Swept the stack-heavy measurement across placements, and qualified its result

[fdesbiens]

#636 reported that a stack in BTCM gave a 3.4× tighter spread than DRAM0 for stack-heavy work. That measurement used a single code placement — the exact methodology #631 and #633 exist to correct. The cache benchmark got an alignment sweep and the interrupt handler got one; this measurement never did.

It surfaced when #637 added two threads to the same image and the figure moved: both spreads came out near 6,500 and the minima rose 15%.

Swept

The recursive body is now generated at four placements, all four measured per placement in one image:

placement	BTCM min / spread	DRAM0 min / spread
offset 0	41,854 / 6,850	42,036 / 6,880
offset 16	48,670 / 1,946	48,792 / 1,978
offset 32	42,388 / 6,914	42,752 / 7,018
offset 48	48,914 / 1,860	49,198 / 6,786

Reproducible across runs to within a few hundred cycles.

What it means

Spread is dominated by code placement, not by the memory holding the stack. It ranges 1,860 to 6,914 depending on where the body falls in a cache line, and placement moves the minimum by 17% (41,854 to 49,214).

Against that, the memory contributes a small but consistent advantage: BTCM's minimum is lower at all four placements, by 0.4% to 0.9%.

BTCM's spread beats DRAM0's decisively at one placement of four — offset 48, 1,860 against 6,786. At the other three they are within 2% of each other.

#636 should be read as qualified by this

The effect it reported is real where it occurs but is not a property of the part. Quoting it as one invited the reader to expect it everywhere. A stack in BTCM buys a small consistent improvement in the best case, and a large improvement in spread at some code placements and not others. Anyone building a determinism argument on it needs the placement sweep in the loop, not a single figure.

Method note

The pad nops that displace each placement execute on every recursion level rather than once, so each placement carries a slightly different constant cost — about 0.6% at the widest pad. That cancels in the BTCM-against-DRAM0 comparison, which is made at the same placement, and does not affect spread within a placement.