Module dispatch 

Runtime-dispatched SIMD abstractions and dynamic facade. Runtime-dispatched SIMD operations.

Monomorphization chain

sum::<f32>(data) -> f32::sum(data) -> sum::dispatch_sum::<f32>(data) -> avx2 kernel.

Modules

argmax

Generic runtime-dispatch argmax kernel.

argmin

Generic runtime-dispatch argmin kernel.

binary

Generic runtime-dispatch elementwise binary kernel.

complex

Generic interleaved complex kernels.

dot

Generic runtime-dispatch dot product kernel.

gemm

Generic runtime-dispatch tiled GEMM kernel.

gemv

Generic runtime-dispatch register-blocked matrix–vector product (y += A · x).

gemv_strided

Generic runtime-dispatch register-blocked sub-matrix GEMV (y += A · x, row stride lda).

gemv_transpose

Generic runtime-dispatch transposed matrix–vector product (y += Aᵀ · x).

gemv_transpose_strided

Generic runtime-dispatch transposed sub-matrix GEMV (y += Aᵀ · x, row stride lda).

masked

Runtime-dispatched masked SIMD operations.

max

Generic runtime-dispatch max-reduction kernel.

min

Generic runtime-dispatch min-reduction kernel.

modular

Modular-arithmetic kernels for transform workloads.

scale

Generic runtime-dispatch in-place scale kernel.

sparse

Runtime-dispatched sparse matrix-vector multiplication.

sum

Generic runtime-dispatch sum kernel.

Traits

SimdOps

Sealed extension trait implementing dynamic runtime SIMD dispatch for any T: Scalar.

Functions

abs_max

Reduces the slice to max |x| (∞-norm accumulator); T::ZERO for empty.

abs_sum

Reduces the slice to Σ |x| (L1-norm accumulator); T::ZERO for empty.

argmax

Returns the first maximum, or None for empty or NaN-containing data.

argmin

Returns the first minimum, or None for empty or NaN-containing data.

axpy

Fused row update out[i] += alpha * x[i] (AXPY) via runtime-dispatched SIMD with no temporary allocation. Errors on length mismatch.

axpy_rows

Fused multi-row update out[row, i] += alphas[row] * x[i] via one runtime-dispatched SIMD kernel. out is a row-major strided window.

axpy_rows_batch

Fused batched multi-row update: out[row, i] += sum_k alphas[k, row] * x_panel[k, i] via one runtime-dispatched SIMD kernel. alphas is depth-major with rows elements per depth, x_panel is depth-major with cols elements per depth, and out is a row-major strided window.

dispatch_reduce_popcount

dispatch_reduce_popcount_and

dispatch_reduce_popcount_or

dispatch_reduce_popcount_xor

dot

Computes the dot product of two slices using runtime-dispatched SIMD.

elementwise_add

Computes the elementwise sum of two slices and writes to out.

elementwise_div

Computes the elementwise quotient of two slices and writes to out.

elementwise_mul

Computes the elementwise multiplication of two slices and writes to out.

elementwise_sub

Computes the elementwise difference of two slices and writes to out.

gemv

Computes register-blocked GEMV y += A · x with runtime backend selection.

gemv_strided

Computes register-blocked sub-matrix GEMV y += A · x with row stride lda, runtime backend selection. A is a row-major nrows × ncols block with leading dimension lda ≥ ncols; lda = ncols is the packed gemv(). Accumulates into y.

gemv_transpose

Computes register-blocked transposed GEMV y += Aᵀ · x with runtime backend selection — the complement of gemv().

gemv_transpose_strided

Computes register-blocked transposed sub-matrix GEMV y += Aᵀ · x with row stride lda, runtime backend selection. lda = ncols is the packed gemv_transpose(). Accumulates into y.

interleaved_complex_dot

Computes an interleaved complex dot product using a monomorphized SIMD architecture.

interleaved_complex_dot_runtime

Computes an interleaved complex dot product using Hermes runtime provider selection.

interleaved_complex_mul_assign

Multiplies interleaved complex values in-place using a monomorphized SIMD architecture.

interleaved_complex_mul_assign_runtime

Multiplies interleaved complex values in-place using Hermes runtime provider selection.

masked_add

Computes the elementwise sum of elements matching a boolean mask.

masked_dot

Computes the dot product of elements matching a boolean mask.

masked_sum

Computes the sum of elements matching a boolean mask.

max

Computes the maximum element of the slice using runtime-dispatched SIMD.

min

Computes the minimum element of the slice using runtime-dispatched SIMD.

ntt_butterfly_stage_u64

Executes one exact modular radix-2 NTT butterfly stage over u64 residues.

reduce_popcount

Computes the horizontal sum of population counts of all elements using runtime-dispatched SIMD.

reduce_popcount_and

Computes the horizontal sum of population counts of a[i] & b[i] using runtime-dispatched SIMD.

reduce_popcount_or

Computes the horizontal sum of population counts of a[i] | b[i] using runtime-dispatched SIMD.

reduce_popcount_xor

Computes the horizontal sum of population counts of a[i] ^ b[i] (Hamming distance) using runtime-dispatched SIMD.

scale

Multiplies every element of data by scalar in-place.

spmv_bcoo

Computes sparse SpMV using const-generic Blocked-COO tiles.

spmv_csr

Computes sparse SpMV using CSR: y += A · x.

spmv_dense_masked

Computes sparse SpMV using Dense-with-Mask.

spmv_sellp

Computes sparse SpMV using const-generic Sliced ELLPACK (SELL-p).

sum

Computes the sum of elements in the slice using runtime-dispatched SIMD.

tiled_gemm

Computes register-blocked tiled GEMM: c += A * B.