Major Features: - Debug counter infrastructure for Refill Stage tracking - Free Pipeline counters (ss_local, ss_remote, tls_sll) - Diagnostic counters for early return analysis - Unified larson.sh benchmark runner with profiles - Phase 6-3 regression analysis documentation Bug Fixes: - Fix SuperSlab disabled by default (HAKMEM_TINY_USE_SUPERSLAB) - Fix profile variable naming consistency - Add .gitignore patterns for large files Performance: - Phase 6-3: 4.79 M ops/s (has OOM risk) - With SuperSlab: 3.13 M ops/s (+19% improvement) This is a clean repository without large log files. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
148 lines
5.1 KiB
C
148 lines
5.1 KiB
C
// Inline helpers for Tiny HotMag (small TLS magazine for hot classes)
|
|
// This header is textually included from hakmem_tiny.c after the following
|
|
// symbols are defined:
|
|
// - static int g_hotmag_enable;
|
|
// - static __thread TinyHotMag g_tls_hot_mag[TINY_NUM_CLASSES];
|
|
// - tiny_mag_init_if_needed(int)
|
|
// - g_tls_sll_head[], g_tls_sll_count[], g_tls_mags[]
|
|
|
|
static inline int hkm_is_hot_class(int class_idx) {
|
|
return class_idx >= 0 && class_idx <= 3 && g_hotmag_class_en[class_idx];
|
|
}
|
|
|
|
static inline uint16_t hotmag_effective_cap(int class_idx) {
|
|
if (!hkm_is_hot_class(class_idx)) return 0;
|
|
int cap = g_hotmag_cap_current[class_idx];
|
|
if (!g_hotmag_cap_locked[class_idx] || cap <= 0) {
|
|
cap = g_hotmag_cap_default;
|
|
}
|
|
if (cap < 16) cap = 16;
|
|
if (cap > 1024) cap = 1024;
|
|
return (uint16_t)cap;
|
|
}
|
|
|
|
static inline uint16_t hotmag_refill_target(int class_idx) {
|
|
if (!hkm_is_hot_class(class_idx)) return 0;
|
|
int refill = g_hotmag_refill_current[class_idx];
|
|
if (!g_hotmag_refill_locked[class_idx] && refill <= 0) {
|
|
refill = g_hotmag_refill_default;
|
|
}
|
|
if (refill < 0) refill = 0;
|
|
uint16_t cap = hotmag_effective_cap(class_idx);
|
|
if ((uint16_t)refill > cap) refill = cap;
|
|
return (uint16_t)refill;
|
|
}
|
|
|
|
static inline void hotmag_init_if_needed(int class_idx) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx)) return;
|
|
TinyHotMag* hm = &g_tls_hot_mag[class_idx];
|
|
uint16_t desired = hotmag_effective_cap(class_idx);
|
|
if (hm->cap != desired) {
|
|
hm->cap = desired;
|
|
if (hm->top > hm->cap) hm->top = hm->cap;
|
|
}
|
|
}
|
|
|
|
static inline void* hotmag_pop(int class_idx) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx)) return NULL;
|
|
TinyHotMag* hm = &g_tls_hot_mag[class_idx];
|
|
if (__builtin_expect(hm->top > 0, 1)) {
|
|
return hm->slots[--hm->top];
|
|
}
|
|
return NULL;
|
|
}
|
|
|
|
static inline int hotmag_push(int class_idx, void* ptr) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx)) return 0;
|
|
TinyHotMag* hm = &g_tls_hot_mag[class_idx];
|
|
hotmag_init_if_needed(class_idx);
|
|
if (__builtin_expect(hm->top < hm->cap, 1)) {
|
|
hm->slots[hm->top++] = ptr;
|
|
return 1;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
static inline int hotmag_refill_from_sll(int class_idx, int max_take) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx) || max_take <= 0) return 0;
|
|
TinyHotMag* hm = &g_tls_hot_mag[class_idx];
|
|
int room = (int)hm->cap - (int)hm->top;
|
|
if (room <= 0) return 0;
|
|
if (max_take > room) max_take = room;
|
|
int taken = 0;
|
|
while (taken < max_take) {
|
|
void* head = g_tls_sll_head[class_idx];
|
|
if (!head) break;
|
|
g_tls_sll_head[class_idx] = *(void**)head;
|
|
if (g_tls_sll_count[class_idx] > 0) g_tls_sll_count[class_idx]--;
|
|
hm->slots[hm->top++] = head;
|
|
taken++;
|
|
}
|
|
return taken;
|
|
}
|
|
|
|
static inline int hotmag_refill_from_mag(int class_idx, int max_take) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx) || max_take <= 0) return 0;
|
|
tiny_mag_init_if_needed(class_idx);
|
|
TinyTLSMag* mag = &g_tls_mags[class_idx];
|
|
TinyHotMag* hm = &g_tls_hot_mag[class_idx];
|
|
int room = (int)hm->cap - (int)hm->top;
|
|
if (room <= 0 || mag->top <= 0) return 0;
|
|
if (max_take > room) max_take = room;
|
|
int take = mag->top < max_take ? mag->top : max_take;
|
|
for (int i = 0; i < take; i++) {
|
|
hm->slots[hm->top++] = mag->items[--mag->top].ptr;
|
|
}
|
|
return take;
|
|
}
|
|
|
|
static inline int hotmag_try_refill(int class_idx, TinyHotMag* hm) {
|
|
if (!g_hotmag_enable || !hkm_is_hot_class(class_idx)) return 0;
|
|
uint16_t target = hotmag_refill_target(class_idx);
|
|
if (target == 0) return 0;
|
|
|
|
int room = (int)hm->cap - (int)hm->top;
|
|
if (room <= 0) return 0;
|
|
if (target > (uint16_t)room) target = (uint16_t)room;
|
|
|
|
int filled = 0;
|
|
|
|
if (g_tls_list_enable && target > 0) {
|
|
TinyTLSList* tls = &g_tls_lists[class_idx];
|
|
uint32_t seq = atomic_load_explicit(&g_tls_param_seq[class_idx], memory_order_relaxed);
|
|
if (__builtin_expect(seq != g_tls_param_seen[class_idx], 0)) {
|
|
tiny_tls_refresh_params(class_idx, tls);
|
|
}
|
|
void* chain_head = NULL;
|
|
void* chain_tail = NULL;
|
|
uint32_t taken = tls_list_bulk_take(tls, target, &chain_head, &chain_tail);
|
|
if (taken > 0u) {
|
|
void* node = chain_head;
|
|
for (uint32_t i = 0; i < taken && node; i++) {
|
|
void* next = *(void**)node;
|
|
hm->slots[hm->top++] = node;
|
|
node = next;
|
|
}
|
|
filled += (int)taken;
|
|
room = (int)hm->cap - (int)hm->top;
|
|
target -= (uint16_t)taken;
|
|
}
|
|
}
|
|
|
|
if (target > 0 && g_tls_sll_enable && room > 0) {
|
|
int from_sll = hotmag_refill_from_sll(class_idx, target);
|
|
filled += from_sll;
|
|
room -= from_sll;
|
|
if (room < 0) room = 0;
|
|
if (target > from_sll) target -= (uint16_t)((from_sll < 0) ? 0 : from_sll);
|
|
else target = 0;
|
|
}
|
|
|
|
if (target > 0 && room > 0) {
|
|
int from_mag = hotmag_refill_from_mag(class_idx, target);
|
|
filled += from_mag;
|
|
}
|
|
|
|
return filled;
|
|
}
|