Faster ascii decoding

Objects/unicodeobject.c

--- a/Objects/unicodeobject.c	Thu Mar 22 22:40:44 2012 -0400
+++ b/Objects/unicodeobject.c	Tue Mar 27 01:49:43 2012 +0300
@@ -6874,7 +6874,6 @@
     Py_ssize_t endinpos;
     Py_ssize_t outpos;
     const char *e;
-    int has_error;
     const unsigned char *p = (const unsigned char *)s;
     const unsigned char *end = p + size;
     const unsigned char *aligned_end = (const unsigned char *) ((size_t) end & ~LONG_PTR_MASK);
@@ -6890,8 +6889,38 @@
     if (size == 1 && (unsigned char)s[0] < 128)
         return get_latin1_char((unsigned char)s[0]);
 
-    has_error = 0;
-    while (p < end && !has_error) {
+    v = PyUnicode_New(size, 127);
+    if (v == NULL)
+        goto onError;
+
+#if SIZEOF_LONG <= SIZEOF_VOID_P
+    if (!((size_t) p & LONG_PTR_MASK)) {
+        /* Fast path, see below in PyUnicode_DecodeUTF8Stateful for
+           an explanation. */
+        /* Help register allocation */
+        register const unsigned char *_p = p;
+        register unsigned char * q = PyUnicode_1BYTE_DATA(v);
+        while (_p < aligned_end) {
+            unsigned long value = *(const unsigned long *) _p;
+            if (value & ASCII_CHAR_MASK)
+                goto hasError;
+            *((unsigned long *)q) = value;
+            _p += SIZEOF_LONG;
+            q += SIZEOF_LONG;
+        }
+        p = _p;
+        while (p < end) {
+            if (*p & 0x80)
+                goto hasError;
+            *q = *p;
+            ++p;
+            ++q;
+        }
+        return v;
+    }
+#endif
+
+    while (p < end) {
         /* Fast path, see below in PyUnicode_DecodeUTF8Stateful for
            an explanation. */
         if (!((size_t) p & LONG_PTR_MASK)) {
@@ -6899,36 +6928,24 @@
             register const unsigned char *_p = p;
             while (_p < aligned_end) {
                 unsigned long value = *(unsigned long *) _p;
-                if (value & ASCII_CHAR_MASK) {
-                    has_error = 1;
-                    break;
-                }
+                if (value & ASCII_CHAR_MASK)
+                    goto hasError;
                 _p += SIZEOF_LONG;
             }
             if (_p == end)
                 break;
-            if (has_error)
-                break;
             p = _p;
         }
-        if (*p & 0x80) {
-            has_error = 1;
-            break;
-        }
-        else {
-            ++p;
-        }
-    }
-    if (!has_error)
-        return unicode_fromascii((const unsigned char *)s, size);
-
-    v = PyUnicode_New(size, 127);
-    if (v == NULL)
-        goto onError;
-    if (size == 0)
-        return v;
-    kind = PyUnicode_KIND(v);
-    data = PyUnicode_DATA(v);
+        if (*p & 0x80)
+            goto hasError;
+        ++p;
+    }
+    memcpy(PyUnicode_1BYTE_DATA(v), s, size);
+    return v;
+
+hasError:
+    kind = PyUnicode_1BYTE_KIND;
+    data = PyUnicode_1BYTE_DATA(v);
     outpos = 0;
     e = s + size;
     while (s < e) {