mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-17 09:19:58 -04:00
Give the freezing of idle memory regions a time budget and back off when a freeze is slow.
A freeze (a VirtualQuery and a VirtualProtect of 2 MiB) takes microseconds as a rule, but in the second after an area load each took 8 to 17 ms. The newly loaded regions all go idle together, so 32 came up in one submission: its mirror begin took 70.5 ms and the frame 78 ms, and a 9 ms freeze followed in each of the next 20 frames. In a play session 363 regions were frozen in the second after Kakariko loaded and 70 to 120 a second for the next 20 seconds, with the frames over 50 ms in the same windows. A pass now stops once it has spent NX2NAVI_FREEZE_BUDGET_US (default 1000; 0 as before), and a freeze over 2 ms holds the passes for 250 ms, doubled for each slow one in a row up to 2 s. On the Kakariko route the time in slow memory calls after the load went from 1374 ms in 170 calls to 115 ms in 18, the 78 ms frame is gone, and the regions freeze some three seconds later, once the calls are fast again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -46,7 +46,7 @@ static const char *const known[] = {
|
||||
"COLOR_PRESET", "COLOR_SATURATION", "CONTACT_DEBUG", "CONTACT_SHADOWS", "CORE_DIAG", "CRIT", "CRIT_OUT",
|
||||
"DESC_TEMPLATE", "DEVICE_PROFILE", "DISCORD_APP_ID", "DISCORD_IMAGE", "DISCORD_RPC", "DLAA", "DLAA_CAM_KEEP", "DLSS_IDLE_FREE", "DLAA_JITTER", "DLAA_MV", "DLAA_MASKDEBUG", "DLAA_MVDEBUG", "DLAA_OBJMV", "DLAA_SURVEY", "DLAA_TARGET_KEEP", "DLAA_TIMES", "DLSS", "DLSS_CHAR_MASK", "DLSS_PRESET", "DLSS_REACTIVE", "DUMP", "DUMP_BUF", "DUMP_DIR",
|
||||
"DUMP_LOG", "DUMP_RAW", "DUMP_SHADERS", "DUMP_FILE", "DS3", "LOG_CS_BINDS", "LOG_TEX_FLAGS", "LOG_MESH", "MESH", "POOL", "NATIVE_TENSOR", "LOG_RCOND", "LOG_RT_SIZES", "DUMP_WHITE", "DUMP_AFTER", "DUMP_MOVIE", "SUBMIT_FINISH", "ACQUIRE_WAIT", "CRASH_FILTER_GUARD", "CS_FALLBACK", "EXEFS", "FEEDBACK", "FIX_C0", "FIX_ZERO", "FLIGHT", "FLIGHT_AT", "FLIGHT_DIR", "FLIGHT_FRAMES",
|
||||
"FFX_PIPE_CACHE", "FLIGHT_SPIKE_MS", "FLIGHT_THUMBS", "FLIGHT_EVERY", "FLUSH_SPAN", "FLUSH_UPLOAD", "FOG_CULL", "FOLIAGE_SHADOW_CULL", "FOLIAGE_SHADOW_VRS", "FOV", "FOV_PATCH", "FG_HUDLESS_SHOW", "FG_MULT", "FPS60", "FPS_LIMIT", "FRAMEGEN", "FRAMEGEN_MULTIPLIER", "FREEZE_IDLE", "FREE_CORES",
|
||||
"FFX_PIPE_CACHE", "FLIGHT_SPIKE_MS", "FLIGHT_THUMBS", "FLIGHT_EVERY", "FLUSH_SPAN", "FLUSH_UPLOAD", "FOG_CULL", "FOLIAGE_SHADOW_CULL", "FOLIAGE_SHADOW_VRS", "FOV", "FOV_PATCH", "FG_HUDLESS_SHOW", "FG_MULT", "FPS60", "FPS_LIMIT", "FRAMEGEN", "FRAMEGEN_MULTIPLIER", "FREEZE_IDLE", "FREEZE_BUDGET_US", "FREE_CORES",
|
||||
"FULLSCREEN", "FXAA", "GAME_ASPECT", "GAME_RES", "GPA", "GPA_EMPTY", "GPA_TWIN", "GPA_TWIN_MAX_KB", "GPL", "GPL_DRAW_BUILD", "GPL_FEEDBACK", "PIPE_AHEAD", "PIPE_FAST_WORKERS", "PIPE_FRAME_WAIT_MS", "PIPE_LINK_DRAW_MS", "PIPE_SKIP_STEADY", "PROG_LEAD", "GPU", "GPU_INDEX", "GPU_PASSES", "GPU_NAME", "HDR", "HDR_BLOWOUT", "HDR_CONTRAST",
|
||||
"HDR_EXPOSURE", "HDR_FLARE", "HDR_GAMMA", "HDR_GAME_NITS", "HDR_GRADE", "HDR_HIGHLIGHT_SAT", "HDR_HIGHLIGHTS", "HDR_HUE",
|
||||
"HDR_HUE_CLIP", "HDR_LIVE", "HEAP_LAZY", "HDR_MATCH_SLOPE", "HDR_MIDGRAY",
|
||||
@@ -128,7 +128,7 @@ static const struct { const char *key, *section; } sections[] = {
|
||||
{"DYN_DELTA", "performance"}, {"STG_ARENA", "performance"}, {"SUBMIT_BATCH", "performance"},
|
||||
{"BUILD_FREE_MB", "performance"}, {"BUILD_RESERVE_MB", "performance"}, {"MEM_NEED_MB", "diagnostics"},
|
||||
{"ASTC_GPU_DESWIZZLE", "performance"}, {"ASTC_SCRATCH_KEEP", "performance"}, {"ASTC_ARENA", "performance"}, {"POOL_CREATE_BUDGET", "performance"},
|
||||
{"POOL_CREATE_MS", "performance"}, {"POOL_FINI_IDLE", "performance"}, {"POOL_LOAD_BACKLOG", "performance"}, {"POOL_LOAD_MS", "performance"},
|
||||
{"POOL_CREATE_MS", "performance"}, {"POOL_FINI_IDLE", "performance"}, {"FREEZE_BUDGET_US", "performance"}, {"POOL_LOAD_BACKLOG", "performance"}, {"POOL_LOAD_MS", "performance"},
|
||||
{"TIMER_1MS", "performance"}, {"PIPE_WAIT", "performance"}, {"PIPE_SKIP", "performance"}, {"ONESHOT_BUDGET", "performance"},
|
||||
{"PIPE_FRAME_WAIT_MS", "performance"}, {"PIPE_SKIP_STEADY", "performance"},
|
||||
{"CACHE_DIE", "performance"}, {"ONE_PER_CORE", "performance"},
|
||||
@@ -258,7 +258,7 @@ static const CfgMeta metas[] = {
|
||||
{"DYN_RES", "0", "0|1", LIVE}, {"DYN_RES_TARGET", "16.7", "1..100", LIVE}, {"DYN_RES_STEP", "45", "8..600", LIVE},
|
||||
// the recording thread: a core of its own (auto: with 12 or more logical processors) and the highest priority
|
||||
// (vk_exec.c rec_claim)
|
||||
{"RECORDER_CORE", "auto", "auto|0|1", RESTART}, {"RECORDER_PRIO", "1", "0|1", RESTART}, {"REPLAY_TIME", "0", "0|1", RESTART}, {"LOCK_TIME", "0", "0|1", RESTART}, {"POOL_CREATE_BUDGET", "16", "1..4096", RESTART}, {"POOL_CREATE_MS", "6", "0..100", RESTART}, {"POOL_FINI_IDLE", "0", "0|1", RESTART}, {"POOL_LOAD_BACKLOG", "512", "0..65536", RESTART}, {"POOL_LOAD_MS", "50", "1..1000", RESTART},
|
||||
{"RECORDER_CORE", "auto", "auto|0|1", RESTART}, {"RECORDER_PRIO", "1", "0|1", RESTART}, {"REPLAY_TIME", "0", "0|1", RESTART}, {"LOCK_TIME", "0", "0|1", RESTART}, {"POOL_CREATE_BUDGET", "16", "1..4096", RESTART}, {"POOL_CREATE_MS", "6", "0..100", RESTART}, {"POOL_FINI_IDLE", "0", "0|1", RESTART}, {"FREEZE_BUDGET_US", "1000", "0..100000", RESTART}, {"POOL_LOAD_BACKLOG", "512", "0..65536", RESTART}, {"POOL_LOAD_MS", "50", "1..1000", RESTART},
|
||||
// held submissions handed to the queue together (vk_sync.c, submit batching)
|
||||
{"SUBMIT_BATCH", "4", "1..16", RESTART},
|
||||
{"CACHE_DIE", "1", "0|1|ab", RESTART}, {"ONE_PER_CORE", "0", "0|1|ab", RESTART}, // perf.c cpu_placement
|
||||
|
||||
@@ -3215,16 +3215,54 @@ void nvn_pool_flushed(const pool_t *p, uint64_t off, uint64_t size) {
|
||||
|
||||
// gen_lock exclusive: freezes up to FREEZE_PER_SUB candidate regions still idle. The query after the
|
||||
// protection collects the writes made before it took effect.
|
||||
// A freeze is a VirtualQuery and a VirtualProtect of 2 MiB, some microseconds as a rule; in the second after an area
|
||||
// load, while the loaders write and the driver imports and locks new memory, each took 8 to 17 ms. The newly loaded
|
||||
// regions all go idle together, so 32 of them came up in one submission: its mirror begin took 70.5 ms and the frame
|
||||
// 78 ms, and one freeze of 9 ms followed in each of the next 20 frames. A freeze only saves later write-watch queries
|
||||
// (about a microsecond per MiB per submission), so it can wait: a pass stops once it has spent freeze_budget_ns, and a
|
||||
// freeze over FREEZE_SLOW_NS holds the passes for FREEZE_BACKOFF_NS, doubled for each slow one in a row up to 2 s.
|
||||
// NX2NAVI_FREEZE_BUDGET_US (default 1000; 0: no budget and no backoff, as before).
|
||||
#define FREEZE_SLOW_NS 2000000ull
|
||||
#define FREEZE_BACKOFF_NS 250000000ull
|
||||
#define FREEZE_BACKOFF_MAX_NS 2000000000ull
|
||||
static uint64_t freeze_budget_ns = ~0ull; // ~0: not read yet
|
||||
static uint64_t frz_hold_until, frz_backoff, frz_held_passes, frz_slow; // gen_lock
|
||||
|
||||
static void freeze_pass(void) {
|
||||
if (freeze_budget_ns == ~0ull) {
|
||||
long long us = cfg_int("NX2NAVI_FREEZE_BUDGET_US");
|
||||
freeze_budget_ns = us > 0 ? (uint64_t)us * 1000 : 0;
|
||||
}
|
||||
uint64_t start = qpc();
|
||||
if (freeze_budget_ns && start < frz_hold_until) {
|
||||
if (nfcand) frz_held_passes++;
|
||||
return;
|
||||
}
|
||||
for (int done = 0; nfcand && done < FREEZE_PER_SUB;) {
|
||||
uint64_t j = fcand[--nfcand], a = j << REG_SHIFT;
|
||||
if (ww_ep - reg_last[j] < (uint32_t)freeze_idle || mem_frozen(a)) continue;
|
||||
done++;
|
||||
uint64_t t0 = qpc();
|
||||
int ok = mem_freeze(a);
|
||||
slow_mem_call("freeze", a, 1ull << REG_SHIFT, 0, qpc() - t0);
|
||||
if (!ok) { n_frz_refused++; continue; }
|
||||
query_stamp(a, a + (1ull << REG_SHIFT), ++gen_now);
|
||||
uint64_t t1 = qpc();
|
||||
slow_mem_call("freeze", a, 1ull << REG_SHIFT, 0, t1 - t0);
|
||||
if (ok) query_stamp(a, a + (1ull << REG_SHIFT), ++gen_now);
|
||||
else n_frz_refused++;
|
||||
if (!freeze_budget_ns) continue;
|
||||
if (t1 - t0 >= FREEZE_SLOW_NS) {
|
||||
frz_slow++;
|
||||
frz_backoff = frz_backoff ? frz_backoff * 2 : FREEZE_BACKOFF_NS;
|
||||
if (frz_backoff > FREEZE_BACKOFF_MAX_NS) frz_backoff = FREEZE_BACKOFF_MAX_NS;
|
||||
frz_hold_until = t1 + frz_backoff;
|
||||
static uint32_t logged;
|
||||
if (++logged <= 40)
|
||||
rt_log("mem: freezing an idle region took %.1f ms: no freezes for %.0f ms (%llu slow freezes so far, %llu "
|
||||
"passes held, %d regions waiting)", (t1 - t0) / 1e6, frz_backoff / 1e6, (unsigned long long)frz_slow,
|
||||
(unsigned long long)frz_held_passes, nfcand);
|
||||
break;
|
||||
}
|
||||
frz_backoff = 0;
|
||||
if (qpc() - start >= freeze_budget_ns) break;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user