Update ConvertType

[LeviPesin]

Related issue: #23601 (comment)

Description

• Remove valueComponents.
• Add arguments' flatting.
• Remove valueClass.
• Add cache.
• Rework ConvertType.
• Fix uint generation in NodeBuilder.

[LeviPesin]

Actually, here can be reused function getValueFromType() from #23663 to remove the valueClass parameter.

Also I suggest adding to the ConvertType some cache - so e.g. if float(0) would be called hundred times it will return the same value. This cannot be done with WeakMap, because it only accepts keys which are objects, so maybe it should be done with just Map.

[LeviPesin]

@sunag What is your opinion about something like vec4(1.0, new Vector2(1.0, 1.0), 1.0)? Should it be supported or rather not?

[sunag]

@sunag What is your opinion about something like vec4(1.0, new Vector2(1.0, 1.0), 1.0)? Should it be supported or rather not?

Sounds good!

Can you do this update without change src(core) code?

[LeviPesin]

This is the part that I called arguments' flatting. I changed the core code to remove valueComponents - I can revert it back.

[LeviPesin]

Done.

[sunag]

Can you check the webgpu_nodes_playground example? Seems that some vectors contains NaN values.

[LeviPesin]

I've fixed the error, but for some reason the models now appear to be white instead of black...

[sunag]

@LeviPesin Seems that vec*() no longer works with scalar values.

See ColorSpaceNode.LinearTosRGB:

const a = sub( mul( pow( value.rgb, vec3( 0.41666 ) ), 1.055 ), vec3( 0.055 ) );
const b = mul( rgb, 12.92 );
const factor = vec3( lessThanEqual( rgb, vec3( 0.0031308 ) ) );

const rgbResult = mix( a, b, factor );

return join( rgbResult.r, rgbResult.g, rgbResult.b, value.a );

nodeVar4 = pow( DiffuseColor.xyz, vec3<f32>( 0.0, 0.0, 0.0 ) );
nodeVar5 = ( nodeVar4 * vec3<f32>( 1.055 ) );
nodeVar6 = ( nodeVar5 - vec3<f32>( 0.0, 0.0, 0.0 ) );
nodeVar7 = ( DiffuseColor.xyz * vec3<f32>( 12.92 ) );
nodeVar8 = mix( nodeVar6, nodeVar7, vec3<f32>( 0.0, 0.0, 0.0 ) );
Output = vec4<f32>( nodeVar8.x, nodeVar8.y, nodeVar8.z, DiffuseColor.w );

[LeviPesin]

Fixed.

[LeviPesin]

Is it better now?

Also, maybe it would be better to remove PI, PI2, PI_HALF, RECIPROCAL_PI, RECIPROCAL_PI2 (but keep EPSILON and add INFINITY = 1 / EPSILON)? Just something like Math.PI or float( Math.PI ) could be used instead when needed - and it will cache.

Also, the cache should be added to the new ConstNode( Number( prop ), 'uint' ) and new ShaderNodeObject( new ConstNode( obj ) );, I think.

[sunag]

Also, maybe it would be better to remove PI, PI2, PI_HALF, RECIPROCAL_PI, RECIPROCAL_PI2 (but keep EPSILON and add INFINITY = 1 / EPSILON)? Just something like Math.PI or float( Math.PI ) could be used instead when needed - and it will cache.

@LeviPesin I think we can keep as is for now...

[sunag]

Thanks! 👍