赛题 #37: ReduceSum, SortPair, TopkPair 算法优化 #8

xingjian · 2025-12-03T15:08:35+08:00

xingjian commented

2025-12-03 15:08:35 +08:00

First-time contributor

赛题 #37: ReduceSum, SortPair, TopkPair 算法优化

性能提升 (对比 Thrust 基准)

算法	1M 数据	128M 数据	512M 数据	1G 数据
ReduceSum	+73%	+16%	+3%	+1%
SortPair	+68%	+9%	+9%	+9%
TopkPair	+75%	+11%	+11%	+11%

优化技术

ReduceSum:
- 自定义 kernel：Warp shuffle reduction
- 向量化加载 (float4)
- 连续内存访问模式
SortPair / TopkPair:
- 使用 mccub DeviceRadixSort (MetaX CUB library)
- 优化的临时缓冲区管理

测试环境

GPU: MetaX C500 (64GB)
MACA: 3.0.0.8
所有正确性测试通过 ✅

关联 Issue

Closes #37

加分项说明

✅ 使用 LLM 辅助生成代码
✅ 代码规范、清晰
✅ 性能优化明显

## 赛题 #37: ReduceSum, SortPair, TopkPair 算法优化 ### 性能提升 (对比 Thrust 基准) | 算法 | 1M 数据 | 128M 数据 | 512M 数据 | 1G 数据 | |------|--------|----------|----------|--------| | **ReduceSum** | +73% | +16% | +3% | +1% | | **SortPair** | +68% | +9% | +9% | +9% | | **TopkPair** | +75% | +11% | +11% | +11% | ### 优化技术 1. **ReduceSum**: - 自定义 kernel：Warp shuffle reduction - 向量化加载 (float4) - 连续内存访问模式 2. **SortPair / TopkPair**: - 使用 mccub DeviceRadixSort (MetaX CUB library) - 优化的临时缓冲区管理 ### 测试环境 - GPU: MetaX C500 (64GB) - MACA: 3.0.0.8 - 所有正确性测试通过 ✅ ### 关联 Issue Closes #37 ### 加分项说明 - ✅ 使用 LLM 辅助生成代码 - ✅ 代码规范、清晰 - ✅ 性能优化明显

xingjian added 4 commits 2025-12-03 15:08:37 +08:00

b1355c19d9 feat(S1/37): Optimize ReduceSum, SortPair, TopkPair algorithms

Performance improvements (vs Thrust baseline):
- ReduceSum: +73% (1M), +16% (128M), +3% (512M), +1% (1G)
- SortPair: +68% (1M), +9% (128M-1G)
- TopkPair: +75% (1M), +11% (128M-1G)

Optimization techniques:
- ReduceSum: Custom kernel with warp shuffle + vectorized float4 load + contiguous memory access
- SortPair: mccub DeviceRadixSort (MetaX CUB library)
- TopkPair: mccub DeviceRadixSort with optimized buffer management

All tests passed on MetaX C500 GPU (MACA 3.0.0.8)

Issue: #37

a087148618 docs(S1/37): Add optimization report with LLM prompts

- Detailed performance analysis for all three algorithms
- Hardware-specific optimization strategies documented
- LLM (Claude) prompts recorded for bonus points
- Experience summary and lessons learned

6aecbb980e docs(S1/37): Update report with reusable LLM prompt templates

- Added 5 general-purpose prompt templates for GPU optimization
- Included prompt usage tips and best practices
- Removed verbose process descriptions
- Templates can be reused for other GPU optimization tasks

ffcd1fadad docs(S1/37): Separate LLM prompt templates into standalone file

- Created LLM_PROMPT_TEMPLATES.md with 5 reusable prompt templates
- Updated OPTIMIZATION_REPORT.md to reference the templates file
- Keep optimization strategies in main report

xingjian added 1 commit 2025-12-03 15:23:17 +08:00

659b02bbe8 docs(S1/37): Remove unused section and clarify prompt usage

- Deleted the unused section on using Claude for assistance
- Streamlined the document for better clarity and focus on prompt templates

This pull request can be merged automatically.

This branch is out-of-date with the base branch

You are not authorized to merge this pull request.

You can also view command line instructions.

Step 1:

From your project repository, check out a new branch and test the changes.

git checkout -b xingjian-main main

git pull main

Step 2:

Merge the changes and update on Gitea.

git checkout main

git merge --no-ff xingjian-main

git push origin main

Sign in to join this conversation.

No reviewers

No Label

No Milestone

No project

No Assignees

1 Participants

Notifications

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: ccf-ai-infra/GPUKernelContest#8