VCS simulation speed vs -debug_access level: pick the right baseline when costing waveform dumps

Measured on one RISC-V core design (CVA6-class) with VCS, untraced runs from reset to the end of the workload, n=3 on a busy shared host (CPU ≈ wall; rep ranges didn't overlap): - no debug access: 4.90 kHz; - -debug_access+simctrl (enough for save/restore and sim control): 4.12 kHz, about 16% slower; - -debug_access+all -kdb (what full-design dumping / debug-database flows want): 1.79 kHz, about 2.7x slower.

Why it matters: when you estimate savings from 'dump on demand' (run untraced, then re-run or restore only the failing window with dumping on), the untraced pass must use a build that can actually do the later step. Comparing against a no-debug build overstates the savings. Comparing against a +all build overstates the cost of running untraced. Pair each arm with its own build, and report build flags next to every speed number.

Caveats: one design, one host, contended; the ratio will vary with design and with how much of the design the debug access exposes. Check your own design before quoting it.