[robandstuff]

What and why

In Red Dead Redemption, characters more than a few metres away were drawn as their hats and props only. Their bodies were missing from the main view, although their shadows had them. The game's character occlusion test (rage::CharacterVisQueries::TestVisiblity) takes end − begin of two SAMPLES_PASSED reports around each character's box. A character at or below its threshold goes on an occluded list, and renDrawBucketContainer::AddToDrawList then drops its low-detail body.

The game never calls ResetCounter: about 35 queries and 324 reports a frame, no resets. On the GPU the counter runs from the queue's creation. The host only started counting at a reset, so every report read 0 and every character counted as hidden.

Now a samples-passed report with no reset before it starts counting. If the open encoder has no visibility buffer, the pass is split there. From then on each report holds the samples since the previous report, added to a running total when its submission completes. post_run does that in submission order under g_post_mu. A reset returns to counting from the reset, as before. DM_NVN_OCC_IMPLICIT=0 turns it off.

Reference entry: docs/reference/rendering.md, "Samples-passed reports with no reset before them". The game's profile isn't upstream yet. It is carried in LNPANDY/NX2Apple and will come in its own PR (tracked in LNPANDY/NX2Apple#999).

Checked

• Tests: make -C dm/host/tests GAME=0100000000010000 run: 49 tests, 2,355 checks, none failed. make -C dm/gfx/tests GAME=0100000000010000 run on Apple silicon (Metal): 146 tests, 71,870 checks. The failures are t_render_counter_overlap (1) and t_render_enable_queue (9), which fail the same way on unchanged main (7f50222); t_feedback's check "two reports in one command buffer read two times" (timestamp reports) fails now and then on unchanged main too (2 of 6 runs), and on this branch in 2 of 3 runs. This change only touches samples-passed reports.
• Games run, with title ID and version: Red Dead Redemption 01007820196A6000 1.0.4, on the Mac from a save in Armadillo:
  • A recorded input replay: the distant characters have bodies at frames 2,280–2,340 (before the change, a floating hat for Jake); frame time unchanged at 33.3 ms; no [RED].
  • Played by hand with the fix: distant characters drawn whole.
• Devices and OS: Apple M4 (32 GB), macOS 27.0.1. Not tested on an iPhone or iPad.
• scripts/regress.py check --games was not run: no other game's saves or recordings were at hand. Games that call ResetCounter before their first report take the old path unchanged.

Docs

• The finding is in its docs/reference/rendering.md entry, and python3 dm/tools/docs_symptoms.py was run (the symptom index)
• No chain tool changed
• No game files, keys or data extracted from a game in the change or the attachments

🤖 Generated with Claude Code

[seedds]

Thank you, landed as cfbdd93 with you as co-author: your diagnosis, reference entry and symptom line are kept.

The code was reworked into one model of the counter instead of a second mode with a switch: post_run keeps a single running total in submission order. Each report adds the slots since the previous report, the slots left at a submission's end are added when it completes, and a reset zeroes the total at its place in the queue. Counting starts at the first reset or, with none before it, at the first report. A game that never resets gets the same values as your version, and DM_NVN_OCC_IMPLICIT is gone.

The rework also fixes a second bug in the old reset path: a report one submission after its reset re-read slots whose visibility buffer a later submission had already reused (64 samples instead of 96 in the new t_render_enable_queue check). The whole gfx run list passes here (149 groups).

Your Red Dead Redemption check was on your first version. A run of the landed commit on your side would confirm it, as would the profile PR you mentioned (LNPANDY/NX2Apple#999) when it comes.

[robandstuff]

Thanks for taking it. Confirmed on Red Dead Redemption 1.0.4 with cfbdd93, on a Mac host built from NX2Apple's pin (c78d737) with the landed commit applied in place of my first version (Apple M4, macOS 27.0.1):

• The log has occlusion queries: samples passed reported with no ResetCounter before it: counting from this report once.
• The recorded replay from Armadillo: at the saloon (frame 2,310) Jake and the people down the street have their bodies again, as with the first version.
• No [RED] through frame 3,600; frame time p50 33.3 ms, as before.

NX2Apple now carries cfbdd93 as a backport in place of its own patch (LNPANDY/NX2Apple#1000), and the profile follows in LNPANDY/NX2Apple#999.