Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit51fc725

Browse files
authored
gh-104584: Baby steps towards generating and executing traces (#105924)
Added a new, experimental, tracing optimizer and interpreter (a.k.a. "tier 2"). This currently pessimizes, so don't use yet -- this is infrastructure so we can experiment with optimizing passes. To enable it, pass ``-Xuops`` or set ``PYTHONUOPS=1``. To get debug output, set ``PYTHONUOPSDEBUG=N`` where ``N`` is a debug level (0-4, where 0 is no debug output and 4 is excessively verbose).All of this code is likely to change dramatically before the 3.13 feature freeze. But this is a first step.
1 parentd3af83b commit51fc725

21 files changed

+2559
-305
lines changed

‎.gitattributes

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -86,6 +86,7 @@ Parser/token.c generated
8686
Programs/test_frozenmain.hgenerated
8787
Python/Python-ast.cgenerated
8888
Python/generated_cases.c.hgenerated
89+
Python/executor_cases.c.hgenerated
8990
Python/opcode_targets.hgenerated
9091
Python/stdlib_module_names.hgenerated
9192
Tools/peg_generator/pegen/grammar_parser.pygenerated

‎Include/cpython/optimizer.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -45,6 +45,7 @@ extern _PyOptimizerObject _PyOptimizer_Default;
4545

4646
/* For testing */
4747
PyAPI_FUNC(PyObject*)PyUnstable_Optimizer_NewCounter(void);
48+
PyAPI_FUNC(PyObject*)PyUnstable_Optimizer_NewUOpOptimizer(void);
4849

4950
#defineOPTIMIZER_BITS_IN_COUNTER 4
5051

‎Include/internal/pycore_uops.h

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
#ifndefPy_INTERNAL_UOPS_H
2+
#definePy_INTERNAL_UOPS_H
3+
#ifdef__cplusplus
4+
extern"C" {
5+
#endif
6+
7+
#ifndefPy_BUILD_CORE
8+
# error "this header requires Py_BUILD_CORE define"
9+
#endif
10+
11+
#define_Py_UOP_MAX_TRACE_LENGTH 16
12+
13+
typedefstruct {
14+
intopcode;
15+
uint64_toperand;// Sometimes oparg, sometimes a cache entry
16+
}_PyUOpInstruction;
17+
18+
typedefstruct {
19+
_PyExecutorObjectbase;
20+
_PyUOpInstructiontrace[_Py_UOP_MAX_TRACE_LENGTH];// TODO: variable length
21+
}_PyUOpExecutorObject;
22+
23+
_PyInterpreterFrame*_PyUopExecute(
24+
_PyExecutorObject*executor,
25+
_PyInterpreterFrame*frame,
26+
PyObject**stack_pointer);
27+
28+
#ifdef__cplusplus
29+
}
30+
#endif
31+
#endif/* !Py_INTERNAL_UOPS_H */

‎Include/pystats.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -71,6 +71,9 @@ typedef struct _object_stats {
7171
uint64_ttype_cache_dunder_misses;
7272
uint64_ttype_cache_collisions;
7373
uint64_toptimization_attempts;
74+
uint64_toptimization_traces_created;
75+
uint64_toptimization_traces_executed;
76+
uint64_toptimization_uops_executed;
7477
}ObjectStats;
7578

7679
typedefstruct_stats {

‎Makefile.pre.in

Lines changed: 9 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -1542,19 +1542,9 @@ regen-opcode-targets:
15421542

15431543
.PHONY: regen-cases
15441544
regen-cases:
1545-
# Regenerate Python/generated_cases.c.h
1546-
# and Python/opcode_metadata.h
1547-
# from Python/bytecodes.c
1548-
# using Tools/cases_generator/generate_cases.py
1545+
# Regenerate various files from Python/bytecodes.c
15491546
PYTHONPATH=$(srcdir)/Tools/cases_generator \
1550-
$(PYTHON_FOR_REGEN) \
1551-
$(srcdir)/Tools/cases_generator/generate_cases.py \
1552-
--emit-line-directives \
1553-
-o $(srcdir)/Python/generated_cases.c.h.new \
1554-
-m $(srcdir)/Python/opcode_metadata.h.new \
1555-
$(srcdir)/Python/bytecodes.c
1556-
$(UPDATE_FILE) $(srcdir)/Python/generated_cases.c.h $(srcdir)/Python/generated_cases.c.h.new
1557-
$(UPDATE_FILE) $(srcdir)/Python/opcode_metadata.h $(srcdir)/Python/opcode_metadata.h.new
1547+
$(PYTHON_FOR_REGEN) $(srcdir)/Tools/cases_generator/generate_cases.py -l
15581548

15591549
Python/compile.o: $(srcdir)/Python/opcode_metadata.h
15601550

@@ -1565,6 +1555,13 @@ Python/ceval.o: \
15651555
$(srcdir)/Python/opcode_metadata.h \
15661556
$(srcdir)/Python/opcode_targets.h
15671557

1558+
Python/flowgraph.o: \
1559+
$(srcdir)/Python/opcode_metadata.h
1560+
1561+
Python/optimizer.o: \
1562+
$(srcdir)/Python/executor_cases.c.h \
1563+
$(srcdir)/Python/opcode_metadata.h
1564+
15681565
Python/frozen.o: $(FROZEN_FILES_OUT)
15691566

15701567
# Generate DTrace probe macros, then rename them (PYTHON_ -> PyDTrace_) to
Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1 @@
1+
Added a new, experimental, tracing optimizer and interpreter (a.k.a. "tier 2"). This currently pessimizes, so don't use yet -- this is infrastructure so we can experiment with optimizing passes. To enable it, pass ``-Xuops`` or set ``PYTHONUOPS=1``. To get debug output, set ``PYTHONUOPSDEBUG=N`` where ``N`` is a debug level (0-4, where 0 is no debug output and 4 is excessively verbose).

‎Modules/_testinternalcapi.c

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -830,6 +830,12 @@ get_counter_optimizer(PyObject *self, PyObject *arg)
830830
returnPyUnstable_Optimizer_NewCounter();
831831
}
832832

833+
staticPyObject*
834+
get_uop_optimizer(PyObject*self,PyObject*arg)
835+
{
836+
returnPyUnstable_Optimizer_NewUOpOptimizer();
837+
}
838+
833839
staticPyObject*
834840
set_optimizer(PyObject*self,PyObject*opt)
835841
{
@@ -994,6 +1000,7 @@ static PyMethodDef module_functions[] = {
9941000
{"get_optimizer",get_optimizer,METH_NOARGS,NULL},
9951001
{"set_optimizer",set_optimizer,METH_O,NULL},
9961002
{"get_counter_optimizer",get_counter_optimizer,METH_NOARGS,NULL},
1003+
{"get_uop_optimizer",get_uop_optimizer,METH_NOARGS,NULL},
9971004
{"pending_threadfunc",_PyCFunction_CAST(pending_threadfunc),
9981005
METH_VARARGS |METH_KEYWORDS},
9991006
// {"pending_fd_identify", pending_fd_identify, METH_VARARGS, NULL},

‎Python/bytecodes.c

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -52,8 +52,6 @@
5252
#definefamily(name, ...) static int family_##name
5353
#definepseudo(name) static int pseudo_##name
5454

55-
typedefPyObject*(*convertion_func_ptr)(PyObject*);
56-
5755
// Dummy variables for stack effects.
5856
staticPyObject*value,*value1,*value2,*left,*right,*res,*sum,*prod,*sub;
5957
staticPyObject*container,*start,*stop,*v,*lhs,*rhs,*res2;
@@ -2182,7 +2180,7 @@ dummy_func(
21822180
frame=executor->execute(executor,frame,stack_pointer);
21832181
if (frame==NULL) {
21842182
frame=cframe.current_frame;
2185-
gotoerror;
2183+
gotoresume_with_error;
21862184
}
21872185
gotoresume_frame;
21882186
}

‎Python/ceval.c

Lines changed: 129 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -22,6 +22,7 @@
2222
#include"pycore_sysmodule.h"// _PySys_Audit()
2323
#include"pycore_tuple.h"// _PyTuple_ITEMS()
2424
#include"pycore_typeobject.h"// _PySuper_Lookup()
25+
#include"pycore_uops.h"// _PyUOpExecutorObject
2526
#include"pycore_emscripten_signal.h"// _Py_CHECK_EMSCRIPTEN_SIGNALS
2627

2728
#include"pycore_dict.h"
@@ -223,14 +224,6 @@ _PyEvalFramePushAndInit_Ex(PyThreadState *tstate, PyFunctionObject *func,
223224
staticvoid
224225
_PyEvalFrameClearAndPop(PyThreadState*tstate,_PyInterpreterFrame*frame);
225226

226-
typedefPyObject*(*convertion_func_ptr)(PyObject*);
227-
228-
staticconstconvertion_func_ptrCONVERSION_FUNCTIONS[4]= {
229-
[FVC_STR]=PyObject_Str,
230-
[FVC_REPR]=PyObject_Repr,
231-
[FVC_ASCII]=PyObject_ASCII
232-
};
233-
234227
#defineUNBOUNDLOCAL_ERROR_MSG \
235228
"cannot access local variable '%s' where it is not associated with a value"
236229
#defineUNBOUNDFREE_ERROR_MSG \
@@ -2771,3 +2764,131 @@ void Py_LeaveRecursiveCall(void)
27712764
{
27722765
_Py_LeaveRecursiveCall();
27732766
}
2767+
2768+
///////////////////// Experimental UOp Interpreter /////////////////////
2769+
2770+
// UPDATE_MISS_STATS (called by DEOPT_IF) uses next_instr
2771+
// TODO: Make it do something useful
2772+
#undef UPDATE_MISS_STATS
2773+
#defineUPDATE_MISS_STATS(INSTNAME) ((void)0)
2774+
2775+
_PyInterpreterFrame*
2776+
_PyUopExecute(_PyExecutorObject*executor,_PyInterpreterFrame*frame,PyObject**stack_pointer)
2777+
{
2778+
#ifdefLLTRACE
2779+
char*uop_debug=Py_GETENV("PYTHONUOPSDEBUG");
2780+
intlltrace=0;
2781+
if (uop_debug!=NULL&&*uop_debug >='0') {
2782+
lltrace=*uop_debug-'0';// TODO: Parse an int and all that
2783+
}
2784+
if (lltrace >=2) {
2785+
PyCodeObject*code=_PyFrame_GetCode(frame);
2786+
_Py_CODEUNIT*instr=frame->prev_instr+1;
2787+
fprintf(stderr,
2788+
"Entering _PyUopExecute for %s (%s:%d) at offset %ld\n",
2789+
PyUnicode_AsUTF8(code->co_qualname),
2790+
PyUnicode_AsUTF8(code->co_filename),
2791+
code->co_firstlineno,
2792+
(long)(instr- (_Py_CODEUNIT*)code->co_code_adaptive));
2793+
}
2794+
#endif
2795+
2796+
PyThreadState*tstate=_PyThreadState_GET();
2797+
_PyUOpExecutorObject*self= (_PyUOpExecutorObject*)executor;
2798+
2799+
// Equivalent to CHECK_EVAL_BREAKER()
2800+
_Py_CHECK_EMSCRIPTEN_SIGNALS_PERIODICALLY();
2801+
if (_Py_atomic_load_relaxed_int32(&tstate->interp->ceval.eval_breaker)) {
2802+
if (_Py_HandlePending(tstate)!=0) {
2803+
gotoerror;
2804+
}
2805+
}
2806+
2807+
OBJECT_STAT_INC(optimization_traces_executed);
2808+
_Py_CODEUNIT*ip_offset= (_Py_CODEUNIT*)_PyFrame_GetCode(frame)->co_code_adaptive-1;
2809+
intpc=0;
2810+
intopcode;
2811+
uint64_toperand;
2812+
intoparg;
2813+
for (;;) {
2814+
opcode=self->trace[pc].opcode;
2815+
operand=self->trace[pc].operand;
2816+
oparg= (int)operand;
2817+
#ifdefLLTRACE
2818+
if (lltrace >=3) {
2819+
constchar*opname=opcode<256 ?_PyOpcode_OpName[opcode] :"";
2820+
intstack_level= (int)(stack_pointer-_PyFrame_Stackbase(frame));
2821+
fprintf(stderr," uop %s %d, operand %"PRIu64", stack_level %d\n",
2822+
opname,opcode,operand,stack_level);
2823+
}
2824+
#endif
2825+
pc++;
2826+
OBJECT_STAT_INC(optimization_uops_executed);
2827+
switch (opcode) {
2828+
2829+
#undef ENABLE_SPECIALIZATION
2830+
#defineENABLE_SPECIALIZATION 0
2831+
#include"executor_cases.c.h"
2832+
2833+
caseSET_IP:
2834+
{
2835+
frame->prev_instr=ip_offset+oparg;
2836+
break;
2837+
}
2838+
2839+
caseEXIT_TRACE:
2840+
{
2841+
_PyFrame_SetStackPointer(frame,stack_pointer);
2842+
Py_DECREF(self);
2843+
returnframe;
2844+
}
2845+
2846+
default:
2847+
{
2848+
fprintf(stderr,"Unknown uop %d, operand %"PRIu64"\n",opcode,operand);
2849+
Py_FatalError("Unknown uop");
2850+
abort();// Unreachable
2851+
for (;;) {}
2852+
// Really unreachable
2853+
}
2854+
2855+
}
2856+
}
2857+
2858+
pop_4_error:
2859+
STACK_SHRINK(1);
2860+
pop_3_error:
2861+
STACK_SHRINK(1);
2862+
pop_2_error:
2863+
STACK_SHRINK(1);
2864+
pop_1_error:
2865+
STACK_SHRINK(1);
2866+
error:
2867+
// On ERROR_IF we return NULL as the frame.
2868+
// The caller recovers the frame from cframe.current_frame.
2869+
#ifdefLLTRACE
2870+
if (lltrace >=2) {
2871+
fprintf(stderr,"Error: [Opcode %d, operand %"PRIu64"]\n",opcode,operand);
2872+
}
2873+
#endif
2874+
_PyFrame_SetStackPointer(frame,stack_pointer);
2875+
Py_DECREF(self);
2876+
returnNULL;
2877+
2878+
PREDICTED(UNPACK_SEQUENCE)
2879+
PREDICTED(COMPARE_OP)
2880+
PREDICTED(LOAD_SUPER_ATTR)
2881+
PREDICTED(STORE_SUBSCR)
2882+
PREDICTED(BINARY_SUBSCR)
2883+
PREDICTED(BINARY_OP)
2884+
// On DEOPT_IF we just repeat the last instruction.
2885+
// This presumes nothing was popped from the stack (nor pushed).
2886+
#ifdefLLTRACE
2887+
if (lltrace >=2) {
2888+
fprintf(stderr,"DEOPT: [Opcode %d, operand %"PRIu64"]\n",opcode,operand);
2889+
}
2890+
#endif
2891+
_PyFrame_SetStackPointer(frame,stack_pointer);
2892+
Py_DECREF(self);
2893+
returnframe;
2894+
}

‎Python/ceval_macros.h

Lines changed: 9 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
// Macros needed by ceval.c and bytecodes.c
1+
// Macrosand other thingsneeded by ceval.c and bytecodes.c
22

33
/* Computed GOTOs, or
44
the-optimization-commonly-but-improperly-known-as-"threaded code"
@@ -339,3 +339,11 @@ do { \
339339
goto error; \
340340
} \
341341
} while (0);
342+
343+
typedefPyObject*(*convertion_func_ptr)(PyObject*);
344+
345+
staticconstconvertion_func_ptrCONVERSION_FUNCTIONS[4]= {
346+
[FVC_STR]=PyObject_Str,
347+
[FVC_REPR]=PyObject_Repr,
348+
[FVC_ASCII]=PyObject_ASCII
349+
};

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp