Name in a lost device's report the mesh draws of each marked pass, a group of draws the replay workers recorded between two marks, and in the fatal line how mesh draws are drawn.

A pass mark now counts its mesh draws and those that take their count from the GPU; the gap before a mark notes a group recorded by the workers, whose passes have no marks, where the report said "barriers and state only"; the fatal line names the mesh path (native, the mesh fallback, the emulation). The comment on the RX 6700 XT report in vk_exec.c called the last marked dispatch the replay's own: it is the game's (op_dispatch, cs d7566eb1).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
chrg6
2026-11-13 12:28:27 -05:00
parent 8df4311723
commit 1f4818fb6b
5 changed files with 39 additions and 11 deletions

View File

@@ -865,6 +865,7 @@ void vk_exec_pass_begin(VkExec *ex) {
VkPassMark *m = ex->mark = rw_chunk ? NULL : vk_pass_mark_begin(ex->cb, VK_MARK_PASS, ex->fence); // one ring
if (m) {
m->step = ex->mark_passes++;
m->mesh = m->mesh_ind = 0;
m->w = ex->pass_w, m->h = ex->pass_h, m->d = ex->pass_layers;
m->layer = ex->depth.image ? ex->depth.base_layer : 0;
m->vw = m->vh = 0;
@@ -1855,8 +1856,10 @@ static void op_one(VkExec *ex, const op_hdr *h) {
// 3 workers: on an RX 7900 XT with a Ryzen 7 5800X3D, Hyrule Field standing went from 41.2 to 50.4 fps (2026-11-12, one
// batch), with the same pictures there and in Kokiri Forest, Kakariko, Goron City, a graveyard house and Gerudo
// Valley; 5 or 7 workers measured the same as 3. Default 0 again after players' games crashed with 3 on by default
// (0.9.1, 0.9.2): an RX 6700 XT lost its device after about 18,000 frames, its last mark a compute dispatch of the
// replay's own, with an occlusion query slot found unavailable after its fence just before; the cause is not known.
// (0.9.1, 0.9.2): an RX 6700 XT lost its device after about 18,000 frames, its last mark the game's own compute
// dispatch (op_dispatch: 20x1x1 groups of cs d7566eb1, a depth probe) and nothing after it marked, since the group of
// draws that followed was recorded by the workers, whose passes have no marks; an occlusion query slot was found
// unavailable after its fence just before. An RX 6600 lost it the same way twice (0.9.2). The cause is not known.
// A submission's draws come in runs between ops that have to stay in the replay's order on its own thread (dispatches,
// copies, texture and buffer clears, events, the game's barriers, calls). Hyrule Field, standing (NX2NAVI_REPLAY_RUNS,
// 2026-11-11): about 3,500 draws a frame in 82 runs, 96% of them in runs of 16 draws or more and 67% in runs of 256 or
@@ -2449,6 +2452,9 @@ static const op_hdr *group_start(VkExec *ex, const op_hdr *h, int depth) {
uint64_t t0 = kern_now_ns();
gfx_tl_note('g', 0, draws, (uint64_t)n);
vk_exec_barrier(ex); // the command buffer so far ends behind a full barrier, every image back in GENERAL
// the chunks' passes get no marks (vk_exec_pass_begin): a lost device's report names the group between the marks.
// After the barrier, whose end of an open pass ends that pass's mark and empties the note.
vk_gpu_gap_note(VK_GAP_GROUP, draws);
g->upd = vk_exec_draw_group_begin(ex);
// the front pass
uint32_t per = (draws + (uint32_t)n - 1) / (uint32_t)n, cd = 0;