[Triton/Gluon] spec decode support for gluon kda files

[omuhamma]

Motivation

Newer vllm recipes use spec decode when concurrency is low, enabling and optimizing my spec decode path to atleast beat tritons

Test Plan

Unit test, e2e accuracy, e2e results. Tested on both 1250 and 950

Test Result

UT pass

*accuracy is same as baseline

*measured on m01-21, runs were captured by doing 1 warmup run and then 3 runs averaged together for the results

1250 runtimes:

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.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:gfx1250-ffm-triton	Run the five-shard gfx1250 FFM Triton test suite
ci:triton-300x	Run an additional Triton test job on MI300X in PRs (added automatically when gfx942 configs change); main branch always runs both MI35X and MI300X
ci:triton-355	Run the full Triton test suite on MI35X, not only the tests the change affects
multigpu	Aiter multi-GPU tests on the 8-GPU runner
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 6034 --add-label <label>

One backend per PR:
A PR changes one kernel backend: [Triton/Gluon] (Triton and Gluon count as one), [HIP], [ASM], [CK], [OPUS] or [FlyDSL]. If the title ends up with two backend tags, split the PR -- as stacked pull requests when one part cannot merge without the other.

PR title tags & labels:
Component tags ([Triton/Gluon], [HIP], [CK], [ASM], ...) are added to the PR title and as PR labels automatically from the changed files and re-synced on every push — change-type tags like [fix]/[Perf], op tags like [MLA], and human labels (ci:*) are left untouched. Add the no-auto-title label to stop the title rewrites; labels stay in sync either way.

[Copilot]

Copilot review overview

🟡 Changes recommended

Speculative decode can access undersized convolution state and leave padded gfx950 outputs uninitialized.

Review effort: Balanced
Findings: 2 · 2

Open (4)

• Advance gfx950 padding mask across speculative tokens · New
• Validate history width for speculative decode accesses · New
• Centralize duplicated host-side config selection · New
• Use resolved configs in Triton/Gluon unit tests · New

What changed in this PR

Adds optimized speculative decoding support to the Gluon KDA path for gfx950 and gfx1250.

Changes:

• Implements speculative convolution-window handling and resident state processing.
• Adds gfx950 tuning for fused multi-token workloads.
• Extends correctness tests across speculative token counts and configurations.

File	Description
aiter/​ops/​triton/​attention/​kda.py	Dispatches architecture-specific configuration and enables speculative convolution state.
aiter/​ops/​triton/​_gluon_kernels/​gfx950/​attention/​kda_decode.py	Adds register-resident speculative decode execution.
aiter/​ops/​triton/​_gluon_kernels/​gfx1250/​attention/​kda_decode.py	Adds speculative convolution-window rolling.
aiter/​ops/​triton/​configs/​gfx950/​gluon/​attention/​kda_decode/​DEFAULT.json	Tunes multi-token and fused launches.
op_tests/​triton_tests/​attention/​test_kda.py	Tests fused speculative decoding and state updates.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.