Crate tenferro_cpu

Expand description

CPU backend, kernels, provider selection, and CPU resource pools.

§Examples

use tenferro_cpu::CpuBackend;
use tenferro_tensor::{BackendSessionHost, Tensor, TensorRead};

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0])?;
let b = Tensor::from_vec_col_major(vec![2], vec![3.0_f64, 4.0])?;
let c = backend
    .with_backend_session(|session| {
        session.add_read(TensorRead::from_tensor(&a), TensorRead::from_tensor(&b))
    })??;
assert_eq!(c.as_slice::<f64>().unwrap(), &[4.0, 6.0]);

The deleted one-shot spellings do not compile on the owner or on a session. Each fixture below fails for that reason and nothing else.

use tenferro_cpu::CpuBackend;
use tenferro_tensor::Tensor;

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
let b = Tensor::from_vec_col_major(vec![2], vec![3.0_f64, 4.0]).unwrap();
let _ = backend.add(&a, &b);

use tenferro_cpu::CpuBackend;
use tenferro_tensor::Tensor;

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
let b = Tensor::from_vec_col_major(vec![2], vec![3.0_f64, 4.0]).unwrap();
let _ = backend.mul(&a, &b);

use tenferro_cpu::CpuBackend;
use tenferro_tensor::Tensor;

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
let _ = backend.exp(&a);

use tenferro_cpu::CpuBackend;
use tenferro_tensor::Tensor;

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
let _ = backend.reduce_sum(&a, &[0]);

use tenferro_cpu::CpuBackend;
use tenferro_tensor::Tensor;

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2, 2], vec![1.0_f64, 2.0, 3.0, 4.0]).unwrap();
let _ = backend.transpose(&a, &[1, 0]);

use tenferro_cpu::CpuBackend;
use tenferro_tensor::{DotGeneralConfig, Tensor};

let mut backend = CpuBackend::new();
let a = Tensor::from_vec_col_major(vec![2, 2], vec![1.0_f64, 2.0, 3.0, 4.0]).unwrap();
let config = DotGeneralConfig {
    lhs_contracting_dims: [1].as_slice().into(),
    rhs_contracting_dims: [0].as_slice().into(),
    lhs_batch_dims: [].as_slice().into(),
    rhs_batch_dims: [].as_slice().into(),
};
let _ = backend.dot_general(&a, &a, &config);

The owner no longer implements the cache-aware contraction entry, so an owner-level BackendCachedDot bound does not hold either:

use tenferro_cpu::CpuBackend;
use tenferro_tensor::BackendCachedDot;

fn requires_cached_dot<B: BackendCachedDot>(_backend: &mut B) {}

let mut backend = CpuBackend::new();
requires_cached_dot(&mut backend);

Re-exports§

pub use affinity::available_parallelism;
pub use affinity::process_cpu_affinity;
pub use affinity::process_cpu_affinity_count;
pub use affinity::CpuAffinityError;
pub use backend::CpuBackend;
pub use backend::CpuBackendError;
pub use backend::CpuBackendKind;
pub use backend::CpuExecutionInfo;
pub use backend::CpuExecutionMode;
pub use backend::CpuRuntimeIdentity;
pub use backend::ExternalCpuDomainRegistryError;
pub use context::CpuContext;
pub use context::CpuContextError;
pub use context::DEFAULT_WORKER_STACK_BYTES;
pub use provider::CpuExecutionContext;
pub use provider::ParallelMode;

Modules§

affinity
backend
context
provider: Object-safe CPU contraction provider contracts.

Macros§

same_variant_pair: Dispatch a same-variant pair of tensors to a typed kernel.
same_variant_unary: Dispatch a single tensor to a typed kernel and erase its result.

Structs§

AddOp: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. Addition.
BufferPoolStats: Snapshot of typed host buffers retained by a BufferPool.
CpuAffinityInput: CPU affinity metadata for one logical operation input.
CpuAffinitySelection: Deterministic CPU affinity selection returned by the pure resolver.
CpuBatchPolicy: A batch strategy together with the thresholds Auto uses.
CpuBatchThresholds: Thresholds that CpuBatchStrategy::Auto applies.
CpuDomainExecutorCapabilities: Immutable construction-time capabilities of a CPU domain executor.
CpuId: An operating-system logical CPU identifier.
CpuNode: One usable OS NUMA node and its process-allowed logical CPUs.
CpuProviderBundle: Immutable direct provider slots installed on a CPU backend.
CpuProviderBundleBuildError: Error returned when a custom CPU provider bundle omits mandatory slots.
CpuProviderBundleBuilder: Construction-time builder for immutable CPU provider slots.
CpuProviderExecutionCapabilities: Immutable execution capabilities declared by one CPU provider.
CpuSet: A sorted, deduplicated, non-empty set of logical CPUs.
CpuTopology: Process-visible CPU topology used for execution placement.
ExternalCpuDomain: Caller-supplied descriptor for one externally managed CPU resource domain.
IndexedPlanCacheLimits: Limits for the CPU indexed-plan cache.
MulOp: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. Multiplication.
NumaNodeId: An operating-system NUMA node identifier.
RayonCpuDomainExecutor: Adapter that executes CPU-domain jobs on one caller-owned Rayon pool.
SubOp: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. Subtraction.

Enums§

CpuAdmissionMode: Admission contract for one CPU resource domain.
CpuAffinityInputError: Failure to derive logical input bytes for CPU affinity resolution.
CpuAffinityPolicy: Policy used to select a CPU execution domain from input affinity metadata.
CpuAffinityResolutionError: Failure to resolve CPU affinity from input metadata.
CpuAffinitySelectionReason: Why the CPU affinity resolver selected a domain.
CpuBatchStrategy: How the items of one batched operation are executed.
CpuDomainExecutorError: Failure at the CPU executor admission or scheduling boundary.
CpuDomainOwnership: Ownership class of a CPU resource domain.
CpuEngineConstructionError: Typed failure raised while constructing a CPU execution engine.
CpuExecutorAffinity: Affinity claim made by a CPU domain executor.
CpuExecutorReentrancy: Re-entry capability of one CPU domain executor.
CpuExecutorShutdown: Ownership of CPU executor shutdown.
CpuInnerParallelism: Inner parallel-region support offered by a CPU domain executor.
CpuPlacement: Requested CPU execution placement.
CpuPlacementControl: Per-call control over where a CPU provider executes.
CpuPlacementError: Failure to resolve a CPU placement for the selected public provider kind.
CpuProviderBundleInstallError: Failure to install a CPU provider bundle for the backend’s domains.
CpuProviderDomainError: Typed incompatibility between a CPU provider and a selected CPU domain.
CpuProviderSlot: Provider slot that failed construction-time domain validation.
CpuSetError: Failure to construct a non-empty CPU set.
CpuThreadCountControl: Per-call control over the maximum number of threads used by a CPU provider.
CpuThreadExecution: What CPU execution the current thread is inside.
CpuTopologyError: Failure to canonicalize discovered NUMA topology.
ExternalCpuDomainError: Typed failure to construct an externally managed CPU resource domain.
GeneralContractionPolicy: Policy applied when the configured general-contraction provider reports a typed capability miss.
ResolvedCpuPlacement: Concrete CPU placement resolved for a managed domain or declared by an external domain.

Traits§

BinaryScalarOp: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. A binary operation between two scalars of the same type.
CpuDomainExecutor: Object-safe synchronous executor for one CPU resource domain.
FaerParallelismExt: Invoke a direct faer operation with the parallelism selected by a CPU session.
ScopedCpuJob: One borrowed job installed synchronously into a CPU domain executor.
ScopedCpuJobs: Synchronously submitted indexed jobs for engine-owned outer scheduling.

Functions§

cpu_capabilities: Return the CPU backend operation capability descriptor table.
current_cpu_execution: Report what CPU execution the current thread is inside, including a managed Rayon worker running a session or scope callback.
discover_cpu_topology: Discover the process-visible CPU and NUMA topology.
resolve_cpu_affinity: Resolve a CPU execution domain from input affinity metadata.
resolve_cpu_affinity_with_override: Resolve CPU affinity with an optional operation-local explicit override.
runtime_engine_id: Return the canonical CPU runtime engine identifier.
runtime_engine_registration: Build a runtime engine registration for a CpuBackend.
runtime_engine_registration_with_id: Build a runtime engine registration for a CpuBackend with a caller-selected engine identifier.
runtime_hardware_class: Return the canonical CPU runtime hardware class.
scalar_binary_into: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. Apply a named binary operation elementwise into a caller-owned destination.
scalar_fold: Ordinary CPU entry points that take a caller-provided destination and the caller’s own arithmetic instead of the typed pool. Fold every element of a caller-owned tensor with a named associative operation, starting from init.
with_batch_policy: Run f on session with policy as the effective batch policy.

Crate tenferro_cpu

Crate tenferro_cpu Copy item path

§Examples

Re-exports§

Modules§

Macros§

Structs§

Enums§

Traits§

Functions§

Crate tenferro_cpu