Issues

• [FRONTEND][ONNX] OpNotImplemented: SimplifiedLayerNormalization and SkipSimplifiedLayerNormalization

  needs-triagetype:rfc-tracking
  apache/tvm#20625 · Filter by author Netra-Srinivasan-McW opened on Oct 9, 2026

• [Bug] LiftTransformParams produces an undefined Var in the rewritten function body, causing Relax VM codegen failure

  needs-triagetype: bug
  apache/tvm#20616 · Filter by author yunligou711-commits opened on Oct 8, 2026

• [Bug] DLight Matmul crashes on fused conv2d-bias-ReLU-pad consumer chain

  apache/tvm#20615 · Filter by author Nanmur opened on Oct 8, 2026

• [Bug] AllocateWorkspace leaves dangling GlobalVar references in nested workspace-using functions

  needs-triagetype: bug
  apache/tvm#20614 · Filter by author yunligou711-commits opened on Oct 8, 2026

• [Bug] LLVM codegen emits invalid FCmp instructions for bool equality comparisons

  needs-triagetype: bug
  apache/tvm#20613 · Filter by author yunligou711-commits opened on Oct 8, 2026

• [Bug][Relax][ONNX] Identity rejects ShapeExpr inputs

  apache/tvm#20611 · Filter by author Nanmur opened on Oct 8, 2026

• [Bug] ConvertLayout segfaults on Tensor + PrimValue

  needs-triagetype: bug
  apache/tvm#20608 · Filter by author tangyixuan01 opened on Oct 8, 2026
  1

• [Bug][Relax] ReorderTakeAfterMatmul fails on vector dot products

  apache/tvm#20595 · Filter by author Yuhx141 opened on Oct 7, 2026

• [Bug][Relax] OptimizeLayoutTransform cancels non-inverse layout transforms

  apache/tvm#20594 · Filter by author Yuhx141 opened on Oct 7, 2026

• [Bug][Relax] InlinePrivateFunctions removes functions referenced by closures

  apache/tvm#20593 · Filter by author Yuhx141 opened on Oct 7, 2026

• [Bug][Relax] RemoveUnusedOutputs drops impure calls when tuple outputs are unused

  apache/tvm#20592 · Filter by author Yuhx141 opened on Oct 7, 2026

• [Bug][Relax] ExpandMatmulOfSum crashes on reduction-axis broadcasting

  apache/tvm#20591 · Filter by author Yuhx141 opened on Oct 7, 2026

• [Bug] FoldBatchnormToConv2D turns a finite float32 output into NaN

  needs-triagetype: bug
  apache/tvm#20584 · Filter by author tangyixuan01 opened on Oct 7, 2026

• [Bug]LLVM codegen crashes on Add over bool (internal ICHECK: MatchesCode(DLDataTypeCode::kDLFloat))

  needs-triagetype: bug
  apache/tvm#20582 · Filter by author yunligou711-commits opened on Oct 7, 2026
  1

• [VOTE] Split TIRx foundation into a separate repo

  apache/tvm#20572 · Filter by author tqchen opened on Oct 6, 2026
  12

• [Bug] FuseTransposeMatmul crashes with AttributeError: 'TupleType' object has no attribute 'dtype' when the transposed operand is used outside the pattern

  needs-triagetype: bug
  apache/tvm#20569 · Filter by author tangyixuan01 opened on Oct 6, 2026

• [Bug] Relax printer emits type annotation that the parser cannot read back (TupleType of PrimType)

  needs-triagetype: bug
  apache/tvm#20559 · Filter by author yunligou711-commits opened on Oct 6, 2026

• [Bug] LLVM codegen aborts on arithmetic over "bool" buffers (Add/Sub/Mul)

  needs-triagetype: bug
  apache/tvm#20558 · Filter by author yunligou711-commits opened on Oct 6, 2026
  1

• [Bug][Relax] FoldBatchnormToConv2D crashes for the documented IOHW kernel layout

  apache/tvm#20534 · Filter by author Yuhx141 opened on Oct 3, 2026
  1

• [Bug][Relax] FoldBatchnormToConv2D crashes on float16 BatchNorm constants

  apache/tvm#20533 · Filter by author Yuhx141 opened on Oct 3, 2026
  1

• [Bug][Relax] FoldBatchnormToConv2D changes semantics for training BatchNorm

  apache/tvm#20532 · Filter by author Yuhx141 opened on Oct 3, 2026
  1

• [Bug][Relax] FoldBatchnormToConv2D changes semantics for a non-channel BatchNorm axis

  apache/tvm#20531 · Filter by author Yuhx141 opened on Oct 3, 2026

• [Bug][Relax] FoldBatchnormToConv2D ignores disabled scale and center attributes

  apache/tvm#20530 · Filter by author Yuhx141 opened on Oct 3, 2026

• [Bug] CUDA: the NVRTC compile path (the default) always passes --use_fast_math; no opt-out like TVM_CUDA_NVCC_NO_FAST_MATH on the nvcc path

  apache/tvm#20527 · Filter by author KitKyoD opened on Oct 3, 2026
  3

• [Bug] TIRx/CUDA: an annotated host-level scalar (c: T.float32 = ...) is allocated on the device and then written from the host (SIGSEGV)

  apache/tvm#20526 · Filter by author KitKyoD opened on Oct 3, 2026
  1