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

Author: ryanfleury

src/artifact_cache/artifact_cache.c

@@ -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

@@ -129,6 +129,7 @@ struct AC_Shared
 
   // rjf: cancel thread
   Thread cancel_thread;
+  Mutex cancel_thread_mutex;
 };
 
 ////////////////////////////////

src/dbg_info/dbg_info.c

@@ -1301,11 +1301,11 @@ di_search_artifact_create(String8 key, B32 *cancel_signal, B32 *retry_out, U64 *
     {
       if(lane_idx() == 0)
       {
-        sort_records = push_array(scratch.arena, SortRecord, sort_records_count);
+        sort_records = push_array_no_zero(scratch.arena, SortRecord, sort_records_count);
       }
       if(lane_idx() == lane_from_task_idx(1))
       {
-        sort_records__swap = push_array(scratch.arena, SortRecord, sort_records_count);
+        sort_records__swap = push_array_no_zero(scratch.arena, SortRecord, sort_records_count);
       }
       lane_sync_u64(&sort_records, 0);
       lane_sync_u64(&sort_records__swap, lane_from_task_idx(1));
@@ -1429,7 +1429,7 @@ di_search_artifact_create(String8 key, B32 *cancel_signal, B32 *retry_out, U64 *
       if(lane_idx() == 0)
       {
         items.count = all_items->total_count;
-        items.v = push_array(arena, DI_SearchItem, items.count);
+        items.v = push_array_no_zero(arena, DI_SearchItem, items.count);
       }
       lane_sync_u64(&items.count, 0);
       lane_sync_u64(&items.v, 0);

src/raddbg/raddbg_core.c

@@ -10308,16 +10308,6 @@ rd_frame(void)
   }
 #endif
 
-  //////////////////////////////
-  //- rjf: [windows] clear pages from working set shortly after startup, many of which will not be needed
-  //
-#if OS_WINDOWS
-  if(rd_state->frame_index == 1) ProfScope("SetProcessWorkingSetSize")
-  {
-    SetProcessWorkingSetSize(GetCurrentProcess(), max_U64, max_U64);
-  }
-#endif
-  
   //////////////////////////////
   //- rjf: do per-frame resets
   //
@@ -16563,6 +16553,19 @@ rd_frame(void)
   U64 frame_time_us = end_time_us-begin_time_us;
   rd_state->frame_time_us_history[rd_state->frame_index%ArrayCount(rd_state->frame_time_us_history)] = frame_time_us;
 
+  //////////////////////////////
+  //- rjf: [windows] clear pages from working set shortly after startup, many of which will not be needed
+  //
+#if OS_WINDOWS
+  if(di_load_count() < 50)
+  {
+    if(rd_state->frame_index == 15) ProfScope("SetProcessWorkingSetSize")
+    {
+      SetProcessWorkingSetSize(GetCurrentProcess(), max_U64, max_U64);
+    }
+  }
+#endif
+  
   //////////////////////////////
   //- rjf: bump frame time counters
   //