Skip to content

Latest commit

 

History

History
116 lines (110 loc) · 18.1 KB

rq3-bug-reports.md

File metadata and controls

116 lines (110 loc) · 18.1 KB
Raw

Bug finding evidence (RQ3)

Note Annotations

  • Status: ✅ fixed; 🚨 high-priority bug; 🔵 explicitly confirmed by developers; ❌ won't fix;
  • Symptom: 💥 Crash or exception; 🧮 Result inconsistency (silent semantic bug); 🧴 Sanitizers;

PyTorch

  1. ✅💥🚨 SIGIOT when running model with conv2d and avgpool2d after optimize_for_inference · Issue #86535 · pytorch/pytorch · GitHub
  2. ✅🧮 optimize_for_inference leads to wrong results for model with conv2d, max and clip · Issue #86556 · pytorch/pytorch · GitHub
  3. ✅💥🚨 RuntimeError: could not construct a memory descriptor using a format tag · Issue #86664 · pytorch/pytorch · GitHub
  4. ✅💥 [NNPack] Runtime error with padded Conv1d and >=16 batch size · Issue #90142 · pytorch/pytorch · GitHub
  5. ✅💥 stable torch.sort crash with expanded tensor · Issue #91420 · pytorch/pytorch · GitHub
  6. ✅💥 [Crash] torch.searchsorted with out-of-bound sorter · Issue #91606 · pytorch/pytorch · GitHub
  7. ✅🧮 index_select with scalar input and 0-dimed vector leads to undeterministic output · Issue #94340 · pytorch/pytorch · GitHub
  8. ✅🧮 index_select with scalar input and 0-dimed vector leads to undeterministic output · Issue #94340 · pytorch/pytorch · GitHub
  9. ✅💥 torch.compile failed on torch.add with a constant python number · Issue #92324 · pytorch/pytorch · GitHub
  10. ✅💥 torch.compile generates wrong profiling program for randn_like · Issue #92368 · pytorch/pytorch · GitHub
  11. ✅💥 torch.compile generates wrong profiling program for function having transpose and lerp · Issue #93229 · pytorch/pytorch · GitHub
  12. ✅💥 torch.compile triggers assertion error when explicitly provide out=None · Issue #92814 · pytorch/pytorch · GitHub
  13. ✅💥 INTERNAL ASSERT FAILED in torch.compile when the input tensor of torch.clamp has requires_grad=True · Issue #93225 · pytorch/pytorch · GitHub
  14. ✅💥 torch.compile failed to run in-place operation unsqueeze_(0) · Issue #93259 · pytorch/pytorch · GitHub
  15. ✅🧮 stack + inplace operator produce wrong results in torch.compile · Issue #93283 · pytorch/pytorch · GitHub
  16. ✅🧮 [pt2] compiled model with cat and expand gives wrong results · Issue #93357 · pytorch/pytorch · GitHub
  17. ✅🧮🚨 [pt2] compiled function with cat and mul gives wrong results · Issue #93365 · pytorch/pytorch · GitHub
  18. ✅🧮 [pt2] cannot compile model with linear layer when the input has rank 1 · Issue #93372 · pytorch/pytorch · GitHub
  19. ✅💥 softmax + transpose + div_ triggers assertion fail in compile mode · Issue #93371 · pytorch/pytorch · GitHub
  20. ✅🧮🚨 [pt2] torch.where gives wrong results with torch.compile · Issue #93374 · pytorch/pytorch · GitHub
  21. ✅💥 torch.rsub with alpha=xxx triggers assertion fail in compile mode · Issue #93376 · pytorch/pytorch · GitHub
  22. ✅🧮 [pt2] compile gives wrong result for function having expand and div_ · Issue #93377 · pytorch/pytorch · GitHub
  23. ✅💥 [pt2] torch._inductor.exc.CppCompileError: C++ compile error when compiling neg and max · Issue #93380 · pytorch/pytorch · GitHub
  24. ✅💥 [pt2] exception when compiling max_pool2d_with_indices · Issue #93384 · pytorch/pytorch · GitHub
  25. ✅💥 [pt2] cannot compile function having gt, expand and add_ · Issue #93386 · pytorch/pytorch · GitHub
  26. ✅💥🚨 torch.compile trigger assertion error when executing histogramdd · Issue #93274 · pytorch/pytorch · GitHub
  27. ✅🧮 [pt2] torch.compile produces wrong results for masked_fill · Issue #93823 · pytorch/pytorch · GitHub
  28. ✅🧮 [pt2] torch.compile produces wrong results for function with reciprocal_ · Issue #93824 · pytorch/pytorch · GitHub
  29. ✅🧮 [pt2] torch.compile produces wrong results for function with neg on uint8 tensor · Issue #93829 · pytorch/pytorch · GitHub
  30. ✅💥 log_softmax + pad triggers assertion fail in compile mode · Issue #93819 · pytorch/pytorch · GitHub
  31. ✅💥 [pt2] Cannot compile model with neg and linear · Issue #93836 · pytorch/pytorch · GitHub
  32. ✅🧮 pad + gt produce wrong results in compile mode · Issue #93351 · pytorch/pytorch · GitHub
  33. ✅💥 [pt2] (interpolate with mode=nearest) + kthvalue triggers assertion fail in compile mode · Issue #93849 · pytorch/pytorch · GitHub
  34. ✅💥 [pt2] torch._inductor.exc.CppCompileError: C++ compile error when compiling argmax and min · Issue #94055 · pytorch/pytorch · GitHub
  35. ✅💥 Tensor.select + add_ triggers C++ Compile Error · Issue #94960 · pytorch/pytorch · GitHub
  36. ✅💥 torch.compile fails when using torch.sub with python constant · Issue #95181 · pytorch/pytorch · GitHub
  37. ✅💥 Tensor.copy_ + moveaxis Trigger Exception in Compile Mode · Issue #95262 · pytorch/pytorch · GitHub
  38. ✅🧮 torch.ge produces wrong results in compile mode when given int tensors · Issue #95695 · pytorch/pytorch · GitHub
  39. ✅💥 [pt2] bitwise_and + clamp_max Triggers Compilation Error · Issue #97968 · pytorch/pytorch · GitHub
  40. ✅🧮 [pt2] add + unfold + abs_ produces wrong results · Issue #98143 · pytorch/pytorch · GitHub
  41. ✅🧮 [pt2] pow + cos produces wrong result · Issue #98149 · pytorch/pytorch · GitHub
  42. ✅💥 torch._C._nn.fractional_max_pool3d Trigger Segmentation Fault · Issue #89648 · pytorch/pytorch · GitHub
  43. ✅💥🚨 torch.nn.functional.embedding_bag Trigger "IOT instruction" Failure · Issue #89677 · pytorch/pytorch · GitHub
  44. ✅🧴 torch.Tensor.index_select Trigger heap-buffer-overflow with AddressSanitizer · Issue #88940 · pytorch/pytorch · GitHub
  45. ✅🧴 nn.utils.rnn.pack_sequence Trigger heap-buffer-overflow with AddressSanitizer · Issue #88334 · pytorch/pytorch · GitHub
  46. ✅🚨🧴 MultiMarginLoss Trigger out-of-bound Read under Compute Sanitizer · Issue #88724 · pytorch/pytorch · GitHub
  47. ✅🧴 nn.functional.max_unpool3d Trigger heap-buffer-overflow with AddressSanitizer · Issue #88032 · pytorch/pytorch · GitHub
  48. ✅🧴 torch.nn.functional.interpolate Trigger heap-buffer-overflow with AddressSanitizer · Issue #88939 · pytorch/pytorch · GitHub
  49. ✅🧴 torch.fft.hfft Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88985 · pytorch/pytorch · GitHub
  50. ✅🧴 torch.nn.functional.interpolate Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88951 · pytorch/pytorch · GitHub
  51. 🔵💥 [JIT] Zero-channel conv2d cannot be applied with optimize_for_inference · Issue #91396 · pytorch/pytorch · GitHub
  52. 🔵💥 [JIT] Applying conv2d over Constants Leads to Exception · Issue #92740 · pytorch/pytorch · GitHub
  53. 🔵💥🚨 torch.compile failed on torch.bitwise_xor with a constant python number · Issue #93224 · pytorch/pytorch · GitHub
  54. 🔵🧮 dstack + reciprocal produce wrong result in compile mode · Issue #93078 · pytorch/pytorch · GitHub
  55. 🔵💥 min reduction on float16 tensor failed on certain shapes · Issue #93249 · pytorch/pytorch · GitHub
  56. 🔵🧮 torch.compile produce wrong result in interpolate when mode=bilinear · Issue #93262 · pytorch/pytorch · GitHub
  57. 🔵💥 argmin + view Trigger Exception in compile mode · Issue #95370 · pytorch/pytorch · GitHub
  58. 🔵🧮 torch.fmod produces inconsistent results in eager and compile mode · Issue #97333 · pytorch/pytorch · GitHub
  59. 🔵💥 [CPU Inductor] Compile error when passing float16 tensors to vector_norm + remainder · Issue #97758 · pytorch/pytorch · GitHub
  60. 🔵💥 [pt2] movedim + add_ + cat triggers exception · Issue #98122 · pytorch/pytorch · GitHub
  61. 🔵💥 torch.Tensor.flatten Trigger Segmentation Fault when trying to provide and output named dim · Issue #89718 · pytorch/pytorch · GitHub
  62. 🔵🧴 nn.functional.embedding_bag Trigger out-of-bound Read under Compute Sanitizer · Issue #88563 · pytorch/pytorch · GitHub
  63. 🔵🧴 torch.nn.CTCLoss Trigger heap-buffer-overflow under AddressSanitizer · Issue #88047 · pytorch/pytorch · GitHub
  64. 🔵🧴 torch.nn.ReplicationPad2D Report "invalid configuration argument" Error under Compute Sanitizer · Issue #89254 · pytorch/pytorch · GitHub
  65. 🔵🧴 torch.nn.LayerNorm Abort with "invalid device ordinal" Error · Issue #89218 · pytorch/pytorch · GitHub
  66. 🔵🧴 torch.svd_lowrank Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88942 · pytorch/pytorch · GitHub
  67. 🔵🧴 torch.linalg.lstsq Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88941 · pytorch/pytorch · GitHub
  68. 💥 Adding a linear layer leads to failure of optimize_for_mobile · Issue #86667 · pytorch/pytorch · GitHub
  69. 💥 [JIT] INTERNAL ASSERT FAILED when dispatching for torch.Tensor.view · Issue #90365 · pytorch/pytorch · GitHub
  70. 💥 [JIT] INTERNAL ASSERT FAILED torch.add with boolean primitive constant · Issue #90367 · pytorch/pytorch · GitHub
  71. 💥 [JIT] INTERNAL ASSERT FAILED torch.mul with boolean primitive constant · Issue #90366 · pytorch/pytorch · GitHub
  72. 💥 [JIT] Wrong type inference leads to misleading error message · Issue #90369 · pytorch/pytorch · GitHub
  73. 💥 [JIT] INTERNAL ASSERT FAILED when Conv2d and clamp used together · Issue #92563 · pytorch/pytorch · GitHub
  74. 💥 [JIT] Inconsistency in tensor shape between eager mode and JIT · Issue #92548 · pytorch/pytorch · GitHub
  75. 💥 [JIT][TracingCheckError] inplace ops incompatible with contiguous(.., channels_last) · Issue #92558 · pytorch/pytorch · GitHub
  76. ❌💥 [JIT] Linear + BatchNorm2d Trigger Inconsistency between Eager Mode and JIT · Issue #92674 · pytorch/pytorch · GitHub
  77. 💥 [JIT] Consecutive use of addmm Leads to Exception · Issue #92742 · pytorch/pytorch · GitHub
  78. ❌💥 [Crash][MKL] torch.linalg.eigvals crash with NaN · Issue #93124 · pytorch/pytorch · GitHub
  79. 🧴 torch.nn.CTCLoss Trigger out-of-bound Read under Compute Sanitizer · Issue #89208 · pytorch/pytorch · GitHub
  80. 🧴 torch.nn.functional.embedding_bag Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88950 · pytorch/pytorch · GitHub
  81. 🧴 torch.set_rng_state Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88949 · pytorch/pytorch · GitHub
  82. 🧴 torch.Tensor.msort Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88947 · pytorch/pytorch · GitHub
  83. 🧴 torch.linalg.eigvals Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88945 · pytorch/pytorch · GitHub
  84. 🧴 torch.topk Trigger RuntimError under UndefinedBehaviorSanitizer · Issue #88944 · pytorch/pytorch · GitHub
  85. 🧴 torch.vander Trigger RuntimeError with UndefinedBehaviorSanitizer · Issue #88943 · pytorch/pytorch · GitHub

TensorFlow

  1. 🔵💥 Inconsistant behavior of Conv2D between eager mode and tracing · Issue #57664 · tensorflow/tensorflow · GitHub
  2. 🔵💥 TFLite fails to run a model with a dense layer following an Add operator · Issue #57697 · tensorflow/tensorflow · GitHub
  3. 🔵💥 TFLite throws an error with certain tensor value · Issue #57708 · tensorflow/tensorflow · GitHub
  4. 🔵🧮 TFLite's max operator has wrong broadcasting behavior · Issue #57759 · tensorflow/tensorflow · GitHub
  5. 🔵💥 Issues · tensorflow/tensorflow · GitHub
  6. 🔵🧮 pow operation gives valid output even the input is invalid · Issue #57757 · tensorflow/tensorflow · GitHub
  7. 🔵🧮 TFLite produce wrong results when add follows a leakyrelu · Issue #57818 · tensorflow/tensorflow · GitHub
  8. 🔵💥 TFLite runner crashes with XOR and squeeze in the model · Issue #57882 · tensorflow/tensorflow · GitHub
  9. 🔵💥 Conv2D with XLA jit_compile=True fails to run · Issue #57748 · tensorflow/tensorflow · GitHub
  10. 🔵🧮 log operator outputs wrong results with XLA compilation · Issue #57744 · tensorflow/tensorflow · GitHub
  11. 🔵🧮 pow operator output nan for valid inputs · Issue #57747 · tensorflow/tensorflow · GitHub
  12. 🔵🧮 LRN operator outputs wrong results with jit_compile=True · Issue #57746 · tensorflow/tensorflow · GitHub
  13. 🔵💥 Conv2D layer fails to run with XLA on CUDA · Issue #57838 · tensorflow/tensorflow · GitHub
  14. 🔵🧴 tf.raw_ops.SegmentMax Behaves Differently Under CPU and GPU · Issue #58469 · tensorflow/tensorflow · GitHub
  15. ❌🧮 Inconsistent behavior of TF eager and XLA in int64 casting · Issue #57883 · tensorflow/tensorflow · GitHub

Note NeuRI or NNSmith is able to find a lot more TensorFlow bugs in addition to these above. Because we want to report bugs more responsibly, we use a buffer size of 15 reports until some of them are fixed. That being said, the 15 reported bugs were not fixed (though confirmed) by TensorFlow developers at the time we were developing NeuRI and consequently we discontinued bug finding for TensorFlow.