Skip to content

Optimization document2nd for the first game - #117

Open
FeelTheBeats wants to merge 3 commits into
ScratchV-Compiler:mainfrom
FeelTheBeats:bisai_op
Open

FeelTheBeats wants to merge 3 commits into
ScratchV-Compiler:mainfrom
FeelTheBeats:bisai_op

Conversation

@FeelTheBeats

@FeelTheBeats FeelTheBeats commented Oct 24, 2026 •

Copy link
Copy Markdown
Contributor

for both agents and students

FeelTheBeats and others added 2 commits October 21, 2026 21:27
self-hosted 池没有在线 runner,main 的 16 次 push 运行全部排队到 24 小时
被取消,CI acceptance 从未真正执行过。统一到 ubuntu-24.04 后,PR 与 main
共用同一台机器、同一个 Python 3.12 和同一种取代码方式。

- test/benchmark 的 runs-on 改为 ubuntu-24.04,超时 20/30 分钟改为 45
- 删除两段 /opt/ScratchV 镜像同步(rm -rf 工作区后 cp 镜像)
- checkout 与 setup-python 不再按事件分流,push 也走 actions/checkout
- 新增 pip 缓存,避免每次运行冷装 torch 等依赖
- 工具链注释改为托管机可 sudo 安装,保留 out-of-band 兜底

验证:tests/test_ci_workflows.py + tests/test_ci_routing.py(161 passed)
@github-actions

github-actions Bot commented Oct 24, 2026 •

Copy link
Copy Markdown

🤖 AI Code Review

Head: b80caf6
PR files: 1; fetched: 1; reviewed: 1; failed/missing: 0; unreviewable: 0.
Coverage: complete. A completed AI response is not approval or proof of correctness.

Full per-file results and batch artifacts

Every fetched file has a review response.

@FeelTheBeats FeelTheBeats changed the title Bisai op Optimization document2nd for the first game Oct 24, 2026
@FeelTheBeats
FeelTheBeats force-pushed the bisai_op branch 3 times, most recently from c0b90c0 to 67a0ff5 Compare October 25, 2026 05:14
DEVELOPMENT.md(+402/-19):补齐学生教程里“离榜首还差多少、差在哪一项”
这一层,并修正与仓库/实测不符的地方:

- 0.4 重写“热循环对齐”:实测带 .option norvc 时 .text 里的 .balign 只填
  4 字节 nop(14 条且全部执行),不产生 R_RISCV_ALIGN;真判 0 分的是
  .skip/.2byte
- 4.0 新增地板一节:三题公式 m+(2mr+nr+1)/(mr·nr)、按分块的地板表、
  指令比与 miss 占比两个判据;4.5 的旧公式改为指向此处
- 2.0 新增产出路径(R1-R5 + X0)与 -O3 差 1.9~3.7 倍的三条原因
- 5.4 新增 FMA:4×4 下 2.81 -> 1.81 条/MAC,只讲原理不贴代码
  (仓库 dtypes.py 仍是 fmul.s + fadd.s)
- 6.3 补分块尺寸预算表(4×4 需 25 个寄存器等);地板推导 2.5 -> 2.81,
  与内循环 2.94、整题 2.98 对账
- 6.4 补相位模型(解释 d_miss 773 vs 768)与数据段对齐写法
- 6.5 新增 i_miss 一节:合并内核体、冷热分离、行数下界、总账判据
- 8.1 三关改为可执行门禁清单(3 条可粘贴命令);3.5 代码与仓库对齐;
  术语表补 FMA/地板/相位/R_RISCV_ALIGN

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant