hakmem/core/front/tiny_ring_cache.h

// tiny_ring_cache.h - Phase 21-1: Array-based hot cache (C2/C3 only)
//
// Goal: Eliminate pointer chasing in TLS SLL by using ring buffer
// Target: +15-20% performance (54.4M → 62-65M ops/s)
//
// Design (ChatGPT feedback):
//   - Ring → SLL → SuperSlab (3-layer hierarchy)
//   - Ring size: 128 slots (ENV: 64/128/256 A/B test)
//   - C2/C3 only (hot classes, 33-128B)
//   - Replaces UltraHot (Phase 19-3: +12.9% by removing UltraHot)
//
// Performance:
//   - Alloc: 1-2 instructions (array access, no pointer chasing)
//   - Free: 1-2 instructions (array write, no pointer chasing)
//   - vs TLS SLL: 3 mem accesses → 2 mem accesses, 1 cache miss → 0
//
// ENV Variables:
//   HAKMEM_TINY_HOT_RING_ENABLE=1  # Enable Ring cache (default: 0)
//   HAKMEM_TINY_HOT_RING_C2=128    # C2 ring size (default: 128)
//   HAKMEM_TINY_HOT_RING_C3=128    # C3 ring size (default: 128)
//   HAKMEM_TINY_HOT_RING_CASCADE=1 # Enable SLL → Ring refill (default: 0)

#ifndef HAK_FRONT_TINY_RING_CACHE_H
#define HAK_FRONT_TINY_RING_CACHE_H

#include <stdint.h>
#include <stdlib.h>
#include <stdio.h>
#include "../hakmem_build_flags.h"

// ============================================================================
// Ring Buffer Structure
// ============================================================================

typedef struct {
    void** slots;      // Dynamic array (allocated at init, power-of-2 size)
    uint16_t head;     // Pop index (consumer)
    uint16_t tail;     // Push index (producer)
    uint16_t capacity; // Ring size (power of 2 for fast modulo: & (capacity-1))
    uint16_t mask;     // Capacity - 1 (for fast modulo)
} TinyRingCache;

// ============================================================================
// External TLS Variables (defined in hakmem_tiny.c)
// ============================================================================

extern __thread TinyRingCache g_ring_cache_c2;
extern __thread TinyRingCache g_ring_cache_c3;

// ============================================================================
// ENV Control (cached, lazy init)
// ============================================================================

// Enable flag (default: 0, OFF)
static inline int ring_cache_enabled(void) {
    static int g_enable = -1;
    if (__builtin_expect(g_enable == -1, 0)) {
        const char* e = getenv("HAKMEM_TINY_HOT_RING_ENABLE");
        g_enable = (e && *e && *e != '0') ? 1 : 0;
#if !HAKMEM_BUILD_RELEASE
        if (g_enable) {
            fprintf(stderr, "[Ring-INIT] ring_cache_enabled() = %d\n", g_enable);
            fflush(stderr);
        }
#endif
    }
    return g_enable;
}

// C2 capacity (default: 128)
static inline size_t ring_capacity_c2(void) {
    static size_t g_cap = 0;
    if (__builtin_expect(g_cap == 0, 0)) {
        const char* e = getenv("HAKMEM_TINY_HOT_RING_C2");
        g_cap = (e && *e) ? (size_t)atoi(e) : 128;  // Default: 128

        // Round up to power of 2 (for fast modulo)
        if (g_cap < 32) g_cap = 32;
        if (g_cap > 256) g_cap = 256;

        // Ensure power of 2
        size_t pow2 = 32;
        while (pow2 < g_cap) pow2 *= 2;
        g_cap = pow2;

#if !HAKMEM_BUILD_RELEASE
        fprintf(stderr, "[Ring-INIT] C2 capacity = %zu (power of 2)\n", g_cap);
        fflush(stderr);
#endif
    }
    return g_cap;
}

// C3 capacity (default: 128)
static inline size_t ring_capacity_c3(void) {
    static size_t g_cap = 0;
    if (__builtin_expect(g_cap == 0, 0)) {
        const char* e = getenv("HAKMEM_TINY_HOT_RING_C3");
        g_cap = (e && *e) ? (size_t)atoi(e) : 128;  // Default: 128

        // Round up to power of 2
        if (g_cap < 32) g_cap = 32;
        if (g_cap > 256) g_cap = 256;

        size_t pow2 = 32;
        while (pow2 < g_cap) pow2 *= 2;
        g_cap = pow2;

#if !HAKMEM_BUILD_RELEASE
        fprintf(stderr, "[Ring-INIT] C3 capacity = %zu (power of 2)\n", g_cap);
        fflush(stderr);
#endif
    }
    return g_cap;
}

// Cascade enable flag (default: 0, OFF)
static inline int ring_cascade_enabled(void) {
    static int g_enable = -1;
    if (__builtin_expect(g_enable == -1, 0)) {
        const char* e = getenv("HAKMEM_TINY_HOT_RING_CASCADE");
        g_enable = (e && *e && *e != '0') ? 1 : 0;
#if !HAKMEM_BUILD_RELEASE
        if (g_enable) {
            fprintf(stderr, "[Ring-INIT] ring_cascade_enabled() = %d\n", g_enable);
            fflush(stderr);
        }
#endif
    }
    return g_enable;
}

// ============================================================================
// Init/Shutdown Forward Declarations (needed by pop/push)
// ============================================================================

void ring_cache_init(void);
void ring_cache_shutdown(void);
void ring_cache_print_stats(void);

// ============================================================================
// Ultra-Fast Pop/Push (1-2 instructions)
// ============================================================================

// Pop from ring (alloc fast path)
// Returns: BASE pointer (caller must convert to USER with +1)
static inline void* ring_cache_pop(int class_idx) {
    // Fast path: Ring disabled or wrong class → return NULL immediately
    if (__builtin_expect(!ring_cache_enabled(), 0)) return NULL;
    if (__builtin_expect(class_idx != 2 && class_idx != 3, 0)) return NULL;

    TinyRingCache* ring = (class_idx == 2) ? &g_ring_cache_c2 : &g_ring_cache_c3;

    // Lazy init check (once per thread)
    if (__builtin_expect(ring->slots == NULL, 0)) {
        ring_cache_init();  // First call in this thread
        // Re-check after init (may fail if allocation failed)
        if (ring->slots == NULL) return NULL;
    }

    // Empty check
    if (__builtin_expect(ring->head == ring->tail, 0)) {
        return NULL;  // Empty
    }

    // Pop from head (consumer)
    void* base = ring->slots[ring->head];
    ring->head = (ring->head + 1) & ring->mask;  // Fast modulo (power of 2)

    return base;  // Return BASE pointer
}

// Push to ring (free fast path)
// Input: BASE pointer (caller must pass BASE, not USER)
// Returns: 1=SUCCESS, 0=FULL
static inline int ring_cache_push(int class_idx, void* base) {
    // Fast path: Ring disabled or wrong class → return 0 (not handled)
    if (__builtin_expect(!ring_cache_enabled(), 0)) return 0;
    if (__builtin_expect(class_idx != 2 && class_idx != 3, 0)) return 0;

    TinyRingCache* ring = (class_idx == 2) ? &g_ring_cache_c2 : &g_ring_cache_c3;

    // Lazy init check (once per thread)
    if (__builtin_expect(ring->slots == NULL, 0)) {
        ring_cache_init();  // First call in this thread
        // Re-check after init (may fail if allocation failed)
        if (ring->slots == NULL) return 0;
    }

    uint16_t next_tail = (ring->tail + 1) & ring->mask;

    // Full check (leave 1 slot empty to distinguish full/empty)
    if (__builtin_expect(next_tail == ring->head, 0)) {
        return 0;  // Full
    }

    // Push to tail (producer)
    ring->slots[ring->tail] = base;
    ring->tail = next_tail;

    return 1;  // SUCCESS
}

// ============================================================================
// Refill from TLS SLL (cascade, Phase 21-1-C)
// ============================================================================

// Forward declaration (defined in tiny_ring_cache.c)
int ring_refill_from_sll(int class_idx, int target_count);

#endif // HAK_FRONT_TINY_RING_CACHE_H