BufferGeometry: Add support for interleaved data in computeVertexNormals().

this code:

					vA = indices[ i + 0 ] * 3;
					vB = indices[ i + 1 ] * 3;
					vC = indices[ i + 2 ] * 3; 

and corresponding un-indexed code:

					pA.fromArray( positions, i );
					pB.fromArray( positions, i + 3 );
					pC.fromArray( positions, i + 6 );

fails to take into account the property stride.

repro steps look a bit like this:

	const meshData = (...)
	var geometry = new BufferGeometry();

	var arrayBufferFloat = new Float32Array(meshData.vertices);
	var arrayBufferByte = new Uint8Array(arrayBufferFloat.buffer); // alias this buffer, but with U8

	const buffer32 = new InterleavedBuffer(arrayBufferFloat, meshData.vertexStride);
	const buffer8 = new InterleavedBuffer(arrayBufferByte, meshData.vertexStride * 4);
	geometry.setAttribute(
		'position',
		new InterleavedBufferAttribute(buffer32, 3, 0));

	geometry.setAttribute(
		'color',
		new InterleavedBufferAttribute(buffer8, 4, meshData.colorOffset * 4, true));

	geometry.setIndex(new BufferAttribute(new Uint32Array(meshData.faces), 1));

	geometry.computeVertexNormals();

The “https://threejs.org/examples/#webgl_buffergeometry_points_interleaved” example would reproduce this if it was modified to draw triangles instead of particles.

[WestLangley]

Exactly. So our work is done. 😃

[Mugen87]

But honestly, if you are using interleaved arrays, you may want your normals to be interleaved as well.

The primary purpose of using interleaving vertex data is the better memory locality. Meaning it’s more likely to have cache hits on the pre-transform vertex cache of the GPU. So without interleaving, you will miss the main purpose of using instances of InterleavedBufferAttribute.

The complexity of BufferGeometry.computeVertexNormals() will be much bigger if the engine should support this use case since you have to generate a new InterleavedBuffer from scratch as well as new InterleavedBufferAttributes.

@mrdoob Should we support this?^^