Dev/fly pa reduce jit build

[Bernard-Liu]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 3944 --add-label <label>

[Copilot]

Pull request overview

This PR updates the Gluon paged-attention PS-reduce path to reduce runtime JIT “build wall” by eagerly compiling FlyDSL partition-count variants and tightening compile-time specialization in the FlyDSL kernel.

Changes:

• Introduces eager precompilation of FlyDSL PS-reduce sibling variants (partition counts 1..8) on first use per config to reduce later mid-run JIT.
• Uses flydsl.expr.const_expr for several kernel conditionals to force compile-time branching, and adjusts one arith.bitcast call to unwrap the source value.
• Refactors PS reduce wrapper selection to try FlyDSL first (when available), then C++ PS-reduce, then Triton.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.