nordic-dev.net/wgpu
nordic-dev.net/wgpu
2
0
Fork 0
mirror of https://github.com/gfx-rs/wgpu.git synced 2024-11-21 22:33:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Tracking Optimization and Rewrite (#2662)

Browse Source 
* Texture state full is dead code

* temp

* temp2 - next is lifetime only

* temp3 - all trackers in place

* temp4 - continue muddling along

* temp5 - doodle doodle

* temp6 - continuous progress is continuous

* temp7 - pro-gress

* temp8 - Add refcounts to trackers

* temp9 - Soundness, generics, refcounts, and epochs

* temp10 - Compiling?!1?!1?!1

* temp11 - Add bind group state optimization

* temp12 - Safety and adding size setting

* temp13 - unsafe

* temp14 - Abstract all the state transitions

* temp15 - It verks! kinda...

* temp16 - it more verks!

* temp17 - debugging and unit tests

* temp18 - fixing unfixed things

* temp19 - hmmm

* temp20 - curious about line deltas

* temp21 - working texture trackers

* temp22 - write merge/update/barrier for buffers

* temp23 - cleanup and buffers

* temp24 - clippy cleanup

* temp25 - Add inline markers on buffer functions

* temp26 - Fix buffer trackers

* temp27 - fixed texture insert to handle both sides

* temp28 - document tracker and usages

* temp29 - document track/mod.rs

* temp30 - convert STORAGE_WRITE to STORAGE_READ_WRITE

* temp31 - Add some debug asserts to make sure we can't insert invalid states

* temp32 - clippy is opinionated sometimes

* temp33 - renaming and documentation

* temp34 - logging
This commit is contained in:
Connor Fitzgerald 2022-05-24 11:36:13 -04:00 committed by GitHub
parent dd6febe309
commit 9114283707
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
38 changed files with 4115 additions and 2369 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

555
Cargo.lock generated
View File

File diff suppressed because it is too large Load Diff

1
wgpu-core/Cargo.toml
View File

@ -26,6 +26,7 @@ vulkan-portability = ["hal/vulkan"]
[dependencies]
arrayvec = "0.7"
bitflags = "1.0"
bit-vec = "0.6"
codespan-reporting = "0.11"
copyless = "0.1"
fxhash = "0.2"

28
wgpu-core/src/binding_model.rs
View File

@ -1,10 +1,10 @@
use crate::{
device::{DeviceError, MissingDownlevelFlags, MissingFeatures, SHADER_STAGE_COUNT},
error::{ErrorFormatter, PrettyError},
hub::Resource,
id::{BindGroupLayoutId, BufferId, DeviceId, SamplerId, TextureViewId, Valid},
hub::{HalApi, Resource},
id::{BindGroupLayoutId, BufferId, DeviceId, SamplerId, TextureId, TextureViewId, Valid},
init_tracker::{BufferInitTrackerAction, TextureInitTrackerAction},
track::{TrackerSet, UsageConflict, DUMMY_SELECTOR},
track::{BindGroupStates, UsageConflict},
validation::{MissingBufferUsageError, MissingTextureUsageError},
FastHashMap, Label, LifeGuard, MultiRefCount, Stored,
};
@ -16,10 +16,7 @@ use serde::Deserialize;
#[cfg(feature = "trace")]
use serde::Serialize;
use std::{
borrow::{Borrow, Cow},
ops::Range,
};
use std::{borrow::Cow, ops::Range};
use thiserror::Error;
@ -63,6 +60,8 @@ pub enum CreateBindGroupError {
InvalidBuffer(BufferId),
#[error("texture view {0:?} is invalid")]
InvalidTextureView(TextureViewId),
#[error("texture {0:?} is invalid")]
InvalidTexture(TextureId),
#[error("sampler {0:?} is invalid")]
InvalidSampler(SamplerId),
#[error(
@ -709,13 +708,12 @@ pub(crate) fn buffer_binding_type_alignment(
}
}
#[derive(Debug)]
pub struct BindGroup<A: hal::Api> {
pub struct BindGroup<A: HalApi> {
pub(crate) raw: A::BindGroup,
pub(crate) device_id: Stored<DeviceId>,
pub(crate) layout_id: Valid<BindGroupLayoutId>,
pub(crate) life_guard: LifeGuard,
pub(crate) used: TrackerSet,
pub(crate) used: BindGroupStates<A>,
pub(crate) used_buffer_ranges: Vec<BufferInitTrackerAction>,
pub(crate) used_texture_ranges: Vec<TextureInitTrackerAction>,
pub(crate) dynamic_binding_info: Vec<BindGroupDynamicBindingData>,
@ -724,7 +722,7 @@ pub struct BindGroup<A: hal::Api> {
pub(crate) late_buffer_binding_sizes: Vec<wgt::BufferSize>,
}
impl<A: hal::Api> BindGroup<A> {
impl<A: HalApi> BindGroup<A> {
pub(crate) fn validate_dynamic_bindings(
&self,
offsets: &[wgt::DynamicOffset],
@ -766,13 +764,7 @@ impl<A: hal::Api> BindGroup<A> {
}
}
impl<A: hal::Api> Borrow<()> for BindGroup<A> {
fn borrow(&self) -> &() {
&DUMMY_SELECTOR
}
}
impl<A: hal::Api> Resource for BindGroup<A> {
impl<A: HalApi> Resource for BindGroup<A> {
const TYPE: &'static str = "BindGroup";
fn life_guard(&self) -> &LifeGuard {

96
wgpu-core/src/command/bundle.rs
View File

@ -34,7 +34,7 @@ invalidations or index format changes.
#![allow(clippy::reversed_empty_ranges)]
use crate::{
binding_model::buffer_binding_type_alignment,
binding_model::{self, buffer_binding_type_alignment},
command::{
BasePass, BindGroupStateChange, DrawError, MapPassErr, PassErrorScope, RenderCommand,
RenderCommandError, StateChange,
@ -48,8 +48,9 @@ use crate::{
hub::{GlobalIdentityHandlerFactory, HalApi, Hub, Resource, Storage, Token},
id,
init_tracker::{BufferInitTrackerAction, MemoryInitKind, TextureInitTrackerAction},
pipeline::PipelineFlags,
track::{TrackerSet, UsageConflict},
pipeline::{self, PipelineFlags},
resource,
track::RenderBundleScope,
validation::check_buffer_usage,
Label, LabelHelpers, LifeGuard, Stored,
};
@ -117,7 +118,7 @@ impl RenderBundleEncoder {
},
sample_count: {
let sc = desc.sample_count;
if sc == 0 || sc > 32 || !conv::is_power_of_two(sc) {
if sc == 0 || sc > 32 || !conv::is_power_of_two_u32(sc) {
return Err(CreateRenderBundleError::InvalidSampleCount(sc));
}
sc
@ -177,20 +178,28 @@ impl RenderBundleEncoder {
/// and accumulate buffer and texture initialization actions.
///
/// [`ExecuteBundle`]: RenderCommand::ExecuteBundle
pub(crate) fn finish<A: hal::Api, G: GlobalIdentityHandlerFactory>(
pub(crate) fn finish<A: HalApi, G: GlobalIdentityHandlerFactory>(
self,
desc: &RenderBundleDescriptor,
device: &Device<A>,
hub: &Hub<A, G>,
token: &mut Token<Device<A>>,
) -> Result<RenderBundle, RenderBundleError> {
) -> Result<RenderBundle<A>, RenderBundleError> {
let (pipeline_layout_guard, mut token) = hub.pipeline_layouts.read(token);
let (bind_group_guard, mut token) = hub.bind_groups.read(&mut token);
let (pipeline_guard, mut token) = hub.render_pipelines.read(&mut token);
let (buffer_guard, _) = hub.buffers.read(&mut token);
let (query_set_guard, mut token) = hub.query_sets.read(&mut token);
let (buffer_guard, mut token) = hub.buffers.read(&mut token);
let (texture_guard, _) = hub.textures.read(&mut token);
let mut state = State {
trackers: TrackerSet::new(self.parent_id.backend()),
trackers: RenderBundleScope::new(
&*buffer_guard,
&*texture_guard,
&*bind_group_guard,
&*pipeline_guard,
&*query_set_guard,
),
index: IndexState::new(),
vertex: (0..hal::MAX_VERTEX_BUFFERS)
.map(|_| VertexState::new())
@ -234,11 +243,11 @@ impl RenderBundleEncoder {
next_dynamic_offset = offsets_range.end;
let offsets = &base.dynamic_offsets[offsets_range.clone()];
let bind_group = state
let bind_group: &binding_model::BindGroup<A> = state
.trackers
.bind_groups
.use_extend(&*bind_group_guard, bind_group_id, (), ())
.map_err(|_| RenderCommandError::InvalidBindGroup(bind_group_id))
.add_single(&*bind_group_guard, bind_group_id)
.ok_or(RenderCommandError::InvalidBindGroup(bind_group_id))
.map_pass_err(scope)?;
if bind_group.dynamic_binding_info.len() != offsets.len() {
return Err(RenderCommandError::InvalidDynamicOffsetCount {
@ -268,10 +277,12 @@ impl RenderBundleEncoder {
texture_memory_init_actions.extend_from_slice(&bind_group.used_texture_ranges);
state.set_bind_group(index, bind_group_id, bind_group.layout_id, offsets_range);
state
.trackers
.merge_extend_stateful(&bind_group.used)
.map_pass_err(scope)?;
unsafe {
state
.trackers
.merge_bind_group(&*texture_guard, &bind_group.used)
.map_pass_err(scope)?
};
//Note: stateless trackers are not merged: the lifetime reference
// is held to the bind group itself.
}
@ -280,11 +291,11 @@ impl RenderBundleEncoder {
state.pipeline = Some(pipeline_id);
let pipeline = state
let pipeline: &pipeline::RenderPipeline<A> = state
.trackers
.render_pipes
.use_extend(&*pipeline_guard, pipeline_id, (), ())
.map_err(|_| RenderCommandError::InvalidPipeline(pipeline_id))
.render_pipelines
.add_single(&*pipeline_guard, pipeline_id)
.ok_or(RenderCommandError::InvalidPipeline(pipeline_id))
.map_pass_err(scope)?;
self.context
@ -320,11 +331,11 @@ impl RenderBundleEncoder {
size,
} => {
let scope = PassErrorScope::SetIndexBuffer(buffer_id);
let buffer = state
let buffer: &resource::Buffer<A> = state
.trackers
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDEX)
.unwrap();
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDEX)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, wgt::BufferUsages::INDEX)
.map_pass_err(scope)?;
@ -347,11 +358,11 @@ impl RenderBundleEncoder {
size,
} => {
let scope = PassErrorScope::SetVertexBuffer(buffer_id);
let buffer = state
let buffer: &resource::Buffer<A> = state
.trackers
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::VERTEX)
.unwrap();
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::VERTEX)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, wgt::BufferUsages::VERTEX)
.map_pass_err(scope)?;
@ -472,11 +483,11 @@ impl RenderBundleEncoder {
.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)
.map_pass_err(scope)?;
let buffer = state
let buffer: &resource::Buffer<A> = state
.trackers
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDIRECT)
.unwrap();
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDIRECT)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, wgt::BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -505,11 +516,10 @@ impl RenderBundleEncoder {
.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)
.map_pass_err(scope)?;
let buffer = state
let buffer: &resource::Buffer<A> = state
.trackers
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDIRECT)
.map_err(|err| RenderCommandError::Buffer(buffer_id, err))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDIRECT)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, wgt::BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -612,24 +622,23 @@ pub type RenderBundleDescriptor<'a> = wgt::RenderBundleDescriptor<Label<'a>>;
//Note: here, `RenderBundle` is just wrapping a raw stream of render commands.
// The plan is to back it by an actual Vulkan secondary buffer, D3D12 Bundle,
// or Metal indirect command buffer.
#[derive(Debug)]
pub struct RenderBundle {
pub struct RenderBundle<A: HalApi> {
// Normalized command stream. It can be executed verbatim,
// without re-binding anything on the pipeline change.
base: BasePass<RenderCommand>,
pub(super) is_ds_read_only: bool,
pub(crate) device_id: Stored<id::DeviceId>,
pub(crate) used: TrackerSet,
pub(crate) used: RenderBundleScope<A>,
pub(super) buffer_memory_init_actions: Vec<BufferInitTrackerAction>,
pub(super) texture_memory_init_actions: Vec<TextureInitTrackerAction>,
pub(super) context: RenderPassContext,
pub(crate) life_guard: LifeGuard,
}
unsafe impl Send for RenderBundle {}
unsafe impl Sync for RenderBundle {}
unsafe impl<A: HalApi> Send for RenderBundle<A> {}
unsafe impl<A: HalApi> Sync for RenderBundle<A> {}
impl RenderBundle {
impl<A: HalApi> RenderBundle<A> {
/// Actually encode the contents into a native command buffer.
///
/// This is partially duplicating the logic of `command_encoder_run_render_pass`.
@ -639,7 +648,7 @@ impl RenderBundle {
/// Note that the function isn't expected to fail, generally.
/// All the validation has already been done by this point.
/// The only failure condition is if some of the used buffers are destroyed.
pub(super) unsafe fn execute<A: HalApi>(
pub(super) unsafe fn execute(
&self,
raw: &mut A::CommandEncoder,
pipeline_layout_guard: &Storage<
@ -828,7 +837,7 @@ impl RenderBundle {
}
}
impl Resource for RenderBundle {
impl<A: HalApi> Resource for RenderBundle<A> {
const TYPE: &'static str = "RenderBundle";
fn life_guard(&self) -> &LifeGuard {
@ -1027,10 +1036,9 @@ struct VertexLimitState {
/// [`SetIndexBuffer`] to the simulated state stored here, and then
/// calls the `flush_foo` methods before draw calls to produce the
/// update commands we actually need.
#[derive(Debug)]
struct State {
struct State<A: HalApi> {
/// Resources used by this bundle. This will become [`RenderBundle::used`].
trackers: TrackerSet,
trackers: RenderBundleScope<A>,
/// The current index buffer. We flush this state before indexed
/// draw commands.
@ -1056,7 +1064,7 @@ struct State {
pipeline: Option<id::RenderPipelineId>,
}
impl State {
impl<A: HalApi> State<A> {
fn vertex_limits(&self) -> VertexLimitState {
let mut vert_state = VertexLimitState {
vertex_limit: u32::MAX,
@ -1235,8 +1243,6 @@ pub(super) enum RenderBundleErrorInner {
#[error(transparent)]
RenderCommand(RenderCommandError),
#[error(transparent)]
ResourceUsageConflict(#[from] UsageConflict),
#[error(transparent)]
Draw(#[from] DrawError),
#[error(transparent)]
MissingDownlevelFlags(#[from] MissingDownlevelFlags),

68
wgpu-core/src/command/clear.rs
View File

@ -4,13 +4,12 @@ use std::{num::NonZeroU32, ops::Range};
use crate::device::trace::Command as TraceCommand;
use crate::{
command::CommandBuffer,
device::Device,
get_lowest_common_denom,
hub::{Global, GlobalIdentityHandlerFactory, HalApi, Resource, Token},
hub::{self, Global, GlobalIdentityHandlerFactory, HalApi, Token},
id::{BufferId, CommandEncoderId, DeviceId, TextureId, Valid},
init_tracker::{MemoryInitKind, TextureInitRange},
resource::{Texture, TextureClearMode},
track::{ResourceTracker, TextureSelector, TextureState},
track::{TextureSelector, TextureTracker},
};
use hal::{auxil::align_to, CommandEncoder as _};
@ -90,8 +89,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (dst_buffer, dst_pending) = cmd_buf
.trackers
.buffers
.use_replace(&*buffer_guard, dst, (), hal::BufferUses::COPY_DST)
.map_err(ClearError::InvalidBuffer)?;
.set_single(&*buffer_guard, dst, hal::BufferUses::COPY_DST)
.ok_or(ClearError::InvalidBuffer(dst))?;
let dst_raw = dst_buffer
.raw
.as_ref()
@ -139,7 +138,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let dst_barrier = dst_pending.map(|pending| pending.into_hal(dst_buffer));
let cmd_buf_raw = cmd_buf.encoder.open();
unsafe {
cmd_buf_raw.transition_buffers(dst_barrier);
cmd_buf_raw.transition_buffers(dst_barrier.into_iter());
cmd_buf_raw.clear_buffer(dst_raw, offset..end);
}
Ok(())
@ -191,13 +190,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
// Check if subresource level range is valid
let subresource_level_end = match subresource_range.mip_level_count {
Some(count) => subresource_range.base_mip_level + count.get(),
None => dst_texture.full_range.levels.end,
None => dst_texture.full_range.mips.end,
};
if dst_texture.full_range.levels.start > subresource_range.base_mip_level
|| dst_texture.full_range.levels.end < subresource_level_end
if dst_texture.full_range.mips.start > subresource_range.base_mip_level
|| dst_texture.full_range.mips.end < subresource_level_end
{
return Err(ClearError::InvalidTextureLevelRange {
texture_level_range: dst_texture.full_range.levels.clone(),
texture_level_range: dst_texture.full_range.mips.clone(),
subresource_base_mip_level: subresource_range.base_mip_level,
subresource_mip_level_count: subresource_range.mip_level_count,
});
@ -217,48 +216,34 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
});
}
let device = &device_guard[cmd_buf.device_id.value];
clear_texture(
&*texture_guard,
Valid(dst),
dst_texture,
TextureInitRange {
mip_range: subresource_range.base_mip_level..subresource_level_end,
layer_range: subresource_range.base_array_layer..subresource_layer_end,
},
cmd_buf.encoder.open(),
&mut cmd_buf.trackers.textures,
&device_guard[cmd_buf.device_id.value],
&device.alignments,
&device.zero_buffer,
)
}
}
pub(crate) fn clear_texture<A: hal::Api>(
pub(crate) fn clear_texture<A: HalApi>(
storage: &hub::Storage<Texture<A>, TextureId>,
dst_texture_id: Valid<TextureId>,
dst_texture: &Texture<A>,
range: TextureInitRange,
encoder: &mut A::CommandEncoder,
texture_tracker: &mut ResourceTracker<TextureState>,
device: &Device<A>,
) -> Result<(), ClearError> {
clear_texture_no_device(
dst_texture_id,
dst_texture,
range,
encoder,
texture_tracker,
&device.alignments,
&device.zero_buffer,
)
}
pub(crate) fn clear_texture_no_device<A: hal::Api>(
dst_texture_id: Valid<TextureId>,
dst_texture: &Texture<A>,
range: TextureInitRange,
encoder: &mut A::CommandEncoder,
texture_tracker: &mut ResourceTracker<TextureState>,
texture_tracker: &mut TextureTracker<A>,
alignments: &hal::Alignments,
zero_buffer: &A::Buffer,
) -> Result<(), ClearError> {
let dst_texture = &storage[dst_texture_id];
let dst_raw = dst_texture
.inner
.as_raw()
@ -277,7 +262,7 @@ pub(crate) fn clear_texture_no_device<A: hal::Api>(
};
let selector = TextureSelector {
levels: range.mip_range.clone(),
mips: range.mip_range.clone(),
layers: range.layer_range.clone(),
};
@ -287,14 +272,13 @@ pub(crate) fn clear_texture_no_device<A: hal::Api>(
// On the other hand, when coming via command_encoder_clear_texture, the life_guard is still there since in order to call it a texture object is needed.
//
// We could in theory distinguish these two scenarios in the internal clear_texture api in order to remove this check and call the cheaper change_replace_tracked whenever possible.
let dst_barrier = if let Some(ref_count) = dst_texture.life_guard().ref_count.as_ref() {
texture_tracker.change_replace(dst_texture_id, ref_count, selector, clear_usage)
} else {
texture_tracker.change_replace_tracked(dst_texture_id, selector, clear_usage)
}
.map(|pending| pending.into_hal(dst_texture));
let dst_barrier = texture_tracker
.set_single(storage, dst_texture_id.0, selector, clear_usage)
.unwrap()
.1
.map(|pending| pending.into_hal(dst_texture));
unsafe {
encoder.transition_textures(dst_barrier);
encoder.transition_textures(dst_barrier.into_iter());
}
// Record actual clearing

101
wgpu-core/src/command/compute.rs
View File

@ -14,8 +14,9 @@ use crate::{
hub::{Global, GlobalIdentityHandlerFactory, HalApi, Storage, Token},
id,
init_tracker::MemoryInitKind,
resource::{Buffer, Texture},
track::{StatefulTrackerSubset, TrackerSet, UsageConflict, UseExtendError},
pipeline,
resource::{self, Buffer, Texture},
track::{Tracker, UsageConflict, UsageScope},
validation::{check_buffer_usage, MissingBufferUsageError},
Label,
};
@ -228,15 +229,14 @@ where
}
}
#[derive(Debug)]
struct State {
struct State<A: HalApi> {
binder: Binder,
pipeline: Option<id::ComputePipelineId>,
trackers: StatefulTrackerSubset,
scope: UsageScope<A>,
debug_scope_depth: u32,
}
impl State {
impl<A: HalApi> State<A> {
fn is_ready(&self) -> Result<(), DispatchError> {
let bind_mask = self.binder.invalid_mask();
if bind_mask != 0 {
@ -253,32 +253,36 @@ impl State {
Ok(())
}
fn flush_states<A: HalApi>(
fn flush_states(
&mut self,
raw_encoder: &mut A::CommandEncoder,
base_trackers: &mut TrackerSet,
base_trackers: &mut Tracker<A>,
bind_group_guard: &Storage<BindGroup<A>, id::BindGroupId>,
buffer_guard: &Storage<Buffer<A>, id::BufferId>,
texture_guard: &Storage<Texture<A>, id::TextureId>,
) -> Result<(), UsageConflict> {
for id in self.binder.list_active() {
self.trackers.merge_extend(&bind_group_guard[id].used)?;
//Note: stateless trackers are not merged: the lifetime reference
unsafe {
self.scope
.merge_bind_group(texture_guard, &bind_group_guard[id].used)?
};
// Note: stateless trackers are not merged: the lifetime reference
// is held to the bind group itself.
}
for id in self.binder.list_active() {
unsafe {
base_trackers.set_and_remove_from_usage_scope_sparse(
texture_guard,
&mut self.scope,
&bind_group_guard[id].used,
)
}
}
log::trace!("Encoding dispatch barriers");
CommandBuffer::insert_barriers(
raw_encoder,
base_trackers,
&self.trackers.buffers,
&self.trackers.textures,
buffer_guard,
texture_guard,
);
self.trackers.clear();
CommandBuffer::drain_barriers(raw_encoder, base_trackers, buffer_guard, texture_guard);
Ok(())
}
}
@ -338,7 +342,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let mut state = State {
binder: Binder::new(),
pipeline: None,
trackers: StatefulTrackerSubset::new(A::VARIANT),
scope: UsageScope::new(&*buffer_guard, &*texture_guard),
debug_scope_depth: 0,
};
let mut temp_offsets = Vec::new();
@ -346,6 +350,18 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let mut string_offset = 0;
let mut active_query = None;
cmd_buf.trackers.set_size(
Some(&*buffer_guard),
Some(&*texture_guard),
None,
None,
Some(&*bind_group_guard),
Some(&*pipeline_guard),
None,
None,
Some(&*query_set_guard),
);
let hal_desc = hal::ComputePassDescriptor { label: base.label };
unsafe {
raw.begin_compute_pass(&hal_desc);
@ -379,11 +395,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
);
dynamic_offset_count += num_dynamic_offsets as usize;
let bind_group = cmd_buf
let bind_group: &BindGroup<A> = cmd_buf
.trackers
.bind_groups
.use_extend(&*bind_group_guard, bind_group_id, (), ())
.map_err(|_| ComputePassErrorInner::InvalidBindGroup(bind_group_id))
.add_single(&*bind_group_guard, bind_group_id)
.ok_or(ComputePassErrorInner::InvalidBindGroup(bind_group_id))
.map_pass_err(scope)?;
bind_group
.validate_dynamic_bindings(&temp_offsets, &cmd_buf.limits)
@ -434,11 +450,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
state.pipeline = Some(pipeline_id);
let pipeline = cmd_buf
let pipeline: &pipeline::ComputePipeline<A> = cmd_buf
.trackers
.compute_pipes
.use_extend(&*pipeline_guard, pipeline_id, (), ())
.map_err(|_| ComputePassErrorInner::InvalidPipeline(pipeline_id))
.compute_pipelines
.add_single(&*pipeline_guard, pipeline_id)
.ok_or(ComputePassErrorInner::InvalidPipeline(pipeline_id))
.map_pass_err(scope)?;
unsafe {
@ -587,11 +603,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)
.map_pass_err(scope)?;
let indirect_buffer = state
.trackers
let indirect_buffer: &Buffer<A> = state
.scope
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDIRECT)
.map_err(|_| ComputePassErrorInner::InvalidIndirectBuffer(buffer_id))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDIRECT)
.map_pass_err(scope)?;
check_buffer_usage(indirect_buffer.usage, wgt::BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -670,16 +685,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
} => {
let scope = PassErrorScope::WriteTimestamp;
let query_set = cmd_buf
let query_set: &resource::QuerySet<A> = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => {
ComputePassErrorInner::InvalidQuerySet(query_set_id)
}
_ => unreachable!(),
})
.add_single(&*query_set_guard, query_set_id)
.ok_or(ComputePassErrorInner::InvalidQuerySet(query_set_id))
.map_pass_err(scope)?;
query_set
@ -692,16 +702,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
} => {
let scope = PassErrorScope::BeginPipelineStatisticsQuery;
let query_set = cmd_buf
let query_set: &resource::QuerySet<A> = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => {
ComputePassErrorInner::InvalidQuerySet(query_set_id)
}
_ => unreachable!(),
})
.add_single(&*query_set_guard, query_set_id)
.ok_or(ComputePassErrorInner::InvalidQuerySet(query_set_id))
.map_pass_err(scope)?;
query_set

14
wgpu-core/src/command/draw.rs
View File

@ -5,7 +5,7 @@ use crate::{
binding_model::{LateMinBufferBindingSizeMismatch, PushConstantUploadError},
error::ErrorFormatter,
id,
track::UseExtendError,
track::UsageConflict,
validation::{MissingBufferUsageError, MissingTextureUsageError},
};
use wgt::{BufferAddress, BufferSize, Color};
@ -13,8 +13,6 @@ use wgt::{BufferAddress, BufferSize, Color};
use std::num::NonZeroU32;
use thiserror::Error;
pub type BufferError = UseExtendError<hal::BufferUses>;
/// Error validating a draw call.
#[derive(Clone, Debug, Error, PartialEq)]
pub enum DrawError {
@ -79,8 +77,8 @@ pub enum RenderCommandError {
IncompatiblePipelineTargets(#[from] crate::device::RenderPassCompatibilityError),
#[error("pipeline writes to depth/stencil, while the pass has read-only depth/stencil")]
IncompatiblePipelineRods,
#[error("buffer {0:?} is in error {1:?}")]
Buffer(id::BufferId, BufferError),
#[error(transparent)]
UsageConflict(#[from] UsageConflict),
#[error("buffer {0:?} is destroyed")]
DestroyedBuffer(id::BufferId),
#[error(transparent)]
@ -106,7 +104,11 @@ impl crate::error::PrettyError for RenderCommandError {
Self::InvalidPipeline(id) => {
fmt.render_pipeline_label(&id);
}
Self::Buffer(id, ..) | Self::DestroyedBuffer(id) => {
Self::UsageConflict(UsageConflict::TextureInvalid { id }) => {
fmt.texture_label(&id);
}
Self::UsageConflict(UsageConflict::BufferInvalid { id })
| Self::DestroyedBuffer(id) => {
fmt.buffer_label(&id);
}
_ => {}

41
wgpu-core/src/command/memory_init.rs
View File

@ -4,11 +4,11 @@ use hal::CommandEncoder;
use crate::{
device::Device,
hub::Storage,
hub::{HalApi, Storage},
id::{self, TextureId},
init_tracker::*,
resource::{Buffer, Texture},
track::{ResourceTracker, TextureState, TrackerSet},
track::{TextureTracker, Tracker},
FastHashMap,
};
@ -121,36 +121,37 @@ impl CommandBufferTextureMemoryActions {
// Utility function that takes discarded surfaces from (several calls to) register_init_action and initializes them on the spot.
// Takes care of barriers as well!
pub(crate) fn fixup_discarded_surfaces<
A: hal::Api,
A: HalApi,
InitIter: Iterator<Item = TextureSurfaceDiscard>,
>(
inits: InitIter,
encoder: &mut A::CommandEncoder,
texture_guard: &Storage<Texture<A>, TextureId>,
texture_tracker: &mut ResourceTracker<TextureState>,
texture_tracker: &mut TextureTracker<A>,
device: &Device<A>,
) {
for init in inits {
clear_texture(
texture_guard,
id::Valid(init.texture),
texture_guard.get(init.texture).unwrap(),
TextureInitRange {
mip_range: init.mip_level..(init.mip_level + 1),
layer_range: init.layer..(init.layer + 1),
},
encoder,
texture_tracker,
device,
&device.alignments,
&device.zero_buffer,
)
.unwrap();
}
}
impl<A: hal::Api> BakedCommands<A> {
impl<A: HalApi> BakedCommands<A> {
// inserts all buffer initializations that are going to be needed for executing the commands and updates resource init states accordingly
pub(crate) fn initialize_buffer_memory(
&mut self,
device_tracker: &mut TrackerSet,
device_tracker: &mut Tracker<A>,
buffer_guard: &mut Storage<Buffer<A>, id::BufferId>,
) -> Result<(), DestroyedBufferError> {
// Gather init ranges for each buffer so we can collapse them.
@ -202,11 +203,11 @@ impl<A: hal::Api> BakedCommands<A> {
// Don't do use_replace since the buffer may already no longer have a ref_count.
// However, we *know* that it is currently in use, so the tracker must already know about it.
let transition = device_tracker.buffers.change_replace_tracked(
id::Valid(buffer_id),
(),
hal::BufferUses::COPY_DST,
);
let transition = device_tracker
.buffers
.set_single(buffer_guard, buffer_id, hal::BufferUses::COPY_DST)
.unwrap()
.1;
let buffer = buffer_guard
.get_mut(buffer_id)
@ -214,8 +215,11 @@ impl<A: hal::Api> BakedCommands<A> {
let raw_buf = buffer.raw.as_ref().ok_or(DestroyedBufferError(buffer_id))?;
unsafe {
self.encoder
.transition_buffers(transition.map(|pending| pending.into_hal(buffer)));
self.encoder.transition_buffers(
transition
.map(|pending| pending.into_hal(buffer))
.into_iter(),
);
}
for range in ranges.iter() {
@ -234,7 +238,7 @@ impl<A: hal::Api> BakedCommands<A> {
// any textures that are left discarded by this command buffer will be marked as uninitialized
pub(crate) fn initialize_texture_memory(
&mut self,
device_tracker: &mut TrackerSet,
device_tracker: &mut Tracker<A>,
texture_guard: &mut Storage<Texture<A>, TextureId>,
device: &Device<A>,
) -> Result<(), DestroyedTextureError> {
@ -274,12 +278,13 @@ impl<A: hal::Api> BakedCommands<A> {
// TODO: Could we attempt some range collapsing here?
for range in ranges.drain(..) {
clear_texture(
texture_guard,
id::Valid(texture_use.id),
&*texture,
range,
&mut self.encoder,
&mut device_tracker.textures,
device,
&device.alignments,
&device.zero_buffer,
)
.unwrap();
}

67
wgpu-core/src/command/mod.rs
View File

@ -10,7 +10,7 @@ mod transfer;
use std::slice;
pub(crate) use self::clear::clear_texture_no_device;
pub(crate) use self::clear::clear_texture;
pub use self::{
bundle::*, clear::ClearError, compute::*, draw::*, query::*, render::*, transfer::*,
};
@ -19,11 +19,11 @@ use self::memory_init::CommandBufferTextureMemoryActions;
use crate::error::{ErrorFormatter, PrettyError};
use crate::init_tracker::BufferInitTrackerAction;
use crate::track::{Tracker, UsageScope};
use crate::{
hub::{Global, GlobalIdentityHandlerFactory, HalApi, Storage, Token},
id,
resource::{Buffer, Texture},
track::{BufferState, ResourceTracker, TextureState, TrackerSet},
Label, Stored,
};
@ -81,10 +81,10 @@ impl<A: hal::Api> CommandEncoder<A> {
}
}
pub struct BakedCommands<A: hal::Api> {
pub struct BakedCommands<A: HalApi> {
pub(crate) encoder: A::CommandEncoder,
pub(crate) list: Vec<A::CommandBuffer>,
pub(crate) trackers: TrackerSet,
pub(crate) trackers: Tracker<A>,
buffer_memory_init_actions: Vec<BufferInitTrackerAction>,
texture_memory_actions: CommandBufferTextureMemoryActions,
}
@ -92,11 +92,11 @@ pub struct BakedCommands<A: hal::Api> {
pub(crate) struct DestroyedBufferError(pub id::BufferId);
pub(crate) struct DestroyedTextureError(pub id::TextureId);
pub struct CommandBuffer<A: hal::Api> {
pub struct CommandBuffer<A: HalApi> {
encoder: CommandEncoder<A>,
status: CommandEncoderStatus,
pub(crate) device_id: Stored<id::DeviceId>,
pub(crate) trackers: TrackerSet,
pub(crate) trackers: Tracker<A>,
buffer_memory_init_actions: Vec<BufferInitTrackerAction>,
texture_memory_actions: CommandBufferTextureMemoryActions,
limits: wgt::Limits,
@ -124,7 +124,7 @@ impl<A: HalApi> CommandBuffer<A> {
},
status: CommandEncoderStatus::Recording,
device_id,
trackers: TrackerSet::new(A::VARIANT),
trackers: Tracker::new(),
buffer_memory_init_actions: Default::default(),
texture_memory_actions: Default::default(),
limits,
@ -138,23 +138,52 @@ impl<A: HalApi> CommandBuffer<A> {
}
}
pub(crate) fn insert_barriers(
pub(crate) fn insert_barriers_from_tracker(
raw: &mut A::CommandEncoder,
base: &mut TrackerSet,
head_buffers: &ResourceTracker<BufferState>,
head_textures: &ResourceTracker<TextureState>,
base: &mut Tracker<A>,
head: &Tracker<A>,
buffer_guard: &Storage<Buffer<A>, id::BufferId>,
texture_guard: &Storage<Texture<A>, id::TextureId>,
) {
profiling::scope!("insert_barriers");
debug_assert_eq!(A::VARIANT, base.backend());
let buffer_barriers = base.buffers.merge_replace(head_buffers).map(|pending| {
let buf = &buffer_guard[pending.id];
base.buffers.set_from_tracker(&head.buffers);
base.textures
.set_from_tracker(&*texture_guard, &head.textures);
Self::drain_barriers(raw, base, buffer_guard, texture_guard);
}
pub(crate) fn insert_barriers_from_scope(
raw: &mut A::CommandEncoder,
base: &mut Tracker<A>,
head: &UsageScope<A>,
buffer_guard: &Storage<Buffer<A>, id::BufferId>,
texture_guard: &Storage<Texture<A>, id::TextureId>,
) {
profiling::scope!("insert_barriers");
base.buffers.set_from_usage_scope(&head.buffers);
base.textures
.set_from_usage_scope(&*texture_guard, &head.textures);
Self::drain_barriers(raw, base, buffer_guard, texture_guard);
}
pub(crate) fn drain_barriers(
raw: &mut A::CommandEncoder,
base: &mut Tracker<A>,
buffer_guard: &Storage<Buffer<A>, id::BufferId>,
texture_guard: &Storage<Texture<A>, id::TextureId>,
) {
profiling::scope!("drain_barriers");
let buffer_barriers = base.buffers.drain().map(|pending| {
let buf = unsafe { &buffer_guard.get_unchecked(pending.id) };
pending.into_hal(buf)
});
let texture_barriers = base.textures.merge_replace(head_textures).map(|pending| {
let tex = &texture_guard[pending.id];
let texture_barriers = base.textures.drain().map(|pending| {
let tex = unsafe { texture_guard.get_unchecked(pending.id) };
pending.into_hal(tex)
});
@ -165,7 +194,7 @@ impl<A: HalApi> CommandBuffer<A> {
}
}
impl<A: hal::Api> CommandBuffer<A> {
impl<A: HalApi> CommandBuffer<A> {
fn get_encoder_mut(
storage: &mut Storage<Self, id::CommandEncoderId>,
id: id::CommandEncoderId,
@ -198,7 +227,7 @@ impl<A: hal::Api> CommandBuffer<A> {
}
}
impl<A: hal::Api> crate::hub::Resource for CommandBuffer<A> {
impl<A: HalApi> crate::hub::Resource for CommandBuffer<A> {
const TYPE: &'static str = "CommandBuffer";
fn life_guard(&self) -> &crate::LifeGuard {
@ -321,7 +350,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
cmd_buf.status = CommandEncoderStatus::Finished;
//Note: if we want to stop tracking the swapchain texture view,
// this is the place to do it.
log::trace!("Command buffer {:?} {:#?}", encoder_id, cmd_buf.trackers);
log::trace!("Command buffer {:?}", encoder_id);
None
}
CommandEncoderStatus::Finished => Some(CommandEncoderError::NotRecording),

21
wgpu-core/src/command/query.rs
View File

@ -8,7 +8,6 @@ use crate::{
id::{self, Id, TypedId},
init_tracker::MemoryInitKind,
resource::QuerySet,
track::UseExtendError,
Epoch, FastHashMap, Index,
};
use std::{iter, marker::PhantomData};
@ -300,11 +299,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let query_set = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => QueryError::InvalidQuerySet(query_set_id),
_ => unreachable!(),
})?;
.add_single(&*query_set_guard, query_set_id)
.ok_or(QueryError::InvalidQuerySet(query_set_id))?;
query_set.validate_and_write_timestamp(raw_encoder, query_set_id, query_index, None)?;
@ -348,17 +344,14 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let query_set = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => QueryError::InvalidQuerySet(query_set_id),
_ => unreachable!(),
})?;
.add_single(&*query_set_guard, query_set_id)
.ok_or(QueryError::InvalidQuerySet(query_set_id))?;
let (dst_buffer, dst_pending) = cmd_buf
.trackers
.buffers
.use_replace(&*buffer_guard, destination, (), hal::BufferUses::COPY_DST)
.map_err(QueryError::InvalidBuffer)?;
.set_single(&*buffer_guard, destination, hal::BufferUses::COPY_DST)
.ok_or(QueryError::InvalidBuffer(destination))?;
let dst_barrier = dst_pending.map(|pending| pending.into_hal(dst_buffer));
if !dst_buffer.usage.contains(wgt::BufferUsages::COPY_DST) {
@ -407,7 +400,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
));
unsafe {
raw_encoder.transition_buffers(dst_barrier);
raw_encoder.transition_buffers(dst_barrier.into_iter());
raw_encoder.copy_query_results(
&query_set.raw,
start_query..end_query,

182
wgpu-core/src/command/render.rs
View File

@ -1,6 +1,7 @@
use crate::{
binding_model::BindError,
command::{
self,
bind::Binder,
end_pipeline_statistics_query,
memory_init::{fixup_discarded_surfaces, SurfacesInDiscardState},
@ -17,8 +18,8 @@ use crate::{
id,
init_tracker::{MemoryInitKind, TextureInitRange, TextureInitTrackerAction},
pipeline::PipelineFlags,
resource::{Texture, TextureView},
track::{StatefulTrackerSubset, TextureSelector, UsageConflict},
resource::{self, Buffer, Texture, TextureView},
track::{TextureSelector, UsageConflict, UsageScope},
validation::{
check_buffer_usage, check_texture_usage, MissingBufferUsageError, MissingTextureUsageError,
},
@ -38,7 +39,6 @@ use serde::Deserialize;
#[cfg(any(feature = "serial-pass", feature = "trace"))]
use serde::Serialize;
use crate::track::UseExtendError;
use std::{borrow::Cow, fmt, iter, marker::PhantomData, mem, num::NonZeroU32, ops::Range, str};
use super::{memory_init::TextureSurfaceDiscard, CommandBufferTextureMemoryActions};
@ -561,9 +561,9 @@ impl<A: hal::Api> TextureView<A> {
const MAX_TOTAL_ATTACHMENTS: usize = hal::MAX_COLOR_TARGETS + hal::MAX_COLOR_TARGETS + 1;
type AttachmentDataVec<T> = ArrayVec<T, MAX_TOTAL_ATTACHMENTS>;
struct RenderPassInfo<'a, A: hal::Api> {
struct RenderPassInfo<'a, A: HalApi> {
context: RenderPassContext,
trackers: StatefulTrackerSubset,
usage_scope: UsageScope<A>,
render_attachments: AttachmentDataVec<RenderAttachment<'a>>, // All render attachments, including depth/stencil
is_ds_read_only: bool,
extent: wgt::Extent3d,
@ -605,7 +605,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
// but recording the discard right away be alright since the texture can't be used during the pass anyways
texture_memory_actions.discard(TextureSurfaceDiscard {
texture: view.parent_id.value.0,
mip_level: view.selector.levels.start,
mip_level: view.selector.mips.start,
layer: view.selector.layers.start,
});
}
@ -618,6 +618,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
depth_stencil_attachment: Option<&RenderPassDepthStencilAttachment>,
cmd_buf: &mut CommandBuffer<A>,
view_guard: &'a Storage<TextureView<A>, id::TextureViewId>,
buffer_guard: &'a Storage<Buffer<A>, id::BufferId>,
texture_guard: &'a Storage<Texture<A>, id::TextureId>,
) -> Result<Self, RenderPassErrorInner> {
profiling::scope!("start", "RenderPassInfo");
@ -699,8 +700,8 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
let view = cmd_buf
.trackers
.views
.use_extend(&*view_guard, at.view, (), ())
.map_err(|_| RenderPassErrorInner::InvalidAttachment(at.view))?;
.add_single(&*view_guard, at.view)
.ok_or(RenderPassErrorInner::InvalidAttachment(at.view))?;
check_multiview(view)?;
add_view(view, "depth")?;
@ -779,7 +780,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
// Both are discarded using the regular path.
discarded_surfaces.push(TextureSurfaceDiscard {
texture: view.parent_id.value.0,
mip_level: view.selector.levels.start,
mip_level: view.selector.mips.start,
layer: view.selector.layers.start,
});
}
@ -808,8 +809,8 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
let color_view = cmd_buf
.trackers
.views
.use_extend(&*view_guard, at.view, (), ())
.map_err(|_| RenderPassErrorInner::InvalidAttachment(at.view))?;
.add_single(&*view_guard, at.view)
.ok_or(RenderPassErrorInner::InvalidAttachment(at.view))?;
check_multiview(color_view)?;
add_view(color_view, "color")?;
@ -838,8 +839,8 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
let resolve_view = cmd_buf
.trackers
.views
.use_extend(&*view_guard, resolve_target, (), ())
.map_err(|_| RenderPassErrorInner::InvalidAttachment(resolve_target))?;
.add_single(&*view_guard, resolve_target)
.ok_or(RenderPassErrorInner::InvalidAttachment(resolve_target))?;
check_multiview(resolve_view)?;
if color_view.extent != resolve_view.extent {
@ -934,7 +935,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
Ok(Self {
context,
trackers: StatefulTrackerSubset::new(A::VARIANT),
usage_scope: UsageScope::new(buffer_guard, texture_guard),
render_attachments,
is_ds_read_only,
extent,
@ -949,7 +950,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
mut self,
raw: &mut A::CommandEncoder,
texture_guard: &Storage<Texture<A>, id::TextureId>,
) -> Result<(StatefulTrackerSubset, SurfacesInDiscardState), RenderPassErrorInner> {
) -> Result<(UsageScope<A>, SurfacesInDiscardState), RenderPassErrorInner> {
profiling::scope!("finish", "RenderPassInfo");
unsafe {
raw.end_render_pass();
@ -963,15 +964,18 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
check_texture_usage(texture.desc.usage, TextureUsages::RENDER_ATTACHMENT)?;
// the tracker set of the pass is always in "extend" mode
self.trackers
.textures
.change_extend(
ra.texture_id.value,
&ra.texture_id.ref_count,
ra.selector.clone(),
ra.usage,
)
.map_err(UsageConflict::from)?;
unsafe {
self.usage_scope
.textures
.merge_single(
&*texture_guard,
ra.texture_id.value,
Some(ra.selector.clone()),
&ra.texture_id.ref_count,
ra.usage,
)
.map_err(UsageConflict::from)?
};
}
// If either only stencil or depth was discarded, we put in a special clear pass to keep the init status of the aspects in sync.
@ -1012,7 +1016,7 @@ impl<'a, A: HalApi> RenderPassInfo<'a, A> {
}
}
Ok((self.trackers, self.pending_discard_init_fixups))
Ok((self.usage_scope, self.pending_discard_init_fixups))
}
}
@ -1047,7 +1051,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let mut token = Token::root();
let (device_guard, mut token) = hub.devices.read(&mut token);
let (trackers, query_reset_state, pending_discard_init_fixups) = {
let (scope, query_reset_state, pending_discard_init_fixups) = {
let (mut cmb_guard, mut token) = hub.command_buffers.write(&mut token);
// Spell out the type, to placate rust-analyzer.
@ -1075,7 +1079,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (bundle_guard, mut token) = hub.render_bundles.read(&mut token);
let (pipeline_layout_guard, mut token) = hub.pipeline_layouts.read(&mut token);
let (bind_group_guard, mut token) = hub.bind_groups.read(&mut token);
let (pipeline_guard, mut token) = hub.render_pipelines.read(&mut token);
let (render_pipeline_guard, mut token) = hub.render_pipelines.read(&mut token);
let (query_set_guard, mut token) = hub.query_sets.read(&mut token);
let (buffer_guard, mut token) = hub.buffers.read(&mut token);
let (texture_guard, mut token) = hub.textures.read(&mut token);
@ -1093,10 +1097,23 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
depth_stencil_attachment,
cmd_buf,
&*view_guard,
&*buffer_guard,
&*texture_guard,
)
.map_pass_err(init_scope)?;
cmd_buf.trackers.set_size(
Some(&*buffer_guard),
Some(&*texture_guard),
Some(&*view_guard),
None,
Some(&*bind_group_guard),
None,
Some(&*render_pipeline_guard),
Some(&*bundle_guard),
Some(&*query_set_guard),
);
let raw = &mut cmd_buf.encoder.raw;
let mut state = State {
@ -1142,17 +1159,19 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let bind_group = cmd_buf
.trackers
.bind_groups
.use_extend(&*bind_group_guard, bind_group_id, (), ())
.map_err(|_| RenderCommandError::InvalidBindGroup(bind_group_id))
.add_single(&*bind_group_guard, bind_group_id)
.ok_or(RenderCommandError::InvalidBindGroup(bind_group_id))
.map_pass_err(scope)?;
bind_group
.validate_dynamic_bindings(&temp_offsets, &cmd_buf.limits)
.map_pass_err(scope)?;
// merge the resource tracker in
info.trackers
.merge_extend(&bind_group.used)
.map_pass_err(scope)?;
unsafe {
info.usage_scope
.merge_bind_group(&*texture_guard, &bind_group.used)
.map_pass_err(scope)?;
}
//Note: stateless trackers are not merged: the lifetime reference
// is held to the bind group itself.
@ -1203,9 +1222,9 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let pipeline = cmd_buf
.trackers
.render_pipes
.use_extend(&*pipeline_guard, pipeline_id, (), ())
.map_err(|_| RenderCommandError::InvalidPipeline(pipeline_id))
.render_pipelines
.add_single(&*render_pipeline_guard, pipeline_id)
.ok_or(RenderCommandError::InvalidPipeline(pipeline_id))
.map_pass_err(scope)?;
info.context
@ -1312,11 +1331,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
size,
} => {
let scope = PassErrorScope::SetIndexBuffer(buffer_id);
let buffer = info
.trackers
let buffer: &Buffer<A> = info
.usage_scope
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDEX)
.map_err(|e| RenderCommandError::Buffer(buffer_id, e))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDEX)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, BufferUsages::INDEX)
.map_pass_err(scope)?;
@ -1359,11 +1377,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
size,
} => {
let scope = PassErrorScope::SetVertexBuffer(buffer_id);
let buffer = info
.trackers
let buffer: &Buffer<A> = info
.usage_scope
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::VERTEX)
.map_err(|e| RenderCommandError::Buffer(buffer_id, e))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::VERTEX)
.map_pass_err(scope)?;
check_buffer_usage(buffer.usage, BufferUsages::VERTEX)
.map_pass_err(scope)?;
@ -1617,11 +1634,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)
.map_pass_err(scope)?;
let indirect_buffer = info
.trackers
let indirect_buffer: &Buffer<A> = info
.usage_scope
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDIRECT)
.map_err(|e| RenderCommandError::Buffer(buffer_id, e))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDIRECT)
.map_pass_err(scope)?;
check_buffer_usage(indirect_buffer.usage, BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -1688,11 +1704,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.require_downlevel_flags(wgt::DownlevelFlags::INDIRECT_EXECUTION)
.map_pass_err(scope)?;
let indirect_buffer = info
.trackers
let indirect_buffer: &Buffer<A> = info
.usage_scope
.buffers
.use_extend(&*buffer_guard, buffer_id, (), hal::BufferUses::INDIRECT)
.map_err(|e| RenderCommandError::Buffer(buffer_id, e))
.merge_single(&*buffer_guard, buffer_id, hal::BufferUses::INDIRECT)
.map_pass_err(scope)?;
check_buffer_usage(indirect_buffer.usage, BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -1702,16 +1717,14 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.ok_or(RenderCommandError::DestroyedBuffer(buffer_id))
.map_pass_err(scope)?;
let count_buffer = info
.trackers
let count_buffer: &Buffer<A> = info
.usage_scope
.buffers
.use_extend(
.merge_single(
&*buffer_guard,
count_buffer_id,
(),
hal::BufferUses::INDIRECT,
)
.map_err(|e| RenderCommandError::Buffer(count_buffer_id, e))
.map_pass_err(scope)?;
check_buffer_usage(count_buffer.usage, BufferUsages::INDIRECT)
.map_pass_err(scope)?;
@ -1814,16 +1827,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
} => {
let scope = PassErrorScope::WriteTimestamp;
let query_set = cmd_buf
let query_set: &resource::QuerySet<A> = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => {
RenderCommandError::InvalidQuerySet(query_set_id)
}
_ => unreachable!(),
})
.add_single(&*query_set_guard, query_set_id)
.ok_or(RenderCommandError::InvalidQuerySet(query_set_id))
.map_pass_err(scope)?;
query_set
@ -1841,16 +1849,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
} => {
let scope = PassErrorScope::BeginPipelineStatisticsQuery;
let query_set = cmd_buf
let query_set: &resource::QuerySet<A> = cmd_buf
.trackers
.query_sets
.use_extend(&*query_set_guard, query_set_id, (), ())
.map_err(|e| match e {
UseExtendError::InvalidResource => {
RenderCommandError::InvalidQuerySet(query_set_id)
}
_ => unreachable!(),
})
.add_single(&*query_set_guard, query_set_id)
.ok_or(RenderCommandError::InvalidQuerySet(query_set_id))
.map_pass_err(scope)?;
query_set
@ -1871,11 +1874,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
}
RenderCommand::ExecuteBundle(bundle_id) => {
let scope = PassErrorScope::ExecuteBundle;
let bundle = cmd_buf
let bundle: &command::RenderBundle<A> = cmd_buf
.trackers
.bundles
.use_extend(&*bundle_guard, bundle_id, (), ())
.map_err(|_| RenderCommandError::InvalidRenderBundle(bundle_id))
.add_single(&*bundle_guard, bundle_id)
.ok_or(RenderCommandError::InvalidRenderBundle(bundle_id))
.map_pass_err(scope)?;
info.context
@ -1913,7 +1916,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
raw,
&*pipeline_layout_guard,
&*bind_group_guard,
&*pipeline_guard,
&*render_pipeline_guard,
&*buffer_guard,
)
}
@ -1927,23 +1930,21 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
})
.map_pass_err(scope)?;
info.trackers
.merge_extend(&bundle.used)
.map_pass_err(scope)?;
// Start tracking the bind groups specifically, as they are the only
// compound resources, to make it easier to update submission indices
// later at submission time.
cmd_buf
.trackers
.bind_groups
.merge_extend(&bundle.used.bind_groups)
.unwrap();
unsafe {
info.usage_scope
.merge_render_bundle(&*texture_guard, &bundle.used)
.map_pass_err(scope)?;
cmd_buf
.trackers
.add_from_render_bundle(&bundle.used)
.map_pass_err(scope)?;
};
state.reset_bundle();
}
}
}
log::trace!("Merging {:?} with the render pass", encoder_id);
log::trace!("Merging renderpass into cmd_buf {:?}", encoder_id);
let (trackers, pending_discard_init_fixups) =
info.finish(raw, &*texture_guard).map_pass_err(init_scope)?;
@ -1977,11 +1978,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.map_err(RenderCommandError::InvalidQuerySet)
.map_pass_err(PassErrorScope::QueryReset)?;
super::CommandBuffer::insert_barriers(
super::CommandBuffer::insert_barriers_from_scope(
transit,
&mut cmd_buf.trackers,
&trackers.buffers,
&trackers.textures,
&scope,
&*buffer_guard,
&*texture_guard,
);

89
wgpu-core/src/command/transfer.rs
View File

@ -1,12 +1,12 @@
#[cfg(feature = "trace")]
use crate::device::trace::Command as TraceCommand;
use crate::{
command::{CommandBuffer, CommandEncoderError},
command::{clear_texture, CommandBuffer, CommandEncoderError},
conv,
device::{Device, MissingDownlevelFlags},
error::{ErrorFormatter, PrettyError},
hub::{Global, GlobalIdentityHandlerFactory, HalApi, Storage, Token},
id::{BufferId, CommandEncoderId, Id, TextureId, Valid},
id::{BufferId, CommandEncoderId, TextureId, Valid},
init_tracker::{
has_copy_partial_init_tracker_coverage, MemoryInitKind, TextureInitRange,
TextureInitTrackerAction,
@ -15,14 +15,13 @@ use crate::{
track::TextureSelector,
};
use arrayvec::ArrayVec;
use hal::CommandEncoder as _;
use thiserror::Error;
use wgt::{BufferAddress, BufferUsages, Extent3d, TextureUsages};
use std::iter;
use super::clear::clear_texture;
pub type ImageCopyBuffer = wgt::ImageCopyBuffer<BufferId>;
pub type ImageCopyTexture = wgt::ImageCopyTexture<TextureId>;
@ -191,7 +190,7 @@ pub(crate) fn extract_texture_selector<A: hal::Api>(
aspect: copy_aspect,
};
let selector = TextureSelector {
levels: copy_texture.mip_level..copy_texture.mip_level + 1,
mips: copy_texture.mip_level..copy_texture.mip_level + 1,
layers,
};
@ -381,14 +380,13 @@ pub(crate) fn validate_texture_copy_range(
Ok((copy_extent, array_layer_count))
}
fn handle_texture_init<A: hal::Api>(
fn handle_texture_init<A: HalApi>(
init_kind: MemoryInitKind,
cmd_buf: &mut CommandBuffer<A>,
device: &Device<A>,
copy_texture: &ImageCopyTexture,
copy_size: &Extent3d,
texture_guard: &Storage<Texture<A>, Id<Texture<hal::api::Empty>>>,
texture: &Texture<A>,
texture_guard: &Storage<Texture<A>, TextureId>,
) {
let init_action = TextureInitTrackerAction {
id: copy_texture.texture,
@ -410,15 +408,16 @@ fn handle_texture_init<A: hal::Api>(
let cmd_buf_raw = cmd_buf.encoder.open();
for init in immediate_inits {
clear_texture(
texture_guard,
Valid(init.texture),
texture,
TextureInitRange {
mip_range: init.mip_level..(init.mip_level + 1),
layer_range: init.layer..(init.layer + 1),
},
cmd_buf_raw,
&mut cmd_buf.trackers.textures,
device,
&device.alignments,
&device.zero_buffer,
)
.unwrap();
}
@ -426,14 +425,14 @@ fn handle_texture_init<A: hal::Api>(
}
// Ensures the source texture of a transfer is in the right initialization state and records the state for after the transfer operation.
fn handle_src_texture_init<A: hal::Api>(
fn handle_src_texture_init<A: HalApi>(
cmd_buf: &mut CommandBuffer<A>,
device: &Device<A>,
source: &ImageCopyTexture,
copy_size: &Extent3d,
texture_guard: &Storage<Texture<A>, TextureId>,
) -> Result<(), TransferError> {
let texture = texture_guard
let _ = texture_guard
.get(source.texture)
.map_err(|_| TransferError::InvalidTexture(source.texture))?;
@ -444,13 +443,12 @@ fn handle_src_texture_init<A: hal::Api>(
source,
copy_size,
texture_guard,
texture,
);
Ok(())
}
// Ensures the destination texture of a transfer is in the right initialization state and records the state for after the transfer operation.
fn handle_dst_texture_init<A: hal::Api>(
fn handle_dst_texture_init<A: HalApi>(
cmd_buf: &mut CommandBuffer<A>,
device: &Device<A>,
destination: &ImageCopyTexture,
@ -480,7 +478,6 @@ fn handle_dst_texture_init<A: hal::Api>(
destination,
copy_size,
texture_guard,
texture,
);
Ok(())
}
@ -521,8 +518,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (src_buffer, src_pending) = cmd_buf
.trackers
.buffers
.use_replace(&*buffer_guard, source, (), hal::BufferUses::COPY_SRC)
.map_err(TransferError::InvalidBuffer)?;
.set_single(&*buffer_guard, source, hal::BufferUses::COPY_SRC)
.ok_or(TransferError::InvalidBuffer(source))?;
let src_raw = src_buffer
.raw
.as_ref()
@ -531,15 +528,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
return Err(TransferError::MissingCopySrcUsageFlag.into());
}
// expecting only a single barrier
let src_barrier = src_pending
.map(|pending| pending.into_hal(src_buffer))
.next();
let src_barrier = src_pending.map(|pending| pending.into_hal(src_buffer));
let (dst_buffer, dst_pending) = cmd_buf
.trackers
.buffers
.use_replace(&*buffer_guard, destination, (), hal::BufferUses::COPY_DST)
.map_err(TransferError::InvalidBuffer)?;
.set_single(&*buffer_guard, destination, hal::BufferUses::COPY_DST)
.ok_or(TransferError::InvalidBuffer(destination))?;
let dst_raw = dst_buffer
.raw
.as_ref()
@ -547,9 +542,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
if !dst_buffer.usage.contains(BufferUsages::COPY_DST) {
return Err(TransferError::MissingCopyDstUsageFlag(Some(destination), None).into());
}
let dst_barrier = dst_pending
.map(|pending| pending.into_hal(dst_buffer))
.next();
let dst_barrier = dst_pending.map(|pending| pending.into_hal(dst_buffer));
if size % wgt::COPY_BUFFER_ALIGNMENT != 0 {
return Err(TransferError::UnalignedCopySize(size).into());
@ -659,8 +652,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (src_buffer, src_pending) = cmd_buf
.trackers
.buffers
.use_replace(&*buffer_guard, source.buffer, (), hal::BufferUses::COPY_SRC)
.map_err(TransferError::InvalidBuffer)?;
.set_single(&*buffer_guard, source.buffer, hal::BufferUses::COPY_SRC)
.ok_or(TransferError::InvalidBuffer(source.buffer))?;
let src_raw = src_buffer
.raw
.as_ref()
@ -668,18 +661,18 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
if !src_buffer.usage.contains(BufferUsages::COPY_SRC) {
return Err(TransferError::MissingCopySrcUsageFlag.into());
}
let src_barriers = src_pending.map(|pending| pending.into_hal(src_buffer));
let src_barrier = src_pending.map(|pending| pending.into_hal(src_buffer));
let (dst_texture, dst_pending) = cmd_buf
.trackers
.textures
.use_replace(
.set_single(
&*texture_guard,
destination.texture,
dst_range,
hal::TextureUses::COPY_DST,
)
.unwrap();
.ok_or(TransferError::InvalidTexture(destination.texture))?;
let dst_raw = dst_texture
.inner
.as_raw()
@ -689,7 +682,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
TransferError::MissingCopyDstUsageFlag(None, Some(destination.texture)).into(),
);
}
let dst_barriers = dst_pending.map(|pending| pending.into_hal(dst_texture));
let dst_barrier = dst_pending.map(|pending| pending.into_hal(dst_texture));
let format_desc = dst_texture.desc.format.describe();
let (hal_copy_size, array_layer_count) = validate_texture_copy_range(
@ -736,8 +729,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let cmd_buf_raw = cmd_buf.encoder.open();
unsafe {
cmd_buf_raw.transition_textures(dst_barriers);
cmd_buf_raw.transition_buffers(src_barriers);
cmd_buf_raw.transition_textures(dst_barrier.into_iter());
cmd_buf_raw.transition_buffers(src_barrier.into_iter());
cmd_buf_raw.copy_buffer_to_texture(src_raw, dst_raw, regions);
}
Ok(())
@ -786,13 +779,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (src_texture, src_pending) = cmd_buf
.trackers
.textures
.use_replace(
.set_single(
&*texture_guard,
source.texture,
src_range,
hal::TextureUses::COPY_SRC,
)
.unwrap();
.ok_or(TransferError::InvalidTexture(source.texture))?;
let src_raw = src_texture
.inner
.as_raw()
@ -800,18 +793,17 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
if !src_texture.desc.usage.contains(TextureUsages::COPY_SRC) {
return Err(TransferError::MissingCopySrcUsageFlag.into());
}
let src_barriers = src_pending.map(|pending| pending.into_hal(src_texture));
let src_barrier = src_pending.map(|pending| pending.into_hal(src_texture));
let (dst_buffer, dst_pending) = cmd_buf
.trackers
.buffers
.use_replace(
.set_single(
&*buffer_guard,
destination.buffer,
(),
hal::BufferUses::COPY_DST,
)
.map_err(TransferError::InvalidBuffer)?;
.ok_or(TransferError::InvalidBuffer(destination.buffer))?;
let dst_raw = dst_buffer
.raw
.as_ref()
@ -821,7 +813,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
TransferError::MissingCopyDstUsageFlag(Some(destination.buffer), None).into(),
);
}
let dst_barriers = dst_pending.map(|pending| pending.into_hal(dst_buffer));
let dst_barrier = dst_pending.map(|pending| pending.into_hal(dst_buffer));
let format_desc = src_texture.desc.format.describe();
let (hal_copy_size, array_layer_count) =
@ -876,8 +868,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
});
let cmd_buf_raw = cmd_buf.encoder.open();
unsafe {
cmd_buf_raw.transition_buffers(dst_barriers);
cmd_buf_raw.transition_textures(src_barriers);
cmd_buf_raw.transition_buffers(dst_barrier.into_iter());
cmd_buf_raw.transition_textures(src_barrier.into_iter());
cmd_buf_raw.copy_texture_to_buffer(
src_raw,
hal::TextureUses::COPY_SRC,
@ -937,13 +929,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (src_texture, src_pending) = cmd_buf
.trackers
.textures
.use_replace(
.set_single(
&*texture_guard,
source.texture,
src_range,
hal::TextureUses::COPY_SRC,
)
.unwrap();
.ok_or(TransferError::InvalidTexture(source.texture))?;
let src_raw = src_texture
.inner
.as_raw()
@ -954,20 +946,21 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
//TODO: try to avoid this the collection. It's needed because both
// `src_pending` and `dst_pending` try to hold `trackers.textures` mutably.
let mut barriers = src_pending
let mut barriers: ArrayVec<_, 2> = src_pending
.map(|pending| pending.into_hal(src_texture))
.collect::<Vec<_>>();
.into_iter()
.collect();
let (dst_texture, dst_pending) = cmd_buf
.trackers
.textures
.use_replace(
.set_single(
&*texture_guard,
destination.texture,
dst_range,
hal::TextureUses::COPY_DST,
)
.unwrap();
.ok_or(TransferError::InvalidTexture(destination.texture))?;
let dst_raw = dst_texture
.inner
.as_raw()

12
wgpu-core/src/conv.rs
View File

@ -1,6 +1,10 @@
use crate::resource;
pub fn is_power_of_two(val: u32) -> bool {
pub fn is_power_of_two_u16(val: u16) -> bool {
val != 0 && (val & (val - 1)) == 0
}
pub fn is_power_of_two_u32(val: u32) -> bool {
val != 0 && (val & (val - 1)) == 0
}
@ -53,7 +57,7 @@ pub fn map_buffer_usage(usage: wgt::BufferUsages) -> hal::BufferUses {
usage.contains(wgt::BufferUsages::UNIFORM),
);
u.set(
hal::BufferUses::STORAGE_READ | hal::BufferUses::STORAGE_WRITE,
hal::BufferUses::STORAGE_READ | hal::BufferUses::STORAGE_READ_WRITE,
usage.contains(wgt::BufferUsages::STORAGE),
);
u.set(
@ -81,7 +85,7 @@ pub fn map_texture_usage(
usage.contains(wgt::TextureUsages::TEXTURE_BINDING),
);
u.set(
hal::TextureUses::STORAGE_READ | hal::TextureUses::STORAGE_WRITE,
hal::TextureUses::STORAGE_READ | hal::TextureUses::STORAGE_READ_WRITE,
usage.contains(wgt::TextureUsages::STORAGE_BINDING),
);
let is_color = aspect.contains(hal::FormatAspects::COLOR);
@ -148,7 +152,7 @@ pub fn check_texture_dimension_size(
return Err(Tde::LimitExceeded { dim, given, limit });
}
}
if sample_size == 0 || sample_size > sample_limit || !is_power_of_two(sample_size) {
if sample_size == 0 || sample_size > sample_limit || !is_power_of_two_u32(sample_size) {
return Err(Tde::InvalidSampleCount(sample_size));
}

33
wgpu-core/src/device/life.rs
View File

@ -7,7 +7,7 @@ use crate::{
},
hub::{GlobalIdentityHandlerFactory, HalApi, Hub, Token},
id, resource,
track::TrackerSet,
track::{BindGroupStates, RenderBundleScope, Tracker},
RefCount, Stored, SubmissionIndex,
};
use smallvec::SmallVec;
@ -68,16 +68,20 @@ impl SuspectedResources {
self.query_sets.extend_from_slice(&other.query_sets);
}
pub(super) fn add_trackers(&mut self, trackers: &TrackerSet) {
pub(super) fn add_render_bundle_scope<A: HalApi>(&mut self, trackers: &RenderBundleScope<A>) {
self.buffers.extend(trackers.buffers.used());
self.textures.extend(trackers.textures.used());
self.bind_groups.extend(trackers.bind_groups.used());
self.render_pipelines
.extend(trackers.render_pipelines.used());
self.query_sets.extend(trackers.query_sets.used());
}
pub(super) fn add_bind_group_states<A: HalApi>(&mut self, trackers: &BindGroupStates<A>) {
self.buffers.extend(trackers.buffers.used());
self.textures.extend(trackers.textures.used());
self.texture_views.extend(trackers.views.used());
self.samplers.extend(trackers.samplers.used());
self.bind_groups.extend(trackers.bind_groups.used());
self.compute_pipelines.extend(trackers.compute_pipes.used());
self.render_pipelines.extend(trackers.render_pipes.used());
self.render_bundles.extend(trackers.bundles.used());
self.query_sets.extend(trackers.query_sets.used());
}
}
@ -273,7 +277,8 @@ pub(super) struct LifetimeTracker<A: hal::Api> {
/// Textures can be used in the upcoming submission by `write_texture`.
pub future_suspected_textures: Vec<Stored<id::TextureId>>,
/// Resources that are suspected for destruction.
/// Resources whose user handle has died (i.e. drop/destroy has been called)
/// and will likely be ready for destruction soon.
pub suspected_resources: SuspectedResources,
/// Resources used by queue submissions still in flight. One entry per
@ -491,7 +496,7 @@ impl<A: HalApi> LifetimeTracker<A> {
pub(super) fn triage_suspected<G: GlobalIdentityHandlerFactory>(
&mut self,
hub: &Hub<A, G>,
trackers: &Mutex<TrackerSet>,
trackers: &Mutex<Tracker<A>>,
#[cfg(feature = "trace")] trace: Option<&Mutex<trace::Trace>>,
token: &mut Token<super::Device<A>>,
) {
@ -510,7 +515,7 @@ impl<A: HalApi> LifetimeTracker<A> {
}
if let Some(res) = hub.render_bundles.unregister_locked(id.0, &mut *guard) {
self.suspected_resources.add_trackers(&res.used);
self.suspected_resources.add_render_bundle_scope(&res.used);
}
}
}
@ -529,7 +534,7 @@ impl<A: HalApi> LifetimeTracker<A> {
}
if let Some(res) = hub.bind_groups.unregister_locked(id.0, &mut *guard) {
self.suspected_resources.add_trackers(&res.used);
self.suspected_resources.add_bind_group_states(&res.used);
self.suspected_resources
.bind_group_layouts
@ -670,7 +675,7 @@ impl<A: HalApi> LifetimeTracker<A> {
let mut trackers = trackers.lock();
for id in self.suspected_resources.compute_pipelines.drain(..) {
if trackers.compute_pipes.remove_abandoned(id) {
if trackers.compute_pipelines.remove_abandoned(id) {
log::debug!("Compute pipeline {:?} will be destroyed", id);
#[cfg(feature = "trace")]
if let Some(t) = trace {
@ -695,7 +700,7 @@ impl<A: HalApi> LifetimeTracker<A> {
let mut trackers = trackers.lock();
for id in self.suspected_resources.render_pipelines.drain(..) {
if trackers.render_pipes.remove_abandoned(id) {
if trackers.render_pipelines.remove_abandoned(id) {
log::debug!("Render pipeline {:?} will be destroyed", id);
#[cfg(feature = "trace")]
if let Some(t) = trace {
@ -829,7 +834,7 @@ impl<A: HalApi> LifetimeTracker<A> {
&mut self,
hub: &Hub<A, G>,
raw: &A::Device,
trackers: &Mutex<TrackerSet>,
trackers: &Mutex<Tracker<A>>,
token: &mut Token<super::Device<A>>,
) -> Vec<super::BufferMapPendingClosure> {
if self.ready_to_map.is_empty() {

196
wgpu-core/src/device/mod.rs
View File

@ -8,7 +8,7 @@ use crate::{
TextureInitTracker, TextureInitTrackerAction,
},
instance, pipeline, present, resource,
track::{BufferState, TextureSelector, TextureState, TrackerSet, UsageConflict},
track::{BindGroupStates, TextureSelector, Tracker},
validation::{self, check_buffer_usage, check_texture_usage},
FastHashMap, Label, LabelHelpers as _, LifeGuard, MultiRefCount, RefCount, Stored,
SubmissionIndex, DOWNLEVEL_ERROR_MESSAGE,
@ -22,7 +22,7 @@ use smallvec::SmallVec;
use thiserror::Error;
use wgt::{BufferAddress, TextureFormat, TextureViewDimension};
use std::{borrow::Cow, iter, marker::PhantomData, mem, num::NonZeroU32, ops::Range, ptr};
use std::{borrow::Cow, iter, mem, num::NonZeroU32, ops::Range, ptr};
mod life;
pub mod queue;
@ -256,7 +256,7 @@ impl<A: hal::Api> CommandAllocator<A> {
/// 1. `life_tracker` is locked after `hub.devices`, enforced by the type system
/// 1. `self.trackers` is locked last (unenforced)
/// 1. `self.trace` is locked last (unenforced)
pub struct Device<A: hal::Api> {
pub struct Device<A: HalApi> {
pub(crate) raw: A::Device,
pub(crate) adapter_id: Stored<id::AdapterId>,
pub(crate) queue: A::Queue,
@ -281,7 +281,7 @@ pub struct Device<A: hal::Api> {
/// All live resources allocated with this [`Device`].
///
/// Has to be locked temporarily only (locked last)
pub(crate) trackers: Mutex<TrackerSet>,
pub(crate) trackers: Mutex<Tracker<A>>,
// Life tracker should be locked right after the device and before anything else.
life_tracker: Mutex<life::LifetimeTracker<A>>,
/// Temporary storage for resource management functions. Cleared at the end
@ -306,7 +306,7 @@ pub enum CreateDeviceError {
FailedToCreateZeroBuffer(#[from] DeviceError),
}
impl<A: hal::Api> Device<A> {
impl<A: HalApi> Device<A> {
pub(crate) fn require_features(&self, feature: wgt::Features) -> Result<(), MissingFeatures> {
if self.features.contains(feature) {
Ok(())
@ -328,7 +328,6 @@ impl<A: hal::Api> Device<A> {
}
impl<A: HalApi> Device<A> {
#[allow(clippy::too_many_arguments)]
pub(crate) fn new(
open: hal::OpenDevice<A>,
adapter_id: Stored<id::AdapterId>,
@ -394,7 +393,7 @@ impl<A: HalApi> Device<A> {
command_allocator: Mutex::new(com_alloc),
active_submission_index: 0,
fence,
trackers: Mutex::new(TrackerSet::new(A::VARIANT)),
trackers: Mutex::new(Tracker::new()),
life_tracker: Mutex::new(life::LifetimeTracker::new()),
temp_suspected: life::SuspectedResources::default(),
#[cfg(feature = "trace")]
@ -495,7 +494,7 @@ impl<A: HalApi> Device<A> {
fn untrack<'this, 'token: 'this, G: GlobalIdentityHandlerFactory>(
&'this mut self,
hub: &Hub<A, G>,
trackers: &TrackerSet,
trackers: &Tracker<A>,
token: &mut Token<'token, Self>,
) {
self.temp_suspected.clear();
@ -536,12 +535,12 @@ impl<A: HalApi> Device<A> {
self.temp_suspected.samplers.push(id);
}
}
for id in trackers.compute_pipes.used() {
for id in trackers.compute_pipelines.used() {
if compute_pipe_guard[id].life_guard.ref_count.is_none() {
self.temp_suspected.compute_pipelines.push(id);
}
}
for id in trackers.render_pipes.used() {
for id in trackers.render_pipelines.used() {
if render_pipe_guard[id].life_guard.ref_count.is_none() {
self.temp_suspected.render_pipelines.push(id);
}
@ -655,7 +654,7 @@ impl<A: HalApi> Device<A> {
desc.array_layer_count(),
),
full_range: TextureSelector {
levels: 0..desc.mip_level_count,
mips: 0..desc.mip_level_count,
layers: 0..desc.array_layer_count(),
},
life_guard: LifeGuard::new(desc.label.borrow_or_default()),
@ -876,7 +875,7 @@ impl<A: HalApi> Device<A> {
_ => texture.desc.array_layer_count(),
},
};
let level_end = texture.full_range.levels.end;
let level_end = texture.full_range.mips.end;
let layer_end = texture.full_range.layers.end;
if required_level_count > level_end {
return Err(resource::CreateTextureViewError::TooManyMipLevels {
@ -927,7 +926,7 @@ impl<A: HalApi> Device<A> {
.array_layer_count
.map_or(layer_end, |_| required_layer_count);
let selector = TextureSelector {
levels: desc.range.base_mip_level..end_level,
mips: desc.range.base_mip_level..end_level,
layers: desc.range.base_array_layer..end_layer,
};
@ -972,11 +971,11 @@ impl<A: HalApi> Device<A> {
wgt::TextureViewDimension::D3 => {
hal::TextureUses::RESOURCE
| hal::TextureUses::STORAGE_READ
| hal::TextureUses::STORAGE_WRITE
| hal::TextureUses::STORAGE_READ_WRITE
}
_ => hal::TextureUses::all(),
};
let mask_mip_level = if selector.levels.end - selector.levels.start != 1 {
let mask_mip_level = if selector.mips.end - selector.mips.start != 1 {
hal::TextureUses::RESOURCE
} else {
hal::TextureUses::all()
@ -1058,7 +1057,8 @@ impl<A: HalApi> Device<A> {
let anisotropy_clamp = if let Some(clamp) = desc.anisotropy_clamp {
let clamp = clamp.get();
let valid_clamp = clamp <= hal::MAX_ANISOTROPY && conv::is_power_of_two(clamp as u32);
let valid_clamp =
clamp <= hal::MAX_ANISOTROPY && conv::is_power_of_two_u32(clamp as u32);
if !valid_clamp {
return Err(resource::CreateSamplerError::InvalidClamp(clamp));
}
@ -1486,7 +1486,6 @@ impl<A: HalApi> Device<A> {
})
}
#[allow(clippy::too_many_arguments)]
fn create_buffer_binding<'a>(
bb: &binding_model::BufferBinding,
binding: u32,
@ -1494,7 +1493,7 @@ impl<A: HalApi> Device<A> {
used_buffer_ranges: &mut Vec<BufferInitTrackerAction>,
dynamic_binding_info: &mut Vec<binding_model::BindGroupDynamicBindingData>,
late_buffer_binding_sizes: &mut FastHashMap<u32, wgt::BufferSize>,
used: &mut TrackerSet,
used: &mut BindGroupStates<A>,
storage: &'a Storage<resource::Buffer<A>, id::BufferId>,
limits: &wgt::Limits,
) -> Result<hal::BufferBinding<'a, A>, binding_model::CreateBindGroupError> {
@ -1525,7 +1524,7 @@ impl<A: HalApi> Device<A> {
if read_only {
hal::BufferUses::STORAGE_READ
} else {
hal::BufferUses::STORAGE_READ | hal::BufferUses::STORAGE_WRITE
hal::BufferUses::STORAGE_READ_WRITE
},
limits.max_storage_buffer_binding_size,
),
@ -1543,8 +1542,8 @@ impl<A: HalApi> Device<A> {
let buffer = used
.buffers
.use_extend(storage, bb.buffer_id, (), internal_use)
.map_err(|_| Error::InvalidBuffer(bb.buffer_id))?;
.add_single(storage, bb.buffer_id, internal_use)
.ok_or(Error::InvalidBuffer(bb.buffer_id))?;
check_buffer_usage(buffer.usage, pub_usage)?;
let raw_buffer = buffer
.raw
@ -1613,26 +1612,26 @@ impl<A: HalApi> Device<A> {
fn create_texture_binding(
view: &resource::TextureView<A>,
texture_guard: &parking_lot::lock_api::RwLockReadGuard<
parking_lot::RawRwLock,
Storage<resource::Texture<A>, id::Id<resource::Texture<hal::api::Empty>>>,
>,
texture_guard: &Storage<resource::Texture<A>, id::TextureId>,
internal_use: hal::TextureUses,
pub_usage: wgt::TextureUsages,
used: &mut TrackerSet,
used: &mut BindGroupStates<A>,
used_texture_ranges: &mut Vec<TextureInitTrackerAction>,
) -> Result<(), binding_model::CreateBindGroupError> {
// Careful here: the texture may no longer have its own ref count,
// if it was deleted by the user.
let texture = &texture_guard[view.parent_id.value];
used.textures
.change_extend(
view.parent_id.value,
&view.parent_id.ref_count,
view.selector.clone(),
let texture = used
.textures
.add_single(
texture_guard,
view.parent_id.value.0,
view.parent_id.ref_count.clone(),
Some(view.selector.clone()),
internal_use,
)
.map_err(UsageConflict::from)?;
.ok_or(binding_model::CreateBindGroupError::InvalidTexture(
view.parent_id.value.0,
))?;
check_texture_usage(texture.desc.usage, pub_usage)?;
used_texture_ranges.push(TextureInitTrackerAction {
@ -1674,7 +1673,7 @@ impl<A: HalApi> Device<A> {
// it needs to be in BGL iteration order, not BG entry order.
let mut late_buffer_binding_sizes = FastHashMap::default();
// fill out the descriptors
let mut used = TrackerSet::new(A::VARIANT);
let mut used = BindGroupStates::new();
let (buffer_guard, mut token) = hub.buffers.read(token);
let (texture_guard, mut token) = hub.textures.read(&mut token); //skip token
@ -1738,8 +1737,8 @@ impl<A: HalApi> Device<A> {
wgt::BindingType::Sampler(ty) => {
let sampler = used
.samplers
.use_extend(&*sampler_guard, id, (), ())
.map_err(|_| Error::InvalidSampler(id))?;
.add_single(&*sampler_guard, id)
.ok_or(Error::InvalidSampler(id))?;
// Allowed sampler values for filtering and comparison
let (allowed_filtering, allowed_comparison) = match ty {
@ -1787,8 +1786,8 @@ impl<A: HalApi> Device<A> {
for &id in bindings_array.iter() {
let sampler = used
.samplers
.use_extend(&*sampler_guard, id, (), ())
.map_err(|_| Error::InvalidSampler(id))?;
.add_single(&*sampler_guard, id)
.ok_or(Error::InvalidSampler(id))?;
hal_samplers.push(&sampler.raw);
}
@ -1797,8 +1796,8 @@ impl<A: HalApi> Device<A> {
Br::TextureView(id) => {
let view = used
.views
.use_extend(&*texture_view_guard, id, (), ())
.map_err(|_| Error::InvalidTextureView(id))?;
.add_single(&*texture_view_guard, id)
.ok_or(Error::InvalidTextureView(id))?;
let (pub_usage, internal_use) = Self::texture_use_parameters(
binding,
decl,
@ -1828,8 +1827,8 @@ impl<A: HalApi> Device<A> {
for &id in bindings_array.iter() {
let view = used
.views
.use_extend(&*texture_view_guard, id, (), ())
.map_err(|_| Error::InvalidTextureView(id))?;
.add_single(&*texture_view_guard, id)
.ok_or(Error::InvalidTextureView(id))?;
let (pub_usage, internal_use) =
Self::texture_use_parameters(binding, decl, view,
"SampledTextureArray, ReadonlyStorageTextureArray or WriteonlyStorageTextureArray")?;
@ -1858,6 +1857,8 @@ impl<A: HalApi> Device<A> {
});
}
used.optimize();
hal_entries.sort_by_key(|entry| entry.binding);
for (a, b) in hal_entries.iter().zip(hal_entries.iter().skip(1)) {
if a.binding == b.binding {
@ -2018,7 +2019,7 @@ impl<A: HalApi> Device<A> {
});
}
let mip_level_count = view.selector.levels.end - view.selector.levels.start;
let mip_level_count = view.selector.mips.end - view.selector.mips.start;
if mip_level_count != 1 {
return Err(Error::InvalidStorageTextureMipLevelCount {
binding,
@ -2027,7 +2028,7 @@ impl<A: HalApi> Device<A> {
}
let internal_use = match access {
wgt::StorageTextureAccess::WriteOnly => hal::TextureUses::STORAGE_WRITE,
wgt::StorageTextureAccess::WriteOnly => hal::TextureUses::STORAGE_READ_WRITE,
wgt::StorageTextureAccess::ReadOnly => {
if !view
.format_features
@ -2047,7 +2048,7 @@ impl<A: HalApi> Device<A> {
return Err(Error::StorageReadNotSupported(view.desc.format));
}
hal::TextureUses::STORAGE_WRITE | hal::TextureUses::STORAGE_READ
hal::TextureUses::STORAGE_READ_WRITE
}
};
Ok((wgt::TextureUsages::STORAGE_BINDING, internal_use))
@ -2551,7 +2552,7 @@ impl<A: HalApi> Device<A> {
let samples = {
let sc = desc.multisample.count;
if sc == 0 || sc > 32 || !conv::is_power_of_two(sc) {
if sc == 0 || sc > 32 || !conv::is_power_of_two_u32(sc) {
return Err(pipeline::CreateRenderPipelineError::InvalidSampleCount(sc));
}
sc
@ -2879,7 +2880,7 @@ impl<A: HalApi> Device<A> {
}
}
impl<A: hal::Api> Device<A> {
impl<A: HalApi> Device<A> {
pub(crate) fn destroy_buffer(&self, buffer: resource::Buffer<A>) {
if let Some(raw) = buffer.raw {
unsafe {
@ -2925,7 +2926,7 @@ impl<A: hal::Api> Device<A> {
}
}
impl<A: hal::Api> crate::hub::Resource for Device<A> {
impl<A: HalApi> crate::hub::Resource for Device<A> {
const TYPE: &'static str = "Device";
fn life_guard(&self) -> &LifeGuard {
@ -2981,7 +2982,7 @@ pub struct ImplicitPipelineIds<'a, G: GlobalIdentityHandlerFactory> {
}
impl<G: GlobalIdentityHandlerFactory> ImplicitPipelineIds<'_, G> {
fn prepare<A: hal::Api>(self, hub: &Hub<A, G>) -> ImplicitPipelineContext {
fn prepare<A: HalApi>(self, hub: &Hub<A, G>) -> ImplicitPipelineContext {
ImplicitPipelineContext {
root_id: hub.pipeline_layouts.prepare(self.root_id).into_id(),
group_ids: self
@ -3184,8 +3185,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.trackers
.lock()
.buffers
.init(id, ref_count, BufferState::with_usage(buffer_use))
.unwrap();
.insert_single(id, ref_count, buffer_use);
return (id.0, None);
};
@ -3483,19 +3484,17 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
Ok(texture) => texture,
Err(error) => break error,
};
let num_levels = texture.full_range.levels.end;
let num_layers = texture.full_range.layers.end;
let ref_count = texture.life_guard.add_ref();
let id = fid.assign(texture, &mut token);
log::info!("Created texture {:?} with {:?}", id, desc);
device
.trackers
.lock()
.textures
.init(id, ref_count, TextureState::new(num_levels, num_layers))
.unwrap();
device.trackers.lock().textures.insert_single(
id.0,
ref_count,
hal::TextureUses::UNINITIALIZED,
);
return (id.0, None);
};
@ -3561,19 +3560,17 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
texture.initialization_status = TextureInitTracker::new(desc.mip_level_count, 0);
let num_levels = texture.full_range.levels.end;
let num_layers = texture.full_range.layers.end;
let ref_count = texture.life_guard.add_ref();
let id = fid.assign(texture, &mut token);
log::info!("Created texture {:?} with {:?}", id, desc);
device
.trackers
.lock()
.textures
.init(id, ref_count, TextureState::new(num_levels, num_layers))
.unwrap();
device.trackers.lock().textures.insert_single(
id.0,
ref_count,
hal::TextureUses::UNINITIALIZED,
);
return (id.0, None);
};
@ -3731,12 +3728,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let ref_count = view.life_guard.add_ref();
let id = fid.assign(view, &mut token);
device
.trackers
.lock()
.views
.init(id, ref_count, PhantomData)
.unwrap();
device.trackers.lock().views.insert_single(id, ref_count);
return (id.0, None);
};
@ -3829,12 +3821,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let ref_count = sampler.life_guard.add_ref();
let id = fid.assign(sampler, &mut token);
device
.trackers
.lock()
.samplers
.init(id, ref_count, PhantomData)
.unwrap();
device.trackers.lock().samplers.insert_single(id, ref_count);
return (id.0, None);
};
@ -4092,18 +4080,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let ref_count = bind_group.life_guard.add_ref();
let id = fid.assign(bind_group, &mut token);
log::debug!(
"Bind group {:?} {:#?}",
id,
hub.bind_groups.read(&mut token).0[id].used
);
log::debug!("Bind group {:?}", id,);
device
.trackers
.lock()
.bind_groups
.init(id, ref_count, PhantomData)
.unwrap();
.insert_single(id, ref_count);
return (id.0, None);
};
@ -4406,16 +4389,11 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
Err(e) => break e,
};
log::debug!("Render bundle {:#?}", render_bundle.used);
log::debug!("Render bundle");
let ref_count = render_bundle.life_guard.add_ref();
let id = fid.assign(render_bundle, &mut token);
device
.trackers
.lock()
.bundles
.init(id, ref_count, PhantomData)
.unwrap();
device.trackers.lock().bundles.insert_single(id, ref_count);
return (id.0, None);
};
@ -4494,8 +4472,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.trackers
.lock()
.query_sets
.init(id, ref_count, PhantomData)
.unwrap();
.insert_single(id, ref_count);
return (id.0, None);
};
@ -4589,9 +4566,9 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
device
.trackers
.lock()
.render_pipes
.init(id, ref_count, PhantomData)
.unwrap();
.render_pipelines
.insert_single(id, ref_count);
return (id.0, None);
};
@ -4730,9 +4707,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
device
.trackers
.lock()
.compute_pipes
.init(id, ref_count, PhantomData)
.unwrap();
.compute_pipelines
.insert_single(id, ref_count);
return (id.0, None);
};
@ -5204,16 +5180,20 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
};
log::debug!("Buffer {:?} map state -> Waiting", buffer_id);
(buffer.device_id.value, buffer.life_guard.add_ref())
let device = &device_guard[buffer.device_id.value];
let ret = (buffer.device_id.value, buffer.life_guard.add_ref());
let mut trackers = device.trackers.lock();
trackers
.buffers
.set_single(&*buffer_guard, buffer_id, internal_use);
trackers.buffers.drain();
ret
};
let device = &device_guard[device_id];
device.trackers.lock().buffers.change_replace(
id::Valid(buffer_id),
&ref_count,
(),
internal_use,
);
device
.lock_life(&mut token)

133
wgpu-core/src/device/queue.rs
View File

@ -192,7 +192,7 @@ impl<A: hal::Api> PendingWrites<A> {
}
}
impl<A: hal::Api> super::Device<A> {
impl<A: HalApi> super::Device<A> {
fn prepare_stage(&mut self, size: wgt::BufferAddress) -> Result<StagingData<A>, DeviceError> {
profiling::scope!("prepare_stage");
let stage_desc = hal::BufferDescriptor {
@ -286,8 +286,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let mut trackers = device.trackers.lock();
let (dst, transition) = trackers
.buffers
.use_replace(&*buffer_guard, buffer_id, (), hal::BufferUses::COPY_DST)
.map_err(TransferError::InvalidBuffer)?;
.set_single(&*buffer_guard, buffer_id, hal::BufferUses::COPY_DST)
.ok_or(TransferError::InvalidBuffer(buffer_id))?;
let dst_raw = dst
.raw
.as_ref()
@ -451,9 +451,9 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.drain(init_layer_range)
.collect::<Vec<std::ops::Range<u32>>>()
{
crate::command::clear_texture_no_device(
crate::command::clear_texture(
&*texture_guard,
id::Valid(destination.texture),
&*dst,
TextureInitRange {
mip_range: destination.mip_level..(destination.mip_level + 1),
layer_range,
@ -473,13 +473,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (dst, transition) = trackers
.textures
.use_replace(
.set_single(
&*texture_guard,
destination.texture,
selector,
hal::TextureUses::COPY_DST,
)
.unwrap();
.ok_or(TransferError::InvalidTexture(destination.texture))?;
let (hal_copy_size, array_layer_count) =
validate_texture_copy_range(destination, &dst.desc, CopySide::Destination, size)?;
@ -561,7 +561,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.ok_or(TransferError::InvalidTexture(destination.texture))?;
unsafe {
encoder.transition_textures(transition.map(|pending| pending.into_hal(dst)));
encoder
.transition_textures(transition.map(|pending| pending.into_hal(dst)).into_iter());
encoder.transition_buffers(iter::once(barrier));
encoder.copy_buffer_to_texture(&stage.buffer, dst_raw, regions);
}
@ -594,7 +595,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
device.active_submission_index += 1;
let submit_index = device.active_submission_index;
let mut active_executions = Vec::new();
let mut used_surface_textures = track::ResourceTracker::new(A::VARIANT);
let mut used_surface_textures = track::TextureUsageScope::new();
{
let (mut command_buffer_guard, mut token) = hub.command_buffers.write(&mut token);
@ -616,12 +617,15 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
//Note: locking the trackers has to be done after the storages
let mut trackers = device.trackers.lock();
used_surface_textures.set_size(texture_guard.len());
//TODO: if multiple command buffers are submitted, we can re-use the last
// native command buffer of the previous chain instead of always creating
// a temporary one, since the chains are not finished.
// finish all the command buffers first
for &cmb_id in command_buffer_ids {
#[allow(unused_mut)]
let mut cmdbuf = match hub
.command_buffers
.unregister_locked(cmb_id, &mut *command_buffer_guard)
@ -642,7 +646,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
}
// optimize the tracked states
cmdbuf.trackers.optimize();
// cmdbuf.trackers.optimize();
// update submission IDs
for id in cmdbuf.trackers.buffers.used() {
@ -669,33 +673,35 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
}
for id in cmdbuf.trackers.textures.used() {
let texture = &mut texture_guard[id];
match texture.inner {
let should_extend = match texture.inner {
TextureInner::Native { raw: None } => {
return Err(QueueSubmitError::DestroyedTexture(id.0));
}
TextureInner::Native { raw: Some(_) } => {}
TextureInner::Native { raw: Some(_) } => false,
TextureInner::Surface {
ref mut has_work, ..
} => {
use track::ResourceState as _;
*has_work = true;
let ref_count = cmdbuf.trackers.textures.get_ref_count(id);
//TODO: better error handling here?
// register it in the temporary tracker.
let mut ts = track::TextureState::default();
let _ = ts.change(
id,
texture.full_range.clone(),
hal::TextureUses::empty(), //present
None,
);
let _ = used_surface_textures.init(id, ref_count.clone(), ts);
true
}
}
};
if !texture.life_guard.use_at(submit_index) {
device.temp_suspected.textures.push(id);
}
if should_extend {
unsafe {
let ref_count = cmdbuf.trackers.textures.get_ref_count(id);
used_surface_textures
.merge_single(
&*texture_guard,
id,
None,
ref_count,
hal::TextureUses::PRESENT,
)
.unwrap();
};
}
}
for id in cmdbuf.trackers.views.used() {
if !texture_view_guard[id].life_guard.use_at(submit_index) {
@ -717,13 +723,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
sampler_guard[sub_id].life_guard.use_at(submit_index);
}
}
assert!(cmdbuf.trackers.samplers.is_empty());
for id in cmdbuf.trackers.compute_pipes.used() {
// assert!(cmdbuf.trackers.samplers.is_empty());
for id in cmdbuf.trackers.compute_pipelines.used() {
if !compute_pipe_guard[id].life_guard.use_at(submit_index) {
device.temp_suspected.compute_pipelines.push(id);
}
}
for id in cmdbuf.trackers.render_pipes.used() {
for id in cmdbuf.trackers.render_pipelines.used() {
if !render_pipe_guard[id].life_guard.use_at(submit_index) {
device.temp_suspected.render_pipelines.push(id);
}
@ -741,12 +747,12 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
// We need to update the submission indices for the contained
// state-less (!) resources as well, excluding the bind groups.
// They don't get deleted too early if the bundle goes out of scope.
for sub_id in bundle.used.compute_pipes.used() {
compute_pipe_guard[sub_id].life_guard.use_at(submit_index);
}
for sub_id in bundle.used.render_pipes.used() {
for sub_id in bundle.used.render_pipelines.used() {
render_pipe_guard[sub_id].life_guard.use_at(submit_index);
}
for sub_id in bundle.used.query_sets.used() {
query_set_guard[sub_id].life_guard.use_at(submit_index);
}
}
let mut baked = cmdbuf.into_baked();
@ -766,11 +772,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.map_err(|err| QueueSubmitError::DestroyedTexture(err.0))?;
//Note: stateless trackers are not merged:
// device already knows these resources exist.
CommandBuffer::insert_barriers(
CommandBuffer::insert_barriers_from_tracker(
&mut baked.encoder,
&mut *trackers,
&baked.trackers.buffers,
&baked.trackers.textures,
&baked.trackers,
&*buffer_guard,
&*texture_guard,
);
@ -788,19 +793,19 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
.begin_encoding(Some("(wgpu internal) Present"))
.map_err(DeviceError::from)?
};
let texture_barriers = trackers
trackers
.textures
.merge_replace(&used_surface_textures)
.map(|pending| {
let tex = &texture_guard[pending.id];
pending.into_hal(tex)
});
.set_from_usage_scope(&*texture_guard, &used_surface_textures);
let texture_barriers = trackers.textures.drain().map(|pending| {
let tex = unsafe { texture_guard.get_unchecked(pending.id) };
pending.into_hal(tex)
});
let present = unsafe {
baked.encoder.transition_textures(texture_barriers);
baked.encoder.end_encoding().unwrap()
};
baked.list.push(present);
used_surface_textures.clear();
used_surface_textures = track::TextureUsageScope::new();
}
// done
@ -810,7 +815,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
});
}
log::trace!("Device after submission {}: {:#?}", submit_index, trackers);
log::trace!("Device after submission {}", submit_index);
}
let super::Device {
@ -830,6 +835,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let (_, mut token) = hub.buffers.read(&mut token); // skip token
let (mut texture_guard, _) = hub.textures.write(&mut token);
used_surface_textures.set_size(texture_guard.len());
for &id in pending_writes.dst_textures.iter() {
let texture = texture_guard.get_mut(id).unwrap();
match texture.inner {
@ -840,39 +847,39 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
TextureInner::Surface {
ref mut has_work, ..
} => {
use track::ResourceState as _;
*has_work = true;
let ref_count = texture.life_guard.add_ref();
//TODO: better error handling here?
// register it in the temporary tracker.
let mut ts = track::TextureState::default();
let _ = ts.change(
id::Valid(id),
texture.full_range.clone(),
hal::TextureUses::empty(), //present
None,
);
let _ = used_surface_textures.init(id::Valid(id), ref_count, ts);
unsafe {
used_surface_textures
.merge_single(
&*texture_guard,
id::Valid(id),
None,
&ref_count,
hal::TextureUses::PRESENT,
)
.unwrap()
};
}
}
}
if !used_surface_textures.is_empty() {
let mut trackers = device.trackers.lock();
let texture_barriers = trackers
trackers
.textures
.merge_replace(&used_surface_textures)
.map(|pending| {
let tex = &texture_guard[pending.id];
pending.into_hal(tex)
});
.set_from_usage_scope(&*texture_guard, &used_surface_textures);
let texture_barriers = trackers.textures.drain().map(|pending| {
let tex = unsafe { texture_guard.get_unchecked(pending.id) };
pending.into_hal(tex)
});
unsafe {
pending_writes
.command_encoder
.transition_textures(texture_barriers);
};
used_surface_textures.clear();
}
}

135
wgpu-core/src/hub.rs
View File

@ -194,6 +194,14 @@ impl<T, I: id::TypedId> Storage<T, I> {
result
}
pub(crate) unsafe fn get_unchecked(&self, id: u32) -> &T {
match self.map[id as usize] {
Element::Occupied(ref v, _) => v,
Element::Vacant => panic!("{}[{}] does not exist", self.kind, id),
Element::Error(_, _) => panic!(""),
}
}
pub(crate) fn label_for_invalid_id(&self, id: I) -> &str {
let (index, _, _) = id.unzip();
match self.map.get(index as usize) {
@ -266,6 +274,10 @@ impl<T, I: id::TypedId> Storage<T, I> {
})
}
pub(crate) fn len(&self) -> usize {
self.map.len()
}
fn generate_report(&self) -> StorageReport {
let mut report = StorageReport {
element_size: mem::size_of::<T>(),
@ -298,56 +310,56 @@ pub enum Root {}
impl Access<Instance> for Root {}
impl Access<Surface> for Root {}
impl Access<Surface> for Instance {}
impl<A: hal::Api> Access<Adapter<A>> for Root {}
impl<A: hal::Api> Access<Adapter<A>> for Surface {}
impl<A: hal::Api> Access<Device<A>> for Root {}
impl<A: hal::Api> Access<Device<A>> for Surface {}
impl<A: hal::Api> Access<Device<A>> for Adapter<A> {}
impl<A: hal::Api> Access<PipelineLayout<A>> for Root {}
impl<A: hal::Api> Access<PipelineLayout<A>> for Device<A> {}
impl<A: hal::Api> Access<PipelineLayout<A>> for RenderBundle {}
impl<A: hal::Api> Access<BindGroupLayout<A>> for Root {}
impl<A: hal::Api> Access<BindGroupLayout<A>> for Device<A> {}
impl<A: hal::Api> Access<BindGroupLayout<A>> for PipelineLayout<A> {}
impl<A: hal::Api> Access<BindGroup<A>> for Root {}
impl<A: hal::Api> Access<BindGroup<A>> for Device<A> {}
impl<A: hal::Api> Access<BindGroup<A>> for BindGroupLayout<A> {}
impl<A: hal::Api> Access<BindGroup<A>> for PipelineLayout<A> {}
impl<A: hal::Api> Access<BindGroup<A>> for CommandBuffer<A> {}
impl<A: hal::Api> Access<CommandBuffer<A>> for Root {}
impl<A: hal::Api> Access<CommandBuffer<A>> for Device<A> {}
impl<A: hal::Api> Access<RenderBundle> for Device<A> {}
impl<A: hal::Api> Access<RenderBundle> for CommandBuffer<A> {}
impl<A: hal::Api> Access<ComputePipeline<A>> for Device<A> {}
impl<A: hal::Api> Access<ComputePipeline<A>> for BindGroup<A> {}
impl<A: hal::Api> Access<RenderPipeline<A>> for Device<A> {}
impl<A: hal::Api> Access<RenderPipeline<A>> for BindGroup<A> {}
impl<A: hal::Api> Access<RenderPipeline<A>> for ComputePipeline<A> {}
impl<A: hal::Api> Access<QuerySet<A>> for Root {}
impl<A: hal::Api> Access<QuerySet<A>> for Device<A> {}
impl<A: hal::Api> Access<QuerySet<A>> for CommandBuffer<A> {}
impl<A: hal::Api> Access<QuerySet<A>> for RenderPipeline<A> {}
impl<A: hal::Api> Access<QuerySet<A>> for ComputePipeline<A> {}
impl<A: hal::Api> Access<QuerySet<A>> for Sampler<A> {}
impl<A: hal::Api> Access<ShaderModule<A>> for Device<A> {}
impl<A: hal::Api> Access<ShaderModule<A>> for BindGroupLayout<A> {}
impl<A: hal::Api> Access<Buffer<A>> for Root {}
impl<A: hal::Api> Access<Buffer<A>> for Device<A> {}
impl<A: hal::Api> Access<Buffer<A>> for BindGroupLayout<A> {}
impl<A: hal::Api> Access<Buffer<A>> for BindGroup<A> {}
impl<A: hal::Api> Access<Buffer<A>> for CommandBuffer<A> {}
impl<A: hal::Api> Access<Buffer<A>> for ComputePipeline<A> {}
impl<A: hal::Api> Access<Buffer<A>> for RenderPipeline<A> {}
impl<A: hal::Api> Access<Buffer<A>> for QuerySet<A> {}
impl<A: hal::Api> Access<Texture<A>> for Root {}
impl<A: hal::Api> Access<Texture<A>> for Device<A> {}
impl<A: hal::Api> Access<Texture<A>> for Buffer<A> {}
impl<A: hal::Api> Access<TextureView<A>> for Root {}
impl<A: hal::Api> Access<TextureView<A>> for Device<A> {}
impl<A: hal::Api> Access<TextureView<A>> for Texture<A> {}
impl<A: hal::Api> Access<Sampler<A>> for Root {}
impl<A: hal::Api> Access<Sampler<A>> for Device<A> {}
impl<A: hal::Api> Access<Sampler<A>> for TextureView<A> {}
impl<A: HalApi> Access<Adapter<A>> for Root {}
impl<A: HalApi> Access<Adapter<A>> for Surface {}
impl<A: HalApi> Access<Device<A>> for Root {}
impl<A: HalApi> Access<Device<A>> for Surface {}
impl<A: HalApi> Access<Device<A>> for Adapter<A> {}
impl<A: HalApi> Access<PipelineLayout<A>> for Root {}
impl<A: HalApi> Access<PipelineLayout<A>> for Device<A> {}
impl<A: HalApi> Access<PipelineLayout<A>> for RenderBundle<A> {}
impl<A: HalApi> Access<BindGroupLayout<A>> for Root {}
impl<A: HalApi> Access<BindGroupLayout<A>> for Device<A> {}
impl<A: HalApi> Access<BindGroupLayout<A>> for PipelineLayout<A> {}
impl<A: HalApi> Access<BindGroup<A>> for Root {}
impl<A: HalApi> Access<BindGroup<A>> for Device<A> {}
impl<A: HalApi> Access<BindGroup<A>> for BindGroupLayout<A> {}
impl<A: HalApi> Access<BindGroup<A>> for PipelineLayout<A> {}
impl<A: HalApi> Access<BindGroup<A>> for CommandBuffer<A> {}
impl<A: HalApi> Access<CommandBuffer<A>> for Root {}
impl<A: HalApi> Access<CommandBuffer<A>> for Device<A> {}
impl<A: HalApi> Access<RenderBundle<A>> for Device<A> {}
impl<A: HalApi> Access<RenderBundle<A>> for CommandBuffer<A> {}
impl<A: HalApi> Access<ComputePipeline<A>> for Device<A> {}
impl<A: HalApi> Access<ComputePipeline<A>> for BindGroup<A> {}
impl<A: HalApi> Access<RenderPipeline<A>> for Device<A> {}
impl<A: HalApi> Access<RenderPipeline<A>> for BindGroup<A> {}
impl<A: HalApi> Access<RenderPipeline<A>> for ComputePipeline<A> {}
impl<A: HalApi> Access<QuerySet<A>> for Root {}
impl<A: HalApi> Access<QuerySet<A>> for Device<A> {}
impl<A: HalApi> Access<QuerySet<A>> for CommandBuffer<A> {}
impl<A: HalApi> Access<QuerySet<A>> for RenderPipeline<A> {}
impl<A: HalApi> Access<QuerySet<A>> for ComputePipeline<A> {}
impl<A: HalApi> Access<QuerySet<A>> for Sampler<A> {}
impl<A: HalApi> Access<ShaderModule<A>> for Device<A> {}
impl<A: HalApi> Access<ShaderModule<A>> for BindGroupLayout<A> {}
impl<A: HalApi> Access<Buffer<A>> for Root {}
impl<A: HalApi> Access<Buffer<A>> for Device<A> {}
impl<A: HalApi> Access<Buffer<A>> for BindGroupLayout<A> {}
impl<A: HalApi> Access<Buffer<A>> for BindGroup<A> {}
impl<A: HalApi> Access<Buffer<A>> for CommandBuffer<A> {}
impl<A: HalApi> Access<Buffer<A>> for ComputePipeline<A> {}
impl<A: HalApi> Access<Buffer<A>> for RenderPipeline<A> {}
impl<A: HalApi> Access<Buffer<A>> for QuerySet<A> {}
impl<A: HalApi> Access<Texture<A>> for Root {}
impl<A: HalApi> Access<Texture<A>> for Device<A> {}
impl<A: HalApi> Access<Texture<A>> for Buffer<A> {}
impl<A: HalApi> Access<TextureView<A>> for Root {}
impl<A: HalApi> Access<TextureView<A>> for Device<A> {}
impl<A: HalApi> Access<TextureView<A>> for Texture<A> {}
impl<A: HalApi> Access<Sampler<A>> for Root {}
impl<A: HalApi> Access<Sampler<A>> for Device<A> {}
impl<A: HalApi> Access<Sampler<A>> for TextureView<A> {}
#[cfg(debug_assertions)]
thread_local! {
@ -614,7 +626,7 @@ impl HubReport {
}
}
pub struct Hub<A: hal::Api, F: GlobalIdentityHandlerFactory> {
pub struct Hub<A: HalApi, F: GlobalIdentityHandlerFactory> {
pub adapters: Registry<Adapter<A>, id::AdapterId, F>,
pub devices: Registry<Device<A>, id::DeviceId, F>,
pub pipeline_layouts: Registry<PipelineLayout<A>, id::PipelineLayoutId, F>,
@ -622,7 +634,7 @@ pub struct Hub<A: hal::Api, F: GlobalIdentityHandlerFactory> {
pub bind_group_layouts: Registry<BindGroupLayout<A>, id::BindGroupLayoutId, F>,
pub bind_groups: Registry<BindGroup<A>, id::BindGroupId, F>,
pub command_buffers: Registry<CommandBuffer<A>, id::CommandBufferId, F>,
pub render_bundles: Registry<RenderBundle, id::RenderBundleId, F>,
pub render_bundles: Registry<RenderBundle<A>, id::RenderBundleId, F>,
pub render_pipelines: Registry<RenderPipeline<A>, id::RenderPipelineId, F>,
pub compute_pipelines: Registry<ComputePipeline<A>, id::ComputePipelineId, F>,
pub query_sets: Registry<QuerySet<A>, id::QuerySetId, F>,
@ -1008,6 +1020,25 @@ pub trait HalApi: hal::Api {
fn get_surface_mut(surface: &mut Surface) -> &mut HalSurface<Self>;
}
impl HalApi for hal::api::Empty {
const VARIANT: Backend = Backend::Empty;
fn create_instance_from_hal(_: &str, _: Self::Instance) -> Instance {
unimplemented!("called empty api")
}
fn instance_as_hal(_: &Instance) -> Option<&Self::Instance> {
unimplemented!("called empty api")
}
fn hub<G: GlobalIdentityHandlerFactory>(_: &Global<G>) -> &Hub<Self, G> {
unimplemented!("called empty api")
}
fn get_surface(_: &Surface) -> &HalSurface<Self> {
unimplemented!("called empty api")
}
fn get_surface_mut(_: &mut Surface) -> &mut HalSurface<Self> {
unimplemented!("called empty api")
}
}
#[cfg(vulkan)]
impl HalApi for hal::api::Vulkan {
const VARIANT: Backend = Backend::Vulkan;

10
wgpu-core/src/id.rs
View File

@ -64,9 +64,9 @@ impl<T> From<SerialId> for Id<T> {
}
impl<T> Id<T> {
#[cfg(test)]
pub(crate) fn dummy() -> Valid<Self> {
Valid(Id(NonZeroId::new(1).unwrap(), PhantomData))
#[allow(dead_code)]
pub(crate) fn dummy(index: u32) -> Valid<Self> {
Valid(Id::zip(index, 1, Backend::Empty))
}
pub fn backend(self) -> Backend {
@ -135,7 +135,7 @@ pub(crate) struct Valid<I>(pub I);
/// Most `wgpu-core` clients should not use this trait. Unusual clients that
/// need to construct `Id` values directly, or access their components, like the
/// WGPU recording player, may use this trait to do so.
pub trait TypedId {
pub trait TypedId: Copy {
fn zip(index: Index, epoch: Epoch, backend: Backend) -> Self;
fn unzip(self) -> (Index, Epoch, Backend);
}
@ -184,7 +184,7 @@ pub type CommandBufferId = Id<crate::command::CommandBuffer<Dummy>>;
pub type RenderPassEncoderId = *mut crate::command::RenderPass;
pub type ComputePassEncoderId = *mut crate::command::ComputePass;
pub type RenderBundleEncoderId = *mut crate::command::RenderBundleEncoder;
pub type RenderBundleId = Id<crate::command::RenderBundle>;
pub type RenderBundleId = Id<crate::command::RenderBundle<Dummy>>;
pub type QuerySetId = Id<crate::resource::QuerySet<Dummy>>;
#[test]

2
wgpu-core/src/init_tracker/texture.rs
View File

@ -26,7 +26,7 @@ pub(crate) fn has_copy_partial_init_tracker_coverage(
impl From<TextureSelector> for TextureInitRange {
fn from(selector: TextureSelector) -> Self {
TextureInitRange {
mip_range: selector.levels,
mip_range: selector.mips,
layer_range: selector.layers,
}
}

3
wgpu-core/src/instance.rs
View File

@ -494,6 +494,9 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
}
#[cfg(dx12)]
/// # Safety
///
/// The visual must be valid and able to be used to make a swapchain with.
pub unsafe fn instance_create_surface_from_visual(
&self,
visual: *mut std::ffi::c_void,

4
wgpu-core/src/lib.rs
View File

@ -4,6 +4,8 @@
*/
#![allow(
// It is much clearer to assert negative conditions with eq! false
clippy::bool_assert_comparison,
// We use loops for getting early-out of scope without closures.
clippy::never_loop,
// We don't use syntax sugar where it's not necessary.
@ -16,6 +18,8 @@
clippy::new_without_default,
// Needless updates are more scaleable, easier to play with features.
clippy::needless_update,
// Need many arguments for some core functions to be able to re-use code in many situations.
clippy::too_many_arguments,
// For some reason `rustc` can warn about these in const generics even
// though they are required.
unused_braces,

19
wgpu-core/src/present.rs
View File

@ -174,7 +174,7 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
initialization_status: TextureInitTracker::new(1, 1),
full_range: track::TextureSelector {
layers: 0..1,
levels: 0..1,
mips: 0..1,
},
life_guard: LifeGuard::new("<Surface>"),
clear_mode: resource::TextureClearMode::RenderPass {
@ -187,20 +187,13 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
let id = fid.assign(texture, &mut token);
{
use track::ResourceState as _;
// register it in the device tracker as uninitialized
let mut trackers = device.trackers.lock();
let mut ts = track::TextureState::default();
let _ = ts.change(
id,
track::TextureSelector {
layers: 0..1,
levels: 0..1,
},
trackers.textures.insert_single(
id.0,
ref_count.clone(),
hal::TextureUses::UNINITIALIZED,
None,
);
let _ = trackers.textures.init(id, ref_count.clone(), ts);
}
if present.acquired_texture.is_some() {
@ -273,6 +266,10 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
// The texture ID got added to the device tracker by `submit()`,
// and now we are moving it away.
log::debug!(
"Removing swapchain texture {:?} from the device tracker",
texture_id.value
);
device.trackers.lock().textures.remove(texture_id.value);
let (texture, _) = hub.textures.unregister(texture_id.value.0, &mut token);

25
wgpu-core/src/resource.rs
View File

@ -3,7 +3,7 @@ use crate::{
hub::{Global, GlobalIdentityHandlerFactory, HalApi, Resource, Token},
id::{AdapterId, DeviceId, SurfaceId, TextureId, Valid},
init_tracker::{BufferInitTracker, TextureInitTracker},
track::{TextureSelector, DUMMY_SELECTOR},
track::TextureSelector,
validation::MissingBufferUsageError,
Label, LifeGuard, RefCount, Stored,
};
@ -149,12 +149,6 @@ impl<A: hal::Api> Resource for Buffer<A> {
}
}
impl<A: hal::Api> Borrow<()> for Buffer<A> {
fn borrow(&self) -> &() {
&DUMMY_SELECTOR
}
}
pub type TextureDescriptor<'a> = wgt::TextureDescriptor<Label<'a>>;
#[derive(Debug)]
@ -448,12 +442,6 @@ impl<A: hal::Api> Resource for TextureView<A> {
}
}
impl<A: hal::Api> Borrow<()> for TextureView<A> {
fn borrow(&self) -> &() {
&DUMMY_SELECTOR
}
}
/// Describes a [`Sampler`]
#[derive(Clone, Debug, PartialEq)]
#[cfg_attr(feature = "trace", derive(serde::Serialize))]
@ -530,11 +518,6 @@ impl<A: hal::Api> Resource for Sampler<A> {
}
}
impl<A: hal::Api> Borrow<()> for Sampler<A> {
fn borrow(&self) -> &() {
&DUMMY_SELECTOR
}
}
#[derive(Clone, Debug, Error)]
pub enum CreateQuerySetError {
#[error(transparent)]
@ -565,12 +548,6 @@ impl<A: hal::Api> Resource for QuerySet<A> {
}
}
impl<A: hal::Api> Borrow<()> for QuerySet<A> {
fn borrow(&self) -> &() {
&DUMMY_SELECTOR
}
}
#[derive(Clone, Debug, Error)]
pub enum DestroyError {
#[error("resource is invalid")]

966
wgpu-core/src/track/buffer.rs
View File

@ -1,236 +1,778 @@
use super::{PendingTransition, ResourceState, Unit};
use crate::id::{BufferId, Valid};
/*! Buffer Trackers
*
* Buffers are represented by a single state for the whole resource,
* a 16 bit bitflag of buffer usages. Because there is only ever
* one subresource, they have no selector.
!*/
use std::{borrow::Cow, marker::PhantomData, vec::Drain};
use super::PendingTransition;
use crate::{
hub,
id::{BufferId, TypedId, Valid},
resource::Buffer,
track::{
invalid_resource_state, iterate_bitvec_indices, skip_barrier, ResourceMetadata,
ResourceMetadataProvider, ResourceUses, UsageConflict,
},
LifeGuard, RefCount,
};
use hal::BufferUses;
pub(crate) type BufferState = Unit<BufferUses>;
impl ResourceUses for BufferUses {
const EXCLUSIVE: Self = Self::EXCLUSIVE;
impl PendingTransition<BufferState> {
fn collapse(self) -> Result<BufferUses, Self> {
if self.usage.start.is_empty()
|| self.usage.start == self.usage.end
|| !BufferUses::EXCLUSIVE.intersects(self.usage.start | self.usage.end)
{
Ok(self.usage.start | self.usage.end)
} else {
Err(self)
}
}
}
impl Default for BufferState {
fn default() -> Self {
Self {
first: None,
last: BufferUses::empty(),
}
}
}
impl BufferState {
pub fn with_usage(usage: BufferUses) -> Self {
Unit::new(usage)
}
}
impl ResourceState for BufferState {
type Id = BufferId;
type Selector = ();
type Usage = BufferUses;
fn query(&self, _selector: Self::Selector) -> Option<Self::Usage> {
Some(self.last)
fn bits(self) -> u16 {
Self::bits(&self)
}
fn change(
&mut self,
id: Valid<Self::Id>,
_selector: Self::Selector,
usage: Self::Usage,
output: Option<&mut Vec<PendingTransition<Self>>>,
) -> Result<(), PendingTransition<Self>> {
let old = self.last;
if old != usage || !BufferUses::ORDERED.contains(usage) {
let pending = PendingTransition {
id,
selector: (),
usage: old..usage,
};
*self = match output {
None => {
assert_eq!(
self.first, None,
"extending a state that is already a transition"
);
Unit::new(pending.collapse()?)
}
Some(transitions) => {
transitions.push(pending);
Unit {
first: self.first.or(Some(old)),
last: usage,
}
}
};
fn all_ordered(self) -> bool {
Self::ORDERED.contains(self)
}
fn any_exclusive(self) -> bool {
self.intersects(Self::EXCLUSIVE)
}
}
/// Stores all the buffers that a bind group stores.
pub(crate) struct BufferBindGroupState<A: hub::HalApi> {
buffers: Vec<(Valid<BufferId>, RefCount, BufferUses)>,
_phantom: PhantomData<A>,
}
impl<A: hub::HalApi> BufferBindGroupState<A> {
pub fn new() -> Self {
Self {
buffers: Vec::new(),
_phantom: PhantomData,
}
}
/// Optimize the buffer bind group state by sorting it by ID.
///
/// When this list of states is merged into a tracker, the memory
/// accesses will be in a constant assending order.
pub(crate) fn optimize(&mut self) {
self.buffers
.sort_unstable_by_key(|&(id, _, _)| id.0.unzip().0);
}
/// Returns a list of all buffers tracked. May contain duplicates.
pub fn used(&self) -> impl Iterator<Item = Valid<BufferId>> + '_ {
self.buffers.iter().map(|&(id, _, _)| id)
}
/// Adds the given resource with the given state.
pub fn add_single<'a>(
&mut self,
storage: &'a hub::Storage<Buffer<A>, BufferId>,
id: BufferId,
state: BufferUses,
) -> Option<&'a Buffer<A>> {
let buffer = storage.get(id).ok()?;
self.buffers
.push((Valid(id), buffer.life_guard.add_ref(), state));
Some(buffer)
}
}
/// Stores all buffer state within a single usage scope.
#[derive(Debug)]
pub(crate) struct BufferUsageScope<A: hub::HalApi> {
state: Vec<BufferUses>,
metadata: ResourceMetadata<A>,
}
impl<A: hub::HalApi> BufferUsageScope<A> {
pub fn new() -> Self {
Self {
state: Vec::new(),
metadata: ResourceMetadata::new(),
}
}
fn debug_assert_in_bounds(&self, index: usize) {
debug_assert!(index < self.state.len());
self.metadata.debug_assert_in_bounds(index);
}
/// Sets the size of all the vectors inside the tracker.
///
/// Must be called with the highest possible Buffer ID before
/// all unsafe functions are called.
pub fn set_size(&mut self, size: usize) {
self.state.resize(size, BufferUses::empty());
self.metadata.set_size(size);
}
/// Extend the vectors to let the given index be valid.
fn allow_index(&mut self, index: usize) {
if index >= self.state.len() {
self.set_size(index + 1);
}
}
/// Returns a list of all buffers tracked.
pub fn used(&self) -> impl Iterator<Item = Valid<BufferId>> + '_ {
self.metadata.used()
}
/// Merge the list of buffer states in the given bind group into this usage scope.
///
/// If any of the resulting states is invalid, stops the merge and returns a usage
/// conflict with the details of the invalid state.
///
/// Because bind groups do not check if the union of all their states is valid,
/// this method is allowed to return Err on the first bind group bound.
///
/// # Safety
///
/// [`Self::set_size`] must be called with the maximum possible Buffer ID before this
/// method is called.
pub unsafe fn merge_bind_group(
&mut self,
bind_group: &BufferBindGroupState<A>,
) -> Result<(), UsageConflict> {
for &(id, ref ref_count, state) in &bind_group.buffers {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
insert_or_merge(
None,
None,
&mut self.state,
&mut self.metadata,
index32,
index,
BufferStateProvider::Direct { state },
ResourceMetadataProvider::Direct {
epoch,
ref_count: Cow::Borrowed(ref_count),
},
)?;
}
Ok(())
}
fn merge(
&mut self,
id: Valid<Self::Id>,
other: &Self,
output: Option<&mut Vec<PendingTransition<Self>>>,
) -> Result<(), PendingTransition<Self>> {
let old = self.last;
let new = other.port();
if old == new && BufferUses::ORDERED.contains(new) {
if output.is_some() && self.first.is_none() {
*self = Unit {
first: Some(old),
last: other.last,
};
}
} else {
let pending = PendingTransition {
id,
selector: (),
usage: old..new,
};
*self = match output {
None => {
assert_eq!(
self.first, None,
"extending a state that is already a transition"
);
Unit::new(pending.collapse()?)
}
Some(transitions) => {
transitions.push(pending);
Unit {
first: self.first.or(Some(old)),
last: other.last,
}
}
/// Merge the list of buffer states in the given usage scope into this UsageScope.
///
/// If any of the resulting states is invalid, stops the merge and returns a usage
/// conflict with the details of the invalid state.
///
/// If the given tracker uses IDs higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn merge_usage_scope(&mut self, scope: &Self) -> Result<(), UsageConflict> {
let incoming_size = scope.state.len();
if incoming_size > self.state.len() {
self.set_size(incoming_size);
}
for index in iterate_bitvec_indices(&scope.metadata.owned) {
self.debug_assert_in_bounds(index);
scope.debug_assert_in_bounds(index);
unsafe {
insert_or_merge(
None,
None,
&mut self.state,
&mut self.metadata,
index as u32,
index,
BufferStateProvider::Indirect {
state: &scope.state,
},
ResourceMetadataProvider::Indirect {
metadata: &scope.metadata,
},
)?;
};
}
Ok(())
}
fn optimize(&mut self) {}
}
/// Merge a single state into the UsageScope.
///
/// If the resulting state is invalid, returns a usage
/// conflict with the details of the invalid state.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn merge_single<'a>(
&mut self,
storage: &'a hub::Storage<Buffer<A>, BufferId>,
id: BufferId,
new_state: BufferUses,
) -> Result<&'a Buffer<A>, UsageConflict> {
let buffer = storage
.get(id)
.map_err(|_| UsageConflict::BufferInvalid { id })?;
#[cfg(test)]
mod test {
use super::*;
use crate::id::Id;
let (index32, epoch, _) = id.unzip();
let index = index32 as usize;
#[test]
fn change_extend() {
let mut bs = Unit {
first: None,
last: BufferUses::INDEX,
};
let id = Id::dummy();
assert_eq!(
bs.change(id, (), BufferUses::STORAGE_WRITE, None),
Err(PendingTransition {
id,
selector: (),
usage: BufferUses::INDEX..BufferUses::STORAGE_WRITE,
}),
);
bs.change(id, (), BufferUses::VERTEX, None).unwrap();
bs.change(id, (), BufferUses::INDEX, None).unwrap();
assert_eq!(bs, Unit::new(BufferUses::VERTEX | BufferUses::INDEX));
}
self.allow_index(index);
#[test]
fn change_replace() {
let mut bs = Unit {
first: None,
last: BufferUses::STORAGE_WRITE,
};
let id = Id::dummy();
let mut list = Vec::new();
bs.change(id, (), BufferUses::VERTEX, Some(&mut list))
.unwrap();
assert_eq!(
&list,
&[PendingTransition {
id,
selector: (),
usage: BufferUses::STORAGE_WRITE..BufferUses::VERTEX,
}],
);
assert_eq!(
bs,
Unit {
first: Some(BufferUses::STORAGE_WRITE),
last: BufferUses::VERTEX,
}
);
self.debug_assert_in_bounds(index);
list.clear();
bs.change(id, (), BufferUses::STORAGE_WRITE, Some(&mut list))
.unwrap();
assert_eq!(
&list,
&[PendingTransition {
id,
selector: (),
usage: BufferUses::VERTEX..BufferUses::STORAGE_WRITE,
}],
);
assert_eq!(
bs,
Unit {
first: Some(BufferUses::STORAGE_WRITE),
last: BufferUses::STORAGE_WRITE,
}
);
}
unsafe {
insert_or_merge(
Some(&buffer.life_guard),
None,
&mut self.state,
&mut self.metadata,
index32,
index,
BufferStateProvider::Direct { state: new_state },
ResourceMetadataProvider::Resource { epoch },
)?;
}
#[test]
fn merge_replace() {
let mut bs = Unit {
first: None,
last: BufferUses::empty(),
};
let other_smooth = Unit {
first: Some(BufferUses::empty()),
last: BufferUses::COPY_DST,
};
let id = Id::dummy();
let mut list = Vec::new();
bs.merge(id, &other_smooth, Some(&mut list)).unwrap();
assert!(list.is_empty());
assert_eq!(
bs,
Unit {
first: Some(BufferUses::empty()),
last: BufferUses::COPY_DST,
}
);
let other_rough = Unit {
first: Some(BufferUses::empty()),
last: BufferUses::UNIFORM,
};
bs.merge(id, &other_rough, Some(&mut list)).unwrap();
assert_eq!(
&list,
&[PendingTransition {
id,
selector: (),
usage: BufferUses::COPY_DST..BufferUses::empty(),
}],
);
assert_eq!(
bs,
Unit {
first: Some(BufferUses::empty()),
last: BufferUses::UNIFORM,
}
);
Ok(buffer)
}
}
/// Stores all buffer state within a command buffer or device.
pub(crate) struct BufferTracker<A: hub::HalApi> {
start: Vec<BufferUses>,
end: Vec<BufferUses>,
metadata: ResourceMetadata<A>,
temp: Vec<PendingTransition<BufferUses>>,
}
impl<A: hub::HalApi> BufferTracker<A> {
pub fn new() -> Self {
Self {
start: Vec::new(),
end: Vec::new(),
metadata: ResourceMetadata::new(),
temp: Vec::new(),
}
}
fn debug_assert_in_bounds(&self, index: usize) {
debug_assert!(index < self.start.len());
debug_assert!(index < self.end.len());
self.metadata.debug_assert_in_bounds(index);
}
/// Sets the size of all the vectors inside the tracker.
///
/// Must be called with the highest possible Buffer ID before
/// all unsafe functions are called.
pub fn set_size(&mut self, size: usize) {
self.start.resize(size, BufferUses::empty());
self.end.resize(size, BufferUses::empty());
self.metadata.set_size(size);
}
/// Extend the vectors to let the given index be valid.
fn allow_index(&mut self, index: usize) {
if index >= self.start.len() {
self.set_size(index + 1);
}
}
/// Returns a list of all buffers tracked.
pub fn used(&self) -> impl Iterator<Item = Valid<BufferId>> + '_ {
self.metadata.used()
}
/// Drains all currently pending transitions.
pub fn drain(&mut self) -> Drain<'_, PendingTransition<BufferUses>> {
self.temp.drain(..)
}
/// Inserts a single buffer and its state into the resource tracker.
///
/// If the resource already exists in the tracker, this will panic.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn insert_single(&mut self, id: Valid<BufferId>, ref_count: RefCount, state: BufferUses) {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
self.allow_index(index);
self.debug_assert_in_bounds(index);
unsafe {
let currently_owned = self.metadata.owned.get(index).unwrap_unchecked();
if currently_owned {
panic!("Tried to insert buffer already tracked");
}
insert(
None,
Some(&mut self.start),
&mut self.end,
&mut self.metadata,
index,
BufferStateProvider::Direct { state },
None,
ResourceMetadataProvider::Direct {
epoch,
ref_count: Cow::Owned(ref_count),
},
)
}
}
/// Sets the state of a single buffer.
///
/// If a transition is needed to get the buffer into the given state, that transition
/// is returned. No more than one transition is needed.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn set_single<'a>(
&mut self,
storage: &'a hub::Storage<Buffer<A>, BufferId>,
id: BufferId,
state: BufferUses,
) -> Option<(&'a Buffer<A>, Option<PendingTransition<BufferUses>>)> {
let value = storage.get(id).ok()?;
let (index32, epoch, _) = id.unzip();
let index = index32 as usize;
self.allow_index(index);
self.debug_assert_in_bounds(index);
unsafe {
insert_or_barrier_update(
Some(&value.life_guard),
Some(&mut self.start),
&mut self.end,
&mut self.metadata,
index32,
index,
BufferStateProvider::Direct { state },
None,
ResourceMetadataProvider::Resource { epoch },
&mut self.temp,
)
};
debug_assert!(self.temp.len() <= 1);
Some((value, self.temp.pop()))
}
/// Sets the given state for all buffers in the given tracker.
///
/// If a transition is needed to get the buffers into the needed state,
/// those transitions are stored within the tracker. A subsequent
/// call to [`Self::drain`] is needed to get those transitions.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn set_from_tracker(&mut self, tracker: &Self) {
let incoming_size = tracker.start.len();
if incoming_size > self.start.len() {
self.set_size(incoming_size);
}
for index in iterate_bitvec_indices(&tracker.metadata.owned) {
self.debug_assert_in_bounds(index);
tracker.debug_assert_in_bounds(index);
unsafe {
insert_or_barrier_update(
None,
Some(&mut self.start),
&mut self.end,
&mut self.metadata,
index as u32,
index,
BufferStateProvider::Indirect {
state: &tracker.start,
},
Some(BufferStateProvider::Indirect {
state: &tracker.end,
}),
ResourceMetadataProvider::Indirect {
metadata: &tracker.metadata,
},
&mut self.temp,
)
}
}
}
/// Sets the given state for all buffers in the given UsageScope.
///
/// If a transition is needed to get the buffers into the needed state,
/// those transitions are stored within the tracker. A subsequent
/// call to [`Self::drain`] is needed to get those transitions.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn set_from_usage_scope(&mut self, scope: &BufferUsageScope<A>) {
let incoming_size = scope.state.len();
if incoming_size > self.start.len() {
self.set_size(incoming_size);
}
for index in iterate_bitvec_indices(&scope.metadata.owned) {
self.debug_assert_in_bounds(index);
scope.debug_assert_in_bounds(index);
unsafe {
insert_or_barrier_update(
None,
Some(&mut self.start),
&mut self.end,
&mut self.metadata,
index as u32,
index,
BufferStateProvider::Indirect {
state: &scope.state,
},
None,
ResourceMetadataProvider::Indirect {
metadata: &scope.metadata,
},
&mut self.temp,
)
}
}
}
/// Iterates through all buffers in the given bind group and adopts
/// the state given for those buffers in the UsageScope. It also
/// removes all touched buffers from the usage scope.
///
/// If a transition is needed to get the buffers into the needed state,
/// those transitions are stored within the tracker. A subsequent
/// call to [`Self::drain`] is needed to get those transitions.
///
/// This is a really funky method used by Compute Passes to generate
/// barriers after a call to dispatch without needing to iterate
/// over all elements in the usage scope. We use each the
/// bind group as a source of which IDs to look at. The bind groups
/// must have first been added to the usage scope.
///
/// # Safety
///
/// [`Self::set_size`] must be called with the maximum possible Buffer ID before this
/// method is called.
pub unsafe fn set_and_remove_from_usage_scope_sparse(
&mut self,
scope: &mut BufferUsageScope<A>,
bind_group_state: &BufferBindGroupState<A>,
) {
let incoming_size = scope.state.len();
if incoming_size > self.start.len() {
self.set_size(incoming_size);
}
for &(id, ref ref_count, _) in bind_group_state.buffers.iter() {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
scope.debug_assert_in_bounds(index);
if !scope.metadata.owned.get(index).unwrap_unchecked() {
continue;
}
insert_or_barrier_update(
None,
Some(&mut self.start),
&mut self.end,
&mut self.metadata,
index as u32,
index,
BufferStateProvider::Indirect {
state: &scope.state,
},
None,
ResourceMetadataProvider::Direct {
epoch,
ref_count: Cow::Borrowed(ref_count),
},
&mut self.temp,
);
scope.metadata.reset(index);
}
}
/// Removes the given resource from the tracker iff we have the last reference to the
/// resource and the epoch matches.
///
/// Returns true if the resource was removed.
///
/// If the ID is higher than the length of internal vectors,
/// false will be returned.
pub fn remove_abandoned(&mut self, id: Valid<BufferId>) -> bool {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
if index > self.metadata.owned.len() {
return false;
}
self.debug_assert_in_bounds(index);
unsafe {
if self.metadata.owned.get(index).unwrap_unchecked() {
let existing_epoch = self.metadata.epochs.get_unchecked_mut(index);
let existing_ref_count = self.metadata.ref_counts.get_unchecked_mut(index);
if *existing_epoch == epoch
&& existing_ref_count.as_mut().unwrap_unchecked().load() == 1
{
self.metadata.reset(index);
return true;
}
}
}
false
}
}
/// Source of Buffer State.
#[derive(Debug, Clone)]
enum BufferStateProvider<'a> {
/// Get a state that was provided directly.
Direct { state: BufferUses },
/// Get a state from an an array of states.
Indirect { state: &'a [BufferUses] },
}
impl BufferStateProvider<'_> {
/// Gets the state from the provider, given a resource ID index.
///
/// # Safety
///
/// Index must be in bounds for the indirect source iff this is in the indirect state.
#[inline(always)]
unsafe fn get_state(&self, index: usize) -> BufferUses {
match *self {
BufferStateProvider::Direct { state } => state,
BufferStateProvider::Indirect { state } => {
debug_assert!(index < state.len());
*state.get_unchecked(index)
}
}
}
}
/// Does an insertion operation if the index isn't tracked
/// in the current metadata, otherwise merges the given state
/// with the current state. If the merging would cause
/// a conflict, returns that usage conflict.
///
/// # Safety
///
/// Indexes must be valid indexes into all arrays passed in
/// to this function, either directly or via metadata or provider structs.
#[inline(always)]
unsafe fn insert_or_merge<A: hub::HalApi>(
life_guard: Option<&LifeGuard>,
start_states: Option<&mut [BufferUses]>,
current_states: &mut [BufferUses],
resource_metadata: &mut ResourceMetadata<A>,
index32: u32,
index: usize,
state_provider: BufferStateProvider<'_>,
metadata_provider: ResourceMetadataProvider<'_, A>,
) -> Result<(), UsageConflict> {
let currently_owned = resource_metadata.owned.get(index).unwrap_unchecked();
if !currently_owned {
insert(
life_guard,
start_states,
current_states,
resource_metadata,
index,
state_provider,
None,
metadata_provider,
);
return Ok(());
}
merge(
current_states,
index32,
index,
state_provider,
metadata_provider,
)
}
/// If the resource isn't tracked
/// - Inserts the given resource.
/// - Uses the `start_state_provider` to populate `start_states`
/// - Uses either `end_state_provider` or `start_state_provider`
/// to populate `current_states`.
/// If the resource is tracked
/// - Inserts barriers from the state in `current_states`
/// to the state provided by `start_state_provider`.
/// - Updates the `current_states` with either the state from
/// `end_state_provider` or `start_state_provider`.
///
/// Any barriers are added to the barrier vector.
///
/// # Safety
///
/// Indexes must be valid indexes into all arrays passed in
/// to this function, either directly or via metadata or provider structs.
#[inline(always)]
unsafe fn insert_or_barrier_update<A: hub::HalApi>(
life_guard: Option<&LifeGuard>,
start_states: Option<&mut [BufferUses]>,
current_states: &mut [BufferUses],
resource_metadata: &mut ResourceMetadata<A>,
index32: u32,
index: usize,
start_state_provider: BufferStateProvider<'_>,
end_state_provider: Option<BufferStateProvider<'_>>,
metadata_provider: ResourceMetadataProvider<'_, A>,
barriers: &mut Vec<PendingTransition<BufferUses>>,
) {
let currently_owned = resource_metadata.owned.get(index).unwrap_unchecked();
if !currently_owned {
insert(
life_guard,
start_states,
current_states,
resource_metadata,
index,
start_state_provider,
end_state_provider,
metadata_provider,
);
return;
}
let update_state_provider = end_state_provider.unwrap_or_else(|| start_state_provider.clone());
barrier(
current_states,
index32,
index,
start_state_provider,
barriers,
);
update(current_states, index, update_state_provider);
}
#[inline(always)]
unsafe fn insert<A: hub::HalApi>(
life_guard: Option<&LifeGuard>,
start_states: Option<&mut [BufferUses]>,
current_states: &mut [BufferUses],
resource_metadata: &mut ResourceMetadata<A>,
index: usize,
start_state_provider: BufferStateProvider<'_>,
end_state_provider: Option<BufferStateProvider<'_>>,
metadata_provider: ResourceMetadataProvider<'_, A>,
) {
let new_start_state = start_state_provider.get_state(index);
let new_end_state = end_state_provider.map_or(new_start_state, |p| p.get_state(index));
// This should only ever happen with a wgpu bug, but let's just double
// check that resource states don't have any conflicts.
debug_assert_eq!(invalid_resource_state(new_start_state), false);
debug_assert_eq!(invalid_resource_state(new_end_state), false);
log::trace!("\tbuf {index}: insert {new_start_state:?}..{new_end_state:?}");
if let Some(&mut ref mut start_state) = start_states {
*start_state.get_unchecked_mut(index) = new_start_state;
}
*current_states.get_unchecked_mut(index) = new_end_state;
let (epoch, ref_count) = metadata_provider.get_own(life_guard, index);
resource_metadata.owned.set(index, true);
*resource_metadata.epochs.get_unchecked_mut(index) = epoch;
*resource_metadata.ref_counts.get_unchecked_mut(index) = Some(ref_count);
}
#[inline(always)]
unsafe fn merge<A: hub::HalApi>(
current_states: &mut [BufferUses],
index32: u32,
index: usize,
state_provider: BufferStateProvider<'_>,
metadata_provider: ResourceMetadataProvider<'_, A>,
) -> Result<(), UsageConflict> {
let current_state = current_states.get_unchecked_mut(index);
let new_state = state_provider.get_state(index);
let merged_state = *current_state | new_state;
if invalid_resource_state(merged_state) {
return Err(UsageConflict::from_buffer(
BufferId::zip(index32, metadata_provider.get_epoch(index), A::VARIANT),
*current_state,
new_state,
));
}
log::trace!("\tbuf {index32}: merge {current_state:?} + {new_state:?}");
*current_state = merged_state;
Ok(())
}
#[inline(always)]
unsafe fn barrier(
current_states: &mut [BufferUses],
index32: u32,
index: usize,
state_provider: BufferStateProvider<'_>,
barriers: &mut Vec<PendingTransition<BufferUses>>,
) {
let current_state = *current_states.get_unchecked(index);
let new_state = state_provider.get_state(index);
if skip_barrier(current_state, new_state) {
return;
}
barriers.push(PendingTransition {
id: index32,
selector: (),
usage: current_state..new_state,
});
log::trace!("\tbuf {index32}: transition {current_state:?} -> {new_state:?}");
}
#[inline(always)]
unsafe fn update(
current_states: &mut [BufferUses],
index: usize,
state_provider: BufferStateProvider<'_>,
) {
let current_state = current_states.get_unchecked_mut(index);
let new_state = state_provider.get_state(index);
*current_state = new_state;
}

1237
wgpu-core/src/track/mod.rs
View File

File diff suppressed because it is too large Load Diff

268
wgpu-core/src/track/range.rs
View File

@ -2,25 +2,19 @@
//TODO: consider getting rid of it.
use smallvec::SmallVec;
use std::{cmp::Ordering, fmt::Debug, iter, ops::Range, slice::Iter};
use std::{fmt::Debug, iter, ops::Range};
/// Structure that keeps track of a I -> T mapping,
/// optimized for a case where keys of the same values
/// are often grouped together linearly.
#[derive(Clone, Debug, PartialEq)]
pub struct RangedStates<I, T> {
pub(crate) struct RangedStates<I, T> {
/// List of ranges, each associated with a singe value.
/// Ranges of keys have to be non-intersecting and ordered.
ranges: SmallVec<[(Range<I>, T); 1]>,
}
impl<I: Copy + PartialOrd, T: Copy + PartialEq> RangedStates<I, T> {
pub fn empty() -> Self {
Self {
ranges: SmallVec::new(),
}
}
impl<I: Copy + Ord, T: Copy + PartialEq> RangedStates<I, T> {
pub fn from_range(range: Range<I>, value: T) -> Self {
Self {
ranges: iter::once((range, value)).collect(),
@ -35,20 +29,12 @@ impl<I: Copy + PartialOrd, T: Copy + PartialEq> RangedStates<I, T> {
}
}
/// Clear all the ranges.
pub fn clear(&mut self) {
self.ranges.clear();
pub fn iter(&self) -> impl Iterator<Item = &(Range<I>, T)> + Clone {
self.ranges.iter()
}
/// Append a range.
///
/// Assumes that the object is being constructed from a set of
/// ranges, and they are given in the ascending order of their keys.
pub fn append(&mut self, index: Range<I>, value: T) {
if let Some(last) = self.ranges.last() {
debug_assert!(last.0.end <= index.start);
}
self.ranges.push((index, value));
pub fn iter_mut(&mut self) -> impl Iterator<Item = &mut (Range<I>, T)> {
self.ranges.iter_mut()
}
/// Check that all the ranges are non-intersecting and ordered.
@ -64,7 +50,6 @@ impl<I: Copy + PartialOrd, T: Copy + PartialEq> RangedStates<I, T> {
}
/// Merge the neighboring ranges together, where possible.
#[allow(clippy::suspicious_operation_groupings)]
pub fn coalesce(&mut self) {
let mut num_removed = 0;
let mut iter = self.ranges.iter_mut();
@ -86,25 +71,18 @@ impl<I: Copy + PartialOrd, T: Copy + PartialEq> RangedStates<I, T> {
}
}
/// Check if all intersecting ranges have the same value, which is returned.
///
/// Returns `None` if no intersections are detected.
/// Returns `Some(Err)` if the intersected values are inconsistent.
pub fn query<U: PartialEq>(
&self,
index: &Range<I>,
fun: impl Fn(&T) -> U,
) -> Option<Result<U, ()>> {
let mut result = None;
for &(ref range, ref value) in self.ranges.iter() {
if range.end > index.start && range.start < index.end {
let old = result.replace(fun(value));
if old.is_some() && old != result {
return Some(Err(()));
}
}
}
result.map(Ok)
pub fn iter_filter<'a>(
&'a self,
range: &'a Range<I>,
) -> impl Iterator<Item = (Range<I>, &T)> + 'a {
self.ranges
.iter()
.filter(move |&&(ref inner, ..)| inner.end > range.start && inner.start < range.end)
.map(move |&(ref inner, ref v)| {
let new_range = inner.start.max(range.start)..inner.end.min(range.end);
(new_range, v)
})
}
/// Split the storage ranges in such a way that there is a linear subset of
@ -176,112 +154,12 @@ impl<I: Copy + PartialOrd, T: Copy + PartialEq> RangedStates<I, T> {
clone.check_sanity();
result
}
/// Produce an iterator that merges two instances together.
///
/// Each range in the returned iterator is a subset of a range in either
/// `self` or `other`, and the value returned as a `Range` from `self` to `other`.
pub fn merge<'a>(&'a self, other: &'a Self, base: I) -> Merge<'a, I, T> {
Merge {
base,
sa: self.ranges.iter().peekable(),
sb: other.ranges.iter().peekable(),
}
}
}
/// A custom iterator that goes through two `RangedStates` and process a merge.
#[derive(Debug)]
pub struct Merge<'a, I, T> {
base: I,
sa: iter::Peekable<Iter<'a, (Range<I>, T)>>,
sb: iter::Peekable<Iter<'a, (Range<I>, T)>>,
}
impl<'a, I: Copy + Debug + Ord, T: Copy + Debug> Iterator for Merge<'a, I, T> {
type Item = (Range<I>, Range<Option<T>>);
fn next(&mut self) -> Option<Self::Item> {
match (self.sa.peek(), self.sb.peek()) {
// we have both streams
(Some(&&(ref ra, va)), Some(&&(ref rb, vb))) => {
let (range, usage) = if ra.start < self.base {
// in the middle of the left stream
let (end, end_value) = if self.base == rb.start {
// right stream is starting
debug_assert!(self.base < ra.end);
(rb.end, Some(vb))
} else {
// right hasn't started yet
debug_assert!(self.base < rb.start);
(rb.start, None)
};
(self.base..ra.end.min(end), Some(va)..end_value)
} else if rb.start < self.base {
// in the middle of the right stream
let (end, start_value) = if self.base == ra.start {
// left stream is starting
debug_assert!(self.base < rb.end);
(ra.end, Some(va))
} else {
// left hasn't started yet
debug_assert!(self.base < ra.start);
(ra.start, None)
};
(self.base..rb.end.min(end), start_value..Some(vb))
} else {
// no active streams
match ra.start.cmp(&rb.start) {
// both are starting
Ordering::Equal => (ra.start..ra.end.min(rb.end), Some(va)..Some(vb)),
// only left is starting
Ordering::Less => (ra.start..rb.start.min(ra.end), Some(va)..None),
// only right is starting
Ordering::Greater => (rb.start..ra.start.min(rb.end), None..Some(vb)),
}
};
self.base = range.end;
if ra.end == range.end {
let _ = self.sa.next();
}
if rb.end == range.end {
let _ = self.sb.next();
}
Some((range, usage))
}
// only right stream
(None, Some(&&(ref rb, vb))) => {
let range = self.base.max(rb.start)..rb.end;
self.base = rb.end;
let _ = self.sb.next();
Some((range, None..Some(vb)))
}
// only left stream
(Some(&&(ref ra, va)), None) => {
let range = self.base.max(ra.start)..ra.end;
self.base = ra.end;
let _ = self.sa.next();
Some((range, Some(va)..None))
}
// done
(None, None) => None,
}
}
}
#[cfg(test)]
mod test {
//TODO: randomized/fuzzy testing
use super::RangedStates;
use std::{fmt::Debug, ops::Range};
fn easy_merge<T: PartialEq + Copy + Debug>(
ra: &[(Range<usize>, T)],
rb: &[(Range<usize>, T)],
) -> Vec<(Range<usize>, Range<Option<T>>)> {
RangedStates::from_slice(ra)
.merge(&RangedStates::from_slice(rb), 0)
.collect()
}
#[test]
fn sane_good() {
@ -311,14 +189,6 @@ mod test {
assert_eq!(rs.ranges.as_slice(), &[(1..5, 9), (5..7, 1), (8..9, 1),]);
}
#[test]
fn query() {
let rs = RangedStates::from_slice(&[(1..4, 1u8), (5..7, 2)]);
assert_eq!(rs.query(&(0..1), |v| *v), None);
assert_eq!(rs.query(&(1..3), |v| *v), Some(Ok(1)));
assert_eq!(rs.query(&(1..6), |v| *v), Some(Err(())));
}
#[test]
fn isolate() {
let rs = RangedStates::from_slice(&[(1..4, 9u8), (4..5, 9), (5..7, 1), (8..9, 1)]);
@ -333,104 +203,4 @@ mod test {
&[(6..7, 1), (7..8, 0), (8..9, 1),]
);
}
#[test]
fn merge_same() {
assert_eq!(
&easy_merge(&[(1..4, 0u8),], &[(1..4, 2u8),],),
&[(1..4, Some(0)..Some(2)),]
);
}
#[test]
fn merge_empty() {
assert_eq!(
&easy_merge(&[(1..2, 0u8),], &[],),
&[(1..2, Some(0)..None),]
);
assert_eq!(
&easy_merge(&[], &[(3..4, 1u8),],),
&[(3..4, None..Some(1)),]
);
}
#[test]
fn merge_separate() {
assert_eq!(
&easy_merge(&[(1..2, 0u8), (5..6, 1u8),], &[(2..4, 2u8),],),
&[
(1..2, Some(0)..None),
(2..4, None..Some(2)),
(5..6, Some(1)..None),
]
);
}
#[test]
fn merge_subset() {
assert_eq!(
&easy_merge(&[(1..6, 0u8),], &[(2..4, 2u8),],),
&[
(1..2, Some(0)..None),
(2..4, Some(0)..Some(2)),
(4..6, Some(0)..None),
]
);
assert_eq!(
&easy_merge(&[(2..4, 0u8),], &[(1..4, 2u8),],),
&[(1..2, None..Some(2)), (2..4, Some(0)..Some(2)),]
);
}
#[test]
fn merge_all() {
assert_eq!(
&easy_merge(&[(1..4, 0u8), (5..8, 1u8),], &[(2..6, 2u8), (7..9, 3u8),],),
&[
(1..2, Some(0)..None),
(2..4, Some(0)..Some(2)),
(4..5, None..Some(2)),
(5..6, Some(1)..Some(2)),
(6..7, Some(1)..None),
(7..8, Some(1)..Some(3)),
(8..9, None..Some(3)),
]
);
}
#[test]
fn merge_complex() {
assert_eq!(
&easy_merge(
&[
(0..8, 0u8),
(8..9, 1),
(9..16, 2),
(16..17, 3),
(17..118, 4),
(118..119, 5),
(119..124, 6),
(124..125, 7),
(125..512, 8),
],
&[(15..16, 10u8), (51..52, 11), (126..127, 12),],
),
&[
(0..8, Some(0)..None),
(8..9, Some(1)..None),
(9..15, Some(2)..None),
(15..16, Some(2)..Some(10)),
(16..17, Some(3)..None),
(17..51, Some(4)..None),
(51..52, Some(4)..Some(11)),
(52..118, Some(4)..None),
(118..119, Some(5)..None),
(119..124, Some(6)..None),
(124..125, Some(7)..None),
(125..126, Some(8)..None),
(126..127, Some(8)..Some(12)),
(127..512, Some(8)..None),
]
);
}
}

209
wgpu-core/src/track/stateless.rs Normal file
View File

@ -0,0 +1,209 @@
/*! Stateless Trackers
*
* Stateless trackers don't have any state, so make no
* distinction between a usage scope and a full tracker.
!*/
use std::marker::PhantomData;
use crate::{
hub,
id::{TypedId, Valid},
track::{iterate_bitvec_indices, ResourceMetadata},
RefCount,
};
/// Stores all the resources that a bind group stores.
pub(crate) struct StatelessBindGroupSate<T, Id: TypedId> {
resources: Vec<(Valid<Id>, RefCount)>,
_phantom: PhantomData<T>,
}
impl<T: hub::Resource, Id: TypedId> StatelessBindGroupSate<T, Id> {
pub fn new() -> Self {
Self {
resources: Vec::new(),
_phantom: PhantomData,
}
}
/// Optimize the buffer bind group state by sorting it by ID.
///
/// When this list of states is merged into a tracker, the memory
/// accesses will be in a constant assending order.
pub(crate) fn optimize(&mut self) {
self.resources
.sort_unstable_by_key(|&(id, _)| id.0.unzip().0);
}
/// Returns a list of all resources tracked. May contain duplicates.
pub fn used(&self) -> impl Iterator<Item = Valid<Id>> + '_ {
self.resources.iter().map(|&(id, _)| id)
}
/// Adds the given resource.
pub fn add_single<'a>(&mut self, storage: &'a hub::Storage<T, Id>, id: Id) -> Option<&'a T> {
let resource = storage.get(id).ok()?;
self.resources
.push((Valid(id), resource.life_guard().add_ref()));
Some(resource)
}
}
/// Stores all resource state within a command buffer or device.
pub(crate) struct StatelessTracker<A: hub::HalApi, T, Id: TypedId> {
metadata: ResourceMetadata<A>,
_phantom: PhantomData<(T, Id)>,
}
impl<A: hub::HalApi, T: hub::Resource, Id: TypedId> StatelessTracker<A, T, Id> {
pub fn new() -> Self {
Self {
metadata: ResourceMetadata::new(),
_phantom: PhantomData,
}
}
fn debug_assert_in_bounds(&self, index: usize) {
self.metadata.debug_assert_in_bounds(index);
}
/// Sets the size of all the vectors inside the tracker.
///
/// Must be called with the highest possible Resource ID of this type
/// before all unsafe functions are called.
pub fn set_size(&mut self, size: usize) {
self.metadata.set_size(size);
}
/// Extend the vectors to let the given index be valid.
fn allow_index(&mut self, index: usize) {
if index >= self.metadata.owned.len() {
self.set_size(index + 1);
}
}
/// Returns a list of all resources tracked.
pub fn used(&self) -> impl Iterator<Item = Valid<Id>> + '_ {
self.metadata.used()
}
/// Inserts a single resource into the resource tracker.
///
/// If the resource already exists in the tracker, it will be overwritten.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn insert_single(&mut self, id: Valid<Id>, ref_count: RefCount) {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
self.allow_index(index);
self.debug_assert_in_bounds(index);
unsafe {
*self.metadata.epochs.get_unchecked_mut(index) = epoch;
*self.metadata.ref_counts.get_unchecked_mut(index) = Some(ref_count);
self.metadata.owned.set(index, true);
}
}
/// Adds the given resource to the tracker.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn add_single<'a>(&mut self, storage: &'a hub::Storage<T, Id>, id: Id) -> Option<&'a T> {
let item = storage.get(id).ok()?;
let (index32, epoch, _) = id.unzip();
let index = index32 as usize;
self.allow_index(index);
self.debug_assert_in_bounds(index);
unsafe {
*self.metadata.epochs.get_unchecked_mut(index) = epoch;
*self.metadata.ref_counts.get_unchecked_mut(index) = Some(item.life_guard().add_ref());
self.metadata.owned.set(index, true);
}
Some(item)
}
/// Adds the given resources from the given tracker.
///
/// If the ID is higher than the length of internal vectors,
/// the vectors will be extended. A call to set_size is not needed.
pub fn add_from_tracker(&mut self, other: &Self) {
let incoming_size = other.metadata.owned.len();
if incoming_size > self.metadata.owned.len() {
self.set_size(incoming_size);
}
for index in iterate_bitvec_indices(&other.metadata.owned) {
self.debug_assert_in_bounds(index);
other.debug_assert_in_bounds(index);
unsafe {
let previously_owned = self.metadata.owned.get(index).unwrap_unchecked();
if !previously_owned {
self.metadata.owned.set(index, true);
let other_ref_count = other
.metadata
.ref_counts
.get_unchecked(index)
.clone()
.unwrap_unchecked();
*self.metadata.ref_counts.get_unchecked_mut(index) = Some(other_ref_count);
let epoch = *other.metadata.epochs.get_unchecked(index);
*self.metadata.epochs.get_unchecked_mut(index) = epoch;
}
}
}
}
/// Removes the given resource from the tracker iff we have the last reference to the
/// resource and the epoch matches.
///
/// Returns true if the resource was removed.
///
/// If the ID is higher than the length of internal vectors,
/// false will be returned.
pub fn remove_abandoned(&mut self, id: Valid<Id>) -> bool {
let (index32, epoch, _) = id.0.unzip();
let index = index32 as usize;
if index > self.metadata.owned.len() {
return false;
}
self.debug_assert_in_bounds(index);
unsafe {
if self.metadata.owned.get(index).unwrap_unchecked() {
let existing_epoch = self.metadata.epochs.get_unchecked_mut(index);
let existing_ref_count = self.metadata.ref_counts.get_unchecked_mut(index);
if *existing_epoch == epoch
&& existing_ref_count.as_mut().unwrap_unchecked().load() == 1
{
self.metadata.reset(index);
return true;
}
}
}
false
}
}

1826
wgpu-core/src/track/texture.rs
View File

File diff suppressed because it is too large Load Diff

2
wgpu-hal/examples/halmark/main.rs
View File

@ -685,7 +685,7 @@ impl<A: hal::Api> Example<A> {
let target_barrier1 = hal::TextureBarrier {
texture: surface_tex.borrow(),
range: wgt::ImageSubresourceRange::default(),
usage: hal::TextureUses::COLOR_TARGET..hal::TextureUses::empty(),
usage: hal::TextureUses::COLOR_TARGET..hal::TextureUses::PRESENT,
};
unsafe {
ctx.encoder.end_render_pass();

4
wgpu-hal/src/dx12/command.rs
View File

@ -287,7 +287,7 @@ impl crate::CommandEncoder<super::Api> for super::CommandEncoder {
StateAfter: s1,
};
self.temp.barriers.push(raw);
} else if barrier.usage.start == crate::BufferUses::STORAGE_WRITE {
} else if barrier.usage.start == crate::BufferUses::STORAGE_READ_WRITE {
let mut raw = d3d12::D3D12_RESOURCE_BARRIER {
Type: d3d12::D3D12_RESOURCE_BARRIER_TYPE_UAV,
Flags: d3d12::D3D12_RESOURCE_BARRIER_FLAG_NONE,
@ -382,7 +382,7 @@ impl crate::CommandEncoder<super::Api> for super::CommandEncoder {
}
}
}
} else if barrier.usage.start == crate::TextureUses::STORAGE_WRITE {
} else if barrier.usage.start == crate::TextureUses::STORAGE_READ_WRITE {
let mut raw = d3d12::D3D12_RESOURCE_BARRIER {
Type: d3d12::D3D12_RESOURCE_BARRIER_TYPE_UAV,
Flags: d3d12::D3D12_RESOURCE_BARRIER_FLAG_NONE,

8
wgpu-hal/src/dx12/conv.rs
View File

@ -3,7 +3,7 @@ use winapi::um::{d3d12, d3dcommon};
pub fn map_buffer_usage_to_resource_flags(usage: crate::BufferUses) -> d3d12::D3D12_RESOURCE_FLAGS {
let mut flags = 0;
if usage.contains(crate::BufferUses::STORAGE_WRITE) {
if usage.contains(crate::BufferUses::STORAGE_READ_WRITE) {
flags |= d3d12::D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
}
flags
@ -33,7 +33,7 @@ pub fn map_texture_usage_to_resource_flags(
flags |= d3d12::D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE;
}
}
if usage.contains(crate::TextureUses::STORAGE_WRITE) {
if usage.contains(crate::TextureUses::STORAGE_READ_WRITE) {
flags |= d3d12::D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
}
@ -130,7 +130,7 @@ pub fn map_buffer_usage_to_state(usage: crate::BufferUses) -> d3d12::D3D12_RESOU
if usage.intersects(Bu::VERTEX | Bu::UNIFORM) {
state |= d3d12::D3D12_RESOURCE_STATE_VERTEX_AND_CONSTANT_BUFFER;
}
if usage.intersects(Bu::STORAGE_WRITE) {
if usage.intersects(Bu::STORAGE_READ_WRITE) {
state |= d3d12::D3D12_RESOURCE_STATE_UNORDERED_ACCESS;
} else if usage.intersects(Bu::STORAGE_READ) {
state |= d3d12::D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE
@ -170,7 +170,7 @@ pub fn map_texture_usage_to_state(usage: crate::TextureUses) -> d3d12::D3D12_RES
if usage.intersects(Tu::DEPTH_STENCIL_WRITE) {
state |= d3d12::D3D12_RESOURCE_STATE_DEPTH_WRITE;
}
if usage.intersects(Tu::STORAGE_READ | Tu::STORAGE_WRITE) {
if usage.intersects(Tu::STORAGE_READ | Tu::STORAGE_READ_WRITE) {
state |= d3d12::D3D12_RESOURCE_STATE_UNORDERED_ACCESS;
}
state

9
wgpu-hal/src/dx12/device.rs
View File

@ -428,7 +428,7 @@ impl crate::Device<super::Api> for super::Device {
|| !desc.usage.intersects(
crate::TextureUses::RESOURCE
| crate::TextureUses::STORAGE_READ
| crate::TextureUses::STORAGE_WRITE,
| crate::TextureUses::STORAGE_READ_WRITE,
) {
auxil::dxgi::conv::map_texture_format(desc.format)
} else {
@ -516,10 +516,9 @@ impl crate::Device<super::Api> for super::Device {
} else {
None
},
handle_uav: if desc
.usage
.intersects(crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_WRITE)
{
handle_uav: if desc.usage.intersects(
crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_READ_WRITE,
) {
let raw_desc = view_desc.to_uav();
let handle = self.srv_uav_pool.lock().alloc_handle();
self.raw.CreateUnorderedAccessView(

12
wgpu-hal/src/gles/command.rs
View File

@ -230,7 +230,11 @@ impl crate::CommandEncoder<super::Api> for super::CommandEncoder {
}
for bar in barriers {
// GLES only synchronizes storage -> anything explicitly
if !bar.usage.start.contains(crate::BufferUses::STORAGE_WRITE) {
if !bar
.usage
.start
.contains(crate::BufferUses::STORAGE_READ_WRITE)
{
continue;
}
self.cmd_buffer
@ -253,7 +257,11 @@ impl crate::CommandEncoder<super::Api> for super::CommandEncoder {
let mut combined_usage = crate::TextureUses::empty();
for bar in barriers {
// GLES only synchronizes storage -> anything explicitly
if !bar.usage.start.contains(crate::TextureUses::STORAGE_WRITE) {
if !bar
.usage
.start
.contains(crate::TextureUses::STORAGE_READ_WRITE)
{
continue;
}
// unlike buffers, there is no need for a concrete texture

8
wgpu-hal/src/gles/queue.rs
View File

@ -790,9 +790,9 @@ impl super::Queue {
if usage.intersects(crate::BufferUses::MAP_READ | crate::BufferUses::MAP_WRITE) {
flags |= glow::BUFFER_UPDATE_BARRIER_BIT;
}
if usage
.intersects(crate::BufferUses::STORAGE_READ | crate::BufferUses::STORAGE_WRITE)
{
if usage.intersects(
crate::BufferUses::STORAGE_READ | crate::BufferUses::STORAGE_READ_WRITE,
) {
flags |= glow::SHADER_STORAGE_BARRIER_BIT;
}
gl.memory_barrier(flags);
@ -803,7 +803,7 @@ impl super::Queue {
flags |= glow::TEXTURE_FETCH_BARRIER_BIT;
}
if usage.intersects(
crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_WRITE,
crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_READ_WRITE,
) {
flags |= glow::SHADER_IMAGE_ACCESS_BARRIER_BIT;
}

74
wgpu-hal/src/lib.rs
View File

@ -628,53 +628,77 @@ bitflags!(
bitflags::bitflags! {
/// Similar to `wgt::BufferUsages` but for internal use.
pub struct BufferUses: u32 {
pub struct BufferUses: u16 {
/// The argument to a read-only mapping.
const MAP_READ = 1 << 0;
/// The argument to a write-only mapping.
const MAP_WRITE = 1 << 1;
/// The source of a hardware copy.
const COPY_SRC = 1 << 2;
/// The destination of a hardware copy.
const COPY_DST = 1 << 3;
/// The index buffer used for drawing.
const INDEX = 1 << 4;
/// A vertex buffer used for drawing.
const VERTEX = 1 << 5;
/// A uniform buffer bound in a bind group.
const UNIFORM = 1 << 6;
/// A read-only storage buffer used in a bind group.
const STORAGE_READ = 1 << 7;
const STORAGE_WRITE = 1 << 8;
/// A read-write or write-only buffer used in a bind group.
const STORAGE_READ_WRITE = 1 << 8;
/// The indirect or count buffer in a indirect draw or dispatch.
const INDIRECT = 1 << 9;
/// The combination of usages that can be used together (read-only).
/// The combination of states that a buffer may be in _at the same time_.
const INCLUSIVE = Self::MAP_READ.bits | Self::COPY_SRC.bits |
Self::INDEX.bits | Self::VERTEX.bits | Self::UNIFORM.bits |
Self::STORAGE_READ.bits | Self::INDIRECT.bits;
/// The combination of exclusive usages (write-only and read-write).
/// These usages may still show up with others, but can't automatically be combined.
const EXCLUSIVE = Self::MAP_WRITE.bits | Self::COPY_DST.bits | Self::STORAGE_WRITE.bits;
/// The combination of states that a buffer must exclusively be in.
const EXCLUSIVE = Self::MAP_WRITE.bits | Self::COPY_DST.bits | Self::STORAGE_READ_WRITE.bits;
/// The combination of all usages that the are guaranteed to be be ordered by the hardware.
/// If a usage is not ordered, then even if it doesn't change between draw calls, there
/// still need to be pipeline barriers inserted for synchronization.
/// If a usage is ordered, then if the buffer state doesn't change between draw calls, there
/// are no barriers needed for synchronization.
const ORDERED = Self::INCLUSIVE.bits | Self::MAP_WRITE.bits;
}
}
bitflags::bitflags! {
/// Similar to `wgt::TextureUsages` but for internal use.
pub struct TextureUses: u32 {
const COPY_SRC = 1 << 0;
const COPY_DST = 1 << 1;
const RESOURCE = 1 << 2;
const COLOR_TARGET = 1 << 3;
const DEPTH_STENCIL_READ = 1 << 4;
const DEPTH_STENCIL_WRITE = 1 << 5;
const STORAGE_READ = 1 << 6;
const STORAGE_WRITE = 1 << 7;
/// The combination of usages that can be used together (read-only).
pub struct TextureUses: u16 {
/// The texture is in unknown state.
const UNINITIALIZED = 1 << 0;
/// Ready to present image to the surface.
const PRESENT = 1 << 1;
/// The source of a hardware copy.
const COPY_SRC = 1 << 2;
/// The destination of a hardware copy.
const COPY_DST = 1 << 3;
/// Read-only sampled or fetched resource.
const RESOURCE = 1 << 4;
/// The color target of a renderpass.
const COLOR_TARGET = 1 << 5;
/// Read-only depth stencil usage.
const DEPTH_STENCIL_READ = 1 << 6;
/// Read-write depth stencil usage
const DEPTH_STENCIL_WRITE = 1 << 7;
/// Read-only storage buffer usage. Corresponds to a UAV in d3d, so is exclusive, despite being read only.
const STORAGE_READ = 1 << 8;
/// Read-write or write-only storage buffer usage.
const STORAGE_READ_WRITE = 1 << 9;
/// The combination of states that a texture may be in _at the same time_.
const INCLUSIVE = Self::COPY_SRC.bits | Self::RESOURCE.bits | Self::DEPTH_STENCIL_READ.bits;
/// The combination of exclusive usages (write-only and read-write).
/// These usages may still show up with others, but can't automatically be combined.
const EXCLUSIVE = Self::COPY_DST.bits | Self::COLOR_TARGET.bits | Self::DEPTH_STENCIL_WRITE.bits | Self::STORAGE_READ.bits | Self::STORAGE_WRITE.bits;
/// The combination of states that a texture must exclusively be in.
const EXCLUSIVE = Self::COPY_DST.bits | Self::COLOR_TARGET.bits | Self::DEPTH_STENCIL_WRITE.bits | Self::STORAGE_READ.bits | Self::STORAGE_READ_WRITE.bits | Self::PRESENT.bits;
/// The combination of all usages that the are guaranteed to be be ordered by the hardware.
/// If a usage is not ordered, then even if it doesn't change between draw calls, there
/// still need to be pipeline barriers inserted for synchronization.
/// If a usage is ordered, then if the texture state doesn't change between draw calls, there
/// are no barriers needed for synchronization.
const ORDERED = Self::INCLUSIVE.bits | Self::COLOR_TARGET.bits | Self::DEPTH_STENCIL_WRITE.bits | Self::STORAGE_READ.bits;
//TODO: remove this
const UNINITIALIZED = 0xFFFF;
/// Flag used by the wgpu-core texture tracker to say a texture is in different states for every sub-resource
const COMPLEX = 1 << 10;
/// Flag used by the wgpu-core texture tracker to say that the tracker does not know the state of the sub-resource.
/// This is different from UNINITIALIZED as that says the tracker does know, but the texture has not been initialized.
const UNKNOWN = 1 << 11;
}
}

6
wgpu-hal/src/metal/conv.rs
View File

@ -9,11 +9,13 @@ pub fn map_texture_usage(usage: crate::TextureUses) -> mtl::MTLTextureUsage {
);
mtl_usage.set(
mtl::MTLTextureUsage::ShaderRead,
usage.intersects(Tu::RESOURCE | Tu::DEPTH_STENCIL_READ | Tu::STORAGE_READ),
usage.intersects(
Tu::RESOURCE | Tu::DEPTH_STENCIL_READ | Tu::STORAGE_READ | Tu::STORAGE_READ_WRITE,
),
);
mtl_usage.set(
mtl::MTLTextureUsage::ShaderWrite,
usage.intersects(Tu::STORAGE_WRITE),
usage.intersects(Tu::STORAGE_READ_WRITE),
);
mtl_usage

18
wgpu-hal/src/vulkan/conv.rs
View File

@ -200,7 +200,7 @@ pub fn derive_image_layout(
vk::ImageLayout::DEPTH_STENCIL_ATTACHMENT_OPTIMAL
}
_ => {
if usage.is_empty() {
if usage == crate::TextureUses::PRESENT {
vk::ImageLayout::PRESENT_SRC_KHR
} else if is_color {
vk::ImageLayout::GENERAL
@ -230,7 +230,7 @@ pub fn map_texture_usage(usage: crate::TextureUses) -> vk::ImageUsageFlags {
) {
flags |= vk::ImageUsageFlags::DEPTH_STENCIL_ATTACHMENT;
}
if usage.intersects(crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_WRITE) {
if usage.intersects(crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_READ_WRITE) {
flags |= vk::ImageUsageFlags::STORAGE;
}
flags
@ -276,12 +276,12 @@ pub fn map_texture_usage_to_barrier(
stages |= shader_stages;
access |= vk::AccessFlags::SHADER_READ;
}
if usage.contains(crate::TextureUses::STORAGE_WRITE) {
if usage.contains(crate::TextureUses::STORAGE_READ_WRITE) {
stages |= shader_stages;
access |= vk::AccessFlags::SHADER_WRITE;
access |= vk::AccessFlags::SHADER_READ | vk::AccessFlags::SHADER_WRITE;
}
if usage == crate::TextureUses::UNINITIALIZED || usage.is_empty() {
if usage == crate::TextureUses::UNINITIALIZED || usage == crate::TextureUses::PRESENT {
(
vk::PipelineStageFlags::TOP_OF_PIPE,
vk::AccessFlags::empty(),
@ -309,7 +309,7 @@ pub fn map_vk_image_usage(usage: vk::ImageUsageFlags) -> crate::TextureUses {
bits |= crate::TextureUses::DEPTH_STENCIL_READ | crate::TextureUses::DEPTH_STENCIL_WRITE;
}
if usage.contains(vk::ImageUsageFlags::STORAGE) {
bits |= crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_WRITE;
bits |= crate::TextureUses::STORAGE_READ | crate::TextureUses::STORAGE_READ_WRITE;
}
bits
}
@ -457,7 +457,7 @@ pub fn map_buffer_usage(usage: crate::BufferUses) -> vk::BufferUsageFlags {
if usage.contains(crate::BufferUses::UNIFORM) {
flags |= vk::BufferUsageFlags::UNIFORM_BUFFER;
}
if usage.intersects(crate::BufferUses::STORAGE_READ | crate::BufferUses::STORAGE_WRITE) {
if usage.intersects(crate::BufferUses::STORAGE_READ | crate::BufferUses::STORAGE_READ_WRITE) {
flags |= vk::BufferUsageFlags::STORAGE_BUFFER;
}
if usage.contains(crate::BufferUses::INDEX) {
@ -505,9 +505,9 @@ pub fn map_buffer_usage_to_barrier(
stages |= shader_stages;
access |= vk::AccessFlags::SHADER_READ;
}
if usage.intersects(crate::BufferUses::STORAGE_WRITE) {
if usage.intersects(crate::BufferUses::STORAGE_READ_WRITE) {
stages |= shader_stages;
access |= vk::AccessFlags::SHADER_WRITE;
access |= vk::AccessFlags::SHADER_READ | vk::AccessFlags::SHADER_WRITE;
}
if usage.contains(crate::BufferUses::INDEX) {
stages |= vk::PipelineStageFlags::VERTEX_INPUT;

2
wgpu-hal/src/vulkan/device.rs
View File

@ -1413,7 +1413,7 @@ impl crate::Device<super::Api> for super::Device {
unsafe fn destroy_shader_module(&self, module: super::ShaderModule) {
match module {
super::ShaderModule::Raw(raw) => {
let _ = self.shared.raw.destroy_shader_module(raw, None);
self.shared.raw.destroy_shader_module(raw, None);
}
super::ShaderModule::Intermediate { .. } => {}
}