Struct CommandEncoder

pub struct CommandEncoder { /* private fields */ }

Encodes a series of GPU operations.

A command encoder can record RenderPasses, ComputePasses, and transfer operations between driver-managed resources like Buffers and Textures.

When finished recording, call CommandEncoder::finish to obtain a CommandBuffer which may be submitted for execution.

Corresponds to WebGPU GPUCommandEncoder.

Implementations

impl CommandEncoder

pub fn finish(self) -> CommandBuffer

Finishes recording and returns a CommandBuffer that can be submitted for execution.

Examples found in repository

examples/app/headless_renderer.rs (line 391)

    fn run(
        &self,
        _graph: &mut RenderGraphContext,
        render_context: &mut RenderContext,
        world: &World,
    ) -> Result<(), NodeRunError> {
        let image_copiers = world.get_resource::<ImageCopiers>().unwrap();
        let gpu_images = world
            .get_resource::<RenderAssets<bevy::render::texture::GpuImage>>()
            .unwrap();

        for image_copier in image_copiers.iter() {
            if !image_copier.enabled() {
                continue;
            }

            let src_image = gpu_images.get(&image_copier.src_image).unwrap();

            let mut encoder = render_context
                .render_device()
                .create_command_encoder(&CommandEncoderDescriptor::default());

            let block_dimensions = src_image.texture_format.block_dimensions();
            let block_size = src_image.texture_format.block_copy_size(None).unwrap();

            // Calculating correct size of image row because
            // copy_texture_to_buffer can copy image only by rows aligned wgpu::COPY_BYTES_PER_ROW_ALIGNMENT
            // That's why image in buffer can be little bit wider
            // This should be taken into account at copy from buffer stage
            let padded_bytes_per_row = RenderDevice::align_copy_bytes_per_row(
                (src_image.size.width as usize / block_dimensions.0 as usize) * block_size as usize,
            );

            encoder.copy_texture_to_buffer(
                src_image.texture.as_image_copy(),
                TexelCopyBufferInfo {
                    buffer: &image_copier.buffer,
                    layout: TexelCopyBufferLayout {
                        offset: 0,
                        bytes_per_row: Some(
                            std::num::NonZero::<u32>::new(padded_bytes_per_row as u32)
                                .unwrap()
                                .into(),
                        ),
                        rows_per_image: None,
                    },
                },
                src_image.size,
            );

            let render_queue = world.get_resource::<RenderQueue>().unwrap();
            render_queue.submit(std::iter::once(encoder.finish()));
        }

        Ok(())
    }

pub fn begin_render_pass<'encoder>( &'encoder mut self, desc: &RenderPassDescriptor<'_>, ) -> RenderPass<'encoder>

Begins recording of a render pass.

This function returns a RenderPass object which records a single render pass.

As long as the returned RenderPass has not ended, any mutating operation on this command encoder causes an error and invalidates it. Note that the 'encoder lifetime relationship protects against this, but it is possible to opt out of it by calling RenderPass::forget_lifetime. This can be useful for runtime handling of the encoder->pass dependency e.g. when pass and encoder are stored in the same data structure.

pub fn begin_compute_pass<'encoder>( &'encoder mut self, desc: &ComputePassDescriptor<'_>, ) -> ComputePass<'encoder>

Begins recording of a compute pass.

This function returns a ComputePass object which records a single compute pass.

As long as the returned ComputePass has not ended, any mutating operation on this command encoder causes an error and invalidates it. Note that the 'encoder lifetime relationship protects against this, but it is possible to opt out of it by calling ComputePass::forget_lifetime. This can be useful for runtime handling of the encoder->pass dependency e.g. when pass and encoder are stored in the same data structure.

Examples found in repository

examples/shader/gpu_readback.rs (lines 211-214)

    fn run(
        &self,
        _graph: &mut render_graph::RenderGraphContext,
        render_context: &mut RenderContext,
        world: &World,
    ) -> Result<(), render_graph::NodeRunError> {
        let pipeline_cache = world.resource::<PipelineCache>();
        let pipeline = world.resource::<ComputePipeline>();
        let bind_group = world.resource::<GpuBufferBindGroup>();

        if let Some(init_pipeline) = pipeline_cache.get_compute_pipeline(pipeline.pipeline) {
            let mut pass =
                render_context
                    .command_encoder()
                    .begin_compute_pass(&ComputePassDescriptor {
                        label: Some("GPU readback compute pass"),
                        ..default()
                    });

            pass.set_bind_group(0, &bind_group.0, &[]);
            pass.set_pipeline(init_pipeline);
            pass.dispatch_workgroups(BUFFER_LEN as u32, 1, 1);
        }
        Ok(())
    }

examples/shader/compute_shader_game_of_life.rs (line 267)

    fn run(
        &self,
        _graph: &mut render_graph::RenderGraphContext,
        render_context: &mut RenderContext,
        world: &World,
    ) -> Result<(), render_graph::NodeRunError> {
        let bind_groups = &world.resource::<GameOfLifeImageBindGroups>().0;
        let pipeline_cache = world.resource::<PipelineCache>();
        let pipeline = world.resource::<GameOfLifePipeline>();

        let mut pass = render_context
            .command_encoder()
            .begin_compute_pass(&ComputePassDescriptor::default());

        // select the pipeline based on the current state
        match self.state {
            GameOfLifeState::Loading => {}
            GameOfLifeState::Init => {
                let init_pipeline = pipeline_cache
                    .get_compute_pipeline(pipeline.init_pipeline)
                    .unwrap();
                pass.set_bind_group(0, &bind_groups[0], &[]);
                pass.set_pipeline(init_pipeline);
                pass.dispatch_workgroups(SIZE.0 / WORKGROUP_SIZE, SIZE.1 / WORKGROUP_SIZE, 1);
            }
            GameOfLifeState::Update(index) => {
                let update_pipeline = pipeline_cache
                    .get_compute_pipeline(pipeline.update_pipeline)
                    .unwrap();
                pass.set_bind_group(0, &bind_groups[index], &[]);
                pass.set_pipeline(update_pipeline);
                pass.dispatch_workgroups(SIZE.0 / WORKGROUP_SIZE, SIZE.1 / WORKGROUP_SIZE, 1);
            }
        }

        Ok(())
    }

pub fn copy_buffer_to_buffer( &mut self, source: &Buffer, source_offset: u64, destination: &Buffer, destination_offset: u64, copy_size: u64, )

Copy data from one buffer to another.

Panics

• Buffer offsets or copy size not a multiple of COPY_BUFFER_ALIGNMENT.
• Copy would overrun buffer.
• Copy within the same buffer.

Examples found in repository

examples/3d/occlusion_culling.rs (lines 465-471)

    fn run<'w>(
        &self,
        _: &mut RenderGraphContext,
        render_context: &mut RenderContext<'w>,
        world: &'w World,
    ) -> Result<(), NodeRunError> {
        // Extract the buffers that hold the GPU indirect draw parameters from
        // the world resources. We're going to read those buffers to determine
        // how many meshes were actually drawn.
        let (Some(indirect_parameters_buffers), Some(indirect_parameters_mapping_buffers)) = (
            world.get_resource::<IndirectParametersBuffers>(),
            world.get_resource::<IndirectParametersStagingBuffers>(),
        ) else {
            return Ok(());
        };

        // Get the indirect parameters buffers corresponding to the opaque 3D
        // phase, since all our meshes are in that phase.
        let Some(phase_indirect_parameters_buffers) =
            indirect_parameters_buffers.get(&TypeId::of::<Opaque3d>())
        else {
            return Ok(());
        };

        // Grab both the buffers we're copying from and the staging buffers
        // we're copying to. Remember that we can't map the indirect parameters
        // buffers directly, so we have to copy their contents to a staging
        // buffer.
        let (
            Some(indexed_data_buffer),
            Some(indexed_batch_sets_buffer),
            Some(indirect_parameters_staging_data_buffer),
            Some(indirect_parameters_staging_batch_sets_buffer),
        ) = (
            phase_indirect_parameters_buffers.indexed.data_buffer(),
            phase_indirect_parameters_buffers
                .indexed
                .batch_sets_buffer(),
            indirect_parameters_mapping_buffers.data.as_ref(),
            indirect_parameters_mapping_buffers.batch_sets.as_ref(),
        )
        else {
            return Ok(());
        };

        // Copy from the indirect parameters buffers to the staging buffers.
        render_context.command_encoder().copy_buffer_to_buffer(
            indexed_data_buffer,
            0,
            indirect_parameters_staging_data_buffer,
            0,
            indexed_data_buffer.size(),
        );
        render_context.command_encoder().copy_buffer_to_buffer(
            indexed_batch_sets_buffer,
            0,
            indirect_parameters_staging_batch_sets_buffer,
            0,
            indexed_batch_sets_buffer.size(),
        );

        Ok(())
    }

pub fn copy_buffer_to_texture( &mut self, source: TexelCopyBufferInfo<&Buffer>, destination: TexelCopyTextureInfo<&Texture>, copy_size: Extent3d, )

Copy data from a buffer to a texture.

pub fn copy_texture_to_buffer( &mut self, source: TexelCopyTextureInfo<&Texture>, destination: TexelCopyBufferInfo<&Buffer>, copy_size: Extent3d, )

Copy data from a texture to a buffer.

Examples found in repository

examples/app/headless_renderer.rs (lines 373-388)

    fn run(
        &self,
        _graph: &mut RenderGraphContext,
        render_context: &mut RenderContext,
        world: &World,
    ) -> Result<(), NodeRunError> {
        let image_copiers = world.get_resource::<ImageCopiers>().unwrap();
        let gpu_images = world
            .get_resource::<RenderAssets<bevy::render::texture::GpuImage>>()
            .unwrap();

        for image_copier in image_copiers.iter() {
            if !image_copier.enabled() {
                continue;
            }

            let src_image = gpu_images.get(&image_copier.src_image).unwrap();

            let mut encoder = render_context
                .render_device()
                .create_command_encoder(&CommandEncoderDescriptor::default());

            let block_dimensions = src_image.texture_format.block_dimensions();
            let block_size = src_image.texture_format.block_copy_size(None).unwrap();

            // Calculating correct size of image row because
            // copy_texture_to_buffer can copy image only by rows aligned wgpu::COPY_BYTES_PER_ROW_ALIGNMENT
            // That's why image in buffer can be little bit wider
            // This should be taken into account at copy from buffer stage
            let padded_bytes_per_row = RenderDevice::align_copy_bytes_per_row(
                (src_image.size.width as usize / block_dimensions.0 as usize) * block_size as usize,
            );

            encoder.copy_texture_to_buffer(
                src_image.texture.as_image_copy(),
                TexelCopyBufferInfo {
                    buffer: &image_copier.buffer,
                    layout: TexelCopyBufferLayout {
                        offset: 0,
                        bytes_per_row: Some(
                            std::num::NonZero::<u32>::new(padded_bytes_per_row as u32)
                                .unwrap()
                                .into(),
                        ),
                        rows_per_image: None,
                    },
                },
                src_image.size,
            );

            let render_queue = world.get_resource::<RenderQueue>().unwrap();
            render_queue.submit(std::iter::once(encoder.finish()));
        }

        Ok(())
    }

pub fn copy_texture_to_texture( &mut self, source: TexelCopyTextureInfo<&Texture>, destination: TexelCopyTextureInfo<&Texture>, copy_size: Extent3d, )

Copy data from one texture to another.

Panics

• Textures are not the same type
• If a depth texture, or a multisampled texture, the entire texture must be copied
• Copy would overrun either texture

pub fn clear_texture( &mut self, texture: &Texture, subresource_range: &ImageSubresourceRange, )

Clears texture to zero.

Note that unlike with clear_buffer, COPY_DST usage is not required.

Implementation notes

• implemented either via buffer copies and render/depth target clear, path depends on texture usages
• behaves like texture zero init, but is performed immediately (clearing is not delayed via marking it as uninitialized)

Panics

• CLEAR_TEXTURE extension not enabled
• Range is out of bounds

pub fn clear_buffer(&mut self, buffer: &Buffer, offset: u64, size: Option<u64>)

Clears buffer to zero.

Panics

• Buffer does not have COPY_DST usage.
• Range is out of bounds

pub fn insert_debug_marker(&mut self, label: &str)

Inserts debug marker.

pub fn push_debug_group(&mut self, label: &str)

Start record commands and group it into debug marker group.

pub fn pop_debug_group(&mut self)

Stops command recording and creates debug group.

pub fn resolve_query_set( &mut self, query_set: &QuerySet, query_range: Range<u32>, destination: &Buffer, destination_offset: u64, )

Resolves a query set, writing the results into the supplied destination buffer.

Occlusion and timestamp queries are 8 bytes each (see crate::QUERY_SIZE). For pipeline statistics queries, see PipelineStatisticsTypes for more information.

pub unsafe fn as_hal_mut<A, F, R>( &mut self, hal_command_encoder_callback: F, ) -> R

where A: HalApi, F: FnOnce(Option<&mut <A as Api>::CommandEncoder>) -> R,

Returns the inner hal CommandEncoder using a callback. The hal command encoder will be None if the backend type argument does not match with this wgpu CommandEncoder

This method will start the wgpu_core level command recording.

Safety

• The raw handle obtained from the hal CommandEncoder must not be manually destroyed

impl CommandEncoder

Features::TIMESTAMP_QUERY_INSIDE_ENCODERS must be enabled on the device in order to call these functions.

pub fn write_timestamp(&mut self, query_set: &QuerySet, query_index: u32)

Issue a timestamp command at this point in the queue. The timestamp will be written to the specified query set, at the specified index.

Must be multiplied by Queue::get_timestamp_period to get the value in nanoseconds. Absolute values have no meaning, but timestamps can be subtracted to get the time it takes for a string of operations to complete.

Attention: Since commands within a command recorder may be reordered, there is no strict guarantee that timestamps are taken after all commands recorded so far and all before all commands recorded after. This may depend both on the backend and the driver.

impl CommandEncoder

Features::EXPERIMENTAL_RAY_TRACING_ACCELERATION_STRUCTURE must be enabled on the device in order to call these functions.

pub fn build_acceleration_structures<'a>( &mut self, blas: impl IntoIterator<Item = &'a BlasBuildEntry<'a>>, tlas: impl IntoIterator<Item = &'a TlasPackage>, )

Build bottom and top level acceleration structures.

Builds the BLASes then the TLASes, but does not build the BLASes into the TLASes, that must be done by setting a TLAS instance in the TLAS package to one that contains the BLAS (and with an appropriate transform)

Validation

• blas: Iterator of bottom level acceleration structure entries to build. For each entry, the provided size descriptor must be strictly smaller or equal to the descriptor given at BLAS creation, this means:
  • Less or equal number of geometries
  • Same kind of geometry (with index buffer or without) (same vertex/index format)
  • Same flags
  • Less or equal number of vertices
  • Less or equal number of indices (if applicable)
• tlas: iterator of top level acceleration structure packages to build For each entry:
  • Each BLAS in each TLAS instance must have been being built in the current call or in a previous call to build_acceleration_structures or build_acceleration_structures_unsafe_tlas
  • The number of TLAS instances must be less than or equal to the max number of tlas instances when creating (if creating a package with TlasPackage::new() this is already satisfied)

If the device the command encoder is created from does not have Features::EXPERIMENTAL_RAY_TRACING_ACCELERATION_STRUCTURE enabled then a validation error is generated

A bottom level acceleration structure may be build and used as a reference in a top level acceleration structure in the same invocation of this function.

Bind group usage

When a top level acceleration structure is used in a bind group, some validation takes place:

• The top level acceleration structure is valid and has been built.
• All the bottom level acceleration structures referenced by the top level acceleration structure are valid and have been built prior, or at same time as the containing top level acceleration structure.

pub unsafe fn build_acceleration_structures_unsafe_tlas<'a>( &mut self, blas: impl IntoIterator<Item = &'a BlasBuildEntry<'a>>, tlas: impl IntoIterator<Item = &'a TlasBuildEntry<'a>>, )

Build bottom and top level acceleration structures. See CommandEncoder::build_acceleration_structures for the safe version and more details. All validation in CommandEncoder::build_acceleration_structures except that listed under tlas applies here as well.

Safety

• The contents of the raw instance buffer must be valid for the underling api.
• All bottom level acceleration structures, referenced in the raw instance buffer must be valid and built, when the corresponding top level acceleration structure is built. (builds may happen in the same invocation of this function).
• At the time when the top level acceleration structure is used in a bind group, all associated bottom level acceleration structures must be valid, and built (no later than the time when the top level acceleration structure was built).

Trait Implementations

impl Debug for CommandEncoder

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Hash for CommandEncoder

fn hash<H>(&self, state: &mut H)

where H: Hasher,

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl Ord for CommandEncoder

fn cmp(&self, other: &CommandEncoder) -> Ordering

This method returns an Ordering between self and other.

fn max(self, other: Self) -> Self

where Self: Sized,

Compares and returns the maximum of two values.

fn min(self, other: Self) -> Self

where Self: Sized,

Compares and returns the minimum of two values.

fn clamp(self, min: Self, max: Self) -> Self

where Self: Sized,

Restrict a value to a certain interval.

impl PartialEq for CommandEncoder

fn eq(&self, other: &CommandEncoder) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl PartialOrd for CommandEncoder

fn partial_cmp(&self, other: &CommandEncoder) -> Option<Ordering>

This method returns an ordering between self and other values if one exists.

fn lt(&self, other: &Rhs) -> bool

Tests less than (for self and other) and is used by the < operator.

fn le(&self, other: &Rhs) -> bool

Tests less than or equal to (for self and other) and is used by the <= operator.

fn gt(&self, other: &Rhs) -> bool

Tests greater than (for self and other) and is used by the > operator.

fn ge(&self, other: &Rhs) -> bool

Tests greater than or equal to (for self and other) and is used by the >= operator.

impl Eq for CommandEncoder