oneflow与pytorch一点算子在int类型上的区别

[xiezipeng-ML]

tensor.abs:

x = torch.tensor([1,-2,3,-4],dtype=torch.int64)
x.abs()
tensor([1, 2, 3, 4])
x = flow.tensor([1,-2,3,-4],dtype=flow.int64)
x.abs()
E0912 21:22:50.176399 3927826 maybe.h:276]
File "/home/xiezipeng/oneflow/oneflow/core/eager/opkernel_instruction_type.cpp", line 449, in Infer
operand->mut_opkernel()->ChooseOpKernel( operand->inputs(), operand->outputs(), operand->consistent_tensor_infer_result())
File "/home/xiezipeng/oneflow/oneflow/user/kernels/stateful_local_opkernel.cpp", line 446, in ChooseOpKernel
user_op::UserOpRegistryMgr::Get().GetOpKernelRegistryResult(op_type_name, *reg_ctx_)
Cannot find the kernel matching Current OperatorConf.
The Info of OperatorConf are
op_name: abs2
op_type_name: abs
DeviceType_Name: kCPU
DataType_Name of x_0: kInt64
DataType_Name of y_0: kInt64
op_kernels_not_found_debug_str: "(((device_tag == gpu [False]) and (data_type of tensor 'x' == kFloat)) and (data_type of tensor 'y' == kFloat))"
op_kernels_not_found_debug_str: "(((device_tag == gpu [False]) and (data_type of tensor 'x' == kDouble)) and (data_type of tensor 'y' == kDouble))"
op_kernels_not_found_debug_str: "(((device_tag == gpu [False]) and (data_type of tensor 'x' == kFloat16)) and (data_type of tensor 'y' == kFloat16))"
op_kernels_not_found_debug_str: "(((device_tag == cpu [True]) and (data_type of tensor 'x' == kFloat [False])) and (data_type of tensor 'y' == kFloat))"
op_kernels_not_found_debug_str: "(((device_tag == cpu [True]) and (data_type of tensor 'x' == kDouble [False])) and (data_type of tensor 'y' == kDouble))"
F0912 21:22:50.178730 3927826 opkernel_instruction_type.cpp:602] Check failed: MaybeIsOk(LocalCallOpKernelUtil::Infer(instruction))
*** Check failure stack trace: ***
@ 0x7f59a6f44d5d google::LogMessage::Fail()
@ 0x7f59a6f46b04 google::LogMessage::SendToLog()
@ 0x7f59a6f4488d google::LogMessage::Flush()
@ 0x7f59a6f47469 google::LogMessageFatal::~LogMessageFatal()
@ 0x7f59a3a7acd3 oneflow::vm::LocalCallOpKernelInstructionType::Compute()
@ 0x7f59a42ad446 oneflow::vm::CpuStreamType::Compute()
@ 0x7f59a42d469e oneflow::vm::StreamType::Run()
@ 0x7f59a42e24bf oneflow::vm::VirtualMachine::DispatchAndPrescheduleInstructions()
@ 0x7f59a42e304e oneflow::vm::VirtualMachine::Schedule()
@ 0x7f59a42c6454 oneflow::OneflowVM::Loop()
@ 0x7f5997b4519d execute_native_thread_routine
@ 0x7f59f9208609 start_thread
@ 0x7f59f912f293 clone
@ (nil) (unknown)
Aborted (core dumped)

tensor.mean:

x = torch.tensor([1,2,3,4],dtype=torch.int32)
x.mean()
Traceback (most recent call last):
File "", line 1, in
RuntimeError: Can only calculate the mean of floating types. Got Long instead.
x = flow.tensor([1,2,3,4],dtype=flow.int32)
x.mean()
tensor(0, dtype=oneflow.int32)

[doombeaker]

第一个问题是缺少对应的 kernel 注册。
第二个是没有对应检查。

两个问题都比较简单（第一个确认，第二个没完全确认），既和 user op/kernel 有关系，又只是完整 user op 中的很小一个环节。
适合想借任务机会了解 user op/kernel 的人练手，有哪位想主动认领的可以留言，然后和我沟通下怎么做（也可以不用直接和我沟通，本贴回复后，直接去 #4936 issue 更新记录就好）

[MARD1NO]

#6293 该PR已解决