wgpu/deno_webgpu/pipeline.rs
Brad Werth 9b7a965667 Add an experimental vertex pulling flag to Metal pipelines.
This proves a flag in msl::PipelineOptions that attempts to write all
Metal vertex entry points to use a vertex pulling technique. It does
this by:

1) Forcing the _buffer_sizes structure to be generated for all vertex
entry points. The structure has additional buffer_size members that
contain the byte sizes of the vertex buffers.
2) Adding new args to vertex entry points for the vertex id and/or
the instance id and for the bound buffers. If there is an existing
@builtin(vertex_index) or @builtin(instance_index) param, then no
duplicate arg is created.
3) Adding code at the beginning of the function for vertex entry points
to compare the vertex id or instance id against the lengths of all the
bound buffers, and force an early-exit if the bounds are violated.
4) Extracting the raw bytes from the vertex buffer(s) and unpacking
those bytes into the bound attributes with the expected types.
5) Replacing the varyings input and instead using the unpacked
attributes to fill any structs-as-args that are rebuilt in the entry
point.

A new naga test is added which exercises this flag and demonstrates the
effect of the transform. The msl generated by this test passes
validation.

Eventually this transformation will be the default, always-on behavior
for Metal pipelines, though the flag may remain so that naga
translation tests can be run with and without the tranformation.
2024-05-30 13:08:59 +02:00

460 lines
14 KiB
Rust

// Copyright 2018-2024 the Deno authors. All rights reserved. MIT license.
use deno_core::error::AnyError;
use deno_core::op2;
use deno_core::OpState;
use deno_core::Resource;
use deno_core::ResourceId;
use serde::Deserialize;
use serde::Serialize;
use std::borrow::Cow;
use std::collections::HashMap;
use std::rc::Rc;
use super::error::WebGpuError;
use super::error::WebGpuResult;
const MAX_BIND_GROUPS: usize = 8;
pub(crate) struct WebGpuPipelineLayout(
pub(crate) crate::Instance,
pub(crate) wgpu_core::id::PipelineLayoutId,
);
impl Resource for WebGpuPipelineLayout {
fn name(&self) -> Cow<str> {
"webGPUPipelineLayout".into()
}
fn close(self: Rc<Self>) {
gfx_select!(self.1 => self.0.pipeline_layout_drop(self.1));
}
}
pub(crate) struct WebGpuComputePipeline(
pub(crate) crate::Instance,
pub(crate) wgpu_core::id::ComputePipelineId,
);
impl Resource for WebGpuComputePipeline {
fn name(&self) -> Cow<str> {
"webGPUComputePipeline".into()
}
fn close(self: Rc<Self>) {
gfx_select!(self.1 => self.0.compute_pipeline_drop(self.1));
}
}
pub(crate) struct WebGpuRenderPipeline(
pub(crate) crate::Instance,
pub(crate) wgpu_core::id::RenderPipelineId,
);
impl Resource for WebGpuRenderPipeline {
fn name(&self) -> Cow<str> {
"webGPURenderPipeline".into()
}
fn close(self: Rc<Self>) {
gfx_select!(self.1 => self.0.render_pipeline_drop(self.1));
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
pub enum GPUAutoLayoutMode {
Auto,
}
#[derive(Deserialize)]
#[serde(untagged)]
pub enum GPUPipelineLayoutOrGPUAutoLayoutMode {
Layout(ResourceId),
Auto(GPUAutoLayoutMode),
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct GpuProgrammableStage {
module: ResourceId,
entry_point: Option<String>,
constants: Option<HashMap<String, f64>>,
}
#[op2]
#[serde]
pub fn op_webgpu_create_compute_pipeline(
state: &mut OpState,
#[smi] device_rid: ResourceId,
#[string] label: Cow<str>,
#[serde] layout: GPUPipelineLayoutOrGPUAutoLayoutMode,
#[serde] compute: GpuProgrammableStage,
) -> Result<WebGpuResult, AnyError> {
let instance = state.borrow::<super::Instance>();
let device_resource = state
.resource_table
.get::<super::WebGpuDevice>(device_rid)?;
let device = device_resource.1;
let pipeline_layout = match layout {
GPUPipelineLayoutOrGPUAutoLayoutMode::Layout(rid) => {
let id = state.resource_table.get::<WebGpuPipelineLayout>(rid)?;
Some(id.1)
}
GPUPipelineLayoutOrGPUAutoLayoutMode::Auto(GPUAutoLayoutMode::Auto) => None,
};
let compute_shader_module_resource = state
.resource_table
.get::<super::shader::WebGpuShaderModule>(compute.module)?;
let descriptor = wgpu_core::pipeline::ComputePipelineDescriptor {
label: Some(label),
layout: pipeline_layout,
stage: wgpu_core::pipeline::ProgrammableStageDescriptor {
module: compute_shader_module_resource.1,
entry_point: compute.entry_point.map(Cow::from),
constants: Cow::Owned(compute.constants.unwrap_or_default()),
zero_initialize_workgroup_memory: true,
vertex_pulling_transform: false,
},
cache: None,
};
let implicit_pipelines = match layout {
GPUPipelineLayoutOrGPUAutoLayoutMode::Layout(_) => None,
GPUPipelineLayoutOrGPUAutoLayoutMode::Auto(GPUAutoLayoutMode::Auto) => {
Some(wgpu_core::device::ImplicitPipelineIds {
root_id: None,
group_ids: &[None; MAX_BIND_GROUPS],
})
}
};
let (compute_pipeline, maybe_err) = gfx_select!(device => instance.device_create_compute_pipeline(
device,
&descriptor,
None,
implicit_pipelines
));
let rid = state
.resource_table
.add(WebGpuComputePipeline(instance.clone(), compute_pipeline));
Ok(WebGpuResult::rid_err(rid, maybe_err))
}
#[derive(Serialize)]
#[serde(rename_all = "camelCase")]
pub struct PipelineLayout {
rid: ResourceId,
label: String,
err: Option<WebGpuError>,
}
#[op2]
#[serde]
pub fn op_webgpu_compute_pipeline_get_bind_group_layout(
state: &mut OpState,
#[smi] compute_pipeline_rid: ResourceId,
index: u32,
) -> Result<PipelineLayout, AnyError> {
let instance = state.borrow::<super::Instance>();
let compute_pipeline_resource = state
.resource_table
.get::<WebGpuComputePipeline>(compute_pipeline_rid)?;
let compute_pipeline = compute_pipeline_resource.1;
let (bind_group_layout, maybe_err) = gfx_select!(compute_pipeline => instance.compute_pipeline_get_bind_group_layout(compute_pipeline, index, None));
let label =
gfx_select!(bind_group_layout => instance.bind_group_layout_label(bind_group_layout));
let rid = state
.resource_table
.add(super::binding::WebGpuBindGroupLayout(
instance.clone(),
bind_group_layout,
));
Ok(PipelineLayout {
rid,
label,
err: maybe_err.map(WebGpuError::from),
})
}
#[derive(Deserialize)]
#[serde(rename_all = "kebab-case")]
pub enum GpuCullMode {
None,
Front,
Back,
}
impl From<GpuCullMode> for Option<wgpu_types::Face> {
fn from(value: GpuCullMode) -> Option<wgpu_types::Face> {
match value {
GpuCullMode::None => None,
GpuCullMode::Front => Some(wgpu_types::Face::Front),
GpuCullMode::Back => Some(wgpu_types::Face::Back),
}
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuPrimitiveState {
topology: wgpu_types::PrimitiveTopology,
strip_index_format: Option<wgpu_types::IndexFormat>,
front_face: wgpu_types::FrontFace,
cull_mode: GpuCullMode,
unclipped_depth: bool,
}
impl From<GpuPrimitiveState> for wgpu_types::PrimitiveState {
fn from(value: GpuPrimitiveState) -> wgpu_types::PrimitiveState {
wgpu_types::PrimitiveState {
topology: value.topology,
strip_index_format: value.strip_index_format,
front_face: value.front_face,
cull_mode: value.cull_mode.into(),
unclipped_depth: value.unclipped_depth,
polygon_mode: Default::default(), // native-only
conservative: false, // native-only
}
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuDepthStencilState {
format: wgpu_types::TextureFormat,
depth_write_enabled: bool,
depth_compare: wgpu_types::CompareFunction,
stencil_front: wgpu_types::StencilFaceState,
stencil_back: wgpu_types::StencilFaceState,
stencil_read_mask: u32,
stencil_write_mask: u32,
depth_bias: i32,
depth_bias_slope_scale: f32,
depth_bias_clamp: f32,
}
impl From<GpuDepthStencilState> for wgpu_types::DepthStencilState {
fn from(state: GpuDepthStencilState) -> wgpu_types::DepthStencilState {
wgpu_types::DepthStencilState {
format: state.format,
depth_write_enabled: state.depth_write_enabled,
depth_compare: state.depth_compare,
stencil: wgpu_types::StencilState {
front: state.stencil_front,
back: state.stencil_back,
read_mask: state.stencil_read_mask,
write_mask: state.stencil_write_mask,
},
bias: wgpu_types::DepthBiasState {
constant: state.depth_bias,
slope_scale: state.depth_bias_slope_scale,
clamp: state.depth_bias_clamp,
},
}
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuVertexBufferLayout {
array_stride: u64,
step_mode: wgpu_types::VertexStepMode,
attributes: Vec<wgpu_types::VertexAttribute>,
}
impl<'a> From<GpuVertexBufferLayout> for wgpu_core::pipeline::VertexBufferLayout<'a> {
fn from(layout: GpuVertexBufferLayout) -> wgpu_core::pipeline::VertexBufferLayout<'a> {
wgpu_core::pipeline::VertexBufferLayout {
array_stride: layout.array_stride,
step_mode: layout.step_mode,
attributes: Cow::Owned(layout.attributes),
}
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuVertexState {
module: ResourceId,
entry_point: Option<String>,
constants: Option<HashMap<String, f64>>,
buffers: Vec<Option<GpuVertexBufferLayout>>,
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuMultisampleState {
count: u32,
mask: u64,
alpha_to_coverage_enabled: bool,
}
impl From<GpuMultisampleState> for wgpu_types::MultisampleState {
fn from(gms: GpuMultisampleState) -> wgpu_types::MultisampleState {
wgpu_types::MultisampleState {
count: gms.count,
mask: gms.mask,
alpha_to_coverage_enabled: gms.alpha_to_coverage_enabled,
}
}
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
struct GpuFragmentState {
targets: Vec<Option<wgpu_types::ColorTargetState>>,
module: u32,
entry_point: Option<String>,
constants: Option<HashMap<String, f64>>,
}
#[derive(Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct CreateRenderPipelineArgs {
device_rid: ResourceId,
label: String,
layout: GPUPipelineLayoutOrGPUAutoLayoutMode,
vertex: GpuVertexState,
primitive: GpuPrimitiveState,
depth_stencil: Option<GpuDepthStencilState>,
multisample: wgpu_types::MultisampleState,
fragment: Option<GpuFragmentState>,
}
#[op2]
#[serde]
pub fn op_webgpu_create_render_pipeline(
state: &mut OpState,
#[serde] args: CreateRenderPipelineArgs,
) -> Result<WebGpuResult, AnyError> {
let instance = state.borrow::<super::Instance>();
let device_resource = state
.resource_table
.get::<super::WebGpuDevice>(args.device_rid)?;
let device = device_resource.1;
let layout = match args.layout {
GPUPipelineLayoutOrGPUAutoLayoutMode::Layout(rid) => {
let pipeline_layout_resource = state.resource_table.get::<WebGpuPipelineLayout>(rid)?;
Some(pipeline_layout_resource.1)
}
GPUPipelineLayoutOrGPUAutoLayoutMode::Auto(GPUAutoLayoutMode::Auto) => None,
};
let vertex_shader_module_resource = state
.resource_table
.get::<super::shader::WebGpuShaderModule>(args.vertex.module)?;
let fragment = if let Some(fragment) = args.fragment {
let fragment_shader_module_resource =
state
.resource_table
.get::<super::shader::WebGpuShaderModule>(fragment.module)?;
Some(wgpu_core::pipeline::FragmentState {
stage: wgpu_core::pipeline::ProgrammableStageDescriptor {
module: fragment_shader_module_resource.1,
entry_point: fragment.entry_point.map(Cow::from),
constants: Cow::Owned(fragment.constants.unwrap_or_default()),
// Required to be true for WebGPU
zero_initialize_workgroup_memory: true,
vertex_pulling_transform: false,
},
targets: Cow::Owned(fragment.targets),
})
} else {
None
};
let vertex_buffers = args
.vertex
.buffers
.into_iter()
.flatten()
.map(Into::into)
.collect();
let descriptor = wgpu_core::pipeline::RenderPipelineDescriptor {
label: Some(Cow::Owned(args.label)),
layout,
vertex: wgpu_core::pipeline::VertexState {
stage: wgpu_core::pipeline::ProgrammableStageDescriptor {
module: vertex_shader_module_resource.1,
entry_point: args.vertex.entry_point.map(Cow::Owned),
constants: Cow::Owned(args.vertex.constants.unwrap_or_default()),
// Required to be true for WebGPU
zero_initialize_workgroup_memory: true,
vertex_pulling_transform: false,
},
buffers: Cow::Owned(vertex_buffers),
},
primitive: args.primitive.into(),
depth_stencil: args.depth_stencil.map(Into::into),
multisample: args.multisample,
fragment,
multiview: None,
cache: None,
};
let implicit_pipelines = match args.layout {
GPUPipelineLayoutOrGPUAutoLayoutMode::Layout(_) => None,
GPUPipelineLayoutOrGPUAutoLayoutMode::Auto(GPUAutoLayoutMode::Auto) => {
Some(wgpu_core::device::ImplicitPipelineIds {
root_id: None,
group_ids: &[None; MAX_BIND_GROUPS],
})
}
};
let (render_pipeline, maybe_err) = gfx_select!(device => instance.device_create_render_pipeline(
device,
&descriptor,
None,
implicit_pipelines
));
let rid = state
.resource_table
.add(WebGpuRenderPipeline(instance.clone(), render_pipeline));
Ok(WebGpuResult::rid_err(rid, maybe_err))
}
#[op2]
#[serde]
pub fn op_webgpu_render_pipeline_get_bind_group_layout(
state: &mut OpState,
#[smi] render_pipeline_rid: ResourceId,
index: u32,
) -> Result<PipelineLayout, AnyError> {
let instance = state.borrow::<super::Instance>();
let render_pipeline_resource = state
.resource_table
.get::<WebGpuRenderPipeline>(render_pipeline_rid)?;
let render_pipeline = render_pipeline_resource.1;
let (bind_group_layout, maybe_err) = gfx_select!(render_pipeline => instance.render_pipeline_get_bind_group_layout(render_pipeline, index, None));
let label =
gfx_select!(bind_group_layout => instance.bind_group_layout_label(bind_group_layout));
let rid = state
.resource_table
.add(super::binding::WebGpuBindGroupLayout(
instance.clone(),
bind_group_layout,
));
Ok(PipelineLayout {
rid,
label,
err: maybe_err.map(WebGpuError::from),
})
}