[Eager, Performance optimization] support multiply( * operator) to sink to Cpp layer (#46326)

* [Eager] math op sink to Cpp level

* fix ci errors

* draft version


* support + and - operator under cpp directly

* add static test

* polish code

* promote types or unify right type to left

* recover static test case

* polish code and fix some ci errors

* support complex and polish code

* fix conflicts

* fix windows ci errors

* fix windows-inference-ci errors

* polish and fix tests

* fix test case

* polish code

* [Eager, Performance optimization] support multiply( * operator) to sink to Cpp layer

* rm useless glog

* polish code

* polish code and fix code-format

* polish code

* fix CI

* polish code

paddle/fluid/pybind/eager_math_op_patch.cc

@@ -103,7 +103,7 @@ void SetDevice(paddle::platform::Place place) {
   if (paddle::platform::is_gpu_place(place)) {
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
     phi::backends::gpu::SetDeviceId(place.device);
-    VLOG(1) << "CurrentDeviceId: " << phi::backends::gpu::GetCurrentDeviceId()
+    VLOG(6) << "CurrentDeviceId: " << phi::backends::gpu::GetCurrentDeviceId()
             << " from " << static_cast<int>(place.device);
 #else
     PADDLE_THROW(paddle::platform::errors::PreconditionNotMet(
@@ -114,7 +114,7 @@ void SetDevice(paddle::platform::Place place) {
   if (paddle::platform::is_custom_place(place)) {
 #if defined(PADDLE_WITH_CUSTOM_DEVICE)
     phi::DeviceManager::SetDevice(place);
-    VLOG(1) << "CurrentDeviceId: "
+    VLOG(6) << "CurrentDeviceId: "
             << phi::DeviceManager::GetDevice(place.GetDeviceType()) << " from "
             << static_cast<int>(place.device);
 #else
@@ -139,6 +139,8 @@ paddle::experimental::Tensor CallScalarFuction(
 
   } else if (op_type == "rsub") {
     ret = scale_ad_func(self_tensor, phi::Scalar(-1.0), other, true);
+  } else if (op_type == "mul") {
+    ret = scale_ad_func(self_tensor, phi::Scalar(other), 0.0, true);
   }
 
   return ret;
@@ -431,6 +433,107 @@ static PyObject* tensor__rsub__method(TensorObject* self,
   EAGER_CATCH_AND_THROW_RETURN_NULL
 }
 
+static PyObject* tensor__mul__method(TensorObject* self,
+                                     PyObject* args,
+                                     PyObject* kwargs) {
+  paddle::platform::RecordEvent pythonc_record_event(
+      "__mul__ pybind_patch_func",
+      paddle::platform::TracerEventType::UserDefined,
+      1);
+
+  EAGER_TRY
+  VLOG(6) << "Running Eager tensor__mul__method";
+
+  // Set Device ID
+  auto place = egr::Controller::Instance().GetExpectedPlace();
+  SetDevice(place);
+
+  paddle::experimental::Tensor ret;
+  paddle::experimental::Tensor self_tensor = self->tensor;
+
+  PyObject* other_obj = PyTuple_GET_ITEM(args, 0);
+
+  // 1. scalar exists cases
+  if ((PyFloat_Check(other_obj) || PyLong_Check(other_obj)) &&
+      !PyBool_Check(other_obj)) {
+    float other = 0.0;
+    if (PyFloat_Check(other_obj)) {
+      other = CastPyArg2AttrFloat(other_obj, 0);
+      if (_supported_int_dtype_.find(self_tensor.dtype()) !=
+          _supported_int_dtype_.end()) {
+        eager_gil_scoped_release guard;
+        self_tensor = cast_ad_func(self_tensor, DataType::FLOAT32);
+      }
+    } else if (PyCheckInteger(other_obj) || IsNumpyType(other_obj)) {
+      other = static_cast<float>(CastPyArg2AttrInt(other_obj, 0));
+    }
+    {
+      eager_gil_scoped_release guard;
+      ret = CallScalarFuction(self_tensor, other, "mul");
+    }
+    return ToPyObject(ret);
+  }
+
+  // 2. create or get tensor for other_obj
+  paddle::experimental::Tensor other_tensor;
+  if (!PyCheckTensor(other_obj)) {
+    paddle::experimental::Scalar value =
+        CastPyArg2Scalar(other_obj, "__mul__", 0);
+    if (PyComplex_Check(other_obj)) {
+      eager_gil_scoped_release guard;
+      other_tensor = full_ad_func({1}, value, DataType::COMPLEX64, place);
+    } else {
+      eager_gil_scoped_release guard;
+      other_tensor =
+          full_ad_func(self_tensor.shape(), value, self_tensor.dtype(), place);
+    }
+  } else {
+    other_tensor = CastPyArg2Tensor(other_obj, 0);
+  }
+
+  // 3. promote types or unify right var type to left var
+  phi::DataType lhs_dtype = self_tensor.dtype();
+  phi::DataType rhs_dtype = other_tensor.dtype();
+  if (lhs_dtype != rhs_dtype) {
+    // note: only op_type in _supported_promote_complex_types_ should promote
+    // dtype
+    if (_complex_dtypes.find(lhs_dtype) != _complex_dtypes.end() ||
+        _complex_dtypes.find(rhs_dtype) != _complex_dtypes.end()) {
+      phi::DataType promote_dtype =
+          framework::TransToPhiDataType(framework::PromoteTypesIfComplexExists(
+              framework::TransToProtoVarType(lhs_dtype),
+              framework::TransToProtoVarType(rhs_dtype)));
+      if (lhs_dtype != promote_dtype) {
+        // cast
+        eager_gil_scoped_release guard;
+        self_tensor = cast_ad_func(self_tensor, promote_dtype);
+      }
+      if (rhs_dtype != promote_dtype) {
+        eager_gil_scoped_release guard;
+        other_tensor = cast_ad_func(other_tensor, promote_dtype);
+      }
+    } else {
+      LOG(WARNING)
+          << "The dtype of left and right Tensor are not the same, left "
+             "dtype is "
+          << lhs_dtype << ", but right dtype is " << rhs_dtype
+          << ", the right dtype will convert to " << lhs_dtype;
+      eager_gil_scoped_release guard;
+      other_tensor = cast_ad_func(other_tensor, lhs_dtype);
+    }
+  }
+
+  // 4. calculation
+  VLOG(6) << "Calling multiply_ad_func in tensor__mul__method";
+  {
+    eager_gil_scoped_release guard;
+    ret = multiply_ad_func(self_tensor, other_tensor);
+  }
+
+  return ToPyObject(ret);
+  EAGER_CATCH_AND_THROW_RETURN_NULL
+}
+
 PyMethodDef math_op_patch_methods[] = {
     {"__add__",
      (PyCFunction)(void (*)(void))tensor__add__method,
@@ -448,6 +551,14 @@ PyMethodDef math_op_patch_methods[] = {
      (PyCFunction)(void (*)(void))tensor__rsub__method,
      METH_VARARGS | METH_KEYWORDS,
      NULL},
+    {"__mul__",
+     (PyCFunction)(void (*)(void))tensor__mul__method,
+     METH_VARARGS | METH_KEYWORDS,
+     NULL},
+    {"__rmul__",
+     (PyCFunction)(void (*)(void))tensor__mul__method,
+     METH_VARARGS | METH_KEYWORDS,
+     NULL},
     {NULL, NULL, 0, NULL}};
 
 }  // namespace pybind

paddle/fluid/pybind/eager_utils.cc

@@ -1300,6 +1300,9 @@ paddle::experimental::Scalar CastPyArg2Scalar(PyObject* obj,
     return paddle::experimental::Scalar(value);
   } else if (type_name.find("numpy") != std::string::npos) {
     return CastNumpy2Scalar(obj, op_type, arg_pos);
+  } else if (PyComplex_Check(obj)) {
+    auto value = CastPyArg2Complex(obj, op_type, arg_pos);
+    return paddle::experimental::Scalar(value);
   } else if (PyObject_CheckLongOrToLong(&obj)) {
     int value = CastPyArg2Int(obj, op_type, arg_pos);
     return paddle::experimental::Scalar(value);

paddle/fluid/pybind/op_function_common.cc

@@ -31,6 +31,7 @@
 #include "paddle/fluid/imperative/type_defs.h"
 #include "paddle/fluid/operators/ops_extra_info.h"
 #include "paddle/fluid/pybind/imperative.h"
+#include "paddle/phi/common/complex.h"
 
 namespace py = pybind11;
 namespace paddle {
@@ -214,6 +215,25 @@ double CastPyArg2Double(PyObject* obj,
   return 0.0;
 }
 
+phi::dtype::complex<float> CastPyArg2Complex(PyObject* obj,
+                                             const std::string& op_type,
+                                             ssize_t arg_pos) {
+  if (PyComplex_Check(obj)) {
+    double real = PyComplex_RealAsDouble(obj);
+    double imag = PyComplex_ImagAsDouble(obj);
+    return phi::dtype::complex<float>(real, imag);
+  } else {
+    PADDLE_THROW(platform::errors::InvalidArgument(
+        "%s(): argument (position %d) must be "
+        "complex, but got %s",
+        op_type,
+        arg_pos + 1,
+        ((PyTypeObject*)obj->ob_type)->tp_name));  // NOLINT
+  }
+
+  return phi::dtype::complex<float>(0, 0);
+}
+
 void CastPyArg2AttrDouble(PyObject* obj,
                           paddle::framework::AttributeMap& attrs,  // NOLINT
                           const std::string& key,

paddle/fluid/pybind/op_function_common.h

@@ -61,6 +61,9 @@ float CastPyArg2Float(PyObject* obj,
 double CastPyArg2Double(PyObject* obj,
                         const std::string& op_type,
                         ssize_t arg_pos);
+phi::dtype::complex<float> CastPyArg2Complex(PyObject* obj,
+                                             const std::string& op_type,
+                                             ssize_t arg_pos);
 std::string CastPyArg2String(PyObject* obj,
                              const std::string& op_type,
                              ssize_t arg_pos);

python/paddle/fluid/dygraph/math_op_patch.py

@@ -389,10 +389,6 @@ def monkey_patch_math_varbase():
         ('ndim', _ndim_),
         ('size', _size_),
         ('T', _T_),
-        ('__mul__',
-         _binary_creator_('__mul__', 'multiply', False, _scalar_mul_, True)),
-        ('__rmul__',
-         _binary_creator_('__rmul__', 'multiply', False, _scalar_mul_, True)),
         ('__div__',
          _binary_creator_('__div__', 'divide', False, _scalar_div_, True)),
         ('__truediv__',
@@ -427,6 +423,8 @@ def monkey_patch_math_varbase():
         "__radd__",
         '__sub__',
         '__rsub__',
+        '__mul__',
+        '__rmul__',
     ]
 
     global _already_patch_varbase

python/paddle/fluid/tests/unittests/test_tensor_type_promotion.py

@@ -62,8 +62,7 @@ class TestTensorTypePromotion(unittest.TestCase):
     def test_operator(self):
         with _test_eager_guard():
             self.setUp()
-            # add and sub has been sunk to cpp level, there is no warnings to catch by this test.
-            self.mul_operator()
+            # add / sub / mul has been sunk to cpp level, there is no warnings to catch by this test.
             self.div_operator()
         self.setUp()
         self.add_operator()