Expand support for GLSL integer attributes

Expanding on the vertexAttribIPointer support added in https://github.com/mrdoob/three.js/pull/19019, I would like to propose a new property:

BufferAttribute.targetType = FloatType (default)

This property would support values of the existing constants FloatType (default) and IntType.

This would expand support for integer attributes (beyond the current int32 and uint32) without changing any existing behavior.

The property would ultimately get passed through to WebGLBindingStates.vertexAttribPointer() where the code impact is minimal:

-function vertexAttribPointer( index, size, type, normalized, stride, offset ) {
+function vertexAttribPointer( index, size, type, normalized, stride, offset, targetType ) {

-    if ( capabilities.isWebGL2 === true && ( type === 5124 || type === 5125 ) ) {
+    if ( capabilities.isWebGL2 === true && ( type === 5124 || type === 5125 || targetType == IntType ) ) {

        gl.vertexAttribIPointer( index, size, type, stride, offset );

    } else {

        gl.vertexAttribPointer( index, size, type, normalized, stride, offset );

    }

}

Alternatively, this could be a boolean attribute such as BufferAttribute.toIntegers = true, as it looks like WebGL only supports vertexAttribPointer (float) and vertexAttribIPointer (int) – not vertexAttribLPointer (double).

Apologies if I’m missing anything obvious. Thank you!

[Mugen87]

Thanks for the additional information! That made it more clear. To sum up:

• The renderer currently only supports Int32BufferAttribute (gl.INT) and Uint32BufferAttribute (gl.UNSIGNED_INT) in context of vertexAttribIPointer(). Other types of integer attributes (gl.BYTE, gl.UNSIGNED_BYTE, gl.SHORT, gl.UNSIGNED_SHORT) are currently not supported.
• The purpose of this issue is to add support for the above types.

How about add a new flag to Int8BufferAttribute, Uint8BufferAttribute, Uint16BufferAttribute and Int16BufferAttribute to control how the data should be interpreted? I would prefer this instead of changing BufferAttribute.

[gkjohnson]

For ‘native’ vertex attributes like position, which our shaders assume to be floats, either make an exception to (1), or explicitly cast from int to float in the shader.

I commented here but to expand a bit on this comment - I’d prefer to not to have this kind of “magic” used internally (ie providing an exception only for position, etc). It feels like a recognition that this capability and flexibility is valuable while precluding users from being able to take advantage of it.