needforspeed: added PY_LOCAL_AGGRESSIVE macro to enable "aggressive"

LOCAL inlining; also added some missing whitespace

diff --git a/Include/pyport.h b/Include/pyport.h
index 9d46470eb1e03221dbc873bbc0760cb79de0202b..07fdf284e547628916890dca378efc43720952fb 100644 (file)
--- a/Include/pyport.h
+++ b/Include/pyport.h
@@ -141,6 +141,10 @@ typedef Py_intptr_t        Py_ssize_t;
  * convention for functions that are local to a given module.  It also enables
  * inlining, where suitable.
  *
+ * If PY_LOCAL_AGGRESSIVE is defined before python.h is included, a more
+ * "aggressive" inlining is enabled.  This may lead to code bloat, and may
+ * slow things down for those reasons.  Use with care.
+ *
  * NOTE: You can only use this for functions that are entirely local to a
  * module; functions that are exported via method tables, callbacks, etc,
  * should keep using static.
@@ -149,6 +153,10 @@ typedef Py_intptr_t        Py_ssize_t;
 #undef USE_INLINE /* XXX - set via configure? */
 
 #if defined(_MSC_VER)
+#if defined(PY_LOCAL_AGGRESSIVE)
+/* enable more aggressive optimization for visual studio */
+#pragma optimize("agtw", on)
+#endif
 /* ignore warnings if the compiler decides not to inline a function */ 
 #pragma warning(disable: 4710)
 /* fastest possible local call under MSVC */

diff --git a/Python/ceval.c b/Python/ceval.c
index 53a263aa162b68c60fc1112c0842a4592eec3b89..da27fff9e277e7019561ddfdc5e10c701f66ce94 100644 (file)
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -6,6 +6,9 @@
    XXX document it!
    */
 
+/* enable more aggressive local inlining (platform dependent) */
+#define PY_LOCAL_AGGRESSIVE
+
 #include "Python.h"
 
 #include "code.h"
@@ -16,6 +19,11 @@
 
 #include <ctype.h>
 
+#if defined(_MSC_VER)
+/* enable more aggressive optimization for visual studio */
+#pragma optimize("agtw", on)
+#endif
+
 #ifndef WITH_TSC 
 
 #define READ_TIMESTAMP(var)
@@ -83,16 +91,16 @@ typedef PyObject *(*callproc)(PyObject *, PyObject *, PyObject *);
 
 /* Forward declarations */
 #ifdef WITH_TSC
-Py_LOCAL(PyObject *)call_function(PyObject ***, int, uint64*, uint64*);
+Py_LOCAL(PyObject *) call_function(PyObject ***, int, uint64*, uint64*);
 #else
-Py_LOCAL(PyObject *)call_function(PyObject ***, int);
+Py_LOCAL(PyObject *) call_function(PyObject ***, int);
 #endif
-Py_LOCAL(PyObject *)fast_function(PyObject *, PyObject ***, int, int, int);
-Py_LOCAL(PyObject *)do_call(PyObject *, PyObject ***, int, int);
-Py_LOCAL(PyObject *)ext_do_call(PyObject *, PyObject ***, int, int, int);
-Py_LOCAL(PyObject *)update_keyword_args(PyObject *, int, PyObject ***,PyObject *);
-Py_LOCAL(PyObject *)update_star_args(int, int, PyObject *, PyObject ***);
-Py_LOCAL(PyObject *)load_args(PyObject ***, int);
+Py_LOCAL(PyObject *) fast_function(PyObject *, PyObject ***, int, int, int);
+Py_LOCAL(PyObject *) do_call(PyObject *, PyObject ***, int, int);
+Py_LOCAL(PyObject *) ext_do_call(PyObject *, PyObject ***, int, int, int);
+Py_LOCAL(PyObject *) update_keyword_args(PyObject *, int, PyObject ***,PyObject *);
+Py_LOCAL(PyObject *) update_star_args(int, int, PyObject *, PyObject ***);
+Py_LOCAL(PyObject *) load_args(PyObject ***, int);
 #define CALL_FLAG_VAR 1
 #define CALL_FLAG_KW 2
 
@@ -108,19 +116,19 @@ Py_LOCAL(void) call_exc_trace(Py_tracefunc, PyObject *, PyFrameObject *);
 Py_LOCAL(int) maybe_call_line_trace(Py_tracefunc, PyObject *,
                                  PyFrameObject *, int *, int *, int *);
 
-Py_LOCAL(PyObject *)apply_slice(PyObject *, PyObject *, PyObject *);
+Py_LOCAL(PyObject *) apply_slice(PyObject *, PyObject *, PyObject *);
 Py_LOCAL(int) assign_slice(PyObject *, PyObject *,
                        PyObject *, PyObject *);
-Py_LOCAL(PyObject *)cmp_outcome(int, PyObject *, PyObject *);
-Py_LOCAL(PyObject *)import_from(PyObject *, PyObject *);
+Py_LOCAL(PyObject *) cmp_outcome(int, PyObject *, PyObject *);
+Py_LOCAL(PyObject *) import_from(PyObject *, PyObject *);
 Py_LOCAL(int) import_all_from(PyObject *, PyObject *);
-Py_LOCAL(PyObject *)build_class(PyObject *, PyObject *, PyObject *);
+Py_LOCAL(PyObject *) build_class(PyObject *, PyObject *, PyObject *);
 Py_LOCAL(int) exec_statement(PyFrameObject *,
                          PyObject *, PyObject *, PyObject *);
 Py_LOCAL(void) set_exc_info(PyThreadState *, PyObject *, PyObject *, PyObject *);
 Py_LOCAL(void) reset_exc_info(PyThreadState *);
 Py_LOCAL(void) format_exc_check_arg(PyObject *, char *, PyObject *);
-Py_LOCAL(PyObject *)string_concatenate(PyObject *, PyObject *,
+Py_LOCAL(PyObject *) string_concatenate(PyObject *, PyObject *,
                                    PyFrameObject *, unsigned char *);
 
 #define NAME_ERROR_MSG \
@@ -476,7 +484,7 @@ enum why_code {
                WHY_YIELD =     0x0040  /* 'yield' operator */
 };
 
-static enum why_code do_raise(PyObject *, PyObject *, PyObject *);
+Py_LOCAL(enum why_code) do_raise(PyObject *, PyObject *, PyObject *);
 Py_LOCAL(int) unpack_iterable(PyObject *, int, PyObject **);
 
 /* for manipulating the thread switch and periodic "stuff" - used to be
@@ -2971,7 +2979,7 @@ reset_exc_info(PyThreadState *tstate)
 
 /* Logic for the raise statement (too complicated for inlining).
    This *consumes* a reference count to each of its arguments. */
-static enum why_code
+Py_LOCAL(enum why_code)
 do_raise(PyObject *type, PyObject *value, PyObject *tb)
 {
        if (type == NULL) {