The beginComputePass() method of the GPUCommandEncoder interface starts encoding a compute pass, returning a GPUComputePassEncoder that can be used to control computation.
beginComputePass()
beginComputePass(descriptor)
The following criteria must be met when calling beginComputePass(), otherwise a GPUValidationError is generated and an invalid GPUComputePassEncoder is returned:
- The
timestamp-query feature is enabled in the GPUDevice. - No two
timestampWrites objects have the same location. In effect, this means that you can only run two timestamp queries per render pass. - For each timestamp query, the
querySet GPUQuerySet.type is "timestamp", and the queryIndex value is less than the GPUQuerySet.count.
In our basic compute demo, several commands are recorded via a GPUCommandEncoder. Most of these commands originate from the GPUComputePassEncoder created via beginComputePass().
const commandEncoder = device.createCommandEncoder();
const passEncoder = commandEncoder.beginComputePass();
passEncoder.setPipeline(computePipeline);
passEncoder.setBindGroup(0, bindGroup);
passEncoder.dispatchWorkgroups(Math.ceil(BUFFER_SIZE / 64));
passEncoder.end();
commandEncoder.copyBufferToBuffer(
output,
0,
stagingBuffer,
0,
BUFFER_SIZE,
);
device.queue.submit([commandEncoder.finish()]);