set wctl.is_cond_neg to 0 to avoid a x defined tmask in sfu in 4 stat…

[JulessD]

When running Vortex on questasim, with 4 state simulation, I came accross an issue where the then_mask went x.
Upon investigation, the cause was traced back to the signal is_cond_neg.
While the value of is_cond_neg has no impact on join result it enters into the logic equation for the then_mask.
If it is left at x, in 4-state simulation, the then_mask is assigned x which creates problems.
Assigning is_cond_neg to 0 adds no logic and ensures the then_mask is not x.

After this fix, I was able to run multiple kernels with branches on questa.

[tinebp]

Verified and merged as 62ef86d, authored to you — thanks for a precise report AND
the right-sized fix. We confirmed the mechanism: the wctl unit folds is_cond_neg into
the join mask as (taken & tmask) | (~taken & tmask), which cancels in 2-state and in
hardware (so no silicon defect and no Verilator symptom) but survives the OR as X in
4-state and drives an all-X tmask into the warp scheduler on every JOIN. Reproduced in
xsim: X before, 0 after, join mask defined; rtlsim cycles bit-identical. Your one-bit
zero at the one leaking arm was exactly right — the remaining wctl fields stay
don't-care on purpose (their consumers qualify them), so we kept your change as-is and
only condensed the comment to house style. Your report also led us to fix one more
declare-before-use straggler (VX_mem_from_axi) that blocked 4-state compilation of the
RTL libs (bcba2ac).