Fix tool issues with automatic variables in PLRU tree

[michael-platzer]

Verilator currently does not properly support automatic variables declared inside a combinatorial block if these are only used in some branches (it incorrectly emits a latch warning for these variables). This PR moves the update logic and the PLRU output logic in plru_tree.sv into two functions to avoid this issue.

Additionally, some linting tools (e.g. Spyglass) complain when individual bits of integers are accessed via indexing (as is commonly done with logic vectors), which occurs within the current implementation of plru_tree.sv in two instances, e.g.:

https://github.com/pulp-platform/common_cells/blob/52f83eed3e92ed4ea5f79774854b0cb68618dcbb/src/plru_tree.sv#L94

Instead of extracting bit 0 of new_index by indexing (i.e., via new_index[0]) this PR casts the integer to a width of 1 (via 1'(new_index)) instead, which is equivalent but does not cause linting issues.

The code is otherwise unchanged and remains logically and functionally equivalent.

[niwis]

Thank you @michael-platzer. We recently had tool-compatibility issues with functions. Do you think the two issues could also be solved by

• adding default assignments for the variables inside the combinatorial block (to fix the latch inference)
• changing the variable type of new_index (and possibly the other variables) to a logic array, which seems to reflect the intent better

[michael-platzer]

Thanks @niwis for the prompt feedback.

Do you think the two issues could also be solved by

• adding default assignments for the variables inside the combinatorial block (to fix the latch inference)
• changing the variable type of new_index (and possibly the other variables) to a logic array, which seems to reflect the intent better

Sure, if that's the preferred solution, then let me change that accordingly.

[michael-platzer]

@niwis I applied the changes you suggested and can confirm that this fixes the issues we had.