ed/raddebugger
1
0
Fork 0
mirror of https://github.com/Ed94/raddebugger.git synced 2026-10-08 17:05:43 +00:00
Code Issues Packages Projects Releases Wiki Activity

better artifact cache cancellation thread work organization - only do more frequent cancellation scans when in a tick, otherwise sleep

Browse Source
This commit is contained in:
Ryan Fleury
2025-10-23 14:39:44 -07:00
parent 721649b7ea
commit ae17084b44
4 changed files with 63 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/artifact_cache/artifact_cache.c
+46 -28
View File
@@ -19,6 +19,8 @@ ac_init(void)
ac_shared->req_batches[idx].arena = arena_alloc();
}
ac_shared->cancel_thread = thread_launch(ac_cancel_thread_entry_point, 0);
ac_shared->cancel_thread_mutex = mutex_alloc();
mutex_take(ac_shared->cancel_thread_mutex);
}
////////////////////////////////
@@ -211,6 +213,14 @@ ac_async_tick(void)
{
Temp scratch = scratch_begin(0, 0);
//////////////////////////////
//- rjf: enable cancellation scanning
//
if(lane_idx() == 0)
{
mutex_drop(ac_shared->cancel_thread_mutex);
}
//////////////////////////////
//- rjf: do eviction pass across all caches
//
@@ -575,6 +585,13 @@ ac_async_tick(void)
}
lane_sync();
//////////////////////////////
//- rjf: disable cancellation scanning
//
if(lane_idx() == 0)
{
mutex_take(ac_shared->cancel_thread_mutex);
}
scratch_end(scratch);
}
@@ -586,46 +603,47 @@ ac_cancel_thread_entry_point(void *p)
{
for(;;)
{
os_sleep_milliseconds(500);
//- rjf: scan in-flight nodes for expiration
for EachIndex(cache_slot_idx, ac_shared->cache_slots_count)
os_sleep_milliseconds(50);
MutexScope(ac_shared->cancel_thread_mutex)
{
Stripe *cache_stripe = stripe_from_slot_idx(&ac_shared->cache_stripes, cache_slot_idx);
RWMutexScope(cache_stripe->rw_mutex, 0)
for EachIndex(cache_slot_idx, ac_shared->cache_slots_count)
{
for EachNode(cache, AC_Cache, ac_shared->cache_slots[cache_slot_idx])
Stripe *cache_stripe = stripe_from_slot_idx(&ac_shared->cache_stripes, cache_slot_idx);
RWMutexScope(cache_stripe->rw_mutex, 0)
{
Rng1U64 slot_range = lane_range(cache->slots_count);
for EachInRange(slot_idx, slot_range)
for EachNode(cache, AC_Cache, ac_shared->cache_slots[cache_slot_idx])
{
AC_Slot *slot = &cache->slots[slot_idx];
Stripe *stripe = stripe_from_slot_idx(&cache->stripes, slot_idx);
for(B32 write_mode = 0; write_mode <= 1; write_mode += 1)
Rng1U64 slot_range = lane_range(cache->slots_count);
for EachInRange(slot_idx, slot_range)
{
B32 slot_has_work = 0;
RWMutexScope(stripe->rw_mutex, write_mode)
AC_Slot *slot = &cache->slots[slot_idx];
Stripe *stripe = stripe_from_slot_idx(&cache->stripes, slot_idx);
for(B32 write_mode = 0; write_mode <= 1; write_mode += 1)
{
for(AC_Node *n = slot->first, *next = 0; n != 0; n = next)
B32 slot_has_work = 0;
RWMutexScope(stripe->rw_mutex, write_mode)
{
next = n->next;
if(access_pt_is_expired(&n->access_pt, .time = n->evict_threshold_us) && ins_atomic_u64_eval(&n->working_count) > 0)
for(AC_Node *n = slot->first, *next = 0; n != 0; n = next)
{
slot_has_work = 1;
if(!write_mode)
next = n->next;
if(access_pt_is_expired(&n->access_pt, .time = n->evict_threshold_us) && ins_atomic_u64_eval(&n->working_count) > 0)
{
break;
}
else
{
n->cancelled = 1;
slot_has_work = 1;
if(!write_mode)
{
break;
}
else
{
n->cancelled = 1;
}
}
}
}
}
if(!slot_has_work)
{
break;
if(!slot_has_work)
{
break;
}
}
}
}
src/artifact_cache/artifact_cache.h
+1
View File
@@ -129,6 +129,7 @@ struct AC_Shared
// rjf: cancel thread
Thread cancel_thread;
Mutex cancel_thread_mutex;
};
////////////////////////////////