bpo-32677: Optimize str.isascii() (GH-5356)

methane · web-flow · commit bea57060c863 · 2018-01-28T09:59:12.000+09:00
diff --git a/Lib/test/string_tests.py b/Lib/test/string_tests.py
@@ -916,6 +916,13 @@ def test_isascii(self):
         self.checkequal(True, '\x00\x7f', 'isascii')
         self.checkequal(False, '\x80', 'isascii')
         self.checkequal(False, '\xe9', 'isascii')
+        # bytes.isascii() and bytearray.isascii() has optimization which
+        # check 4 or 8 bytes at once.  So check some alignments.
+        for p in range(8):
+            self.checkequal(True, ' '*p + '\x7f', 'isascii')
+            self.checkequal(False, ' '*p + '\x80', 'isascii')
+            self.checkequal(True, ' '*p + '\x7f' + ' '*8, 'isascii')
+            self.checkequal(False, ' '*p + '\x80' + ' '*8, 'isascii')
 
     def test_isdigit(self):
         self.checkequal(False, '', 'isdigit')
diff --git a/Objects/bytes_methods.c b/Objects/bytes_methods.c
@@ -98,19 +98,51 @@ PyDoc_STRVAR_shared(_Py_isascii__doc__,
 Return True if B is empty or all characters in B are ASCII,\n\
 False otherwise.");
 
+// Optimization is copied from ascii_decode in unicodeobject.c
+/* Mask to quickly check whether a C 'long' contains a
+   non-ASCII, UTF8-encoded char. */
+#if (SIZEOF_LONG == 8)
+# define ASCII_CHAR_MASK 0x8080808080808080UL
+#elif (SIZEOF_LONG == 4)
+# define ASCII_CHAR_MASK 0x80808080UL
+#else
+# error C 'long' size should be either 4 or 8!
+#endif
+
 PyObject*
 _Py_bytes_isascii(const char *cptr, Py_ssize_t len)
 {
-    const unsigned char *p = (unsigned char *) cptr;
-    const unsigned char *e = p + len;
-    for (; p < e; p++) {
-        if (*p >= 128) {
+    const char *p = cptr;
+    const char *end = p + len;
+    const char *aligned_end = (const char *) _Py_ALIGN_DOWN(end, SIZEOF_LONG);
+
+    while (p < end) {
+        /* Fast path, see in STRINGLIB(utf8_decode) in stringlib/codecs.h
+           for an explanation. */
+        if (_Py_IS_ALIGNED(p, SIZEOF_LONG)) {
+            /* Help allocation */
+            const char *_p = p;
+            while (_p < aligned_end) {
+                unsigned long value = *(unsigned long *) _p;
+                if (value & ASCII_CHAR_MASK) {
+                    Py_RETURN_FALSE;
+                }
+                _p += SIZEOF_LONG;
+            }
+            p = _p;
+            if (_p == end)
+                break;
+        }
+        if ((unsigned char)*p & 0x80) {
             Py_RETURN_FALSE;
         }
+        p++;
     }
     Py_RETURN_TRUE;
 }
 
+#undef ASCII_CHAR_MASK
+
 
 PyDoc_STRVAR_shared(_Py_isdigit__doc__,
 "B.isdigit() -> bool\n\
