[mypyc] feat: further optimize equality check with string literals [1/1] (#19883)

This PR further optimizes string equality checks against literals by
getting rid of the PyUnicode_GET_LENGTH call against the literal value,
which is not necessary since the value is known at compile-time

I think this optimization will be helpful in cases where the non-literal
string DOES match but is actually a subtype of string (actual strings
instances that match would be caught by the identity check), or in cases
where an exact string does NOT match.

Author: BobTheBuidler

mypyc/irbuild/ll_builder.py

@@ -8,7 +8,8 @@
 
 import sys
 from collections.abc import Sequence
-from typing import Callable, Final, Optional
+from typing import Callable, Final, Optional, cast
+from typing_extensions import TypeGuard
 
 from mypy.argmap import map_actuals_to_formals
 from mypy.nodes import ARG_POS, ARG_STAR, ARG_STAR2, ArgKind
@@ -185,6 +186,7 @@
 from mypyc.primitives.str_ops import (
     str_check_if_true,
     str_eq,
+    str_eq_literal,
     str_ssize_t_size_op,
     unicode_compare,
 )
@@ -1551,9 +1553,33 @@ def check_tagged_short_int(self, val: Value, line: int, negated: bool = False) -
     def compare_strings(self, lhs: Value, rhs: Value, op: str, line: int) -> Value:
         """Compare two strings"""
         if op == "==":
+            # We can specialize this case if one or both values are string literals
+            literal_fastpath = False
+
+            def is_string_literal(value: Value) -> TypeGuard[LoadLiteral]:
+                return isinstance(value, LoadLiteral) and is_str_rprimitive(value.type)
+
+            if is_string_literal(lhs):
+                if is_string_literal(rhs):
+                    # we can optimize out the check entirely in some constant-folded cases
+                    return self.true() if lhs.value == rhs.value else self.false()
+
+                # if lhs argument is string literal, switch sides to match specializer C api
+                lhs, rhs = rhs, lhs
+                literal_fastpath = True
+            elif is_string_literal(rhs):
+                literal_fastpath = True
+
+            if literal_fastpath:
+                literal_string = cast(str, cast(LoadLiteral, rhs).value)
+                literal_length = Integer(len(literal_string), c_pyssize_t_rprimitive, line)
+                return self.primitive_op(str_eq_literal, [lhs, rhs, literal_length], line)
+
             return self.primitive_op(str_eq, [lhs, rhs], line)
+
         elif op == "!=":
-            eq = self.primitive_op(str_eq, [lhs, rhs], line)
+            # perform a standard equality check, then negate
+            eq = self.compare_strings(lhs, rhs, "==", line)
             return self.add(ComparisonOp(eq, self.false(), ComparisonOp.EQ, line))
 
         # TODO: modify 'str' to use same interface as 'compare_bytes' as it would avoid

mypyc/lib-rt/CPy.h

@@ -735,6 +735,7 @@ static inline char CPyDict_CheckSize(PyObject *dict, Py_ssize_t size) {
 #define BOTHSTRIP  2
 
 char CPyStr_Equal(PyObject *str1, PyObject *str2);
+char CPyStr_EqualLiteral(PyObject *str, PyObject *literal_str, Py_ssize_t literal_length);
 PyObject *CPyStr_Build(Py_ssize_t len, ...);
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index);
 PyObject *CPyStr_GetItemUnsafe(PyObject *str, Py_ssize_t index);

mypyc/lib-rt/str_ops.c

@@ -64,20 +64,33 @@ make_bloom_mask(int kind, const void* ptr, Py_ssize_t len)
 #undef BLOOM_UPDATE
 }
 
-// Adapted from CPython 3.13.1 (_PyUnicode_Equal)
-char CPyStr_Equal(PyObject *str1, PyObject *str2) {
-    if (str1 == str2) {
-        return 1;
-    }
-    Py_ssize_t len = PyUnicode_GET_LENGTH(str1);
-    if (PyUnicode_GET_LENGTH(str2) != len)
+static inline char _CPyStr_Equal_NoIdentCheck(PyObject *str1, PyObject *str2, Py_ssize_t str2_length) {
+    // This helper function only exists to deduplicate code in CPyStr_Equal and CPyStr_EqualLiteral
+    Py_ssize_t str1_length = PyUnicode_GET_LENGTH(str1);
+    if (str1_length != str2_length)
         return 0;
     int kind = PyUnicode_KIND(str1);
     if (PyUnicode_KIND(str2) != kind)
         return 0;
     const void *data1 = PyUnicode_DATA(str1);
     const void *data2 = PyUnicode_DATA(str2);
-    return memcmp(data1, data2, len * kind) == 0;
+    return memcmp(data1, data2, str1_length * kind) == 0;
+}
+
+// Adapted from CPython 3.13.1 (_PyUnicode_Equal)
+char CPyStr_Equal(PyObject *str1, PyObject *str2) {
+    if (str1 == str2) {
+        return 1;
+    }
+    Py_ssize_t str2_length = PyUnicode_GET_LENGTH(str2);
+    return _CPyStr_Equal_NoIdentCheck(str1, str2, str2_length);
+}
+
+char CPyStr_EqualLiteral(PyObject *str, PyObject *literal_str, Py_ssize_t literal_length) {
+    if (str == literal_str) {
+        return 1;
+    }
+    return _CPyStr_Equal_NoIdentCheck(str, literal_str, literal_length);
 }
 
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index) {

mypyc/primitives/str_ops.py

@@ -88,6 +88,14 @@
     error_kind=ERR_NEVER,
 )
 
+str_eq_literal = custom_primitive_op(
+    name="str_eq_literal",
+    c_function_name="CPyStr_EqualLiteral",
+    arg_types=[str_rprimitive, str_rprimitive, c_pyssize_t_rprimitive],
+    return_type=bool_rprimitive,
+    error_kind=ERR_NEVER,
+)
+
 unicode_compare = custom_op(
     arg_types=[str_rprimitive, str_rprimitive],
     return_type=c_int_rprimitive,

mypyc/test-data/irbuild-classes.test

@@ -2325,15 +2325,15 @@ def SetAttr.__setattr__(self, key, val):
     r12 :: bit
 L0:
     r0 = 'regular_attr'
-    r1 = CPyStr_Equal(key, r0)
+    r1 = CPyStr_EqualLiteral(key, r0, 12)
     if r1 goto L1 else goto L2 :: bool
 L1:
     r2 = unbox(int, val)
     self.regular_attr = r2; r3 = is_error
     goto L6
 L2:
     r4 = 'class_var'
-    r5 = CPyStr_Equal(key, r4)
+    r5 = CPyStr_EqualLiteral(key, r4, 9)
     if r5 goto L3 else goto L4 :: bool
 L3:
     r6 = builtins :: module
@@ -2468,15 +2468,15 @@ def SetAttr.__setattr__(self, key, val):
     r12 :: bit
 L0:
     r0 = 'regular_attr'
-    r1 = CPyStr_Equal(key, r0)
+    r1 = CPyStr_EqualLiteral(key, r0, 12)
     if r1 goto L1 else goto L2 :: bool
 L1:
     r2 = unbox(int, val)
     self.regular_attr = r2; r3 = is_error
     goto L6
 L2:
     r4 = 'class_var'
-    r5 = CPyStr_Equal(key, r4)
+    r5 = CPyStr_EqualLiteral(key, r4, 9)
     if r5 goto L3 else goto L4 :: bool
 L3:
     r6 = builtins :: module

mypyc/test-data/irbuild-dict.test

@@ -410,7 +410,7 @@ L2:
     k = r8
     v = r7
     r9 = 'name'
-    r10 = CPyStr_Equal(k, r9)
+    r10 = CPyStr_EqualLiteral(k, r9, 4)
     if r10 goto L3 else goto L4 :: bool
 L3:
     name = v

mypyc/test-data/irbuild-str.test

@@ -740,3 +740,34 @@ L2:
 L3:
     keep_alive x
     return r2
+
+[case testStrEqLiteral]
+from typing import Final
+literal: Final = "literal"
+def literal_rhs(x: str) -> bool:
+    return x == literal
+def literal_lhs(x: str) -> bool:
+    return literal == x
+def literal_both() -> bool:
+    return literal == "literal"
+[out]
+def literal_rhs(x):
+    x, r0 :: str
+    r1 :: bool
+L0:
+    r0 = 'literal'
+    r1 = CPyStr_EqualLiteral(x, r0, 7)
+    return r1
+def literal_lhs(x):
+    x, r0 :: str
+    r1 :: bool
+L0:
+    r0 = 'literal'
+    r1 = CPyStr_EqualLiteral(x, r0, 7)
+    return r1
+def literal_both():
+    r0, r1 :: str
+L0:
+    r0 = 'literal'
+    r1 = 'literal'
+    return 1

mypyc/test-data/irbuild-unreachable.test

@@ -20,7 +20,7 @@ L0:
     r2 = CPyObject_GetAttr(r0, r1)
     r3 = cast(str, r2)
     r4 = 'x'
-    r5 = CPyStr_Equal(r3, r4)
+    r5 = CPyStr_EqualLiteral(r3, r4, 1)
     if r5 goto L2 else goto L1 :: bool
 L1:
     r6 = r5
@@ -54,7 +54,7 @@ L0:
     r2 = CPyObject_GetAttr(r0, r1)
     r3 = cast(str, r2)
     r4 = 'x'
-    r5 = CPyStr_Equal(r3, r4)
+    r5 = CPyStr_EqualLiteral(r3, r4, 1)
     if r5 goto L2 else goto L1 :: bool
 L1:
     r6 = r5