Optimization: Fixup performance regressions incurred

* Calling the specialized helper for same types is not
  fast enough, so inline that operation manually there.

* When there is specialization for same types in a function,
  the fallback path cost a lot of performance, so force it to
  a separate function if it's present, and inline otherwise.

* Some of these lessons are still to be applied to binary
  operation code.

nuitka/build/static_src/HelpersOperationInplaceMatmult.c

@@ -45,6 +45,9 @@ static inline bool _BINARY_OPERATION_MATMULT_LONG_LONG_INPLACE(PyObject **operan
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = &PyLong_Type;
+    PyTypeObject *type2 = &PyLong_Type;
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -55,9 +58,6 @@ static inline bool _BINARY_OPERATION_MATMULT_LONG_LONG_INPLACE(PyObject **operan
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = &PyLong_Type;
-    PyTypeObject *type2 = &PyLong_Type;
-
     // No inplace number slot nb_inplace_matrix_multiply available for this type.
     assert(type2->tp_as_number == NULL || type2->tp_as_number->nb_inplace_matrix_multiply == NULL);
 
@@ -101,6 +101,9 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_LONG_INPLACE(PyObject **oper
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = Py_TYPE(*operand1);
+    PyTypeObject *type2 = &PyLong_Type;
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -111,9 +114,6 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_LONG_INPLACE(PyObject **oper
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = Py_TYPE(*operand1);
-    PyTypeObject *type2 = &PyLong_Type;
-
     binaryfunc islot = (type1->tp_as_number != NULL && NEW_STYLE_NUMBER_TYPE(type1))
                            ? type1->tp_as_number->nb_inplace_matrix_multiply
                            : NULL;
@@ -197,6 +197,9 @@ static inline bool _BINARY_OPERATION_MATMULT_LONG_OBJECT_INPLACE(PyObject **oper
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = &PyLong_Type;
+    PyTypeObject *type2 = Py_TYPE(operand2);
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -207,9 +210,6 @@ static inline bool _BINARY_OPERATION_MATMULT_LONG_OBJECT_INPLACE(PyObject **oper
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = &PyLong_Type;
-    PyTypeObject *type2 = Py_TYPE(operand2);
-
     // No inplace number slot nb_inplace_matrix_multiply available for this type.
     assert(type2->tp_as_number == NULL || type2->tp_as_number->nb_inplace_matrix_multiply == NULL);
 
@@ -290,6 +290,9 @@ static inline bool _BINARY_OPERATION_MATMULT_FLOAT_FLOAT_INPLACE(PyObject **oper
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = &PyFloat_Type;
+    PyTypeObject *type2 = &PyFloat_Type;
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -300,9 +303,6 @@ static inline bool _BINARY_OPERATION_MATMULT_FLOAT_FLOAT_INPLACE(PyObject **oper
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = &PyFloat_Type;
-    PyTypeObject *type2 = &PyFloat_Type;
-
     // No inplace number slot nb_inplace_matrix_multiply available for this type.
     assert(type2->tp_as_number == NULL || type2->tp_as_number->nb_inplace_matrix_multiply == NULL);
 
@@ -342,6 +342,9 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_FLOAT_INPLACE(PyObject **ope
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = Py_TYPE(*operand1);
+    PyTypeObject *type2 = &PyFloat_Type;
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -352,9 +355,6 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_FLOAT_INPLACE(PyObject **ope
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = Py_TYPE(*operand1);
-    PyTypeObject *type2 = &PyFloat_Type;
-
     binaryfunc islot = (type1->tp_as_number != NULL && NEW_STYLE_NUMBER_TYPE(type1))
                            ? type1->tp_as_number->nb_inplace_matrix_multiply
                            : NULL;
@@ -434,6 +434,9 @@ static inline bool _BINARY_OPERATION_MATMULT_FLOAT_OBJECT_INPLACE(PyObject **ope
         // execute stuff in-place.
     }
 
+    PyTypeObject *type1 = &PyFloat_Type;
+    PyTypeObject *type2 = Py_TYPE(operand2);
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -444,9 +447,6 @@ static inline bool _BINARY_OPERATION_MATMULT_FLOAT_OBJECT_INPLACE(PyObject **ope
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = &PyFloat_Type;
-    PyTypeObject *type2 = Py_TYPE(operand2);
-
     // No inplace number slot nb_inplace_matrix_multiply available for this type.
     assert(type2->tp_as_number == NULL || type2->tp_as_number->nb_inplace_matrix_multiply == NULL);
 
@@ -578,6 +578,9 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_OBJECT_INPLACE(PyObject **op
     if (Py_TYPE(*operand1) == Py_TYPE(operand2)) {
     }
 
+    PyTypeObject *type1 = Py_TYPE(*operand1);
+    PyTypeObject *type2 = Py_TYPE(operand2);
+
 #ifdef _MSC_VER
 #pragma warning(push)
 #pragma warning(disable : 4101)
@@ -588,9 +591,6 @@ static inline bool _BINARY_OPERATION_MATMULT_OBJECT_OBJECT_INPLACE(PyObject **op
 #pragma warning(pop)
 #endif
 
-    PyTypeObject *type1 = Py_TYPE(*operand1);
-    PyTypeObject *type2 = Py_TYPE(operand2);
-
     binaryfunc islot = (type1->tp_as_number != NULL && NEW_STYLE_NUMBER_TYPE(type1))
                            ? type1->tp_as_number->nb_inplace_matrix_multiply
                            : NULL;