Realization.Nvidia.SM86.LinearStepSM86

Controls. Fixed-latency forms: stall 15 with no scoreboard use, the over-stall control the momentum realization proved on the 3070 for a dependent FFMA chain. Variable-latency forms (S2R, LDG) name a write barrier and their first consumer waits on it: S2R on SB0, waited by the ISETP; every LDG on SB1, waited by the first FFMA. Without the waits the consumers read the registers' previous contents -- the RTX 3090 returned a record of zeros for the first version of this program (2026-09-22), and SM86.Scoreboard now refuses it at the checker.

54def lsControlWith =
55  (lambda unrestricted write : (family SM86Barrier) . (lambda unrestricted wait : Nat .
56    (constructor SM86Control SM86ControlValue
57      (byte 15)
58      (constructor SM86YieldMode SM86Continue)
59      write
60      (constructor SM86Barrier SM86BarrierNone)
61      (nat-to-byte wait)
62      (byte 0))))