[Python-checkins] r46438 - in python/trunk: Include/pyport.h Python/ceval.c
fredrik.lundh
python-checkins at python.org
Sat May 27 12:39:49 CEST 2006
Author: fredrik.lundh
Date: Sat May 27 12:39:48 2006
New Revision: 46438
Modified:
python/trunk/Include/pyport.h
python/trunk/Python/ceval.c
Log:
needforspeed: backed out the Py_LOCAL-isation of ceval; the massive in-
lining killed performance on certain Intel boxes, and the "aggressive"
macro itself gives most of the benefits on others.
Modified: python/trunk/Include/pyport.h
==============================================================================
--- python/trunk/Include/pyport.h (original)
+++ python/trunk/Include/pyport.h Sat May 27 12:39:48 2006
@@ -137,13 +137,17 @@
# endif
#endif
-/* PY_LOCAL can be used instead of static to get the fastest possible calling
- * convention for functions that are local to a given module. It also enables
- * inlining, where suitable.
+/* Py_LOCAL can be used instead of static to get the fastest possible calling
+ * convention for functions that are local to a given module.
*
- * 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.
+ * Py_LOCAL_INLINE does the same thing, and also explicitly requests inlining,
+ * for platforms that support that.
+ *
+ * If PY_LOCAL_AGGRESSIVE is defined before python.h is included, more
+ * "aggressive" inlining/optimizaion is enabled for the entire module. This
+ * may lead to code bloat, and may slow things down for those reasons. It may
+ * also lead to errors, if the code relies on pointer aliasing. 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,
@@ -160,11 +164,14 @@
/* ignore warnings if the compiler decides not to inline a function */
#pragma warning(disable: 4710)
/* fastest possible local call under MSVC */
-#define Py_LOCAL(type) static __inline type __fastcall
+#define Py_LOCAL(type) static type __fastcall
+#define Py_LOCAL_INLINE(type) static __inline type __fastcall
#elif defined(USE_INLINE)
-#define Py_LOCAL(type) static inline type
+#define Py_LOCAL(type) static type
+#define Py_LOCAL_INLINE(type) static inline type
#else
#define Py_LOCAL(type) static type
+#define Py_LOCAL_INLINE(type) static type
#endif
#include <stdlib.h>
Modified: python/trunk/Python/ceval.c
==============================================================================
--- python/trunk/Python/ceval.c (original)
+++ python/trunk/Python/ceval.c Sat May 27 12:39:48 2006
@@ -6,7 +6,7 @@
XXX document it!
*/
-/* enable more aggressive local inlining (platform dependent) */
+/* enable more aggressive intra-module optimizations, where available */
#define PY_LOCAL_AGGRESSIVE
#include "Python.h"
@@ -19,11 +19,6 @@
#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)
@@ -38,7 +33,7 @@
#define READ_TIMESTAMP(var) ppc_getcounter(&var)
-Py_LOCAL(void)
+static void
ppc_getcounter(uint64 *v)
{
register unsigned long tbu, tb, tbu2;
@@ -91,44 +86,44 @@
/* Forward declarations */
#ifdef WITH_TSC
-Py_LOCAL(PyObject *) call_function(PyObject ***, int, uint64*, uint64*);
+static PyObject * call_function(PyObject ***, int, uint64*, uint64*);
#else
-Py_LOCAL(PyObject *) call_function(PyObject ***, int);
+static 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);
+static PyObject * fast_function(PyObject *, PyObject ***, int, int, int);
+static PyObject * do_call(PyObject *, PyObject ***, int, int);
+static PyObject * ext_do_call(PyObject *, PyObject ***, int, int, int);
+static PyObject * update_keyword_args(PyObject *, int, PyObject ***,PyObject *);
+static PyObject * update_star_args(int, int, PyObject *, PyObject ***);
+static PyObject * load_args(PyObject ***, int);
#define CALL_FLAG_VAR 1
#define CALL_FLAG_KW 2
#ifdef LLTRACE
static int lltrace;
-Py_LOCAL(int) prtrace(PyObject *, char *);
+static int prtrace(PyObject *, char *);
#endif
-Py_LOCAL(int) call_trace(Py_tracefunc, PyObject *, PyFrameObject *,
+static int call_trace(Py_tracefunc, PyObject *, PyFrameObject *,
int, PyObject *);
-Py_LOCAL(void) call_trace_protected(Py_tracefunc, PyObject *,
+static void call_trace_protected(Py_tracefunc, PyObject *,
PyFrameObject *, int, PyObject *);
-Py_LOCAL(void) call_exc_trace(Py_tracefunc, PyObject *, PyFrameObject *);
-Py_LOCAL(int) maybe_call_line_trace(Py_tracefunc, PyObject *,
+static void call_exc_trace(Py_tracefunc, PyObject *, PyFrameObject *);
+static int maybe_call_line_trace(Py_tracefunc, PyObject *,
PyFrameObject *, int *, int *, int *);
-Py_LOCAL(PyObject *) apply_slice(PyObject *, PyObject *, PyObject *);
-Py_LOCAL(int) assign_slice(PyObject *, PyObject *,
+static PyObject * apply_slice(PyObject *, PyObject *, PyObject *);
+static int assign_slice(PyObject *, PyObject *,
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(int) exec_statement(PyFrameObject *,
+static PyObject * cmp_outcome(int, PyObject *, PyObject *);
+static PyObject * import_from(PyObject *, PyObject *);
+static int import_all_from(PyObject *, PyObject *);
+static PyObject * build_class(PyObject *, PyObject *, PyObject *);
+static 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 *,
+static void set_exc_info(PyThreadState *, PyObject *, PyObject *, PyObject *);
+static void reset_exc_info(PyThreadState *);
+static void format_exc_check_arg(PyObject *, char *, PyObject *);
+static PyObject * string_concatenate(PyObject *, PyObject *,
PyFrameObject *, unsigned char *);
#define NAME_ERROR_MSG \
@@ -484,8 +479,8 @@
WHY_YIELD = 0x0040 /* 'yield' operator */
};
-Py_LOCAL(enum why_code) do_raise(PyObject *, PyObject *, PyObject *);
-Py_LOCAL(int) unpack_iterable(PyObject *, int, PyObject **);
+static enum why_code do_raise(PyObject *, PyObject *, PyObject *);
+static int unpack_iterable(PyObject *, int, PyObject **);
/* for manipulating the thread switch and periodic "stuff" - used to be
per thread, now just a pair o' globals */
@@ -2902,7 +2897,7 @@
*/
-Py_LOCAL(void)
+static void
set_exc_info(PyThreadState *tstate,
PyObject *type, PyObject *value, PyObject *tb)
{
@@ -2947,7 +2942,7 @@
PySys_SetObject("exc_traceback", tb);
}
-Py_LOCAL(void)
+static void
reset_exc_info(PyThreadState *tstate)
{
PyFrameObject *frame;
@@ -2994,7 +2989,7 @@
/* Logic for the raise statement (too complicated for inlining).
This *consumes* a reference count to each of its arguments. */
-Py_LOCAL(enum why_code)
+static enum why_code
do_raise(PyObject *type, PyObject *value, PyObject *tb)
{
if (type == NULL) {
@@ -3103,7 +3098,7 @@
/* Iterate v argcnt times and store the results on the stack (via decreasing
sp). Return 1 for success, 0 if error. */
-Py_LOCAL(int)
+static int
unpack_iterable(PyObject *v, int argcnt, PyObject **sp)
{
int i = 0;
@@ -3150,7 +3145,7 @@
#ifdef LLTRACE
-Py_LOCAL(int)
+static int
prtrace(PyObject *v, char *str)
{
printf("%s ", str);
@@ -3161,7 +3156,7 @@
}
#endif
-Py_LOCAL(void)
+static void
call_exc_trace(Py_tracefunc func, PyObject *self, PyFrameObject *f)
{
PyObject *type, *value, *traceback, *arg;
@@ -3187,7 +3182,7 @@
}
}
-Py_LOCAL(void)
+static void
call_trace_protected(Py_tracefunc func, PyObject *obj, PyFrameObject *frame,
int what, PyObject *arg)
{
@@ -3204,7 +3199,7 @@
}
}
-Py_LOCAL(int)
+static int
call_trace(Py_tracefunc func, PyObject *obj, PyFrameObject *frame,
int what, PyObject *arg)
{
@@ -3239,7 +3234,7 @@
return result;
}
-Py_LOCAL(int)
+static int
maybe_call_line_trace(Py_tracefunc func, PyObject *obj,
PyFrameObject *frame, int *instr_lb, int *instr_ub,
int *instr_prev)
@@ -3467,7 +3462,7 @@
}
}
-Py_LOCAL(void)
+static void
err_args(PyObject *func, int flags, int nargs)
{
if (flags & METH_NOARGS)
@@ -3514,7 +3509,7 @@
x = call; \
}
-Py_LOCAL(PyObject *)
+static PyObject *
call_function(PyObject ***pp_stack, int oparg
#ifdef WITH_TSC
, uint64* pintr0, uint64* pintr1
@@ -3605,7 +3600,7 @@
done before evaluating the frame.
*/
-Py_LOCAL(PyObject *)
+static PyObject *
fast_function(PyObject *func, PyObject ***pp_stack, int n, int na, int nk)
{
PyCodeObject *co = (PyCodeObject *)PyFunction_GET_CODE(func);
@@ -3658,7 +3653,7 @@
PyFunction_GET_CLOSURE(func));
}
-Py_LOCAL(PyObject *)
+static PyObject *
update_keyword_args(PyObject *orig_kwdict, int nk, PyObject ***pp_stack,
PyObject *func)
{
@@ -3698,7 +3693,7 @@
return kwdict;
}
-Py_LOCAL(PyObject *)
+static PyObject *
update_star_args(int nstack, int nstar, PyObject *stararg,
PyObject ***pp_stack)
{
@@ -3723,7 +3718,7 @@
return callargs;
}
-Py_LOCAL(PyObject *)
+static PyObject *
load_args(PyObject ***pp_stack, int na)
{
PyObject *args = PyTuple_New(na);
@@ -3738,7 +3733,7 @@
return args;
}
-Py_LOCAL(PyObject *)
+static PyObject *
do_call(PyObject *func, PyObject ***pp_stack, int na, int nk)
{
PyObject *callargs = NULL;
@@ -3774,7 +3769,7 @@
return result;
}
-Py_LOCAL(PyObject *)
+static PyObject *
ext_do_call(PyObject *func, PyObject ***pp_stack, int flags, int na, int nk)
{
int nstar = 0;
@@ -3886,7 +3881,7 @@
PyType_HasFeature((x)->ob_type, Py_TPFLAGS_HAVE_INDEX) \
&& (x)->ob_type->tp_as_number->nb_index))
-Py_LOCAL(PyObject *)
+static PyObject *
apply_slice(PyObject *u, PyObject *v, PyObject *w) /* return u[v:w] */
{
PyTypeObject *tp = u->ob_type;
@@ -3912,7 +3907,7 @@
}
}
-Py_LOCAL(int)
+static int
assign_slice(PyObject *u, PyObject *v, PyObject *w, PyObject *x)
/* u[v:w] = x */
{
@@ -3946,7 +3941,7 @@
}
}
-Py_LOCAL(PyObject *)
+static PyObject *
cmp_outcome(int op, register PyObject *v, register PyObject *w)
{
int res = 0;
@@ -3979,7 +3974,7 @@
return v;
}
-Py_LOCAL(PyObject *)
+static PyObject *
import_from(PyObject *v, PyObject *name)
{
PyObject *x;
@@ -3993,7 +3988,7 @@
return x;
}
-Py_LOCAL(int)
+static int
import_all_from(PyObject *locals, PyObject *v)
{
PyObject *all = PyObject_GetAttrString(v, "__all__");
@@ -4050,7 +4045,7 @@
return err;
}
-Py_LOCAL(PyObject *)
+static PyObject *
build_class(PyObject *methods, PyObject *bases, PyObject *name)
{
PyObject *metaclass = NULL, *result, *base;
@@ -4102,7 +4097,7 @@
return result;
}
-Py_LOCAL(int)
+static int
exec_statement(PyFrameObject *f, PyObject *prog, PyObject *globals,
PyObject *locals)
{
@@ -4198,7 +4193,7 @@
return 0;
}
-Py_LOCAL(void)
+static void
format_exc_check_arg(PyObject *exc, char *format_str, PyObject *obj)
{
char *obj_str;
@@ -4213,7 +4208,7 @@
PyErr_Format(exc, format_str, obj_str);
}
-Py_LOCAL(PyObject *)
+static PyObject *
string_concatenate(PyObject *v, PyObject *w,
PyFrameObject *f, unsigned char *next_instr)
{
@@ -4288,7 +4283,7 @@
#ifdef DYNAMIC_EXECUTION_PROFILE
-Py_LOCAL(PyObject *)
+static PyObject *
getarray(long a[256])
{
int i;
More information about the Python-checkins
mailing list