Two beginner questions #5133
-
|
I am learning the basics of webgpu, but have two minor questions:
I find it confusing in the spec's Introduction as well:
|
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 4 replies
-
|
Beta Was this translation helpful? Give feedback.
-
|
My confusion here was not so much about what the writeBuffer does, which is what you cite (and useful) but about the concept of a queue. I understand that it is a set of operations, but, for example, how do these interact with operations in a command buffer? are they always required to be completed beforehand? and such. |
Beta Was this translation helpful? Give feedback.
-
|
The queue is the thing that actually executes commands. A command buffer is just a buffer full of commands. You get those commands executed by submitting them the queue const encoder = device.createCommandEncoder();
...
const commandBuffer = encoder.finish();
// at this point, no commands have been executed. They've just been put into a buffer
device.queue.submit([commandBuffer]); // executes the commandsHaving the queue lets you choose the order things in the queue are executed.
I'm not sure what you're asking here. Except for object creation (like device.queue.writeBuffer(...)
device.queue.submit([someComamndBuffer]);
device.queue.writeTexture(...)
device.queue.copyExternalImageToTexture(...);
device.queue.submit([someOtherComamndBuffer]); |
Beta Was this translation helpful? Give feedback.
-
|
It roughly mirrors Vulkan's
In WebGPU terms, the Queues are essentially a way to enqueue work to be performed on the GPU. |
Beta Was this translation helpful? Give feedback.
-
|
Interesting, thanks! Along with your comment I got some insight from this post:
It seems to me that the adapter then is a translation layer but maps to a single GPU, but each application calls to a slice of the gpu by using those logical devices. So request device is "request logical device" a bit like you indicated. I'd appreciate any correction. |
Beta Was this translation helpful? Give feedback.
-
|
The naming is confusing - we kind of picked arbitrary names for the two concepts. :) That's essentially correct, yes. An adapter is generally some pair of (GPU hardware + implementation of WebGPU on top of it). For example it might be WebGPU->D3D12->integrated, WebGPU->D3D12->discrete, WebGPU->Vulkan->discrete, WebGPU->CPU fallback. Those could theoretically all exist at the same time, though in practice they wouldn't. And the logical device is just a multitasking abstraction, like how the OS can run many processes on one CPU. |
Beta Was this translation helpful? Give feedback.
It roughly mirrors Vulkan's
PhysicalDevice,Device, andQueue:https://docs.vulkan.org/spec/latest/chapters/devsandqueues.html
In WebGPU terms, the
Adapteris your physical hardware and theDeviceis the "logical instance".Queues are essentially a way to enqueue work to be performed on the GPU.