Fix crash in ConvertParallelToGPU1 pass on grid×block loops with multiple inner parallels

[xys-syx]

The input IRs:

scf.parallel (%bx,%by,%bz, %tx,%ty,%tz) {
    scf.parallel (%i) = 0 to 8 { ... }
    scf.parallel (%j) = 0 to 8 { ... }
    scf.parallel (%k) = 0 to 8 { ... }
}

ConvertParallelToGPU1 pass crashed with llvm_unreachable("Unhandled case") (in ParallelizeBlockOps)

Fix with if the GPUDIMs equals 6 (grid -> [0, 3), block -> [3,6)), InnerParallelSerialization will serialize a parallel loop into scf.for.

[vimarsh6739]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b62ef6353f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve serialization for lower-dimensional launches

This guard only serializes inner sibling scf.parallel loops when the enclosing loops have six IVs total, so a valid lower-dimensional grid/block nest such as a 1D grid loop containing a 1D block loop is no longer handled. ParallelToGPULaunch pads missing dimensions to 1, and the previous depth-based check serialized these cases; with this change, multiple sibling inner parallels remain in the block body and ParallelizeBlockOps can still hit its llvm_unreachable("Unhandled case") on the second sibling.

Useful? React with 👍 / 👎.