Module prelude 

Common imports for users of atomr-accel-cuda.

Re-exports

pub use crate::completion::CompletionStrategy;

pub use crate::completion::HostFnCompletion;

pub use crate::device::ContextActor;

pub use crate::device::ContextMsg;

pub use crate::device::DeviceActor;

pub use crate::device::DeviceConfig;

pub use crate::device::DeviceLoad;

pub use crate::device::DeviceMsg;

pub use crate::device::DeviceState;

pub use crate::device::EnabledLibraries;

pub use crate::device::HostBuf;

pub use crate::device::KernelChildren;

pub use crate::device::SgemmRequest;

pub use crate::dispatcher::GpuDispatcher;

pub use crate::error::decider;

pub use crate::error::device_supervisor_strategy;

pub use crate::error::DeviceSupervisor;

pub use crate::error::GpuError;

pub use crate::gpu_ref::GpuRef;

pub use crate::graph::FftR2COp;

pub use crate::graph::GraphOpLegacy;Deprecated

pub use crate::graph::RngFillUniformOp;

pub use crate::graph::GraphActor;

pub use crate::graph::GraphHandle;

pub use crate::graph::GraphMsg;

pub use crate::graph::GraphOp;

pub use crate::graph::GraphRecordCtx;

pub use crate::graph::MemcpyOp;

pub use crate::graph::SgemmOp;

pub use crate::host::PinnedBuf;

pub use crate::host::PinnedBufferPool;

pub use crate::host::PinnedBufferPoolConfig;

pub use crate::host::PinnedPoolMsg;

pub use crate::host::PinnedPoolStats;

pub use crate::kernel::dispatch::BlasLtDispatch;

pub use crate::kernel::dispatch::BlasLtDispatchCtx;

pub use crate::kernel::dispatch::CollectiveDispatch;

pub use crate::kernel::dispatch::CollectiveDispatchCtx;

pub use crate::kernel::dispatch::CudnnDispatch;

pub use crate::kernel::dispatch::CudnnDispatchCtx;

pub use crate::kernel::dispatch::DevSliceArg;

pub use crate::kernel::dispatch::GemmDispatch;

pub use crate::kernel::dispatch::GemmDispatchCtx;

pub use crate::kernel::dispatch::NvrtcDispatchCtx;

pub use crate::kernel::dispatch::NvrtcLaunchDispatch;

pub use crate::kernel::dispatch::RngDispatch;

pub use crate::kernel::dispatch::ScalarArg;

pub use crate::kernel::dispatch::FftDispatch;

pub use crate::kernel::dispatch::FftDispatchCtx;

pub use crate::kernel::dispatch::SendSparseHandle;

pub use crate::kernel::dispatch::SparseDispatch;

pub use crate::kernel::dispatch::SparseDispatchCtx;

pub use crate::kernel::dispatch::SparseOp;

pub use crate::kernel::dispatch::TensorDispatch;

pub use crate::kernel::dispatch::TensorDispatchCtx;

pub use crate::kernel::envelope;

pub use crate::kernel::record::RecordMode;

pub use crate::kernel::BlasActor;

pub use crate::kernel::BlasMsg;

pub use crate::memory::ManagedAllocatorActor;

pub use crate::memory::ManagedFlags;

pub use crate::memory::ManagedMsg;

pub use crate::memory::ManagedRef;

pub use crate::memory::ManagedStats;

pub use crate::p2p::P2pGraph;

pub use crate::p2p::P2pMsg;

pub use crate::p2p::P2pTopology;

pub use crate::pipeline::run_pipeline;

pub use crate::pipeline::spawn_pipeline;

pub use crate::pipeline::BoxedStage;

pub use crate::pipeline::PipelineExecutor;

pub use crate::pipeline::PipelineExecutorN;

pub use crate::pipeline::PipelineSink;

pub use crate::pipeline::PipelineSource;

pub use crate::pipeline::PipelineStage;

pub use crate::pipeline::StageBox;

pub use crate::placement::DeviceChoice;

pub use crate::placement::LeastLoadedPolicy;

pub use crate::placement::PlacementActor;

pub use crate::placement::PlacementHints;

pub use crate::placement::PlacementMsg;

pub use crate::placement::PlacementPolicy;

pub use crate::placement::RoundRobinPolicy;

pub use crate::replay::replay_via_sink;

pub use crate::replay::JournalEntry;

pub use crate::replay::ReplayHarness;

pub use crate::replay::ReplayMode;

pub use crate::replay::ReplayMsg;

pub use crate::replay::ReplaySink;

pub use crate::dtype::CudaDtype;

pub use crate::dtype::SolverSupported;

pub use crate::kernel::CholeskyRequest;

pub use crate::kernel::GesvdjBatchedRequest;

pub use crate::kernel::GetrfBatchedRequest;

pub use crate::kernel::HegvdRequest;

pub use crate::kernel::LuRequest;

pub use crate::kernel::LuSolveRequest;

pub use crate::kernel::PotrfBatchedRequest;

pub use crate::kernel::QrRequest;

pub use crate::kernel::SolverActor;

pub use crate::kernel::SolverDispatch;

pub use crate::kernel::SolverMsg;

pub use crate::kernel::SvdRequest;

pub use crate::kernel::SyevdRequest;

pub use crate::kernel::SygvdRequest;

pub use crate::kernel::Uplo;

pub use crate::kernel::CsrMatrix;

pub use crate::kernel::SparseActor;

pub use crate::kernel::SparseMsg;

pub use crate::dtype::TensorSupported;

pub use crate::kernel::ComputeDesc;

pub use crate::kernel::ContractRequest;

pub use crate::kernel::ElementwiseBinaryRequest;

pub use crate::kernel::ElementwiseTrinaryRequest;

pub use crate::kernel::OperandSpec;

pub use crate::kernel::PermutationRequest;

pub use crate::kernel::ReductionRequest;

pub use crate::kernel::TensorActor;

pub use crate::kernel::TensorMsg;

pub use crate::kernel::TensorSpec;

pub use crate::kernel::BlasLtActor;

pub use crate::kernel::BlasLtMsg;

pub use crate::kernel::BlasLtWorkspacePool;

pub use crate::kernel::Epilogue;

pub use crate::kernel::HeuristicCacheRef;

pub use crate::kernel::MatmulRequest;

pub use crate::kernel::ScaleSet;

pub use crate::kernel::KernelArg;

pub use crate::kernel::KernelHandle;

pub use crate::kernel::NvrtcActor;

pub use crate::kernel::NvrtcMsg;

pub use crate::kernel::NvrtcOpts;

pub use crate::kernel::AllGatherRequest;

pub use crate::kernel::AllReduceRequest;

pub use crate::kernel::AllToAllRequest;

pub use crate::kernel::AllToAllvRequest;

pub use crate::kernel::BroadcastRequest;

pub use crate::kernel::CollectiveActor;

pub use crate::kernel::CollectiveMsg;

pub use crate::kernel::GroupGuard;

pub use crate::kernel::NcclCapabilities;

pub use crate::kernel::NcclReduceSupported;

pub use crate::kernel::PreMulSumOp;

pub use crate::kernel::RecvRequest;

pub use crate::kernel::ReduceRequest;

pub use crate::kernel::ReduceScatterRequest;

pub use crate::kernel::SendRequest;

pub use crate::multi_device::NcclWorldActor;

pub use crate::multi_device::NcclWorldConfig;

pub use crate::multi_device::NcclWorldMsg;

pub use crate::stream::ActorHints;

pub use crate::stream::PerActorAllocator;

pub use crate::stream::PooledAllocator;

pub use crate::stream::Priority;

pub use crate::stream::SingleStreamAllocator;

pub use crate::stream::StreamAllocator;

pub use crate::stream::WorkloadKind;

pub use crate::kernel::ActivationFwdRequest;

pub use crate::kernel::ActivationKind;

pub use crate::kernel::ActivationRequest;

pub use crate::kernel::AttentionMask;

pub use crate::kernel::AttentionParams;

pub use crate::kernel::BatchNormRequest;

pub use crate::kernel::ConvBwdDataRequest;

pub use crate::kernel::ConvBwdFilterRequest;

pub use crate::kernel::ConvDescParams;

pub use crate::kernel::ConvForwardRequest;

pub use crate::kernel::ConvFwdRequest;

pub use crate::kernel::ConvParams;

pub use crate::kernel::CudnnActor;

pub use crate::kernel::CudnnMsg;

pub use crate::kernel::DropoutFwdRequest;

pub use crate::kernel::EpilogueKind;

pub use crate::kernel::GroupNormRequest;

pub use crate::kernel::InstanceNormRequest;

pub use crate::kernel::LayerNormRequest;

pub use crate::kernel::LrnFwdRequest;

pub use crate::kernel::LrnParams;

pub use crate::kernel::MultiHeadAttnBwdRequest;

pub use crate::kernel::MultiHeadAttnFwdRequest;

pub use crate::kernel::NormBwdRequest;

pub use crate::kernel::NormMode;

pub use crate::kernel::NormPhase;

pub use crate::kernel::PoolBwdRequest;

pub use crate::kernel::PoolFwdRequest;

pub use crate::kernel::PoolMode;

pub use crate::kernel::PoolParams;

pub use crate::kernel::RnnBwdRequest;

pub use crate::kernel::RnnDirection;

pub use crate::kernel::RnnFwdRequest;

pub use crate::kernel::RnnMode;

pub use crate::kernel::RnnParams;

pub use crate::kernel::SoftmaxFwdRequest;

pub use crate::kernel::SoftmaxMode;

pub use crate::kernel::SoftmaxRequest;

pub use crate::kernel::TensorLayout;

pub use crate::kernel::FftActor;

pub use crate::kernel::FftCallbackKind;

pub use crate::kernel::FftDirection;

pub use crate::kernel::FftKind;

pub use crate::kernel::FftMsg;

pub use crate::kernel::FftPlan;

pub use crate::kernel::FftPlanMany;

pub use crate::kernel::FftRequest;

pub use crate::kernel::PlanKey;

pub use crate::kernel::Distribution;

pub use crate::kernel::FillRequest;

pub use crate::kernel::RngActor;

pub use crate::kernel::RngGeneratorKind;

pub use crate::kernel::RngMsg;

Enums

Activation

Available activation for kernel fusing in matmul

ReduceOp