Issues

• [Bug Skip]: [New Case] segment_reduce accumulates in the input dtype, failing test_reduced_precision_accumulation for float16/bfloat16

  new_case_failureskipped_bmgskipped
  intel/torch-xpu-ops#6182 · Filter by author torchxpubot opened on Nov 12, 2026 · UT Backlog
  1

• [Bug Skip]: [Regression] segment_reduce float16/bfloat16 no longer matches CPU for the lengths and offsets variants

  regressionskipped_bmgskipped
  intel/torch-xpu-ops#6181 · Filter by author torchxpubot opened on Nov 12, 2026 · UT Backlog
  1

• [SparseCsrXPU] mm/addmm with out= ignore non-contiguous out.values()

  intel/torch-xpu-ops#6173 · Filter by author pearu opened on Nov 12, 2026

• [upstream_ut] NVFP4 scaled_mm tests skipped on Xe3 iGPU due to Xe2_Or_Later gate (pass when gate bypassed)

  test: ut
  intel/torch-xpu-ops#6172 · Filter by author mengfei25 opened on Nov 12, 2026
  

• [upstream_ut] test_scaled_mm_deepseek_error_messages never runs on XPU; actual BlockWise1x128/128x128 behavior is inconsistent

  test: ut
  intel/torch-xpu-ops#6171 · Filter by author mengfei25 opened on Nov 12, 2026
  

• [Feature][Windows] Enable the static XPU Triton launcher (zebin kernel format + StaticXpuLauncher) on Windows

  enhancementmodule: inductoros: Windows
  Feature

  intel/torch-xpu-ops#6167 · Filter by author BartoszKokoszko opened on Nov 12, 2026

• [BMG] [upstream_ut][Failed to collect] UT collection failures

  intel/torch-xpu-ops#6165 · Filter by author kaileiyx opened on Nov 12, 2026
  

• Track remaining large-index paths in nearest upsample kernels

  intel/torch-xpu-ops#6164 · Filter by author jiafuzha opened on Nov 12, 2026

• [B60] Track: Inductor/Triton compile is pathologically slow on BMG (root cause of 600s+ UT timeouts)

  dependency component: IGChw: BMGmodule: inductorperformance
  intel/torch-xpu-ops#6163 · Filter by author ZhaoqiongZ opened on Nov 12, 2026
  

• [xpu-alignment] [26-11-11] XPU grid_sampler_2d backward accumulates the fp16/bf16 grid gradient in the input dtype: 2048 unit contributions sum to 1024 (fp16) and 128 (bf16)

  ai_generated
  intel/torch-xpu-ops#6162 · Filter by author torchxpubot opened on Nov 12, 2026

• [wontfix] test_property_on_cuda_device_properties uses CUDA-only torch.cuda.get_device_properties on XPU

  test: ut
  intel/torch-xpu-ops#6159 · Filter by author daisyden opened on Nov 11, 2026

• [Skills] .claude/skills: contradictions, stale references and structure issues needing maintainer decision

  documentation
  intel/torch-xpu-ops#6156 · Filter by author EikanWang opened on Nov 11, 2026
  1

  

• [PVC] timm_models convnextv2_nano.fcmae_ft_in22k_in1k training/bfloat16 got fail_accuracy on Linux PVC(1100)

  hw: PVCneed_splittest: e2e
  intel/torch-xpu-ops#6155 · Filter by author kaileiyx opened on Nov 11, 2026

• [upstream_ut] Converting copy (int32/bool -> int64/float64) uses UnrolledElementwiseKernel instead of the vectorized kernel on XPU

  dtype: float64dtype: int32dtype: int64module: eltwiseperformancetest: ut
  Bug

  intel/torch-xpu-ops#6150 · Filter by author daisyden opened on Nov 11, 2026
  1

• [OneDNN 3.14.2][BMG] E2E Performance drops compared with OneDNN 3.13.3

  dependency component: oneDNNdtype: amp_fp16hw: BMGinferencemodule: gemmperformanceregressiontest: e2e
  Bug

  intel/torch-xpu-ops#6145 · Filter by author kaileiyx opened on Nov 11, 2026
  2

  

• [xccl] ProcessGroup collective hooks never fire on XPU, so NanCheckHook does not detect NaNs

  bugfunctionalitymodule: distributedtest: ut
  Bug

  intel/torch-xpu-ops#6144 · Filter by author madhumitha0102 opened on Nov 11, 2026 · PT2.16
  1

• [Bug Skip]: [Failed to collect] [Regression] test_cpp_api_parity_xpu.py fails to import: No module named 'cpp_api_parity'

  functionalitymodule: infraregressionskipped_bmgskippedtest: ut
  Bug

  intel/torch-xpu-ops#6143 · Filter by author torchxpubot opened on Nov 10, 2026 · UT Backlog
  1

  

• [Bug Skip]: [Failed to collect] [Regression] dynamo/test_misc_xpu.py fails to import: cannot import name 'TestTracer' from dynamo.test_misc

  functionalitymodule: infraregressionskipped_bmgskippedtest: ut
  Bug

  intel/torch-xpu-ops#6142 · Filter by author torchxpubot opened on Nov 10, 2026 · UT Backlog
  2

  

• [Bug Skip]: [Regression] test_transformers, test_profiler and test_nn XPU tests fail: module 'torch.jit' has no attribute 'script' / 'trace'

  duplicatefunctionalitymodule: infraregressionskipped_bmgskippedtest: ut
  Bug

  intel/torch-xpu-ops#6141 · Filter by author torchxpubot opened on Nov 10, 2026 · UT Backlog
  2

  

• [Bug Skip]: [Regression] torchvision op regression tests fail at torch.jit.script: module 'torch.jit' has no attribute 'script'

  functionalitymodule: infraregressionskipped_bmgskippedtest: ut
  Bug

  intel/torch-xpu-ops#6140 · Filter by author torchxpubot opened on Nov 10, 2026 · UT Backlog
  2

  

• [upstream_ut] aten::flip is not implemented for the QuantizedXPU backend (test_flip_xpu_float32)

  dtype: float32functionalitymodule: aotest: ut
  Bug

  intel/torch-xpu-ops#6137 · Filter by author daisyden opened on Nov 10, 2026
  1

• [XPU] Enable test_reused_subgraph_square_backward on device-generic test suite

  module: inductortest: ut
  Task

  intel/torch-xpu-ops#6136 · Filter by author gplutop7 opened on Nov 10, 2026
  1

  

• [PyTorch CI][BMG] Invalid ZE_AFFINITY_MASK crashes zeInit in OMIX Ubuntu 24.04

  dependency component: Level_Zerofunctionalityhw: BMGmodule: coreos: Linuxpytorch-ci-failuretest: ut
  Bug

  intel/torch-xpu-ops#6131 · Filter by author chuanqi129 opened on Nov 10, 2026
  1

• [xpu-alignment] [26-11-09] XPU Tensor.renorm divides by norm + 1e-7, so renormalised rows land below max_norm (33% short at max_norm=1e-7)

  Accuracyai_generateddtype: float32dtype: float64module: opstest: ut
  Bug

  intel/torch-xpu-ops#6128 · Filter by author torchxpubot opened on Nov 10, 2026
  1

• [xpu-alignment] [26-11-09] XPU segment_reduce accumulates fp16/bf16 in the input dtype: 4096 ones sum to 2048 (fp16) and 256 (bf16)

  Accuracyai_generateddtype: bfloat16dtype: float16module: reductiontest: ut
  Bug

  intel/torch-xpu-ops#6127 · Filter by author torchxpubot opened on Nov 10, 2026
  1