Issues

• [XLA:CPU] while-loop compile time ~5x slower since small-while-loop threshold default went to 1e9

  openxla/xla#52122 · Filter by author BJMCox opened on Nov 2, 2026

• [XLA:GPU] Concurrent command buffers can stall when collectives together need more SMs than the GPU has

  openxla/xla#52099 · Filter by author loupicaaa opened on Nov 2, 2026

• [XLA:CPU] YNN reduce fusion after a batched trsm adds a layout copy and runs slower than the loop fusion

  openxla/xla#52091 · Filter by author BJMCox opened on Nov 2, 2026

• dot keeps avoidable transposes for w @ x under nested vmap and for unvmapped x @ w.T

  CPUenhancementerr:performance
  openxla/xla#52070 · Filter by author mvanhorn opened on Nov 1, 2026

• [GPU] spmd-simplify inflates the per-device memory of a collective-free shard_map (OOM and 8x compile time)

  bugGPUNVIDIA-GPUsharding
  openxla/xla#52041 · Filter by author PhilipVinc opened on Oct 31, 2026

• WaitAtBarrier without participating_tasks waits for every configured task, not "all the connected tasks" as documented

  CPUquestion
  openxla/xla#52040 · Filter by author vasilisnasopoulos opened on Oct 31, 2026

• GetAliveTasks (live_devices) with recoverable tasks: a call made before a restart counts for the restarted process

  bugerr: Runtime
  openxla/xla#52039 · Filter by author vasilisnasopoulos opened on Oct 31, 2026
  1

• [XLA:CPU] fft_thunk.cc RET_CHECK IsMonotonicWithDim0Major: FFT on a sharded dimension gets a non-default operand layout after SPMD all-gather

  bugCPUerr: Runtimeshardingstat:awaiting openxla-eng
  openxla/xla#51931 · Filter by author jbelofamd opened on Oct 30, 2026
  1

• [GPU] Processes sharing a GPU crash with Xid 13 SKEDCHECK22_INVALIDATE_ACTIVE_QMD under concurrent command-buffer scheduling

  bugerr: RuntimeGPUNVIDIA-GPU
  openxla/xla#51930 · Filter by author carlosgmartin opened on Oct 30, 2026

• [XLA:GPU] CUDA_ERROR_ILLEGAL_ADDRESS during GPU autotuning on RTX PRO 6000 Blackwell (sm_120); isolated GEMMs pass, autotune_level=0 + fusion autotuner off works

  bugerr: RuntimeGPUNVIDIA-GPU
  openxla/xla#51900 · Filter by author hz-xiaxz opened on Oct 29, 2026
  2

• [XLA] float6_e2m3fn / float6_e3m2fn: comparing with a constant aborts the process; convert and arithmetic fail to compile

  bugCPUerr: RuntimeGPUNVIDIA-GPU
  openxla/xla#51829 · Filter by author SomyaPadhy4501 opened on Oct 28, 2026
  1

• [XLA:CPU] s8/u8 add, subtract and multiply saturate instead of wrapping when fused with a max/min reduce (YNN fusion)

  bugCPU
  openxla/xla#51826 · Filter by author SomyaPadhy4501 opened on Oct 28, 2026
  1

• [GPU] Wrong f64 forward-over-reverse Hessian on GH200 with Triton GEMM enabled

  bugGPUNVIDIA-GPU
  openxla/xla#51786 · Filter by author dcyang opened on Oct 28, 2026

• [XLA:TPU] Scatter drops updates when an elementwise op on its updates is fused into it

  bugerr: RuntimeGoogle-TPU
  openxla/xla#51765 · Filter by author apullin opened on Oct 28, 2026
  1

• [XLA:GPU] SIGSEGV in libcuda with --xla_gpu_enable_command_buffer=+WHILE when another process shares an A100

  bugerr: RuntimeGPUNVIDIA-GPU
  openxla/xla#51747 · Filter by author micdoh opened on Oct 27, 2026

• [ROCm] Loop-emitter transpose that keeps the minor dim runs at ~14% of copy bandwidth on gfx1151 when the row stride is a multiple of 4 KiB

  AMD-GPUerr:performanceGPU
  openxla/xla#51666 · Filter by author FIM43-Redeye opened on Oct 27, 2026

• GPU: algebraic simplifier gives wrong results for an outer-product-mean + triangle-multiplication graph (jit != CPU/eager)

  bugGPUNVIDIA-GPU
  openxla/xla#51647 · Filter by author maraxen opened on Oct 26, 2026
  1

• Duplicate scatter computations are not CSE'd when accumulated into separate while loop carries

  bugerr:performanceGPUNVIDIA-GPUstat:awaiting openxla-eng
  openxla/xla#51576 · Filter by author ryan112358 opened on Oct 25, 2026
  1

• [XLA] AllReduceSimplifier misses eliminating product all-reduce on replicated inputs

  enhancementGPUNVIDIA-GPUstat:awaiting openxla-eng
  openxla/xla#51565 · Filter by author nikokurumi opened on Oct 24, 2026
  1

• [XLA] Wrong results when buffer assignment aliases a fusion's output with an operand it also reads through a bitcast

  bugCPUerr: RuntimeGoogle-TPUGPUNVIDIA-GPUstat:awaiting openxla-eng
  openxla/xla#51490 · Filter by author dennis-ant opened on Oct 23, 2026
  2

  

• [XLA:TPU] f32 ragged_dot with Precision.HIGHEST fails to compile on v6e at m=16388, k=8192, n=2048 (E1001 CompileTimeScopedVmemOom)

  bugerr: RuntimeGoogle-TPU
  openxla/xla#51481 · Filter by author boshjerns opened on Oct 22, 2026

• [XLA] GpuAllReduceCombiner misses combining smaller all-reduces after a threshold overflow

  bugerr:performanceGPUNVIDIA-GPU
  openxla/xla#51444 · Filter by author nikokurumi opened on Oct 22, 2026

• [XLA] SPMD partitioner aborts on a ()-shaped custom-call whose sharding is propagated from the output (HloSharding::GetUniqueDevice, hlo_sharding.cc:880)

  bugGPUNVIDIA-GPUsharding
  openxla/xla#51433 · Filter by author YuanchengJiang opened on Oct 22, 2026
  1

• [XLA:GPU] HloRematerialization host-offload strategy aborts with "Have nothing to place this before!" when the offloaded value's only user is an async-start (hlo_rematerialization.cc:2207)

  bugGPUNVIDIA-GPU
  openxla/xla#51432 · Filter by author YuanchengJiang opened on Oct 22, 2026

• TPU v5p: async all-to-all and all-gather never overlap independent compute (reduce-scatter / all-reduce do, via SparseCore offload)

  err:performanceGoogle-TPUstat:awaiting response from contributor
  openxla/xla#51400 · Filter by author bignamehyp opened on Oct 22, 2026
  2