Module einsum2

Expand description

Binary einsum contractions on strided views.

Modules§

backend: Backend abstraction for batched GEMM dispatch. Backend abstraction for batched GEMM dispatch.
bgemm_faer: Batched GEMM backend using the [faer] library. faer-backed batched GEMM kernel on strided views.
bgemm_naive: Batched GEMM fallback using explicit loops. Naive batched GEMM kernel on strided views.
contiguous: GEMM-ready operand types and preparation functions for contiguous data. GEMM-ready operand types and preparation functions for contiguous data.
dot_general: Axis-based general dot product API. Axis-based general dot product API.
plan: Contraction planning: axis classification and permutation computation. Einsum2 plan: axis classification and permutation computation.
raw_bgemm: Raw borrowed-layout batched GEMM entry points. Raw borrowed-layout batched GEMM entry points.
trace: Trace-axis reduction (summing axes that appear only in one operand). Trace-axis reduction for einsum operands.
util: Shared helpers (permutation inversion, multi-index iteration, dimension fusion). Shared helpers for strided-einsum2.

DotGeneralConfig: DotGeneral dimension configuration.
Einsum2Plan: Pre-computed execution plan for a binary einsum contraction.
RawStridedMut: Borrowed raw strided output layout.
RawStridedRef: Borrowed raw strided input layout.
StridedView: Dynamic-rank immutable strided view with lazy element operations.
StridedViewMut: Dynamic-rank mutable strided view.

AxisId: Trait alias for axis label types.
Backend: Trait for backends that can execute batched GEMM on contiguous operands.
Scalar: Trait alias for element types supported by the faer active backend.
ScalarBase: Shared trait bounds for all element types usable with einsum, independent of GEMM backend.

bgemm_raw_strided_into: Batched strided GEMM on raw borrowed layout metadata using the active backend.
bgemm_raw_strided_into_unchecked^⚠: Batched strided GEMM on raw borrowed layout metadata without validation.
bgemm_raw_with_backend_into: Batched strided GEMM on raw borrowed metadata using an explicit backend.
bgemm_raw_with_backend_into_unchecked^⚠: Unchecked variant of bgemm_raw_with_backend_into.
col_major_strides: Compute column-major strides (Julia default: first index varies fastest).
dot_general_into: Compute C = alpha * dot_general(A, B) + beta * C with the active backend.
dot_general_with_backend_into: Compute C = alpha * dot_general(A, B) + beta * C with an explicit backend.
einsum2_into: Binary einsum contraction: C = alpha * contract(A, B) + beta * C.
einsum2_into_owned: Binary einsum accepting owned inputs for zero-copy optimization.
einsum2_naive_into: Binary einsum for custom scalar types using naive GEMM.
einsum2_with_backend_into: Binary einsum with a pluggable GEMM backend.