Issues

• tileiras 13.4 emits misaligned f16 row loads after reshape/broadcast on sm_89 (13.3 passes)

  NVIDIA/cuda-tile#26 · Filter by author hhdri opened on Oct 10, 2026

• tileiras 13.4.92 miscompiles a loop exit value on sm_90a / sm_100a (wrong result at -O1 and above)

  NVIDIA/cuda-tile#25 · Filter by author dawnop opened on Oct 5, 2026
  2

• mma_scaled accumulated over a K loop faults with cudaErrorMisalignedAddress on sm_103

  bug
  NVIDIA/cuda-tile#24 · Filter by author elogir opened on Aug 27, 2026
  3

• Host SIGSEGV in cuLaunchKernel when cuModuleUnload runs concurrently, with Tile IR modules only.

  bug
  NVIDIA/cuda-tile#23 · Filter by author elogir opened on Jul 30, 2026
  1

• Feature request: make_im2col_view

  enhancement
  NVIDIA/cuda-tile#22 · Filter by author Baxi-codes opened on Jul 24, 2026
  1

• CUTILE IR Feature request: smem atomics

  enhancement
  NVIDIA/cuda-tile#21 · Filter by author spapiniattestable opened on May 30, 2026
  3

  

• [FEA] Enable CACHE for LLVM_GIT_REPO and LLVM_BUILD_COMMIT_HASH.

  enhancement
  NVIDIA/cuda-tile#11 · Filter by author Peter9606 opened on Feb 5, 2026