Replace block_ptr in HSTU attn kernel

[scxiao]

Motivation

block pointer was deprecated from Triton 3.8, this PR is to replace the block pointer in the HSTU attention kernel with normal pointers.

Technical Details

None

Test Plan

Existing correctness test can pass and no performance regression

Test Result

Correctness test passed, saw perf uplift with some tuning parameter changes, see the following table:

  ┌────────────┬─────────────┬──────────┬───────┬──────────┬────────────┬─────────────┬────────────┬────────────┐ 
  │ batch_size │ max_seq_len │ sparsity │ heads │ attn_dim │ hidden_dim │ Triton 3.7  │ Triton 3.7 │ Triton 3.8 │
  │            │             │          │       │          │            │(noPR)TFLOPS │ (PR)TFLOPS │ (PR)TFLOPS │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │         32 │        1024 │    0.500 │     4 │      128 │        128 │     130.309 │    156.241 │    160.427 │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤
  │         32 │        2048 │    0.500 │     4 │      128 │        128 │     218.606 │    252.746 │    261.425 │ 
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤
  │         32 │        4096 │    0.500 │     4 │      128 │        128 │     344.016 │    382.364 │    393.997 │ 
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │         32 │        8192 │    0.500 │     4 │      128 │        128 │     427.342 │    462.575 │    478.822 │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │         32 │       16384 │    0.500 │     4 │      128 │        128 │     450.008 │    493.805 │    512.873 │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │        512 │         512 │    0.970 │     4 │      128 │        128 │     200.882 │    253.690 │    252.419 │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │        512 │        3072 │    0.366 │     4 │      128 │        128 │     393.988 │    450.191 │    447.854 │
  ├────────────┼─────────────┼──────────┼───────┼──────────┼────────────┼─────────────┼────────────┼────────────┤ 
  │       1024 │        1024 │    0.768 │     4 │      128 │        128 │     200.937 │    258.980 │    257.411 │
  └────────────┴─────────────┴──────────┴───────┴──────────┴────────────┴─────────────┴────────────┴────────────┘ 

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: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; 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 4736 --add-label <label>