appnope segfaults on Apple Silicon

[geohot]

I haven't root caused why, but this PR disables it for arm64 chip. Otherwise on M1 the kernels segfault on startup.

[mwidjaja1]

Can confirm, this change fixes Jupyter for me on an ARM Macbook Air!

[erykoff]

Note that this solution is just a band-aid on a much deeper wound, see #562 . Something is not working with ctypes calls on Apple Silicon that is touched by appnope.

[geohot]

As far as I can tell, it's just appnope itself that's broken, nothing deeper. Has to do with whatever the implicit argtypes is for ctypes. Also objc_getClass returning None because Foundation wasn't loaded. Fix here:

minrk/appnope#7

[mwidjaja1]

I don't think this change is necessary anymore because as noted in the issue @geohot linked, the issue has been resolved in appnope. I updated appnope today, reverted this MR in my copy of iPython, and I was able to run Jupyterlab natively.

minrk/appnope#7 (comment)

[geohot]

Yes, should be good with appnope 0.1.1, which will already be installed. Closing PR.

[mtoseef99]

@geohot this update isn't working for me anymore, tried it many times after creating new envs.

Here is the sample output of installed packages and GPU installation:

• macOS Big Sur 11.6, M1 MacbookPro

• Python 3.9.7 | packaged by conda-forge | (default, Sep 29 2021, 19:24:02)

• GPU is available

• Num GPUs Available: 1

• Metal device set to: Apple M1

• systemMemory: 16.00 GB

• maxCacheSize: 5.33 GB

2021-10-20 22:11:58.637125: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2021-10-20 22:11:58.637250: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
Epoch 1/12
2021-10-20 22:11:58.825292: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)
2021-10-20 22:11:58.826147: W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz
2021-10-20 22:11:58.826201: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-10-20 22:11:58.856 python[44484:5804473] -[MPSGraph adamUpdateWithLearningRateTensor:beta1Tensor:beta2Tensor:epsilonTensor:beta1PowerTensor:beta2PowerTensor:valuesTensor:momentumTensor:velocityTensor:maximumVelocityTensor:gradientTensor:name:]: unrecognized selector sent to instance 0x168eff7d0
2021-10-20 22:11:58.869 python[44484:5804473] *** Terminating app due to uncaught exception 'NSInvalidArgumentException', reason: '-[MPSGraph adamUpdateWithLearningRateTensor:beta1Tensor:beta2Tensor:epsilonTensor:beta1PowerTensor:beta2PowerTensor:valuesTensor:momentumTensor:velocityTensor:maximumVelocityTensor:gradientTensor:name:]: unrecognized selector sent to instance 0x168eff7d0'
*** First throw call stack:
(
	0   CoreFoundation                      0x000000019bb8b838 __exceptionPreprocess + 240
	1   libobjc.A.dylib                     0x000000019b8b50a8 objc_exception_throw + 60
	2   CoreFoundation                      0x000000019bc1c694 -[NSObject(NSObject) __retain_OA] + 0
	3   CoreFoundation                      0x000000019baeccd4 ___forwarding___ + 1444
	4   CoreFoundation                      0x000000019baec670 _CF_forwarding_prep_0 + 96
	5   libmetal_plugin.dylib               0x0000000150b82290 _ZN12metal_plugin14MPSApplyAdamOpIfEC2EPNS_20OpKernelConstructionE + 656
	6   libmetal_plugin.dylib               0x0000000150b81ebc _ZN12metal_pluginL14CreateOpKernelINS_14MPSApplyAdamOpIfEEEEPvP23TF_OpKernelConstruction + 52
	7   libtensorflow_framework.2.dylib     0x00000001221685d4 _ZN10tensorflow12_GLOBAL__N_120KernelBuilderFactory6CreateEPNS_20OpKernelConstructionE + 88
	8   libtensorflow_framework.2.dylib     0x00000001221ea158 _ZN10tensorflow14CreateOpKernelENS_10DeviceTypeEPNS_10DeviceBaseEPNS_9AllocatorEPNS_22FunctionLibraryRuntimeEPNS_11ResourceMgrERKNSt3__110shared_ptrIKNS_14NodePropertiesEEEiPPNS_8OpKernelE + 784
	9   libtensorflow_framework.2.dylib     0x00000001223c52b8 _ZN10tensorflow21CreateNonCachedKernelEPNS_6DeviceEPNS_22FunctionLibraryRuntimeERKNSt3__110shared_ptrIKNS_14NodePropertiesEEEiPPNS_8OpKernelE + 272
	10  libtensorflow_framework.2.dylib     0x000000012236fc20 _ZN10tensorflow26FunctionLibraryRuntimeImpl12CreateKernelERKNSt3__110shared_ptrIKNS_14NodePropertiesEEEPNS_22FunctionLibraryRuntimeEPPNS_8OpKernelE + 600
	11  libtensorflow_framework.2.dylib     0x00000001223da430 _ZN10tensorflow22ImmutableExecutorState10InitializeERKNS_5GraphE + 1192
	12  libtensorflow_framework.2.dylib     0x00000001223c5064 _ZN10tensorflow16NewLocalExecutorERKNS_19LocalExecutorParamsERKNS_5GraphEPPNS_8ExecutorE + 304
	13  libtensorflow_framework.2.dylib     0x00000001223d2e6c _ZN10tensorflow12_GLOBAL__N_124DefaultExecutorRegistrar7Factory11NewExecutorERKNS_19LocalExecutorParamsERKNS_5GraphEPNSt3__110unique_ptrINS_8ExecutorENS9_14default_deleteISB_EEEE + 48
	14  libtensorflow_framework.2.dylib     0x00000001223d37e8 _ZN10tensorflow11NewExecutorERKNSt3__112basic_stringIcNS0_11char_traitsIcEENS0_9allocatorIcEEEERKNS_19LocalExecutorParamsERKNS_5GraphEPNS0_10unique_ptrINS_8ExecutorENS0_14default_deleteISG_EEEE + 92
	15  libtensorflow_framework.2.dylib     0x0000000122372278 _ZN10tensorflow26FunctionLibraryRuntimeImpl10CreateItemEPPNS0_4ItemE + 2676
	16  libtensorflow_framework.2.dylib     0x000000012237306c _ZN10tensorflow26FunctionLibraryRuntimeImpl3RunERKNS_22FunctionLibraryRuntime7OptionsEyN4absl12lts_202103244SpanIKNS_6TensorEEEPNSt3__16vectorIS8_NSB_9allocatorIS8_EEEENSB_8functionIFvRKNS_6StatusEEEE + 676
	17  libtensorflow_framework.2.dylib     0x00000001223810c0 _ZNK10tensorflow29ProcessFunctionLibraryRuntime14RunMultiDeviceERKNS_22FunctionLibraryRuntime7OptionsEyPNSt3__16vectorIN4absl12lts_202103247variantIJNS_6TensorENS_11TensorShapeEEEENS5_9allocatorISC_EEEEPNS6_INS5_10unique_ptrINS0_11CleanUpItemENS5_14default_deleteISI_EEEENSD_ISL_EEEENS5_8functionIFvRKNS_6StatusEEEENSP_IFSQ_RKNS0_21ComponentFunctionDataEPNS0_12InternalArgsEEEE + 2640
	18  libtensorflow_framework.2.dylib     0x0000000122384098 _ZNK10tensorflow29ProcessFunctionLibraryRuntime3RunERKNS_22FunctionLibraryRuntime7OptionsEyN4absl12lts_202103244SpanIKNS_6TensorEEEPNSt3__16vectorIS8_NSB_9allocatorIS8_EEEENSB_8functionIFvRKNS_6StatusEEEE + 2012
	19  libtensorflow_framework.2.dylib     0x0000000122384868 _ZNK10tensorflow29ProcessFunctionLibraryRuntime7RunSyncERKNS_22FunctionLibraryRuntime7OptionsEyN4absl12lts_202103244SpanIKNS_6TensorEEEPNSt3__16vectorIS8_NSB_9allocatorIS8_EEEE + 160
	20  _pywrap_tensorflow_internal.so      0x0000000134175554 _ZN10tensorflow19KernelAndDeviceFunc3RunEPNS_19ScopedStepContainerERKNS_15EagerKernelArgsEPNSt3__16vectorIN4absl12lts_202103247variantIJNS_6TensorENS_11TensorShapeEEEENS6_9allocatorISD_EEEEPNS_19CancellationManagerERKNS9_8optionalINS_25EagerRemoteFunctionParamsEEERKNSK_INS_17ManagedStackTraceEEE + 516
	21  _pywrap_tensorflow_internal.so      0x000000013413fd60 _ZN10tensorflow18EagerKernelExecuteEPNS_12EagerContextERKN4absl12lts_2021032413InlinedVectorIPNS_12TensorHandleELm4ENSt3__19allocatorIS6_EEEERKNS3_8optionalINS_25EagerRemoteFunctionParamsEEERKNS7_10unique_ptrINS_15KernelAndDeviceENS_4core15RefCountDeleterEEEPNS_14GraphCollectorEPNS_19CancellationManagerENS3_4SpanIS6_EERKNSD_INS_17ManagedStackTraceEEE + 372
	22  _pywrap_tensorflow_internal.so      0x00000001341463c4 _ZN10tensorflow11ExecuteNode3RunEv + 396
	23  _pywrap_tensorflow_internal.so      0x0000000134481764 _ZN10tensorflow13EagerExecutor11SyncExecuteEPNS_9EagerNodeE + 172
	24  _pywrap_tensorflow_internal.so      0x000000013413f89c _ZN10tensorflow12_GLOBAL__N_117EagerLocalExecuteEPNS_14EagerOperationEPPNS_12TensorHandleEPi + 1976
	25  _pywrap_tensorflow_internal.so      0x000000013413da44 _ZN10tensorflow12EagerExecuteEPNS_14EagerOperationEPPNS_12TensorHandleEPi + 296
	26  _pywrap_tensorflow_internal.so      0x0000000133da2ba4 _ZN10tensorflow14EagerOperation7ExecuteEN4absl12lts_202103244SpanIPNS_20AbstractTensorHandleEEEPi + 192
	27  _pywrap_tensorflow_internal.so      0x000000013417b92c _ZN10tensorflow21CustomDeviceOpHandler7ExecuteEPNS_27ImmediateExecutionOperationEPPNS_30ImmediateExecutionTensorHandleEPi + 468
	28  _pywrap_tensorflow_internal.so      0x00000001309c7f38 TFE_Execute + 80
	29  _pywrap_tensorflow_internal.so      0x0000000130944ac0 _Z24TFE_Py_ExecuteCancelableP11TFE_ContextPKcS2_PN4absl12lts_2021032413InlinedVectorIP16TFE_TensorHandleLm4ENSt3__19allocatorIS7_EEEEP7_objectP23TFE_CancellationManagerPNS5_IS7_Lm2ESA_EEP9TF_Status + 616
	30  _pywrap_tfe.so                      0x000000012540641c _ZN10tensorflow32TFE_Py_ExecuteCancelable_wrapperERKN8pybind116handleEPKcS5_S3_S3_PNS_19CancellationManagerES3_ + 160
	31  _pywrap_tfe.so                      0x0000000125437208 _ZZN8pybind1112cpp_function10initializeIZL25pybind11_init__pywrap_tfeRNS_7module_EE4$_44NS_6objectEJRKNS_6handleEPKcSA_S8_S8_S8_EJNS_4nameENS_5scopeENS_7siblingEEEEvOT_PFT0_DpT1_EDpRKT2_ENUlRNS_6detail13function_callEE_8__invokeESR_ + 184
	32  _pywrap_tfe.so                      0x00000001254190e0 _ZN8pybind1112cpp_function10dispatcherEP7_objectS2_S2_ + 3216
	33  python                              0x0000000104b73398 cfunction_call + 80
	34  python                              0x0000000104b1f1e8 _PyObject_MakeTpCall + 340
	35  python                              0x0000000104c2f6ac call_function + 724
	36  python                              0x0000000104c2bd44 _PyEval_EvalFrameDefault + 29268
	37  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	38  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	39  python                              0x0000000104c2f614 call_function + 572
	40  python                              0x0000000104c2be40 _PyEval_EvalFrameDefault + 29520
	41  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	42  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	43  python                              0x0000000104b22cf0 method_vectorcall + 164
	44  python                              0x0000000104c2f614 call_function + 572
	45  python                              0x0000000104c2be40 _PyEval_EvalFrameDefault + 29520
	46  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	47  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	48  python                              0x0000000104b22cf0 method_vectorcall + 164
	49  python                              0x0000000104c2f614 call_function + 572
	50  python                              0x0000000104c2be40 _PyEval_EvalFrameDefault + 29520
	51  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	52  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	53  python                              0x0000000104b1f468 _PyObject_FastCallDictTstate + 320
	54  python                              0x0000000104b201e0 _PyObject_Call_Prepend + 164
	55  python                              0x0000000104b9735c slot_tp_call + 376
	56  python                              0x0000000104b1fc34 _PyObject_Call + 156
	57  python                              0x0000000104c2c078 _PyEval_EvalFrameDefault + 30088
	58  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	59  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	60  python                              0x0000000104b22e50 method_vectorcall + 516
	61  python                              0x0000000104c2c078 _PyEval_EvalFrameDefault + 30088
	62  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	63  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	64  python                              0x0000000104b1f468 _PyObject_FastCallDictTstate + 320
	65  python                              0x0000000104b201e0 _PyObject_Call_Prepend + 164
	66  python                              0x0000000104b9735c slot_tp_call + 376
	67  python                              0x0000000104b1f1e8 _PyObject_MakeTpCall + 340
	68  python                              0x0000000104c2f6ac call_function + 724
	69  python                              0x0000000104c2bd44 _PyEval_EvalFrameDefault + 29268
	70  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	71  python                              0x0000000104b1fe64 _PyFunction_Vectorcall + 240
	72  python                              0x0000000104b22cf0 method_vectorcall + 164
	73  python                              0x0000000104c2f614 call_function + 572
	74  python                              0x0000000104c2be40 _PyEval_EvalFrameDefault + 29520
	75  python                              0x0000000104c244a8 _PyEval_EvalCode + 2968
	76  python                              0x0000000104c87834 pyrun_file + 376
	77  python                              0x0000000104c86d48 PyRun_SimpleFileExFlags + 816
	78  python                              0x0000000104ca9e84 Py_RunMain + 2916
	79  python                              0x0000000104cab018 pymain_main + 1272
	80  python                              0x0000000104ac5ddc main + 56
	81  libdyld.dylib                       0x000000019ba2d430 start + 4
)
libc++abi: terminating with uncaught exception of type NSException

[erykoff]

@mtoseef99 That appears to be a bug in tensorflow or somewhere else, it has no relation to ipython or appnope.

[mtoseef99]

@erykoff I tried different solutions, installing, re-installing miniforge and TensorFlow but nothing worked for me except changing the optimizer to sgd from adam.

Someone just talked about this on stack exchange today:
https://datascience.stackexchange.com/questions/102435/could-not-identify-numa-node-of-platform-gpu-id-0-on-m1-macbook/103338#103338

But using sgd is not helpful at all, accuracy is quite low in my case. I will further check it with different optimizers and architectures.