mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-16 11:49:59 -04:00
Let the wait after a frame's first submission cover the write-backs under shadow storage.
With shadow storage (the Linux default) what the GPU wrote reaches guest memory in the fence's completion callback, later than an import's store on the GPU. The wait after a frame's first submission returned at the executed timeline, before that, so the game rewrote what those passes had written and the write-back then put the GPU's older bytes back over the rewrite. The whole 3D scene came out brighter for single frames, each of them a frame whose main work was held at its CPU event wait and replayed late. Standing outdoors with NX2NAVI_SHADOW=1 on an RTX 5090: 695, 624, 443, 323 and 2 flashed frames of 1,800 in five runs with the split, none in eight runs without it. The frame rate did not drop (74.6 fps median with the fix; runs of the build before it gave 71 to 72). NX2NAVI_WB_SPLIT=2 keeps the split with shadow storage, to measure the fault. That this is the flicker a player reported on Linux (RX 7800 XT, RADV) is not confirmed on their machine; NX2NAVI_WB_SPLIT=0 gives 0.7.1 the same wait there. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
This commit is contained in:
@@ -272,7 +272,10 @@ writes come back at its fence. Pages both sides write keep a base copy, so each
|
||||
a VRAM mirror's write-back merge marks the pages it changed, and only those are compared at the fence; a lookup that
|
||||
reads guest memory itself (the uniform snapshot, `vk_mem_lookup`) does not queue its pool for the copy-in. In Kokiri
|
||||
Forest (1.1.1, RTX 4090) that costs about 1 ms a frame (copy-in 0.5 ms, write-back 0.6 ms), and the copies hold about
|
||||
5 GB. `NX2NAVI_SHADOW=1` turns it on under Windows too. The Steam Deck's RADV has `VK_EXT_external_memory_host` as
|
||||
5 GB. The GPU's bytes reach guest memory in the fence's completion callback, later than an import's, so the wait
|
||||
after a frame's first submission covers the write-backs there (`runtime/vk_sync.c`, "the executed timeline"): the game
|
||||
rewrites what those passes wrote without a fence, and a write-back that landed after the rewrite left the GPU adding
|
||||
light onto its own last output, the whole scene brighter for a frame. `NX2NAVI_SHADOW=1` turns it on under Windows too. The Steam Deck's RADV has `VK_EXT_external_memory_host` as
|
||||
well: the shadow copies are chosen for the measured cost above, not for a missing extension.
|
||||
|
||||
The device address table (`runtime/vk_gpa.c`) names a pool's shadow copy as it names an import, so shaders follow guest
|
||||
|
||||
@@ -279,7 +279,7 @@ static const CfgMeta metas[] = {
|
||||
{"VK_NOOPT", "0", "0|1", RESTART}, // vk_pipeline.c noopt_flag
|
||||
// diagnostics
|
||||
{"TRACE", "", NULL, RESTART}, {"FLIGHT", "0", "0|1", RESTART}, {"FLIGHT_SPIKE_MS", "100", "0..60000", RESTART},
|
||||
{"WB_FENCE_LIMIT", "65536", "0..1073741824", RESTART}, {"WB_MERGE", "1", "0|1", RESTART}, {"WB_EXACT", "1", "0|1", RESTART}, {"MEM_CENSUS", "0", "0|1", RESTART}, {"WB_NO_REUP", "1", "0|1", RESTART}, {"WB_PROBE", "0", "0|1", RESTART}, {"WB_SPLIT", "1", "0|1", RESTART}, {"WB_LAZY", WB_LAZY_DEFAULT, "0|1", RESTART}, {"WB_LAZY_TOUCH", "0", "0..16", RESTART}, {"WB_LAZY_CHUNK", "1", "0|1", RESTART}, {"FLUSH_UPLOAD", "0", "0|1", RESTART}, {"GPA_TWIN_MAX_KB", "0", "0..1048576", RESTART},
|
||||
{"WB_FENCE_LIMIT", "65536", "0..1073741824", RESTART}, {"WB_MERGE", "1", "0|1", RESTART}, {"WB_EXACT", "1", "0|1", RESTART}, {"MEM_CENSUS", "0", "0|1", RESTART}, {"WB_NO_REUP", "1", "0|1", RESTART}, {"WB_PROBE", "0", "0|1", RESTART}, {"WB_SPLIT", "1", "0|1|2", RESTART}, {"WB_LAZY", WB_LAZY_DEFAULT, "0|1", RESTART}, {"WB_LAZY_TOUCH", "0", "0..16", RESTART}, {"WB_LAZY_CHUNK", "1", "0|1", RESTART}, {"FLUSH_UPLOAD", "0", "0|1", RESTART}, {"GPA_TWIN_MAX_KB", "0", "0..1048576", RESTART},
|
||||
{"WB_MERGE_MAX", "32", "1..256", RESTART}, {"TEXUP_LOG_MS", "0", "0..60000", RESTART}, // vk_texture.c vk_texstat_frame
|
||||
{"VMA_BLOCK_MB", "64", "16..256", RESTART}, // vk_headroom.c vk_vma_block_size
|
||||
{"ASTC_BC_PROBE", "0", "0|1", RESTART}, // vk_texture.c astc_probe_alpha
|
||||
|
||||
@@ -82,12 +82,35 @@ static int hold_max = -1;
|
||||
// write-backs follow in a part of their own, which signals the timeline as before. A submission without that split
|
||||
// signals both at its end. vk_sync_exec_wait waits on exec_tl; every other wait, the fences the game asks for among
|
||||
// them, still covers the write-backs. NX2NAVI_WB_SPLIT=0: no split, and the wait is the fence's.
|
||||
// Not with shadow storage (vk_memory.c): there the game does need the write-backs before it goes on. With an import a
|
||||
// write-back stores into guest memory on the GPU, right behind the passes. With a shadow copy it stores into the copy,
|
||||
// and the bytes reach guest memory in the fence's completion callback (shadow_wb_cb), after the write-back part, the
|
||||
// fence and the callbacks queued before it, while the game is already rewriting for its next frame what those passes
|
||||
// wrote. A write-back that lands after such a rewrite puts the GPU's older bytes over it in guest memory and moves the
|
||||
// page's base copy along, so the next copy-in finds nothing the CPU changed; and a rewrite with the value the base
|
||||
// copy holds (a reset to zero) is no difference at any time. The GPU then goes on from its own last output, as the
|
||||
// game's lighting accumulators do, which the shaders add to and the CPU resets.
|
||||
// Measured: with the split the whole 3D scene came out brighter for single frames (3.8 of 255, the interface as it
|
||||
// was), each of them a frame whose main work was held at its CPU event wait and replayed late, together with the next
|
||||
// frame's: standing outdoors, 695, 624, 443, 323 and 2 of 1,800 frames in five runs, and none in eight runs without
|
||||
// the split (RTX 5090 with NX2NAVI_SHADOW=1, 2026-11-10). A player on Linux (RX 7800 XT, RADV) saw the open field lit
|
||||
// in one frame and flat in the next; that it is the same fault is not confirmed on their machine. Which words the
|
||||
// write-backs overtook was not traced: NX2NAVI_SHADOW_CHECK counts no byte that both sides changed to different
|
||||
// values, which fits a reset to the base copy's value.
|
||||
// NX2NAVI_WB_SPLIT=2 keeps the split with shadow storage too, to measure that fault.
|
||||
static VkSemaphore exec_tl;
|
||||
static uint64_t exec_sig; // sub_lock: the newest value a submitted part signals on exec_tl
|
||||
static int wb_split = -1;
|
||||
|
||||
int vk_sync_wb_split(void) {
|
||||
if (wb_split < 0) wb_split = cfg_bool("NX2NAVI_WB_SPLIT");
|
||||
if (wb_split < 0) {
|
||||
const char *e = cfg_get("NX2NAVI_WB_SPLIT");
|
||||
int forced = e && !strcmp(e, "2");
|
||||
wb_split = cfg_bool("NX2NAVI_WB_SPLIT") && (forced || !vk_mem_shadow_storage());
|
||||
if (vk_mem_shadow_storage())
|
||||
rt_log("vk: the wait after a frame's first submission %s (shadow storage; NX2NAVI_WB_SPLIT=2 leaves them out)",
|
||||
wb_split ? "leaves out its write-backs" : "covers its write-backs and their copies into guest memory");
|
||||
}
|
||||
return wb_split && exec_tl != VK_NULL_HANDLE;
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user