scratchpad
Untracked location for all `working out` code.
- License 许可
- MIT
- 最近更新
- 2026-11-03
- Stars
- ★ 13
- Forks
- 3
01 前言 Front Matter
- name
- scratchpad
- description
- Untracked location for all `working out` code.
03 SKILL.md 正文
该 Skill 的授权仅允许展示来源链接,正文请前往 GitHub 查看。
查看 SKILL.md ↗04 同仓库更多 Skills
sycl-optimization
Intel GPU SYCL kernel optimization toolkit. USE FOR: writing high-performance SYCL kernels on Intel Xe2-HPG GPUs, performance analysis, roofline modeling, memory access optimization, sub-group operations, SLM tiling, occupancy tuning. Delegates to hardware-info (device specs & roofline) and software-info (optimization techniques & benchmarks).
software-info
SYCL software optimization techniques for Intel Xe2-HPG GPUs. USE FOR: memory access patterns (vec_load, coalesced access, prefetch), sub-group operations, SLM tiling, occupancy tuning, kernel launch configuration. Includes benchmarks to measure real throughput for each technique.
register-info
Diagnose register pressure, spill, and GRF allocation for Intel GPU SYCL kernels. USE FOR: checking register spill count, comparing large vs small GRF modes, dumping ISA assembly, analyzing compiler register allocation decisions, fixing register spill performance issues on Xe2-HPG.
reference-use
Explore repositories under /reference/{repository_name}. Prefer over `webfetch` for source code exploration where applicable.
reference-update
Update every git repository under reference/ to its latest upstream commit.
hardware-info
Measure Intel GPU hardware specifications using SYCL. USE FOR: querying GPU device info, memory hierarchy, bandwidth estimates, compute throughput, roofline analysis ridge points, architecture constants for Xe2-HPG. Compiles and runs device_info.cpp to produce a full hardware report.