cpython/Python/optimizer_analysis.c at main · python/cpython · GitHub

History

831 lines (757 loc) · 26.4 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

#ifdef _Py_TIER2

* This file contains the support code for CPython's uops optimizer.

* It also performs some simple optimizations.

* It performs a traditional data-flow analysis[1] over the trace of uops.

* Using the information gained, it chooses to emit, or skip certain instructions

* if possible.

* [1] For information on data-flow analysis, please see

* https://clang.llvm.org/docs/DataFlowAnalysisIntro.html

* */

#include "Python.h"

#include "opcode.h"

#include "pycore_dict.h"

#include "pycore_interp.h"

#include "pycore_opcode_metadata.h"

#include "pycore_opcode_utils.h"

#include "pycore_pystate.h" // _PyInterpreterState_GET()

#include "pycore_pyatomic_ft_wrappers.h" // FT_ATOMIC_*

#include "pycore_tstate.h" // _PyThreadStateImpl

#include "pycore_uop_metadata.h"

#include "pycore_long.h"

#include "pycore_interpframe.h" // _PyFrame_GetCode

#include "pycore_optimizer.h"

#include "pycore_object.h"

#include "pycore_function.h"

#include "pycore_uop_ids.h"

#include "pycore_range.h"

#include "pycore_unicodeobject.h"

#include "pycore_ceval.h"

#include "pycore_floatobject.h"

#include "pycore_setobject.h"

#include "pycore_typeobject.h"

#include <stdarg.h>

#include <stdbool.h>

#include <stdint.h>

#include <stddef.h>

#ifdef Py_DEBUG

extern const char *_PyUOpName(int index);

extern void _PyUOpPrint(const _PyUOpInstruction *uop);

extern void _PyUOpSymPrint(JitOptRef ref);

static const char *const DEBUG_ENV = "PYTHON_OPT_DEBUG";

static inline int get_lltrace(void) {

char *uop_debug = Py_GETENV(DEBUG_ENV);

int lltrace = 0;

if (uop_debug != NULL && *uop_debug >= '0') {

lltrace = *uop_debug - '0'; // TODO: Parse an int and all that

}

return lltrace;

}

#define DPRINTF(level, ...) \

if (get_lltrace() >= (level)) { printf(__VA_ARGS__); }

static void

dump_abstract_stack(_Py_UOpsAbstractFrame *frame, JitOptRef *stack_pointer)

{

printf(" locals=[");

for (int i = 0 ; i < frame->locals_len; i++) {

if (i > 0) {

printf(", ");

}

_PyUOpSymPrint(frame->locals[i]);

}

printf("]\n");

if (stack_pointer < frame->stack) {

printf(" stack=%d\n", (int)(stack_pointer - frame->stack));

}

else {

printf(" stack=[");

for (JitOptRef *ptr = frame->stack; ptr < stack_pointer; ptr++) {

if (ptr != frame->stack) {

printf(", ");

}

_PyUOpSymPrint(*ptr);

}

printf("]\n");

}

fflush(stdout);

}

static void

dump_uop(JitOptContext *ctx, const char *label, int index,

const _PyUOpInstruction *instr, JitOptRef *stack_pointer)

{

if (get_lltrace() >= 3) {

printf("%4d %s: ", index, label);

_PyUOpPrint(instr);

printf("\n");

if (get_lltrace() >= 5 && ctx->frame->code != ((PyCodeObject *)&_Py_InitCleanup)) {

dump_abstract_stack(ctx->frame, stack_pointer);

}

static void

dump_uops(JitOptContext *ctx, const char *label,

_PyUOpInstruction *start, JitOptRef *stack_pointer)

{

int current_len = uop_buffer_length(&ctx->out_buffer);

int added_count = (int)(ctx->out_buffer.next - start);

for (int j = 0; j < added_count; j++) {

dump_uop(ctx, label, current_len - added_count + j, &start[j], stack_pointer);

}

#define DUMP_UOP dump_uop

#define DUMP_UOPS dump_uops

#else

#define DPRINTF(level, ...)

#define DUMP_UOP(ctx, label, index, instr, stack_pointer)

#define DUMP_UOPS(ctx, label, start, stack_pointer)

#endif

static int

get_mutations(PyObject* dict) {

assert(PyDict_CheckExact(dict));

PyDictObject *d = (PyDictObject *)dict;

uint64_t tag = FT_ATOMIC_LOAD_UINT64_RELAXED(d->_ma_watcher_tag);

return (tag >> DICT_MAX_WATCHERS) & ((1 << DICT_WATCHED_MUTATION_BITS) - 1);

}

static void

increment_mutations(PyObject* dict) {

assert(PyDict_CheckExact(dict));

PyDictObject *d = (PyDictObject *)dict;

FT_ATOMIC_ADD_UINT64(d->_ma_watcher_tag, 1ULL << DICT_MAX_WATCHERS);

}

/* The first two dict watcher IDs are reserved for CPython,

* so we don't need to check that they haven't been used */

#define BUILTINS_WATCHER_ID 0

#define GLOBALS_WATCHER_ID 1

#define TYPE_WATCHER_ID 0

static int

globals_watcher_callback(PyDict_WatchEvent event, PyObject* dict,

PyObject* key, PyObject* new_value)

{

RARE_EVENT_STAT_INC(watched_globals_modification);

assert(get_mutations(dict) < _Py_MAX_ALLOWED_GLOBALS_MODIFICATIONS);

_Py_Executors_InvalidateDependency(_PyInterpreterState_GET(), dict, 1);

increment_mutations(dict);

PyDict_Unwatch(GLOBALS_WATCHER_ID, dict);

return 0;

}

static int

type_watcher_callback(PyTypeObject* type)

{

_Py_Executors_InvalidateDependency(_PyInterpreterState_GET(), type, 1);

PyType_Unwatch(TYPE_WATCHER_ID, (PyObject *)type);

return 0;

}

static int

_setup_optimizer_watchers(void *Py_UNUSED(arg))

{

PyInterpreterState *interp = _PyInterpreterState_GET();

FT_ATOMIC_STORE_PTR_RELEASE(

interp->dict_state.watchers[GLOBALS_WATCHER_ID],

globals_watcher_callback);

interp->type_watchers[TYPE_WATCHER_ID] = type_watcher_callback;

return 0;

}

static void

watch_type(PyTypeObject *type, _PyBloomFilter *filter)

{

if (_Py_IsImmortal(type) && (type->tp_flags & Py_TPFLAGS_IMMUTABLETYPE)) {

return;

}

PyType_Watch(TYPE_WATCHER_ID, (PyObject *)type);

_Py_BloomFilter_Add(filter, type);

}

static PyObject *

convert_global_to_const(_PyUOpInstruction *inst, PyObject *obj)

{

assert(inst->opcode == _LOAD_GLOBAL_MODULE || inst->opcode == _LOAD_GLOBAL_BUILTINS || inst->opcode == _LOAD_ATTR_MODULE);

assert(PyDict_CheckExact(obj));

PyDictObject *dict = (PyDictObject *)obj;

assert(dict->ma_keys->dk_kind == DICT_KEYS_UNICODE);

PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dict->ma_keys);

int64_t index = inst->operand1;

assert(index <= UINT16_MAX);

if ((int)index >= dict->ma_keys->dk_nentries) {

return NULL;

}

PyDictKeysObject *keys = dict->ma_keys;

if (keys->dk_version != inst->operand0) {

return NULL;

}

PyObject *res = entries[index].me_value;

if (res == NULL) {

return NULL;

}

if (_Py_IsImmortal(res)) {

inst->opcode = _LOAD_CONST_INLINE_BORROW;

} else {

inst->opcode = _LOAD_CONST_INLINE;

}

inst->operand0 = (uint64_t)res;

return res;

}

static bool

incorrect_keys(PyObject *obj, uint32_t version)

{

if (!PyDict_CheckExact(obj)) {

return true;

}

PyDictObject *dict = (PyDictObject *)obj;

return dict->ma_keys->dk_version != version;

}

#define STACK_LEVEL() ((int)(stack_pointer - ctx->frame->stack))

#define STACK_SIZE() ((int)(ctx->frame->stack_len))

static inline int

is_terminator_uop(const _PyUOpInstruction *uop)

{

int opcode = uop->opcode;

return (

opcode == _EXIT_TRACE ||

opcode == _JUMP_TO_TOP ||

opcode == _DYNAMIC_EXIT ||

opcode == _DEOPT

);

}

#define CURRENT_FRAME_IS_INIT_SHIM() (ctx->frame->code == ((PyCodeObject *)&_Py_InitCleanup))

#define GETLOCAL(idx) ((ctx->frame->locals[idx]))

#define REPLACE_OP(INST, OP, ARG, OPERAND) \

(INST)->opcode = OP; \

(INST)->oparg = ARG; \

(INST)->operand0 = OPERAND;

#define ADD_OP(OP, ARG, OPERAND) add_op(ctx, this_instr, (OP), (ARG), (OPERAND))

static inline void

add_op(JitOptContext *ctx, _PyUOpInstruction *this_instr,

uint16_t opcode, uint16_t oparg, uintptr_t operand0)

{

_PyUOpInstruction *out = ctx->out_buffer.next;

assert(out < ctx->out_buffer.end);

out->opcode = (opcode);

out->format = this_instr->format;

out->oparg = (oparg);

out->target = this_instr->target;

out->operand0 = (operand0);

out->operand1 = this_instr->operand1;

#ifdef Py_STATS

out->fitness = this_instr->fitness;

#endif

ctx->out_buffer.next++;

}

/* Shortened forms for convenience, used in optimizer_bytecodes.c */

#define sym_is_not_null _Py_uop_sym_is_not_null

#define sym_is_const _Py_uop_sym_is_const

#define sym_is_safe_const _Py_uop_sym_is_safe_const

#define sym_is_not_container _Py_uop_sym_is_not_container

#define sym_get_const _Py_uop_sym_get_const

#define sym_new_const_steal _Py_uop_sym_new_const_steal

#define sym_get_const_as_stackref _Py_uop_sym_get_const_as_stackref

#define sym_new_unknown _Py_uop_sym_new_unknown

#define sym_new_not_null _Py_uop_sym_new_not_null

#define sym_new_type _Py_uop_sym_new_type

#define sym_is_null _Py_uop_sym_is_null

#define sym_new_const _Py_uop_sym_new_const

#define sym_new_null _Py_uop_sym_new_null

#define sym_has_type _Py_uop_sym_has_type

#define sym_get_type _Py_uop_sym_get_type

#define sym_get_probable_type _Py_uop_sym_get_probable_type

#define sym_matches_type _Py_uop_sym_matches_type

#define sym_matches_type_version _Py_uop_sym_matches_type_version

#define sym_get_type_version _Py_uop_sym_get_type_version

#define sym_set_null(SYM) _Py_uop_sym_set_null(ctx, SYM)

#define sym_set_non_null(SYM) _Py_uop_sym_set_non_null(ctx, SYM)

#define sym_set_type(SYM, TYPE) _Py_uop_sym_set_type(ctx, SYM, TYPE)

#define sym_set_type_version(SYM, VERSION) _Py_uop_sym_set_type_version(ctx, SYM, VERSION)

#define sym_set_const(SYM, CNST) _Py_uop_sym_set_const(ctx, SYM, CNST)

#define sym_set_compact_int(SYM) _Py_uop_sym_set_compact_int(ctx, SYM)

#define sym_is_bottom _Py_uop_sym_is_bottom

#define sym_truthiness _Py_uop_sym_truthiness

#define frame_new _Py_uop_frame_new

#define frame_new_from_symbol _Py_uop_frame_new_from_symbol

#define frame_pop _Py_uop_frame_pop

#define sym_new_tuple _Py_uop_sym_new_tuple

#define sym_tuple_getitem _Py_uop_sym_tuple_getitem

#define sym_tuple_length _Py_uop_sym_tuple_length

#define sym_is_immortal _Py_uop_symbol_is_immortal

#define sym_is_compact_int _Py_uop_sym_is_compact_int

#define sym_new_compact_int _Py_uop_sym_new_compact_int

#define sym_new_truthiness _Py_uop_sym_new_truthiness

#define sym_new_predicate _Py_uop_sym_new_predicate

#define sym_apply_predicate_narrowing _Py_uop_sym_apply_predicate_narrowing

#define sym_set_recorded_type(SYM, TYPE) _Py_uop_sym_set_recorded_type(ctx, SYM, TYPE)

#define sym_set_recorded_value(SYM, VAL) _Py_uop_sym_set_recorded_value(ctx, SYM, VAL)

#define sym_set_recorded_gen_func(SYM, VAL) _Py_uop_sym_set_recorded_gen_func(ctx, SYM, VAL)

#define sym_get_probable_func_code _Py_uop_sym_get_probable_func_code

#define sym_get_probable_value _Py_uop_sym_get_probable_value

#define sym_set_stack_depth(DEPTH, SP) _Py_uop_sym_set_stack_depth(ctx, DEPTH, SP)

/* Comparison oparg masks */

#define COMPARE_LT_MASK 2

#define COMPARE_GT_MASK 4

#define COMPARE_EQ_MASK 8

#define JUMP_TO_LABEL(label) goto label;

static int

check_stack_bounds(JitOptContext *ctx, JitOptRef *stack_pointer, int offset, int opcode)

{

int stack_level = (int)(stack_pointer + (offset) - ctx->frame->stack);

int should_check = !CURRENT_FRAME_IS_INIT_SHIM() ||

(opcode == _RETURN_VALUE) ||

(opcode == _RETURN_GENERATOR) ||

(opcode == _YIELD_VALUE);

if (should_check && (stack_level < 0 || stack_level > STACK_SIZE() + MAX_CACHED_REGISTER)) {

ctx->contradiction = true;

ctx->done = true;

return 1;

}

return 0;

}

#define CHECK_STACK_BOUNDS(offset) \

if (check_stack_bounds(ctx, stack_pointer, offset, opcode)) { \

break; \

} \

static int

optimize_to_bool(

_PyUOpInstruction *this_instr,

JitOptContext *ctx,

JitOptRef value,

JitOptRef *result_ptr,

uint16_t prefix, uint16_t suffix)

{

if (sym_matches_type(value, &PyBool_Type)) {

ADD_OP(_NOP, 0, 0);

*result_ptr = value;

return 1;

}

int truthiness = sym_truthiness(ctx, value);

if (truthiness >= 0) {

PyObject *load = truthiness ? Py_True : Py_False;

if (prefix != _NOP) {

ADD_OP(prefix, 0, 0);

}

ADD_OP(_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)load);

if (suffix != _NOP) {

ADD_OP(suffix, 2, 0);

}

*result_ptr = sym_new_const(ctx, load);

return 1;

}

return 0;

}

static void

optimize_dict_known_hash(

JitOptContext *ctx, _PyBloomFilter *dependencies, _PyUOpInstruction *this_instr,

PyObject *sub, uint16_t opcode)

{

if (PyUnicode_CheckExact(sub) || PyLong_CheckExact(sub) || PyBytes_CheckExact(sub)

|| PyFloat_CheckExact(sub) || PyComplex_CheckExact(sub)) {

// PyObject_Hash can't fail on these types

ADD_OP(opcode, 0, PyObject_Hash(sub));

}

else if (PyTuple_CheckExact(sub)) {

// only use known hash variant when hash of tuple is already computed

// since computing it can call arbitrary code

Py_hash_t hash = ((PyTupleObject *)sub)->ob_hash;

if (hash != -1) {

ADD_OP(opcode, 0, hash);

}

else if (Py_TYPE(sub)->tp_hash == PyBaseObject_Type.tp_hash) {

// for user-defined objects which don't override tp_hash

Py_hash_t hash = PyObject_Hash(sub);

ADD_OP(opcode, 0, hash);

watch_type(Py_TYPE(sub), dependencies);

}

static void

eliminate_pop_guard(_PyUOpInstruction *this_instr, JitOptContext *ctx, bool exit)

{

ADD_OP(_POP_TOP, 0, 0);

if (exit) {

REPLACE_OP((this_instr+1), _EXIT_TRACE, 0, 0);

this_instr[1].target = this_instr->target;

}

static JitOptRef

lookup_attr(JitOptContext *ctx, _PyBloomFilter *dependencies, _PyUOpInstruction *this_instr,

PyTypeObject *type, PyObject *name,

uint16_t prefix, uint16_t suffix)

{

// The cached value may be dead, so we need to do the lookup again... :(

if (type && PyType_Check(type)) {

PyObject *lookup = _PyType_Lookup(type, name);

if (lookup) {

bool immortal = _Py_IsImmortal(lookup) || (type->tp_flags & Py_TPFLAGS_IMMUTABLETYPE);

if (prefix != _NOP) {

ADD_OP(prefix, 0, 0);

}

ADD_OP(immortal ? _LOAD_CONST_INLINE_BORROW : _LOAD_CONST_INLINE,

0, (uintptr_t)lookup);

if (suffix != _NOP) {

ADD_OP(suffix, 2, 0);

}

if ((type->tp_flags & Py_TPFLAGS_IMMUTABLETYPE) == 0) {

watch_type(type, dependencies);

}

return sym_new_const(ctx, lookup);

}

return sym_new_not_null(ctx);

}

static void

optimize_pop_top(JitOptContext *ctx, _PyUOpInstruction *this_instr, JitOptRef value)

{

PyTypeObject *typ = sym_get_type(value);

if (PyJitRef_IsBorrowed(value) ||

sym_is_immortal(PyJitRef_Unwrap(value)) ||

sym_is_null(value)) {

ADD_OP(_POP_TOP_NOP, 0, 0);

}

else if (typ == &PyLong_Type) {

ADD_OP(_POP_TOP_INT, 0, 0);

}

else if (typ == &PyFloat_Type) {

ADD_OP(_POP_TOP_FLOAT, 0, 0);

}

else if (typ == &PyUnicode_Type) {

ADD_OP(_POP_TOP_UNICODE, 0, 0);

}

else {

ADD_OP(_POP_TOP, 0, 0);

}

/* Look up name via super (normal case from supercheck where

su_obj_type = Py_TYPE(obj)). */

static JitOptRef

lookup_super_attr(JitOptContext *ctx, _PyBloomFilter *dependencies,

_PyUOpInstruction *this_instr,

PyTypeObject *su_type, PyTypeObject *obj_type,

PyObject *name,

uint16_t immortal, uint16_t mortal, uint16_t suffix)

{

if (su_type == NULL || obj_type == NULL) {

return sym_new_not_null(ctx);

}

/* Normal case: obj_type must be a subtype of su_type */

if (!PyType_IsSubtype(obj_type, su_type)) {

return sym_new_not_null(ctx);

}

PyObject *lookup = _PySuper_LookupDescr(su_type, obj_type, name);

if (lookup == NULL) {

if (PyErr_Occurred()) {

PyErr_Clear();

}

return sym_new_not_null(ctx);

}

if ((Py_TYPE(lookup)->tp_flags & Py_TPFLAGS_METHOD_DESCRIPTOR) == 0) {

Py_DECREF(lookup);

return sym_new_not_null(ctx);

}

int opcode = mortal;

if (_Py_IsImmortal(lookup) || (obj_type->tp_flags & Py_TPFLAGS_IMMUTABLETYPE)) {

opcode = immortal;

}

ADD_OP(_SWAP, 3, 0);

ADD_OP(_POP_TOP, 0, 0);

ADD_OP(opcode, 0, (uintptr_t)lookup);

if (suffix != _NOP) {

ADD_OP(suffix, 2, 0);

}

// if obj_type is immutable, then all its superclasses are immutable

if ((obj_type->tp_flags & Py_TPFLAGS_IMMUTABLETYPE) == 0) {

watch_type(su_type, dependencies);

watch_type(obj_type, dependencies);

}

return sym_new_const_steal(ctx, lookup);

}

static

PyCodeObject *

get_current_code_object(JitOptContext *ctx)

{

return (PyCodeObject *)ctx->frame->code;

}

static PyObject *

get_co_name(JitOptContext *ctx, int index)

{

return PyTuple_GET_ITEM(get_current_code_object(ctx)->co_names, index);

}

static int

get_test_bit_for_bools(void) {

#ifdef Py_STACKREF_DEBUG

uintptr_t false_bits = _Py_STACKREF_FALSE_INDEX;

uintptr_t true_bits = _Py_STACKREF_TRUE_INDEX;

#else

uintptr_t false_bits = (uintptr_t)&_Py_FalseStruct;

uintptr_t true_bits = (uintptr_t)&_Py_TrueStruct;

#endif

for (int i = 4; i < 8; i++) {

if ((true_bits ^ false_bits) & (uintptr_t)(1 << i)) {

return i;

}

return 0;

}

static int

test_bit_set_in_true(int bit) {

#ifdef Py_STACKREF_DEBUG

uintptr_t true_bits = _Py_STACKREF_TRUE_INDEX;

#else

uintptr_t true_bits = (uintptr_t)&_Py_TrueStruct;

#endif

assert((true_bits ^ ((uintptr_t)&_Py_FalseStruct)) & (uintptr_t)(1 << bit));

return true_bits & (uintptr_t)(1 << bit);

}

#ifdef Py_DEBUG

void

_Py_opt_assert_within_stack_bounds(

_Py_UOpsAbstractFrame *frame, JitOptRef *stack_pointer,

const char *filename, int lineno

) {

if (frame->code == ((PyCodeObject *)&_Py_InitCleanup)) {

return;

}

int level = (int)(stack_pointer - frame->stack);

if (level < 0) {

printf("Stack underflow (depth = %d) at %s:%d\n", level, filename, lineno);

fflush(stdout);

abort();

}

int size = (int)(frame->stack_len) + MAX_CACHED_REGISTER;

if (level > size) {

printf("Stack overflow (depth = %d) at %s:%d\n", level, filename, lineno);

fflush(stdout);

abort();

}

#endif

#ifdef Py_DEBUG

#define ASSERT_WITHIN_STACK_BOUNDS(F, L) _Py_opt_assert_within_stack_bounds(ctx->frame, stack_pointer, (F), (L))

#else

#define ASSERT_WITHIN_STACK_BOUNDS(F, L) (void)0

#endif

/* >0 (length) for success, 0 for not ready, clears all possible errors. */

static int

optimize_uops(

_PyThreadStateImpl *tstate,

_PyUOpInstruction *trace,

int trace_len,

int curr_stacklen,

_PyUOpInstruction *output,

_PyBloomFilter *dependencies

)

{

assert(!PyErr_Occurred());

assert(tstate->jit_tracer_state != NULL);

PyFunctionObject *func = tstate->jit_tracer_state->initial_state.func;

JitOptContext *ctx = &tstate->jit_tracer_state->opt_context;

uint32_t opcode = UINT16_MAX;

uop_buffer_init(&ctx->out_buffer, output, UOP_MAX_TRACE_LENGTH);

// Make sure that watchers are set up

PyInterpreterState *interp = _PyInterpreterState_GET();

_PyOnceFlag_CallOnce(&interp->dict_state.watcher_setup_once,

_setup_optimizer_watchers, NULL);

_Py_uop_abstractcontext_init(ctx, dependencies);

_Py_UOpsAbstractFrame *frame = _Py_uop_frame_new(ctx, (PyCodeObject *)func->func_code, NULL, 0);

if (frame == NULL) {

return 0;

}

frame->func = func;

ctx->curr_frame_depth++;

ctx->frame = frame;

_Py_uop_sym_set_stack_depth(ctx, curr_stacklen, frame->stack_pointer);

_PyUOpInstruction *this_instr = NULL;

JitOptRef *stack_pointer = ctx->frame->stack_pointer;

for (int i = 0; i < trace_len; i++) {

this_instr = &trace[i];

if (ctx->done) {

// Don't do any more optimization, but

// we still need to reach a terminator for corrctness.

*(ctx->out_buffer.next++) = *this_instr;

if (is_terminator_uop(this_instr)) {

break;

}

continue;

}

int oparg = this_instr->oparg;

opcode = this_instr->opcode;

if (!CURRENT_FRAME_IS_INIT_SHIM()) {

stack_pointer = ctx->frame->stack_pointer;

}

DUMP_UOP(ctx, "abs", (int)(this_instr - trace), this_instr, stack_pointer);

_PyUOpInstruction *out_ptr = ctx->out_buffer.next;

switch (opcode) {

#include "optimizer_cases.c.h"

default:

DPRINTF(1, "\nUnknown opcode in abstract interpreter\n");

Py_UNREACHABLE();

}

// If no ADD_OP was called during this iteration, copy the original instruction

if (ctx->out_buffer.next == out_ptr) {

*(ctx->out_buffer.next++) = *this_instr;

}

assert(ctx->frame != NULL);

DUMP_UOPS(ctx, "out", out_ptr, stack_pointer);

if (!CURRENT_FRAME_IS_INIT_SHIM() && !ctx->done) {

DPRINTF(3, " stack_level %d\n", STACK_LEVEL());

ctx->frame->stack_pointer = stack_pointer;

assert(STACK_LEVEL() >= 0);

}

if (ctx->out_of_space) {

DPRINTF(3, "\n");

DPRINTF(1, "Out of space in abstract interpreter\n");

}

if (ctx->contradiction) {

// Attempted to push a "bottom" (contradiction) symbol onto the stack.

// This means that the abstract interpreter has optimized to trace

// to an unreachable estate.

// We *could* generate an _EXIT_TRACE or _FATAL_ERROR here, but hitting

// bottom usually indicates an optimizer bug, so we are probably better off

// retrying later.

DPRINTF(3, "\n");

DPRINTF(1, "Hit bottom in abstract interpreter\n");

_Py_uop_abstractcontext_fini(ctx);

OPT_STAT_INC(optimizer_contradiction);

return 0;

}

/* Either reached the end or cannot optimize further, but there

* would be no benefit in retrying later */

_Py_uop_abstractcontext_fini(ctx);

// Check that the trace ends with a proper terminator

if (uop_buffer_length(&ctx->out_buffer) > 0) {

assert(is_terminator_uop(uop_buffer_last(&ctx->out_buffer)));

}

return uop_buffer_length(&ctx->out_buffer);

error:

DPRINTF(3, "\n");

DPRINTF(1, "Encountered error in abstract interpreter\n");

if (opcode <= MAX_UOP_ID) {

OPT_ERROR_IN_OPCODE(opcode);

}

_Py_uop_abstractcontext_fini(ctx);

assert(PyErr_Occurred());

PyErr_Clear();

return 0;

}

const uint16_t op_without_push[MAX_UOP_ID + 1] = {

[_COPY] = _NOP,

[_LOAD_CONST_INLINE] = _NOP,

[_LOAD_CONST_INLINE_BORROW] = _NOP,

[_LOAD_FAST] = _NOP,

[_LOAD_FAST_BORROW] = _NOP,

[_LOAD_SMALL_INT] = _NOP,

[_PUSH_NULL] = _NOP,

};

const bool op_skip[MAX_UOP_ID + 1] = {

[_NOP] = true,

[_CHECK_VALIDITY] = true,

[_CHECK_PERIODIC] = true,

[_SET_IP] = true,

};

const uint16_t op_without_pop[MAX_UOP_ID + 1] = {

[_POP_TOP] = _NOP,

[_POP_TOP_NOP] = _NOP,

[_POP_TOP_INT] = _NOP,

[_POP_TOP_FLOAT] = _NOP,

[_POP_TOP_UNICODE] = _NOP,

};

static int

remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)

{

/* Remove _SET_IP and _CHECK_VALIDITY where possible.

* _SET_IP is needed if the following instruction escapes or

* could error. _CHECK_VALIDITY is needed if the previous

* instruction could have escaped. */

int last_set_ip = -1;

bool may_have_escaped = true;

for (int pc = 0; pc < buffer_size; pc++) {

int opcode = buffer[pc].opcode;

switch (opcode) {

case _START_EXECUTOR:

may_have_escaped = false;

break;

case _SET_IP:

buffer[pc].opcode = _NOP;

last_set_ip = pc;

break;

case _CHECK_VALIDITY:

if (may_have_escaped) {

may_have_escaped = false;

}

else {

buffer[pc].opcode = _NOP;

}

break;

case _EXIT_TRACE:

default:

{

// Cancel out pushes and pops, repeatedly. So:

// _LOAD_FAST + _POP_TOP + _POP_TOP + _LOAD_CONST_INLINE_BORROW + _POP_TOP

// ...becomes:

// _NOP + _NOP + _POP_TOP + _NOP + _NOP

while (op_without_pop[opcode]) {

_PyUOpInstruction *last = &buffer[pc - 1];

while (op_skip[last->opcode]) {

last--;

}

if (op_without_push[last->opcode] && op_without_pop[opcode]) {

last->opcode = op_without_push[last->opcode];

opcode = buffer[pc].opcode = op_without_pop[opcode];

if (op_without_pop[last->opcode]) {

opcode = last->opcode;

pc = (int)(last - buffer);

}

else {

break;

}

/* _PUSH_FRAME doesn't escape or error, but it

* does need the IP for the return address */

bool needs_ip = (opcode == _PUSH_FRAME || opcode == _YIELD_VALUE || opcode == _DYNAMIC_EXIT || opcode == _EXIT_TRACE);

if (_PyUop_Flags[opcode] & HAS_ESCAPES_FLAG) {

needs_ip = true;

may_have_escaped = true;

}

if (needs_ip && last_set_ip >= 0) {

assert(buffer[last_set_ip].opcode == _NOP);

buffer[last_set_ip].opcode = _SET_IP;

last_set_ip = -1;

}

if (opcode == _EXIT_TRACE) {

return pc + 1;

}

break;

}

case _JUMP_TO_TOP:

case _DYNAMIC_EXIT:

case _DEOPT:

return pc + 1;

}

Py_UNREACHABLE();

}

// 0 - failure, no error raised, just fall back to Tier 1

// -1 - failure, and raise error

// > 0 - length of optimized trace

int

_Py_uop_analyze_and_optimize(

_PyThreadStateImpl *tstate,

_PyUOpInstruction *buffer,

int length,

int curr_stacklen,

_PyUOpInstruction *output,

_PyBloomFilter *dependencies

)

{

OPT_STAT_INC(optimizer_attempts);

length = optimize_uops(

tstate, buffer, length, curr_stacklen, output, dependencies);

if (length == 0) {

return length;

}

assert(length > 0);

length = remove_unneeded_uops(output, length);

assert(length > 0);

OPT_STAT_INC(optimizer_successes);

return length;

}

#endif /* _Py_TIER2 */

cpython/Python/optimizer_analysis.c at main · python/cpython · GitHub

Uh oh!

Uh oh!

cpython/Python/optimizer_analysis.c at main · python/cpython · GitHub

Uh oh!

Uh oh!

Expand file tree

Breadcrumbs

optimizer_analysis.c

Latest commit

History

Breadcrumbs

optimizer_analysis.c

File metadata and controls