Vulkan bindings

[burnt barn]

https://github.com/Snektron/vulkan-zig

[fiery folio]

Great Vulkan bindings! Got my seal of approval!

[full gust]

I am currently using these and they are pretty lit

[midnight wyvern]

self.vkb = vk.BaseWrapper.load(c.SDL_Vulkan_GetVkGetInstanceProcAddr);

Error when building:

.zig-cache/o/d55ae902be81ed03dcf52b64a96575e6/vk.zig:27022:33: error: expected 0 argument(s), found 2
                const cmd_ptr = loader(Instance.null_handle, field.name.ptr) orelse undefined;
                                ^~~~~~
/home/leeb/Projects/Nashi/.zig-cache/o/ffe70c69b6b234637678234c67df6247/cimport.zig:7476:12: note: function declared here
pub extern fn SDL_Vulkan_GetVkGetInstanceProcAddr() SDL_FunctionPointer;

Is there a fix for this?

[burnt barn]

Yeah you need to call the SDL function i think

[burnt barn]

https://github.com/Snektron/vulkan-zig/issues/171 see for more info

https://github.com/Snektron/vulkan-zig/issues/170 and this one is also related

self.vkb = vk.BaseWrapper.load(@as(vk.PfnGetInstanceProcAddr, @ptrCast(c.SDL_Vulkan_GetVkGetInstanceProcAddr)));

[midnight wyvern]

self.vkb = vk.BaseWrapper.load(@as(vk.PfnGetInstanceProcAddr, @ptrCast(c.SDL_Vulkan_GetVkGetInstanceProcAddr())));

->

thread 26701 panic: cast causes pointer to be null
/home/leeb/Projects/Nashi/src/renderer_vk.zig:28:66: 0x122c48c in prepare_vulkan (zig_sdl3_cross_template)
        vkb = vk.BaseWrapper.load(@as(vk.PfnGetInstanceProcAddr, @ptrCast(c.SDL_Vulkan_GetVkGetInstanceProcAddr())));

oh, nvm brain fart. It was just this:

self.vkb = vk.BaseWrapper.load(@as(vk.PfnGetInstanceProcAddr, @ptrCast(&c.SDL_Vulkan_GetVkGetInstanceProcAddr)));

Thx

[midnight wyvern]

But, still there's one issue when compiling to android:

.zig-cache/o/d55ae902be81ed03dcf52b64a96575e6/vk.zig:11:55: error: no field or member function named 'isARM' in 'Target.Cpu.Arch'
else if (builtin.abi == .android and (builtin.cpu.arch.isARM() or builtin.cpu.arch.isThumb()) and std.Target.arm.featureSetHas(builtin.cpu.features, .has_v7) and builtin.cpu.arch.ptrBitWidth() == 32)

[midnight wyvern]

I'm using Zig 0.14.1 via the zig-0.14-compat branch

[burnt barn]

Oh that seems like a bug

Can you submit an issue? I can't fix that rn

Or maybe a pr

[midnight wyvern]

alr

[burnt barn]

That function needs to be replaced with whatever the current one is

[midnight wyvern]

it's isArm();

[burnt barn]

Seems like a simple enough fix

[midnight wyvern]

ye

[burnt barn]

I can press the merge button from my phone but i can't edit

[midnight wyvern]

did

[burnt barn]

I see it

I think you made the pr from the master branch

[midnight wyvern]

nope

right?

[burnt barn]

I mean the commits in your branch are from master

So you should cherry pick your commit onto the 0.14 branch

But first you should retarget your pr to the master branch because the fix is needed there too

If that makes sense

[midnight wyvern]

Yeah, to be honest I'm quite new to a lot of git stuff. Just so you know. But I'll try

[burnt barn]

No problem, I'll help you through it

So first you should edit your pull request to target master instead of 0.14-compat

[midnight wyvern]

What I'm confused about is why only the main branch in my fork is showing

[burnt barn]

Hmm you may need explicitly check it out

Try git checkout zig-0.14-compat

[midnight wyvern]

That didnt work, it literally doesn't show up on github.

[burnt barn]

You may have checked some button to only clone master, i think there is one

In that case use git remote add upstream https://github.com/Snektron/vulkan-zig

Followed by git fetch --all

[midnight wyvern]

yeah, that did it

[midnight wyvern]

Made one for zig-0.14-compat too

[burnt barn]

Excellent, thanks!

[midnight wyvern]

No problem (and thanks for your incredible patience!

And that fixed the error, thanks!

[midnight wyvern]

Now I'm having an issue with android again:

fn create_instance(self: *renderer) !void {
    const allocator = std.heap.c_allocator;

    const app_info: vk.ApplicationInfo = .{
        .p_application_name = "Nashi",
        .application_version = @bitCast(vk.makeApiVersion(1, 0, 0, 0)),
        .p_engine_name = "Nashi",
        .engine_version = @bitCast(vk.makeApiVersion(0, 0, 1, 0)),
        .api_version = @bitCast(vk.API_VERSION_1_2),
    };    

    var extension_names = std.ArrayList([*:0]const u8).init(allocator);
    defer extension_names.deinit();

    try extension_names.append("VK_KHR_portability_enumeration");

    var sdl_exts_count: u32 = 0;
    const sdl_exts = c.SDL_Vulkan_GetInstanceExtensions(&sdl_exts_count);

    try extension_names.appendSlice(@ptrCast(sdl_exts[0..sdl_exts_count]));

    for (extension_names.items) |e| {
        std.log.info("Extension: {s}", .{ e });
    }

    const validation_layer_name = "VK_LAYER_KHRONOS_validation"; // Null-terminated C string
    const validation_layers = [_][*:0]const u8{
        validation_layer_name,
    };


    const create_info: vk.InstanceCreateInfo = .{
        .p_application_info = &app_info,
        .enabled_extension_count = @intCast(extension_names.items.len),
        .pp_enabled_extension_names = extension_names.items.ptr,
        .enabled_layer_count = @intCast(validation_layers.len),
        .pp_enabled_layer_names = validation_layers[0..],
        .flags = .{ .enumerate_portability_bit_khr = true },
    }; 

    const instance = try self.vkb.createInstance(&create_info, null);

    const vki = try allocator.create(vk.InstanceWrapper);
    errdefer allocator.destroy(vki);
    vki.* = vk.InstanceWrapper.load(instance, self.vkb.dispatch.vkGetInstanceProcAddr.?);
    self.vk_instance = vk.InstanceProxy.init(instance, vki);
    errdefer self.vk_instance.destroyInstance(null);
}

This is what I'm doing and it's crashing after 1 second or so...

It's working just fine for Dekstop

[burnt barn]

Can you try removing the validation layer?

If youre testing as cli program then that won't work

But i don't recall if that makes it crash

Otherwise i don't know

[midnight wyvern]

that didn't work, what I do for android in build.zig:

const vulkan = b.dependency("vulkan", .{
    .registry = b.dependency("vulkan_headers", .{}).path("registry/vk.xml"),
});
exe.root_module.addImport("vulkan", vulkan.module("vulkan-zig"));

[burnt barn]

I can't help you rn sorry

[midnight wyvern]

alr

[midnight wyvern]

Will there be a fix?

[burnt barn]

When i get home maybe

[midnight wyvern]

u on vacation?

[burnt barn]

Ye but going home today

[midnight wyvern]

Yo, have fun, dude!

[burnt barn]

I get the feeling that it's probably related to your own logic than to the library, though

[midnight wyvern]

maybe

[burnt barn]

did you check if portability is actually available?

I think it depends a bit on your phone, some android vendors are lacking a bit

i recommend installing the vulkan caps viewer app so that you can double check

if that's not the problem, then perhaps you can share the entire project with me so that i can build it from scratch

[midnight wyvern]

It's both not working on the emulator and my actual physical phone

[burnt barn]

you got a repo?

[midnight wyvern]

Yeah, sorry for such a delayed response. My repo: https://github.com/mozbeel/Nashi

[burnt barn]

How do I build for android?

[midnight wyvern]

zig build -Dandroid

you need zig 0.15 (nightly) though

[burnt barn]

that shouldnt be a problem

[burnt barn]

the android sdk on the other hand...

[burnt barn]

@midnight wyvern

error: ld.lld: /nix/store/28hvz5qsipq2n92qbrr8d0xlacs312fh-libglvnd-1.7.0/lib/libGLESv1_CM.so is incompatible with armelf_linux_eabi
error: ld.lld: /nix/store/28hvz5qsipq2n92qbrr8d0xlacs312fh-libglvnd-1.7.0/lib/libGLESv2.so is incompatible with armelf_linux_eabi

hmm looks like an issue in your SDL bindings, for some reason it prefers my system gles rather than the sdk one... I removed the system one and now it builds

[burnt barn]

• you didnt create sdl with the right flag (SDL_INIT_VIDEO)
• you didnt call SDL_Vulkan_LoadLibrary
• you should check if portability is available before adding it (my phone doesnt have it)

i recommend starting with vulkan on desktop before moving onto android because its significantly easier to debug

[midnight wyvern]

alright, thank you!

[nimble cobalt]

Hey there, I'm trying out the vulkan-zig bindings in an embedded project, so we don't have X/wayland in there. vkcube runs fine on it and basically uses direct display rendering. Due to this, we're basically not using any SDL/GLFW bindings and need to load vkGetInstanceProcAddr ourselves. We seem to be having some issues with it, here's current simplified code:

const std = @import("std");
const vk = @import("vulkan");

pub fn main() !void {
    var lib = try std.DynLib.openZ("libvulkan.so.1");
    defer lib.close();

    const vkGetInstanceLookup = lib.lookup(vk.PfnGetInstanceProcAddr, "vkGetInstanceProcAddr")
        orelse return error.SymbolNotFound;

    const vkb = vk.BaseWrapper.load(vkGetInstanceLookup);
    ...
}

This segfaults on const vkb = vk.BaseWrapper.load(vkGetInstanceLookup);:

/usr/bin/zig build run
Segmentation fault at address 0x0
???:?:?: 0x0 in ??? (???)```

Any recommendations on what would you use to load this or what the issue might be? Thanks!

Zig version 0.14.1, vulkan-zig branch zig-0.14-compat, Vulkan-Headers v1.4.317

[nimble cobalt]

Solved, in case anyone finds this in the future: exe.linkLibC();

[hollow jasper]

I know this is too early.
but i just woke up on this https://ziglang.org/download/0.15.1/release-notes.html

[bronze plover]

Seems like parser borks on 1.4.339: https://github.com/Snektron/vulkan-zig/issues/220

Hmm wanted to do a quick test before going to bed but none of the hunks I suspected we're the culprit

[burnt barn]

yeah sorry about that

i need to get on that

its been broken for a while

[burnt barn]

ah they finally made the <proto> definitions a bit better

[dusty oriole]

Hi, I am trying to use push descriptors and distinct immutable samplers and images. Any idea why ⁨⁨cmd.pushDescriptorSet⁩⁩ segfaults?
⁨⁨```
fn pushDescriptorSets (
self: *UiRenderer, cmd: vk.CommandBufferProxy
) void {
const sampler_write = vk.WriteDescriptorSet {
.descriptor_type = .sampler,

    .dst_binding = 0,
    .descriptor_count = 1,
    .dst_array_element = 0,

    .p_buffer_info = undefined,
    .p_image_info = undefined,
    .p_texel_buffer_view = undefined,
    .dst_set = .null_handle
};

const txr_atlas_info = vk.DescriptorImageInfo {
    .image_view = self.txr_atlas.img_view,
    .image_layout = .shader_read_only_optimal,
    .sampler = .null_handle,
};

const txr_atlas_write = vk.WriteDescriptorSet {
    .descriptor_type = .sampled_image,

    .dst_binding = 1,
    .descriptor_count = 1,
    .dst_array_element = 0,

    .p_image_info = @ptrCast (&txr_atlas_info),

    .p_buffer_info = undefined,
    .p_texel_buffer_view = undefined,
    .dst_set = .null_handle
};

const write_array = [_] vk.WriteDescriptorSet {
    sampler_write, txr_atlas_write
};

cmd.pushDescriptorSet (
    .graphics, self.ppl_layout, 0, @intCast (write_array.len), &write_array
);

}

Here's the create function too:
⁨```
fn createDescriptorSetLayout (self: *UiRenderer, gpu: Gpu) !void {
const sampler_bdg = vk.DescriptorSetLayoutBinding {
.binding = 0,
.descriptor_count = 1,
.descriptor_type = .sampler,
.p_immutable_samplers = @ptrCast (&self.sampler),
.stage_flags = .{.fragment_bit = true}
};

const txr_atlas_bdg = vk.DescriptorSetLayoutBinding {
    .binding = 1,
    .descriptor_count = 1,
    .descriptor_type = .sampled_image,
    .stage_flags = .{.fragment_bit = true}
};

const bindings = [_] vk.DescriptorSetLayoutBinding {sampler_bdg, txr_atlas_bdg};

const info = vk.DescriptorSetLayoutCreateInfo {
    .flags = .{.push_descriptor_bit = true},

    .binding_count = @intCast (bindings.len),
    .p_bindings = &bindings
};

self.desc_layout = try gpu.dev.createDescriptorSetLayout (&info, null);

}

[quiet vector]

Do you have the extension enabled?

[dusty oriole]

Do I need it if I enabled the feature and the vulkan 1.4 api?

[quiet vector]

Oh then you shouldn't

Have you tried running under validation layers? Also a debugger

Oh maybe make a zig-help thread

[dusty oriole]

It actually segfaults in the layer

Maybe I'll try without when I get home. I wonder if it expects null for those values instead of undefined...

[dusty oriole]

Supplying a value dummy value for p_image_info fixed it like this:

const sampler_info = vk.DescriptorImageInfo {
    .image_layout = .undefined,
    .image_view = .null_handle,
    .sampler = .null_handle
    };

I think this might be a bug though. The docs say that this value should be ignored:

To push an immutable sampler, use a VkWriteDescriptorSet with dstBinding and dstArrayElement selecting the immutable sampler’s binding. If the descriptor type is VK_DESCRIPTOR_TYPE_SAMPLER, the pImageInfo parameter is ignored and the immutable sampler is taken from the push descriptor set layout in the pipeline layout. If the descriptor type is VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, the sampler member of the pImageInfo parameter is ignored and the immutable sampler is taken from the push descriptor set layout in the pipeline layout.

[tulip cloud]

i have been using these bindings for months now, works great, thanks!

Now to wait until zig has full support for shader land and it's going to be even better.

[tulip cloud]

Hello.

don't have a github account, but i think i found a bug with the library.

wrapper.dispatch.vkGetDeviceFaultReportsKHR is null even when enabling the device fault extension and feature.

[burnt barn]

Does your driver have that available?