fix(gnovm): correct softfloat add/sub for normal operands cancelling to subnormal

[omarsy]

Description

Fixes #5806.

gno run and go run disagree on float64 +/- for certain operands: when two near-equal, opposite-sign normal float64 values cancel into a subnormal result (|x+y| < 2.2e-308), gno returns a wrongly-scaled value (off by ~6 orders of magnitude), e.g. the issue's MRE returns 202154086 instead of 847895691526144.

Root cause

gno's softfloat (runtime_softfloat64.go) is a verbatim copy of Go's runtime/softfloat64.go, and the bug is in upstream's fpack64. Under heavy cancellation, fadd64 hands fpack64 a mantissa far below 1<<mantbits64 while the exponent is still normal-range. The denormal branch resets to the un-normalized (mant0, exp0) and only right-shifts to align to the subnormal exponent — the wrong direction in this case — so it returns the un-normalized cancellation mantissa at the wrong scale instead of the correctly-rounded subnormal.

This is latent in upstream Go too (it only executes on software-float targets like GOMIPS=softfloat); gno hits it because it always uses softfloat for deterministic, hardware-independent results.

Fix

Re-normalize the mantissa (left-shift) before the subnormal alignment loop — restoring the normalization invariant fpack64 already applies on its normal path (the identical loop runs at the top of the function). It is a no-op for already-normalized callers (fmul64/fdiv64/conversions), so they are unaffected; only the fadd64/fsub64 cancellation case changes. fpack32 carries the identical latent bug and is fixed for parity.

Because the file is generated (DO NOT EDIT, copied from $GOROOT), the fix is applied through the generator so it survives go generate, with an anchor guard that fails loudly if a future Go toolchain changes or fixes the upstream code.

Changes

• Bug fix — gnovm/pkg/gnolang/internal/softfloat/runtime_softfloat64.go: left-normalization loop added to the denormal branch of fpack64 and fpack32.
• Tooling — gnovm/pkg/gnolang/internal/softfloat/gen/main.go: applies the fix during go generate (with a strings.Count==1 anchor guard) plus injects the regression case into the generated test.
• Test — gnovm/pkg/gnolang/internal/softfloat/runtime_softfloat64_test.go: two cancellation operands added to TestFloat64's all-pairs (soft-vs-hardware) base slice.

Verification

• The issue's add/sub operands now return 847895691526144, matching go run.
• 0 mismatches vs hardware across 40M cancellation pairs (16M subnormal) and 20M random pairs over + - * /; the previous code was wrong on >50% of cancellation-to-subnormal pairs.
• The reachable fpack32 path (f64->f32 narrowing) and full conversion sweeps match hardware; the new loop is a verified no-op for non-add/sub callers.
• The same fix corrects a real GOMIPS=softfloat go1.22.3 build under qemu-mips (202154086 -> 847895691526144).

This bug has also been reported upstream to Go; once fixed there, the generator's anchor guard will trip on the next toolchain bump so the local workaround can be removed.

[Gno2D2]

🛠 PR Checks Summary

🔴 Pending initial approval by a review team member, or review from tech-staff

Manual Checks (for Reviewers):

• IGNORE the bot requirements for this PR (force green CI check)

Read More

🤖 This bot helps streamline PR reviews by verifying automated checks and providing guidance for contributors and reviewers.

✅ Automated Checks (for Contributors):

🟢 Maintainers must be able to edit this pull request (more info)
🔴 Pending initial approval by a review team member, or review from tech-staff

☑️ Contributor Actions:

1. Fix any issues flagged by automated checks.
2. Follow the Contributor Checklist to ensure your PR is ready for review.
   • Add new tests, or document why they are unnecessary.
   • Provide clear examples/screenshots, if necessary.
   • Update documentation, if required.
   • Ensure no breaking changes, or include BREAKING CHANGE notes.
   • Link related issues/PRs, where applicable.

☑️ Reviewer Actions:

1. Complete manual checks for the PR, including the guidelines and additional checks if applicable.

📚 Resources:

• Report a bug with the bot.
• View the bot’s configuration file.

Debug

Automated Checks

Maintainers must be able to edit this pull request (more info)

If

🟢 Condition met
└── 🟢 And
    ├── 🟢 The base branch matches this pattern: ^master$
    └── 🟢 The pull request was created from a fork (head branch repo: omarsy/gno)

Then

🟢 Requirement satisfied
└── 🟢 Maintainer can modify this pull request

Pending initial approval by a review team member, or review from tech-staff

If

🟢 Condition met
└── 🟢 And
    ├── 🟢 The base branch matches this pattern: ^master$
    └── 🟢 Not (🔴 Pull request author is a member of the team: tech-staff)

Then

🔴 Requirement not satisfied
└── 🔴 If
    ├── 🔴 Condition
    │   └── 🔴 Or
    │       ├── 🔴 At least one of these user(s) reviewed the pull request: [aronpark1007 davd-gzl jefft0 notJoon omarsy MikaelVallenet] (with state "APPROVED")
    │       ├── 🔴 At least 1 user(s) of the team tech-staff reviewed pull request
    │       └── 🔴 This pull request is a draft
    └── 🔴 Else
        └── 🔴 And
            ├── 🟢 This label is applied to pull request: review/triage-pending
            └── 🔴 On no pull request

Manual Checks

**IGNORE** the bot requirements for this PR (force green CI check)

If

🟢 Condition met
└── 🟢 On every pull request

Can be checked by

• Any user with comment edit permission