ollama/llama/ggml-impl.h

/**
 * llama.cpp - commit 3f1ae2e32cde00c39b96be6d01c2997c29bae555 - do not edit this file
 *
 * MIT License
 *
 * Copyright (c) 2023-2024 The ggml authors
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#pragma once

// GGML internal header

#include "ggml.h"

#include <assert.h>
#include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
#include <stdbool.h>
#include <stdint.h>

#ifdef __cplusplus
extern "C" {
#endif

#undef MIN
#undef MAX

#define MIN(a, b) ((a) < (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))

// static_assert should be a #define, but if it's not,
// fall back to the _Static_assert C11 keyword.
// if C99 - static_assert is noop
// ref: https://stackoverflow.com/a/53923785/4039976
#ifndef __cplusplus
#ifndef static_assert
#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201100L)
#define static_assert(cond, msg) _Static_assert(cond, msg)
#else
#define static_assert(cond, msg) struct global_scope_noop_trick
#endif
#endif
#endif

// bitset

typedef uint32_t ggml_bitset_t;

static_assert(sizeof(ggml_bitset_t) == 4, "bitset_t constants must be updated");
#define BITSET_SHR 5 // log2(sizeof(ggml_bitset_t)*8)
#define BITSET_MASK (sizeof(ggml_bitset_t)*8 - 1)

static size_t ggml_bitset_size(size_t n) {
    return (n + BITSET_MASK) >> BITSET_SHR;
}

static inline bool ggml_bitset_get(const ggml_bitset_t * bitset, size_t i) {
    return !!(bitset[i >> BITSET_SHR] & (1u << (i & BITSET_MASK)));
}

static inline void ggml_bitset_set(ggml_bitset_t * bitset, size_t i) {
    bitset[i >> BITSET_SHR] |= (1u << (i & BITSET_MASK));
}

static inline void ggml_bitset_clear(ggml_bitset_t * bitset, size_t i) {
    bitset[i >> BITSET_SHR] &= ~(1u << (i & BITSET_MASK));
}

// hash set

#define GGML_HASHSET_FULL ((size_t)-1)
#define GGML_HASHSET_ALREADY_EXISTS ((size_t)-2)

struct ggml_hash_set {
    size_t size;
    ggml_bitset_t * used;       // whether or not the keys are in use i.e. set
    struct ggml_tensor ** keys; // actual tensors in the set, keys[i] is only defined if ggml_bitset_get(used, i)
};

struct ggml_hash_set ggml_hash_set_new(size_t size);
void                 ggml_hash_set_free(struct ggml_hash_set * hash_set);

// returns the minimum size for a hash set that can hold min_sz elements
size_t ggml_hash_size(size_t min_sz);

// remove all elements from the hash set
void ggml_hash_set_reset(struct ggml_hash_set * hash_set);

// returns true if key is in the hash set
static bool ggml_hash_contains(const struct ggml_hash_set * hash_set, struct ggml_tensor * key);

// returns GGML_HASHSET_FULL if table is full, otherwise the current index of the key or where it should be inserted
static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, struct ggml_tensor * key);

// returns GGML_HASHSET_ALREADY_EXISTS if key already exists, index otherwise, asserts if table is full
static size_t ggml_hash_insert(struct ggml_hash_set * hash_set, struct ggml_tensor * key);

// return index, asserts if table is full
static size_t ggml_hash_find_or_insert(struct ggml_hash_set * hash_set, struct ggml_tensor * key);

// hash function for ggml_tensor
static inline size_t ggml_hash(const struct ggml_tensor * p) {
    // the last 4 bits are always zero due to alignment
    return (size_t)(uintptr_t)p >> 4;
}

static size_t ggml_hash_find(const struct ggml_hash_set * hash_set, struct ggml_tensor * key) {
    size_t h = ggml_hash(key) % hash_set->size;

    // linear probing
    size_t i = h;
    while (ggml_bitset_get(hash_set->used, i) && hash_set->keys[i] != key) {
        i = (i + 1) % hash_set->size;
        if (i == h) {
            // visited all hash table entries -> not found
            return GGML_HASHSET_FULL;
        }
    }
    return i;
}

static bool ggml_hash_contains(const struct ggml_hash_set * hash_set, struct ggml_tensor * key) {
    size_t i = ggml_hash_find(hash_set, key);
    return i != GGML_HASHSET_FULL && ggml_bitset_get(hash_set->used, i);
}

static size_t ggml_hash_insert(struct ggml_hash_set * hash_set, struct ggml_tensor * key) {
    size_t h = ggml_hash(key) % hash_set->size;

    // linear probing
    size_t i = h;
    do {
        if (!ggml_bitset_get(hash_set->used, i)) {
            ggml_bitset_set(hash_set->used, i);
            hash_set->keys[i] = key;
            return i;
        }
        if (hash_set->keys[i] == key) {
            return GGML_HASHSET_ALREADY_EXISTS;
        }
        i = (i + 1) % hash_set->size;
    } while (i != h);

    // visited all hash table entries -> not found
    GGML_ABORT("fatal error");
}

static size_t ggml_hash_find_or_insert(struct ggml_hash_set * hash_set, struct ggml_tensor * key) {
    size_t h = ggml_hash(key) % hash_set->size;

    // linear probing
    size_t i = h;
    do {
        if (!ggml_bitset_get(hash_set->used, i)) {
            ggml_bitset_set(hash_set->used, i);
            hash_set->keys[i] = key;
            return i;
        }
        if (hash_set->keys[i] == key) {
            return i;
        }
        i = (i + 1) % hash_set->size;
    } while (i != h);

    // visited all hash table entries -> not found
    GGML_ABORT("fatal error");
}

// computation graph

enum ggml_cgraph_eval_order {
    GGML_CGRAPH_EVAL_ORDER_LEFT_TO_RIGHT = 0,
    GGML_CGRAPH_EVAL_ORDER_RIGHT_TO_LEFT,
    GGML_CGRAPH_EVAL_ORDER_COUNT
};

struct ggml_cgraph {
    int size;
    int n_nodes;
    int n_leafs;

    struct ggml_tensor ** nodes;
    struct ggml_tensor ** grads;
    struct ggml_tensor ** leafs;

    struct ggml_hash_set visited_hash_set;

    enum ggml_cgraph_eval_order order;
};

struct ggml_cgraph ggml_graph_view(struct ggml_cgraph * cgraph, int i0, int i1);

#ifdef __cplusplus
}
#endif