Where communities thrive

Join over 1.5M+ people
Join over 100K+ communities
Free without limits
Create your own community

Explore more communities

MaikKlein/ash

People

Repo info

Activity

Chehui Chou

@deadshot465_gitlab

If there's anyone, does someone know how to use cmd_push_constants?

It expects a &[u8] but I don't know how to convert my own struct to an &[u8]

matrixbot

@matrixbot

Ralith try the bytemuck crate

Chehui Chou

@deadshot465_gitlab

So should I implement Pod trait for my own struct?

matrixbot

@matrixbot

Ralith you should do whatever the crate's docs tell you to do

Chehui Chou

@deadshot465_gitlab

thanks

Chehui Chou

@deadshot465_gitlab

Does anyone have no problems in debug build but have problems in release build?

I haven't finished my program but I tried to use cargo build --release to make a release build, and suddenly there are tons of error messages about RenderPass and GraphicsPipelines

But I didn't change the code and there are no error messages in debug build

There are errors like

[2020-09-20T17:52:55Z ERROR demo_game_rs::game::graphics::vk::graphics] Validation Error: [ VUID-VkRenderPassCreateInfo-attachment-00834 ] Object 0: handle = 0x16629e6ed68, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0xfca478c2 | Resolve: vkCreateRenderPass() attachment 38 must be less than the total number of attachments 3. The Vulkan spec states: If the attachment member of any element of pInputAttachments, pColorAttachments, pResolveAttachments or pDepthStencilAttachment, or any element of pPreserveAttachments in any element of pSubpasses is not VK_ATTACHMENT_UNUSED, it must be less than attachmentCount (https://vulkan.lunarg.com/doc/view/1.2.148.0/windows/1.2-extensions/vkspec.html#VUID-VkRenderPassCreateInfo-attachment-00834)

But I never set the number of attachment to 38 or whatsoever

matrixbot

@matrixbot

Ralith sounds like you're invoking UB

Ralith a typical cause would be passing references to temporaries which no longer exist

Chehui Chou

@deadshot465_gitlab

Yeah I was explicitly calling build() on create infos and they worked fine in debug builds, but didn't work in release builds

After avoiding most explicit build() and making sure references exist until the structure is created everything works fine now

Thank you

matrixbot

@matrixbot

Ralith yep, that's the right pattern to follow

Chehui Chou

@deadshot465_gitlab

But why does it work in debug builds though

Shouldn't lifetime problems happen regardless of build mode?

matrixbot

@matrixbot

Ralith dereferencing a dangling raw pointer is undefined behavior

Ralith undefined behavior is, by definition, unpredictable

Ralith it can do different things depending on compile mode, target environment, phase of moon, etc

Chehui Chou

@deadshot465_gitlab

So it worked only because of...coincidence or the situation permits that?

matrixbot

@matrixbot

Ralith undefined behavior doing what you intended by coincidence is common, yeah; typically followed by changing when you least expect it

Mike Krasnenkov

@agmike

Hello. I'm trying to properly handle 1.2 features which reside in pNext chain, like VkPhysicalDeviceImagelessFramebufferFeatures. Is there a way to parse such chain returned by vulkan?

Or do I need to walk through it manually

I've seen internals such as ptr_chain_iter, but they are not exposed

matrixbot

@matrixbot

Ralith vulkan never creates pNext chains

Ralith it's up to you to construct them and pass them in

Ralith otherwise the memory management situation would be a nightmare

Mike Krasnenkov

@agmike

But isn't that how vkGetPhysicalDeviceProperties2 works?

I would imagine the pointers are all pointing to static data somewhere in the driver

it looks like vulkan-hpp provides the getters for them https://github.com/KhronosGroup/Vulkan-Hpp#structure-pointer-chains

Mike Krasnenkov

@agmike

Aaah, I see, I got it all backwards

So to get the properties and features I construct the chain and then vulkan fills it in.

Mike Krasnenkov

@agmike

Well then, sorry for the disturbance :)

matrixbot

@matrixbot

Ralith np

Chehui Chou

@deadshot465_gitlab

What's the correct way to use extension methods like cmd_set_checkpoint_nv?

Like for DebugUtils we can use ash::vk::extensions::DebugUtils, but it seems that I can't find similar thing for NVIDIA extensions, so right now I don't know how to use those extensions in ash...

Chehui Chou

@deadshot465_gitlab

ash::vk::NvDeviceDiagnosticCheckpointsFn::cmd_set_checkpoint_nv takes &self and I don't know where to get that &self

matrixbot

@matrixbot

Ralith functions that take &self are called methods; the &self comes from the value that you call them on

Chehui Chou

@deadshot465_gitlab

Yes, I mean I didn't know how to get an instance of NvDeviceDiagnosticCheckpointsFn

Thanks though

Chehui Chou

@deadshot465_gitlab

Has anyone tried to compile and run their Vulkan applications on macOS recently? After it compiles, I tried to run it on macOS (both in x86_64 and via Rosetta 2), but whenever it reaches vkGetPhysicalDeviceFeatures I will get a segmentation fault. Not sure if it's a probelm in VulkanSDK for macOS, or a problem in ash.

Chehui Chou

@deadshot465_gitlab

madmaxio

@madmaxio

There is metal backend there?

Not vulkan

Chehui Chou

@deadshot465_gitlab

It's MoltenVK

Chehui Chou

@deadshot465_gitlab

Nevermind I found the culprit

left_adjoint

@leftadjoint

Since an ash::Instance underlying value is a handle, is there some issue with adding an Eq implementation for it?