[jordanella]

Comparing a switch's two values meant two launches: a minute apart, another walk to the scene, other warm caches, and differences of a few percent lost in that. This makes a switch changeable while the game runs, so both values are measured in one session, alternating, with the game's own numbers for each half. The dev-loop proposal's in-session A/B (#2636, D4/D5 review); the first switch is #2871's RX_VK_OCC_QUERIES.

Stacked on #2871: review the last commit; this rebases to it alone once #2871 is in.

Change

• gpu/vk/vk_live.h (new): live switches. Each starts as its environment variable says (hookshot_env.h's rule). The control bridge asks for another value with the command it already has: hookshot_ctl.py option set RX_VK_OCC_QUERIES 1. The bridge's settings list (menu_setting) ends with the live switches (tab live, key env.<NAME>); they are not in the menu and not saved to hookshot_options.ini. No interface change: setting/setting_set carry them.
• The present takes an ask at its next frame (live_tick, in status_tick) and first logs the time since the last ask:
  vk: live: 590 frames in 9.5 s with RX_VK_OCC_QUERIES=0: 62.1 fps (1% low 51.0), GPU 8.12 ms a frame, 1830 draws a frame
  The first half second after a change is left out (frames already on their way; GPU times are read when their slots come round). An ask for the value a switch already has changes nothing and still ends the period, so a run can start clean.
• A switch is listed only once its reader takes it where a change is safe. RX_VK_OCC_QUERIES is asked once a command buffer, before its dry pass (replay_handle's occ_room), so the dry pass and its workers agree:
  • turned off, the run already open goes past its limit and its report stays "visible", as without the switch;
  • turned on, the next samples-passed reset opens a run;
  • its query pool is made the first time it is on.
• hookshot_ctl.py ab NAME [--every 10] [--rounds 4]:
  • It starts from the switch's value with a mark, then flips it every N seconds, R times each way, and ends on the start value.
  • It prints each half, each value's medians (fps, 1% low, GPU ms, draws), and the difference.
  • It also prints the fps spread between halves of the same value, and says when the difference is within it.
  • A half in which the game left play or changed map is marked (a status before and after each ask).

In use

Windows 11, RTX 3080, Castle Town Market from a save, one headless launch, ab RX_VK_OCC_QUERIES --every 10 --rounds 4 standing, then again with the camera turning:

	fps (median)	spread	GPU ms	draws
standing, off	18.6	3.9	27.2	2750
standing, on	20.0	3.5	25.4	2325
turning, off	16.3	7.1	30.5	1482
turning, on	14.2	6.0	35.8	1364

• On, the game leaves out about 15% of its draws standing. Turning, the GPU time goes up about 17%.
• This agrees in direction with the four separate launches on fix(vk): give samples-passed reports the GPU's counts (RX_VK_OCC_QUERIES) #2871 (turning: GPU time +25%). It came from one launch, a few minutes after reaching the scene.
• The fps differences are within the spread between halves of the same value, and ab says so.
• In the log: 16 changes taken, the query pool made once (at the first "on"), and no errors.

Testing

• tests/test_live_switches.py (new):
  • vk_live.h, live_tick, live_fill and menu_setting_set's env. branch are cut out and run against stub counters: the start value from the variable; an ask taken at the next frame, not before; the settle time left out; the frame counts, fps, 1% low, GPU ms and draws of each half, read back with hookshot_ctl's own pattern; a mark; a half too short.
  • ab against a fake bridge: the order of the asks, the halves kept, the medians and spread, a half with a map change marked, a menu row refused.
  • In the source: the tick is in the present, the listing comes after the rows, the ask is not saved and takes no menu lock, and only live_tick writes what the readers take.
  • Two deliberate bugs fail it: no settle time, and an ask applied before its half is logged.
• tests/test_vk_occlusion_queries.py: the switch is in vk_live.h now, and occ_queries_on asks it before the pool is made.
• Every check that reads the changed files (179), through run_all.py, Windows: 176 pass. test_linux_driver_fault_guard and test_readback_wait_bounded fail the same way on main here, and test_input_background fails the same way at fix(vk): give samples-passed reports the GPU's counts (RX_VK_OCC_QUERIES) #2871's head (its older base; it passes on today's main).
• Not tested: Linux.

See #2636, #2871.

🤖 Generated with Claude Code