cp_fp_reg_edges: honor _frm suffix on fs2 and fs3 generators

[Aditya Amol Kuchekar (dev-aditya-hub)]

Summary

make_fs1_edges in cp_fp_reg_edges.py honors a _frm suffix on the coverpoint name and crosses each edge value with all six rounding modes. make_fs2_edges and make_fs3_edges ignore the suffix and emit one frm-unspecified test per edge value.

Problem

Because the registry uses longest-prefix matching, a coverpoint named cp_fs2_edges_frm routes to make_fs2_edges, which silently drops the rounding-mode cross. The label says the suite covered the cross-product; the generated assembly did not. A DUT with a rounding-path bug that only triggers for specific fs2 or fs3 edge values (e.g., subnormal fs2 with RTZ on fmadd.s) is never exercised, while the symmetric fs1 case is.

Fix

Mirror the cross_frm / frm_modes handling from make_fs1_edges into make_fs2_edges and make_fs3_edges, including the bin-name and description suffixes so the generated labels stay distinct per frm mode.

Risk

Low. When the coverpoint does not end in _frm, frm_modes = [None] and the new loop runs exactly once per edge value with frm=None, producing the same output as before. Only coverpoint names containing _frm are affected.

[Copilot]

Pull request overview

This PR fixes a gap in the floating-point coverpoint generators where cp_fs2_edges* and cp_fs3_edges* variants containing _frm were not actually crossing edge-value tests with all rounding modes, despite the coverpoint name implying that behavior. It aligns make_fs2_edges/make_fs3_edges behavior with the already-correct make_fs1_edges implementation.

Changes:

• Add _frm detection to make_fs2_edges and make_fs3_edges and iterate edge values across ("dyn", "rdn", "rmm", "rne", "rtz", "rup") when requested.
• Make bin names and descriptions frm-specific so per-frm generated cases remain distinct and accurately labeled.

[Aditya Amol Kuchekar (dev-aditya-hub)]

Heads-up on landing order for this one. I have two other PRs touching the same file/concept:

• Share cp_fs{1,2,3}_edges implementation; enable cross_frm for fs2/fs3 #1469 refactors the three make_fs{1,2,3}_edges generators into a single shared helper. If that lands first, this PR's diff collapses into the shared helper.
• frm coverpoints: sweep all 5 fcsr.frm values for dyn rounding mode #1467 expands the dyn rounding mode into 5 explicit fcsr.frm sweep variants. If that lands first (or the precedent it sets is accepted), the frm_modes tuple here should follow the same pattern for the dyn case.

Happy to sequence however you prefer. My read is #1469 first (pure refactor, no behavior change), then #1554, then #1467 on top of the deduped helper. But I'll defer to whatever order makes the review easier.

[David Harris (davidharrishmc)]

I am concerned this change skips the tests entirely for cp_fs2_edges (without _frm) because frm_modes is set to None, and then the for loop doesn't iterate over any modes and doesn't print anything. Could be mistaken, but hard to be certain without looking at the emitted code.

The floating-point tests are placeholders and will be superseded this summer by more comprehensive tests based on the IBM coverpoints. Not sure that it's worth investing more time in tweaking and reviewing the placeholders.

[Aditya Amol Kuchekar (dev-aditya-hub)]

On (1): frm_modes is [None] (a 1-element list), not None the loop runs once with frm_mode = None, producing the same single test chunk per edge value as the previous make_fs2_edges. The cp_fs2_edges (no _frm) output is byte-identical to pre-refactor happy to attach a diff of generated tests if it'd help confirm.

On (2): fair point. #1469 (the refactor) already merged, so the shared helper is in tree. This PR is just the small follow-up that lets fs2/fs3 actually honor the _frm suffix (today they silently fall through to a single frm mode). If the IBM-coverpoint replacement is landing soon and you'd rather not carry even this small delta, happy to close your call.

[Jordan Carlin (jordancarlin)]

As far as I can tell, this was fully done in #1469. Please update this PR if there is anything else that actually needs to change. For now, I’m going to close it.