Optimize bytes.translate() by deferring change detection

Move the equality check out of the hot loop to allow better compiler
optimization. Instead of checking each byte during translation, perform
a single memcmp at the end to determine if the input can be returned
unchanged.

This allows compilers to unroll and pipeline the loops, resulting in ~2x
throughput improvement for medium-to-large inputs (tested on an AMD zen2).
No change observed on small inputs.

It will also be faster for bytes subclasses as those do not need change
detection.

Author: gpshead

Objects/bytesobject.c

@@ -2237,11 +2237,15 @@ bytes_translate_impl(PyBytesObject *self, PyObject *table,
         /* If no deletions are required, use faster code */
         for (i = inlen; --i >= 0; ) {
             c = Py_CHARMASK(*input++);
-            if (Py_CHARMASK((*output++ = table_chars[c])) != c)
-                changed = 1;
-        }
-        if (!changed && PyBytes_CheckExact(input_obj)) {
-            Py_SETREF(result, Py_NewRef(input_obj));
+            *output++ = table_chars[c];
+        }
+        /* Check if anything changed (for returning original object) */
+        /* We save this check until the end so that the compiler will */
+        /* unroll the loop above leading to MUCH faster code. */
+        if (PyBytes_CheckExact(input_obj)) {
+            if (memcmp(PyBytes_AS_STRING(input_obj), output_start, inlen) == 0) {
+                Py_SETREF(result, Py_NewRef(input_obj));
+            }
         }
         PyBuffer_Release(&del_table_view);
         PyBuffer_Release(&table_view);