Increasing access
Unsure
Most appropriate sub-area of p5.js?
Feature request details
Current behavior
p5.StorageBuffer.update() always rewrites the entire buffer from index 0. If a user wants to change a single element for example, one particle out of 10,000, they currently have to:
await buf.read() - full GPU→CPU sync- modify the element in the JS array
buf.update(data) - rewrite the whole buffer
This is slow and awkward for simple updates.
Proposed feature
Add a set(index, value) method to p5.StorageBuffer that updates a single element at a given index without affecting the rest of the buffer.
// Float buffer
let buf = createStorage(new Float32Array([1, 2, 3, 4]));
buf.set(2, 9.5); // only updates index 2
// Struct buffer
let particles = createStorage(makeParticles());
particles.set(42, { position: createVector(0, 0), velocity: createVector(1, 0) });Why this helps
WebGPU's GPUQueue.writeBuffer() natively supports a byte offset, so a partial write doesn’t need a readback, staging buffer, or GPU sync. The cost is proportional to one element instead of the whole buffer.
Implementation notes
The byte offset for index i in a struct buffer is i * schema.stride. For a float buffer it's i * 4. The existing _packField logic already handles all field types, so packing a single element just reuses it with baseOffset = 0, then calls:
device.queue.writeBuffer(this.buffer, byteOffset, packedElement);
Scope for v1:
- Full element replacement only (no partial field updates)
- Works for both float and struct buffers
- Bounds check with a friendly error
This complements the .read() method added recently.
Related to #8722
This idea came up in discussion with @davepagurek after the .read() implementation.
Increasing access
Unsure
Most appropriate sub-area of p5.js?
Feature request details
Current behavior
p5.StorageBuffer.update()always rewrites the entire buffer from index 0. If a user wants to change a single element for example, one particle out of 10,000, they currently have to:await buf.read()- full GPU→CPU syncbuf.update(data)- rewrite the whole bufferThis is slow and awkward for simple updates.
Proposed feature
Add a
set(index, value)method top5.StorageBufferthat updates a single element at a given index without affecting the rest of the buffer.Why this helps
WebGPU's
GPUQueue.writeBuffer()natively supports a byte offset, so a partial write doesn’t need a readback, staging buffer, or GPU sync. The cost is proportional to one element instead of the whole buffer.Implementation notes
The byte offset for index
iin a struct buffer isi * schema.stride. For a float buffer it'si * 4. The existing_packFieldlogic already handles all field types, so packing a single element just reuses it withbaseOffset = 0, then calls:Scope for v1:
This complements the
.read()method added recently.Related to #8722
This idea came up in discussion with @davepagurek after the
.read()implementation.