bpo-46072: Merge dxpairs into py_stats. (GH-31197)

Author: markshannon

Misc/NEWS.d/next/Core and Builtins/2022-02-07-14-38-54.bpo-46072.6ebLyN.rst

@@ -0,0 +1,2 @@
+Opcode pair stats are now gathered with ``--enable-pystats``. Defining
+``DYNAMIC_EXECUTION_PROFILE`` or  ``DXPAIRS`` no longer has any effect.

Python/ceval.c

@@ -114,16 +114,6 @@ _PyEvalFrameClearAndPop(PyThreadState *tstate, InterpreterFrame *frame);
     "cannot access free variable '%s' where it is not associated with a" \
     " value in enclosing scope"
 
-/* Dynamic execution profile */
-#ifdef DYNAMIC_EXECUTION_PROFILE
-#ifdef DXPAIRS
-static long dxpairs[257][256];
-#define dxp dxpairs[256]
-#else
-static long dxp[256];
-#endif
-#endif
-
 #ifndef NDEBUG
 /* Ensure that tstate is valid: sanity check for PyEval_AcquireThread() and
    PyEval_RestoreThread(). Detect if tstate memory was freed. It can happen
@@ -1238,10 +1228,6 @@ eval_frame_handle_pending(PyThreadState *tstate)
    faster than the normal "switch" version, depending on the compiler and the
    CPU architecture.
 
-   We disable the optimization if DYNAMIC_EXECUTION_PROFILE is defined,
-   because it would render the measurements invalid.
-
-
    NOTE: care must be taken that the compiler doesn't try to "optimize" the
    indirect jumps by sharing them between all opcodes. Such optimizations
    can be disabled on gcc by using the -fno-gcse flag (or possibly
@@ -1253,23 +1239,13 @@ eval_frame_handle_pending(PyThreadState *tstate)
  * We want to be sure that the compiler knows this before it generates
  * the CFG.
  */
-#ifdef LLTRACE
-#define LLTRACE_INSTR() if (lltrace) { lltrace_instruction(frame, opcode, oparg); }
-#else
-#define LLTRACE_INSTR() ((void)0)
-#endif
 
 #ifdef WITH_DTRACE
 #define OR_DTRACE_LINE | (PyDTrace_LINE_ENABLED() ? 255 : 0)
 #else
 #define OR_DTRACE_LINE
 #endif
 
-#ifdef DYNAMIC_EXECUTION_PROFILE
-#undef USE_COMPUTED_GOTOS
-#define USE_COMPUTED_GOTOS 0
-#endif
-
 #ifdef HAVE_COMPUTED_GOTOS
     #ifndef USE_COMPUTED_GOTOS
     #define USE_COMPUTED_GOTOS 1
@@ -1283,7 +1259,14 @@ eval_frame_handle_pending(PyThreadState *tstate)
 #endif
 
 #ifdef Py_STATS
-#define INSTRUCTION_START(op) frame->f_lasti = INSTR_OFFSET(); next_instr++; OPCODE_EXE_INC(op);
+#define INSTRUCTION_START(op) \
+    do { \
+        frame->f_lasti = INSTR_OFFSET(); \
+        next_instr++; \
+        OPCODE_EXE_INC(op); \
+        _py_stats.opcode_stats[lastopcode].pair_count[op]++; \
+        lastopcode = op; \
+    } while (0)
 #else
 #define INSTRUCTION_START(op) frame->f_lasti = INSTR_OFFSET(); next_instr++
 #endif
@@ -1296,34 +1279,12 @@ eval_frame_handle_pending(PyThreadState *tstate)
 #define DISPATCH_GOTO() goto dispatch_opcode
 #endif
 
-/* RECORD_DXPROFILE()  records the dxprofile information, if enabled. Normally a no-op */
-#ifdef DYNAMIC_EXECUTION_PROFILE
-#ifdef DXPAIRS
-#define RECORD_DXPROFILE() \
-    do { \
-        dxpairs[lastopcode][opcode]++; \
-        lastopcode = opcode; \
-        dxp[opcode]++; \
-    } while (0)
-#else
-  #define RECORD_DXPROFILE() \
-    do { \
-        dxp[opcode]++; \
-    } while (0)
-#endif
+/* PRE_DISPATCH_GOTO() does lltrace if enabled. Normally a no-op */
+#ifdef LLTRACE
+#define PRE_DISPATCH_GOTO() if (lltrace) { lltrace_instruction(frame, opcode, oparg); }
 #else
-#define RECORD_DXPROFILE() ((void)0)
-#endif
-
-/* PRE_DISPATCH_GOTO() does lltrace and dxprofile if either is enabled. Normally a no-op */
-#ifndef LLTRACE
-#ifndef DYNAMIC_EXECUTION_PROFILE
 #define PRE_DISPATCH_GOTO() ((void)0)
 #endif
-#endif
-#ifndef PRE_DISPATCH_GOTO
-#define PRE_DISPATCH_GOTO() do { LLTRACE_INSTR(); RECORD_DXPROFILE(); } while (0)
-#endif
 
 #define NOTRACE_DISPATCH() \
     { \
@@ -1403,7 +1364,7 @@ eval_frame_handle_pending(PyThreadState *tstate)
 
 #define PREDICT_ID(op)          PRED_##op
 
-#if defined(DYNAMIC_EXECUTION_PROFILE) || USE_COMPUTED_GOTOS
+#if USE_COMPUTED_GOTOS
 #define PREDICT(op)             if (0) goto PREDICT_ID(op)
 #else
 #define PREDICT(op) \
@@ -1653,7 +1614,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
 #include "opcode_targets.h"
 #endif
 
-#ifdef DXPAIRS
+#ifdef Py_STATS
     int lastopcode = 0;
 #endif
     int opcode;        /* Current opcode */
@@ -7489,16 +7450,16 @@ format_awaitable_error(PyThreadState *tstate, PyTypeObject *type, int prevprevpr
     }
 }
 
-#ifdef DYNAMIC_EXECUTION_PROFILE
+#ifdef Py_STATS
 
 static PyObject *
-getarray(long a[256])
+getarray(uint64_t a[256])
 {
     int i;
     PyObject *l = PyList_New(256);
     if (l == NULL) return NULL;
     for (i = 0; i < 256; i++) {
-        PyObject *x = PyLong_FromLong(a[i]);
+        PyObject *x = PyLong_FromUnsignedLongLong(a[i]);
         if (x == NULL) {
             Py_DECREF(l);
             return NULL;
@@ -7513,22 +7474,18 @@ getarray(long a[256])
 PyObject *
 _Py_GetDXProfile(PyObject *self, PyObject *args)
 {
-#ifndef DXPAIRS
-    return getarray(dxp);
-#else
     int i;
     PyObject *l = PyList_New(257);
     if (l == NULL) return NULL;
     for (i = 0; i < 257; i++) {
-        PyObject *x = getarray(dxpairs[i]);
+        PyObject *x = getarray(_py_stats.opcode_stats[i].pair_count);
         if (x == NULL) {
             Py_DECREF(l);
             return NULL;
         }
         PyList_SET_ITEM(l, i, x);
     }
     return l;
-#endif
 }
 
 #endif

Python/sysmodule.c

@@ -1923,7 +1923,7 @@ sys__debugmallocstats_impl(PyObject *module)
 extern PyObject *_Py_GetObjects(PyObject *, PyObject *);
 #endif
 
-#ifdef DYNAMIC_EXECUTION_PROFILE
+#ifdef Py_STATS
 /* Defined in ceval.c because it uses static globals in that file */
 extern PyObject *_Py_GetDXProfile(PyObject *,  PyObject *);
 #endif
@@ -1992,7 +1992,7 @@ static PyMethodDef sys_methods[] = {
     SYS_GETDEFAULTENCODING_METHODDEF
     SYS_GETDLOPENFLAGS_METHODDEF
     SYS_GETALLOCATEDBLOCKS_METHODDEF
-#ifdef DYNAMIC_EXECUTION_PROFILE
+#ifdef Py_STATS
     {"getdxp",          _Py_GetDXProfile, METH_VARARGS},
 #endif
     SYS_GETFILESYSTEMENCODING_METHODDEF