Sibling transitions are now detected in smf_set_state() when source and
destination share the same parent, removing the need for states to flag
the sibling transition explicitly.
Restore include/zephyr/smf.h and lib/smf/Kconfig to match main so this
change only carries the functional update in lib/smf/smf.c, and add a
micro-benchmark test. For accurate statistics, run the test on a board.
Micro-benchmark setup
- Board: NUCLEO-F746ZG (STM32F746, ~192 MHz)
- 200000 iterations per state transition
- Simple HSM (letters are nodes; R is root):
=== SMF transition micro-benchmark ===
R
/ \
A B
/ \ / \
C D E F
/ \ \
G H J
Current smf_set_state:
======================
Sibling Transitions
(A->B) : 459 cycles/transition (2390 ns)
(C->D) : 482 cycles/transition (2510 ns)
(G->H) : 522 cycles/transition (2718 ns)
Other Transitions
(G<->G) : 237 cycles/transition (1234 ns)
(C->G) : 343 cycles/transition (1786 ns)
(A->H) : 452 cycles/transition (2354 ns)
(G->D) : 651 cycles/transition (3390 ns)
(D->E) : 752 cycles/transition (3916 ns)
(J->D) : 893 cycles/transition (4651 ns)
(J->G) : 1077 cycles/transition (5609 ns)
New smf_set_state:
==================
Sibling Transitions
(A->B) : 356 cycles/transition (1854 ns)(22% faster)
(C->D) : 356 cycles/transition (1854 ns)(26% faster)
(G->H) : 356 cycles/transition (1854 ns)(32% faster)
Other Transitions
(G<->G) : 258 cycles/transition (1343 ns)(9% slower)
(C->G) : 356 cycles/transition (1854 ns)(4% slower)
(A->H) : 464 cycles/transition (2416 ns)(3% slower)
(G->D) : 707 cycles/transition (3682 ns)(9% slower)
(D->E) : 797 cycles/transition (4151 ns)(6% slower)
(J->D) : 970 cycles/transition (5052 ns)(9% slower)
(J->G) : 1157 cycles/transition (6026 ns)(8% slower)
This change makes sibling transitions
deterministic and cheaper (356 cycles on nsim),
at the cost of a small increase for deeper/LCA
transitions (~7–9%). This is a net win for state machines
that mostly hop between siblings.
Signed-off-by: Vladislav Kulikov <vlad_kulikov_c@pm.me>
286cefbb30