Buffers

Use Buffer when a kernel needs an array of values.

import { Buffer } from '@volten/core';

const values = new Buffer([1, 2, 3, 4], 'f32');
const output = new Buffer([0, 0, 0, 0], 'f32', 'rw');

The second argument is the element type. Volten uses it to pack the data and generate the shader binding.

Access Mode

The third argument controls shader access:

const input = new Buffer([1, 2, 3, 4], 'f32', 'r');
const inout = new Buffer([1, 2, 3, 4], 'f32', 'rw');

'r' generates a read-only storage binding. 'rw' generates a read-write storage binding.

If omitted, the access mode defaults to 'rw'.

This is shader access, not CPU access. JavaScript can still upload data with set() or update().

Element Count and Byte Length

Buffers have a fixed element count.

const values = new Buffer([1, 2, 3, 4], 'f32');

values.count; // 4
values.stride; // bytes per element
values.byteLength; // count * stride

Fixed size

set() and update() can change contents, but they cannot resize the buffer. Create a new Buffer when the element count changes.

WebGPU mapping

stride includes WGSL layout padding. For example, vec3f has a 16-byte stride when stored in an array.

Updating Data

Use set() to replace the full contents without changing the element count:

values.set([10, 20, 30, 40]);

Use update() to replace a contiguous range. The offset is measured in elements:

values.update([99, 100], 1);

Labels

Labels help when inspecting GPU resources:

const positions = new Buffer([[0, 0, 0]], 'vec3f', 'rw', {
    label: 'particle positions'
});