feat(heap,pool): optimizations

Author: singaraiona

core/heap.h

@@ -31,9 +31,9 @@
 #define MAX_BLOCK_ORDER 25  // 2^25 = 32MB
 #define MAX_POOL_ORDER 38   // 2^38 = 256GB
 
-// Small object cache (slab) for sizes 32, 64, 128, 256 bytes
-#define SLAB_CACHE_SIZE 32
-#define SLAB_ORDERS 4  // orders 5, 6, 7, 8
+// Small object cache (slab) for sizes 32, 64, 128, 256, 512 bytes
+#define SLAB_CACHE_SIZE 64
+#define SLAB_ORDERS 5  // orders 5, 6, 7, 8, 9
 
 // Memory modes
 #define MMOD_INTERNAL 0xff

core/pool.c

@@ -180,8 +180,20 @@ raw_p executor_run(raw_p arg) {
     __atomic_store_n(&executor->vm, vm, __ATOMIC_RELAXED);
 
     for (;;) {
+        // Adaptive spin-wait: spin briefly before sleeping (reduces wake latency)
+        i64_t spin_rounds = 0;
+        while (spin_rounds < 1000) {
+            if (__atomic_load_n(&executor->pool->tasks_count, __ATOMIC_ACQUIRE) > 0)
+                break;
+            if (__atomic_load_n(&executor->pool->state, __ATOMIC_ACQUIRE) == RUN_STATE_STOPPED)
+                break;
+            backoff_spin(&spin_rounds);
+        }
+
         mutex_lock(&executor->pool->mutex);
-        cond_wait(&executor->pool->run, &executor->pool->mutex);
+        // Re-check after acquiring lock
+        if (executor->pool->tasks_count == 0 && executor->pool->state == RUN_STATE_RUNNING)
+            cond_wait(&executor->pool->run, &executor->pool->mutex);
 
         if (executor->pool->state == RUN_STATE_STOPPED) {
             mutex_unlock(&executor->pool->mutex);