Struct TaskScope 

pub struct TaskScope(/* private fields */);

Implementations

impl TaskScope

pub fn alloc<T>(&self, capacity: usize) -> Buffer<T, Self>

Allocates a buffer in this scope on the device.

This call is not blocking. Upon successful completion, it will return a buffer with a memory that is guaranteed to be available in the scope of the task but without any absolute guarantee relative to the host or any other task.

Other tasks may try to allocate memory concurrently. In order to guarantee enough memory for all expected work, the user must ensure some limit on task calls by e.g. using a semaphore.

pub fn try_alloc<T>( &self, capacity: usize, ) -> Result<Buffer<T, Self>, TryReserveError>

Tries to allocate a buffer in this scope on the device.

pub unsafe fn launch_host_fn( &self, host_fn: unsafe extern "C" fn(*mut c_void), data: *mut c_void, ) -> Result<(), CudaError>

Launches a host function in this task.

Safety

The function essentially executes an extern call in C. The safety assumption of an extern.
The user must ensure the pointer is valid and that the data remains valid as this call will be asynchronous.

pub unsafe fn launch_kernel( &self, kernel: KernelPtr, grid_dim: impl Into<Dim3>, block_dim: impl Into<Dim3>, args: &[*mut c_void], shared_mem: usize, ) -> Result<(), CudaError>

Launches a kernel in this task.

Safety

The caller must ensure that:

• The kernel ptr is valid.
• The arguments are passed correctly across the FFI interface.
• The data lives whitin the scope of the current task.

pub fn sleep(&self, time: Duration)

Sends the CUDA task to sleep for at least the given duration.

This function will not block the calling host thread. The function does a small allocation so the sleep time might be slightly longer than the given duration for very short times.

pub unsafe fn copy<T: DeviceCopy>( &self, dst: &mut Slice<T, Self>, src: &Slice<T, Self>, ) -> Result<(), CopyError>

Copies data between slices using CudaMemCpyAsync

Safety

The caller must ensure that the data is valid and that the data remains valid as this call

pub async fn synchronize(&self) -> Result<(), CudaError>

Waits for all work enqueued so far in this task to finish.

This function can be useful in case there is work to be enqueued but for some reason this work cannot be done using Self::launch_host_fn.

pub fn into_device<T: IntoDevice>( &self, data: T, ) -> Result<T::Output, CopyError>

Copies data from the host to the device.

pub fn to_device<T: ToDevice>(&self, data: &T) -> Result<T::Output, CopyError>

pub fn synchronize_blocking(&self) -> Result<(), CudaError>

Waits for all work enqueued so far in this task to finish.

This function can be useful in case there is work to be enqueued but for some reason this work cannot be done using Self::launch_host_fn.

pub unsafe fn handle(&self) -> CudaStreamHandle

Safety

pub fn owner(&self) -> TaskPool

pub fn spawn<F, Fut>(&self, f: F) -> SpawnHandle<Fut::Output>

where F: FnOnce(TaskScope) -> Fut + Send + 'static, Fut: Future + Send + 'static, Fut::Output: CudaSend + 'static,

Spawns a new task from the current task pool.

The task starting point will have a “happens before” relationship with the current task when the spawn is called. The handle can be used to wait for the child task to finish.

pub async fn run_in_place<F, Fut>(&self, f: F) -> Result<Fut::Output, CudaError>

where F: FnOnce(TaskScope) -> Fut, Fut: Future, Fut::Output: CudaSend,

Runs a task in place in a new stream.

Awaiting this task will peform the device calls and synchronize the end of this task to the parent, but does not do host synchronization.

Trait Implementations

impl Allocator for TaskScope

unsafe fn allocate(&self, layout: Layout) -> Result<NonNull<[u8]>, AllocError>

Attempts to allocate a block of memory.

unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout)

Deallocates the memory referenced by ptr.

fn by_ref(&self) -> &Self

where Self: Sized,

impl Backend for TaskScope

fn copy_from<B, T>( &self, data: T, ) -> Result<<T as CopyIntoBackend<Self, B>>::Output, CopyError>

where B: Backend, T: HasBackend + CopyIntoBackend<Self, B>,

impl Clone for TaskScope

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TaskScope

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

Formats the value using the given formatter.

impl Deref for TaskScope

type Target = Task

The resulting type after dereferencing.

fn deref(&self) -> &Self::Target

Dereferences the value.

impl DeviceMemory for TaskScope

unsafe fn copy_nonoverlapping( &self, src: *const u8, dst: *mut u8, size: usize, direction: CopyDirection, ) -> Result<(), CopyError>

Safety

unsafe fn write_bytes( &self, dst: *mut u8, value: u8, size: usize, ) -> Result<(), CopyError>

TODO

impl DeviceSumKernel<BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn partial_sum_kernel() -> KernelPtr

fn block_sum_kernel() -> KernelPtr

impl DeviceSumKernel<KoalaBear> for TaskScope

fn partial_sum_kernel() -> KernelPtr

fn block_sum_kernel() -> KernelPtr

impl DeviceTransposeKernel<[KoalaBear; 8]> for TaskScope

fn transpose_kernel() -> KernelPtr

impl DeviceTransposeKernel<[u32; 8]> for TaskScope

fn transpose_kernel() -> KernelPtr

impl DeviceTransposeKernel<BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn transpose_kernel() -> KernelPtr

impl DeviceTransposeKernel<KoalaBear> for TaskScope

fn transpose_kernel() -> KernelPtr

impl DeviceTransposeKernel<u32> for TaskScope

fn transpose_kernel() -> KernelPtr

impl DotKernel<BinomialExtensionField<KoalaBear, 4>, BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn partial_dot_kernel_last_dim() -> KernelPtr

fn dot_along_short_dimension_kernel() -> KernelPtr

impl DotKernel<KoalaBear, BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn partial_dot_kernel_last_dim() -> KernelPtr

fn dot_along_short_dimension_kernel() -> KernelPtr

impl DotKernel<KoalaBear, KoalaBear> for TaskScope

fn partial_dot_kernel_last_dim() -> KernelPtr

fn dot_along_short_dimension_kernel() -> KernelPtr

impl FoldKernel<BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn fold_kernel() -> KernelPtr

impl FoldKernel<KoalaBear> for TaskScope

fn fold_kernel() -> KernelPtr

impl IntoFuture for TaskScope

type Output = Result<(), CudaError>

The output that the future will produce on completion.

type IntoFuture = StreamCallbackFuture<TaskScope>

Which kind of future are we turning this into?

fn into_future(self) -> Self::IntoFuture

Creates a future from a value.

impl<F: Field> MleBaseBackend<F> for TaskScope

fn uninit_mle( &self, num_polynomials: usize, num_non_zero_entries: usize, ) -> Tensor<F, Self>

Creates an uninitialized MLE tensor.

fn num_polynomials(guts: &Tensor<F, Self>) -> usize

Returns the number of polynomials in the batch.

fn num_variables(guts: &Tensor<F, Self>) -> u32

Returns the number of variables in the polynomials.

fn num_non_zero_entries(guts: &Tensor<F, Self>) -> usize

Number of non-zero entries in the MLE.

impl MleFixLastVariableKernel<BinomialExtensionField<KoalaBear, 4>, BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn mle_fix_last_variable_kernel() -> KernelPtr

fn mle_fix_last_variable_constant_padding_kernel() -> KernelPtr

impl MleFixLastVariableKernel<KoalaBear, BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn mle_fix_last_variable_kernel() -> KernelPtr

fn mle_fix_last_variable_constant_padding_kernel() -> KernelPtr

impl PartialGeqKernel<KoalaBear> for TaskScope

fn partial_geq_kernel() -> KernelPtr

impl PartialLagrangeKernel<BinomialExtensionField<KoalaBear, 4>> for TaskScope

fn partial_lagrange_kernel() -> KernelPtr

impl PartialLagrangeKernel<KoalaBear> for TaskScope

fn partial_lagrange_kernel() -> KernelPtr

impl ScanKernel<KoalaBear> for TaskScope

fn single_block_scan_kernel_large_bb31_septic_curve() -> KernelPtr

fn scan_kernel_large_bb31_septic_curve() -> KernelPtr

impl TracegenPreprocessedRecursionBaseAluKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_base_alu_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionConvertKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_convert_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionExtAluKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_ext_alu_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionLinearLayerKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_linear_layer_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionPoseidon2WideKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_poseidon2_wide_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionSBoxKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_sbox_kernel() -> KernelPtr

impl TracegenPreprocessedRecursionSelectKernel<KoalaBear> for TaskScope

fn tracegen_preprocessed_recursion_select_kernel() -> KernelPtr

impl TracegenRecursionBaseAluKernel<KoalaBear> for TaskScope

fn tracegen_recursion_base_alu_kernel() -> KernelPtr

impl TracegenRecursionConvertKernel<KoalaBear> for TaskScope

fn tracegen_recursion_convert_kernel() -> KernelPtr

impl TracegenRecursionExtAluKernel<KoalaBear> for TaskScope

fn tracegen_recursion_ext_alu_kernel() -> KernelPtr

impl TracegenRecursionLinearLayerKernel<KoalaBear> for TaskScope

fn tracegen_recursion_linear_layer_kernel() -> KernelPtr

impl TracegenRecursionPoseidon2WideKernel<KoalaBear> for TaskScope

fn tracegen_recursion_poseidon2_wide_kernel() -> KernelPtr

impl TracegenRecursionPrefixSumChecksKernel<KoalaBear> for TaskScope

fn tracegen_recursion_prefix_sum_checks_kernel() -> KernelPtr

impl TracegenRecursionSBoxKernel<KoalaBear> for TaskScope

fn tracegen_recursion_sbox_kernel() -> KernelPtr

impl TracegenRecursionSelectKernel<KoalaBear> for TaskScope

fn tracegen_recursion_select_kernel() -> KernelPtr

impl TracegenRiscvGlobalKernel<KoalaBear> for TaskScope

fn tracegen_riscv_global_decompress_kernel() -> KernelPtr

fn tracegen_riscv_global_finalize_kernel() -> KernelPtr