mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-17 15:49:57 -04:00
Run the game on the die with the larger L3 cache on processors whose dies differ (AMD's two-die X3D processors).
Windows places a game on the V-Cache die only when it recognizes the exe as a game, and it does not know this one, so the threads were spread over both dies by frequency preference. cache_die = 1 (the default) narrows the process to the logical processors that share the largest L3; it never widens the set the process was given, and does nothing where every L3 has the same size. The recorder's reserved core is then chosen inside that set. Co-Authored-By: Claude Fable 5.2 <[EMAIL]>
This commit is contained in:
@@ -482,24 +482,49 @@ int rt_knob(const char *name, const char *def) {
|
||||
// render thread, the guest cores, the job workers) never share a core's execution units with another; "ab" switches
|
||||
// it with the A/B blocks. Off by default until measured here; the same rule
|
||||
// is reported to give +4-10% on an 8-core SMT CPU.
|
||||
//
|
||||
// NX2NAVI_CACHE_DIE=1, the default: on a processor whose dies have L3 caches of different sizes (AMD's two-die
|
||||
// X3D processors, 3D V-Cache on one die) the process runs on the die with the larger cache, so the guest cores, the
|
||||
// render thread and the recorder share one L3 and never cross dies. Windows' own placement of games on that die
|
||||
// (game mode with AMD's driver) depends on recognizing the exe as a game. 0 leaves the placement to the OS; "ab"
|
||||
// switches it with the A/B blocks. No change on one-die and Intel processors.
|
||||
//
|
||||
// The placement is set at start (perf_cpu_start, before the thread pools count the usable processors) and checked at
|
||||
// every present for the A/B switches. The rules narrow the processors the process was given (start /affinity, a
|
||||
// launcher's or a benchmark's pin), never widen them; with both off nothing is set.
|
||||
static int one_mode = -1, die_mode;
|
||||
static uint64_t cpu_all, cpu_one, cpu_die, cpu_applied; // cpu_applied: the mask set last, 0 none
|
||||
|
||||
static void cpu_placement(void) {
|
||||
static int mode = -1, applied = -1;
|
||||
static uint64_t all, one;
|
||||
if (mode < 0) {
|
||||
mode = rt_knob("ONE_PER_CORE", "0");
|
||||
int cores = plat_cpu_cores(&all, &one);
|
||||
if (mode && one && one != all)
|
||||
rt_log("cpu: %d cores, %d logical processors: one per core%s", cores, __builtin_popcountll(all),
|
||||
mode == 2 ? " in the B blocks" : "");
|
||||
if (one_mode < 0) {
|
||||
one_mode = rt_knob("ONE_PER_CORE", "0");
|
||||
die_mode = rt_knob("CACHE_DIE", "1");
|
||||
int cores = plat_cpu_cores(&cpu_all, &cpu_one);
|
||||
uint32_t kb = 0;
|
||||
cpu_die = plat_cpu_cache_die(&kb);
|
||||
if (one_mode && cpu_one && cpu_one != cpu_all)
|
||||
rt_log("cpu: %d cores, %d logical processors: one per core%s", cores, __builtin_popcountll(cpu_all),
|
||||
one_mode == 2 ? " in the B blocks" : "");
|
||||
else
|
||||
mode = 0;
|
||||
one_mode = 0;
|
||||
if (cpu_die && cpu_die != cpu_all)
|
||||
rt_log("cpu: logical processors %#llx of %#llx share the largest L3 cache (%u MiB)%s",
|
||||
(unsigned long long)cpu_die, (unsigned long long)cpu_all, kb >> 10,
|
||||
!die_mode ? "; cache_die = 0: not used" : die_mode == 2 ? ": the process runs there in the B blocks"
|
||||
: ": the process runs there");
|
||||
if (!cpu_die || cpu_die == cpu_all) die_mode = 0;
|
||||
}
|
||||
int on = RT_KNOB_ON(mode);
|
||||
if (on == applied) return;
|
||||
applied = on;
|
||||
plat_process_affinity(on ? one : all);
|
||||
uint64_t want = cpu_all;
|
||||
if (RT_KNOB_ON(die_mode)) want &= cpu_die;
|
||||
if (RT_KNOB_ON(one_mode)) want &= cpu_one;
|
||||
if (!want) want = cpu_all;
|
||||
if (want == (cpu_applied ? cpu_applied : cpu_all)) return; // nothing new; never restricted: as it was given
|
||||
cpu_applied = want;
|
||||
if (!plat_process_affinity(want)) rt_log("cpu: could not move the process to logical processors %#llx", (unsigned long long)want);
|
||||
}
|
||||
|
||||
void perf_cpu_start(void) { cpu_placement(); }
|
||||
|
||||
// NX2NAVI_CRASH_FILTER_GUARD (default 1; 0 off): once a second, the crash report's last-chance exception filter is put
|
||||
// back if a DLL loaded after the start (a GPU driver, an overlay) set its own in its place. Without it, a fault in such
|
||||
// a DLL ends the process with no FATAL line in the log. The first eight replacements are logged with the module.
|
||||
|
||||
Reference in New Issue
Block a user