servo/components/webgpu/lib.rs
2020-06-04 01:19:13 +05:30

639 lines
25 KiB
Rust

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at https://mozilla.org/MPL/2.0/. */
#[macro_use]
extern crate log;
#[macro_use]
pub extern crate wgpu_core as wgpu;
pub extern crate wgpu_types as wgt;
pub mod identity;
use arrayvec::ArrayVec;
use identity::{IdentityRecyclerFactory, WebGPUMsg};
use ipc_channel::ipc::{self, IpcReceiver, IpcSender};
use malloc_size_of::{MallocSizeOf, MallocSizeOfOps};
use serde::{Deserialize, Serialize};
use servo_config::pref;
use smallvec::SmallVec;
use std::ffi::CString;
use std::ptr;
use wgpu::{
binding_model::{
BindGroupDescriptor, BindGroupEntry, BindGroupLayoutDescriptor, BindGroupLayoutEntry,
},
id,
instance::RequestAdapterOptions,
};
#[derive(Debug, Deserialize, Serialize)]
pub enum WebGPUResponse {
RequestAdapter {
adapter_name: String,
adapter_id: WebGPUAdapter,
channel: WebGPU,
},
RequestDevice {
device_id: WebGPUDevice,
queue_id: WebGPUQueue,
_descriptor: wgt::DeviceDescriptor,
},
}
pub type WebGPUResponseResult = Result<WebGPUResponse, String>;
#[derive(Debug, Deserialize, Serialize)]
pub enum WebGPURequest {
CommandEncoderFinish {
command_encoder_id: id::CommandEncoderId,
// TODO(zakorgy): Serialize CommandBufferDescriptor in wgpu-core
// wgpu::command::CommandBufferDescriptor,
},
CopyBufferToBuffer {
command_encoder_id: id::CommandEncoderId,
source_id: id::BufferId,
source_offset: wgt::BufferAddress,
destination_id: id::BufferId,
destination_offset: wgt::BufferAddress,
size: wgt::BufferAddress,
},
CreateBindGroup {
device_id: id::DeviceId,
bind_group_id: id::BindGroupId,
bind_group_layout_id: id::BindGroupLayoutId,
bindings: Vec<BindGroupEntry>,
},
CreateBindGroupLayout {
device_id: id::DeviceId,
bind_group_layout_id: id::BindGroupLayoutId,
bindings: Vec<BindGroupLayoutEntry>,
},
CreateBuffer {
device_id: id::DeviceId,
buffer_id: id::BufferId,
descriptor: wgt::BufferDescriptor<String>,
},
CreateCommandEncoder {
device_id: id::DeviceId,
// TODO(zakorgy): Serialize CommandEncoderDescriptor in wgpu-core
// wgpu::command::CommandEncoderDescriptor,
command_encoder_id: id::CommandEncoderId,
},
CreateComputePipeline {
device_id: id::DeviceId,
compute_pipeline_id: id::ComputePipelineId,
pipeline_layout_id: id::PipelineLayoutId,
program_id: id::ShaderModuleId,
entry_point: String,
},
CreatePipelineLayout {
device_id: id::DeviceId,
pipeline_layout_id: id::PipelineLayoutId,
bind_group_layouts: Vec<id::BindGroupLayoutId>,
},
CreateRenderPipeline {
device_id: id::DeviceId,
render_pipeline_id: id::RenderPipelineId,
pipeline_layout_id: id::PipelineLayoutId,
vertex_module: id::ShaderModuleId,
vertex_entry_point: String,
fragment_module: Option<id::ShaderModuleId>,
fragment_entry_point: Option<String>,
primitive_topology: wgt::PrimitiveTopology,
rasterization_state: wgt::RasterizationStateDescriptor,
color_states: ArrayVec<[wgt::ColorStateDescriptor; wgpu::device::MAX_COLOR_TARGETS]>,
depth_stencil_state: Option<wgt::DepthStencilStateDescriptor>,
vertex_state: (
wgt::IndexFormat,
Vec<(u64, wgt::InputStepMode, Vec<wgt::VertexAttributeDescriptor>)>,
),
sample_count: u32,
sample_mask: u32,
alpha_to_coverage_enabled: bool,
},
CreateSampler {
device_id: id::DeviceId,
sampler_id: id::SamplerId,
descriptor: wgt::SamplerDescriptor<String>,
},
CreateShaderModule {
device_id: id::DeviceId,
program_id: id::ShaderModuleId,
program: Vec<u32>,
},
CreateTexture {
device_id: id::DeviceId,
texture_id: id::TextureId,
descriptor: wgt::TextureDescriptor<String>,
},
CreateTextureView {
texture_id: id::TextureId,
texture_view_id: id::TextureViewId,
descriptor: wgt::TextureViewDescriptor<String>,
},
DestroyBuffer(id::BufferId),
DestroyTexture(id::TextureId),
Exit(IpcSender<()>),
RequestAdapter {
sender: IpcSender<WebGPUResponseResult>,
options: RequestAdapterOptions,
ids: SmallVec<[id::AdapterId; 4]>,
},
RequestDevice {
sender: IpcSender<WebGPUResponseResult>,
adapter_id: WebGPUAdapter,
descriptor: wgt::DeviceDescriptor,
device_id: id::DeviceId,
},
RunComputePass {
command_encoder_id: id::CommandEncoderId,
pass_data: Vec<u8>,
},
RunRenderPass {
command_encoder_id: id::CommandEncoderId,
pass_data: Vec<u8>,
},
Submit {
queue_id: id::QueueId,
command_buffers: Vec<id::CommandBufferId>,
},
UnmapBuffer {
device_id: id::DeviceId,
buffer_id: id::BufferId,
array_buffer: Vec<u8>,
},
}
#[derive(Clone, Debug, Deserialize, Serialize)]
pub struct WebGPU(pub IpcSender<WebGPURequest>);
impl WebGPU {
pub fn new() -> Option<(Self, IpcReceiver<WebGPUMsg>)> {
if !pref!(dom.webgpu.enabled) {
return None;
}
let (sender, receiver) = match ipc::channel() {
Ok(sender_and_receiver) => sender_and_receiver,
Err(e) => {
warn!(
"Failed to create sender and receiver for WGPU thread ({})",
e
);
return None;
},
};
let sender_clone = sender.clone();
let (script_sender, script_recv) = match ipc::channel() {
Ok(sender_and_receiver) => sender_and_receiver,
Err(e) => {
warn!(
"Failed to create receiver and sender for WGPU thread ({})",
e
);
return None;
},
};
if let Err(e) = std::thread::Builder::new()
.name("WGPU".to_owned())
.spawn(move || {
WGPU::new(receiver, sender_clone, script_sender).run();
})
{
warn!("Failed to spwan WGPU thread ({})", e);
return None;
}
Some((WebGPU(sender), script_recv))
}
pub fn exit(&self, sender: IpcSender<()>) -> Result<(), &'static str> {
self.0
.send(WebGPURequest::Exit(sender))
.map_err(|_| "Failed to send Exit message")
}
}
struct WGPU {
receiver: IpcReceiver<WebGPURequest>,
sender: IpcSender<WebGPURequest>,
script_sender: IpcSender<WebGPUMsg>,
global: wgpu::hub::Global<IdentityRecyclerFactory>,
adapters: Vec<WebGPUAdapter>,
devices: Vec<WebGPUDevice>,
// Track invalid adapters https://gpuweb.github.io/gpuweb/#invalid
_invalid_adapters: Vec<WebGPUAdapter>,
}
impl WGPU {
fn new(
receiver: IpcReceiver<WebGPURequest>,
sender: IpcSender<WebGPURequest>,
script_sender: IpcSender<WebGPUMsg>,
) -> Self {
let factory = IdentityRecyclerFactory {
sender: script_sender.clone(),
};
WGPU {
receiver,
sender,
script_sender,
global: wgpu::hub::Global::new("wgpu-core", factory),
adapters: Vec::new(),
devices: Vec::new(),
_invalid_adapters: Vec::new(),
}
}
fn run(mut self) {
while let Ok(msg) = self.receiver.recv() {
match msg {
WebGPURequest::CommandEncoderFinish { command_encoder_id } => {
let global = &self.global;
let _ = gfx_select!(command_encoder_id => global.command_encoder_finish(
command_encoder_id,
&wgt::CommandBufferDescriptor::default()
));
},
WebGPURequest::CopyBufferToBuffer {
command_encoder_id,
source_id,
source_offset,
destination_id,
destination_offset,
size,
} => {
let global = &self.global;
let _ = gfx_select!(command_encoder_id => global.command_encoder_copy_buffer_to_buffer(
command_encoder_id,
source_id,
source_offset,
destination_id,
destination_offset,
size
));
},
WebGPURequest::CreateBindGroup {
device_id,
bind_group_id,
bind_group_layout_id,
bindings,
} => {
let global = &self.global;
let descriptor = BindGroupDescriptor {
layout: bind_group_layout_id,
entries: bindings.as_ptr(),
entries_length: bindings.len(),
label: ptr::null(),
};
let _ = gfx_select!(bind_group_id =>
global.device_create_bind_group(device_id, &descriptor, bind_group_id));
},
WebGPURequest::CreateBindGroupLayout {
device_id,
bind_group_layout_id,
bindings,
} => {
let global = &self.global;
let descriptor = BindGroupLayoutDescriptor {
entries: bindings.as_ptr(),
entries_length: bindings.len(),
label: ptr::null(),
};
let _ = gfx_select!(bind_group_layout_id =>
global.device_create_bind_group_layout(device_id, &descriptor, bind_group_layout_id));
},
WebGPURequest::CreateBuffer {
device_id,
buffer_id,
descriptor,
} => {
let global = &self.global;
let st = CString::new(descriptor.label.as_bytes()).unwrap();
let _ = gfx_select!(buffer_id =>
global.device_create_buffer(device_id, &descriptor.map_label(|_| st.as_ptr()), buffer_id));
},
WebGPURequest::CreateCommandEncoder {
device_id,
command_encoder_id,
} => {
let global = &self.global;
let _ = gfx_select!(command_encoder_id =>
global.device_create_command_encoder(device_id, &Default::default(), command_encoder_id));
},
WebGPURequest::CreateComputePipeline {
device_id,
compute_pipeline_id,
pipeline_layout_id,
program_id,
entry_point,
} => {
let global = &self.global;
let entry_point = std::ffi::CString::new(entry_point).unwrap();
let descriptor = wgpu_core::pipeline::ComputePipelineDescriptor {
layout: pipeline_layout_id,
compute_stage: wgpu_core::pipeline::ProgrammableStageDescriptor {
module: program_id,
entry_point: entry_point.as_ptr(),
},
};
let _ = gfx_select!(compute_pipeline_id =>
global.device_create_compute_pipeline(device_id, &descriptor, compute_pipeline_id));
},
WebGPURequest::CreatePipelineLayout {
device_id,
pipeline_layout_id,
bind_group_layouts,
} => {
let global = &self.global;
let descriptor = wgpu_core::binding_model::PipelineLayoutDescriptor {
bind_group_layouts: bind_group_layouts.as_ptr(),
bind_group_layouts_length: bind_group_layouts.len(),
};
let _ = gfx_select!(pipeline_layout_id =>
global.device_create_pipeline_layout(device_id, &descriptor, pipeline_layout_id));
},
//TODO: consider https://github.com/gfx-rs/wgpu/issues/684
WebGPURequest::CreateRenderPipeline {
device_id,
render_pipeline_id,
pipeline_layout_id,
vertex_module,
vertex_entry_point,
fragment_module,
fragment_entry_point,
primitive_topology,
rasterization_state,
color_states,
depth_stencil_state,
vertex_state,
sample_count,
sample_mask,
alpha_to_coverage_enabled,
} => {
let global = &self.global;
let vertex_ep = std::ffi::CString::new(vertex_entry_point).unwrap();
let frag_stage = match fragment_module {
Some(frag) => {
let frag_ep =
std::ffi::CString::new(fragment_entry_point.unwrap()).unwrap();
let frag_module = wgpu_core::pipeline::ProgrammableStageDescriptor {
module: frag,
entry_point: frag_ep.as_ptr(),
};
Some(frag_module)
},
None => None,
};
let descriptor = wgpu_core::pipeline::RenderPipelineDescriptor {
layout: pipeline_layout_id,
vertex_stage: wgpu_core::pipeline::ProgrammableStageDescriptor {
module: vertex_module,
entry_point: vertex_ep.as_ptr(),
},
fragment_stage: frag_stage
.as_ref()
.map_or(ptr::null(), |fs| fs as *const _),
primitive_topology,
rasterization_state: &rasterization_state as *const _,
color_states: color_states.as_ptr(),
color_states_length: color_states.len(),
depth_stencil_state: depth_stencil_state
.as_ref()
.map_or(ptr::null(), |dss| dss as *const _),
vertex_state: wgpu_core::pipeline::VertexStateDescriptor {
index_format: vertex_state.0,
vertex_buffers_length: vertex_state.1.len(),
vertex_buffers: vertex_state
.1
.iter()
.map(|buffer| wgpu_core::pipeline::VertexBufferLayoutDescriptor {
array_stride: buffer.0,
step_mode: buffer.1,
attributes_length: buffer.2.len(),
attributes: buffer.2.as_ptr(),
})
.collect::<Vec<_>>()
.as_ptr(),
},
sample_count,
sample_mask,
alpha_to_coverage_enabled,
};
let _ = gfx_select!(render_pipeline_id =>
global.device_create_render_pipeline(device_id, &descriptor, render_pipeline_id));
},
WebGPURequest::CreateSampler {
device_id,
sampler_id,
descriptor,
} => {
let global = &self.global;
let st = CString::new(descriptor.label.as_bytes()).unwrap();
let _ = gfx_select!(sampler_id =>
global.device_create_sampler(device_id, &descriptor.map_label(|_| st.as_ptr()), sampler_id));
},
WebGPURequest::CreateShaderModule {
device_id,
program_id,
program,
} => {
let global = &self.global;
let descriptor = wgpu_core::pipeline::ShaderModuleDescriptor {
code: wgpu_core::U32Array {
bytes: program.as_ptr(),
length: program.len(),
},
};
let _ = gfx_select!(program_id =>
global.device_create_shader_module(device_id, &descriptor, program_id));
},
WebGPURequest::CreateTexture {
device_id,
texture_id,
descriptor,
} => {
let global = &self.global;
let st = CString::new(descriptor.label.as_bytes()).unwrap();
let _ = gfx_select!(texture_id =>
global.device_create_texture(device_id, &descriptor.map_label(|_| st.as_ptr()), texture_id));
},
WebGPURequest::CreateTextureView {
texture_id,
texture_view_id,
descriptor,
} => {
let global = &self.global;
let st = CString::new(descriptor.label.as_bytes()).unwrap();
let _ = gfx_select!(texture_view_id => global.texture_create_view(
texture_id,
Some(&descriptor.map_label(|_| st.as_ptr())),
texture_view_id
));
},
WebGPURequest::DestroyBuffer(buffer) => {
let global = &self.global;
gfx_select!(buffer => global.buffer_destroy(buffer));
},
WebGPURequest::DestroyTexture(texture) => {
let global = &self.global;
gfx_select!(texture => global.texture_destroy(texture));
},
WebGPURequest::Exit(sender) => {
if let Err(e) = self.script_sender.send(WebGPUMsg::Exit) {
warn!("Failed to send WebGPUMsg::Exit to script ({})", e);
}
drop(self.global);
if let Err(e) = sender.send(()) {
warn!("Failed to send response to WebGPURequest::Exit ({})", e)
}
return;
},
WebGPURequest::RequestAdapter {
sender,
options,
ids,
} => {
let adapter_id = match self.global.pick_adapter(
&options,
wgpu::instance::AdapterInputs::IdSet(&ids, |id| id.backend()),
) {
Some(id) => id,
None => {
if let Err(e) =
sender.send(Err("Failed to get webgpu adapter".to_string()))
{
warn!(
"Failed to send response to WebGPURequest::RequestAdapter ({})",
e
)
}
return;
},
};
let adapter = WebGPUAdapter(adapter_id);
self.adapters.push(adapter);
let global = &self.global;
let info = gfx_select!(adapter_id => global.adapter_get_info(adapter_id));
if let Err(e) = sender.send(Ok(WebGPUResponse::RequestAdapter {
adapter_name: info.name,
adapter_id: adapter,
channel: WebGPU(self.sender.clone()),
})) {
warn!(
"Failed to send response to WebGPURequest::RequestAdapter ({})",
e
)
}
},
WebGPURequest::RequestDevice {
sender,
adapter_id,
descriptor,
device_id,
} => {
let global = &self.global;
let id = gfx_select!(device_id => global.adapter_request_device(
adapter_id.0,
&descriptor,
None,
device_id
));
let device = WebGPUDevice(id);
// Note: (zakorgy) Note sure if sending the queue is needed at all,
// since wgpu-core uses the same id for the device and the queue
let queue = WebGPUQueue(id);
self.devices.push(device);
if let Err(e) = sender.send(Ok(WebGPUResponse::RequestDevice {
device_id: device,
queue_id: queue,
_descriptor: descriptor,
})) {
warn!(
"Failed to send response to WebGPURequest::RequestDevice ({})",
e
)
}
},
WebGPURequest::RunComputePass {
command_encoder_id,
pass_data,
} => {
let global = &self.global;
gfx_select!(command_encoder_id => global.command_encoder_run_compute_pass(
command_encoder_id,
&pass_data
));
},
WebGPURequest::RunRenderPass {
command_encoder_id,
pass_data,
} => {
let global = &self.global;
gfx_select!(command_encoder_id => global.command_encoder_run_render_pass(
command_encoder_id,
&pass_data
));
},
WebGPURequest::Submit {
queue_id,
command_buffers,
} => {
let global = &self.global;
let _ = gfx_select!(queue_id => global.queue_submit(
queue_id,
&command_buffers
));
},
WebGPURequest::UnmapBuffer {
device_id,
buffer_id,
array_buffer,
} => {
let global = &self.global;
gfx_select!(buffer_id => global.device_set_buffer_sub_data(
device_id,
buffer_id,
0,
array_buffer.as_slice()
));
},
}
}
}
}
macro_rules! webgpu_resource {
($name:ident, $id:ty) => {
#[derive(Clone, Copy, Debug, Deserialize, Hash, PartialEq, Serialize)]
pub struct $name(pub $id);
impl MallocSizeOf for $name {
fn size_of(&self, _ops: &mut MallocSizeOfOps) -> usize {
0
}
}
impl Eq for $name {}
};
}
webgpu_resource!(WebGPUAdapter, id::AdapterId);
webgpu_resource!(WebGPUBindGroup, id::BindGroupId);
webgpu_resource!(WebGPUBindGroupLayout, id::BindGroupLayoutId);
webgpu_resource!(WebGPUBuffer, id::BufferId);
webgpu_resource!(WebGPUCommandBuffer, id::CommandBufferId);
webgpu_resource!(WebGPUCommandEncoder, id::CommandEncoderId);
webgpu_resource!(WebGPUComputePipeline, id::ComputePipelineId);
webgpu_resource!(WebGPUDevice, id::DeviceId);
webgpu_resource!(WebGPUPipelineLayout, id::PipelineLayoutId);
webgpu_resource!(WebGPUQueue, id::QueueId);
webgpu_resource!(WebGPURenderPipeline, id::RenderPipelineId);
webgpu_resource!(WebGPUSampler, id::SamplerId);
webgpu_resource!(WebGPUShaderModule, id::ShaderModuleId);
webgpu_resource!(WebGPUTexture, id::TextureId);
webgpu_resource!(WebGPUTextureView, id::TextureViewId);