feat(perf): show GPU render and compute loadHEADmaster

xe exposes no busy percentage, but its DRM fdinfo carries per-engine
cycle counters. vram.sh becomes gpu.sh: the same scan also sums
drm-cycles-rcs/ccs per client and prints them with the GPU's cycle
counter, and Perf.js turns two scans into a percentage per engine.

The GPU card's headline is now the busier engine, with render (games)
and compute (local LLMs) as bars, then VRAM, temperature and fan.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

1 files changed, 6 insertions, 4 deletions

diff --git a/desktop/modules/perf/README.md b/desktop/modules/perf/README.md
index d5ae0b3..9fdbc6e 100644
--- a/desktop/modules/perf/README.md
+++ b/desktop/modules/perf/README.md
@@ -2,10 +2,12 @@
 
 The figures the conky dashboard used to show, as a drawer page of cards, one
 per row; the page scrolls. CPU load with a bar per core, the CPU model and
-RAM; every temperature as a tile; the Arc B580's VRAM and fan (no
-utilisation, xe exposes none); the five mounts with their usage; and
+RAM; every temperature as a tile; the Arc B580's render and compute load
+(games and local LLMs, from fdinfo's cycle counters), VRAM, temperature and
+fan; the five mounts with their usage; and
 `~/.cache` with its five largest children. `Perf.qml` reads them; `Perf.js`
-holds the parsing, tested by `node desktop/modules/perf/test_perf.js`.
+holds the parsing, tested by `node desktop/modules/perf/test_perf.js`;
+`gpu.sh` sums the fdinfo scan, tested by `sh desktop/modules/perf/test_gpu.sh`.
 
 The tile shows CPU and RAM, and a dot for the worst of these, yellow or red:
 
@@ -27,7 +29,7 @@ into the LAN depends on the ISP's fixed address. The fetch is
 never a mismatch: the old value goes stale after 4 hours and shows as a dash.
 
 Everything is sampled always, including per-core load, the network rate,
-VRAM (a scan of every fdinfo, ~14ms every 2s) and the fan. The network rate
+GPU load and VRAM (a scan of every fdinfo, ~14ms every 2s) and the fan. The network rate
 is drawn by the network page, which reads it from here, so its graph opens
 on the last minute of history instead of empty. The mount list, `br0`, the GPU name and the 12 GiB VRAM
 size are constants in `Perf.qml`.