Issues

• [BUG] Same-name PrimFunc Vars duplicate generated host-wrapper parameters

  tile-ai/tilelang#3413 · Filter by author dajiaohuang opened on Oct 2, 2026
  1

• [BUG] Autotuner input-cache compatibility check ignores tensor rank

  tile-ai/tilelang#3412 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] T.Pipelined(start, stop) treats stop as extent for nonzero start

  tile-ai/tilelang#3411 · Filter by author dajiaohuang opened on Oct 2, 2026
  1

• [BUG] Profiler validation unconditionally synchronizes CUDA

  tile-ai/tilelang#3410 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] Default do_bench cache flush stays on cuda:0 for nonzero current devices

  tile-ai/tilelang#3409 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] NVRTC and CuTeDSL can launch on the current device's stream instead of the tensor device

  tile-ai/tilelang#3408 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] Dynamic output dimensions represented by PrimExpr are not evaluated

  tile-ai/tilelang#3407 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG][Cython/PTO] Output shape from a scalar parameter is not resolved

  tile-ai/tilelang#3406 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] Dynamic output allocation conflates distinct Vars with the same name

  tile-ai/tilelang#3404 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] Index promotion removes explicit narrowing casts

  tile-ai/tilelang#3403 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] MFMA swizzle layout divides by zero when the row is shorter than vecSize

  tile-ai/tilelang#3402 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] ROCm auto target detection always reads GPU 0

  tile-ai/tilelang#3401 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] CUDA 12.9 SM100 256-bit vector copy lacks a 32-byte alignment guard

  tile-ai/tilelang#3399 · Filter by author dajiaohuang opened on Oct 2, 2026
  5

• [BUG] ROCm MFMA GEMM can omit output columns for non-atom warp tiles

  tile-ai/tilelang#3398 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] HIP atomic_addx2 lacks the documented FP16 and BF16 overloads

  tile-ai/tilelang#3397 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] SM100 two-CTA TMA cluster barrier may undercount expected bytes

  tile-ai/tilelang#3396 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] SM100 2-CTA TMA barrier expectation appears to overcount bytes

  tile-ai/tilelang#3395 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] HIP T.round default appears to use ties-away instead of ties-to-even

  tile-ai/tilelang#3394 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG][Ascend] T.copy from an FP4 view with an odd source offset floors to the wrong byte

  tile-ai/tilelang#3391 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG][Ascend] Runtime rejects scalar bool kernel parameters

  tile-ai/tilelang#3389 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] VerifyBufferInit misses T.any_of/T.all_of buffer reads

  tile-ai/tilelang#3388 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] PTODSL Philox rand_uniform can round to 1.0

  tile-ai/tilelang#3387 · Filter by author dajiaohuang opened on Oct 2, 2026

• [BUG] CuTeDSL sync_grid shares barrier state across concurrent launches

  tile-ai/tilelang#3386 · Filter by author dajiaohuang opened on Oct 2, 2026

• T.print on a global buffer repeats the full dump from every thread

  tile-ai/tilelang#3383 · Filter by author dajiaohuang opened on Oct 2, 2026

• Predicated LDG can ignore the enclosing store predicate

  tile-ai/tilelang#3381 · Filter by author dajiaohuang opened on Oct 2, 2026