Optimize pickle.load() and pickle.loads()

Issue #27056: Optimize pickle.load() and pickle.loads(), up to 10% faster to
deserialize a lot of small objects.

3 files changed, 33 insertions, 20 deletions

diff --git a/Doc/whatsnew/3.6.rst b/Doc/whatsnew/3.6.rst
index 52be134..67fd50f 100644
--- a/Doc/whatsnew/3.6.rst
+++ b/Doc/whatsnew/3.6.rst
@@ -467,6 +467,9 @@ Optimizations
   with a short lifetime, and use :c:func:`malloc` for larger memory blocks.
   (Contributed by Victor Stinner in :issue:`26249`).
 
+* :func:`pickle.load` and :func:`pickle.loads` are now up to 10% faster when
+  deserializing many small objects (Contributed by Victor Stinner in
+  :issue:`27056`).
 
 Build and C API Changes
 =======================
diff --git a/Misc/NEWS b/Misc/NEWS
index 94e508f..ba66c4e 100644
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -16,6 +16,9 @@ Core and Builtins
 Library
 -------
 
+- Issue #27056: Optimize pickle.load() and pickle.loads(), up to 10% faster
+  to deserialize a lot of small objects.
+
 
 What's New in Python 3.6.0 alpha 1?
 ===================================
@@ -341,7 +344,7 @@ Library
 - Issue #26977: Removed unnecessary, and ignored, call to sum of squares helper
   in statistics.pvariance.
 
-- Issue #26002: Use bisect in statistics.median instead of a linear search. 
+- Issue #26002: Use bisect in statistics.median instead of a linear search.
   Patch by Upendra Kuma.
 
 - Issue #25974: Make use of new Decimal.as_integer_ratio() method in statistics
diff --git a/Modules/_pickle.c b/Modules/_pickle.c
index fdd60e0..e3aa7c5 100644
--- a/Modules/_pickle.c
+++ b/Modules/_pickle.c
@@ -1197,21 +1197,9 @@ _Unpickler_ReadFromFile(UnpicklerObject *self, Py_ssize_t n)
     return read_size;
 }
 
-/* Read `n` bytes from the unpickler's data source, storing the result in `*s`.
-
-   This should be used for all data reads, rather than accessing the unpickler's
-   input buffer directly. This method deals correctly with reading from input
-   streams, which the input buffer doesn't deal with.
-
-   Note that when reading from a file-like object, self->next_read_idx won't
-   be updated (it should remain at 0 for the entire unpickling process). You
-   should use this function's return value to know how many bytes you can
-   consume.
-
-   Returns -1 (with an exception set) on failure. On success, return the
-   number of chars read. */
+/* Don't call it directly: use _Unpickler_Read() */
 static Py_ssize_t
-_Unpickler_Read(UnpicklerObject *self, char **s, Py_ssize_t n)
+_Unpickler_ReadImpl(UnpicklerObject *self, char **s, Py_ssize_t n)
 {
     Py_ssize_t num_read;
 
@@ -1222,11 +1210,10 @@ _Unpickler_Read(UnpicklerObject *self, char **s, Py_ssize_t n)
                         "read would overflow (invalid bytecode)");
         return -1;
     }
-    if (self->next_read_idx + n <= self->input_len) {
-        *s = self->input_buffer + self->next_read_idx;
-        self->next_read_idx += n;
-        return n;
-    }
+
+    /* This case is handled by the _Unpickler_Read() macro for efficiency */
+    assert(self->next_read_idx + n > self->input_len);
+
     if (!self->read) {
         PyErr_Format(PyExc_EOFError, "Ran out of input");
         return -1;
@@ -1243,6 +1230,26 @@ _Unpickler_Read(UnpicklerObject *self, char **s, Py_ssize_t n)
     return n;
 }
 
+/* Read `n` bytes from the unpickler's data source, storing the result in `*s`.
+
+   This should be used for all data reads, rather than accessing the unpickler's
+   input buffer directly. This method deals correctly with reading from input
+   streams, which the input buffer doesn't deal with.
+
+   Note that when reading from a file-like object, self->next_read_idx won't
+   be updated (it should remain at 0 for the entire unpickling process). You
+   should use this function's return value to know how many bytes you can
+   consume.
+
+   Returns -1 (with an exception set) on failure. On success, return the
+   number of chars read. */
+#define _Unpickler_Read(self, s, n) \
+    (((self)->next_read_idx + (n) <= (self)->input_len)      \
+     ? (*(s) = (self)->input_buffer + (self)->next_read_idx, \
+        (self)->next_read_idx += (n),                        \
+        (n))                                                 \
+     : _Unpickler_ReadImpl(self, (s), (n)))
+
 static Py_ssize_t
 _Unpickler_CopyLine(UnpicklerObject *self, char *line, Py_ssize_t len,
                     char **result)