Fix #1317: Variable written in an always_ff should not driven by other process

[MrCookieeeee]

This PR fix #1371

This PR adds a check for variables written by always_ff processes, following the SystemVerilog single-writer rule for always_ff.

Implementation

1. First, it scans all always_ff processes and collects the Nexus objects written by each process through nex_output(). A map is used to record which always_ff process first writes each Nexus. If another always_ff process writes the same Nexus, an error is reported.

2. Second, it scans the remaining non-always_ff processes and checks whether any of them also write a Nexus that was already written by an always_ff process. If so, an error is reported. Compiler-scheduled initialization processes marked with _ivl_schedule_init are skipped, so declaration initializers are not treated as conflicting procedural writers.

Overhead

For time complexity, this check walks the process list twice and calls nex_output() once for each relevant process. The map/set lookups are logarithmic, so the cost is roughly proportional to the total number of procedural write targets, with log(N) overhead from the containers. I think this should be acceptable for normal elaboration-time checking, but I would appreciate feedback if there is a better or more idiomatic way to do this in Icarus.

Discussion

One limitation is that this check currently works at the Nexus level and does not explicitly distinguish disjoint bit/part-select writes. For example:

always_ff @(posedge clk)
    q[3:0] <= a;

always_ff @(posedge clk)
    q[7:4] <= b;

I did not add special handling for this case yet. Please let me know whether this is necessary. If so, I could add some check on it. :D

[MrCookieeeee]

Hi,

I have updated the code and resubmitted the PR.

The changes are:

1. In elaborate.cc, I removed the redundant if (outputs.size() == 0) check and changed the iterator declaration to use auto cur to keep the line shorter.

2. I added the new regression test always_ff_lval_fail to regress-vvp.list, together with the corresponding always_ff_lval_fail.json and always_ff_lval_fail-iverilog-stderr.gold files.

Thanks.

[MrCookieeeee]

Copy, i will fix it real quick.