PEP 465: a dedicated infix operator for matrix multiplication (closes #21176)

Doc/c-api/number.rst

@@ -30,6 +30,14 @@ Number Protocol
    the equivalent of the Python expression ``o1 * o2``.
 
 
+.. c:function:: PyObject* PyNumber_MatrixMultiply(PyObject *o1, PyObject *o2)
+
+   Returns the result of matrix multiplication on *o1* and *o2*, or *NULL* on
+   failure.  This is the equivalent of the Python expression ``o1 @ o2``.
+
+   .. versionadded:: 3.5
+
+
 .. c:function:: PyObject* PyNumber_FloorDivide(PyObject *o1, PyObject *o2)
 
    Return the floor of *o1* divided by *o2*, or *NULL* on failure.  This is
@@ -146,6 +154,15 @@ Number Protocol
    the Python statement ``o1 *= o2``.
 
 
+.. c:function:: PyObject* PyNumber_InPlaceMatrixMultiply(PyObject *o1, PyObject *o2)
+
+   Returns the result of matrix multiplication on *o1* and *o2*, or *NULL* on
+   failure.  The operation is done *in-place* when *o1* supports it.  This is
+   the equivalent of the Python statement ``o1 @= o2``.
+
+   .. versionadded:: 3.5
+
+
 .. c:function:: PyObject* PyNumber_InPlaceFloorDivide(PyObject *o1, PyObject *o2)
 
    Returns the mathematical floor of dividing *o1* by *o2*, or *NULL* on failure.

Doc/c-api/typeobj.rst

@@ -1121,6 +1121,9 @@ Number Object Structures
             binaryfunc nb_inplace_true_divide;
 
             unaryfunc nb_index;
+
+            binaryfunc nb_matrix_multiply;
+            binaryfunc nb_inplace_matrix_multiply;
        } PyNumberMethods;
 
    .. note::

Doc/library/dis.rst

@@ -364,6 +364,11 @@ result back on the stack.
    Implements ``TOS = TOS1 * TOS``.
 
 
+.. opcode:: BINARY_MATRIX_MULTIPLY
+
+   Implements ``TOS = TOS1 @ TOS``.
+
+
 .. opcode:: BINARY_FLOOR_DIVIDE
 
    Implements ``TOS = TOS1 // TOS``.
@@ -436,6 +441,11 @@ the original TOS1.
    Implements in-place ``TOS = TOS1 * TOS``.
 
 
+.. opcode:: INPLACE_MATRIX_MULTIPLY
+
+   Implements in-place ``TOS = TOS1 @ TOS``.
+
+
 .. opcode:: INPLACE_FLOOR_DIVIDE
 
    Implements in-place ``TOS = TOS1 // TOS``.

Doc/library/operator.rst

@@ -138,6 +138,14 @@ The mathematical and bitwise operations are the most numerous:
    Return ``a * b``, for *a* and *b* numbers.
 
 
+.. function:: matmul(a, b)
+              __matmul__(a, b)
+
+   Return ``a @ b``.
+
+   .. versionadded:: 3.5
+
+
 .. function:: neg(obj)
               __neg__(obj)
 
@@ -400,6 +408,8 @@ Python syntax and the functions in the :mod:`operator` module.
 +-----------------------+-------------------------+---------------------------------------+
 | Multiplication        | ``a * b``               | ``mul(a, b)``                         |
 +-----------------------+-------------------------+---------------------------------------+
+| Matrix Multiplication | ``a @ b``               | ``matmul(a, b)``                      |
++-----------------------+-------------------------+---------------------------------------+
 | Negation (Arithmetic) | ``- a``                 | ``neg(a)``                            |
 +-----------------------+-------------------------+---------------------------------------+
 | Negation (Logical)    | ``not a``               | ``not_(a)``                           |
@@ -508,6 +518,14 @@ will perform the update, so no subsequent assignment is necessary:
    ``a = imul(a, b)`` is equivalent to ``a *= b``.
 
 
+.. function:: imatmul(a, b)
+              __imatmul__(a, b)
+
+   ``a = imatmul(a, b)`` is equivalent to ``a @= b``.
+
+   .. versionadded:: 3.5
+
+
 .. function:: ior(a, b)
               __ior__(a, b)
 

Doc/library/token.rst

@@ -93,6 +93,7 @@ The token constants are:
           DOUBLESLASH
           DOUBLESLASHEQUAL
           AT
+          ATEQUAL
           RARROW
           ELLIPSIS
           OP

Doc/reference/datamodel.rst

@@ -1970,6 +1970,7 @@ left undefined.
 .. method:: object.__add__(self, other)
             object.__sub__(self, other)
             object.__mul__(self, other)
+            object.__matmul__(self, other)
             object.__truediv__(self, other)
             object.__floordiv__(self, other)
             object.__mod__(self, other)
@@ -1986,15 +1987,16 @@ left undefined.
       builtin: pow
       builtin: pow
 
-   These methods are called to implement the binary arithmetic operations (``+``,
+   These methods are called to implement the binary arithmetic operations
-   ``-``, ``*``, ``/``, ``//``, ``%``, :func:`divmod`, :func:`pow`, ``**``, ``<<``,
+   (``+``, ``-``, ``*``, ``@``, ``/``, ``//``, ``%``, :func:`divmod`,
-   ``>>``, ``&``, ``^``, ``|``).  For instance, to evaluate the expression
+   :func:`pow`, ``**``, ``<<``, ``>>``, ``&``, ``^``, ``|``).  For instance, to
-   ``x + y``, where *x* is an instance of a class that has an :meth:`__add__`
+   evaluate the expression ``x + y``, where *x* is an instance of a class that
-   method, ``x.__add__(y)`` is called.  The :meth:`__divmod__` method should be the
+   has an :meth:`__add__` method, ``x.__add__(y)`` is called.  The
-   equivalent to using :meth:`__floordiv__` and :meth:`__mod__`; it should not be
+   :meth:`__divmod__` method should be the equivalent to using
-   related to :meth:`__truediv__`.  Note that :meth:`__pow__` should be defined
+   :meth:`__floordiv__` and :meth:`__mod__`; it should not be related to
-   to accept an optional third argument if the ternary version of the built-in
+   :meth:`__truediv__`.  Note that :meth:`__pow__` should be defined to accept
-   :func:`pow` function is to be supported.
+   an optional third argument if the ternary version of the built-in :func:`pow`
+   function is to be supported.
 
    If one of those methods does not support the operation with the supplied
    arguments, it should return ``NotImplemented``.
@@ -2003,6 +2005,7 @@ left undefined.
 .. method:: object.__radd__(self, other)
             object.__rsub__(self, other)
             object.__rmul__(self, other)
+            object.__rmatmul__(self, other)
             object.__rtruediv__(self, other)
             object.__rfloordiv__(self, other)
             object.__rmod__(self, other)
@@ -2018,14 +2021,14 @@ left undefined.
       builtin: divmod
       builtin: pow
 
-   These methods are called to implement the binary arithmetic operations (``+``,
+   These methods are called to implement the binary arithmetic operations
-   ``-``, ``*``, ``/``, ``//``, ``%``, :func:`divmod`, :func:`pow`, ``**``,
+   (``+``, ``-``, ``*``, ``@``, ``/``, ``//``, ``%``, :func:`divmod`,
-   ``<<``, ``>>``, ``&``, ``^``, ``|``) with reflected (swapped) operands.
+   :func:`pow`, ``**``, ``<<``, ``>>``, ``&``, ``^``, ``|``) with reflected
-   These functions are only called if the left operand does not support the
+   (swapped) operands.  These functions are only called if the left operand does
-   corresponding operation and the operands are of different types. [#]_  For
+   not support the corresponding operation and the operands are of different
-   instance, to evaluate the expression ``x - y``, where *y* is an instance of
+   types. [#]_ For instance, to evaluate the expression ``x - y``, where *y* is
-   a class that has an :meth:`__rsub__` method, ``y.__rsub__(x)`` is called if
+   an instance of a class that has an :meth:`__rsub__` method, ``y.__rsub__(x)``
-   ``x.__sub__(y)`` returns *NotImplemented*.
+   is called if ``x.__sub__(y)`` returns *NotImplemented*.
 
    .. index:: builtin: pow
 
@@ -2043,6 +2046,7 @@ left undefined.
 .. method:: object.__iadd__(self, other)
             object.__isub__(self, other)
             object.__imul__(self, other)
+            object.__imatmul__(self, other)
             object.__itruediv__(self, other)
             object.__ifloordiv__(self, other)
             object.__imod__(self, other)
@@ -2054,17 +2058,17 @@ left undefined.
             object.__ior__(self, other)
 
    These methods are called to implement the augmented arithmetic assignments
-   (``+=``, ``-=``, ``*=``, ``/=``, ``//=``, ``%=``, ``**=``, ``<<=``, ``>>=``,
+   (``+=``, ``-=``, ``*=``, ``@=``, ``/=``, ``//=``, ``%=``, ``**=``, ``<<=``,
-   ``&=``, ``^=``, ``|=``).  These methods should attempt to do the operation
+   ``>>=``, ``&=``, ``^=``, ``|=``).  These methods should attempt to do the
-   in-place (modifying *self*) and return the result (which could be, but does
+   operation in-place (modifying *self*) and return the result (which could be,
-   not have to be, *self*).  If a specific method is not defined, the augmented
+   but does not have to be, *self*).  If a specific method is not defined, the
-   assignment falls back to the normal methods.  For instance, if *x* is an
+   augmented assignment falls back to the normal methods.  For instance, if *x*
-   instance of a class with an :meth:`__iadd__` method, ``x += y`` is equivalent
+   is an instance of a class with an :meth:`__iadd__` method, ``x += y`` is
-   to ``x = x.__iadd__(y)`` . Otherwise, ``x.__add__(y)`` and ``y.__radd__(x)``
+   equivalent to ``x = x.__iadd__(y)`` . Otherwise, ``x.__add__(y)`` and
-   are considered, as with the evaluation of ``x + y``. In certain situations,
+   ``y.__radd__(x)`` are considered, as with the evaluation of ``x + y``. In
-   augmented assignment can result in unexpected errors (see
+   certain situations, augmented assignment can result in unexpected errors (see
-   :ref:`faq-augmented-assignment-tuple-error`), but this behavior is in
+   :ref:`faq-augmented-assignment-tuple-error`), but this behavior is in fact
-   fact part of the data model.
+   part of the data model.
 
 
 .. method:: object.__neg__(self)

Doc/reference/expressions.rst

@@ -892,8 +892,9 @@ from the power operator, there are only two levels, one for multiplicative
 operators and one for additive operators:
 
 .. productionlist::
-   m_expr: `u_expr` | `m_expr` "*" `u_expr` | `m_expr` "//" `u_expr` | `m_expr` "/" `u_expr`
+   m_expr: `u_expr` | `m_expr` "*" `u_expr` | `m_expr` "@" `m_expr` |
-         : | `m_expr` "%" `u_expr`
+         : `m_expr` "//" `u_expr`| `m_expr` "/" `u_expr` |
+         : `m_expr` "%" `u_expr`
    a_expr: `m_expr` | `a_expr` "+" `m_expr` | `a_expr` "-" `m_expr`
 
 .. index:: single: multiplication
@@ -904,6 +905,13 @@ the other must be a sequence. In the former case, the numbers are converted to a
 common type and then multiplied together.  In the latter case, sequence
 repetition is performed; a negative repetition factor yields an empty sequence.
 
+.. index:: single: matrix multiplication
+
+The ``@`` (at) operator is intended to be used for matrix multiplication.  No
+builtin Python types implement this operator.
+
+.. versionadded:: 3.5
+
 .. index::
    exception: ZeroDivisionError
    single: division
@@ -1346,8 +1354,9 @@ groups from right to left).
 +-----------------------------------------------+-------------------------------------+
 | ``+``, ``-``                                  | Addition and subtraction            |
 +-----------------------------------------------+-------------------------------------+
-| ``*``, ``/``, ``//``, ``%``                   | Multiplication, division, remainder |
+| ``*``, ``@``, ``/``, ``//``, ``%``            | Multiplication, matrix              |
-|                                               | [#]_                                |
+|                                               | multiplication division,            |
+|                                               | remainder [#]_                      |
 +-----------------------------------------------+-------------------------------------+
 | ``+x``, ``-x``, ``~x``                        | Positive, negative, bitwise NOT     |
 +-----------------------------------------------+-------------------------------------+

Doc/reference/simple_stmts.rst

@@ -267,7 +267,7 @@ operation and an assignment statement:
 .. productionlist::
    augmented_assignment_stmt: `augtarget` `augop` (`expression_list` | `yield_expression`)
    augtarget: `identifier` | `attributeref` | `subscription` | `slicing`
-   augop: "+=" | "-=" | "*=" | "/=" | "//=" | "%=" | "**="
+   augop: "+=" | "-=" | "*=" | "@=" | "/=" | "//=" | "%=" | "**="
         : | ">>=" | "<<=" | "&=" | "^=" | "|="
 
 (See section :ref:`primaries` for the syntax definitions for the last three

Grammar/Grammar

@@ -40,7 +40,7 @@ small_stmt: (expr_stmt | del_stmt | pass_stmt | flow_stmt |
 expr_stmt: testlist_star_expr (augassign (yield_expr|testlist) |
                      ('=' (yield_expr|testlist_star_expr))*)
 testlist_star_expr: (test|star_expr) (',' (test|star_expr))* [',']
-augassign: ('+=' | '-=' | '*=' | '/=' | '%=' | '&=' | '|=' | '^=' |
+augassign: ('+=' | '-=' | '*=' | '@=' | '/=' | '%=' | '&=' | '|=' | '^=' |
             '<<=' | '>>=' | '**=' | '//=')
 # For normal assignments, additional restrictions enforced by the interpreter
 del_stmt: 'del' exprlist
@@ -97,7 +97,7 @@ xor_expr: and_expr ('^' and_expr)*
 and_expr: shift_expr ('&' shift_expr)*
 shift_expr: arith_expr (('<<'|'>>') arith_expr)*
 arith_expr: term (('+'|'-') term)*
-term: factor (('*'|'/'|'%'|'//') factor)*
+term: factor (('*'|'@'|'/'|'%'|'//') factor)*
 factor: ('+'|'-'|'~') factor | power
 power: atom trailer* ['**' factor]
 atom: ('(' [yield_expr|testlist_comp] ')' |

Include/Python-ast.h

@@ -15,9 +15,9 @@ typedef struct _slice *slice_ty;
 
 typedef enum _boolop { And=1, Or=2 } boolop_ty;
 
-typedef enum _operator { Add=1, Sub=2, Mult=3, Div=4, Mod=5, Pow=6, LShift=7,
+typedef enum _operator { Add=1, Sub=2, Mult=3, MatMult=4, Div=5, Mod=6, Pow=7,
-                         RShift=8, BitOr=9, BitXor=10, BitAnd=11, FloorDiv=12 }
+                         LShift=8, RShift=9, BitOr=10, BitXor=11, BitAnd=12,
-                         operator_ty;
+                         FloorDiv=13 } operator_ty;
 
 typedef enum _unaryop { Invert=1, Not=2, UAdd=3, USub=4 } unaryop_ty;
 

Include/abstract.h

@@ -658,6 +658,12 @@ xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx*/
      o1*o2.
        */
 
+     PyAPI_FUNC(PyObject *) PyNumber_MatrixMultiply(PyObject *o1, PyObject *o2);
+
+       /*
+     This is the equivalent of the Python expression: o1 @ o2.
+       */
+
      PyAPI_FUNC(PyObject *) PyNumber_FloorDivide(PyObject *o1, PyObject *o2);
 
        /*
@@ -832,6 +838,12 @@ xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx*/
      o1 *= o2.
        */
 
+     PyAPI_FUNC(PyObject *) PyNumber_InPlaceMatrixMultiply(PyObject *o1, PyObject *o2);
+
+       /*
+     This is the equivalent of the Python expression: o1 @= o2.
+       */
+
      PyAPI_FUNC(PyObject *) PyNumber_InPlaceFloorDivide(PyObject *o1,
                                                         PyObject *o2);
 

Include/object.h

@@ -275,6 +275,9 @@ typedef struct {
     binaryfunc nb_inplace_true_divide;
 
     unaryfunc nb_index;
+
+    binaryfunc nb_matrix_multiply;
+    binaryfunc nb_inplace_matrix_multiply;
 } PyNumberMethods;
 
 typedef struct {

Include/opcode.h

@@ -20,6 +20,9 @@ extern "C" {
 
 #define UNARY_INVERT    15
 
+#define BINARY_MATRIX_MULTIPLY 16
+#define INPLACE_MATRIX_MULTIPLY 17
+
 #define BINARY_POWER    19
 
 #define BINARY_MULTIPLY 20

Include/token.h

@@ -58,13 +58,14 @@ extern "C" {
 #define DOUBLESTAREQUAL	46
 #define DOUBLESLASH	47
 #define DOUBLESLASHEQUAL 48
-#define AT              49	
+#define AT              49
-#define RARROW          50
+#define ATEQUAL		50
-#define ELLIPSIS        51
+#define RARROW          51
+#define ELLIPSIS        52
 /* Don't forget to update the table _PyParser_TokenNames in tokenizer.c! */
-#define OP		52
+#define OP		53
-#define ERRORTOKEN	53
+#define ERRORTOKEN	54
-#define N_TOKENS	54
+#define N_TOKENS	55
 
 /* Special definitions for cooperation with parser */
 

Include/typeslots.h

@@ -74,3 +74,5 @@
 #define Py_tp_members 72
 #define Py_tp_getset 73
 #define Py_tp_free 74
+#define Py_nb_matrix_multiply 75
+#define Py_nb_inplace_matrix_multiply 76

Lib/importlib/_bootstrap.py

@@ -419,12 +419,13 @@ def _call_with_frames_removed(f, *args, **kwds):
 #     Python 3.4a4  3290 (changes to __qualname__ computation)
 #     Python 3.4a4  3300 (more changes to __qualname__ computation)
 #     Python 3.4rc2 3310 (alter __qualname__ computation)
+#     Python 3.5a0  3320 (matrix multiplication operator)
 #
 # MAGIC must change whenever the bytecode emitted by the compiler may no
 # longer be understood by older implementations of the eval loop (usually
 # due to the addition of new opcodes).
 
-MAGIC_NUMBER = (3310).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3320).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'

Lib/opcode.py

@@ -70,6 +70,9 @@ def_op('UNARY_NOT', 12)
 
 def_op('UNARY_INVERT', 15)
 
+def_op('BINARY_MATRIX_MULTIPLY', 16)
+def_op('INPLACE_MATRIX_MULTIPLY', 17)
+
 def_op('BINARY_POWER', 19)
 def_op('BINARY_MULTIPLY', 20)
 

Lib/operator.py

@@ -105,6 +105,10 @@ def mul(a, b):
     "Same as a * b."
     return a * b
 
+def matmul(a, b):
+    "Same as a @ b."
+    return a @ b
+
 def neg(a):
     "Same as -a."
     return -a
@@ -326,6 +330,11 @@ def imul(a, b):
     a *= b
     return a
 
+def imatmul(a, b):
+    "Same as a @= b."
+    a @= b
+    return a
+
 def ior(a, b):
     "Same as a |= b."
     a |= b
@@ -383,6 +392,7 @@ __invert__ = invert
 __lshift__ = lshift
 __mod__ = mod
 __mul__ = mul
+__matmul__ = matmul
 __neg__ = neg
 __or__ = or_
 __pos__ = pos
@@ -403,6 +413,7 @@ __ifloordiv__ = ifloordiv
 __ilshift__ = ilshift
 __imod__ = imod
 __imul__ = imul
+__imatmul__ = imatmul
 __ior__ = ior
 __ipow__ = ipow
 __irshift__ = irshift

Lib/test/test_augassign.py

@@ -136,6 +136,14 @@ class AugAssignTest(unittest.TestCase):
                 output.append("__imul__ called")
                 return self
 
+            def __matmul__(self, val):
+                output.append("__matmul__ called")
+            def __rmatmul__(self, val):
+                output.append("__rmatmul__ called")
+            def __imatmul__(self, val):
+                output.append("__imatmul__ called")
+                return self
+
             def __div__(self, val):
                 output.append("__div__ called")
             def __rdiv__(self, val):
@@ -233,6 +241,10 @@ class AugAssignTest(unittest.TestCase):
         1 * x
         x *= 1
 
+        x @ 1
+        1 @ x
+        x @= 1
+
         x / 1
         1 / x
         x /= 1
@@ -279,6 +291,9 @@ __isub__ called
 __mul__ called
 __rmul__ called
 __imul__ called
+__matmul__ called
+__rmatmul__ called
+__imatmul__ called
 __truediv__ called
 __rtruediv__ called
 __itruediv__ called

Lib/test/test_capi.py

@@ -150,6 +150,23 @@ class CAPITest(unittest.TestCase):
         self.assertEqual(_testcapi.docstring_with_signature_and_extra_newlines.__text_signature__,
             "($module, /, parameter)")
 
+    def test_c_type_with_matrix_multiplication(self):
+        M = _testcapi.matmulType
+        m1 = M()
+        m2 = M()
+        self.assertEqual(m1 @ m2, ("matmul", m1, m2))
+        self.assertEqual(m1 @ 42, ("matmul", m1, 42))
+        self.assertEqual(42 @ m1, ("matmul", 42, m1))
+        o = m1
+        o @= m2
+        self.assertEqual(o, ("imatmul", m1, m2))
+        o = m1
+        o @= 42
+        self.assertEqual(o, ("imatmul", m1, 42))
+        o = 42
+        o @= m1
+        self.assertEqual(o, ("matmul", 42, m1))
+
 
 @unittest.skipUnless(threading, 'Threading required for this test.')
 class TestPendingCalls(unittest.TestCase):

Lib/test/test_descr.py

@@ -4160,6 +4160,7 @@ order (MRO) for bases """
                 ('__add__',      'x + y',                   'x += y'),
                 ('__sub__',      'x - y',                   'x -= y'),
                 ('__mul__',      'x * y',                   'x *= y'),
+                ('__matmul__',   'x @ y',                   'x @= y'),
                 ('__truediv__',  'operator.truediv(x, y)',  None),
                 ('__floordiv__', 'operator.floordiv(x, y)', None),
                 ('__div__',      'x / y',                   'x /= y'),

Lib/test/test_grammar.py

@@ -985,6 +985,20 @@ class GrammarTests(unittest.TestCase):
         self.assertFalse((False is 2) is 3)
         self.assertFalse(False is 2 is 3)
 
+    def test_matrix_mul(self):
+        # This is not intended to be a comprehensive test, rather just to be few
+        # samples of the @ operator in test_grammar.py.
+        class M:
+            def __matmul__(self, o):
+                return 4
+            def __imatmul__(self, o):
+                self.other = o
+                return self
+        m = M()
+        self.assertEqual(m @ m, 4)
+        m @= 42
+        self.assertEqual(m.other, 42)
+
 
 def test_main():
     run_unittest(TokenTests, GrammarTests)

Lib/test/test_operator.py

@@ -203,6 +203,15 @@ class OperatorTestCase:
         self.assertRaises(TypeError, operator.mul, None, None)
         self.assertTrue(operator.mul(5, 2) == 10)
 
+    def test_matmul(self):
+        operator = self.module
+        self.assertRaises(TypeError, operator.matmul)
+        self.assertRaises(TypeError, operator.matmul, 42, 42)
+        class M:
+            def __matmul__(self, other):
+                return other - 1
+        self.assertEqual(M() @ 42, 41)
+
     def test_neg(self):
         operator = self.module
         self.assertRaises(TypeError, operator.neg)
@@ -416,6 +425,7 @@ class OperatorTestCase:
             def __ilshift__  (self, other): return "ilshift"
             def __imod__     (self, other): return "imod"
             def __imul__     (self, other): return "imul"
+            def __imatmul__  (self, other): return "imatmul"
             def __ior__      (self, other): return "ior"
             def __ipow__     (self, other): return "ipow"
             def __irshift__  (self, other): return "irshift"
@@ -430,6 +440,7 @@ class OperatorTestCase:
         self.assertEqual(operator.ilshift  (c, 5), "ilshift")
         self.assertEqual(operator.imod     (c, 5), "imod")
         self.assertEqual(operator.imul     (c, 5), "imul")
+        self.assertEqual(operator.imatmul  (c, 5), "imatmul")
         self.assertEqual(operator.ior      (c, 5), "ior")
         self.assertEqual(operator.ipow     (c, 5), "ipow")
         self.assertEqual(operator.irshift  (c, 5), "irshift")

Lib/test/test_sys.py

@@ -952,7 +952,7 @@ class SizeofTest(unittest.TestCase):
         check(int, s)
         # (PyTypeObject + PyNumberMethods + PyMappingMethods +
         #  PySequenceMethods + PyBufferProcs + 4P)
-        s = vsize('P2n15Pl4Pn9Pn11PIP') + struct.calcsize('34P 3P 10P 2P 4P')
+        s = vsize('P2n17Pl4Pn9Pn11PIP') + struct.calcsize('34P 3P 10P 2P 4P')
         # Separate block for PyDictKeysObject with 4 entries
         s += struct.calcsize("2nPn") + 4*struct.calcsize("n2P")
         # class

Lib/test/test_tokenize.py

@@ -464,7 +464,7 @@ Additive
 
 Multiplicative
 
-    >>> dump_tokens("x = 1//1*1/5*12%0x12")
+    >>> dump_tokens("x = 1//1*1/5*12%0x12@42")
     ENCODING   'utf-8'       (0, 0) (0, 0)
     NAME       'x'           (1, 0) (1, 1)
     OP         '='           (1, 2) (1, 3)
@@ -479,6 +479,8 @@ Multiplicative
     NUMBER     '12'          (1, 13) (1, 15)
     OP         '%'           (1, 15) (1, 16)
     NUMBER     '0x12'        (1, 16) (1, 20)
+    OP         '@'           (1, 20) (1, 21)
+    NUMBER     '42'          (1, 21) (1, 23)
 
 Unary
 
@@ -1154,6 +1156,7 @@ class TestTokenize(TestCase):
         self.assertExactTypeEqual('//', token.DOUBLESLASH)
         self.assertExactTypeEqual('//=', token.DOUBLESLASHEQUAL)
         self.assertExactTypeEqual('@', token.AT)
+        self.assertExactTypeEqual('@=', token.ATEQUAL)
 
         self.assertExactTypeEqual('a**2+b**2==c**2',
                                   NAME, token.DOUBLESTAR, NUMBER,

Lib/token.py

@@ -60,11 +60,12 @@ DOUBLESTAREQUAL = 46
 DOUBLESLASH = 47
 DOUBLESLASHEQUAL = 48
 AT = 49
-RARROW = 50
+ATEQUAL = 50
-ELLIPSIS = 51
+RARROW = 51
-OP = 52
+ELLIPSIS = 52
-ERRORTOKEN = 53
+OP = 53
-N_TOKENS = 54
+ERRORTOKEN = 54
+N_TOKENS = 55
 NT_OFFSET = 256
 #--end constants--
 

Lib/tokenize.py

@@ -91,7 +91,8 @@ EXACT_TOKEN_TYPES = {
     '**=': DOUBLESTAREQUAL,
     '//':  DOUBLESLASH,
     '//=': DOUBLESLASHEQUAL,
-    '@':   AT
+    '@':   AT,
+    '@=':  ATEQUAL,
 }
 
 class TokenInfo(collections.namedtuple('TokenInfo', 'type string start end line')):
@@ -150,7 +151,7 @@ String = group(StringPrefix + r"'[^\n'\\]*(?:\\.[^\n'\\]*)*'",
 # recognized as two instances of =).
 Operator = group(r"\*\*=?", r">>=?", r"<<=?", r"!=",
                  r"//=?", r"->",
-                 r"[+\-*/%&|^=<>]=?",
+                 r"[+\-*/%&@|^=<>]=?",
                  r"~")
 
 Bracket = '[][(){}]'

Misc/NEWS

@@ -10,6 +10,8 @@ Release date: TBA
 Core and Builtins
 -----------------
 
+- PEP 465 and Issue #21176: Add the '@' operator for matrix multiplication.
+
 - Issue #21134: Fix segfault when str is called on an uninitialized
   UnicodeEncodeError, UnicodeDecodeError, or UnicodeTranslateError object.
 

Modules/_operator.c

@@ -69,6 +69,7 @@ spami(truth            , PyObject_IsTrue)
 spam2(op_add           , PyNumber_Add)
 spam2(op_sub           , PyNumber_Subtract)
 spam2(op_mul           , PyNumber_Multiply)
+spam2(op_matmul        , PyNumber_MatrixMultiply)
 spam2(op_floordiv      , PyNumber_FloorDivide)
 spam2(op_truediv       , PyNumber_TrueDivide)
 spam2(op_mod           , PyNumber_Remainder)
@@ -86,6 +87,7 @@ spam2(op_or_           , PyNumber_Or)
 spam2(op_iadd          , PyNumber_InPlaceAdd)
 spam2(op_isub          , PyNumber_InPlaceSubtract)
 spam2(op_imul          , PyNumber_InPlaceMultiply)
+spam2(op_imatmul       , PyNumber_InPlaceMatrixMultiply)
 spam2(op_ifloordiv     , PyNumber_InPlaceFloorDivide)
 spam2(op_itruediv      , PyNumber_InPlaceTrueDivide)
 spam2(op_imod          , PyNumber_InPlaceRemainder)
@@ -343,6 +345,7 @@ spam2o(index, "index(a) -- Same as a.__index__()")
 spam2(add, "add(a, b) -- Same as a + b.")
 spam2(sub, "sub(a, b) -- Same as a - b.")
 spam2(mul, "mul(a, b) -- Same as a * b.")
+spam2(matmul, "matmul(a, b) -- Same as a @ b.")
 spam2(floordiv, "floordiv(a, b) -- Same as a // b.")
 spam2(truediv, "truediv(a, b) -- Same as a / b.")
 spam2(mod, "mod(a, b) -- Same as a % b.")
@@ -360,6 +363,7 @@ spam2(or_, "or_(a, b) -- Same as a | b.")
 spam2(iadd, "a = iadd(a, b) -- Same as a += b.")
 spam2(isub, "a = isub(a, b) -- Same as a -= b.")
 spam2(imul, "a = imul(a, b) -- Same as a *= b.")
+spam2(imatmul, "a = imatmul(a, b) -- Same as a @= b.")
 spam2(ifloordiv, "a = ifloordiv(a, b) -- Same as a //= b.")
 spam2(itruediv, "a = itruediv(a, b) -- Same as a /= b")
 spam2(imod, "a = imod(a, b) -- Same as a %= b.")

Modules/_testcapimodule.c

@@ -3298,6 +3298,109 @@ static PyTypeObject test_structmembersType = {
 };
 
 
+typedef struct {
+    PyObject_HEAD
+} matmulObject;
+
+static PyObject *
+matmulType_matmul(PyObject *self, PyObject *other)
+{
+    return Py_BuildValue("(sOO)", "matmul", self, other);
+}
+
+static PyObject *
+matmulType_imatmul(PyObject *self, PyObject *other)
+{
+    return Py_BuildValue("(sOO)", "imatmul", self, other);
+}
+
+static void
+matmulType_dealloc(PyObject *self)
+{
+    return Py_TYPE(self)->tp_free(self);
+}
+
+static PyNumberMethods matmulType_as_number = {
+    0,                          /* nb_add */
+    0,                          /* nb_subtract */
+    0,                          /* nb_multiply */
+    0,                          /* nb_remainde r*/
+    0,                          /* nb_divmod */
+    0,                          /* nb_power */
+    0,                          /* nb_negative */
+    0,                          /* tp_positive */
+    0,                          /* tp_absolute */
+    0,                          /* tp_bool */
+    0,                          /* nb_invert */
+    0,                          /* nb_lshift */
+    0,                          /* nb_rshift */
+    0,                          /* nb_and */
+    0,                          /* nb_xor */
+    0,                          /* nb_or */
+    0,                          /* nb_int */
+    0,                          /* nb_reserved */
+    0,                          /* nb_float */
+    0,                          /* nb_inplace_add */
+    0,                          /* nb_inplace_subtract */
+    0,                          /* nb_inplace_multiply */
+    0,                          /* nb_inplace_remainder */
+    0,                          /* nb_inplace_power */
+    0,                          /* nb_inplace_lshift */
+    0,                          /* nb_inplace_rshift */
+    0,                          /* nb_inplace_and */
+    0,                          /* nb_inplace_xor */
+    0,                          /* nb_inplace_or */
+    0,                          /* nb_floor_divide */
+    0,                          /* nb_true_divide */
+    0,                          /* nb_inplace_floor_divide */
+    0,                          /* nb_inplace_true_divide */
+    0,                          /* nb_index */
+    matmulType_matmul,        /* nb_matrix_multiply */
+    matmulType_imatmul        /* nb_matrix_inplace_multiply */
+};
+
+static PyTypeObject matmulType = {
+    PyVarObject_HEAD_INIT(NULL, 0)
+    "matmulType",
+    sizeof(matmulObject),               /* tp_basicsize */
+    0,                                  /* tp_itemsize */
+    matmulType_dealloc,                 /* destructor tp_dealloc */
+    0,                                  /* tp_print */
+    0,                                  /* tp_getattr */
+    0,                                  /* tp_setattr */
+    0,                                  /* tp_reserved */
+    0,                                  /* tp_repr */
+    &matmulType_as_number,              /* tp_as_number */
+    0,                                  /* tp_as_sequence */
+    0,                                  /* tp_as_mapping */
+    0,                                  /* tp_hash */
+    0,                                  /* tp_call */
+    0,                                  /* tp_str */
+    PyObject_GenericGetAttr,            /* tp_getattro */
+    PyObject_GenericSetAttr,            /* tp_setattro */
+    0,                                  /* tp_as_buffer */
+    0,                                  /* tp_flags */
+    "C level type with matrix operations defined",
+    0,                                  /* traverseproc tp_traverse */
+    0,                                  /* tp_clear */
+    0,                                  /* tp_richcompare */
+    0,                                  /* tp_weaklistoffset */
+    0,                                  /* tp_iter */
+    0,                                  /* tp_iternext */
+    0,                                  /* tp_methods */
+    0,                                  /* tp_members */
+    0,
+    0,
+    0,
+    0,
+    0,
+    0,
+    0,
+    0,
+    PyType_GenericNew,                  /* tp_new */
+    PyObject_Del,                       /* tp_free */
+};
+
 
 static struct PyModuleDef _testcapimodule = {
     PyModuleDef_HEAD_INIT,
@@ -3327,6 +3430,10 @@ PyInit__testcapi(void)
     /* don't use a name starting with "test", since we don't want
        test_capi to automatically call this */
     PyModule_AddObject(m, "_test_structmembersType", (PyObject *)&test_structmembersType);
+    if (PyType_Ready(&matmulType) < 0)
+        return NULL;
+    Py_INCREF(&matmulType);
+    PyModule_AddObject(m, "matmulType", (PyObject *)&matmulType);
 
     PyModule_AddObject(m, "CHAR_MAX", PyLong_FromLong(CHAR_MAX));
     PyModule_AddObject(m, "CHAR_MIN", PyLong_FromLong(CHAR_MIN));

Objects/abstract.c

@@ -931,6 +931,12 @@ PyNumber_Multiply(PyObject *v, PyObject *w)
     return result;
 }
 
+PyObject *
+PyNumber_MatrixMultiply(PyObject *v, PyObject *w)
+{
+    return binary_op(v, w, NB_SLOT(nb_matrix_multiply), "@");
+}
+
 PyObject *
 PyNumber_FloorDivide(PyObject *v, PyObject *w)
 {
@@ -1012,6 +1018,7 @@ INPLACE_BINOP(PyNumber_InPlaceAnd, nb_inplace_and, nb_and, "&=")
 INPLACE_BINOP(PyNumber_InPlaceLshift, nb_inplace_lshift, nb_lshift, "<<=")
 INPLACE_BINOP(PyNumber_InPlaceRshift, nb_inplace_rshift, nb_rshift, ">>=")
 INPLACE_BINOP(PyNumber_InPlaceSubtract, nb_inplace_subtract, nb_subtract, "-=")
+INPLACE_BINOP(PyNumber_InMatrixMultiply, nb_inplace_matrix_multiply, nb_matrix_multiply, "@=")
 
 PyObject *
 PyNumber_InPlaceFloorDivide(PyObject *v, PyObject *w)
@@ -1077,6 +1084,13 @@ PyNumber_InPlaceMultiply(PyObject *v, PyObject *w)
     return result;
 }
 
+PyObject *
+PyNumber_InPlaceMatrixMultiply(PyObject *v, PyObject *w)
+{
+    return binary_iop(v, w, NB_SLOT(nb_inplace_matrix_multiply),
+                      NB_SLOT(nb_matrix_multiply), "@=");
+}
+
 PyObject *
 PyNumber_InPlaceRemainder(PyObject *v, PyObject *w)
 {

Objects/typeobject.c

@@ -4469,6 +4469,8 @@ inherit_slots(PyTypeObject *type, PyTypeObject *base)
         COPYNUM(nb_inplace_true_divide);
         COPYNUM(nb_inplace_floor_divide);
         COPYNUM(nb_index);
+        COPYNUM(nb_matrix_multiply);
+        COPYNUM(nb_inplace_matrix_multiply);
     }
 
     if (type->tp_as_sequence != NULL && base->tp_as_sequence != NULL) {
@@ -5605,6 +5607,7 @@ slot_mp_ass_subscript(PyObject *self, PyObject *key, PyObject *value)
 SLOT1BIN(slot_nb_add, nb_add, "__add__", "__radd__")
 SLOT1BIN(slot_nb_subtract, nb_subtract, "__sub__", "__rsub__")
 SLOT1BIN(slot_nb_multiply, nb_multiply, "__mul__", "__rmul__")
+SLOT1BIN(slot_nb_matrix_multiply, nb_matrix_multiply, "__matmul__", "__rmatmul__")
 SLOT1BIN(slot_nb_remainder, nb_remainder, "__mod__", "__rmod__")
 SLOT1BIN(slot_nb_divmod, nb_divmod, "__divmod__", "__rdivmod__")
 
@@ -5698,6 +5701,7 @@ SLOT0(slot_nb_float, "__float__")
 SLOT1(slot_nb_inplace_add, "__iadd__", PyObject *, "O")
 SLOT1(slot_nb_inplace_subtract, "__isub__", PyObject *, "O")
 SLOT1(slot_nb_inplace_multiply, "__imul__", PyObject *, "O")
+SLOT1(slot_nb_inplace_matrix_multiply, "__imatmul__", PyObject *, "O")
 SLOT1(slot_nb_inplace_remainder, "__imod__", PyObject *, "O")
 /* Can't use SLOT1 here, because nb_inplace_power is ternary */
 static PyObject *
@@ -6278,6 +6282,12 @@ static slotdef slotdefs[] = {
            "__index__($self, /)\n--\n\n"
            "Return self converted to an integer, if self is suitable"
            "for use as an index into a list."),
+    BINSLOT("__matmul__", nb_matrix_multiply, slot_nb_matrix_multiply,
+            "@"),
+    RBINSLOT("__rmatmul__", nb_matrix_multiply, slot_nb_matrix_multiply,
+             "@"),
+    IBSLOT("__imatmul__", nb_inplace_matrix_multiply, slot_nb_inplace_matrix_multiply,
+           wrap_binaryfunc, "@="),
     MPSLOT("__len__", mp_length, slot_mp_length, wrap_lenfunc,
            "__len__($self, /)\n--\n\nReturn len(self)."),
     MPSLOT("__getitem__", mp_subscript, slot_mp_subscript,

Objects/typeslots.inc

@@ -73,3 +73,5 @@ offsetof(PyHeapTypeObject, ht_type.tp_traverse),
 offsetof(PyHeapTypeObject, ht_type.tp_members),
 offsetof(PyHeapTypeObject, ht_type.tp_getset),
 offsetof(PyHeapTypeObject, ht_type.tp_free),
+offsetof(PyHeapTypeObject, as_number.nb_matrix_multiply),
+offsetof(PyHeapTypeObject, as_number.nb_inplace_matrix_multiply),

Parser/Python.asdl

@@ -91,7 +91,7 @@ module Python
 
     boolop = And | Or 
 
-    operator = Add | Sub | Mult | Div | Mod | Pow | LShift 
+    operator = Add | Sub | Mult | MatMult | Div | Mod | Pow | LShift 
                  | RShift | BitOr | BitXor | BitAnd | FloorDiv
 
     unaryop = Invert | Not | UAdd | USub

Parser/tokenizer.c

@@ -98,6 +98,7 @@ const char *_PyParser_TokenNames[] = {
     "DOUBLESLASH",
     "DOUBLESLASHEQUAL",
     "AT",
+    "ATEQUAL",
     "RARROW",
     "ELLIPSIS",
     /* This table must match the #defines in token.h! */
@@ -1131,7 +1132,7 @@ PyToken_OneChar(int c)
     case '}':           return RBRACE;
     case '^':           return CIRCUMFLEX;
     case '~':           return TILDE;
-    case '@':       return AT;
+    case '@':           return AT;
     default:            return OP;
     }
 }
@@ -1207,6 +1208,11 @@ PyToken_TwoChars(int c1, int c2)
         case '=':               return CIRCUMFLEXEQUAL;
         }
         break;
+    case '@':
+        switch (c2) {
+        case '=':               return ATEQUAL;
+        }
+        break;
     }
     return OP;
 }

Python/Python-ast.c

@@ -349,13 +349,14 @@ static PyTypeObject *And_type;
 static PyTypeObject *Or_type;
 static PyTypeObject *operator_type;
 static PyObject *Add_singleton, *Sub_singleton, *Mult_singleton,
-*Div_singleton, *Mod_singleton, *Pow_singleton, *LShift_singleton,
+*MatMult_singleton, *Div_singleton, *Mod_singleton, *Pow_singleton,
-*RShift_singleton, *BitOr_singleton, *BitXor_singleton, *BitAnd_singleton,
+*LShift_singleton, *RShift_singleton, *BitOr_singleton, *BitXor_singleton,
-*FloorDiv_singleton;
+*BitAnd_singleton, *FloorDiv_singleton;
 static PyObject* ast2obj_operator(operator_ty);
 static PyTypeObject *Add_type;
 static PyTypeObject *Sub_type;
 static PyTypeObject *Mult_type;
+static PyTypeObject *MatMult_type;
 static PyTypeObject *Div_type;
 static PyTypeObject *Mod_type;
 static PyTypeObject *Pow_type;
@@ -970,6 +971,10 @@ static int init_types(void)
     if (!Mult_type) return 0;
     Mult_singleton = PyType_GenericNew(Mult_type, NULL, NULL);
     if (!Mult_singleton) return 0;
+    MatMult_type = make_type("MatMult", operator_type, NULL, 0);
+    if (!MatMult_type) return 0;
+    MatMult_singleton = PyType_GenericNew(MatMult_type, NULL, NULL);
+    if (!MatMult_singleton) return 0;
     Div_type = make_type("Div", operator_type, NULL, 0);
     if (!Div_type) return 0;
     Div_singleton = PyType_GenericNew(Div_type, NULL, NULL);
@@ -3232,6 +3237,9 @@ PyObject* ast2obj_operator(operator_ty o)
         case Mult:
             Py_INCREF(Mult_singleton);
             return Mult_singleton;
+        case MatMult:
+            Py_INCREF(MatMult_singleton);
+            return MatMult_singleton;
         case Div:
             Py_INCREF(Div_singleton);
             return Div_singleton;
@@ -6175,6 +6183,14 @@ obj2ast_operator(PyObject* obj, operator_ty* out, PyArena* arena)
         *out = Mult;
         return 0;
     }
+    isinstance = PyObject_IsInstance(obj, (PyObject *)MatMult_type);
+    if (isinstance == -1) {
+        return 1;
+    }
+    if (isinstance) {
+        *out = MatMult;
+        return 0;
+    }
     isinstance = PyObject_IsInstance(obj, (PyObject *)Div_type);
     if (isinstance == -1) {
         return 1;
@@ -6956,6 +6972,8 @@ PyInit__ast(void)
     if (PyDict_SetItemString(d, "Add", (PyObject*)Add_type) < 0) return NULL;
     if (PyDict_SetItemString(d, "Sub", (PyObject*)Sub_type) < 0) return NULL;
     if (PyDict_SetItemString(d, "Mult", (PyObject*)Mult_type) < 0) return NULL;
+    if (PyDict_SetItemString(d, "MatMult", (PyObject*)MatMult_type) < 0) return
+        NULL;
     if (PyDict_SetItemString(d, "Div", (PyObject*)Div_type) < 0) return NULL;
     if (PyDict_SetItemString(d, "Mod", (PyObject*)Mod_type) < 0) return NULL;
     if (PyDict_SetItemString(d, "Pow", (PyObject*)Pow_type) < 0) return NULL;

Python/ast.c

@@ -825,6 +825,8 @@ get_operator(const node *n)
             return Sub;
         case STAR:
             return Mult;
+        case AT:
+            return MatMult;
         case SLASH:
             return Div;
         case DOUBLESLASH:
@@ -1030,6 +1032,8 @@ ast_for_augassign(struct compiling *c, const node *n)
                 return Pow;
             else
                 return Mult;
+        case '@':
+            return MatMult;
         default:
             PyErr_Format(PyExc_SystemError, "invalid augassign: %s", STR(n));
             return (operator_ty)0;
@@ -2266,7 +2270,7 @@ ast_for_expr(struct compiling *c, const node *n)
        and_expr: shift_expr ('&' shift_expr)*
        shift_expr: arith_expr (('<<'|'>>') arith_expr)*
        arith_expr: term (('+'|'-') term)*
-       term: factor (('*'|'/'|'%'|'//') factor)*
+       term: factor (('*'|'@'|'/'|'%'|'//') factor)*
        factor: ('+'|'-'|'~') factor | power
        power: atom trailer* ('**' factor)*
     */
@@ -2577,7 +2581,7 @@ ast_for_expr_stmt(struct compiling *c, const node *n)
     /* expr_stmt: testlist_star_expr (augassign (yield_expr|testlist)
                 | ('=' (yield_expr|testlist))*)
        testlist_star_expr: (test|star_expr) (',' test|star_expr)* [',']
-       augassign: '+=' | '-=' | '*=' | '/=' | '%=' | '&=' | '|=' | '^='
+       augassign: '+=' | '-=' | '*=' | '@=' | '/=' | '%=' | '&=' | '|=' | '^='
                 | '<<=' | '>>=' | '**=' | '//='
        test: ... here starts the operator precendence dance
      */

Python/ceval.c

@@ -1495,6 +1495,18 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             DISPATCH();
         }
 
+        TARGET(BINARY_MATRIX_MULTIPLY) {
+            PyObject *right = POP();
+            PyObject *left = TOP();
+            PyObject *res = PyNumber_MatrixMultiply(left, right);
+            Py_DECREF(left);
+            Py_DECREF(right);
+            SET_TOP(res);
+            if (res == NULL)
+                goto error;
+            DISPATCH();
+        }
+
         TARGET(BINARY_TRUE_DIVIDE) {
             PyObject *divisor = POP();
             PyObject *dividend = TOP();
@@ -1685,6 +1697,18 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             DISPATCH();
         }
 
+        TARGET(INPLACE_MATRIX_MULTIPLY) {
+            PyObject *right = POP();
+            PyObject *left = TOP();
+            PyObject *res = PyNumber_InPlaceMatrixMultiply(left, right);
+            Py_DECREF(left);
+            Py_DECREF(right);
+            SET_TOP(res);
+            if (res == NULL)
+                goto error;
+            DISPATCH();
+        }
+
         TARGET(INPLACE_TRUE_DIVIDE) {
             PyObject *divisor = POP();
             PyObject *dividend = TOP();

Python/compile.c

@@ -881,6 +881,7 @@ PyCompile_OpcodeStackEffect(int opcode, int oparg)
 
         case BINARY_POWER:
         case BINARY_MULTIPLY:
+        case BINARY_MATRIX_MULTIPLY:
         case BINARY_MODULO:
         case BINARY_ADD:
         case BINARY_SUBTRACT:
@@ -895,6 +896,7 @@ PyCompile_OpcodeStackEffect(int opcode, int oparg)
         case INPLACE_ADD:
         case INPLACE_SUBTRACT:
         case INPLACE_MULTIPLY:
+        case INPLACE_MATRIX_MULTIPLY:
         case INPLACE_MODULO:
             return -1;
         case STORE_SUBSCR:
@@ -2625,6 +2627,8 @@ binop(struct compiler *c, operator_ty op)
         return BINARY_SUBTRACT;
     case Mult:
         return BINARY_MULTIPLY;
+    case MatMult:
+        return BINARY_MATRIX_MULTIPLY;
     case Div:
         return BINARY_TRUE_DIVIDE;
     case Mod:
@@ -2689,6 +2693,8 @@ inplace_binop(struct compiler *c, operator_ty op)
         return INPLACE_SUBTRACT;
     case Mult:
         return INPLACE_MULTIPLY;
+    case MatMult:
+        return INPLACE_MATRIX_MULTIPLY;
     case Div:
         return INPLACE_TRUE_DIVIDE;
     case Mod:

Python/opcode_targets.h

@@ -15,8 +15,8 @@ static void *opcode_targets[256] = {
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&TARGET_UNARY_INVERT,
-    &&_unknown_opcode,
+    &&TARGET_BINARY_MATRIX_MULTIPLY,
-    &&_unknown_opcode,
+    &&TARGET_INPLACE_MATRIX_MULTIPLY,
     &&_unknown_opcode,
     &&TARGET_BINARY_POWER,
     &&TARGET_BINARY_MULTIPLY,