CMake.git - CMake is a cross-platform, open-source build system generator.

path: root/Help/guide/tutorial/Step8/MathFunctions/MathFunctions.h

diff options

author	Kyle Edwards <kyle.edwards@kitware.com>	2020-01-10 21:31:29 (GMT)
committer	Kyle Edwards <kyle.edwards@kitware.com>	2020-01-10 21:31:29 (GMT)
commit	e8032e202e23436cc4301b7ded23c3fdce1161b2 (patch)
tree	9269877224bb3d6d70972887965d5d331fe11e3b /Help/guide/tutorial/Step8/MathFunctions/MathFunctions.h
parent	75e109a5b304ed502894f5e0ea791b327f6d1963 (diff)
download	CMake-e8032e202e23436cc4301b7ded23c3fdce1161b2.zip CMake-e8032e202e23436cc4301b7ded23c3fdce1161b2.tar.gz CMake-e8032e202e23436cc4301b7ded23c3fdce1161b2.tar.bz2

Ninja Multi-Config: Make cross-config building opt-in

Many users will want to use the Ninja Multi-Config generator like a traditional Visual Studio-style multi-config generator, which doesn't mix configurations - custom commands are built using target executables of the same configuration the command is for. We do not want to force these people to generate an N*N build matrix when they only need N*1, especially if they have lots of targets. Add a new variable, CMAKE_NINJA_CROSS_CONFIG_ENABLE, to opt-in to the cross-config build matrix.

Diffstat (limited to 'Help/guide/tutorial/Step8/MathFunctions/MathFunctions.h')

0 files changed, 0 insertions, 0 deletions

id='n154' href='#n154'>154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187

#include "Python.h"
#include "opcode.h"
#include "pycore_interp.h"
#include "pycore_bitutils.h"        // _Py_popcount32()
#include "pycore_opcode_metadata.h" // _PyOpcode_OpName[]
#include "pycore_opcode_utils.h"  // MAX_REAL_OPCODE
#include "pycore_optimizer.h"     // _Py_uop_analyze_and_optimize()
#include "pycore_pystate.h"       // _PyInterpreterState_GET()
#include "pycore_uop_ids.h"
#include "pycore_jit.h"
#include "cpython/optimizer.h"
#include <stdbool.h>
#include <stdint.h>
#include <stddef.h>

#define NEED_OPCODE_METADATA
#include "pycore_uop_metadata.h" // Uop tables
#undef NEED_OPCODE_METADATA

#define MAX_EXECUTORS_SIZE 256


static bool
has_space_for_executor(PyCodeObject *code, _Py_CODEUNIT *instr)
{
    if (instr->op.code == ENTER_EXECUTOR) {
        return true;
    }
    if (code->co_executors == NULL) {
        return true;
    }
    return code->co_executors->size < MAX_EXECUTORS_SIZE;
}

static int32_t
get_index_for_executor(PyCodeObject *code, _Py_CODEUNIT *instr)
{
    if (instr->op.code == ENTER_EXECUTOR) {
        return instr->op.arg;
    }
    _PyExecutorArray *old = code->co_executors;
    int size = 0;
    int capacity = 0;
    if (old != NULL) {
        size = old->size;
        capacity = old->capacity;
        assert(size < MAX_EXECUTORS_SIZE);
    }
    assert(size <= capacity);
    if (size == capacity) {
        /* Array is full. Grow array */
        int new_capacity = capacity ? capacity * 2 : 4;
        _PyExecutorArray *new = PyMem_Realloc(
            old,
            offsetof(_PyExecutorArray, executors) +
            new_capacity * sizeof(_PyExecutorObject *));
        if (new == NULL) {
            return -1;
        }
        new->capacity = new_capacity;
        new->size = size;
        code->co_executors = new;
    }
    assert(size < code->co_executors->capacity);
    return size;
}

static void
insert_executor(PyCodeObject *code, _Py_CODEUNIT *instr, int index, _PyExecutorObject *executor)
{
    Py_INCREF(executor);
    if (instr->op.code == ENTER_EXECUTOR) {
        assert(index == instr->op.arg);
        _Py_ExecutorClear(code->co_executors->executors[index]);
    }
    else {
        assert(code->co_executors->size == index);
        assert(code->co_executors->capacity > index);
        code->co_executors->size++;
    }
    executor->vm_data.opcode = instr->op.code;
    executor->vm_data.oparg = instr->op.arg;
    executor->vm_data.code = code;
    executor->vm_data.index = (int)(instr - _PyCode_CODE(code));
    code->co_executors->executors[index] = executor;
    assert(index < MAX_EXECUTORS_SIZE);
    instr->op.code = ENTER_EXECUTOR;
    instr->op.arg = index;
}

int
PyUnstable_Replace_Executor(PyCodeObject *code, _Py_CODEUNIT *instr, _PyExecutorObject *new)
{
    if (instr->op.code != ENTER_EXECUTOR) {
        PyErr_Format(PyExc_ValueError, "No executor to replace");
        return -1;
    }
    int index = instr->op.arg;
    assert(index >= 0);
    insert_executor(code, instr, index, new);
    return 0;
}

static int
never_optimize(
    _PyOptimizerObject* self,
    _PyInterpreterFrame *frame,
    _Py_CODEUNIT *instr,
    _PyExecutorObject **exec,
    int Py_UNUSED(stack_entries))
{
    /* Although it should be benign for this to be called,
     * it shouldn't happen, so fail in debug builds. */
    assert(0 && "never optimize should never be called");
    return 0;
}

PyTypeObject _PyDefaultOptimizer_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    .tp_name = "noop_optimizer",
    .tp_basicsize = sizeof(_PyOptimizerObject),
    .tp_itemsize = 0,
    .tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
};

static _PyOptimizerObject _PyOptimizer_Default = {
    PyObject_HEAD_INIT(&_PyDefaultOptimizer_Type)
    .optimize = never_optimize,
    .resume_threshold = OPTIMIZER_UNREACHABLE_THRESHOLD,
    .backedge_threshold = OPTIMIZER_UNREACHABLE_THRESHOLD,
};

static uint32_t
shift_and_offset_threshold(uint16_t threshold)
{
    return (threshold << OPTIMIZER_BITS_IN_COUNTER) + (1 << 15);
}

_PyOptimizerObject *
PyUnstable_GetOptimizer(void)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    if (interp->optimizer == &_PyOptimizer_Default) {
        return NULL;
    }
    assert(interp->optimizer_backedge_threshold ==
           shift_and_offset_threshold(interp->optimizer->backedge_threshold));
    assert(interp->optimizer_resume_threshold ==
           shift_and_offset_threshold(interp->optimizer->resume_threshold));
    Py_INCREF(interp->optimizer);
    return interp->optimizer;
}

_PyOptimizerObject *
_Py_SetOptimizer(PyInterpreterState *interp, _PyOptimizerObject *optimizer)
{
    if (optimizer == NULL) {
        optimizer = &_PyOptimizer_Default;
    }
    _PyOptimizerObject *old = interp->optimizer;
    Py_INCREF(optimizer);
    interp->optimizer = optimizer;
    interp->optimizer_backedge_threshold = shift_and_offset_threshold(optimizer->backedge_threshold);
    interp->optimizer_resume_threshold = shift_and_offset_threshold(optimizer->resume_threshold);
    return old;
}

void
PyUnstable_SetOptimizer(_PyOptimizerObject *optimizer)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    _PyOptimizerObject *old = _Py_SetOptimizer(interp, optimizer);
    Py_DECREF(old);
}

int
_PyOptimizer_Optimize(_PyInterpreterFrame *frame, _Py_CODEUNIT *start, PyObject **stack_pointer)
{
    PyCodeObject *code = (PyCodeObject *)frame->f_executable;
    assert(PyCode_Check(code));
    PyInterpreterState *interp = _PyInterpreterState_GET();
    if (!has_space_for_executor(code, start)) {
        return 0;
    }
    _PyOptimizerObject *opt = interp->optimizer;
    _PyExecutorObject *executor = NULL;
    int err = opt->optimize(opt, frame, start, &executor, (int)(stack_pointer - _PyFrame_Stackbase(frame)));
    if (err <= 0) {
        assert(executor == NULL);
        return err;
    }
    int index = get_index_for_executor(code, start);
    if (index < 0) {
        /* Out of memory. Don't raise and assume that the
         * error will show up elsewhere.
         *
         * If an optimizer has already produced an executor,
         * it might get confused by the executor disappearing,
         * but there is not much we can do about that here. */
        Py_DECREF(executor);
        return 0;
    }
    insert_executor(code, start, index, executor);
    Py_DECREF(executor);
    return 1;
}

_PyExecutorObject *
PyUnstable_GetExecutor(PyCodeObject *code, int offset)
{
    int code_len = (int)Py_SIZE(code);
    for (int i = 0 ; i < code_len;) {
        if (_PyCode_CODE(code)[i].op.code == ENTER_EXECUTOR && i*2 == offset) {
            int oparg = _PyCode_CODE(code)[i].op.arg;
            _PyExecutorObject *res = code->co_executors->executors[oparg];
            Py_INCREF(res);
            return res;
        }
        i += _PyInstruction_GetLength(code, i);
    }
    PyErr_SetString(PyExc_ValueError, "no executor at given byte offset");
    return NULL;
}

static PyObject *
is_valid(PyObject *self, PyObject *Py_UNUSED(ignored))
{
    return PyBool_FromLong(((_PyExecutorObject *)self)->vm_data.valid);
}

static PyMethodDef executor_methods[] = {
    { "is_valid", is_valid, METH_NOARGS, NULL },
    { NULL, NULL },
};

///////////////////// Experimental UOp Optimizer /////////////////////

static void
uop_dealloc(_PyExecutorObject *self) {
    _Py_ExecutorClear(self);
#ifdef _Py_JIT
    _PyJIT_Free(self);
#endif
    PyObject_Free(self);
}

const char *
_PyUOpName(int index)
{
    return _PyOpcode_uop_name[index];
}

static Py_ssize_t
uop_len(_PyExecutorObject *self)
{
    return Py_SIZE(self);
}

static PyObject *
uop_item(_PyExecutorObject *self, Py_ssize_t index)
{
    Py_ssize_t len = uop_len(self);
    if (index < 0 || index >= len) {
        PyErr_SetNone(PyExc_IndexError);
        return NULL;
    }
    const char *name = _PyUOpName(self->trace[index].opcode);
    if (name == NULL) {
        name = "<nil>";
    }
    PyObject *oname = _PyUnicode_FromASCII(name, strlen(name));
    if (oname == NULL) {
        return NULL;
    }
    PyObject *oparg = PyLong_FromUnsignedLong(self->trace[index].oparg);
    if (oparg == NULL) {
        Py_DECREF(oname);
        return NULL;
    }
    PyObject *operand = PyLong_FromUnsignedLongLong(self->trace[index].operand);
    if (operand == NULL) {
        Py_DECREF(oparg);
        Py_DECREF(oname);
        return NULL;
    }
    PyObject *args[3] = { oname, oparg, operand };
    return _PyTuple_FromArraySteal(args, 3);
}

PySequenceMethods uop_as_sequence = {
    .sq_length = (lenfunc)uop_len,
    .sq_item = (ssizeargfunc)uop_item,
};

PyTypeObject _PyUOpExecutor_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    .tp_name = "uop_executor",
    .tp_basicsize = offsetof(_PyExecutorObject, trace),
    .tp_itemsize = sizeof(_PyUOpInstruction),
    .tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
    .tp_dealloc = (destructor)uop_dealloc,
    .tp_as_sequence = &uop_as_sequence,
    .tp_methods = executor_methods,
};

/* TO DO -- Generate these tables */
static const uint16_t
_PyUOp_Replacements[MAX_UOP_ID + 1] = {
    [_ITER_JUMP_RANGE] = _GUARD_NOT_EXHAUSTED_RANGE,
    [_ITER_JUMP_LIST] = _GUARD_NOT_EXHAUSTED_LIST,
    [_ITER_JUMP_TUPLE] = _GUARD_NOT_EXHAUSTED_TUPLE,
    [_FOR_ITER] = _FOR_ITER_TIER_TWO,
};

static const uint16_t
BRANCH_TO_GUARD[4][2] = {
    [POP_JUMP_IF_FALSE - POP_JUMP_IF_FALSE][0] = _GUARD_IS_TRUE_POP,
    [POP_JUMP_IF_FALSE - POP_JUMP_IF_FALSE][1] = _GUARD_IS_FALSE_POP,
    [POP_JUMP_IF_TRUE - POP_JUMP_IF_FALSE][0] = _GUARD_IS_FALSE_POP,
    [POP_JUMP_IF_TRUE - POP_JUMP_IF_FALSE][1] = _GUARD_IS_TRUE_POP,
    [POP_JUMP_IF_NONE - POP_JUMP_IF_FALSE][0] = _GUARD_IS_NOT_NONE_POP,
    [POP_JUMP_IF_NONE - POP_JUMP_IF_FALSE][1] = _GUARD_IS_NONE_POP,
    [POP_JUMP_IF_NOT_NONE - POP_JUMP_IF_FALSE][0] = _GUARD_IS_NONE_POP,
    [POP_JUMP_IF_NOT_NONE - POP_JUMP_IF_FALSE][1] = _GUARD_IS_NOT_NONE_POP,
};

#define CONFIDENCE_RANGE 1000
#define CONFIDENCE_CUTOFF 333

#ifdef Py_DEBUG
#define DPRINTF(level, ...) \
    if (lltrace >= (level)) { printf(__VA_ARGS__); }
#else
#define DPRINTF(level, ...)
#endif


#define ADD_TO_TRACE(OPCODE, OPARG, OPERAND, TARGET) \
    DPRINTF(2, \
            "  ADD_TO_TRACE(%s, %d, %" PRIu64 ", %d)\n", \
            _PyUOpName(OPCODE), \
            (OPARG), \
            (uint64_t)(OPERAND), \
            TARGET); \
    assert(trace_length < max_length); \
    trace[trace_length].opcode = (OPCODE); \
    trace[trace_length].oparg = (OPARG); \
    trace[trace_length].operand = (OPERAND); \
    trace[trace_length].target = (TARGET); \
    trace_length++;

#define INSTR_IP(INSTR, CODE) \
    ((uint32_t)((INSTR) - ((_Py_CODEUNIT *)(CODE)->co_code_adaptive)))

// Reserve space for n uops
#define RESERVE_RAW(n, opname) \
    if (trace_length + (n) > max_length) { \
        DPRINTF(2, "No room for %s (need %d, got %d)\n", \
                (opname), (n), max_length - trace_length); \
        OPT_STAT_INC(trace_too_long); \
        goto done; \
    }

// Reserve space for N uops, plus 3 for _SET_IP, _CHECK_VALIDITY and _EXIT_TRACE
#define RESERVE(needed) RESERVE_RAW((needed) + 3, _PyUOpName(opcode))

// Trace stack operations (used by _PUSH_FRAME, _POP_FRAME)
#define TRACE_STACK_PUSH() \
    if (trace_stack_depth >= TRACE_STACK_SIZE) { \
        DPRINTF(2, "Trace stack overflow\n"); \
        OPT_STAT_INC(trace_stack_overflow); \
        ADD_TO_TRACE(_EXIT_TRACE, 0, 0, 0); \
        goto done; \
    } \
    assert(func->func_code == (PyObject *)code); \
    trace_stack[trace_stack_depth].func = func; \
    trace_stack[trace_stack_depth].instr = instr; \
    trace_stack_depth++;
#define TRACE_STACK_POP() \
    if (trace_stack_depth <= 0) { \
        Py_FatalError("Trace stack underflow\n"); \
    } \
    trace_stack_depth--; \
    func = trace_stack[trace_stack_depth].func; \
    code = (PyCodeObject *)trace_stack[trace_stack_depth].func->func_code; \
    instr = trace_stack[trace_stack_depth].instr;

/* Returns 1 on success,
 * 0 if it failed to produce a worthwhile trace,
 * and -1 on an error.
 */
static int
translate_bytecode_to_trace(
    _PyInterpreterFrame *frame,
    _Py_CODEUNIT *instr,
    _PyUOpInstruction *trace,
    int buffer_size,
    _PyBloomFilter *dependencies)
{
    bool progress_needed = true;
    PyCodeObject *code = (PyCodeObject *)frame->f_executable;
    PyFunctionObject *func = (PyFunctionObject *)frame->f_funcobj;
    assert(PyFunction_Check(func));
    PyCodeObject *initial_code = code;
    _Py_BloomFilter_Add(dependencies, initial_code);
    _Py_CODEUNIT *initial_instr = instr;
    int trace_length = 0;
    int max_length = buffer_size;
    struct {
        PyFunctionObject *func;
        _Py_CODEUNIT *instr;
    } trace_stack[TRACE_STACK_SIZE];
    int trace_stack_depth = 0;
    int confidence = CONFIDENCE_RANGE;  // Adjusted by branch instructions

#ifdef Py_DEBUG
    char *python_lltrace = Py_GETENV("PYTHON_LLTRACE");
    int lltrace = 0;
    if (python_lltrace != NULL && *python_lltrace >= '0') {
        lltrace = *python_lltrace - '0';  // TODO: Parse an int and all that
    }
#endif

    DPRINTF(4,
            "Optimizing %s (%s:%d) at byte offset %d\n",
            PyUnicode_AsUTF8(code->co_qualname),
            PyUnicode_AsUTF8(code->co_filename),
            code->co_firstlineno,
            2 * INSTR_IP(initial_instr, code));
    uint32_t target = 0;

top:  // Jump here after _PUSH_FRAME or likely branches
    for (;;) {
        target = INSTR_IP(instr, code);
        RESERVE_RAW(2, "epilogue");  // Always need space for _SET_IP, _CHECK_VALIDITY and _EXIT_TRACE
        ADD_TO_TRACE(_CHECK_VALIDITY_AND_SET_IP, 0, (uintptr_t)instr, target);

        uint32_t opcode = instr->op.code;
        uint32_t oparg = instr->op.arg;
        uint32_t extended = 0;

        if (opcode == ENTER_EXECUTOR) {
            assert(oparg < 256);
            _PyExecutorObject *executor = code->co_executors->executors[oparg];
            opcode = executor->vm_data.opcode;
            DPRINTF(2, "  * ENTER_EXECUTOR -> %s\n",  _PyOpcode_OpName[opcode]);
            oparg = executor->vm_data.oparg;
        }

        if (opcode == EXTENDED_ARG) {
            instr++;
            extended = 1;
            opcode = instr->op.code;
            oparg = (oparg << 8) | instr->op.arg;
            if (opcode == EXTENDED_ARG) {
                instr--;
                goto done;
            }
        }
        assert(opcode != ENTER_EXECUTOR && opcode != EXTENDED_ARG);

        /* Special case the first instruction,
         * so that we can guarantee forward progress */
        if (progress_needed) {
            progress_needed = false;
            if (opcode == JUMP_BACKWARD || opcode == JUMP_BACKWARD_NO_INTERRUPT) {
                instr += 1 + _PyOpcode_Caches[opcode] - (int32_t)oparg;
                initial_instr = instr;
                continue;
            }
            else {
                if (OPCODE_HAS_DEOPT(opcode)) {
                    opcode = _PyOpcode_Deopt[opcode];
                }
                assert(!OPCODE_HAS_DEOPT(opcode));
            }
        }

        switch (opcode) {
            case POP_JUMP_IF_NONE:
            case POP_JUMP_IF_NOT_NONE:
            case POP_JUMP_IF_FALSE:
            case POP_JUMP_IF_TRUE:
            {
                RESERVE(1);
                int counter = instr[1].cache;
                int bitcount = _Py_popcount32(counter);
                int jump_likely = bitcount > 8;
                if (jump_likely) {
                    confidence = confidence * bitcount / 16;
                }
                else {
                    confidence = confidence * (16 - bitcount) / 16;
                }
                if (confidence < CONFIDENCE_CUTOFF) {
                    DPRINTF(2, "Confidence too low (%d)\n", confidence);
                    OPT_STAT_INC(low_confidence);
                    goto done;
                }
                uint32_t uopcode = BRANCH_TO_GUARD[opcode - POP_JUMP_IF_FALSE][jump_likely];
                DPRINTF(2, "%s(%d): counter=%x, bitcount=%d, likely=%d, confidence=%d, uopcode=%s\n",
                        _PyOpcode_OpName[opcode], oparg,
                        counter, bitcount, jump_likely, confidence, _PyUOpName(uopcode));
                _Py_CODEUNIT *next_instr = instr + 1 + _PyOpcode_Caches[_PyOpcode_Deopt[opcode]];
                _Py_CODEUNIT *target_instr = next_instr + oparg;
                if (jump_likely) {
                    DPRINTF(2, "Jump likely (%x = %d bits), continue at byte offset %d\n",
                            instr[1].cache, bitcount, 2 * INSTR_IP(target_instr, code));
                    instr = target_instr;
                    ADD_TO_TRACE(uopcode, max_length, 0, INSTR_IP(next_instr, code));
                    goto top;
                }
                ADD_TO_TRACE(uopcode, max_length, 0, INSTR_IP(target_instr, code));
                break;
            }

            case JUMP_BACKWARD:
            case JUMP_BACKWARD_NO_INTERRUPT:
            {
                _Py_CODEUNIT *target = instr + 1 + _PyOpcode_Caches[opcode] - (int)oparg;
                if (target == initial_instr) {
                    /* We have looped round to the start */
                    RESERVE(1);
                    ADD_TO_TRACE(_JUMP_TO_TOP, 0, 0, 0);
                }
                else {
                    OPT_STAT_INC(inner_loop);
                    DPRINTF(2, "JUMP_BACKWARD not to top ends trace\n");
                }
                goto done;
            }

            case JUMP_FORWARD:
            {
                RESERVE(0);
                // This will emit two _SET_IP instructions; leave it to the optimizer
                instr += oparg;
                break;
            }

            default:
            {
                const struct opcode_macro_expansion *expansion = &_PyOpcode_macro_expansion[opcode];
                if (expansion->nuops > 0) {
                    // Reserve space for nuops (+ _SET_IP + _EXIT_TRACE)
                    int nuops = expansion->nuops;
                    RESERVE(nuops);
                    if (expansion->uops[nuops-1].uop == _POP_FRAME) {
                        // Check for trace stack underflow now:
                        // We can't bail e.g. in the middle of
                        // LOAD_CONST + _POP_FRAME.
                        if (trace_stack_depth == 0) {
                            DPRINTF(2, "Trace stack underflow\n");
                            OPT_STAT_INC(trace_stack_underflow);
                            goto done;
                        }
                    }
                    uint32_t orig_oparg = oparg;  // For OPARG_TOP/BOTTOM
                    for (int i = 0; i < nuops; i++) {
                        oparg = orig_oparg;
                        uint32_t uop = expansion->uops[i].uop;
                        uint64_t operand = 0;
                        // Add one to account for the actual opcode/oparg pair:
                        int offset = expansion->uops[i].offset + 1;
                        switch (expansion->uops[i].size) {
                            case OPARG_FULL:
                                assert(opcode != JUMP_BACKWARD_NO_INTERRUPT && opcode != JUMP_BACKWARD);
                                break;
                            case OPARG_CACHE_1:
                                operand = read_u16(&instr[offset].cache);
                                break;
                            case OPARG_CACHE_2:
                                operand = read_u32(&instr[offset].cache);
                                break;
                            case OPARG_CACHE_4:
                                operand = read_u64(&instr[offset].cache);
                                break;
                            case OPARG_TOP:  // First half of super-instr
                                oparg = orig_oparg >> 4;
                                break;
                            case OPARG_BOTTOM:  // Second half of super-instr
                                oparg = orig_oparg & 0xF;
                                break;
                            case OPARG_SAVE_RETURN_OFFSET:  // op=_SAVE_RETURN_OFFSET; oparg=return_offset
                                oparg = offset;
                                assert(uop == _SAVE_RETURN_OFFSET);
                                break;
                            case OPARG_REPLACED:
                                uop = _PyUOp_Replacements[uop];
                                assert(uop != 0);
                                if (uop == _FOR_ITER_TIER_TWO) {
                                    target += 1 + INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 2 + extended;
                                    assert(_PyCode_CODE(code)[target-2].op.code == END_FOR ||
                                            _PyCode_CODE(code)[target-2].op.code == INSTRUMENTED_END_FOR);
                                    assert(_PyCode_CODE(code)[target-1].op.code == POP_TOP);
                                }
                                break;
                            default:
                                fprintf(stderr,
                                        "opcode=%d, oparg=%d; nuops=%d, i=%d; size=%d, offset=%d\n",
                                        opcode, oparg, nuops, i,
                                        expansion->uops[i].size,
                                        expansion->uops[i].offset);
                                Py_FatalError("garbled expansion");
                        }
                        ADD_TO_TRACE(uop, oparg, operand, target);
                        if (uop == _POP_FRAME) {
                            TRACE_STACK_POP();
                            /* Set the operand to the function object returned to,
                             * to assist optimization passes */
                            trace[trace_length-1].operand = (uintptr_t)func;
                            DPRINTF(2,
                                "Returning to %s (%s:%d) at byte offset %d\n",
                                PyUnicode_AsUTF8(code->co_qualname),
                                PyUnicode_AsUTF8(code->co_filename),
                                code->co_firstlineno,
                                2 * INSTR_IP(instr, code));
                            goto top;
                        }
                        if (uop == _PUSH_FRAME) {
                            assert(i + 1 == nuops);
                            int func_version_offset =
                                offsetof(_PyCallCache, func_version)/sizeof(_Py_CODEUNIT)
                                // Add one to account for the actual opcode/oparg pair:
                                + 1;
                            uint32_t func_version = read_u32(&instr[func_version_offset].cache);
                            PyFunctionObject *new_func = _PyFunction_LookupByVersion(func_version);
                            DPRINTF(3, "Function object: %p\n", func);
                            if (new_func != NULL) {
                                PyCodeObject *new_code = (PyCodeObject *)PyFunction_GET_CODE(new_func);
                                if (new_code == code) {
                                    // Recursive call, bail (we could be here forever).
                                    DPRINTF(2, "Bailing on recursive call to %s (%s:%d)\n",
                                            PyUnicode_AsUTF8(new_code->co_qualname),
                                            PyUnicode_AsUTF8(new_code->co_filename),
                                            new_code->co_firstlineno);
                                    OPT_STAT_INC(recursive_call);
                                    ADD_TO_TRACE(_EXIT_TRACE, 0, 0, 0);
                                    goto done;
                                }
                                if (new_code->co_version != func_version) {
                                    // func.__code__ was updated.
                                    // Perhaps it may happen again, so don't bother tracing.
                                    // TODO: Reason about this -- is it better to bail or not?
                                    DPRINTF(2, "Bailing because co_version != func_version\n");
                                    ADD_TO_TRACE(_EXIT_TRACE, 0, 0, 0);
                                    goto done;
                                }
                                // Increment IP to the return address
                                instr += _PyOpcode_Caches[_PyOpcode_Deopt[opcode]] + 1;
                                TRACE_STACK_PUSH();
                                _Py_BloomFilter_Add(dependencies, new_code);
                                /* Set the operand to the callee's code object,
                                * to assist optimization passes */
                                trace[trace_length-1].operand = (uintptr_t)new_func;
                                code = new_code;
                                func = new_func;
                                instr = _PyCode_CODE(code);
                                DPRINTF(2,
                                    "Continuing in %s (%s:%d) at byte offset %d\n",
                                    PyUnicode_AsUTF8(code->co_qualname),
                                    PyUnicode_AsUTF8(code->co_filename),
                                    code->co_firstlineno,
                                    2 * INSTR_IP(instr, code));
                                goto top;
                            }
                            ADD_TO_TRACE(_EXIT_TRACE, 0, 0, 0);
                            goto done;
                        }
                    }
                    break;
                }
                DPRINTF(2, "Unsupported opcode %s\n", _PyOpcode_OpName[opcode]);
                OPT_UNSUPPORTED_OPCODE(opcode);
                goto done;  // Break out of loop
            }  // End default

        }  // End switch (opcode)

        instr++;
        // Add cache size for opcode
        instr += _PyOpcode_Caches[_PyOpcode_Deopt[opcode]];
    }  // End for (;;)

done:
    while (trace_stack_depth > 0) {
        TRACE_STACK_POP();
    }
    assert(code == initial_code);
    // Skip short traces like _SET_IP, LOAD_FAST, _SET_IP, _EXIT_TRACE
    if (progress_needed || trace_length < 5) {
        OPT_STAT_INC(trace_too_short);
        DPRINTF(4,
                "No trace for %s (%s:%d) at byte offset %d\n",
                PyUnicode_AsUTF8(code->co_qualname),
                PyUnicode_AsUTF8(code->co_filename),
                code->co_firstlineno,
                2 * INSTR_IP(initial_instr, code));
        return 0;
    }
    ADD_TO_TRACE(_EXIT_TRACE, 0, 0, target);
    DPRINTF(1,
            "Created a trace for %s (%s:%d) at byte offset %d -- length %d\n",
            PyUnicode_AsUTF8(code->co_qualname),
            PyUnicode_AsUTF8(code->co_filename),
            code->co_firstlineno,
            2 * INSTR_IP(initial_instr, code),
            trace_length);
    OPT_HIST(trace_length + buffer_size - max_length, trace_length_hist);
    return 1;
}

#undef RESERVE
#undef RESERVE_RAW
#undef INSTR_IP
#undef ADD_TO_TRACE
#undef DPRINTF

#define UNSET_BIT(array, bit) (array[(bit)>>5] &= ~(1<<((bit)&31)))
#define SET_BIT(array, bit) (array[(bit)>>5] |= (1<<((bit)&31)))
#define BIT_IS_SET(array, bit) (array[(bit)>>5] & (1<<((bit)&31)))

/* Count the number of used uops, and mark them in the bit vector `used`.
 * This can be done in a single pass using simple reachability analysis,
 * as there are no backward jumps.
 * NOPs are excluded from the count.
*/
static int
compute_used(_PyUOpInstruction *buffer, uint32_t *used)
{
    int count = 0;
    SET_BIT(used, 0);
    for (int i = 0; i < UOP_MAX_TRACE_LENGTH; i++) {
        if (!BIT_IS_SET(used, i)) {
            continue;
        }
        count++;
        int opcode = buffer[i].opcode;
        if (opcode == _JUMP_TO_TOP || opcode == _EXIT_TRACE) {
            continue;
        }
        /* All other micro-ops fall through, so i+1 is reachable */
        SET_BIT(used, i+1);
        assert(opcode <= MAX_UOP_ID);
        if (_PyUop_Flags[opcode] & HAS_JUMP_FLAG) {
            /* Mark target as reachable */
            SET_BIT(used, buffer[i].oparg);
        }
        if (opcode == NOP) {
            count--;
            UNSET_BIT(used, i);
        }
    }
    return count;
}

/* Makes an executor from a buffer of uops.
 * Account for the buffer having gaps and NOPs by computing a "used"
 * bit vector and only copying the used uops. Here "used" means reachable
 * and not a NOP.
 */
static _PyExecutorObject *
make_executor_from_uops(_PyUOpInstruction *buffer, _PyBloomFilter *dependencies)
{
    uint32_t used[(UOP_MAX_TRACE_LENGTH + 31)/32] = { 0 };
    int length = compute_used(buffer, used);
    _PyExecutorObject *executor = PyObject_NewVar(_PyExecutorObject, &_PyUOpExecutor_Type, length);
    if (executor == NULL) {
        return NULL;
    }
    int dest = length - 1;
    /* Scan backwards, so that we see the destinations of jumps before the jumps themselves. */
    for (int i = UOP_MAX_TRACE_LENGTH-1; i >= 0; i--) {
        if (!BIT_IS_SET(used, i)) {
            continue;
        }
        executor->trace[dest] = buffer[i];
        int opcode = buffer[i].opcode;
        if (opcode == _POP_JUMP_IF_FALSE ||
            opcode == _POP_JUMP_IF_TRUE)
        {
            /* The oparg of the target will already have been set to its new offset */
            int oparg = executor->trace[dest].oparg;
            executor->trace[dest].oparg = buffer[oparg].oparg;
        }
        /* Set the oparg to be the destination offset,
         * so that we can set the oparg of earlier jumps correctly. */
        buffer[i].oparg = dest;
        dest--;
    }
    assert(dest == -1);
    _Py_ExecutorInit(executor, dependencies);
#ifdef Py_DEBUG
    char *python_lltrace = Py_GETENV("PYTHON_LLTRACE");
    int lltrace = 0;
    if (python_lltrace != NULL && *python_lltrace >= '0') {
        lltrace = *python_lltrace - '0';  // TODO: Parse an int and all that
    }
    if (lltrace >= 2) {
        printf("Optimized executor (length %d):\n", length);
        for (int i = 0; i < length; i++) {
            printf("%4d %s(%d, %d, %" PRIu64 ")\n",
                   i,
                   _PyUOpName(executor->trace[i].opcode),
                   executor->trace[i].oparg,
                   executor->trace[i].target,
                   executor->trace[i].operand);
        }
    }
#endif
#ifdef _Py_JIT
    executor->jit_code = NULL;
    executor->jit_size = 0;
    if (_PyJIT_Compile(executor, executor->trace, Py_SIZE(executor))) {
        Py_DECREF(executor);
        return NULL;
    }
#endif
    return executor;
}

static int
uop_optimize(
    _PyOptimizerObject *self,
    _PyInterpreterFrame *frame,
    _Py_CODEUNIT *instr,
    _PyExecutorObject **exec_ptr,
    int curr_stackentries)
{
    _PyBloomFilter dependencies;
    _Py_BloomFilter_Init(&dependencies);
    _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
    int err = translate_bytecode_to_trace(frame, instr, buffer, UOP_MAX_TRACE_LENGTH, &dependencies);
    if (err <= 0) {
        // Error or nothing translated
        return err;
    }
    OPT_STAT_INC(traces_created);
    char *uop_optimize = Py_GETENV("PYTHONUOPSOPTIMIZE");
    if (uop_optimize == NULL || *uop_optimize > '0') {
        err = _Py_uop_analyze_and_optimize(frame, buffer,
                                           UOP_MAX_TRACE_LENGTH,
                                           curr_stackentries, &dependencies);
        if (err <= 0) {
            return err;
        }
    }
    assert(err == 1);
    _PyExecutorObject *executor = make_executor_from_uops(buffer, &dependencies);
    if (executor == NULL) {
        return -1;
    }
    OPT_HIST(Py_SIZE(executor), optimized_trace_length_hist);
    *exec_ptr = executor;
    return 1;
}

static void
uop_opt_dealloc(PyObject *self) {
    PyObject_Free(self);
}

PyTypeObject _PyUOpOptimizer_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    .tp_name = "uop_optimizer",
    .tp_basicsize = sizeof(_PyOptimizerObject),
    .tp_itemsize = 0,
    .tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
    .tp_dealloc = uop_opt_dealloc,
};

PyObject *
PyUnstable_Optimizer_NewUOpOptimizer(void)
{
    _PyOptimizerObject *opt = PyObject_New(_PyOptimizerObject, &_PyUOpOptimizer_Type);
    if (opt == NULL) {
        return NULL;
    }
    opt->optimize = uop_optimize;
    opt->resume_threshold = OPTIMIZER_UNREACHABLE_THRESHOLD;
    // Need a few iterations to settle specializations,
    // and to ammortize the cost of optimization.
    opt->backedge_threshold = 16;
    return (PyObject *)opt;
}

static void
counter_dealloc(_PyExecutorObject *self) {
    PyObject *opt = (PyObject *)self->trace[0].operand;
    Py_DECREF(opt);
    uop_dealloc(self);
}

PyTypeObject _PyCounterExecutor_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    .tp_name = "counting_executor",
    .tp_basicsize = offsetof(_PyExecutorObject, trace),
    .tp_itemsize = sizeof(_PyUOpInstruction),
    .tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
    .tp_dealloc = (destructor)counter_dealloc,
    .tp_methods = executor_methods,
};

static int
counter_optimize(
    _PyOptimizerObject* self,
    _PyInterpreterFrame *frame,
    _Py_CODEUNIT *instr,
    _PyExecutorObject **exec_ptr,
    int Py_UNUSED(curr_stackentries)
)
{
    PyCodeObject *code = (PyCodeObject *)frame->f_executable;
    int oparg = instr->op.arg;
    while (instr->op.code == EXTENDED_ARG) {
        instr++;
        oparg = (oparg << 8) | instr->op.arg;
    }
    if (instr->op.code != JUMP_BACKWARD) {
        /* Counter optimizer can only handle backward edges */
        return 0;
    }
    _Py_CODEUNIT *target = instr + 1 + _PyOpcode_Caches[JUMP_BACKWARD] - oparg;
    _PyUOpInstruction buffer[3] = {
        { .opcode = _LOAD_CONST_INLINE_BORROW, .operand = (uintptr_t)self },
        { .opcode = _INTERNAL_INCREMENT_OPT_COUNTER },
        { .opcode = _EXIT_TRACE, .target = (uint32_t)(target - _PyCode_CODE(code)) }
    };
    _PyBloomFilter empty;
    _Py_BloomFilter_Init(&empty);
    _PyExecutorObject *executor = make_executor_from_uops(buffer, &empty);
    if (executor == NULL) {
        return -1;
    }
    Py_INCREF(self);
    Py_SET_TYPE(executor, &_PyCounterExecutor_Type);
    *exec_ptr = executor;
    return 1;
}

static PyObject *
counter_get_counter(PyObject *self, PyObject *args)
{
    return PyLong_FromLongLong(((_PyCounterOptimizerObject *)self)->count);
}

static PyMethodDef counter_optimizer_methods[] = {
    { "get_count", counter_get_counter, METH_NOARGS, NULL },
    { NULL, NULL },
};

PyTypeObject _PyCounterOptimizer_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    .tp_name = "Counter optimizer",
    .tp_basicsize = sizeof(_PyCounterOptimizerObject),
    .tp_itemsize = 0,
    .tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
    .tp_methods = counter_optimizer_methods,
    .tp_dealloc = (destructor)PyObject_Del,
};

PyObject *
PyUnstable_Optimizer_NewCounter(void)
{
    _PyCounterOptimizerObject *opt = (_PyCounterOptimizerObject *)_PyObject_New(&_PyCounterOptimizer_Type);
    if (opt == NULL) {
        return NULL;
    }
    opt->base.optimize = counter_optimize;
    opt->base.resume_threshold = OPTIMIZER_UNREACHABLE_THRESHOLD;
    opt->base.backedge_threshold = 0;
    opt->count = 0;
    return (PyObject *)opt;
}


/*****************************************
 *        Executor management
 ****************************************/

/* We use a bloomfilter with k = 6, m = 256
 * The choice of k and the following constants
 * could do with a more rigourous analysis,
 * but here is a simple analysis:
 *
 * We want to keep the false positive rate low.
 * For n = 5 (a trace depends on 5 objects),
 * we expect 30 bits set, giving a false positive
 * rate of (30/256)**6 == 2.5e-6 which is plenty
 * good enough.
 *
 * However with n = 10 we expect 60 bits set (worst case),
 * giving a false positive of (60/256)**6 == 0.0001
 *
 * We choose k = 6, rather than a higher number as
 * it means the false positive rate grows slower for high n.
 *
 * n = 5, k = 6 => fp = 2.6e-6
 * n = 5, k = 8 => fp = 3.5e-7
 * n = 10, k = 6 => fp = 1.6e-4
 * n = 10, k = 8 => fp = 0.9e-4
 * n = 15, k = 6 => fp = 0.18%
 * n = 15, k = 8 => fp = 0.23%
 * n = 20, k = 6 => fp = 1.1%
 * n = 20, k = 8 => fp = 2.3%
 *
 * The above analysis assumes perfect hash functions,
 * but those don't exist, so the real false positive
 * rates may be worse.
 */

#define K 6

#define SEED 20221211

/* TO DO -- Use more modern hash functions with better distribution of bits */
static uint64_t
address_to_hash(void *ptr) {
    assert(ptr != NULL);
    uint64_t uhash = SEED;
    uintptr_t addr = (uintptr_t)ptr;
    for (int i = 0; i < SIZEOF_VOID_P; i++) {
        uhash ^= addr & 255;
        uhash *= (uint64_t)_PyHASH_MULTIPLIER;
        addr >>= 8;
    }
    return uhash;
}

void
_Py_BloomFilter_Init(_PyBloomFilter *bloom)
{
    for (int i = 0; i < BLOOM_FILTER_WORDS; i++) {
        bloom->bits[i] = 0;
    }
}

/* We want K hash functions that each set 1 bit.
 * A hash function that sets 1 bit in M bits can be trivially
 * derived from a log2(M) bit hash function.
 * So we extract 8 (log2(256)) bits at a time from
 * the 64bit hash. */
void
_Py_BloomFilter_Add(_PyBloomFilter *bloom, void *ptr)
{
    uint64_t hash = address_to_hash(ptr);
    assert(K <= 8);
    for (int i = 0; i < K; i++) {
        uint8_t bits = hash & 255;
        bloom->bits[bits >> 5] |= (1 << (bits&31));
        hash >>= 8;
    }
}

static bool
bloom_filter_may_contain(_PyBloomFilter *bloom, _PyBloomFilter *hashes)
{
    for (int i = 0; i < BLOOM_FILTER_WORDS; i++) {
        if ((bloom->bits[i] & hashes->bits[i]) != hashes->bits[i]) {
            return false;
        }
    }
    return true;
}

static void
link_executor(_PyExecutorObject *executor)
{
    PyInterpreterState *interp = _PyInterpreterState_GET();
    _PyExecutorLinkListNode *links = &executor->vm_data.links;
    _PyExecutorObject *head = interp->executor_list_head;
    if (head == NULL) {
        interp->executor_list_head = executor;
        links->previous = NULL;
        links->next = NULL;
    }
    else {
        _PyExecutorObject *next = head->vm_data.links.next;
        links->previous = head;
        links->next = next;
        if (next != NULL) {
            next->vm_data.links.previous = executor;
        }
        head->vm_data.links.next = executor;
    }
    executor->vm_data.valid = true;
    /* executor_list_head must be first in list */
    assert(interp->executor_list_head->vm_data.links.previous == NULL);
}

static void
unlink_executor(_PyExecutorObject *executor)
{
    if (!executor->vm_data.valid) {
        return;
    }
    _PyExecutorLinkListNode *links = &executor->vm_data.links;
    _PyExecutorObject *next = links->next;
    _PyExecutorObject *prev = links->previous;
    if (next != NULL) {
        next->vm_data.links.previous = prev;
    }
    if (prev != NULL) {
        prev->vm_data.links.next = next;
    }
    else {
        // prev == NULL implies that executor is the list head
        PyInterpreterState *interp = PyInterpreterState_Get();
        assert(interp->executor_list_head == executor);
        interp->executor_list_head = next;
    }
    executor->vm_data.valid = false;
}

/* This must be called by optimizers before using the executor */
void
_Py_ExecutorInit(_PyExecutorObject *executor, _PyBloomFilter *dependency_set)
{
    executor->vm_data.valid = true;
    for (int i = 0; i < BLOOM_FILTER_WORDS; i++) {
        executor->vm_data.bloom.bits[i] = dependency_set->bits[i];
    }
    link_executor(executor);
}

/* This must be called by executors during dealloc */
void
_Py_ExecutorClear(_PyExecutorObject *executor)
{
    unlink_executor(executor);
    PyCodeObject *code = executor->vm_data.code;
    if (code == NULL) {
        return;
    }
    _Py_CODEUNIT *instruction = &_PyCode_CODE(code)[executor->vm_data.index];
    assert(instruction->op.code == ENTER_EXECUTOR);
    int index = instruction->op.arg;
    assert(code->co_executors->executors[index] == executor);
    instruction->op.code = executor->vm_data.opcode;
    instruction->op.arg = executor->vm_data.oparg;
    executor->vm_data.code = NULL;
    Py_CLEAR(code->co_executors->executors[index]);
}

void
_Py_Executor_DependsOn(_PyExecutorObject *executor, void *obj)
{
    assert(executor->vm_data.valid);
    _Py_BloomFilter_Add(&executor->vm_data.bloom, obj);
}

/* Invalidate all executors that depend on `obj`
 * May cause other executors to be invalidated as well
 */
void
_Py_Executors_InvalidateDependency(PyInterpreterState *interp, void *obj)
{
    _PyBloomFilter obj_filter;
    _Py_BloomFilter_Init(&obj_filter);
    _Py_BloomFilter_Add(&obj_filter, obj);
    /* Walk the list of executors */
    /* TO DO -- Use a tree to avoid traversing as many objects */
    for (_PyExecutorObject *exec = interp->executor_list_head; exec != NULL;) {
        assert(exec->vm_data.valid);
        _PyExecutorObject *next = exec->vm_data.links.next;
        if (bloom_filter_may_contain(&exec->vm_data.bloom, &obj_filter)) {
            _Py_ExecutorClear(exec);
        }
        exec = next;
    }
}

/* Invalidate all executors */
void
_Py_Executors_InvalidateAll(PyInterpreterState *interp)
{
    while (interp->executor_list_head) {
        _PyExecutorObject *executor = interp->executor_list_head;
        if (executor->vm_data.code) {
            // Clear the entire code object so its co_executors array be freed:
            _PyCode_Clear_Executors(executor->vm_data.code);
        }
        else {
            _Py_ExecutorClear(executor);
        }
    }
}


context:
space:
mode: