[prim,ipgen] Fix up some hierarchies of fusesoc cores #25703
Conversation
Place the prim_mubi_pkg package into a separate prim_mubi_pkg fusesoc core, so it can be used freely for interfaces without pulling in the RTL. This helps avoid dependency cycles in the build graph. Adjust tlul:headers to depend only on the prim_mubi_pkg core. Signed-off-by: Alexander Williams <[email protected]>
| @@ -10,6 +10,7 @@ virtual: | |||
| filesets: | |||
| files_rtl: | |||
| depend: | |||
| - lowrisc:prim:subreg | |||
| - lowrisc:tlul:headers | |||
There was a problem hiding this comment.
Funny this depends only on tlul:headers and clkmgr on ip:tlul. It seems it should only one of the two uniformly.
There was a problem hiding this comment.
Since the reg_top is here, it should be ip:tlul. Nice catch!
Relatedly, I think I ought to move rstmgr_reg_pkg.sv over to the rstmgr_pkg core, so it can avoid pulling in RTL cores that can create dependency cycles. (similar to what has been done for the pwrmgr_pkg core)
Ensure packages with constants are at the leaves of the dependency tree, so other cores may depend on them without creating cycles. Adjust the pwrmgr_reg core to depend on the pwrmgr_pkg core, with pwrmgr_reg_pkg moved to the pwrmgr_pkg core. Signed-off-by: Alexander Williams <[email protected]>
Have prim_generic_flash only depend directly on the tlul headers to avoid dependency cycles. Internally, it only uses tlul_pkg for the port data types. A different core implements the reg_top with tlul logic, and prim_generic_flash's dependency to it is virtual. Signed-off-by: Alexander Williams <[email protected]>
They were all missing prim_subreg_pkg in their dependencies. Signed-off-by: Alexander Williams <[email protected]>
The adapter_sram core did not claim dependencies on prim_flop and prim_sec_anchor, even though it uses them in its RTL. Signed-off-by: Alexander Williams <[email protected]>
a-will
force-pushed
the
hierarchy-fixes
branch
from
1e0efec to
56f72ed
Compare
|
Well, I'd love for CI to not get broken from this, but I can't debug it, due to the closed-source repo that's required just to call the tool. The terms for ascentlint are antithetical to open-source hardware development, and I feel like we ought to just chuck it from the CI flow. (i.e. relegate it to sign-off of development stages only, and use the less problematic, if less complete tools for lint for PRs) |
Ensure packages with constants are at the leaves of the dependency tree, so other cores may depend on them without creating cycles. Adjust the pwrmgr_reg core to depend on the pwrmgr_pkg core, with pwrmgr_reg_pkg moved to the pwrmgr_pkg core.
Also split the mubi RTL implementations from the package with declarations and standalone utility functions.
In addition, fix up the reg tops for various ipgen'd cores to include a dependency on
prim:subreg.