Extend two existing opt passes and add one new pass to collapse
serial/dynamic-index structures that were leaving high logic depth:
- opt_first_fit_alloc: recognize the "coalesce-matrix" first-fit
allocator variant (same_cat[i][k] coalescing gated on the leader's
enable, driven from a raw input enable). Rewrite both the lane_slot
allocation and the xbar field gather from one shared log-depth scan.
- opt_prienc: detect round-robin / rotated-priority scans (req scanned
from idx_last downward with wraparound) and rewrite the depth-N
idx--/req[idx] mux chain to rotate -> log-depth priority-encode ->
unrotate.
- opt_priokey (new): recognize priority-by-key one-hot accumulators and
replace each dynamic taken[key] read ($shiftx/$bmux) with the
equivalent pairwise-key-compare reduction, dropping the wide dynamic
indexing. Supports -strict for full-key-range formal validation.
Each includes self-contained tests (equiv_opt / sat -prove-asserts,
mux-bound and negative cases) in tests/opt/.
Co-authored-by: Cursor <cursoragent@cursor.com>
Address review feedback on the fixpoint loop:
- Scope the per-sweep cleanup to the module under rewrite via
Pass::call_on_module(..., "clean -purge") instead of running clean over
the whole design selection. This avoids O(N^2) work across modules and
keeps untouched modules' dangling cells until their own sweep, matching
the original single-call behavior.
- Emit a log_warning if a module fails to reach a fixpoint within
max_sweeps, so silent truncation of compaction is visible.
Co-authored-by: Cursor <cursoragent@cursor.com>
- pack_lanes: assert elem_w < 32 and pack the full element width instead
of silently dropping bits >= 31.
- Remove the dead `cell` struct member and its unused assignment in run()
(every emit helper shadows it with its own local `cell`).
- Decorrelate the pseudo-random bc bits from en (independent mix) so they
no longer share an LFSR bit (e.g. en[7]/bc[0] for n=8).
- Add purpose comments to fingerprint_dsel and lane_of_bit.
Declined the std::stoi/std::stoll arg-parsing suggestion: it matches the
established convention in sibling passes (opt_argmax, opt_priority_onehot).
Co-authored-by: Cursor <cursoragent@cursor.com>
The tests/silimate suite (which aborted the Run tests CI job) exposed two
issues in the generalized passes:
- opt_compact_prefix: the forward dense pack regressions at 64 and 128 bits
no longer rewrote. The ConstEval fingerprint was uint64_t-based (capped at
62 bits) and the per-cone cell cap (max_width*96) was below the O(width^2)
cell count of a wide pack. The fingerprint now drives whole-width Const
bit patterns (no width cap) and the cone cap scales quadratically; total
work stays bounded by the shared walk/eval budgets.
- opt_priority_onehot: the "max-width below lane count" negative test set
max_width=8 on a 16-lane design expecting no rewrite, but the generalized
matcher legitimately (and equivalence-provably) rewrites the 8-lane
sub-region. The test now uses max_width=3 (below min_width 4) to verify
the width gate suppresses all matching.
Co-authored-by: Cursor <cursoragent@cursor.com>
- update CMakeLists.txt to load two new files:
- SilimateConfig.cmake: sets Silimate configuration options and defaults
- SilimateVerific.cmake: compiles Verific library, optionally with Silimate modifications
- include silimate tests in test Makefile
