Nalgebra vector fails to be recongized

[marstaik]

I tried using nalgebra (no-std) for types in shaders, but the simplest shaders fails to compile.

use nalgebra::Vector4;
use spirv_std::spirv;

#[allow(dead_code)]
#[spirv(fragment)]
pub fn main_fs(output: &mut Vector4<f32>) {
	*output = Vector4::new(1.0, 0.0, 0.0, 1.0);
}

#[allow(dead_code)]
#[spirv(vertex)]
pub fn main_vs(
	#[spirv(vertex_index)] vert_id: i32,
	#[spirv(position, invariant)] out_pos: &mut Vector4<f32>,
) {
	*out_pos = Vector4::new(
		(vert_id - 1) as f32,
		((vert_id & 1) * 2 - 1) as f32,
		0.0,
		1.0,
	);
}

Is the module I am trying to compile, and the error received is:

  error: [VUID-Position-Position-04321] According to the Vulkan spec BuiltIn Position variable needs to be a 4-component 32-bit float vector. ID <3> (OpVariable) is not a float vector.
    |
    = note: module `/home/marios/proj/hestia_vk/target/spirv-builder/spirv-unknown-vulkan1.2/release/deps/shaders.spvs/mesh-main_vs.spv`

  warning: an unknown error occurred
    |
    = note: spirv-opt failed, leaving as unoptimized
    = note: module `/home/marios/proj/hestia_vk/target/spirv-builder/spirv-unknown-vulkan1.2/release/deps/shaders.spvs/mesh-main_vs.spv`

  error: error:0:0 - [VUID-Position-Position-04321] According to the Vulkan spec BuiltIn Position variable needs to be a 4-component 32-bit float vector. ID <3> (OpVariable) is not a float vector.
           %50 = OpInBoundsAccessChain %_ptr_Output__arr_float_uint_4 %out_pos %uint_0 %uint_0 %uint_0
    |
    = note: spirv-val failed

If I comment out the main_vs function all is well.

It seems like the Vector4 type isnt being interpreted as a [f32;4]

I am wondering if this is because of the underlying type: pub struct ArrayStorage<T, const R: usize, const C: usize>(pub [[T; R]; C]);

It evaluates with R=4, C=1, making the inner type [[f32; 4],1]

[Turtyo]

Could it because it's a wrapper ? Have you tried defining a struct with just a [f32; 4] inside and see if that works ?
Might just also be that it's indeed a [[f32; 4], 1] and it doesn't cast to [f32; 4] directly

[marstaik]

Strangely, both [f32;4] and [[f32;4];1] don't work. The shader compilation seems to fail silently.

[LegNeato]

I believe for inputs and outputs it is hardcoded to needglam. Glam also uses spirv intrinsics on vulkan so it will be faster currently.

We've had discussions on the best way to support any implementation (mint, features, a trait, etc). We haven't landed on one yet but I personally believe it is important to be able to bring your own library in this space.