generates wrong sized initializations on its own generated localparams

[pbreuer]

I'm seeing a lot of:

        // removed localparam type sv2v_uu_icache_SelHPTW
        localparam [0:0] sv2v_uu_icache_ext_SelHPTW_0 = 1'sb0;
        localparam sv2v_uu_icache_WORDLEN = 32;
        // removed localparam type sv2v_uu_icache_ByteMask
        localparam [3:0] sv2v_uu_icache_ext_ByteMask_0 = 1'sb0;
        localparam sv2v_uu_icache_LOGBWPL = AHBWLOGBWPL;

I don't know exactly what is going on here but the initialization constants are the wrong size for some of these declarations, as seen above for [3:0].

Regards

PTB

[zachjs]

Thank you for filing this issue! For these localparams, sv2v relies on the sign and size extension of assignment like contexts defined in the Verilog standard to simplify its conversion. That is,

localparam [3:0] sv2v_uu_icache_ext_ByteMask_0 = 1'sb0;

is behaviorally equivalent to

localparam [3:0] sv2v_uu_icache_ext_ByteMask_0 = 3'b000;

Do you agree? If not, what am I missing?

[pbreuer]

Hi ...

On Sun, 22 Dec 2024 07:12:17 -0800 Zachary Snow ***@***.***> wrote: Thank you for filing this issue! For these localparams, sv2v relies on the sign and size extension of assignment like contexts defined in the Verilog standard to simplify its conversion. That is, ```Verilog localparam [3:0] sv2v_uu_icache_ext_ByteMask_0 = 1'sb0; ``` is behaviorally equivalent to ```Verilog localparam [3:0] sv2v_uu_icache_ext_ByteMask_0 = 3'b000; ``` Do you agree? If not, what am I missing?

Well, the problem is that verilator complains about it and since comments are elided I can't turn the warnings off as verilator reads comments to know that an issue has been spotted and deemed nondangerous. So sv2v is generating hundreds of verilator warning messages that come from the conversion alone, making them useless for diagnostics, and hiding any actual problems. I'm afraid one has to do this. The behavioral semantics of code is not quite the full semantics. Is there difficulty in using the same "[3:0]" type that is output to generate "4'b" instead of "1's" ? It looks like practically a one-line change in the haskell. best regards PTB

[zachjs]

Is there difficulty in using the same "[3:0]" type that is output to generate "4'b" instead of "1's" ? It looks like practically a one-line change in the haskell.

Indeed, commenting out the following logic in src/Convert/UnbasedUnsized.hs should yield the behavior you desire.

convertAsgn (lhs, UU bit) =
    (lhs, literalFor bit)

It is technically possible to add a command line flag to toggle this behavior, but I fear this is but one of the many arbitrary decisions a linter could make. verilator itself has the option to disable this warning, but I'm guessing that's not what you want. Before proceeding, I'm at least curious if you could try that local sv2v modification above and see if it actually yields the overall behavior you want.

If the complaints are isolated to the sv2v_uu localparam declarations, it may be possible to use sed to post-process the output to convince verilator we know what we're doing for these particular snippets.