[RFQ] Bazel orfs beta test#4163
Conversation
Updates rules-base.json for: asap7/aes-block, asap7/riscv32i, asap7/mock-alu, gf180/aes-hybrid, gf180/uart-blocks, nangate45/tinyRocket, nangate45/ariane133, nangate45/ariane136, nangate45/bp_be_top, nangate45/bp_fe_top, nangate45/swerv_wrapper, sky130hd/microwatt.
- Bump bazel-orfs to d998bbff78c4557 - Add bazel-orfs-verilog transitive dependency - Update rules_python to 1.8.5, remove compatibility_level - Bump ORFS Docker image to 26Q1-816-gf40d2f346 - Add variables_yaml to orfs.default() - Add orfs_designs() repo rule for all 6 public-PDK platforms - Add commented-out OpenROAD-from-source instructions
Add orfs_design() targets and exports_files() to all design directories across asap7, gf180, ihp-sg13g2, nangate45, sky130hd, and sky130hs platforms, plus their source directories.
Add bazel-orfs.md with quick start guide, available targets, build results for 59 designs across 6 platforms, and status of blocked designs.
Add transitive dependency overrides required by the new bazel-orfs: openroad, qt-bazel, mock-klayout, glpk, and toolchains_llvm. The new bazel-orfs builds OpenROAD from source by default. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
Use local_path_override to point the openroad module at the tools/OpenROAD submodule rather than fetching from GitHub. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
The rules_python bootstrap wrapper sets RUNFILES_DIR to its own runfiles tree. When run_command.py spawned OpenROAD, it inherited that variable and looked for Tcl init files in the Python runfiles instead of its own, causing "application-specific initialization failed" in Bazel builds. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
The `include` filegroups at rtl/BUILD.bazel and rtl/register_interface/include/BUILD.bazel only globbed their own directory, so headers referenced via relative include paths from subdirectories (e.g. `common_cells/registers.svh`, `axi/typedef.svh`, `register_interface/typedef.svh`) were missing from the yosys sandbox. Explicitly list the .svh files that live in sibling/child packages so VERILOG_INCLUDE_DIRS resolution finds them. Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
`export OPENROAD_EXE := $(OPENROAD_EXE)` rebinds the variable to origin "file", which puts it on the UNSET_VARIABLES_NAMES list that UNSET_AND_MAKE clears before invoking a sub-make. In sub-make the variable is gone and the `?=` fallback resolves to the in-tree `tools/install/OpenROAD/bin/openroad` path that does not exist in a Bazel sandbox, so every stage invoked via UNSET_AND_MAKE (floorplan, place, route, final) fails with "openroad: No such file or directory". Using a bare `export` preserves the original origin (environment when bazel-orfs supplies it, file when the local default fills in) so the env value survives the UNSET_VARS step and sub-makes see the right binary. Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request integrates bazel-orfs into the OpenROAD-flow-scripts, enabling Bazel-based builds for various designs across multiple PDK platforms. It adds extensive BUILD.bazel files, updates dependencies and toolchains in MODULE.bazel, and provides new documentation. Feedback suggests deduplicating commit hashes in MODULE.bazel to reduce maintenance burden and using a more robust lookup method for PDK extensions in flow/BUILD.bazel to prevent potential build failures.
| git_override( | ||
| module_name = "bazel-orfs", | ||
| commit = "f8a4b694b37c8f5322323eba9a9ae37f9541ee17", | ||
| commit = "f8b36bf4a92c6e84317a70bab215bdcd2e25a26e", |
There was a problem hiding this comment.
The commit hash f8b36bf4a92c6e84317a70bab215bdcd2e25a26e is duplicated in multiple places throughout this file (lines 20, 28, 37, 38, 45, 46). This redundancy increases the maintenance burden and the risk of partial updates. If the bump tool mentioned on line 10 does not automatically handle all these locations, consider documenting the manual update process or investigating if the overrides can be simplified.
| "nangate45": ["cfg", "gds", "lef", "lib", "lyt", "mk", "rules", "tcl", "v"], | ||
| "sky130hd": ["gds", "lef", "lib", "lyt", "mk", "rules", "tcl", "tlef", "v"], | ||
| "sky130hs": ["gds", "lef", "lib", "lyt", "mk", "rules", "tcl", "tlef", "v"], | ||
| }[pdk] |
There was a problem hiding this comment.
Using direct indexing [pdk] instead of .get(pdk, []) makes the build fragile. If a new PDK is added to the list at the end of the file but omitted from this dictionary, the build will fail with a KeyError during the loading phase. Reverting to .get() with a default value is more robust against such omissions.
| }[pdk] | |
| }.get(pdk, []) |
|
out of sight, out of mind for now. |
1 participant
No description provided.