OoT's Presentation Thread wakes when the main thread hands it the frame and needs a
virtual core for the guest code before its submit; the six BaseProcExecutors, woken by
the same hand-off, hold all six cores, and it waited ~1.2 ms for one twice in every slow
frame pair at Lon Lon Ranch, on the frame's critical path (HOS_CHAIN).

HOS_VCORE_FIRST=<name part>,...: those threads rank above every guest priority for the
virtual cores only (a waiter asks a holder for its core at once; no waiter takes theirs).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
(cherry picked from commit f9404ccd578592525692b7b686e983f3845e96cb)

@@ -264,6 +264,7 @@ Pipeline-cache file ownership, temporary storage and failure behavior are descri
 | `RX_VK_NAN_GUARD=0` | No NaN guard on float colour outputs (on by default on every card: one NaN texel spreads through bloom into glowing blocks) |
 | `RX_VK_FIFO=1`, `RX_VK_IMMEDIATE=1` | V-Sync On (FIFO) or Off (immediate) for one run, over the menu's V-Sync row |
 | `HOS_VCORES=0`, `HOS_CORES=<1..8>`, `HOS_RESERVE_CORE=0` | No virtual cores; the cores the game sees (6); no core kept for the replay (one is, with 8 or more logical processors) |
+| `HOS_VCORE_FIRST=<name part>,...` | The threads whose name holds one of the parts take a virtual core before any other thread (above every guest priority, for the virtual cores only; their yields and time slices still follow guest priority). For OoT, `Presentation`: its Presentation Thread no longer waits behind the six BaseProcExecutors (off by default) |

@@ -653,1 +657,4 @@
 
+/* HOS_VCORE_FIRST=<name part>,... (experiment, off by default): the threads whose name holds one of the parts get a
+   virtual core before any other, as if their priority were above every guest priority (-1; Horizon's are 0-63): a
+   waiter asks a holder for its core at once, and no waiter takes theirs. The guest's own priority (GetThreadPriority)

@@ -654,0 +669,4 @@
+        const size_t n = end ? (size_t)(end - p) : strlen(p);
+        char part[64];
+        snprintf(part, sizeof part, "%.*s", (int)(n < sizeof part ? n : sizeof part - 1), p);
+        if (part[0] && strstr(name, part)) return 1;

@@ -1641,0 +1667,4 @@
+/* w may take t's core in a turn (a yield of t's, or after the time slice): as good in that order, or, t being a first
+   thread, as good by guest priority: a first thread comes first, but its yields and slices stay as they were (a first
+   thread waiting by yielding for a job thread pinned to its own core kept that core until the 2 s escape) */
+static int vc_turn(const HosThread *w, const HosThread *t) {

@@ -1724,3 +1756,3 @@
         for (int k = 0; k < vc_cores(); k++) {
             HosThread *h = vc_owner[k];
-            if ((m >> k & 1) && h && h->priority > t->priority && (!worst || h->priority > worst->priority)) worst = h;
+            if ((m >> k & 1) && h && vc_prio(h) > vc_prio(t) && (!worst || vc_prio(h) > vc_prio(worst))) worst = h;