// mid_tcache.h - Mid-size TLS tcache (classes 4..7)
// Box: Per-thread, per-class singly-linked stack. No atomics, no sharing.
// Goal: O(1) pop/push on alloc/free for 128..1024B while keeping Box boundaries.

#pragma once
#include <stdint.h>
#include <stdlib.h>

#ifndef TINY_NUM_CLASSES
#define TINY_NUM_CLASSES 8
#endif

static inline int midtc_enabled(void) {
    static int en = -1;
    if (__builtin_expect(en == -1, 0)) {
        const char* s = getenv("HAKMEM_MID_TC");
        en = (s && *s && *s != '0') ? 1 : 0;
    }
    return en;
}

static inline int midtc_cap_global(void) {
    static int cap = -1;
    if (__builtin_expect(cap == -1, 0)) {
        const char* s = getenv("HAKMEM_MID_TC_CAP");
        int v = (s && *s) ? atoi(s) : 32; // conservative default
        if (v < 0) v = 0;
        if (v > 1024) v = 1024;
        cap = v;
    }
    return cap;
}

// Per-thread state
static __thread void* g_midtc_head[TINY_NUM_CLASSES];
static __thread uint16_t g_midtc_count[TINY_NUM_CLASSES];

// Push returns 1 on accept, 0 to fallback to existing path.
static inline int midtc_push(int class_idx, void* p) {
    if (!midtc_enabled()) return 0;
    if (class_idx < 4 || class_idx >= TINY_NUM_CLASSES) return 0; // only 128..1024B
    uint16_t cnt = g_midtc_count[class_idx];
    int cap = midtc_cap_global();
    if (cnt >= (uint16_t)cap) return 0;
    *(void**)p = g_midtc_head[class_idx];
    g_midtc_head[class_idx] = p;
    g_midtc_count[class_idx] = (uint16_t)(cnt + 1);
    return 1;
}

static inline void* midtc_pop(int class_idx) {
    if (!midtc_enabled()) return NULL;
    if (class_idx < 4 || class_idx >= TINY_NUM_CLASSES) return NULL;
    void* h = g_midtc_head[class_idx];
    if (!h) return NULL;
    g_midtc_head[class_idx] = *(void**)h;
    if (g_midtc_count[class_idx] > 0) g_midtc_count[class_idx]--;
    return h;
}