[Dy2St] pir dy2st unittest verification - Part -3

[DrRyanHuang]

PR types

Others

PR changes

Others

Description

状态	单测	错误类别	备注	报错信息
✅93	test_pir_selectedrows
❌94	test_bert {test_train, test_train_composite}	?		段错误
95	test_tensor_memcpy_on_cpu
❌96	test_program_translator {TestIfElseEarlyReturn}	API		__eq__ Value & int
❌97	test_grad {TestGrad}	?		Segmentation fault
❌97	test_grad {TestGradLinear}	Save/Load

相关链接:

• PIR 动转静理想态单测推全验证任务列表（一期） #58633

[DrRyanHuang]

test_program_translator 无法开启理想态

class TestIfElseEarlyReturn(Dy2StTestBase):
    def test_ifelse_early_return1(self):
        answer = np.zeros([2, 2]) + 1
        static_func = paddle.jit.to_static(dyfunc_with_if_else_early_return1)
        out = static_func()
        np.testing.assert_allclose(answer, out[0].numpy(), rtol=1e-05)

    def test_ifelse_early_return2(self):
        answer = np.zeros([2, 2]) + 3
        static_func = paddle.jit.to_static(dyfunc_with_if_else_early_return2)
        out = static_func()
        np.testing.assert_allclose(answer, out[0].numpy(), rtol=1e-05)

    File "/home/aistudio/giao/Paddle/build/python/paddle/jit/sot/symbolic/interpreter.py", line 178, in wrapper
        return interpreter.run_sir(name, state)
    File "/home/aistudio/giao/Paddle/build/python/paddle/jit/sot/symbolic/interpreter.py", line 116, in run_sir
        outs = getattr(self, stmt.type)(stmt, inputs)
    File "/home/aistudio/giao/Paddle/build/python/paddle/jit/sot/symbolic/interpreter.py", line 150, in method
        return getattr(var, stmt.method)(*args[1:], **kwargs)

    TypeError: __eq__(): incompatible function arguments. The following argument types are supported:
    1. (self: paddle.base.libpaddle.pir.Value, arg0: paddle.base.libpaddle.pir.Value) -> bool

Invoked with: Value(define_op_name=pd_op.assign_value_, index=0, dtype=pd_op.tensor<1xi64>), 0

[DrRyanHuang]

test_bert 无法开启理想态
test_train 和 test_train_composite 都是

The owner op type is:pd_op.full_like
*** Check failure stack trace: ***
    @     0x7f869968e27d  google::LogMessage::Fail()
    @     0x7f869969052d  google::LogMessage::SendToLog()
    @     0x7f869968dd6d  google::LogMessage::Flush()
    @     0x7f8699691029  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f869af62e07  pir::detail::OpResultImpl::~OpResultImpl()
    @     0x7f869af66803  pir::Operation::Destroy()
    @     0x7f869af4ae26  pir::Block::erase()
    @     0x7f869948937b  _ZZN8pybind1112cpp_function10initializeIZN6paddle6pybind9BindBlockEPNS_7module_EEUlRN3pir5BlockEPNS6_9OperationEE5_vJS8_SA_EJNS_4nameENS_9is_methodENS_7siblingEA172_cEEEvOT_PFT0_DpT1_EDpRKT2_ENUlRNS_6detail13function_callEE1_4_FUNEST_
    @     0x7f8699038528  pybind11::cpp_function::dispatcher()
    @           0x5d80be  (unknown)
    @           0x5d8d8c  _PyObject_MakeTpCall
    @           0x4fedd4  (unknown)
    @           0x47c416  (unknown)
    @           0x499082  (unknown)
    @           0x5d8868  _PyFunction_Vectorcall
    @           0x47c416  (unknown)
    @           0x4990a5  (unknown)
    @           0x55cd91  _PyEval_EvalCodeWithName
    @           0x5d8941  _PyFunction_Vectorcall
    @           0x47c416  (unknown)
    @           0x4997f3  (unknown)
    @           0x5d8868  _PyFunction_Vectorcall
    @           0x47c416  (unknown)
    @           0x4990a5  (unknown)
    @           0x55cd91  _PyEval_EvalCodeWithName
    @           0x5d8941  _PyFunction_Vectorcall
    @           0x47c416  (unknown)
    @           0x4997f3  (unknown)
    @           0x5d8868  _PyFunction_Vectorcall
    @           0x5d8506  PyVectorcall_Call
    @           0x55f797  _PyEval_EvalFrameDefault
    @           0x55d078  _PyEval_EvalCodeWithName

[DrRyanHuang]

test_grad 无法开启理想态

--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0   paddle::pybind::GetValueDtype(pir::Value)
1   pir::Type::type_id()

----------------------
Error Message Summary:
----------------------
FatalError: `Segmentation fault` is detected by the operating system.
  [TimeInfo: *** Aborted at 1701419815 (unix time) try "date -d @1701419815" if you are using GNU date ***]
  [SignalInfo: *** SIGSEGV (@0x0) received by PID 1772305 (TID 0x7f0573b5e740) from PID 0 ***]

[SigureMo]

test_pir_selectedrows 可以改写下，该单测写法依赖于整图训练 train step（即 to_static 装饰包含前反向），但这个单测是不依赖于这个特性的，所以可以改写成 to_static 仅装饰 net 前向部分

[SigureMo]

LGTMeow