Triforce v0.4.0

What was merged from NX2navi (DLSS frame generation, AMD FSR 4 in Setup builds, texture detail by video memory, sprint, volume, frame caps, safer saves, the above-60 fixes), the ultrawide HUD at the screen edges, camera speed, movies without the AV1 Video Extension, the 5120x2880 + RenoDX HDR start fix, the renderer work of 11-14 and the first float changes of the lift.
This commit is contained in:
ruckus7382sfgsgdf
2026-11-15 06:39:00 -06:00
parent 183b05d8c4
commit 93eeb714c2
87 changed files with 5947 additions and 153 deletions

View File

@@ -1188,6 +1188,16 @@ int hos_vcores_on(void) {
static int vc_cores(void) { return hos_cores() > 8 ? 8 : hos_cores(); }
static unsigned vc_allowed(HosThread *t) {
const unsigned all = (1u << vc_cores()) - 1;
/* (Triforce, 11-14, the lab) HOS_VCORE_ANY=1: a thread takes any free core, not only the ones its affinity mask names.
What the game needs of the cores is that no two threads run under one core number at a time (its per-core data has no
locks); the masks are the console's placement. OoT pins nearly every thread to one core (six BaseProc executors, five
module workers, the main and Presentation threads on six cores), and here a thread waits for ITS core while others
stand free: at the Market ~61 waits a frame of ~0.1 ms each, 7 ms of waiting a frame summed over the threads, on the
path the frame now waits for (the game's own CPU work, 13-14 ms a frame with the replay on its own thread). The risk:
a thread that reads its core number, blocks, and goes on under another */
static int any = -1;
if (any < 0) any = getenv("HOS_VCORE_ANY") && atoi(getenv("HOS_VCORE_ANY")) != 0;
if (any) return all;
unsigned m = (unsigned)(t->core_mask & all);
if (!m) m = 1u << hos_thread_core(t);
return m;