Phoenix PM table 0x4C0007: per-core CPU temperature offset — floats [529..536]

[WRCX]

While reverse-engineering the Phoenix APU PM table, I mapped the per-core CPU temperatures on PM-table version 0x4C0007 (Ryzen 7 7840HS, "Phoenix", Zen 4 mobile). They don't seem to be mapped here (there's no 0x4C0007 pm_table_* function) nor in RyzenAdj, so sharing in case it's useful.

Finding

Per-core CPU temperatures = 8 consecutive float32 at indices [529..536] — i.e. physical core N → float [529 + N].

Verified three independent ways on a 7840HS (8 physical cores)

1. Load correlation. The 8 values are °C-plausible, rise together under sustained all-core load (~45 → ~67 °C) and the hottest converges with the package Tctl (SMN 0x00059800).
2. Affinity causality. Pinning a stress load to a single physical core moves only its float:
   • physical core 0 → only [529] rose (+9.4 °C; others within ±2 °C)
   • physical core 4 → only [533] rose (+7.2 °C; others within ±2.6 °C)
3. Cross-check vs HWiNFO. In stable conditions, HWiNFO's Core0..Core7 match [529..536] to within 0.4–0.5 °C, core-for-core, in the same order.

For context, the rest of 0x4C0007 (classic APU header, matches RyzenAdj)

field	float index	notes
fast PPT	[2] limit / [3] value	W (value == socket power)
VDD TDC	[8] limit / [9] value	A
THM limit	[16]	100 °C
per-core temp	[529..536]	°C, this issue

Only verified on 0x4C0007 (Phoenix); the sibling versions 0x4C0006/0x4C0008/0x4C0009 (Hawk Point etc.) may place it differently, so I wouldn't assume it there without a dump.

Method: read the full table via read_pm_table, sample under variable / single-core-affinity load, diff idle vs hot. Happy to share raw dumps if helpful.