Search results

perf: avoid finalize copy in einsum2 by deferring output permutation (lazy permute)

Status: Open.

#128 In tensor4all/strided-rs;

· shinaoka opened on Feb 28, 2026

Extract transpose-flip helper in bgemm_blas.rs

Status: Open.

#127 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Simplify axis classification: reduce 6 groups to 4 + lazy trace

Status: Open.

#126 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Use cblas_dgemm_batch when available instead of manual batch loop

Status: Open.

#125 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

BufferPool: make it user-controllable via eval() API

Status: Open.

#124 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

BufferPool: use best-fit (>=) instead of exact-size matching

Status: Open.

#123 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Merge BackendConfig and BgemmBackend into a single trait

Status: Open.

#122 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Extract common pipeline from einsum2 entry points in lib.rs

Status: Open.

#121 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Reduce code duplication in contiguous.rs prepare functions

Status: Open.

#120 In tensor4all/strided-rs;

· shinaoka opened on Feb 20, 2026

Adopt OMEinsum.jl's pre-permutation dimension fusion in einsum2

Status: Open.

#116 In tensor4all/strided-rs;

· shinaoka opened on Feb 19, 2026

einsum2 batched GEMM has 3.5x overhead vs Julia for small-m contractions

Status: Open.

#115 In tensor4all/strided-rs;

· shinaoka opened on Feb 19, 2026

strided_perm::copy_into is 4x slower than Julia permutedims! for high-rank binary tensors

Status: Open.

#114 In tensor4all/strided-rs;

· shinaoka opened on Feb 19, 2026