[OP][Optimization] Remove ENABLE_PREFILL template parameter in multi_query_append_attention_warp1_4_kernel

[zhoutianzi666]

Motivation

💡 If this PR is a Cherry Pick, the PR title needs to follow the format by adding the [Cherry-Pick] label at the very beginning and appending the original PR ID at the end. For example, [Cherry-Pick][CI] Add check trigger and logic(#5191)

💡 如若此PR是Cherry Pick，PR标题需遵循格式，在最开始加上[Cherry-Pick]标签，以及最后面加上原PR ID，例如[Cherry-Pick][CI] Add check trigger and logic(#5191)

Modifications

Usage or Command

Accuracy Tests

Checklist

• Add at least a tag in the PR title.
  • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
  • You can add new tags based on the PR content, but the semantics must be clear.
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 66.66667% with 2 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@da3dfe1). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...el_executor/layers/attention/flash_attn_backend.py	50.00%	0 Missing and 1 partial ⚠️
...ecutor/layers/attention/flash_mask_attn_backend.py	50.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #7201   +/-   ##
==========================================
  Coverage           ?   73.84%           
==========================================
  Files              ?      376           
  Lines              ?    52947           
  Branches           ?     8263           
==========================================
  Hits               ?    39101           
  Misses             ?    11111           
  Partials           ?     2735           

Flag	Coverage Δ
GPU	73.84% <66.66%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[fastdeploy-bot]

🤖 AI Code Review | 2026-04-07 10:33 CST

📋 Review 摘要

PR 概述：删除 ENABLE_PREFILL 模板参数，统一 attention kernel 的内存布局
变更范围：custom_ops/gpu_ops/append_attn/、fastdeploy/model_executor/layers/attention/
影响面 Tag：OP Speculative Decoding

📝 PR 规范检查

PR 标题和描述不符合规范：

1. 标题缺少有效的 Tag 标签
2. 描述中 Motivation 和 Modifications 部分未填写

标题建议（可直接复制）：

[OP][Optimization] Remove ENABLE_PREFILL template parameter in attention kernel

描述建议：

## Motivation
简化 attention kernel 代码，删除不再需要的 `ENABLE_PREFILL` 模板参数分支，统一使用 `speculate_max_draft_token_num` 的内存布局。

## Modifications
1. 删除 `multiquery_attention_c16_impl.cuh` 中 `multi_query_append_attention_warp1_4_kernel` 的 `ENABLE_PREFILL` 模板参数
2. 简化内存分配逻辑，统一使用 `speculate_max_draft_token_num * bsz` 的计算方式
3. Python 端（append_attn_backend.py, flash_attn_backend.py, flash_mask_attn_backend.py）在非 speculative 模式下将 `speculate_max_draft_token_num` 设为 0

问题

级别	文件	概述
❓ 疑问	multiquery_attention_c16_impl.cuh:433	c4/c8 版本未同步删除 ENABLE_PREFILL，是否有意为之？
🟡 建议	append_attn_backend.py:150	建议添加注释说明与 CUDA 端的关联

总体评价

代码逻辑正确：Python 端设置 speculate_max_draft_token_num=0（传给 CUDA 时 +1=1），与删除 ENABLE_PREFILL 后的统一内存布局兼容。建议补充 PR 规范信息，并确认 c4/c8 版本是否需要同步修改。

[fastdeploy-bot]

❓ 疑问 只有 multi_query_append_attention_warp1_4_kernel 删除了 ENABLE_PREFILL 模板参数

注意到 multiquery_attention_c4_impl.cuh 和 multiquery_attention_c8_impl.cuh 中仍然保留了 ENABLE_PREFILL 参数。这是否是有意为之（分阶段重构），还是遗漏？

如果是有意为之，建议在 PR 描述中说明原因。

[zhoutianzi666]

分阶段重构

[fastdeploy-bot]

🟡 建议 建议添加注释说明此设置的目的

这个修改确保非 speculative 模式下 speculate_max_draft_token_num=0，与 CUDA 端删除 ENABLE_PREFILL 后的统一内存布局配合。建议添加注释说明这个关联，方便后续维护。

# When not using speculative decoding, set to 0. The CUDA kernel will
# receive (speculate_max_draft_token_num + 1) = 1, which matches the
# simplified memory layout after removing ENABLE_PREFILL branches.
if self.speculative_method is None:
    self.speculate_max_draft_token_num = 0

[chang-wenbin]

LGTM