Files
hakmem/core/pool_tls_remote.c

129 lines
4.0 KiB
C
Raw Normal View History

#include "pool_tls_remote.h"
#include <pthread.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <unistd.h>
#include <stdatomic.h>
Front-Direct implementation: SS→FC direct refill + SLL complete bypass ## Summary Implemented Front-Direct architecture with complete SLL bypass: - Direct SuperSlab → FastCache refill (1-hop, bypasses SLL) - SLL-free allocation/free paths when Front-Direct enabled - Legacy path sealing (SLL inline opt-in, SFC cascade ENV-only) ## New Modules - core/refill/ss_refill_fc.h (236 lines): Standard SS→FC refill entry point - Remote drain → Freelist → Carve priority - Header restoration for C1-C6 (NOT C0/C7) - ENV: HAKMEM_TINY_P0_DRAIN_THRESH, HAKMEM_TINY_P0_NO_DRAIN - core/front/fast_cache.h: FastCache (L1) type definition - core/front/quick_slot.h: QuickSlot (L0) type definition ## Allocation Path (core/tiny_alloc_fast.inc.h) - Added s_front_direct_alloc TLS flag (lazy ENV check) - SLL pop guarded by: g_tls_sll_enable && !s_front_direct_alloc - Refill dispatch: - Front-Direct: ss_refill_fc_fill() → fastcache_pop() (1-hop) - Legacy: sll_refill_batch_from_ss() → SLL → FC (2-hop, A/B only) - SLL inline pop sealed (requires HAKMEM_TINY_INLINE_SLL=1 opt-in) ## Free Path (core/hakmem_tiny_free.inc, core/hakmem_tiny_fastcache.inc.h) - FC priority: Try fastcache_push() first (same-thread free) - tiny_fast_push() bypass: Returns 0 when s_front_direct_free || !g_tls_sll_enable - Fallback: Magazine/slow path (safe, bypasses SLL) ## Legacy Sealing - SFC cascade: Default OFF (ENV-only via HAKMEM_TINY_SFC_CASCADE=1) - Deleted: core/hakmem_tiny_free.inc.bak, core/pool_refill_legacy.c.bak - Documentation: ss_refill_fc_fill() promoted as CANONICAL refill entry ## ENV Controls - HAKMEM_TINY_FRONT_DIRECT=1: Enable Front-Direct (SS→FC direct) - HAKMEM_TINY_P0_DIRECT_FC_ALL=1: Same as above (alt name) - HAKMEM_TINY_REFILL_BATCH=1: Enable batch refill (also enables Front-Direct) - HAKMEM_TINY_SFC_CASCADE=1: Enable SFC cascade (default OFF) - HAKMEM_TINY_INLINE_SLL=1: Enable inline SLL pop (default OFF, requires AGGRESSIVE_INLINE) ## Benchmarks (Front-Direct Enabled) ```bash ENV: HAKMEM_BENCH_FAST_FRONT=1 HAKMEM_TINY_FRONT_DIRECT=1 HAKMEM_TINY_REFILL_BATCH=1 HAKMEM_TINY_P0_DIRECT_FC_ALL=1 HAKMEM_TINY_REFILL_COUNT_HOT=256 HAKMEM_TINY_REFILL_COUNT_MID=96 HAKMEM_TINY_BUMP_CHUNK=256 bench_random_mixed (16-1040B random, 200K iter): 256 slots: 1.44M ops/s (STABLE, 0 SEGV) 128 slots: 1.44M ops/s (STABLE, 0 SEGV) bench_fixed_size (fixed size, 200K iter): 256B: 4.06M ops/s (has debug logs, expected >10M without logs) 128B: Similar (debug logs affect) ``` ## Verification - TRACE_RING test (10K iter): **0 SLL events** detected ✅ - Complete SLL bypass confirmed when Front-Direct=1 - Stable execution: 200K iterations × multiple sizes, 0 SEGV ## Next Steps - Disable debug logs in hak_alloc_api.inc.h (call_num 14250-14280 range) - Re-benchmark with clean Release build (target: 10-15M ops/s) - 128/256B shortcut path optimization (FC hit rate improvement) Co-Authored-By: ChatGPT <chatgpt@openai.com> Suggested-By: ultrathink
2025-11-14 05:41:49 +09:00
#include "box/tiny_next_ptr_box.h" // Box API: preserve header by using class-aware next offset
#define REMOTE_BUCKETS 256
// Lock-free MPSC queue per class
typedef struct RemoteQueue {
_Atomic(void*) head; // Atomic head for lock-free push (LIFO)
_Atomic uint32_t count; // Approximate count
} RemoteQueue;
typedef struct RemoteRec {
int tid;
RemoteQueue queues[7]; // One queue per class (lock-free)
struct RemoteRec* next;
} RemoteRec;
static RemoteRec* g_buckets[REMOTE_BUCKETS];
static pthread_mutex_t g_locks[REMOTE_BUCKETS]; // Only for RemoteRec creation
static pthread_once_t g_once = PTHREAD_ONCE_INIT;
static void rq_init(void){
for (int i=0;i<REMOTE_BUCKETS;i++) pthread_mutex_init(&g_locks[i], NULL);
}
static inline unsigned hb(int tid){ return (unsigned)tid & (REMOTE_BUCKETS-1); }
int pool_remote_push(int class_idx, void* ptr, int owner_tid){
if (class_idx < 0 || class_idx > 6 || ptr == NULL) return 0;
pthread_once(&g_once, rq_init);
unsigned b = hb(owner_tid);
// Find or create RemoteRec (only this part needs mutex)
RemoteRec* r = g_buckets[b];
while (r && r->tid != owner_tid) r = r->next;
if (!r){
pthread_mutex_lock(&g_locks[b]);
// Double-check after acquiring lock
r = g_buckets[b];
while (r && r->tid != owner_tid) r = r->next;
if (!r){
r = (RemoteRec*)calloc(1, sizeof(RemoteRec));
r->tid = owner_tid;
r->next = g_buckets[b];
g_buckets[b] = r;
}
pthread_mutex_unlock(&g_locks[b]);
}
// Lock-free push using CAS (this is the hot path!)
RemoteQueue* q = &r->queues[class_idx];
void* old_head = atomic_load_explicit(&q->head, memory_order_relaxed);
do {
// Link new node to current head using Box API (preserves header)
tiny_next_write(class_idx, ptr, old_head);
} while (!atomic_compare_exchange_weak_explicit(
&q->head, &old_head, ptr,
memory_order_release, memory_order_relaxed));
atomic_fetch_add_explicit(&q->count, 1, memory_order_relaxed);
return 1;
}
// Drain up to a small batch for this thread and class
int pool_remote_pop_chain(int class_idx, int max_take, void** out_chain){
if (class_idx < 0 || class_idx > 6 || out_chain==NULL) return 0;
pthread_once(&g_once, rq_init);
int mytid = (int)syscall(SYS_gettid);
unsigned b = hb(mytid);
// Find my RemoteRec (no lock needed for reading)
RemoteRec* r = g_buckets[b];
while (r && r->tid != mytid) r = r->next;
if (!r) return 0; // No remote queue for this thread
// Lock-free pop using atomic exchange
RemoteQueue* q = &r->queues[class_idx];
void* head = atomic_exchange_explicit(&q->head, NULL, memory_order_acquire);
if (!head) return 0; // Queue was empty
// Count nodes and take up to max_take (traverse LIFO chain)
if (max_take <= 0) max_take = 32;
void* chain = NULL;
void* tail = NULL;
int batch = 0;
// Pop up to max_take from the stolen chain
while (head && batch < max_take){
void* nxt = tiny_next_read(class_idx, head);
// Build output chain (reverse for FIFO order)
if (!chain){
chain = head;
tail = head;
} else {
tiny_next_write(class_idx, tail, head);
tail = head;
}
head = nxt;
batch++;
}
// If we didn't take all nodes, push remainder back (lock-free)
if (head){
void* old_head = atomic_load_explicit(&q->head, memory_order_relaxed);
do {
// Find tail of remainder chain
void* remainder_tail = head;
while (tiny_next_read(class_idx, remainder_tail)) {
remainder_tail = tiny_next_read(class_idx, remainder_tail);
}
// Link remainder to current head
tiny_next_write(class_idx, remainder_tail, old_head);
} while (!atomic_compare_exchange_weak_explicit(
&q->head, &old_head, head,
memory_order_release, memory_order_relaxed));
}
atomic_fetch_sub_explicit(&q->count, batch, memory_order_relaxed);
*out_chain = chain;
return batch;
}