[mypyc] Support the i16 native integer type (#15464)

The `i16` type behaves the same as `i64` and `i32`, but is obviously
smaller. The PR is big, but it's mostly because of test cases, which are
adapted from `i32` test cases.

Also fix error handling in unboxing of native int and float types (i.e.
all types with overlapping error values).

Author: JukkaL

mypy/types.py

@@ -150,7 +150,11 @@
 )
 
 # Mypyc fixed-width native int types (compatible with builtins.int)
-MYPYC_NATIVE_INT_NAMES: Final = ("mypy_extensions.i64", "mypy_extensions.i32")
+MYPYC_NATIVE_INT_NAMES: Final = (
+    "mypy_extensions.i64",
+    "mypy_extensions.i32",
+    "mypy_extensions.i16",
+)
 
 DATACLASS_TRANSFORM_NAMES: Final = (
     "typing.dataclass_transform",

mypyc/codegen/emit.py

@@ -35,6 +35,7 @@
     is_dict_rprimitive,
     is_fixed_width_rtype,
     is_float_rprimitive,
+    is_int16_rprimitive,
     is_int32_rprimitive,
     is_int64_rprimitive,
     is_int_rprimitive,
@@ -900,28 +901,35 @@ def emit_unbox(
             self.emit_line(f"    {dest} = 1;")
         elif is_int64_rprimitive(typ):
             # Whether we are borrowing or not makes no difference.
+            assert not optional  # Not supported for overlapping error values
             if declare_dest:
                 self.emit_line(f"int64_t {dest};")
             self.emit_line(f"{dest} = CPyLong_AsInt64({src});")
-            # TODO: Handle 'optional'
-            # TODO: Handle 'failure'
+            if not isinstance(error, AssignHandler):
+                self.emit_unbox_failure_with_overlapping_error_value(dest, typ, failure)
         elif is_int32_rprimitive(typ):
             # Whether we are borrowing or not makes no difference.
+            assert not optional  # Not supported for overlapping error values
             if declare_dest:
                 self.emit_line(f"int32_t {dest};")
             self.emit_line(f"{dest} = CPyLong_AsInt32({src});")
-            # TODO: Handle 'optional'
-            # TODO: Handle 'failure'
+            if not isinstance(error, AssignHandler):
+                self.emit_unbox_failure_with_overlapping_error_value(dest, typ, failure)
+        elif is_int16_rprimitive(typ):
+            # Whether we are borrowing or not makes no difference.
+            assert not optional  # Not supported for overlapping error values
+            if declare_dest:
+                self.emit_line(f"int16_t {dest};")
+            self.emit_line(f"{dest} = CPyLong_AsInt16({src});")
+            if not isinstance(error, AssignHandler):
+                self.emit_unbox_failure_with_overlapping_error_value(dest, typ, failure)
         elif is_float_rprimitive(typ):
+            assert not optional  # Not supported for overlapping error values
             if declare_dest:
                 self.emit_line("double {};".format(dest))
             # TODO: Don't use __float__ and __index__
             self.emit_line(f"{dest} = PyFloat_AsDouble({src});")
-            self.emit_lines(
-                f"if ({dest} == -1.0 && PyErr_Occurred()) {{", f"{dest} = -113.0;", "}"
-            )
-            # TODO: Handle 'optional'
-            # TODO: Handle 'failure'
+            self.emit_lines(f"if ({dest} == -1.0 && PyErr_Occurred()) {{", failure, "}")
         elif isinstance(typ, RTuple):
             self.declare_tuple_struct(typ)
             if declare_dest:
@@ -1006,7 +1014,7 @@ def emit_box(
             self.emit_lines(f"{declaration}{dest} = Py_None;")
             if not can_borrow:
                 self.emit_inc_ref(dest, object_rprimitive)
-        elif is_int32_rprimitive(typ):
+        elif is_int32_rprimitive(typ) or is_int16_rprimitive(typ):
             self.emit_line(f"{declaration}{dest} = PyLong_FromLong({src});")
         elif is_int64_rprimitive(typ):
             self.emit_line(f"{declaration}{dest} = PyLong_FromLongLong({src});")
@@ -1137,6 +1145,13 @@ def _emit_traceback(
         if DEBUG_ERRORS:
             self.emit_line('assert(PyErr_Occurred() != NULL && "failure w/o err!");')
 
+    def emit_unbox_failure_with_overlapping_error_value(
+        self, dest: str, typ: RType, failure: str
+    ) -> None:
+        self.emit_line(f"if ({dest} == {self.c_error_value(typ)} && PyErr_Occurred()) {{")
+        self.emit_line(failure)
+        self.emit_line("}")
+
 
 def c_array_initializer(components: list[str], *, indented: bool = False) -> str:
     """Construct an initializer for a C array variable.

mypyc/doc/float_operations.rst

@@ -14,6 +14,7 @@ Construction
 * ``float(x: int)``
 * ``float(x: i64)``
 * ``float(x: i32)``
+* ``float(x: i16)``
 * ``float(x: str)``
 * ``float(x: float)`` (no-op)
 
@@ -28,8 +29,9 @@ Functions
 ---------
 
 * ``int(f)``
-* ``i32(f)`` (convert to ``i32``)
-* ``i64(f)`` (convert to ``i64``)
+* ``i64(f)`` (convert to 64-bit signed integer)
+* ``i32(f)`` (convert to 32-bit signed integer)
+* ``i16(f)`` (convert to 16-bit signed integer)
 * ``abs(f)``
 * ``math.sin(f)``
 * ``math.cos(f)``

mypyc/doc/int_operations.rst

@@ -8,14 +8,16 @@ Mypyc supports these integer types:
 * ``int`` (arbitrary-precision integer)
 * ``i64`` (64-bit signed integer)
 * ``i32`` (32-bit signed integer)
+* ``i16`` (16-bit signed integer)
 
-``i64`` and ``i32`` are *native integer types* and must be imported
+``i64``, ``i32`` and ``i16`` are *native integer types* and must be imported
 from the ``mypy_extensions`` module. ``int`` corresponds to the Python
 ``int`` type, but uses a more efficient runtime representation (tagged
-pointer). Native integer types are value types. All integer types have
-optimized primitive operations, but the native integer types are more
-efficient than ``int``, since they don't require range or bounds
-checks.
+pointer). Native integer types are value types.
+
+All integer types have optimized primitive operations, but the native
+integer types are more efficient than ``int``, since they don't
+require range or bounds checks.
 
 Operations on integers that are listed here have fast, optimized
 implementations. Other integer operations use generic implementations
@@ -31,6 +33,7 @@ Construction
 * ``int(x: float)``
 * ``int(x: i64)``
 * ``int(x: i32)``
+* ``int(x: i16)``
 * ``int(x: str)``
 * ``int(x: str, base: int)``
 * ``int(x: int)`` (no-op)
@@ -40,6 +43,7 @@ Construction
 * ``i64(x: int)``
 * ``i64(x: float)``
 * ``i64(x: i32)``
+* ``i64(x: i16)``
 * ``i64(x: str)``
 * ``i64(x: str, base: int)``
 * ``i64(x: i64)`` (no-op)
@@ -49,10 +53,21 @@ Construction
 * ``i32(x: int)``
 * ``i32(x: float)``
 * ``i32(x: i64)`` (truncate)
+* ``i32(x: i16)``
 * ``i32(x: str)``
 * ``i32(x: str, base: int)``
 * ``i32(x: i32)`` (no-op)
 
+``i16`` type:
+
+* ``i16(x: int)``
+* ``i16(x: float)``
+* ``i16(x: i64)`` (truncate)
+* ``i16(x: i32)`` (truncate)
+* ``i16(x: str)``
+* ``i16(x: str, base: int)``
+* ``i16(x: i16)`` (no-op)
+
 Conversions from ``int`` to a native integer type raise
 ``OverflowError`` if the value is too large or small. Conversions from
 a wider native integer type to a narrower one truncate the value and never

mypyc/doc/using_type_annotations.rst

@@ -32,6 +32,7 @@ implementations:
 * ``int`` (:ref:`native operations <int-ops>`)
 * ``i64`` (:ref:`documentation <native-ints>`, :ref:`native operations <int-ops>`)
 * ``i32`` (:ref:`documentation <native-ints>`, :ref:`native operations <int-ops>`)
+* ``i16`` (:ref:`documentation <native-ints>`, :ref:`native operations <int-ops>`)
 * ``float`` (:ref:`native operations <float-ops>`)
 * ``bool`` (:ref:`native operations <bool-ops>`)
 * ``str`` (:ref:`native operations <str-ops>`)
@@ -342,13 +343,14 @@ Examples::
 Native integer types
 --------------------
 
-You can use the native integer types ``i64`` (64-bit signed integer)
-and ``i32`` (32-bit signed integer) if you know that integer values
-will always fit within fixed bounds. These types are faster than the
-arbitrary-precision ``int`` type, since they don't require overflow
-checks on operations. ``i32`` may also use less memory than ``int``
-values. The types are imported from the ``mypy_extensions`` module
-(installed via ``pip install mypy_extensions``).
+You can use the native integer types ``i64`` (64-bit signed integer),
+``i32`` (32-bit signed integer), and ``i16`` (16-bit signed integer)
+if you know that integer values will always fit within fixed
+bounds. These types are faster than the arbitrary-precision ``int``
+type, since they don't require overflow checks on operations. ``i32``
+and ``i16`` may also use less memory than ``int`` values. The types
+are imported from the ``mypy_extensions`` module (installed via ``pip
+install mypy_extensions``).
 
 Example::
 

mypyc/ir/rtypes.py

@@ -206,7 +206,7 @@ def __init__(
         self.error_overlap = error_overlap
         if ctype == "CPyTagged":
             self.c_undefined = "CPY_INT_TAG"
-        elif ctype in ("int32_t", "int64_t"):
+        elif ctype in ("int16_t", "int32_t", "int64_t"):
             # This is basically an arbitrary value that is pretty
             # unlikely to overlap with a real value.
             self.c_undefined = "-113"
@@ -290,6 +290,16 @@ def __hash__(self) -> int:
 
 # Low level integer types (correspond to C integer types)
 
+int16_rprimitive: Final = RPrimitive(
+    "int16",
+    is_unboxed=True,
+    is_refcounted=False,
+    is_native_int=True,
+    is_signed=True,
+    ctype="int16_t",
+    size=2,
+    error_overlap=True,
+)
 int32_rprimitive: Final = RPrimitive(
     "int32",
     is_unboxed=True,
@@ -432,6 +442,10 @@ def is_short_int_rprimitive(rtype: RType) -> bool:
     return rtype is short_int_rprimitive
 
 
+def is_int16_rprimitive(rtype: RType) -> bool:
+    return rtype is int16_rprimitive
+
+
 def is_int32_rprimitive(rtype: RType) -> bool:
     return rtype is int32_rprimitive or (
         rtype is c_pyssize_t_rprimitive and rtype._ctype == "int32_t"
@@ -445,7 +459,7 @@ def is_int64_rprimitive(rtype: RType) -> bool:
 
 
 def is_fixed_width_rtype(rtype: RType) -> bool:
-    return is_int32_rprimitive(rtype) or is_int64_rprimitive(rtype)
+    return is_int64_rprimitive(rtype) or is_int32_rprimitive(rtype) or is_int16_rprimitive(rtype)
 
 
 def is_uint32_rprimitive(rtype: RType) -> bool:
@@ -536,6 +550,8 @@ def visit_rprimitive(self, t: RPrimitive) -> str:
             return "8"  # "8 byte integer"
         elif t._ctype == "int32_t":
             return "4"  # "4 byte integer"
+        elif t._ctype == "int16_t":
+            return "2"  # "2 byte integer"
         elif t._ctype == "double":
             return "F"
         assert not t.is_unboxed, f"{t} unexpected unboxed type"

mypyc/irbuild/ll_builder.py

@@ -104,6 +104,7 @@
     is_dict_rprimitive,
     is_fixed_width_rtype,
     is_float_rprimitive,
+    is_int16_rprimitive,
     is_int32_rprimitive,
     is_int64_rprimitive,
     is_int_rprimitive,
@@ -146,6 +147,9 @@
     py_vectorcall_op,
 )
 from mypyc.primitives.int_ops import (
+    int16_divide_op,
+    int16_mod_op,
+    int16_overflow,
     int32_divide_op,
     int32_mod_op,
     int32_overflow,
@@ -456,6 +460,10 @@ def coerce_int_to_fixed_width(self, src: Value, target_type: RType, line: int) -
             # Slow path just always generates an OverflowError
             self.call_c(int32_overflow, [], line)
             self.add(Unreachable())
+        elif is_int16_rprimitive(target_type):
+            # Slow path just always generates an OverflowError
+            self.call_c(int16_overflow, [], line)
+            self.add(Unreachable())
         else:
             assert False, target_type
 
@@ -469,7 +477,7 @@ def coerce_short_int_to_fixed_width(self, src: Value, target_type: RType, line:
         assert False, (src.type, target_type)
 
     def coerce_fixed_width_to_int(self, src: Value, line: int) -> Value:
-        if is_int32_rprimitive(src.type) and PLATFORM_SIZE == 8:
+        if (is_int32_rprimitive(src.type) and PLATFORM_SIZE == 8) or is_int16_rprimitive(src.type):
             # Simple case -- just sign extend and shift.
             extended = self.add(Extend(src, c_pyssize_t_rprimitive, signed=True))
             return self.int_op(
@@ -2038,6 +2046,8 @@ def fixed_width_int_op(self, type: RType, lhs: Value, rhs: Value, op: int, line:
                 prim = int64_divide_op
             elif is_int32_rprimitive(type):
                 prim = int32_divide_op
+            elif is_int16_rprimitive(type):
+                prim = int16_divide_op
             else:
                 assert False, type
             return self.call_c(prim, [lhs, rhs], line)
@@ -2050,6 +2060,8 @@ def fixed_width_int_op(self, type: RType, lhs: Value, rhs: Value, op: int, line:
                 prim = int64_mod_op
             elif is_int32_rprimitive(type):
                 prim = int32_mod_op
+            elif is_int16_rprimitive(type):
+                prim = int16_mod_op
             else:
                 assert False, type
             return self.call_c(prim, [lhs, rhs], line)

mypyc/irbuild/mapper.py

@@ -32,6 +32,7 @@
     bytes_rprimitive,
     dict_rprimitive,
     float_rprimitive,
+    int16_rprimitive,
     int32_rprimitive,
     int64_rprimitive,
     int_rprimitive,
@@ -102,6 +103,8 @@ def type_to_rtype(self, typ: Type | None) -> RType:
                 return int64_rprimitive
             elif typ.type.fullname == "mypy_extensions.i32":
                 return int32_rprimitive
+            elif typ.type.fullname == "mypy_extensions.i16":
+                return int16_rprimitive
             else:
                 return object_rprimitive
         elif isinstance(typ, TupleType):