cpython/Python/specialize.c

#include "Python.h"
#include "pycore_code.h"
#include "pycore_dict.h"
#include "pycore_moduleobject.h"
#include "opcode.h"
#include "structmember.h"         // struct PyMemberDef, T_OFFSET_EX


/* We layout the quickened data as a bi-directional array:
 * Instructions upwards, cache entries downwards.
 * first_instr is aligned to a SpecializedCacheEntry.
 * The nth instruction is located at first_instr[n]
 * The nth cache is located at ((SpecializedCacheEntry *)first_instr)[-1-n]
 * The first (index 0) cache entry is reserved for the count, to enable finding
 * the first instruction from the base pointer.
 * The cache_count argument must include space for the count.
 * We use the SpecializedCacheOrInstruction union to refer to the data
 * to avoid type punning.

 Layout of quickened data, each line 8 bytes for M cache entries and N instructions:

 <cache_count>                              <---- co->co_quickened
 <cache M-1>
 <cache M-2>
 ...
 <cache 0>
 <instr 0> <instr 1> <instr 2> <instr 3>    <--- co->co_first_instr
 <instr 4> <instr 5> <instr 6> <instr 7>
 ...
 <instr N-1>
*/

Py_ssize_t _Py_QuickenedCount = 0;
#if SPECIALIZATION_STATS
SpecializationStats _specialization_stats = { 0 };

#define PRINT_STAT(name) fprintf(stderr, #name " : %" PRIu64" \n", _specialization_stats.name);
void
_Py_PrintSpecializationStats(void)
{
    PRINT_STAT(specialization_success);
    PRINT_STAT(specialization_failure);
    PRINT_STAT(loadattr_hit);
    PRINT_STAT(loadattr_deferred);
    PRINT_STAT(loadattr_miss);
    PRINT_STAT(loadattr_deopt);
}

#endif

static SpecializedCacheOrInstruction *
allocate(int cache_count, int instruction_count)
{
    assert(sizeof(SpecializedCacheOrInstruction) == 2*sizeof(int32_t));
    assert(sizeof(SpecializedCacheEntry) == 2*sizeof(int32_t));
    assert(cache_count > 0);
    assert(instruction_count > 0);
    int count = cache_count + (instruction_count + INSTRUCTIONS_PER_ENTRY -1)/INSTRUCTIONS_PER_ENTRY;
    SpecializedCacheOrInstruction *array = (SpecializedCacheOrInstruction *)
        PyMem_Malloc(sizeof(SpecializedCacheOrInstruction) * count);
    if (array == NULL) {
        PyErr_NoMemory();
        return NULL;
    }
    _Py_QuickenedCount++;
    array[0].entry.zero.cache_count = cache_count;
    return array;
}

static int
get_cache_count(SpecializedCacheOrInstruction *quickened) {
    return quickened[0].entry.zero.cache_count;
}

/* Map from opcode to adaptive opcode.
  Values of zero are ignored. */
static uint8_t adaptive_opcodes[256] = {
    [LOAD_ATTR] = LOAD_ATTR_ADAPTIVE,
};

/* The number of cache entries required for a "family" of instructions. */
static uint8_t cache_requirements[256] = {
    [LOAD_ATTR] = 2,
};

/* Return the oparg for the cache_offset and instruction index.
 *
 * If no cache is needed then return the original oparg.
 * If a cache is needed, but cannot be accessed because
 * oparg would be too large, then return -1.
 *
 * Also updates the cache_offset, as it may need to be incremented by
 * more than the cache requirements, if many instructions do not need caches.
 *
 * See pycore_code.h for details of how the cache offset,
 * instruction index and oparg are related */
static int
oparg_from_instruction_and_update_offset(int index, int opcode, int original_oparg, int *cache_offset) {
    /* The instruction pointer in the interpreter points to the next
     * instruction, so we compute the offset using nexti (index + 1) */
    int nexti = index + 1;
    uint8_t need = cache_requirements[opcode];
    if (need == 0) {
        return original_oparg;
    }
    assert(adaptive_opcodes[opcode] != 0);
    int oparg = oparg_from_offset_and_nexti(*cache_offset, nexti);
    assert(*cache_offset == offset_from_oparg_and_nexti(oparg, nexti));
    /* Some cache space is wasted here as the minimum possible offset is (nexti>>1) */
    if (oparg < 0) {
        oparg = 0;
        *cache_offset = offset_from_oparg_and_nexti(oparg, nexti);
    }
    else if (oparg > 255) {
        return -1;
    }
    *cache_offset += need;
    return oparg;
}

static int
entries_needed(_Py_CODEUNIT *code, int len)
{
    int cache_offset = 0;
    int previous_opcode = -1;
    for (int i = 0; i < len; i++) {
        uint8_t opcode = _Py_OPCODE(code[i]);
        if (previous_opcode != EXTENDED_ARG) {
            oparg_from_instruction_and_update_offset(i, opcode, 0, &cache_offset);
        }
        previous_opcode = opcode;
    }
    return cache_offset + 1;   // One extra for the count entry
}

static inline _Py_CODEUNIT *
first_instruction(SpecializedCacheOrInstruction *quickened)
{
    return &quickened[get_cache_count(quickened)].code[0];
}

/** Insert adaptive instructions and superinstructions.
 *
 * Skip instruction preceded by EXTENDED_ARG for adaptive
 * instructions as those are both very rare and tricky
 * to handle.
 */
static void
optimize(SpecializedCacheOrInstruction *quickened, int len)
{
    _Py_CODEUNIT *instructions = first_instruction(quickened);
    int cache_offset = 0;
    int previous_opcode = -1;
    for(int i = 0; i < len; i++) {
        int opcode = _Py_OPCODE(instructions[i]);
        int oparg = _Py_OPARG(instructions[i]);
        uint8_t adaptive_opcode = adaptive_opcodes[opcode];
        if (adaptive_opcode && previous_opcode != EXTENDED_ARG) {
            int new_oparg = oparg_from_instruction_and_update_offset(
                i, opcode, oparg, &cache_offset
            );
            if (new_oparg < 0) {
                /* Not possible to allocate a cache for this instruction */
                previous_opcode = opcode;
                continue;
            }
            instructions[i] = _Py_MAKECODEUNIT(adaptive_opcode, new_oparg);
            previous_opcode = adaptive_opcode;
            int entries_needed = cache_requirements[opcode];
            if (entries_needed) {
                /* Initialize the adpative cache entry */
                int cache0_offset = cache_offset-entries_needed;
                SpecializedCacheEntry *cache =
                    _GetSpecializedCacheEntry(instructions, cache0_offset);
                cache->adaptive.original_oparg = oparg;
                cache->adaptive.counter = 0;
            }
        }
        else {
            /* Super instructions don't use the cache,
             * so no need to update the offset. */
            switch (opcode) {
                case JUMP_ABSOLUTE:
                    instructions[i] = _Py_MAKECODEUNIT(JUMP_ABSOLUTE_QUICK, oparg);
                    break;
                /* Insert superinstructions here
                 E.g.
                case LOAD_FAST:
                    if (previous_opcode == LOAD_FAST)
                        instructions[i-1] = _Py_MAKECODEUNIT(LOAD_FAST__LOAD_FAST, oparg);
                 */
            }
            previous_opcode = opcode;
        }
    }
    assert(cache_offset+1 == get_cache_count(quickened));
}

int
_Py_Quicken(PyCodeObject *code) {
    if (code->co_quickened) {
        return 0;
    }
    Py_ssize_t size = PyBytes_GET_SIZE(code->co_code);
    int instr_count = (int)(size/sizeof(_Py_CODEUNIT));
    if (instr_count > MAX_SIZE_TO_QUICKEN) {
        code->co_warmup = QUICKENING_WARMUP_COLDEST;
        return 0;
    }
    int entry_count = entries_needed(code->co_firstinstr, instr_count);
    SpecializedCacheOrInstruction *quickened = allocate(entry_count, instr_count);
    if (quickened == NULL) {
        return -1;
    }
    _Py_CODEUNIT *new_instructions = first_instruction(quickened);
    memcpy(new_instructions, code->co_firstinstr, size);
    optimize(quickened, instr_count);
    code->co_quickened = quickened;
    code->co_firstinstr = new_instructions;
    return 0;
}

static int
specialize_module_load_attr(
    PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
    _PyAdaptiveEntry *cache0, _PyLoadAttrCache *cache1)
{
    PyModuleObject *m = (PyModuleObject *)owner;
    PyObject *value = NULL;
    PyObject *getattr;
    _Py_IDENTIFIER(__getattr__);
    PyDictObject *dict = (PyDictObject *)m->md_dict;
    if (dict == NULL) {
        return -1;
    }
    if (dict->ma_keys->dk_kind != DICT_KEYS_UNICODE) {
        return -1;
    }
    getattr = _PyUnicode_FromId(&PyId___getattr__); /* borrowed */
    if (getattr == NULL) {
        PyErr_Clear();
        return -1;
    }
    Py_ssize_t index = _PyDict_GetItemHint(dict, getattr, -1,  &value);
    assert(index != DKIX_ERROR);
    if (index != DKIX_EMPTY) {
        return -1;
    }
    index = _PyDict_GetItemHint(dict, name, -1, &value);
    assert (index != DKIX_ERROR);
    if (index != (uint16_t)index) {
        return -1;
    }
    uint32_t keys_version = _PyDictKeys_GetVersionForCurrentState(dict);
    if (keys_version == 0) {
        return -1;
    }
    cache1->dk_version_or_hint = keys_version;
    cache0->index = (uint16_t)index;
    *instr = _Py_MAKECODEUNIT(LOAD_ATTR_MODULE, _Py_OPARG(*instr));
    return 0;
}

int
_Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache)
{
    _PyAdaptiveEntry *cache0 = &cache->adaptive;
    _PyLoadAttrCache *cache1 = &cache[-1].load_attr;
    if (PyModule_CheckExact(owner)) {
        int err = specialize_module_load_attr(owner, instr, name, cache0, cache1);
        if (err) {
            goto fail;
        }
        goto success;
    }
    PyTypeObject *type = Py_TYPE(owner);
    if (type->tp_getattro != PyObject_GenericGetAttr) {
        goto fail;
    }
    if (type->tp_dict == NULL) {
        if (PyType_Ready(type) < 0) {
            return -1;
        }
    }
    PyObject *descr = _PyType_Lookup(type, name);
    if (descr != NULL) {
        // We found an attribute with a data-like descriptor.
        PyTypeObject *dtype = Py_TYPE(descr);
        if (dtype != &PyMemberDescr_Type) {
            goto fail;
        }
        // It's a slot
        PyMemberDescrObject *member = (PyMemberDescrObject *)descr;
        struct PyMemberDef *dmem = member->d_member;
        if (dmem->type != T_OBJECT_EX) {
            // It's a slot of a different type.  We don't handle those.
            goto fail;
        }
        Py_ssize_t offset = dmem->offset;
        if (offset != (uint16_t)offset) {
            goto fail;
        }
        assert(offset > 0);
        cache0->index = (uint16_t)offset;
        cache1->tp_version = type->tp_version_tag;
        *instr = _Py_MAKECODEUNIT(LOAD_ATTR_SLOT, _Py_OPARG(*instr));
        goto success;
    }
    // No desciptor
    if (type->tp_dictoffset <= 0) {
        // No dictionary, or computed offset dictionary
        goto fail;
    }
    PyObject **dictptr = (PyObject **) ((char *)owner + type->tp_dictoffset);
    if (*dictptr == NULL || !PyDict_CheckExact(*dictptr)) {
        goto fail;
    }
    // We found an instance with a __dict__.
    PyDictObject *dict = (PyDictObject *)*dictptr;
    if ((type->tp_flags & Py_TPFLAGS_HEAPTYPE)
        && dict->ma_keys == ((PyHeapTypeObject*)type)->ht_cached_keys
    ) {
        // Keys are shared
        assert(PyUnicode_CheckExact(name));
        Py_hash_t hash = PyObject_Hash(name);
        if (hash == -1) {
            return -1;
        }
        PyObject *value;
        Py_ssize_t index = _Py_dict_lookup(dict, name, hash, &value);
        assert (index != DKIX_ERROR);
        if (index != (uint16_t)index) {
            goto fail;
        }
        uint32_t keys_version = _PyDictKeys_GetVersionForCurrentState(dict);
        if (keys_version == 0) {
            goto fail;
        }
        cache1->dk_version_or_hint = keys_version;
        cache1->tp_version = type->tp_version_tag;
        cache0->index = (uint16_t)index;
        *instr = _Py_MAKECODEUNIT(LOAD_ATTR_SPLIT_KEYS, _Py_OPARG(*instr));
        goto success;
    }
    else {
        PyObject *value = NULL;
        Py_ssize_t hint =
            _PyDict_GetItemHint(dict, name, -1, &value);
        if (hint != (uint32_t)hint) {
            goto fail;
        }
        cache1->dk_version_or_hint = (uint32_t)hint;
        cache1->tp_version = type->tp_version_tag;
        *instr = _Py_MAKECODEUNIT(LOAD_ATTR_WITH_HINT, _Py_OPARG(*instr));
        goto success;
    }

fail:
    STAT_INC(specialization_failure);
    assert(!PyErr_Occurred());
    cache_backoff(cache0);
    return 0;
success:
    STAT_INC(specialization_success);
    assert(!PyErr_Occurred());
    cache0->counter = saturating_start();
    return 0;
}