MoltenVK Enhancement Roadmap

[billhollings]

This list provides a place to consolidate the priorities and status of some of the larger MoltenVK enhancements whose development is either underway, or will be soon. This list will also help to prevent multiple MoltenVK devs unknowingly working independently on significant enhancements.

If you are, or soon plan to be, working on a significant MoltenVK enhancement, please comment below, and we'll add it to the list (not every enhancement needs to go on this list, if you are banging out a small enhancement). If you want to take over one of the un-started enhancements, let us know below and it can be assigned to you.

If you are waiting on a significant enhancement, and want to request that it be added to the list, please request that in the issue for that enhancement. This list is not meant to be a broad wish list, but instead it represents an upcoming priorities list.

Please note that we are tracking Vulkan 1.3 requirements separately in issue #1930. An item from there can be added here when a developer takes it upon themself to plan to work on it, and want's to let us know here.

The list is in order of attack. This list does not include bug fixing work.

	In Dev	Merged	Developer	Enhancement	Deps	Issue#	PR#
1	✅		@spnda	VK_KHR_maintenance4			#2116
2			@billhollings	Install Validation Layer testing
3			@billhollings @aitor-lunarg	Remaining CTS fails for Vulkan 1.0
4	✅	✅	@billhollings	Metal 3 Arg Buffs (enhancing bindless resources)			#2260
5				VK_EXT_descriptor_buffer VK_EXT_mutable_descriptor_type	#(4)	#1776 #2158
6				GL_EXT_debug_printf		#2252
7	✅		@gpx1000	VK_EXT_transform_feedback			#1943
8	✅		@AntarticCoder	Vulkan Acceleration Structures		#1956	#1967
9	✅		@billhollings	Faster pipeline caching using MTLBinaryArchives		#1765
10	✅			Vulkan 1.3		#1930
11				(macOS 14+ only) Integrate Apple's Shader Converter tech (including Geom & Tess shaders)	#(4)

[QuantumDeveloper]

I think would be nice to include here VK_KHR_maintenance4 extension if that possible

[AntarticCoder]

If anyone's interested in implementing this, the spec is here. VK_KHR_maintenance4 is also considered core in Vulkan 1.3, so implementing it is almost a no brainer.

[billhollings]

VK_KHR_maintenance4 is also considered core in Vulkan 1.3

Yes. Vulkan 1.3 requirements are being tracked separately in issue #1930. I've clarified that above, along with how this list interacts with that list. See for example, VK_KHR_synchronization2 appears in both lists, because I have plans to work on it above.

[oscarbg]

Hi,
Seems geo and tf are part of the present crossover 23 beta1 release, which says,
“We’re pleased to announce that this release also includes initial support for geometry shaders and transform feedback on macOS”..

I’m pretty curious to see how this geo shader implementation compares to “Apple's Shader Converter tech” both performance wise and compatibility wise once the two ara available on MoltenVK.. at least on MacOS 14+..
Exciting times..

[AndreaCicca]

Hi, Seems geo and tf are part of the present crossover 23 beta1 release, which says, “We’re pleased to announce that this release also includes initial support for geometry shaders and transform feedback on macOS”..

I’m pretty curious to see how this geo shader implementation compares to “Apple's Shader Converter tech” both performance wise and compatibility wise once the two ara available on MoltenVK.. at least on MacOS 14+.. Exciting times..

@js6i has a PR (#1815) where he said that "Currently we're focused on preparing the next CrossOver release, there will be updates after it's out."

I don't know what will be the plan for that PR if the project want to use Apple's Shader Converter for Geometry shader

[Zingam]

VK_EXT_shader_object looks like an extension that should be included sooner rather than later.
https://www.khronos.org/blog/you-can-use-vulkan-without-pipelines-today

[K0bin]

Metal 3 Arg Buffs (enhancing bindless resources)

It would be very nice if MoltenVK could also properly support "aliasing descriptors" with the reworked binding code.

The shader compiler really hates this pattern right now:

layout(set = 0, binding = 0) uniform sampler2D s0_2d;
layout(set = 0, binding = 0) uniform sampler3D s0_3d;