All pull requests

Pull requests

• [Test] Eliminate unnecessary compilation time

  #3517·penguin-wwy opened Oct 14, 2026Collaborator·

  ·
  2

• [Ascend] Support the tvm_ffi execution backend for the PTO target

  #3516·liggest opened Oct 14, 2026Contributor·

  ·
  3

• [Ascend] Add SS buffer support

  #3514·Denverjin opened Oct 14, 2026Collaborator·

  ·
  2

  

• [Ascend][PTO] Vectorize T.Parallel to PTO VMI: pipeline framework and element-wise support

  #3512·KurrinQu opened Oct 14, 2026·

  ·
  11

• [Ascend] Support partial L0 copies with explicit pitch

  #3511·Denverjin opened Oct 14, 2026Collaborator·

  ·
  12

  

• [Typing] Improve bundled TVM discovery and Python DSL declarations

  #3508·sepcnt opened Oct 13, 2026Contributor·

  ·
  5

• [Fix][Eager] Handle _eval_type type_params across Python versions

  #3505·sepcnt opened Oct 13, 2026Contributor·

  ·
  1

  2

• [CUDA][WS] Run a TMA-only producer loop in warp 0 only

  #3503·AkshayM5 opened Oct 13, 2026·

  ·
  1

  7

• [BugFix][Debug] Separate host and device sources in lower trace

  #3496·erhsh opened Oct 12, 2026Contributor·

  ·
  2

• [CUDA] Support SM120 MXFP8 and MXFP4 block-scaled MMA

  #3495·christopherowen opened Oct 12, 2026Contributor·

  ·
  3

• [Ascend] Fix SimtVF remat_idx EvalSet / vector gather pack

  ascend
  #3492·peanutchan opened Oct 12, 2026·

  ·
  1

  2

• [CUDA] Add L2 prefetching instructions with optional eviction policy

  #3490·dwindz opened Oct 12, 2026·

  ·
  7

• [transform] Flatten safe guards for positive constant divisors

  #3488·LLMZhangYC opened Oct 12, 2026Contributor·

  ·
  2

• [BugFix][Profiler] Validate output indices consistently

  #3486·dc-tw opened Oct 12, 2026Contributor·

  ·
  1

  4

• [BugFix] Only record validated TMA descriptor names

  #3484·adenzhou1350 opened Oct 12, 2026Contributor·

  ·
  3

• [WIP] Initialize sparse backward atomic gradient accumulators

  #3483·cla7aye15I4nd opened Oct 11, 2026Contributor·

  ·
  2

• [WIP] Fix stale routed expert accumulation across MoE forwards

  #3482·cla7aye15I4nd opened Oct 11, 2026Contributor·

  ·
  2

• [Experimental][MIR] Direct Host IR JIT and native cache

  #3481·sepcnt opened Oct 11, 2026Contributor·

  ·
  11

• Fix CUDA scalar reductions for partial warps

  #3478·1sgtpepper opened Oct 10, 2026·

  ·
  2

• [BugFix][JIT] Preserve cast operand grouping in host expressions

  #3477·adenzhou1350 opened Oct 10, 2026Contributor·

  ·
  2

• perf(autotuner): compile independent JIT configurations concurrently

  #3476·ZenAlexa opened Oct 10, 2026Contributor·

  ·
  4

  

• [Refactor][JIT] Unify CUDA and CuTeDSL execution through Host IR

  #3473·sepcnt opened Oct 10, 2026Contributor·

  ·
  2

  12

• [BugFix][Transform] Expand pipeline epilogues whose trip count is only known at runtime

  #3472·bigSheep123 opened Oct 10, 2026Contributor·

  ·
  2

• [Ascend]feat : support L0C fp32->f16/bf16 dual_copy on-path cast

  ascend
  #3468·yyyy999 opened Oct 9, 2026Contributor·

  ·
  18

  

• [Ascend] Support scalar sqrt outside VF scopes

  ascend
  #3466·dangyw18 opened Oct 9, 2026·

  ·
  7