Issues

• Tiled replay placement forks to the token budget, not to the width its measured rate asks for (#364 open item)

  tensor4all/strided-rs#430 · Filter by author shinaoka opened on Nov 9, 2026
  

• View-path copies tile layouts that the plan/replay path now declines as having no reusable cache line

  tensor4all/strided-rs#428 · Filter by author Ryo-wtnb11 opened on Nov 9, 2026
  1

• ConcatenatePlan's whole-plan parallel path walks flat element ranges, untiled

  tensor4all/strided-rs#427 · Filter by author Ryo-wtnb11 opened on Nov 9, 2026
  1

• The shared block model sizes every operand with one element size; Strided.jl uses each array's own

  tensor4all/strided-rs#426 · Filter by author Ryo-wtnb11 opened on Nov 9, 2026
  1

  

• plan cost: cost a CopyPlan by segments and footprint, not by a flat per-element class rate (#363)

  tensor4all/strided-rs#364 · Filter by author shinaoka opened on Nov 4, 2026
  

• KernelClass::Copy rate is 4-19x below the measured cost of strided transposes on a Xeon AVX-512 host, so their width is chosen for a much smaller job

  tensor4all/strided-rs#363 · Filter by author Ryo-wtnb11 opened on Nov 4, 2026
  3

  

• Perf: 1T typed max/min (all, axis0) ~1.4x slower than erased; Zen5 axis-1 rows still unverified

  tensor4all/strided-rs#361 · Filter by author shinaoka opened on Nov 2, 2026

• strided-nn: evaluate stable softmax/loss gradients from arXiv:2610.31642

  tensor4all/strided-rs#335 · Filter by author shinaoka opened on Oct 25, 2026

• Version core types/traits separately from the permutation engine and stabilize the core

  tensor4all/strided-rs#296 · Filter by author shinaoka opened on Oct 12, 2026
  1

• compute_block_sizes: transposed-operand tile is ~1.5x off a hand-tiled optimum

  tensor4all/strided-rs#287 · Filter by author shinaoka opened on Oct 3, 2026

• Umbrella (long-term): grow strided-rs into an embeddable engine (C API, crate layout, docs)

  tensor4all/strided-rs#234 · Filter by author shinaoka opened on Aug 17, 2026

• Long-term: evolve fused elementwise plans into a backend-neutral Strided IR

  enhancement
  tensor4all/strided-rs#233 · Filter by author shinaoka opened on Aug 15, 2026

• Umbrella: audit and burn down all open issues (2026-08-10)

  tensor4all/strided-rs#230 · Filter by author shinaoka opened on Aug 10, 2026
  3

• Marker remediation: SAFETY comments, #[allow] rationales, INVARIANT markers

  tensor4all/strided-rs#219 · Filter by author shinaoka opened on Aug 2, 2026
  1

  

• Test organization: extract inline cfg(test) modules; resolve strided-perm's test-gated implementation modules

  tensor4all/strided-rs#218 · Filter by author shinaoka opened on Aug 2, 2026
  3

  

• strided-kernel: make copy_into_col_major policy-aware or remove the duplicate API

  tensor4all/strided-rs#212 · Filter by author shinaoka opened on Aug 2, 2026
  3

  

• Unify the parallel threshold and extend the ExecutionPolicy contract and rayon guard test to strided-perm

  tensor4all/strided-rs#209 · Filter by author shinaoka opened on Aug 2, 2026
  2

  

• strided-kernel: destination injectivity for elementwise RMW ops; consolidate validate_destination_layout; dead cost==0 split guard

  tensor4all/strided-rs#206 · Filter by author shinaoka opened on Aug 2, 2026
  2

  

• Track Faer typed MaybeUninit overwrite API for strided GEMM

  tensor4all/strided-rs#198 · Filter by author shinaoka opened on Jul 31, 2026
  7