64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
193 Commits 49 Branches 0 Tags
d46418f1c32857327a5c9bae3e1ca87e2a982476
Commit Graph

28 Commits

Author SHA1 Message Date
CGH0S7
39450228f5 Accelerate Shell-Patch interpolation fast paths 2026-05-08 13:26:16 +08:00
CGH0S7
063f28b3b4 Add Shell-Patch GPU runtime fast paths 2026-05-08 09:26:36 +08:00
CGH0S7
1064a68d16 Optimize BSSN-EM 8th-order AMR transfers 2026-05-07 21:38:16 +08:00
CGH0S7
dcc83bafcb Support 2nd and 8th order CUDA AMR paths 2026-05-07 20:31:26 +08:00
CGH0S7
c4d8d41b25 Cover Z4C CUDA AMR restrict prolong 2026-05-07 19:49:09 +08:00
CGH0S7
9ff2f065be Apply BSSN AMR sync default to EScalar 2026-05-07 17:12:33 +08:00
CGH0S7
2317e4abde Fix BSSN GPU resident AMR sync default 2026-05-07 17:11:09 +08:00
CGH0S7
96829d0441 Optimize Z4C GPU runtime defaults 2026-05-07 15:37:09 +08:00
CGH0S7
83afaf19ce Skip zero EM resident downloads 2026-05-07 13:04:46 +08:00
CGH0S7
cb911dec06 Add EM GPU fast paths and defaults 2026-05-07 12:18:56 +08:00
CGH0S7
ffa0d801ed Default Python GPU runner to EScalar fast path 2026-05-06 00:12:46 +08:00
CGH0S7
85fe29cc2e Optimize BSSN-EScalar CUDA path 2026-05-05 10:47:46 +08:00
ianchb
06f62dee36 Switch back to Intel toolchain as the default option 
Seems that Intel MPI also supports CUDA-aware by setting I_MPI_OFFLOAD to 1. Besides, I_MPI_OFFLOAD_IPC=0 is needed to avoid segfaults.
 2026-05-01 21:59:13 +08:00
CGH0S7
a9a3809148 Default Python launcher to fast GPU path 2026-04-30 20:15:34 +08:00
CGH0S7
e0d0673c8e Enable optimized GPU runs from Python launcher 2026-04-30 18:31:31 +08:00
ianchb
c689cc8dc9 [WIP] Add CUDA support for Z4C 
Rewritten done by Codex.
This still has errors, do not pick this one now.
 2026-04-27 11:58:43 +08:00
ianchb
53c55451b3 Update makefile and scripts for CUDA BSSN configuration and build commands 2026-04-25 09:19:50 +08:00
ianchb
86a683de26 Replace legacy ABEGPU stack with ABE_CUDA backend 2026-04-12 21:19:14 +08:00
CGH0S7
44efb2e08c 预赛最终版本v1.0.0: 确定PGO和原负载均衡方案在当前版本造成负优化已经回退 2026-03-01 18:04:25 +08:00
CGH0S7
1eba73acbe 先关闭绑核心，发现速度对比：不绑定核心+SCX>绑核心+SCX 2026-02-28 23:27:44 +08:00
CGH0S7
e0b5e012df 引入 PGO 式两遍编译流程，将 Interp_Points 负载均衡优化合法化 
背景：
上一个 commit 中同事实现的热点 block 拆分与 rank 重映射取得了显著
加速效果，但其中硬编码了 heavy ranks (27/28/35/36) 和重映射表，
属于针对特定测例的优化，违反竞赛规则第 6 条（不允许针对参数或测例
的专门优化）。

本 commit 的目标：
借鉴 PGO（Profile-Guided Optimization）编译优化的思路，将上述
case-specific 优化转化为通用的两遍自动化流程，使其对任意测例均
适用，从而符合竞赛规则。

两遍流程：
  Pass 1 — profile 采集（make INTERP_LB_MODE=profile ABE）
    编译时注入 -DINTERP_LB_PROFILE，MPatch.C 中 Interp_Points
    在首次调用时用 MPI_Wtime 计时 + MPI_Gather 汇总各 rank 耗时，
    识别超过均值 2.5 倍的热点 rank，写入 interp_lb_profile.bin。

  中间步骤 — 生成编译时头文件
    python3 gen_interp_lb_header.py 读取 profile.bin，自动计算
    拆分策略和重映射表，生成 interp_lb_profile_data.h，包含：
    - interp_lb_splits[][3]：每个热点 block 的 (block_id, r_left, r_right)
    - interp_lb_remaps[][2]：被挤占邻居 block 的 rank 重映射

  Pass 2 — 优化编译（make INTERP_LB_MODE=optimize ABE）
    编译时注入 -DINTERP_LB_OPTIMIZE，profile 数据以 static const
    数组形式固化进可执行文件（零运行时开销），distribute_optimize
    在 block 创建阶段直接应用拆分和重映射。

具体改动：
- makefile.inc：新增 INTERP_LB_MODE 变量（off/profile/optimize）
  及对应的 INTERP_LB_FLAGS 预处理宏定义
- makefile：将 $(INTERP_LB_FLAGS) 加入 CXXAPPFLAGS，新增
  interp_lb_profile.o 编译目标
- gen_interp_lb_header.py：profile.bin → interp_lb_profile_data.h
  的自动转换脚本
- interp_lb_profile_data.h：自动生成的编译时常量头文件
- interp_lb_profile.bin：profile 采集阶段生成的二进制数据
- AMSS_NCKU_Program.py：构建时自动拷贝 profile.bin 到运行目录
- makefile_and_run.py：默认构建命令切换为 INTERP_LB_MODE=optimize

通用性说明：
整个流程不依赖任何硬编码的 rank 编号或测例参数。对于不同的网格
配置、进程数或物理问题，只需重新执行 Pass 1 采集 profile，即可
自动生成对应的优化方案。这与 PGO 编译优化的理念完全一致——先
profile 再优化，是一种通用的性能优化方法论。
 2026-02-27 15:10:22 +08:00
CGH0S7
e2bc472845 优化绑核逻辑，取消硬编码改为智能识别 2026-02-25 10:59:32 +08:00
CGH0S7
4bb6c03013 makefile setting updated 2026-02-08 16:14:43 +08:00
ianchb
03d501db04 Display the runtime of TwoPunctures 2026-02-07 14:45:16 +08:00
CGH0S7
223ec17a54 input updated 2026-02-06 13:57:48 +08:00
CGH0S7
3a7bce3af2 Update Intel oneAPI configuration and CPU binding settings 
- Update makefile.inc with Intel oneAPI compiler flags and oneMKL linking
   - Configure taskset CPU binding to use nohz_full cores (4-55, 60-111)
   - Set build parallelism to 104 jobs for faster compilation
   - Update MPI process count to 48 in input configuration
 2026-01-17 20:41:02 +08:00
CGH0S7
7a76cbaafd Add numactl CPU binding to avoid cores 0-3 and 56-59 
Bind all computation processes (ABE, ABEGPU, TwoPunctureABE) to
   CPU cores 4-55 and 60-111 using numactl --physcpubind to prevent
   interference with system processes on reserved cores.
 2026-01-16 10:24:46 +08:00
CGH0S7
f2fc9af70e asc26 amss-ncku initialized 2026-01-13 15:01:15 +08:00