Fix ConvertType

[LeviPesin]

Related issue: #24964 (comment)

Description

I think this should fix it...

[LeviPesin]

It seems that actually that bug is two bugs, and this PR fixes only one of them... With this PR WebGPU complains on uvec4( index, index, index, uint( 1 ) ) like so:

1 error(s) generated while compiling the shader:
:32:49 error: no matching initializer for vec4<u32>(vec3<u32>, abstract-float)

10 candidate initializers:
  vec4(xyz: vec3<T>, w: T) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(xy: vec2<T>, zw: vec2<T>) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(x: T, zyw: vec3<T>) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(vec4<T>) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(T) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4<T>() -> vec4<T>  where: T is f32, f16, i32, u32 or bool
  vec4(xy: vec2<T>, z: T, w: T) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(x: T, yz: vec2<T>, w: T) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(x: T, y: T, zw: vec2<T>) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool
  vec4(x: T, y: T, z: T, w: T) -> vec4<T>  where: T is abstract-int, abstract-float, f32, f16, i32, u32 or bool

5 candidate conversions:
  vec4<T>(vec4<U>) -> vec4<f32>  where: T is f32, U is abstract-int, abstract-float, i32, f16, u32 or bool
  vec4<T>(vec4<U>) -> vec4<f16>  where: T is f16, U is abstract-int, abstract-float, f32, i32, u32 or bool
  vec4<T>(vec4<U>) -> vec4<i32>  where: T is i32, U is abstract-int, abstract-float, f32, f16, u32 or bool
  vec4<T>(vec4<U>) -> vec4<u32>  where: T is u32, U is abstract-int, abstract-float, f32, f16, i32 or bool
  vec4<T>(vec4<U>) -> vec4<bool>  where: T is bool, U is abstract-int, abstract-float, f32, f16, i32 or u32

  NodeBuffer_291.nodeUniform0[ instanceIndex ] = vec4<u32>( vec4<u32>( instanceIndex, instanceIndex, instanceIndex, 1u ).xyz, 1.0 );
                                                 ^^^^

Note there are no vec4<f32> there anymore.

[LeviPesin]

The second bug is very strange - debugging it, it seems that builder.format( 'vec4<u32>( instanceIndex, instanceIndex, instanceIndex, 1u )', 'uvec4', 'vec3' ) and then builder.format( 'vec4<u32>( instanceIndex, instanceIndex, instanceIndex, 1u ).xyz', 'vec3', 'uvec4' ) are called...
This is a bug in my own code 🤦‍♂️. So this fix seems sufficient.

[sunag]

Thanks! :)