The case works fine in TVM 0.6, but occured Check failed in TVM 0.8.
The recurrence step in TVM 0.8 is similar to the following:
-
add a function in Stage such as:
// include/tvm/te/schedule.h TVM_DLL Stage& TestArrayArrayPrimExpr(Array<Array<Expr>> input); // src/te/schedule/schedule_lang.cc Stage& Stage::TestArrayArrayPrimExpr(Array<Array<PrimExpr>> input){ for (const auto out : input) { for (const auto in : out) { std::cout << __FILE__ << ": " << __LINE__ << ": " << __FUNCTION__ << ": in is " << in << std::endl; } } return *this; }
The parameter type is nested Array: Array<Array<PrimExpr>>
-
Then register function as global function for calling in python, such as
TVM_REGISTER_GLOBAL("te.TestArrayArrayPrimExpr").set_body_method(&Stage::TestArrayArrayPrimExpr); -
Then wrap the function in tvm.te.schedule.Stage (tvm/te/schedule.py), such as
def test_aa(self, *args): _ffi_api.TestArrayArrayPrimExpr(self, *args) -
Create a python test and use the func, such as
def test_nest_array_case(self): data = tvm.placeholder((32,32), name="data", dtype="float16") out = tvm.compute((32,32), lambda *indice: data(*indice) + 1, name="out") # schedule s = tvm.create_schedule(out.op) dataL = s.cache_read(data, "shared", [out]) outL = s.cache_write(out, "local") s[outL].test_aa([['a', 'b'], ['c', 'd'], ['e', 123]]) -
Then I got stack trace and Check failed message below:
tvm._ffi.base.TVMError: Traceback (most recent call last): 1: TVMFuncCall 0: std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::runtime::TypedPackedFunc<tvm::te::Stage& (tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)>::AssignTypedLambda<tvm::runtime::Registry::set_body_method<tvm::te::Stage, tvm::te::Stage&, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> >(tvm::te::Stage& (tvm::te::Stage::*)(tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>))::{lambda(tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)#1}>(tvm::runtime::Registry::set_body_method<tvm::te::Stage, tvm::te::Stage&, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> >(tvm::te::Stage& (tvm::te::Stage::*)(tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>))::{lambda(tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)#1}, std::string)::{lambda(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*)#1}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&) 2: TVMFuncCall 1: std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::runtime::TypedPackedFunc<tvm::te::Stage& (tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)>::AssignTypedLambda<tvm::runtime::Registry::set_body_method<tvm::te::Stage, tvm::te::Stage&, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> >(tvm::te::Stage& (tvm::te::Stage::*)(tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>))::{lambda(tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)#1}>(tvm::runtime::Registry::set_body_method<tvm::te::Stage, tvm::te::Stage&, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> >(tvm::te::Stage& (tvm::te::Stage::*)(tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>))::{lambda(tvm::te::Stage, tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void>)#1}, std::string)::{lambda(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*)#1}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&) 0: tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> tvm::runtime::TVMPODValue_::AsObjectRef<tvm::runtime::Array<tvm::runtime::Array<tvm::PrimExpr, void>, void> >() const File "compiler_depend.ts", line 716 TVMError: In function te.TestArrayArrayPrimExpr: error while converting argument 1: [21:04:30] compiler_depend.ts:1593: --------------------------------------------------------------- An error occurred during the execution of TVM. For more information, please see: https://tvm.apache.org/docs/errors.html --------------------------------------------------------------- Check failed: (!checked_type.defined()) is false: Expected Array[Array[PrimExpr]], but got Array[index 0: Array[index 0: Array]]According to stack trace and Check failed info:
Expected Array[Array[PrimExpr]], but got Array[index 0: Array[index 0: Array]]
I found TVMPODValue_::AsObjectRef in code include/tvm/runtime/packed_func.h
if (type_code_ == kTVMObjectHandle) { // normal object type check. Object* ptr = static_cast<Object*>(value_.v_handle); Optional<String> checked_type = ObjectTypeChecker<TObjectRef>::CheckAndGetMismatch(ptr); ICHECK(!checked_type.defined()) << "Expected " << ObjectTypeChecker<TObjectRef>::TypeName() << ", but got " << checked_type.value(); return TObjectRef(GetObjectPtr<Object>(ptr)); }and ObjectTypeChecker::CheckAndGetMismatch in struct ObjectTypeChecker<Array>
struct ObjectTypeChecker<Array<T>> { static Optional<String> CheckAndGetMismatch(const Object* ptr) { if (ptr == nullptr) { return NullOpt; } if (!ptr->IsInstance<ArrayNode>()) { return String(ptr->GetTypeKey()); } const ArrayNode* n = static_cast<const ArrayNode*>(ptr); for (size_t i = 0; i < n->size(); i++) { const ObjectRef& p = (*n)[i]; Optional<String> check_subtype = ObjectTypeChecker<T>::CheckAndGetMismatch(p.get()); if (check_subtype.defined()) { return String("Array[index " + std::to_string(i) + ": " + check_subtype.value() + "]"); } } return NullOpt; } blablabla.... };I do not really understand interaction between c++ and python.
Why add ‘Array[index’ in return string when check_subtype.defined() in CheckAndGetMismatch?
How can I transfer param type as Array<Array<PrimExpr>> to c++ function from python in TVM 0.8 just like Array<Array<Expr>> in TVM 0.6 ?
Thanks a lot.