/* Execute compiled code */

/* XXX TO DO:

XXX speed up searching for keywords by using a dictionary

XXX document it!

*/

/* enable more aggressive intra-module optimizations, where available */

#define PY_LOCAL_AGGRESSIVE

#include "Python.h"

#include "code.h"

#include "dictobject.h"

#include "frameobject.h"

#include "opcode.h"

#include "setobject.h"

#include "structmember.h"

#include <ctype.h>

#ifndef WITH_TSC

#define READ_TIMESTAMP(var)

#else

typedef unsigned long long uint64;

/* PowerPC support.

"__ppc__" appears to be the preprocessor definition to detect on OS X, whereas

"__powerpc__" appears to be the correct one for Linux with GCC

*/

#if defined(__ppc__) || defined (__powerpc__)

#define READ_TIMESTAMP(var) ppc_getcounter(&var)

static void

ppc_getcounter(uint64 *v)

{

unsigned long tbu, tb, tbu2;

loop:

asm volatile ("mftbu %0" : "=r" (tbu) );

asm volatile ("mftb %0" : "=r" (tb) );

asm volatile ("mftbu %0" : "=r" (tbu2));

if (__builtin_expect(tbu != tbu2, 0)) goto loop;

/* The slightly peculiar way of writing the next lines is

compiled better by GCC than any other way I tried. */

((long*)(v))[0] = tbu;

((long*)(v))[1] = tb;

}

#elif defined(__i386__)

/* this is for linux/x86 (and probably any other GCC/x86 combo) */

#define READ_TIMESTAMP(val) \

__asm__ __volatile__("rdtsc" : "=A" (val))

#elif defined(__x86_64__)

/* for gcc/x86_64, the "A" constraint in DI mode means *either* rax *or* rdx;

not edx:eax as it does for i386. Since rdtsc puts its result in edx:eax

even in 64-bit mode, we need to use "a" and "d" for the lower and upper

32-bit pieces of the result. */

#define READ_TIMESTAMP(val) do { \

unsigned int h, l; \

__asm__ __volatile__("rdtsc" : "=a" (l), "=d" (h)); \

(val) = ((uint64)l) | (((uint64)h) << 32); \

} while(0)

#else

#error "Don't know how to implement timestamp counter for this architecture"

#endif

void dump_tsc(int opcode, int ticked, uint64 inst0, uint64 inst1,

uint64 loop0, uint64 loop1, uint64 intr0, uint64 intr1)

{

uint64 intr, inst, loop;

PyThreadState *tstate = PyThreadState_Get();

if (!tstate->interp->tscdump)

return;

intr = intr1 - intr0;

inst = inst1 - inst0 - intr;

loop = loop1 - loop0 - intr;

fprintf(stderr, "opcode=%03d t=%d inst=%06lld loop=%06lld\n",

opcode, ticked, inst, loop);

}

#endif

/* Turn this on if your compiler chokes on the big switch: */

/* #define CASE_TOO_BIG 1 */

#ifdef Py_DEBUG

/* For debugging the interpreter: */

#define LLTRACE 1 /* Low-level trace feature */

#define CHECKEXC 1 /* Double-check exception checking */

#endif

typedef PyObject *(*callproc)(PyObject *, PyObject *, PyObject *);

/* Forward declarations */

#ifdef WITH_TSC

static PyObject * call_function(PyObject ***, Py_ssize_t, PyObject *, uint64*, uint64*);

#else

static PyObject * call_function(PyObject ***, Py_ssize_t, PyObject *);

#endif

static PyObject * fast_function(PyObject *, PyObject ***, Py_ssize_t, PyObject *);

static PyObject * do_call(PyObject *, PyObject ***, Py_ssize_t, PyObject *);

static PyObject * do_call_core(PyObject *, PyObject *, PyObject *);

static PyObject * create_keyword_args(PyObject *, PyObject ***, PyObject *);

static PyObject * load_args(PyObject ***, Py_ssize_t);

#ifdef LLTRACE

static int lltrace;

static int prtrace(PyObject *, const char *);

#endif

static int call_trace(Py_tracefunc, PyObject *,

PyThreadState *, PyFrameObject *,

int, PyObject *);

static int call_trace_protected(Py_tracefunc, PyObject *,

PyThreadState *, PyFrameObject *,

int, PyObject *);

static void call_exc_trace(Py_tracefunc, PyObject *,

PyThreadState *, PyFrameObject *);

static int maybe_call_line_trace(Py_tracefunc, PyObject *,

PyThreadState *, PyFrameObject *, int *, int *, int *);

static PyObject * cmp_outcome(int, PyObject *, PyObject *);

static PyObject * import_name(PyFrameObject *, PyObject *, PyObject *, PyObject *);

static PyObject * import_from(PyObject *, PyObject *);

static int import_all_from(PyObject *, PyObject *);

static void format_exc_check_arg(PyObject *, const char *, PyObject *);

static void format_exc_unbound(PyCodeObject *co, int oparg);

static PyObject * unicode_concatenate(PyObject *, PyObject *,

PyFrameObject *, const unsigned short *);

static PyObject * special_lookup(PyObject *, _Py_Identifier *);

#define NAME_ERROR_MSG \

"name '%.200s' is not defined"

#define UNBOUNDLOCAL_ERROR_MSG \

"local variable '%.200s' referenced before assignment"

#define UNBOUNDFREE_ERROR_MSG \

"free variable '%.200s' referenced before assignment" \

" 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

/* Function call profile */

#ifdef CALL_PROFILE

#define PCALL_NUM 11

static int pcall[PCALL_NUM];

#define PCALL_ALL 0

#define PCALL_FUNCTION 1

#define PCALL_FAST_FUNCTION 2

#define PCALL_FASTER_FUNCTION 3

#define PCALL_METHOD 4

#define PCALL_BOUND_METHOD 5

#define PCALL_CFUNCTION 6

#define PCALL_TYPE 7

#define PCALL_GENERATOR 8

#define PCALL_OTHER 9

#define PCALL_POP 10

/* Notes about the statistics

PCALL_FAST stats

FAST_FUNCTION means no argument tuple needs to be created.

FASTER_FUNCTION means that the fast-path frame setup code is used.

If there is a method call where the call can be optimized by changing

the argument tuple and calling the function directly, it gets recorded

twice.

As a result, the relationship among the statistics appears to be

PCALL_ALL == PCALL_FUNCTION + PCALL_METHOD - PCALL_BOUND_METHOD +

PCALL_CFUNCTION + PCALL_TYPE + PCALL_GENERATOR + PCALL_OTHER

PCALL_FUNCTION > PCALL_FAST_FUNCTION > PCALL_FASTER_FUNCTION

PCALL_METHOD > PCALL_BOUND_METHOD

*/

#define PCALL(POS) pcall[POS]++

PyObject *

PyEval_GetCallStats(PyObject *self)

{

return Py_BuildValue("iiiiiiiiiii",

pcall[0], pcall[1], pcall[2], pcall[3],

pcall[4], pcall[5], pcall[6], pcall[7],

pcall[8], pcall[9], pcall[10]);

}

#else

#define PCALL(O)

PyObject *

PyEval_GetCallStats(PyObject *self)

{

Py_INCREF(Py_None);

return Py_None;

}

#endif

#ifdef WITH_THREAD

#define GIL_REQUEST _Py_atomic_load_relaxed(&gil_drop_request)

#else

#define GIL_REQUEST 0

#endif

/* This can set eval_breaker to 0 even though gil_drop_request became

1. We believe this is all right because the eval loop will release

the GIL eventually anyway. */

#define COMPUTE_EVAL_BREAKER() \

_Py_atomic_store_relaxed( \

&eval_breaker, \

GIL_REQUEST | \

_Py_atomic_load_relaxed(&pendingcalls_to_do) | \

pending_async_exc)

#ifdef WITH_THREAD

#define SET_GIL_DROP_REQUEST() \

do { \

_Py_atomic_store_relaxed(&gil_drop_request, 1); \

_Py_atomic_store_relaxed(&eval_breaker, 1); \

} while (0)

#define RESET_GIL_DROP_REQUEST() \

do { \

_Py_atomic_store_relaxed(&gil_drop_request, 0); \

COMPUTE_EVAL_BREAKER(); \

} while (0)

#endif

/* Pending calls are only modified under pending_lock */

#define SIGNAL_PENDING_CALLS() \

do { \

_Py_atomic_store_relaxed(&pendingcalls_to_do, 1); \

_Py_atomic_store_relaxed(&eval_breaker, 1); \

} while (0)

#define UNSIGNAL_PENDING_CALLS() \

do { \

_Py_atomic_store_relaxed(&pendingcalls_to_do, 0); \

COMPUTE_EVAL_BREAKER(); \

} while (0)

#define SIGNAL_ASYNC_EXC() \

do { \

pending_async_exc = 1; \

_Py_atomic_store_relaxed(&eval_breaker, 1); \

} while (0)

#define UNSIGNAL_ASYNC_EXC() \

do { pending_async_exc = 0; COMPUTE_EVAL_BREAKER(); } while (0)

#ifdef WITH_THREAD

#ifdef HAVE_ERRNO_H

#include <errno.h>

#endif

#include "pythread.h"

static PyThread_type_lock pending_lock = 0; /* for pending calls */

static long main_thread = 0;

/* This single variable consolidates all requests to break out of the fast path

in the eval loop. */

static _Py_atomic_int eval_breaker = {0};

/* Request for dropping the GIL */

static _Py_atomic_int gil_drop_request = {0};

/* Request for running pending calls. */

static _Py_atomic_int pendingcalls_to_do = {0};

/* Request for looking at the `async_exc` field of the current thread state.

Guarded by the GIL. */

static int pending_async_exc = 0;

#include "ceval_gil.h"

int

PyEval_ThreadsInitialized(void)

{

return gil_created();

}

void

PyEval_InitThreads(void)

{

if (gil_created())

return;

create_gil();

take_gil(PyThreadState_GET());

main_thread = PyThread_get_thread_ident();

if (!pending_lock)

pending_lock = PyThread_allocate_lock();

}

void

_PyEval_FiniThreads(void)

{

if (!gil_created())

return;

destroy_gil();

assert(!gil_created());

}

void

PyEval_AcquireLock(void)

{

PyThreadState *tstate = PyThreadState_GET();

if (tstate == NULL)

Py_FatalError("PyEval_AcquireLock: current thread state is NULL");

take_gil(tstate);

}

void

PyEval_ReleaseLock(void)

{

/* This function must succeed when the current thread state is NULL.

We therefore avoid PyThreadState_GET() which dumps a fatal error

in debug mode.

*/

drop_gil((PyThreadState*)_Py_atomic_load_relaxed(

&_PyThreadState_Current));

}

void

PyEval_AcquireThread(PyThreadState *tstate)

{

if (tstate == NULL)

Py_FatalError("PyEval_AcquireThread: NULL new thread state");

/* Check someone has called PyEval_InitThreads() to create the lock */

assert(gil_created());

take_gil(tstate);

if (PyThreadState_Swap(tstate) != NULL)

Py_FatalError(

"PyEval_AcquireThread: non-NULL old thread state");

}

void

PyEval_ReleaseThread(PyThreadState *tstate)

{

if (tstate == NULL)

Py_FatalError("PyEval_ReleaseThread: NULL thread state");

if (PyThreadState_Swap(NULL) != tstate)

Py_FatalError("PyEval_ReleaseThread: wrong thread state");

drop_gil(tstate);

}

/* This function is called from PyOS_AfterFork to destroy all threads which are

* not running in the child process, and clear internal locks which might be

* held by those threads. (This could also be done using pthread_atfork

* mechanism, at least for the pthreads implementation.) */

void

PyEval_ReInitThreads(void)

{

_Py_IDENTIFIER(_after_fork);

PyObject *threading, *result;

PyThreadState *current_tstate = PyThreadState_GET();

if (!gil_created())

return;

recreate_gil();

pending_lock = PyThread_allocate_lock();

take_gil(current_tstate);

main_thread = PyThread_get_thread_ident();

/* Update the threading module with the new state.

*/

threading = PyMapping_GetItemString(current_tstate->interp->modules,

"threading");

if (threading == NULL) {

/* threading not imported */

PyErr_Clear();

return;

}

result = _PyObject_CallMethodId(threading, &PyId__after_fork, NULL);

if (result == NULL)

PyErr_WriteUnraisable(threading);

else

Py_DECREF(result);

Py_DECREF(threading);

/* Destroy all threads except the current one */

_PyThreadState_DeleteExcept(current_tstate);

}

#else

static _Py_atomic_int eval_breaker = {0};

static int pending_async_exc = 0;

#endif /* WITH_THREAD */

/* This function is used to signal that async exceptions are waiting to be

raised, therefore it is also useful in non-threaded builds. */

void

_PyEval_SignalAsyncExc(void)

{

SIGNAL_ASYNC_EXC();

}

/* Functions save_thread and restore_thread are always defined so

dynamically loaded modules needn't be compiled separately for use

with and without threads: */

PyThreadState *

PyEval_SaveThread(void)

{

PyThreadState *tstate = PyThreadState_Swap(NULL);

if (tstate == NULL)

Py_FatalError("PyEval_SaveThread: NULL tstate");

#ifdef WITH_THREAD

if (gil_created())

drop_gil(tstate);

#endif

return tstate;

}

void

PyEval_RestoreThread(PyThreadState *tstate)

{

if (tstate == NULL)

Py_FatalError("PyEval_RestoreThread: NULL tstate");

#ifdef WITH_THREAD

if (gil_created()) {

int err = errno;

take_gil(tstate);

/* _Py_Finalizing is protected by the GIL */

if (_Py_Finalizing && tstate != _Py_Finalizing) {

drop_gil(tstate);

PyThread_exit_thread();

assert(0); /* unreachable */

}

errno = err;

}

#endif

PyThreadState_Swap(tstate);

}

/* Mechanism whereby asynchronously executing callbacks (e.g. UNIX

signal handlers or Mac I/O completion routines) can schedule calls

to a function to be called synchronously.

The synchronous function is called with one void* argument.

It should return 0 for success or -1 for failure -- failure should

be accompanied by an exception.

If registry succeeds, the registry function returns 0; if it fails

(e.g. due to too many pending calls) it returns -1 (without setting

an exception condition).

Note that because registry may occur from within signal handlers,

or other asynchronous events, calling malloc() is unsafe!

#ifdef WITH_THREAD

Any thread can schedule pending calls, but only the main thread

will execute them.

There is no facility to schedule calls to a particular thread, but

that should be easy to change, should that ever be required. In

that case, the static variables here should go into the python

threadstate.

#endif

*/

#ifdef WITH_THREAD

/* The WITH_THREAD implementation is thread-safe. It allows

scheduling to be made from any thread, and even from an executing

callback.

*/

#define NPENDINGCALLS 32

static struct {

int (*func)(void *);

void *arg;

} pendingcalls[NPENDINGCALLS];

static int pendingfirst = 0;

static int pendinglast = 0;

int

Py_AddPendingCall(int (*func)(void *), void *arg)

{

int i, j, result=0;

PyThread_type_lock lock = pending_lock;

/* try a few times for the lock. Since this mechanism is used

* for signal handling (on the main thread), there is a (slim)

* chance that a signal is delivered on the same thread while we

* hold the lock during the Py_MakePendingCalls() function.

* This avoids a deadlock in that case.

* Note that signals can be delivered on any thread. In particular,

* on Windows, a SIGINT is delivered on a system-created worker

* thread.

* We also check for lock being NULL, in the unlikely case that

* this function is called before any bytecode evaluation takes place.

*/

if (lock != NULL) {

for (i = 0; i<100; i++) {

if (PyThread_acquire_lock(lock, NOWAIT_LOCK))

break;

}

if (i == 100)

return -1;

}

i = pendinglast;

j = (i + 1) % NPENDINGCALLS;

if (j == pendingfirst) {

result = -1; /* Queue full */

} else {

pendingcalls[i].func = func;

pendingcalls[i].arg = arg;

pendinglast = j;

}

/* signal main loop */

SIGNAL_PENDING_CALLS();

if (lock != NULL)

PyThread_release_lock(lock);

return result;

}

int

Py_MakePendingCalls(void)

{

static int busy = 0;

int i;

int r = 0;

if (!pending_lock) {

/* initial allocation of the lock */

pending_lock = PyThread_allocate_lock();

if (pending_lock == NULL)

return -1;

}

/* only service pending calls on main thread */

if (main_thread && PyThread_get_thread_ident() != main_thread)

return 0;

/* don't perform recursive pending calls */

if (busy)

return 0;

busy = 1;

/* perform a bounded number of calls, in case of recursion */

for (i=0; i<NPENDINGCALLS; i++) {

int j;

int (*func)(void *);

void *arg = NULL;

/* pop one item off the queue while holding the lock */

PyThread_acquire_lock(pending_lock, WAIT_LOCK);

j = pendingfirst;

if (j == pendinglast) {

func = NULL; /* Queue empty */

} else {

func = pendingcalls[j].func;

arg = pendingcalls[j].arg;

pendingfirst = (j + 1) % NPENDINGCALLS;

}

if (pendingfirst != pendinglast)

SIGNAL_PENDING_CALLS();

else

UNSIGNAL_PENDING_CALLS();

PyThread_release_lock(pending_lock);

/* having released the lock, perform the callback */

if (func == NULL)

break;

r = func(arg);

if (r)

break;

}

busy = 0;

return r;

}

#else /* if ! defined WITH_THREAD */

/*

WARNING! ASYNCHRONOUSLY EXECUTING CODE!

This code is used for signal handling in python that isn't built

with WITH_THREAD.

Don't use this implementation when Py_AddPendingCalls() can happen

on a different thread!

There are two possible race conditions:

(1) nested asynchronous calls to Py_AddPendingCall()

(2) AddPendingCall() calls made while pending calls are being processed.

(1) is very unlikely because typically signal delivery

is blocked during signal handling. So it should be impossible.

(2) is a real possibility.

The current code is safe against (2), but not against (1).

The safety against (2) is derived from the fact that only one

thread is present, interrupted by signals, and that the critical

section is protected with the "busy" variable. On Windows, which

delivers SIGINT on a system thread, this does not hold and therefore

Windows really shouldn't use this version.

The two threads could theoretically wiggle around the "busy" variable.

*/

#define NPENDINGCALLS 32

static struct {

int (*func)(void *);

void *arg;

} pendingcalls[NPENDINGCALLS];

static volatile int pendingfirst = 0;

static volatile int pendinglast = 0;

static _Py_atomic_int pendingcalls_to_do = {0};

int

Py_AddPendingCall(int (*func)(void *), void *arg)

{

static volatile int busy = 0;

int i, j;

/* XXX Begin critical section */

if (busy)

return -1;

busy = 1;

i = pendinglast;

j = (i + 1) % NPENDINGCALLS;

if (j == pendingfirst) {

busy = 0;

return -1; /* Queue full */

}

pendingcalls[i].func = func;

pendingcalls[i].arg = arg;

pendinglast = j;

SIGNAL_PENDING_CALLS();

busy = 0;

/* XXX End critical section */

return 0;

}

int

Py_MakePendingCalls(void)

{

static int busy = 0;

if (busy)

return 0;

busy = 1;

UNSIGNAL_PENDING_CALLS();

for (;;) {

int i;

int (*func)(void *);

void *arg;

i = pendingfirst;

if (i == pendinglast)

break; /* Queue empty */

func = pendingcalls[i].func;

arg = pendingcalls[i].arg;

pendingfirst = (i + 1) % NPENDINGCALLS;

if (func(arg) < 0) {

busy = 0;

SIGNAL_PENDING_CALLS(); /* We're not done yet */

return -1;

}

}

busy = 0;

return 0;

}

#endif /* WITH_THREAD */

/* The interpreter's recursion limit */

#ifndef Py_DEFAULT_RECURSION_LIMIT

#define Py_DEFAULT_RECURSION_LIMIT 1000

#endif

static int recursion_limit = Py_DEFAULT_RECURSION_LIMIT;

int _Py_CheckRecursionLimit = Py_DEFAULT_RECURSION_LIMIT;

int

Py_GetRecursionLimit(void)

{

return recursion_limit;

}

void

Py_SetRecursionLimit(int new_limit)

{

recursion_limit = new_limit;

_Py_CheckRecursionLimit = recursion_limit;

}

/* the macro Py_EnterRecursiveCall() only calls _Py_CheckRecursiveCall()

if the recursion_depth reaches _Py_CheckRecursionLimit.

If USE_STACKCHECK, the macro decrements _Py_CheckRecursionLimit

to guarantee that _Py_CheckRecursiveCall() is regularly called.

Without USE_STACKCHECK, there is no need for this. */

int

_Py_CheckRecursiveCall(const char *where)

{

PyThreadState *tstate = PyThreadState_GET();

#ifdef USE_STACKCHECK

if (PyOS_CheckStack()) {

--tstate->recursion_depth;

PyErr_SetString(PyExc_MemoryError, "Stack overflow");

return -1;

}

#endif

_Py_CheckRecursionLimit = recursion_limit;

if (tstate->recursion_critical)

/* Somebody asked that we don't check for recursion. */

return 0;

if (tstate->overflowed) {

if (tstate->recursion_depth > recursion_limit + 50) {

/* Overflowing while handling an overflow. Give up. */

Py_FatalError("Cannot recover from stack overflow.");

}

return 0;

}

if (tstate->recursion_depth > recursion_limit) {

--tstate->recursion_depth;

tstate->overflowed = 1;

PyErr_Format(PyExc_RecursionError,

"maximum recursion depth exceeded%s",

where);

return -1;

}

return 0;

}

/* Status code for main loop (reason for stack unwind) */

enum why_code {

WHY_NOT = 0x0001, /* No error */

WHY_EXCEPTION = 0x0002, /* Exception occurred */

WHY_RETURN = 0x0008, /* 'return' statement */

WHY_BREAK = 0x0010, /* 'break' statement */

WHY_CONTINUE = 0x0020, /* 'continue' statement */

WHY_YIELD = 0x0040, /* 'yield' operator */

WHY_SILENCED = 0x0080 /* Exception silenced by 'with' */

};

static void save_exc_state(PyThreadState *, PyFrameObject *);

static void swap_exc_state(PyThreadState *, PyFrameObject *);

static void restore_and_clear_exc_state(PyThreadState *, PyFrameObject *);

static int do_raise(PyObject *, PyObject *);

static int unpack_iterable(PyObject *, int, int, PyObject **);

/* Records whether tracing is on for any thread. Counts the number of

threads for which tstate->c_tracefunc is non-NULL, so if the value

is 0, we know we don't have to check this thread's c_tracefunc.

This speeds up the if statement in PyEval_EvalFrameEx() after

fast_next_opcode*/

static int _Py_TracingPossible = 0;

PyObject *

PyEval_EvalCode(PyObject *co, PyObject *globals, PyObject *locals)

{

return PyEval_EvalCodeEx(co,

globals, locals,

(PyObject **)NULL, 0,

(PyObject **)NULL, 0,

(PyObject **)NULL, 0,

NULL, NULL);

}

/* Interpreter main loop */

PyObject *

PyEval_EvalFrame(PyFrameObject *f) {

/* This is for backward compatibility with extension modules that

used this API; core interpreter code should call

PyEval_EvalFrameEx() */

return PyEval_EvalFrameEx(f, 0);

}

PyObject *

PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)

{

PyThreadState *tstate = PyThreadState_GET();

return tstate->interp->eval_frame(f, throwflag);

}

PyObject *

_PyEval_EvalFrameDefault(PyFrameObject *f, int throwflag)

{

#ifdef DXPAIRS

int lastopcode = 0;

#endif

PyObject **stack_pointer; /* Next free slot in value stack */

const unsigned short *next_instr;

int opcode; /* Current opcode */

int oparg; /* Current opcode argument, if any */

enum why_code why; /* Reason for block stack unwind */

PyObject **fastlocals, **freevars;

PyObject *retval = NULL; /* Return value */

PyThreadState *tstate = PyThreadState_GET();

PyCodeObject *co;

/* when tracing we set things up so that

not (instr_lb <= current_bytecode_offset < instr_ub)

is true when the line being executed has changed. The

initial values are such as to make this false the first

time it is tested. */

int instr_ub = -1, instr_lb = 0, instr_prev = -1;

const unsigned short *first_instr;

PyObject *names;

PyObject *consts;

#ifdef LLTRACE

_Py_IDENTIFIER(__ltrace__);

#endif

/* Computed GOTOs, or

the-optimization-commonly-but-improperly-known-as-"threaded code"

using gcc's labels-as-values extension

(http://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html).

The traditional bytecode evaluation loop uses a "switch" statement, which

decent compilers will optimize as a single indirect branch instruction

combined with a lookup table of jump addresses. However, since the

indirect jump instruction is shared by all opcodes, the CPU will have a

hard time making the right prediction for where to jump next (actually,

it will be always wrong except in the uncommon case of a sequence of

several identical opcodes).

"Threaded code" in contrast, uses an explicit jump table and an explicit

indirect jump instruction at the end of each opcode. Since the jump

instruction is at a different address for each opcode, the CPU will make a

separate prediction for each of these instructions, which is equivalent to

predicting the second opcode of each opcode pair. These predictions have

a much better chance to turn out valid, especially in small bytecode loops.

A mispredicted branch on a modern CPU flushes the whole pipeline and

can cost several CPU cycles (depending on the pipeline depth),

and potentially many more instructions (depending on the pipeline width).

A correctly predicted branch, however, is nearly free.

At the time of this writing, the "threaded code" version is up to 15-20%

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

-fno-crossjumping).

*/

#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

#endif

#else

#if defined(USE_COMPUTED_GOTOS) && USE_COMPUTED_GOTOS

#error "Computed gotos are not supported on this compiler."

#endif

#undef USE_COMPUTED_GOTOS

#define USE_COMPUTED_GOTOS 0

#endif

#if USE_COMPUTED_GOTOS

/* Import the static jump table */

#include "opcode_targets.h"

#define TARGET(op) \

TARGET_##op: \

case op:

#define DISPATCH() \

{ \

if (!_Py_atomic_load_relaxed(&eval_breaker)) { \

FAST_DISPATCH(); \

} \

continue; \

}

#ifdef LLTRACE

#define FAST_DISPATCH() \

{ \

if (!lltrace && !_Py_TracingPossible) { \

f->f_lasti = INSTR_OFFSET(); \

NEXTOPARG(); \

goto *opcode_targets[opcode]; \

} \

goto fast_next_opcode; \

}

#else

#define FAST_DISPATCH() \

{ \

if (!_Py_TracingPossible) { \

f->f_lasti = INSTR_OFFSET(); \

NEXTOPARG(); \

goto *opcode_targets[opcode]; \

} \

goto fast_next_opcode; \

}

#endif

#else

#define TARGET(op) \

case op:

#define DISPATCH() continue

#define FAST_DISPATCH() goto fast_next_opcode

#endif

/* Tuple access macros */

#ifndef Py_DEBUG

#define GETITEM(v, i) PyTuple_GET_ITEM((PyTupleObject *)(v), (i))

#else

#define GETITEM(v, i) PyTuple_GetItem((v), (i))

#endif

#ifdef WITH_TSC

/* Use Pentium timestamp counter to mark certain events:

inst0 -- beginning of switch statement for opcode dispatch

inst1 -- end of switch statement (may be skipped)

loop0 -- the top of the mainloop

loop1 -- place where control returns again to top of mainloop

(may be skipped)

intr1 -- beginning of long interruption

intr2 -- end of long interruption

Many opcodes call out to helper C functions. In some cases, the

time in those functions should be counted towards the time for the

opcode, but not in all cases. For example, a CALL_FUNCTION opcode

calls another Python function; there's no point in charge all the

bytecode executed by the called function to the caller.

It's hard to make a useful judgement statically. In the presence

of operator overloading, it's impossible to tell if a call will

execute new Python code or not.

It's a case-by-case judgement. I'll use intr1 for the following

cases:

IMPORT_STAR

IMPORT_FROM

CALL_FUNCTION (and friends)

*/

uint64 inst0, inst1, loop0, loop1, intr0 = 0, intr1 = 0;

int ticked = 0;

READ_TIMESTAMP(inst0);

READ_TIMESTAMP(inst1);

READ_TIMESTAMP(loop0);

READ_TIMESTAMP(loop1);

/* shut up the compiler */

opcode = 0;

#endif

/* Code access macros */

#ifdef WORDS_BIGENDIAN

#define OPCODE(word) ((word) >> 8)

#define OPARG(word) ((word) & 255)

#else

#define OPCODE(word) ((word) & 255)

#define OPARG(word) ((word) >> 8)

#endif

/* The integer overflow is checked by an assertion below. */

#define INSTR_OFFSET() (2*(int)(next_instr - first_instr))

#define NEXTOPARG() do { \

unsigned short word = *next_instr; \

opcode = OPCODE(word); \

oparg = OPARG(word); \

next_instr++; \

} while (0)