Code and functional coverage with VCS

[davidharrishmc]

Adds code and functional coverage on Synopsys VCS. wsim --sim vcs --ccov/--fcov (with lockstep for --ccov) writes per-run .vdb databases under sim/vcs, make -C sim VcsCodeCoverage/VcsFunctCoverage merge them with urg, and regression-wally --coveragesim vcs runs the same coverage regressions as Questa. Code coverage uses line, cond (observability), branch and fsm on the core; functional coverage compiles the same riscv-arch-test covergroups. Rather than keep a second exclusion list, sim/vcs/exclusions2el translates coverage-exclusions-rv64gc.do and the RTL // coverage off pragmas into a URG exclusion file. It logs the exclusions VCS has no exact counterpart for (43 of 299, mostly where Questa excludes one row of a vector VCS scores as a whole), and, like Questa, fails on an exclusion that names a line with nothing to exclude.

Closes #1179.

🤖 Generated with Claude Code

[davidharrishmc]

Review notes: one fix before taking this out of draft. Questa, Verilator and non-coverage VCS runs are unaffected.

• Over-exclusion on VCS. sim/questa/coverage-exclusions-rv64gc.do:71-72 uses -linerange … -code b -allfalse, which on Questa excludes only the implicit-else branch of the ifs at src/fpu/fctrl.sv:155,165. sim/vcs/exclusions2el:673-682 ignores -code and -allfalse and treats -linerange without -item as all of s/b/c/e on the line, so it also excludes the reachable fmv true arms and line 165's condition vectors, and logs both as PORTED. Honour -code/-allfalse, or report the form as NOT PORTED, and audit the other options the parser drops.
• No staleness check on VCS. Questa fails on an exclusion that matches nothing (the vish-4014 grep), but exclusions2el turns "nothing on line L" into N/A (:489, :720) and VcsCodeCoverage (sim/Makefile:81-92) never fails on NOT PORTED or N/A, so a drifted exclusion passes silently on a VCS-only machine.
• exclusions2el:825 runs tclsh with PATH=/usr/bin:/bin, and the install script doesn't install tclsh. shutil.which on the caller's PATH would be more portable.
• The "41 of 296" and "scores match Questa" figures predate main's later exclusion and RTL changes.
• Main's --ccov currently fails at the exclusion step: coverage-exclusions-rv64gc.do:488,490 still look up ~ITLBMissF & … in ifu.sv, which ifu: don't fetch while an ITLB A-update walk is pending #1934 renamed to ~ITLBMissOrUpdateRawF (ifu.sv:289-290). Because exclusions2el evaluates the same file, the VCS flow hits it too until main is fixed.

🤖 Generated with Claude Code

[davidharrishmc]

Review items addressed in 527d837:

• -code b -allfalse now excludes only the all-false branch (the implicit else, or a case's missing default). On fctrl.sv:155,165 that is one VCS vector each, where it used to be 13 objects including the reachable fmv arms. -code limits -linerange to the kinds it names, and any option the script doesn't translate is now logged NOT PORTED rather than silently dropped. The current .do file only uses -scope, -linerange, -item, -code, -allfalse, -fec*row, -ftrans and -fstate.
• New STALE status: an exclusion that matches nothing in VCS and whose source line holds nothing of the kind it excludes fails exclusions2el, and so VcsCodeCoverage, matching Questa's behaviour on an exclusion that matches nothing.
• tclsh is looked up on the caller's PATH, and wally-package-install.sh now installs tcl.
• -vcs_assert off turns off only VCS's own internal checks: a failing immediate assert still reports under it.
• The description now gives the current translation counts (43 of 299 not or partly ported).

The stale ~ITLBMissF patterns in coverage-exclusions-rv64gc.do are being fixed in a separate PR against main.

🤖 Generated with Claude Code