feat(flydsl): Add HSTU Forward kernel

[damien-lejeune]

Motivation

The purpose of this work is to implement a FlyDSL HSTU forward kernel.

Jira ID : SILOTIGER-617

Technical Details

The HSTU block is defined here: https://arxiv.org/pdf/2402.17152v2
The full HSTU layer, implemented here (paper Eq. 1–3) is:

$$U, V, Q, K = \text{Split}\big(\phi_1(f_1(X))\big)$$

$$A(X),V(X) = \phi_2!\big(Q K^\top + \text{rab}^{p,t}\big),V$$

$$Y(X) = f_2\big(\text{Norm}(A(X)V(X)) \odot U(X)\big)$$

where $f_1, f_2$ are linear layers, $\phi_1,\phi_2$ are SiLU, $\odot$ is elementwise product, and $\text{rab}^{p,t}$ is a relative attention bias (positional + temporal).

Test Plan

• Unit tests were written validating correctness on the input shapes of interest. The tests check the FlyDSL kernel against a PyTorch oracle across a handful of shapes and each feature (causal, targets, sliding window, contextual prefix, and various head/hidden dims). They also cover input validation (rejecting bad device/shape/rank/dtype/target-length) and tuned-CSV config loading.
• Tested support and performance on gfx942. Performance was externally validated to match requirements.
• Tested correctness on gfx950

Test Result

• All unit tests are (and should be) passing

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 4354 --add-label <label>