[mypyc] Add faster primitive for string equality

This speeds up self check by ~1.4%. String equality is one of the top five most common primitive function calls in self check. We previously used a string comparison primitive that calculated the relative order of two strings. Usually we only care about equality, which we can do quicker since we can fast path using a length check, for example. I checked the CPython implementation of string equality in 3.9 (lowest supported Python version) and 3.13, and both of them had a fast path based on string object kind, and equality checks overall have the same semantics.
python · JukkaL · Jul 9, 2025 · Jul 7, 2025 · Jul 8, 2025 · Jul 8, 2025
commit 914757b5396b6a4e3faa5b631e9238c57fc2df7d
diff --git a/mypyc/irbuild/ll_builder.py b/mypyc/irbuild/ll_builder.py
@@ -175,7 +175,12 @@
     unary_ops,
 )
 from mypyc.primitives.set_ops import new_set_op
-from mypyc.primitives.str_ops import str_check_if_true, str_ssize_t_size_op, unicode_compare
+from mypyc.primitives.str_ops import (
+    str_check_if_true,
+    str_eq,
+    str_ssize_t_size_op,
+    unicode_compare,
+)
 from mypyc.primitives.tuple_ops import list_tuple_op, new_tuple_op, new_tuple_with_length_op
 from mypyc.rt_subtype import is_runtime_subtype
 from mypyc.sametype import is_same_type
@@ -1471,6 +1476,11 @@ 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 == "==":
+            return self.primitive_op(str_eq, [lhs, rhs], line)
+        elif op == "!=":
+            eq = self.primitive_op(str_eq, [lhs, rhs], line)
+            return self.add(ComparisonOp(eq, self.false(), ComparisonOp.EQ, line))
         compare_result = self.call_c(unicode_compare, [lhs, rhs], line)
         error_constant = Integer(-1, c_int_rprimitive, line)
         compare_error_check = self.add(

diff --git a/mypyc/lib-rt/CPy.h b/mypyc/lib-rt/CPy.h
@@ -726,6 +726,7 @@ static inline char CPyDict_CheckSize(PyObject *dict, CPyTagged size) {
 #define RIGHTSTRIP 1
 #define BOTHSTRIP  2
 
+char CPyStr_Equal(PyObject *str1, PyObject *str2);
 PyObject *CPyStr_Build(Py_ssize_t len, ...);
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index);
 CPyTagged CPyStr_Find(PyObject *str, PyObject *substr, CPyTagged start, int direction);

diff --git a/mypyc/lib-rt/str_ops.c b/mypyc/lib-rt/str_ops.c
@@ -64,6 +64,22 @@ 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)
+        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;
+}
+
 PyObject *CPyStr_GetItem(PyObject *str, CPyTagged index) {
     if (PyUnicode_READY(str) != -1) {
         if (CPyTagged_CheckShort(index)) {

diff --git a/mypyc/primitives/str_ops.py b/mypyc/primitives/str_ops.py
@@ -21,6 +21,7 @@
     ERR_NEG_INT,
     binary_op,
     custom_op,
+    custom_primitive_op,
     function_op,
     load_address_op,
     method_op,
@@ -69,6 +70,15 @@
     steals=[True, False],
 )
 
+# str1 == str2 (very common operation, so we provide our own)
+str_eq = custom_primitive_op(
+    name="str_eq",
+    c_function_name="CPyStr_Equal",
+    arg_types=[str_rprimitive, str_rprimitive],
+    return_type=bool_rprimitive,
+    error_kind=ERR_NEVER,
+)
+
 unicode_compare = custom_op(
     arg_types=[str_rprimitive, str_rprimitive],
     return_type=c_int_rprimitive,