Clock period improvements left on the table due to SYNTH_MINIMUM_KEEP_SIZE counting macros
Los mantenedores suelen responder en 1 día
@joaomai ya está trabajando en esto.
Desde el 22/10/2026.
Evaluación
Este issue todavía no se ha evaluado.
Descripción
Description
SYNTH_MINIMUM_KEEP_SIZE decides which modules hierarchical synthesis keeps. The expectation is that it measures the logic yosys works on: since @povik's #2552 the decision is keep_hierarchy -min_cost on yosys's cost estimate of each module, in NAND2 equivalents.
A macro is no work for yosys, but it counts. keep_hierarchy -min_cost needs a cost for every instanced blackbox, and convert_liberty_areas gives each Liberty cell, SRAM macros included, the gate equivalent of its area; since #4650 an AUTO_MEMORIES macro gets its generated view's area the same way. A small wrapper around macros therefore looks like a large module of logic and is kept, and its boundary stays wherever a path runs through it.
Example, asap7/riscv32i (SYNTH_MINIMUM_KEEP_SIZE=10000): the only module kept is dmem, four fakeram7_256x32 instances (1,376 µm² of Liberty area) and 56 µm² of address decode. Its boundary sits on the worst path, between the ALU's address adder and the SRAMs' enable decode, which synthesis then optimises apart. Through global route at 780 ps:
| min reg2reg period | reg2reg endpoints within 10 % of it: faster / slower | |
|---|---|---|
today: dmem kept |
835.6 ps | – |
dmem flattened, which is what the rule decides with macro cost 0 |
791.8 ps | 995 / 3 |
Details and the measurement in #4713 (closed, FYI only).
Suggested Solution
Make the rule measure what it is for: a blackbox macro, Liberty or AUTO_MEMORIES, costs 0 (or 1) gate equivalents towards its parent's size. keep_hierarchy -min_cost still has a cost for every blackbox, so the error that made a cost necessary does not come back.
The designs this can change are those that synthesise hierarchically with a keep threshold and have macros (make print-* on master; the gf12 designs not checked): asap7/coralnpu, asap7/riscv32i; nangate45/ariane133, ariane136, black_parrot, bp_be_top, bp_fe_top, bp_multi_top, bp_quad, mempool_group, tinyRocket. Their keep decisions change only where a macro sits under a module the rule would otherwise flatten, and their QoR may move either way.
Additional Context
variables.yaml describes the threshold as an area in NAND2 units. Stating there which is intended, the logic yosys works on or area including macros, would make the behaviour less surprising either way.
- Lenguaje dominante
- Verilog
- Estrellas
- 767
- Forks
- 544
- Merge medio
- 2 d 3 h
- PR fusionados (30 d)
- 96
Preparar el entorno
- Incluye un Dockerfile o un archivo de Docker Compose
- Sin plantilla de pull request
- Sin guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de The-OpenROAD-Project/OpenROAD-flow-scripts
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 90/100
The-OpenROAD-Project/OpenROAD-flow-scripts#4717 ·
Los mantenedores suelen responder en 1 día
-
Make SYNTH_HDL_FRONTEND=slang defaultQuizá libre de nuevo @jeffng-or la tomó hace 31 días y no hay ningún pull request abierto. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 68/100
The-OpenROAD-Project/OpenROAD-flow-scripts#4548 · 1 comentario · 1 asignado ·
Los mantenedores suelen responder en 1 día
-
Use percent of minimum clock periodPosiblemente ocupada @migueldalberto la tomó hace 22 días. Abierto
Dificultad 3/5 1-2 días Aptitud para principiantes 55/100
The-OpenROAD-Project/OpenROAD-flow-scripts#4583 · 7 comentarios · 1 asignado ·
Los mantenedores suelen responder en 1 día
-
asap7 cva6 fails in PSM with demo constraint filePosiblemente ocupada @gadfort la tomó hace 37 días. Abiertopsm
The-OpenROAD-Project/OpenROAD-flow-scripts#4539 · 3 comentarios · 1 asignado ·
Los mantenedores suelen responder en 1 día
-
GF180 platform tech LEF omits Nwell, causing SRAM OBS truncation on OpenDB importPosiblemente ocupada @bandhambalaji la tomó hace 4 días. Abierto
Dificultad 4/5 3-5 días Aptitud para principiantes 48/100
The-OpenROAD-Project/OpenROAD-flow-scripts#4496 · 2 comentarios ·
Los mantenedores suelen responder en 1 día
Todos los issues de The-OpenROAD-Project/OpenROAD-flow-scripts