draft: hardware partition and XCP capacity - #17

Draft

fishman wants to merge 12 commits into

Project-HAMi:mainfrom

fishman:partitioning

Commits on Aug 19, 2026

testdata: add MI355X node fixtures for hardware-free regression tests
fishman
committed
5615f76
amdgpu: configurable sysfs root, amdsmi memory-partition and profile lookups
fishman
committed
4c6e4b1
plugin: injectable AMD SMI seams, fixture-backed registration test
fishman
committed
f201f2e
README: document partition profiles and mode-change semantics
fishman
committed
8231d94
allocator, utils, labeller: nil-guard stale device ids, dual allocation keys
fishman
committed
e09a30a
plugin, amdgpu: advertise per-XCP capacity as an even share of the GPU
fishman
committed
b4cdf74
testdata, README: xgmi/topology fixtures, per-XCP capacity, memory-partition rationale
fishman
committed
ee530f3
amdgpu, plugin: boot-time partition template catalog in memory
fishman
committed
85bd2ef
plugin: register spx whole GPUs, tolerate stale kubelet ids in GPA
fishman
committed
3c2ab69
feat(partitioning): operating-mode switch and API-driven compute partition flip
fishman
committed
9096b6a
plugin: tolerate mixed partition styles on a node
fishman
committed
b8e1781
allocator: warn prominently when GPA sees unknown device ids
fishman
committed
c11882c