mirror of
https://git.triforceproject.com/triforce/source
synced 2026-11-22 17:19:58 -04:00
Triforce v0.7.0 "Sage Of Courage"
The 0.6 crashes after warps and area changes fixed (no write-back into memory the capture leaves out), invisible objects fixed (occlusion gate), DLSS frame generation flicker fixed, achievements, the renderer built with a play profile, faster translated game code, helper threads judged on settled scenes, memory, shader preparation, audio effects, launcher, controls and Linux fixes.
This commit is contained in:
@@ -5,6 +5,11 @@
|
||||
none of them crosses dies. Windows' own placement of games on that die (Game Mode with AMD's chipset driver) depends
|
||||
on its taking the program for a game.
|
||||
|
||||
(Triforce, 11-22, the performance audit's THREAD-49, after Source A's) The replay's core is kept on the die only where
|
||||
the die's cores have more than one logical processor each (SMT on). With SMT off a kept core left the game 7 of the
|
||||
die's 8 cores: the replay ran faster and the frame slower (hos_x3d_keep_replay_core below). HOS_RESERVE_CORE=1 keeps
|
||||
one there too, as before.
|
||||
|
||||
Nothing is done:
|
||||
- with every L3 cache the same size: one die (a 7800X3D), two dies alike (a 7950X, a part with the cache on both), an
|
||||
Intel processor of two kinds of core, whose cores differ in their L2 caches and share one L3;
|
||||
@@ -96,4 +101,24 @@ static inline uint64_t hos_x3d_thread_mask(uint64_t process, uint64_t die, uint6
|
||||
if (on & ~reserved) return reserved || on != process ? on & ~reserved : 0;
|
||||
return on != process ? on : 0;
|
||||
}
|
||||
|
||||
/* (Triforce, 11-22, the performance audit's THREAD-49, after Source A's) Whether a core of the die is kept for the replay.
|
||||
die: hos_x3d_pick's die, 0 for none; cores: the cores in it; asked: HOS_RESERVE_CORE, -1 when it is not set, else 0 or
|
||||
1; own_thread: the replay has a thread of its own (RX_VK_ASYNC=1), which is pinned to the kept core.
|
||||
Source A measured a 9950X3D with SMT off: with one of the die's 8 cores kept the replay took 9.8-10.0 ms a frame and
|
||||
the game ran at 56.9-57.4 fps; with none kept, 10.1-10.5 ms and 59.8-59.9 fps. The game runs six cores' worth of
|
||||
threads beside the replay's, and the eighth core was worth more to them than to the replay alone. With SMT on (16
|
||||
logical processors on the die's 8 cores) it was never measured, so the core stays kept there, as before; so it does
|
||||
with no such die, where this rule changes nothing. The switch, when set, wins either way */
|
||||
static inline int hos_x3d_keep_replay_core(uint64_t die, int cores, int asked, int own_thread) {
|
||||
if (asked >= 0) return asked != 0;
|
||||
if (!die || cores <= 0 || own_thread) return 1;
|
||||
return __builtin_popcountll(die) > cores;
|
||||
}
|
||||
|
||||
/* The core kept for the replay (hos_reserved_cpus): fast, the last core of the fastest kind among the die's cores when
|
||||
there is a die and among all otherwise (pick_reserved_core's, 0 for none), unless the rule above keeps none */
|
||||
static inline uint64_t hos_x3d_reserved(uint64_t fast, uint64_t die, int cores, int asked, int own_thread) {
|
||||
return hos_x3d_keep_replay_core(die, cores, asked, own_thread) ? fast : 0;
|
||||
}
|
||||
#endif
|
||||
|
||||
Reference in New Issue
Block a user