oh!

marty-johnson59 commented Apr 24, 2025

Uh oh!

gpx1000 commented Dec 31, 2025

Uh oh!

gpx1000 commented Dec 31, 2025

CLAassistant commented Mar 20, 2024 •

edited

Loading

Patryk-Jastrzebski-Mobica commented Oct 16, 2024 •

edited

Loading

cforfang commented Oct 23, 2024 •

edited

Loading