Repository navigation
Conversation
Transfer the removed child shared pointer into the newly split CUGR tree node at its last use. This avoids an unnecessary reference-counted copy while preserving the child ownership needed after parent removal. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>
…cal-builder tst: add physical design builders for C++ unit tests
ant: return structured antenna check results
The wheel kept repository paths (python/openroad/, src/odb/odb.py), so `import openroad` failed after `pip install`, and the per-tool modules the tests import (odb, utl, ant, grt, ...) existed only as shims that test/bazel_test.sh wrote at test time. - python/modules holds the per-tool modules as files. openroadpy depends on them, the wheel installs them and the openroad package at the top level, and bazel_test.sh puts them on the path instead of writing shims. - odb_py, utl_py and gpl_py no longer add their source directories to sys.path. A py_test's `import odb` had loaded a second copy of the module next to the src.odb.odb that openroadpy uses, so isinstance failed across the two. - Hermetic py_test examples: openroad_test (unittest) and wheel_test, which installs the wheel into an empty directory and imports it in an isolated interpreter. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
… Tech From Python, set_thread_count() and thread_count() act on the OpenRoad instance owned by the current Tech. When none is alive (never created, or already garbage collected) they dereferenced a null pointer. Raise ORD-0105 instead; Tcl always has an instance and is unaffected. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Every caller of ord::getOpenRoad() dereferences the result, so raise ORD-0105 there rather than only in the thread count functions. This covers get_db() and the other module-level Python functions, and getLogger() now goes through it. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Add cgt, dpl, exa, fin, par, ppl, psm, rsz and wmk, matching the modules `openroad -python` registers, so the wheel and tests can import them. wmk_py no longer puts src/wmk on sys.path, as for odb. syn, ram and web have no Python extension in the Bazel build and are left out. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
rmp.py fell back to openroadpy when rmp_py was missing, but openroadpy has no rmp API, so in the wheel (which did not ship rmp_py) `import rmp` succeeded with none of rmp's classes. Re-export src.rmp.rmp_py like the other modules, ship it through openroadpy, and keep src/rmp off sys.path. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
openroadpy now re-exports odb, utl and rmp, which stay separate extensions because other extensions import them, so `from openroadpy import *` is the whole API and every python/modules file is that one line. wmk adds an alias: in the shared namespace CtsOptions is the cts class, and the watermark options are WatermarkCtsOptions. gpl's classes were built twice, in openroadpy and in a separate _gpl.so, giving two PlaceOptions and Replace classes. Drop the separate extension (25 MB of the wheel), and drop src/wmk/wmk.py, which python/modules/wmk.py replaces. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Signed-off-by: Peter Gadfort <gadfort@zeroasic.com>
Signed-off-by: Peter Gadfort <gadfort@zeroasic.com>
Signed-off-by: osamahammad21 <osama21@aucegypt.edu>
- Guard openroadpy's re-export of odb, utl and rmp with #ifdef BAZEL. CMake registers them as top-level modules, so the src.* imports broke `openroad -python` at startup. - //:openroad_py depends on //python/modules, so `import odb` still works for downstream users of the public target. - bazel_test.sh no longer puts every src/* directory on PYTHONPATH, which allowed a second copy of the extension modules to load. - Correct the comment and ORD-0105 message: destroying the current Tech clears the instance even while older Techs are alive. - openroad_test frees its Tech explicitly instead of relying on immediate collection; wheel_test installs the wheel with pip rather than unzipping. - Drop a stale comment left from the removed gpl extension. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
…eel-packaging python: make the openroad wheel installable and add Python unit tests
Add a py_wrap cc_library (alwayslink) in each tool package that compiles the tool's own SWIG Python wrapper, as CMake builds one static library per module. The next commit links them into one extension library. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
The Bazel Python bindings compiled every tool into one combined module, openroadpy, with odb, utl and rmp as separate extensions. The combined module flattened all tools into one namespace (cts and wmk both define CtsOptions, so wmk's was renamed), and each separate extension carried its own static copy of odb, utl and the logger. Link every tool's SWIG module (ord_py, odb, utl, ant_py, cts_py, ...) into one library, _openroad.so. python/modules/_openroad_loader.py loads each module's init function from that library with ExtensionFileLoader, the counterpart of CMake's PyImport_AppendInittab: the tools keep separate namespaces while the dynamic loader maps the library once, so they share one copy of the C++ state. `import openroad` imports every tool module, as `openroad -python` does at startup, so objects Design returns get their tool's class. The flat openroadpy module is gone; its one user, test/downstream, now imports from openroad. - The wheel ships _openroad.so instead of _openroadpy.so, _odb.so, _utl.so and _rmp_py.so (52 MB, from 64 MB). - odb_py, utl_py and rmp_py are aliases of //:openroad_py. - par's vector templates were skipped under Bazel to avoid duplicates in the combined module; as its own module par needs them. - ord::getLogger() falls back to the default logger when no Tech exists, so utl's Python logging works before a Tech is created and otherwise uses the Tech's logger. - test/helpers.py takes PlaceOptions from gpl under Bazel too. - modules_test checks the separate namespaces, the single library mapping, objects crossing modules and the shared database. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
odb, utl, gpl and wmk were the only Bazel modules not named <tool>_py, a leftover from when they were standalone extensions. Name them odb_py, utl_py, gpl_py and wmk_py as CMake does, so every extension in _openroad.so is _<tool>_py and the loader needs no special cases. The package paths stay (src.odb.odb_py), since other generated wrappers import odb and utl through them; `import odb` and `import utl` are unchanged. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
The finder matched the last component of any module name, so after `import openroad` another package's _ant_py (or a top-level one) would be loaded from _openroad.so. Match the exact names the generated wrappers import: _ord_py and src.<tool>._<tool>_py. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Every tool's py_wrap repeated the same defines, visibility, alwayslink and common deps. python_wrap_library in bazel/python_wrap_cc.bzl holds them, so each tool lists only what is its own. modules_test's /proc/self/maps check is Linux-only, so it is now a separate test skipped elsewhere; the check that every module loads from the one library still runs everywhere. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
syn: give the exported block the tech's DEF units
odb: check for rounding issues in DEF in and warn on doubles in int places
What: UnbufferMt, the Mt form of buffer removal, joins the PARALLEL_EVAL WNS tail. Why: removing a buffer on the worst path is a gain the other Mt moves cannot make. How: the main thread checks the buffer and estimates the saved delay; the tail removes buffers only in single-move batches, since a removal cannot be reverted in place. LEGACY is unchanged. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>
What: SizeDownFanoutMt, the Mt form of SizeDownFanout, joins the PARALLEL_EVAL WNS tail, and every Mt generator follows the same steps: isApplicable() and prepareInputs() on the main thread, generate() and estimate() on the workers. Why: the last serial move with an Mt form, and one rule for which code may read live STA. How: SizeDownFanoutMt measures the step-down cell per off-path load; the virtual MoveGenerator::prepareInputs(Target&) replaces the per-move prepare branches of the policy. PARALLEL_EVAL also stops at the walked -max_utilization area. LEGACY is unchanged. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>
What: profile mode 3 records what every move did to timing, and report_delay_estimator_accuracy_for_rebuffering compares the buffer-tree estimate of a driver with STA. Why: to find the moves whose estimate misled the repair and to measure the buffer-tree estimator. How: RepairSetupAnalysis snapshots each move and samples timing after it; the report plans the trees as RebufferMt does, applies the best one in an ECO journal and undoes it. Lower profile modes do not change. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>
What: architecture.md describes the setup repair code of this stack: the policy model of LEGACY and PARALLEL_EVAL, the Mt generator steps and threading, the PARALLEL_EVAL batch loop, batch outcome, stop rules, WNS tail stage and effort levels, and the profiler. Why: the document still described the code before the stack. How: documentation only. Signed-off-by: Jaehyun Kim <jhkim@precisioninno.com>
…z-estimator-side-models rsz: Model side paths in the Mt delay estimator
…z-rebuffer-mt rsz: Add the RebufferMt buffer-tree move
…z-parallel-eval rsz: Add the experimental PARALLEL_EVAL setup repair policy
…ute reroutes, not only the edited ones Signed-off-by: Eder Monteiro <emrmonteiro@precisioninno.com>
…z-split-load-mt rsz: Add SplitLoadMt to PARALLEL_EVAL
Signed-off-by: Eder Monteiro <emrmonteiro@precisioninno.com>
…z-unbuffer-mt rsz: Add UnbufferMt to the PARALLEL_EVAL WNS tail
…z-size-down-mt rsz: Add SizeDownFanoutMt to the PARALLEL_EVAL WNS tail
…z-move-analysis rsz: Add the per-move analysis and the rebuffer accuracy report
…z-architecture-doc doc: Update the rsz setup repair architecture
…into HEAD Signed-off-by: Jorge Ferreira <jorge.ferreira@precisioninno.com> # Conflicts: # src/web/src/web.cpp # src/web/test/bundle_test.py
Signed-off-by: Jorge Ferreira <jorge.ferreira@precisioninno.com>
…mental-reroute-parasitics
…I-bundle-5-cmake-dist web: embed the checked-in bundle in the CMake build
…-report Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
With -bloat_occupied_layers, a layer whose only shape is an obstruction with MINSPACING or DESIGNRULEWIDTH was not marked occupied, so it did not receive the full-layer cover. Include modified obstructions in the occupancy map when bloating while still writing them individually. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
…I-bundle-6-report web: build the saved report from the bundle
…pins ant: use per-layer and hierarchical pin antenna data
The sizing leakage limit is relative to the current cell, so a cell with unusually low leakage can have every larger size excluded. In sky130hd, o211ai_1 leaks 4.9x less than o211ai_2, so with the default 4x limit repair_timing could never upsize it. A weak o211ai_1 driving a long fanout-1 net then stayed the worst path through the whole flow, since BufferMove does not apply to fanout-1 nets either. When no larger cell passes the leakage limit, keep the smallest larger cell that passes every other restriction. Cells that already have an allowed larger size are unchanged. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Find the smallest larger leakage-limited cell, and whether any larger cell is already swappable, during the main loop. Insert the fallback at the position it held in the equivalent cell order instead of rebuilding the list. No change in the swappable cells. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
…difiers odb: preserve LEF box spacing modifiers
Allowing one upsize step under the leakage limit improves timing on aes_sky130hd (DRT WNS -1.24 -> -1.11 ns, TNS -222 -> -184 ns, hold WNS -0.09 -> -0.06 ns). Post-route max slew and capacitance slack get worse (-10% -> -20% and -10% -> -22%), which is outside the -12% limits. The limits and metrics are regenerated from the bazel run. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
Cells with no larger size under the leakage limit can now take one upsize step, which changes repair_timing results: - repair_setup_path14: WNS -0.885 -> -0.888, TNS -2.3 -> -3.0 - repair_setup_path15: WNS -1.281 -> -1.263, TNS -6.8 -> -6.9 - repair_setup_path16: WNS -6.018 -> -6.007, TNS -1853.5 -> -1852.0 - repair_setup_path19: WNS -1.816 -> -1.755, TNS -57.7 -> -56.0 Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
…ge-limit-min-upsize rsz: always allow one upsize step under the leakage limit
repair_antennas -ratio_margin tightens the limits so repair leaves headroom, but only the partial ratios (PAR/PSR) used it; the cumulative ones (CAR/CSR) were checked against the full limit. Apply the margin to all four. check_antennas always uses a margin of 0, so its results don't change. Signed-off-by: Matt Liberty <mliberty@precisioninno.com>
…argin ant: apply ratio_margin to cumulative ratios too
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
[Describe your changes here]
Type of Change
Impact
[How does this change the tool's behavior?]
Verification
./etc/Build.sh).Related Issues
[Link issues here]