research: preserve solver experiments and reproducible benchmarks
This commit is contained in:
parent
12e4266950
commit
e93bc20b0a
85 changed files with 518443 additions and 0 deletions
76
research/experiments/normal_search/PIPELINE_REPORT.md
Normal file
76
research/experiments/normal_search/PIPELINE_REPORT.md
Normal file
|
|
@ -0,0 +1,76 @@
|
|||
# 普通搜索：坐标类复用及只展开有根类
|
||||
|
||||
2026-10-11（Asia/Hong_Kong）；RTX 3060 Laptop GPU。接续 [16 候选成组筛选实验](REPORT.md)。补扫未修改、未运行。
|
||||
|
||||
结论：复用已接入可执行实验流程，计入分组、根生成、候选展开、必要传输、搜索和结果回传后仍有效。收益取决于同批坐标类重复率及有根类比例，不是统一倍数。对几乎无重复类的散点，直接成组筛选更快。
|
||||
|
||||
## 四条路径
|
||||
|
||||
- 模式 0：每个位置生成根，旧差过滤＋区间求交。
|
||||
- 模式 1：每个位置生成根，上一轮 16 候选成组筛选。
|
||||
- 模式 2：坐标按 `(x mod 65536,z mod 131072)` 分类，每类生成一次根，再在 GPU 上展开任务，成组筛选。
|
||||
- 模式 3：先分类、只构造代表位置；筛出类根后，仅构造和上传有根类的具体位置，再展开任务，成组筛选。
|
||||
|
||||
四条路径都重算本轮的类根，没有拿前一轮缓存假装免费。模式 2/3 使用各类排序后的完整根列表和位置偏移，在 GPU 上生成 `(具体位置,root)` 任务。模式 3 压缩位置后，将结果还原到原始输入索引。
|
||||
|
||||
程序检查根缓冲区和命中缓冲区容量，超限明确失败，不截断。单批支持最多 65,536 个位置、4,194,304 个展开后的根任务；较大任务仍需外层分批。当前是批次级调度实验，不是已经具备全世界断点续跑能力的正式扫描器。
|
||||
|
||||
## 实测批次耗时
|
||||
|
||||
预热一次、三轮轮换模式顺序。下表为 CPU 单调时钟测量的批次耗时中位数，单位 ms。
|
||||
|
||||
| 输入 | 位置数 / 类数 | 模式 0 | 模式 1 | 模式 2 | 模式 3 |
|
||||
|---|---:|---:|---:|---:|---:|
|
||||
| 已知位置所属类的全部世界内位置 | 1653 / 1 | 364.023 | 62.016 | 55.373 | 56.867 |
|
||||
| 32 个有根类的全部世界内位置 | 52213 / 32 | 4691.311 | 915.446 | 656.710 | 676.742 |
|
||||
| 512 个随机类，每类抽 64 个世界内位置 | 32768 / 512 | 172.222 | 141.138 | 14.235 | 9.586 |
|
||||
| 4096 个均匀随机散点 | 4096 / 4096 | 24.022 | 17.902 | 19.239 | 19.502 |
|
||||
|
||||
**计入**：本批次位置项构造、分类、代表类筛根、类根回传/整理、偏移构造、必要 H2D 传输、GPU 根任务展开、每根公共项预计算、搜索、命中回传及排序、模式 3 结果索引还原。
|
||||
|
||||
**不计入**：输入文件读取、一次性 CUDA 初始化/工作区申请、专用于对照验证的完整任务回传和排序、日志打印。一次性准备另记为 `setup_ms`，这四批约 217～242 ms。预热复用了显存；新增预计算仍每轮执行。这里是长期运行时可摊薄初始化后的批次耗时，不是启动一次命令的进程总耗时。
|
||||
|
||||
本轮直接使用 host 时间，不与前一轮的 CUDA event 时间混算。
|
||||
|
||||
## 收益在哪里
|
||||
|
||||
以随机 512 类批次为例：
|
||||
|
||||
- 原先要为 32,768 个位置重复筛根；现在仅筛 512 个代表。
|
||||
- 其中 19 个类有根，只有 1,216 个具体位置需要构造坐标项并传入后续搜索。
|
||||
- 展开后的完整任务仍是相同的 15,424 个 `(位置,root)`，没有省略任何普通分支候选。
|
||||
- 模式 3 的阶段中位数：初始分类/代表构造约 0.416 ms，类根生成约 2.049 ms，整理/展开约 0.690 ms，搜索及回传约 6.775 ms。各阶段中位数不必精确相加等于总耗时中位数。
|
||||
- 相比上一轮成组筛选流程，本批再加速 14.72×；相对模式 0 为 17.97×。
|
||||
|
||||
32 个有根类的压力批次有 1,693,228 个根任务，全部类都必须展开，计算主要花在后续搜索。模式 3 相对模式 1 仅加速 1.35×；模式 2 略优，因为无需压缩位置和还原索引。
|
||||
|
||||
随机散点没有类复用收益，分类、回传和整理增加开销；模式 3 比模式 1 多耗时约 8.9%。因此应按类组织大范围遍历；不能对任意散点输入无条件启用该流程。
|
||||
|
||||
## 正确性和边界检查
|
||||
|
||||
每一轮、每一种模式均将**全部展开根任务列表**排序后与模式 0 逐项比较，并比较完整命中列表，不只是比较数量。
|
||||
|
||||
- 四批完整根列表和命中列表全部一致。
|
||||
- 已知类批次保留原有 24 个命中，其他三批无命中。
|
||||
- 全无根类的全部世界内位置：四条路径都生成 0 个任务、0 个命中；模式 3 正确处理没有具体位置可展开的情况。
|
||||
- 原 21 个压力/边界位置：覆盖原点、世界边界和 int32 边界，四条路径相同的 24 个命中。
|
||||
- 随机打乱 32,768 个混合位置的输入顺序：完整根和结果列表依然一致，位置压缩/还原不依赖同类位置连续排列。
|
||||
|
||||
原始记录在 `pipeline_results/*.jsonl`，汇总、源码哈希在 `pipeline_results/summary.json`；补充检查在 `pipeline_results/edge_validation.json`。
|
||||
|
||||
## 对全世界估算的影响
|
||||
|
||||
本轮证明此前“复用低位根”的前提可以落地，且筛类后只展开有根位置能控制部分开销。但不能把本轮 14.72× 再乘到此前约 50 天的估算上：此前约 50 天已经假定根充分复用，主要计算的是后续搜索。
|
||||
|
||||
本轮仅抽取 512 个随机类，根分布样本不足以重新估计全世界平均耗时；32 个有根类更是有意选择的压力集。维持此前约 50 天作为普通搜索阶段的乐观吞吐量级，不将其升级为完整全世界工期承诺。
|
||||
|
||||
下一步若继续研究算法，应优先减少剩余中间位组的筛选工作，或研究不同实际位置之间能否共享更高位的必要条件。当前低位根复用只消除了根生成重复，不能推导出高位结果也相同。
|
||||
|
||||
## 复现
|
||||
|
||||
```
|
||||
python3 research/experiments/normal_search/run_pipeline.py
|
||||
python3 research/experiments/normal_search/verify_pipeline_edges.py
|
||||
```
|
||||
|
||||
两个脚本都会编译 GPU 基准程序。需要 NVIDIA GPU、nvcc 和 g++ 环境。实现入口为 `pipeline_bench.cu`，复用上一轮 `cuda_bench.cu` 中的内核。原正式入口未替换。
|
||||
87
research/experiments/normal_search/REPORT.md
Normal file
87
research/experiments/normal_search/REPORT.md
Normal file
|
|
@ -0,0 +1,87 @@
|
|||
# 普通搜索：16 个中间位候选成组筛选
|
||||
|
||||
2026-10-11；RTX 3060 Laptop GPU；CUDA sm_86。仅研究普通首次输出分支，未修改或测试补扫。
|
||||
|
||||
结论：保留公共项预计算和 16 候选成组筛选。相对上一轮最优的“输出差过滤＋区间求交”，两个独立的 65,536 随机坐标批次，搜索阶段分别加速 6.51×、6.56×。本轮代码在独立实验入口，未替换正式程序。
|
||||
|
||||
## 同任务实测
|
||||
|
||||
预热后轮换模式顺序，七轮中位数。模式 4～7 每轮重新执行 GPU 预计算，耗时已包括预计算 kernel。所有模式使用相同完整低位根列表，完整输出列表逐项比较。
|
||||
|
||||
| 输入 | 旧最优（模式 3） | 32 位预计算（4） | 成组筛选（7） | 3/7 |
|
||||
|---|---:|---:|---:|---:|
|
||||
| 已知命中位置 | 0.323584 ms | 0.268288 ms | 0.156672 ms | 2.07× |
|
||||
| 21 压力/边界位置 | 1.288192 ms | 0.926720 ms | 0.223232 ms | 5.77× |
|
||||
| 4096 随机位置 | 8.857600 ms | 5.859328 ms | 1.141760 ms | 7.76× |
|
||||
| 65536 随机位置，原批次 | 130.136063 ms | 98.654205 ms | 19.982336 ms | 6.51× |
|
||||
| 65536 随机位置，独立新批次 | 134.184967 ms | 102.582275 ms | 20.440063 ms | 6.56× |
|
||||
|
||||
另外测试了半差模 5（模式 5），以及半差模 5 配合角落优先检查（模式 6）。均未优于模式 7，不作为最终方案。原批次模式 5/6 分别为 114.548737/108.295166 ms。
|
||||
|
||||
这是搜索阶段的 CUDA event 时间，**不包括低位根生成、初始化、显存分配、排序及传输**。同次运行原批次根生成 kernel 为 267.922424 ms；若每个坐标仍单独生成根，则根生成加搜索的 kernel 时间之和从 398.058487 ms 降到 287.904760 ms，仅 1.38×。这个和是阶段计时相加，不是完整应用耗时。
|
||||
|
||||
此前已经验证坐标类之间可以复用低位根，但本实验没有实现全世界分组调度；不可把 6.5× 写成现有完整程序的加速倍数。
|
||||
|
||||
## 公共项预计算
|
||||
|
||||
使用归一化种子，令首格坐标项为 0。固定低 18 位 `root`，中间 17 位记为 `m`：
|
||||
|
||||
```
|
||||
a_i = root + term_i
|
||||
h_i = a_i >> 18
|
||||
l_i = (a_i XOR C) & (2^18 - 1)
|
||||
b_i = (A*l_i + 11) >> 17
|
||||
c = C >> 18 = 0x1790d
|
||||
q_i(m) = [b_i + 2*A*((m+h_i) XOR c)] mod 2^31
|
||||
```
|
||||
|
||||
`A=0x5DEECE66D`，`C=0x5E434E432`。`h_i,b_i` 对同一个根的全部 131,072 个 m 共用，每根准备一次。最后公式按 uint32 运算取模即可，无需每个候选重复 64 位乘法。
|
||||
|
||||
## 为什么可以 16 个一起筛
|
||||
|
||||
`c=0x1790d` 的第 4～7 位为 0（最低位编号为 0）。将全部 m 唯一划分为：
|
||||
|
||||
```
|
||||
base = (g >> 4)*256 + (g & 15), g=0..8191
|
||||
m = base + 16*j, j=0..15
|
||||
```
|
||||
|
||||
每组仅变化 m 的第 4～7 位：
|
||||
|
||||
- 首格 term 为 0，因此 `(m XOR c)` 随 j 线性增加，不会跨 256 边界。
|
||||
- 对任意其他格，`m+h_i` 在这段长度 240 的范围内最多跨一次 256 边界。
|
||||
- 同一段内，两格的输出都有相同线性增量。因此 `(q_i-q_0) mod 2^31` 是常数；整组最多两段。
|
||||
- 只需计算首尾两处差值，并由进位位置求分界，即可准确判断 16 个候选哪些满足 `delta % 10 ∈ {0,8}`。
|
||||
|
||||
用 16 位 mask 表示一组的候选，逐格与允许 mask 求交，变成 0 就整组退出。最终剩余的具体 m 才进入原有高位区间求交及精确 Java 验证。
|
||||
|
||||
每根筛选线程从 131,072 个减少到 8,192 个。仍覆盖所有 m，每个线程所做的工作有所增加，因此实际收益不是理论 16×。这也没有消除中间位维度，只是利用局部等价批量处理。
|
||||
|
||||
## 正确性验证
|
||||
|
||||
- 5,000,000 次随机完整输出对照：预计算公式和原 48 位状态公式一致。
|
||||
- 200,000 个随机组，约束数量循环覆盖 1～25：成组 mask 和逐个检查 16 个候选的差过滤完全一致。包括 159,120 个通过的候选，避免只对照空集。
|
||||
- 已知位置全部 84 个低位根 × 8192 组，与原逐候选差过滤完全一致，共覆盖 11,010,048 个中间位候选，保留 36 个差过滤候选。
|
||||
- 以上共验证 888,128 个完整组 mask。
|
||||
- 五组 GPU 基准，全部模式预热与七轮测量均逐项比较完整结果列表；已知位置与压力批次保留原有 24 个命中；随机批次均无命中。
|
||||
- 每批最多 64 个坐标的完整低位根列表与 CPU 对照；压力集包括世界边界、int32 边界及原点。
|
||||
|
||||
普通分支原有的拒绝重抽覆盖范围未改变。本实验不声称单独覆盖所有 Java 重试解。
|
||||
|
||||
## 全世界量级与下一步
|
||||
|
||||
若沿用之前的乐观假设：低位根充分复用、初始化充分摊薄，只用本次搜索阶段吞吐外推 `3,750,000²` 个锚点，则两批分别约 **49.6 天、50.8 天**，相比之前约 0.9 年有明显改善。这不是完整全世界运行时间，不含全世界分组、根缓存展开及传输，也不含补扫；不能据此承诺两个月完成。
|
||||
|
||||
接下来普通流程优先项：实际接上低位坐标类复用调度，计入分组、候选展开、传输的总成本。随后再研究是否能合并更多中间位组，而不是继续仅压缩高位求解的少量幸存者。
|
||||
|
||||
## 复现
|
||||
|
||||
```
|
||||
python3 research/experiments/normal_search/run.py
|
||||
```
|
||||
|
||||
需要可访问 NVIDIA GPU、nvcc、g++。脚本编译程序、运行 CPU 验证、生成独立随机坐标、顺序运行五组 GPU 基准，并写入 `results/`。不会调用补扫 kernel。
|
||||
|
||||
实现：`grouped.hpp`（预计算和组 mask）、`cuda_bench.cu`（基准与分组 kernel）、`cpu_verify.cpp`（独立对照）。原始七轮数据在 `results/*.jsonl`，汇总与源码哈希在 `results/summary.json`。
|
||||
|
||||
后续已完成计入分组、展开及传输的坐标类复用实验，见 [完整批次流程报告](PIPELINE_REPORT.md)。
|
||||
51
research/experiments/normal_search/cpu_verify.cpp
Normal file
51
research/experiments/normal_search/cpu_verify.cpp
Normal file
|
|
@ -0,0 +1,51 @@
|
|||
#include "grouped.hpp"
|
||||
#include <random>
|
||||
#include <iostream>
|
||||
#include <stdexcept>
|
||||
namespace sr=slime_research;
|
||||
namespace ns=normal_search;
|
||||
int main(){
|
||||
std::mt19937_64 rng(2026101103);
|
||||
uint64_t qchecks=0,groups=0,partialSurvivors=0;
|
||||
for(int test=0;test<200000;++test){
|
||||
sr::CellTerm cells[25];
|
||||
int32_t x=int32_t(rng()%3750000)-1875000,z=int32_t(rng()%3750000)-1875000;
|
||||
const uint64_t anchor=sr::coordinateTerm(x,z);
|
||||
for(int i=0;i<25;++i)cells[i]={(sr::coordinateTerm(x+i%5,z+i/5)-anchor)&sr::kMask48,sr::kSlime};
|
||||
uint32_t root=rng()&((1U<<18)-1);
|
||||
// Full-width random middle values also test carry and modulo behavior.
|
||||
uint32_t m=rng()&((1U<<17)-1);
|
||||
ns::Prepared p;ns::prepare(root,cells,p);
|
||||
for(int i=0;i<25;++i){
|
||||
uint32_t expected=sr::nextStateFromTerm(root+(uint64_t(m)<<18),cells[i].term)>>17;
|
||||
if(ns::preparedQ(p,i,m)!=expected)throw std::runtime_error("q mismatch");
|
||||
++qchecks;
|
||||
}
|
||||
// groupMask matches the difference filter even on odd roots; final
|
||||
// solver parity checks remain responsible for discarding those roots.
|
||||
int count=1+test%25;
|
||||
uint32_t base=m&~240U,expected=0;
|
||||
for(uint32_t j=0;j<16;++j)
|
||||
if(full_search_experiment::differenceFilter(root+(uint64_t(base+16*j)<<18),cells,count))expected|=1U<<j;
|
||||
partialSurvivors+=__builtin_popcount(expected);
|
||||
if(ns::groupMask(p,base,count)!=expected)throw std::runtime_error("group mask mismatch");
|
||||
++groups;
|
||||
}
|
||||
// Positive low18 roots and all middle values: check complete filter masks.
|
||||
sr::CellTerm cells[25];const uint64_t anchor=sr::coordinateTerm(51069,419124);
|
||||
for(int i=0;i<25;++i)cells[i]={(sr::coordinateTerm(51069+i%5,419124+i/5)-anchor)&sr::kMask48,sr::kSlime};
|
||||
uint64_t accepted=0;
|
||||
for(uint32_t root=0;root<(1U<<18);++root){
|
||||
if(!sr::partialPatternPossible(root,18,cells,25))continue;
|
||||
ns::Prepared p;ns::prepare(root,cells,p);
|
||||
for(uint32_t g=0;g<8192;++g){
|
||||
uint32_t base=((g>>4)<<8)+(g&15),expected=0;
|
||||
for(uint32_t j=0;j<16;++j)if(full_search_experiment::differenceFilter(root+(uint64_t(base+16*j)<<18),cells,25))expected|=1U<<j;
|
||||
const uint32_t actual=ns::groupMask(p,base);
|
||||
if(actual!=expected)throw std::runtime_error("positive group mask mismatch");
|
||||
accepted+=__builtin_popcount(actual);++groups;
|
||||
}
|
||||
}
|
||||
if(!accepted)throw std::runtime_error("no positive candidates tested");
|
||||
std::cout<<"{\"q_checks\":"<<qchecks<<",\"group_masks\":"<<groups<<",\"partial_filter_survivors\":"<<partialSurvivors<<",\"known_filter_survivors\":"<<accepted<<",\"passed\":true}\n";
|
||||
}
|
||||
215
research/experiments/normal_search/cuda_bench.cu
Normal file
215
research/experiments/normal_search/cuda_bench.cu
Normal file
|
|
@ -0,0 +1,215 @@
|
|||
#include "grouped.hpp"
|
||||
#include <cuda_runtime.h>
|
||||
#include <algorithm>
|
||||
#include <array>
|
||||
#include <chrono>
|
||||
#include <cstdint>
|
||||
#include <fstream>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <map>
|
||||
#include <numeric>
|
||||
#include <stdexcept>
|
||||
#include <string>
|
||||
#include <tuple>
|
||||
#include <vector>
|
||||
|
||||
namespace sr = slime_research;
|
||||
namespace fs = full_search_experiment;
|
||||
using Clock = std::chrono::steady_clock;
|
||||
#define CUDA(call) do { auto e = (call); if (e != cudaSuccess) throw std::runtime_error(std::string(#call)+": "+cudaGetErrorString(e)); } while (0)
|
||||
struct Position { int32_t x, z; uint64_t anchor; sr::CellTerm cells[25]; };
|
||||
struct Job { uint32_t position, root; };
|
||||
struct Match { uint64_t position, seed; bool operator<(const Match& o)const{return std::tie(position,seed)<std::tie(o.position,o.seed);} bool operator==(const Match& o)const{return position==o.position&&seed==o.seed;} };
|
||||
constexpr uint32_t MATCH_CAPACITY = 1U << 20;
|
||||
constexpr uint32_t JOB_CAPACITY = 1U << 22;
|
||||
|
||||
Position makePosition(int32_t x,int32_t z) {
|
||||
Position p{}; p.x=x;p.z=z;p.anchor=sr::coordinateTerm(x,z);
|
||||
for(int j=0;j<5;++j)for(int i=0;i<5;++i)
|
||||
p.cells[j*5+i]={(sr::coordinateTerm(x+i,z+j)-p.anchor)&sr::kMask48,sr::kSlime};
|
||||
return p;
|
||||
}
|
||||
|
||||
template<int Mode>
|
||||
__global__ void rootKernel(const Position* positions,uint32_t count,const uint32_t* good,
|
||||
uint32_t* counts,Job* jobs,unsigned long long* total,bool collect) {
|
||||
uint64_t idx=uint64_t(blockIdx.x)*blockDim.x+threadIdx.x;
|
||||
if(idx >= (uint64_t(count)<<18)) return;
|
||||
const uint32_t pos=uint32_t(idx>>18), root=uint32_t(idx)&((1U<<18)-1);
|
||||
const Position& p=positions[pos];
|
||||
for(int i=0;i<25;++i){
|
||||
const uint32_t in=root+uint32_t(p.cells[i].term);
|
||||
if constexpr(Mode==0){
|
||||
if((((in^uint32_t(sr::kCombinedXor))*uint32_t(sr::kJavaMultiplier)+11U)>>17)&1U)return;
|
||||
}else{
|
||||
const uint32_t k=in&((1U<<18)-1);
|
||||
if(((good[k>>5]>>(k&31))&1U)==0)return;
|
||||
}
|
||||
}
|
||||
atomicAdd(counts+pos,1U);
|
||||
const auto at=atomicAdd(total,1ULL);
|
||||
if(collect&&at<JOB_CAPACITY)jobs[at]={pos,root};
|
||||
}
|
||||
|
||||
struct Emitter {
|
||||
uint32_t position; uint64_t anchor; Match* out; unsigned long long* count;
|
||||
__device__ void operator()(uint64_t s){
|
||||
auto at=atomicAdd(count,1ULL);
|
||||
if(at<MATCH_CAPACITY)out[at]={position,(s-anchor)&sr::kMask48};
|
||||
}
|
||||
};
|
||||
|
||||
using normal_search::Prepared;
|
||||
using normal_search::preparedQ;
|
||||
__global__ void prepareKernel(const Position* positions,const Job* jobs,uint32_t count,Prepared* out){
|
||||
const uint32_t j=blockIdx.x*blockDim.x+threadIdx.x;
|
||||
if(j<count)normal_search::prepare(jobs[j].root,positions[jobs[j].position].cells,out[j]);
|
||||
}
|
||||
template<int Mode>
|
||||
__global__ void searchKernel(const Position* positions,const Job* jobs,uint32_t count,
|
||||
Match* out,unsigned long long* total,const Prepared* prepared) {
|
||||
uint64_t idx=uint64_t(blockIdx.x)*blockDim.x+threadIdx.x;
|
||||
constexpr int bits=Mode==7?13:17;
|
||||
if(idx >= (uint64_t(count)<<bits))return;
|
||||
const Job job=jobs[idx>>bits];
|
||||
const uint32_t m=idx&((1U<<bits)-1);
|
||||
uint64_t low=job.root+(uint64_t(m)<<18);
|
||||
const Position& p=positions[job.position];
|
||||
Emitter emit{job.position,p.anchor,out,total};
|
||||
if constexpr(Mode==7){
|
||||
const uint32_t base=((m>>4)<<8)+(m&15);
|
||||
uint32_t mask=normal_search::groupMask(prepared[idx>>13],base);
|
||||
while(mask){
|
||||
const uint32_t j=__ffs(mask)-1;mask &= mask-1;
|
||||
const uint64_t candidate=job.root+(uint64_t(base+16*j)<<18);
|
||||
fs::solveIntervals(candidate,p.cells,25,emit);
|
||||
}
|
||||
}
|
||||
else if constexpr(Mode==3){if(fs::differenceFilter(low,p.cells,25))fs::solveIntervals(low,p.cells,25,emit);}
|
||||
else {
|
||||
const Prepared& t=prepared[idx>>17];
|
||||
const uint32_t q0=preparedQ(t,0,m);
|
||||
const uint32_t u0=(m+t.h[0])^uint32_t(sr::kCombinedXor>>18);
|
||||
const int order[24]={24,4,20,12,2,10,14,22,6,8,16,18,1,3,5,9,15,19,21,23,7,11,13,17};
|
||||
for(int k=1;k<25;++k){
|
||||
const int i=Mode==6?order[k-1]:k;
|
||||
if constexpr(Mode==4){
|
||||
const uint32_t r=((preparedQ(t,i,m)-q0)&0x7fffffffU)%10;
|
||||
if(r!=0&&r!=8)return;
|
||||
}else{
|
||||
const uint32_t ui=(m+t.h[i])^uint32_t(sr::kCombinedXor>>18);
|
||||
const uint32_t d=((t.b[i]>>1)-(t.b[0]>>1)+uint32_t(sr::kJavaMultiplier)*(ui-u0))&((1U<<30)-1);
|
||||
const uint32_t r=d%5;
|
||||
if(r!=0&&r!=4)return;
|
||||
}
|
||||
}
|
||||
fs::solveIntervals(low,p.cells,25,emit);
|
||||
}
|
||||
}
|
||||
struct Device {
|
||||
Prepared* prepared=nullptr; Position* positions=nullptr; Job* jobs=nullptr; Match* matches=nullptr;
|
||||
uint32_t* counts=nullptr;uint32_t* good=nullptr;unsigned long long* total=nullptr;
|
||||
cudaEvent_t start,stop;
|
||||
Device(const std::vector<Position>& p){
|
||||
CUDA(cudaMalloc(&positions,p.size()*sizeof(Position)));CUDA(cudaMemcpy(positions,p.data(),p.size()*sizeof(Position),cudaMemcpyHostToDevice));
|
||||
CUDA(cudaMalloc(&jobs,JOB_CAPACITY*sizeof(Job)));CUDA(cudaMalloc(&matches,MATCH_CAPACITY*sizeof(Match)));
|
||||
CUDA(cudaMalloc(&counts,p.size()*sizeof(uint32_t)));CUDA(cudaMalloc(&total,sizeof(*total)));
|
||||
std::array<uint32_t,8192> bits{};
|
||||
for(uint32_t k=0;k<(1U<<18);++k)if(!((((k^uint32_t(sr::kCombinedXor))*uint32_t(sr::kJavaMultiplier)+11U)>>17)&1U))bits[k>>5]|=1U<<(k&31);
|
||||
CUDA(cudaMalloc(&good,sizeof(bits)));CUDA(cudaMemcpy(good,bits.data(),sizeof(bits),cudaMemcpyHostToDevice));
|
||||
CUDA(cudaEventCreate(&start));CUDA(cudaEventCreate(&stop));
|
||||
}
|
||||
~Device(){cudaFree(prepared);cudaFree(positions);cudaFree(jobs);cudaFree(matches);cudaFree(counts);cudaFree(total);cudaFree(good);cudaEventDestroy(start);cudaEventDestroy(stop);}
|
||||
float finish(){CUDA(cudaGetLastError());CUDA(cudaEventRecord(stop));CUDA(cudaEventSynchronize(stop));float ms;CUDA(cudaEventElapsedTime(&ms,start,stop));return ms;}
|
||||
uint64_t totalCount(){unsigned long long n;CUDA(cudaMemcpy(&n,total,sizeof(n),cudaMemcpyDeviceToHost));return n;}
|
||||
float roots(uint32_t n,int mode,bool collect){
|
||||
CUDA(cudaMemset(total,0,sizeof(*total)));CUDA(cudaMemset(counts,0,n*sizeof(uint32_t)));CUDA(cudaEventRecord(start));
|
||||
unsigned blocks=unsigned(((uint64_t(n)<<18)+255)/256);
|
||||
if(mode==0)rootKernel<0><<<blocks,256>>>(positions,n,good,counts,jobs,total,collect);
|
||||
else rootKernel<1><<<blocks,256>>>(positions,n,good,counts,jobs,total,collect);
|
||||
return finish();
|
||||
}
|
||||
float search(uint32_t n,int mode){
|
||||
if(!prepared&&n)CUDA(cudaMalloc(&prepared,n*sizeof(Prepared)));
|
||||
CUDA(cudaMemset(total,0,sizeof(*total)));CUDA(cudaEventRecord(start));
|
||||
unsigned blocks=unsigned(((uint64_t(n)<<17)+255)/256);
|
||||
if(blocks){
|
||||
if(mode>=4)prepareKernel<<<(n+127)/128,128>>>(positions,jobs,n,prepared);
|
||||
if(mode==3)searchKernel<3><<<blocks,256>>>(positions,jobs,n,matches,total,prepared);
|
||||
if(mode==4)searchKernel<4><<<blocks,256>>>(positions,jobs,n,matches,total,prepared);
|
||||
if(mode==5)searchKernel<5><<<blocks,256>>>(positions,jobs,n,matches,total,prepared);
|
||||
if(mode==6)searchKernel<6><<<blocks,256>>>(positions,jobs,n,matches,total,prepared);
|
||||
if(mode==7)searchKernel<7><<<unsigned(((uint64_t(n)<<13)+255)/256),256>>>(positions,jobs,n,matches,total,prepared);
|
||||
}
|
||||
return finish();
|
||||
}
|
||||
std::vector<Match> results(){
|
||||
uint64_t n=totalCount();if(n>MATCH_CAPACITY)throw std::runtime_error("match overflow");
|
||||
std::vector<Match> r(n);CUDA(cudaMemcpy(r.data(),matches,n*sizeof(Match),cudaMemcpyDeviceToHost));std::sort(r.begin(),r.end());return r;
|
||||
}
|
||||
};
|
||||
|
||||
int main(int argc,char**argv){
|
||||
try{
|
||||
if(argc<3)throw std::runtime_error("usage: cuda_bench positions.txt normal [repeats=7]");
|
||||
std::ifstream in(argv[1]);if(!in)throw std::runtime_error("cannot open positions");
|
||||
int32_t x,z;std::vector<Position> positions;while(in>>x>>z)positions.push_back(makePosition(x,z));
|
||||
if(positions.empty()||positions.size()>65536)throw std::runtime_error("position count must be 1..65536");
|
||||
std::string task=argv[2];int repeats=argc>3?std::stoi(argv[3]):5;
|
||||
if(task!="normal"||repeats<1)throw std::runtime_error("expected normal and positive repeat count");
|
||||
auto processStart=Clock::now();Device d(positions);
|
||||
std::cout<<std::setprecision(10);
|
||||
cudaDeviceProp prop{};CUDA(cudaGetDeviceProperties(&prop,0));std::cout<<"{\"type\":\"device\",\"name\":\""<<prop.name<<"\",\"positions\":"<<positions.size()<<"}"<<std::endl;
|
||||
std::vector<uint32_t> referenceCounts;
|
||||
std::vector<Job> referenceJobs;
|
||||
auto collectRoots=[&](int mode){
|
||||
float ms=d.roots(positions.size(),mode,true);uint64_t total=d.totalCount();if(total>JOB_CAPACITY)throw std::runtime_error("job overflow");
|
||||
std::vector<uint32_t> counts(positions.size());CUDA(cudaMemcpy(counts.data(),d.counts,counts.size()*sizeof(uint32_t),cudaMemcpyDeviceToHost));
|
||||
std::vector<Job> jobs(total);CUDA(cudaMemcpy(jobs.data(),d.jobs,total*sizeof(Job),cudaMemcpyDeviceToHost));
|
||||
std::sort(jobs.begin(),jobs.end(),[](Job a,Job b){return std::tie(a.position,a.root)<std::tie(b.position,b.root);});
|
||||
if(referenceCounts.empty()){referenceCounts=counts;referenceJobs=jobs;}
|
||||
else {
|
||||
if(counts!=referenceCounts||jobs.size()!=referenceJobs.size())throw std::runtime_error("root mismatch");
|
||||
for(size_t i=0;i<jobs.size();++i)if(jobs[i].position!=referenceJobs[i].position||jobs[i].root!=referenceJobs[i].root)throw std::runtime_error("root list mismatch");
|
||||
}
|
||||
return ms;
|
||||
};
|
||||
{
|
||||
float rootMs=collectRoots(0);
|
||||
CUDA(cudaMemcpy(d.jobs,referenceJobs.data(),referenceJobs.size()*sizeof(Job),cudaMemcpyHostToDevice));
|
||||
std::vector<Match> expected;
|
||||
const std::vector<int> modes=std::vector<int>{3,4,5,6,7};
|
||||
bool first=true;
|
||||
for(int m:modes){d.search(referenceJobs.size(),m);auto r=d.results();if(first){expected=r;first=false;}else if(r!=expected)throw std::runtime_error("warmup result mismatch");}
|
||||
for(int r=0;r<repeats;++r)for(size_t k=0;k<modes.size();++k){
|
||||
int mode=modes[(k+r)%modes.size()];auto wallStart=Clock::now();float ms=d.search(referenceJobs.size(),mode);auto result=d.results();
|
||||
if(result!=expected)throw std::runtime_error("search result mismatch");
|
||||
std::cout<<"{\"type\":\"search_timing\",\"mode\":"<<mode<<",\"round\":"<<r<<",\"kernel_ms\":"<<ms<<",\"scan_ms\":"<<std::chrono::duration<double,std::milli>(Clock::now()-wallStart).count()<<",\"matches\":"<<result.size()<<"}"<<std::endl;
|
||||
}
|
||||
std::cout<<"{\"type\":\"search_info\",\"root_ms\":"<<rootMs<<",\"jobs\":"<<referenceJobs.size()<<",\"lower35\":"<<(uint64_t(referenceJobs.size())<<17)<<",\"matches\":[";
|
||||
for(size_t i=0;i<expected.size();++i){if(i)std::cout<<",";const auto& p=positions[expected[i].position];std::cout<<"["<<expected[i].seed<<","<<p.x<<","<<p.z<<"]";}std::cout<<"]}"<<std::endl;
|
||||
}
|
||||
// CPU checks of complete low18 root lists, including varied coordinates.
|
||||
for(size_t n=0;n<std::min<size_t>(64,positions.size());++n){
|
||||
size_t pos=n*positions.size()/std::min<size_t>(64,positions.size());std::vector<uint32_t> cpu,gpu;
|
||||
for(uint32_t s=0;s<(1U<<18);++s)if(sr::partialPatternPossible(s,18,positions[pos].cells,25))cpu.push_back(s);
|
||||
for(auto j:referenceJobs)if(j.position==pos)gpu.push_back(j.root);
|
||||
if(cpu!=gpu)throw std::runtime_error("CPU/GPU roots mismatch");
|
||||
}
|
||||
std::map<uint64_t,std::vector<uint32_t>> classes;
|
||||
size_t offset=0;
|
||||
for(size_t pos=0;pos<positions.size();++pos){
|
||||
uint64_t key=(uint64_t(uint32_t(positions[pos].x)&65535)<<17)|(uint32_t(positions[pos].z)&131071);
|
||||
std::vector<uint32_t> roots;
|
||||
while(offset<referenceJobs.size()&&referenceJobs[offset].position==pos)roots.push_back(referenceJobs[offset++].root);
|
||||
auto it=classes.find(key);
|
||||
if(it==classes.end())classes.emplace(key,std::move(roots));
|
||||
else if(it->second!=roots)throw std::runtime_error("periodic class root list mismatch");
|
||||
}
|
||||
std::cout<<"{\"type\":\"periodic_validation\",\"classes\":"<<classes.size()<<",\"positions\":"<<positions.size()<<",\"equal\":true}"<<std::endl;
|
||||
std::cout<<"{\"type\":\"root_counts\",\"counts\":[";for(size_t i=0;i<referenceCounts.size();++i){if(i)std::cout<<",";std::cout<<referenceCounts[i];}std::cout<<"]}"<<std::endl;
|
||||
std::cout<<"{\"type\":\"process\",\"ms\":"<<std::chrono::duration<double,std::milli>(Clock::now()-processStart).count()<<"}"<<std::endl;
|
||||
return 0;
|
||||
}catch(const std::exception&e){std::cerr<<e.what()<<"\n";return 1;}
|
||||
}
|
||||
65536
research/experiments/normal_search/fresh65536.txt
Normal file
65536
research/experiments/normal_search/fresh65536.txt
Normal file
File diff suppressed because it is too large
Load diff
42
research/experiments/normal_search/grouped.hpp
Normal file
42
research/experiments/normal_search/grouped.hpp
Normal file
|
|
@ -0,0 +1,42 @@
|
|||
#pragma once
|
||||
#include "../full_search_feasibility/interval_solver.hpp"
|
||||
#if defined(__CUDACC__)
|
||||
#define NS_HD __host__ __device__ __forceinline__
|
||||
#else
|
||||
#define NS_HD inline
|
||||
#endif
|
||||
namespace normal_search {
|
||||
namespace sr=slime_research;
|
||||
struct Prepared { uint32_t h[25], b[25]; };
|
||||
NS_HD void prepare(uint32_t root,const sr::CellTerm* cells,Prepared& out){
|
||||
for(int i=0;i<25;++i){
|
||||
const uint64_t a=root+cells[i].term;
|
||||
out.h[i]=uint32_t(a>>18);
|
||||
const uint64_t lo=(a^sr::kCombinedXor)&((1U<<18)-1);
|
||||
out.b[i]=uint32_t((lo*sr::kJavaMultiplier+11)>>17);
|
||||
}
|
||||
}
|
||||
NS_HD uint32_t preparedQ(const Prepared& p,int i,uint32_t m){
|
||||
const uint32_t u=(m+p.h[i])^uint32_t(sr::kCombinedXor>>18);
|
||||
return (p.b[i]+2U*uint32_t(sr::kJavaMultiplier)*u)&0x7fffffffU;
|
||||
}
|
||||
|
||||
// Bits 4..7 of kCombinedXor >> 18 are zero. For m=base+16*j,
|
||||
// j in [0,16), m XOR C is affine. Each shifted input crosses at
|
||||
// most one 256 boundary, so each output difference has at most two values.
|
||||
// base bits 4..7 must be zero; cells[0].term must be zero (normalized anchor).
|
||||
NS_HD uint32_t groupMask(const Prepared& p,uint32_t base,int count=25){
|
||||
uint32_t mask=65535;
|
||||
const uint32_t q0=preparedQ(p,0,base),qend=preparedQ(p,0,base+240);
|
||||
for(int i=1;i<count&&mask;++i){
|
||||
const uint32_t n=(256-((base+p.h[i])&255)+15)>>4;
|
||||
const uint32_t before=(1U<<n)-1;
|
||||
const uint32_t r0=((preparedQ(p,i,base)-q0)&0x7fffffffU)%10;
|
||||
const uint32_t r1=((preparedQ(p,i,base+240)-qend)&0x7fffffffU)%10;
|
||||
const uint32_t allowed=((r0==0||r0==8)?before:0U)|((r1==0||r1==8)?(65535U^before):0U);
|
||||
mask &= allowed;
|
||||
}
|
||||
return mask;
|
||||
}
|
||||
}
|
||||
#undef NS_HD
|
||||
114
research/experiments/normal_search/pipeline_bench.cu
Normal file
114
research/experiments/normal_search/pipeline_bench.cu
Normal file
|
|
@ -0,0 +1,114 @@
|
|||
#define main component_benchmark_main
|
||||
#include "cuda_bench.cu"
|
||||
#undef main
|
||||
|
||||
__global__ void expandJobs(const uint32_t* positionClass,const uint32_t* offsets,
|
||||
const uint32_t* classOffsets,const uint32_t* roots,
|
||||
uint32_t n,Job* jobs){
|
||||
uint32_t p=blockIdx.x*blockDim.x+threadIdx.x;
|
||||
if(p>=n)return;
|
||||
const uint32_t c=positionClass[p];
|
||||
for(uint32_t i=classOffsets[c];i<classOffsets[c+1];++i)
|
||||
jobs[offsets[p]+i-classOffsets[c]]={p,roots[i]};
|
||||
}
|
||||
struct Extra {
|
||||
Position* reps;uint32_t *positionClass,*offsets,*classOffsets,*roots;
|
||||
Extra(size_t n){
|
||||
CUDA(cudaMalloc(&reps,n*sizeof(Position)));
|
||||
CUDA(cudaMalloc(&positionClass,n*sizeof(uint32_t)));
|
||||
CUDA(cudaMalloc(&offsets,(n+1)*sizeof(uint32_t)));
|
||||
CUDA(cudaMalloc(&classOffsets,(n+1)*sizeof(uint32_t)));
|
||||
CUDA(cudaMalloc(&roots,JOB_CAPACITY*sizeof(uint32_t)));
|
||||
}
|
||||
~Extra(){cudaFree(reps);cudaFree(positionClass);cudaFree(offsets);cudaFree(classOffsets);cudaFree(roots);}
|
||||
};
|
||||
struct Run {double hostMs,buildMs,rootMs,expandMs,searchMs;uint32_t classes,jobs,activePositions;std::vector<Match> matches;std::vector<Job> rootList;};
|
||||
Run run(const std::vector<std::pair<int32_t,int32_t>>& coords,Device& d,Extra& extra,int mode){
|
||||
const auto start=Clock::now();
|
||||
auto ms=[](auto t){return std::chrono::duration<double,std::milli>(Clock::now()-t).count();};
|
||||
std::vector<Position> positions;positions.reserve(coords.size());
|
||||
if(mode!=3)for(auto [x,z]:coords)positions.push_back(makePosition(x,z));
|
||||
std::vector<Position> reps;std::vector<uint32_t> pc;
|
||||
if(mode>=2){
|
||||
std::map<uint64_t,uint32_t> map;
|
||||
for(auto [x,z]:coords){
|
||||
uint64_t key=(uint64_t(uint32_t(x)&65535)<<17)|(uint32_t(z)&131071);
|
||||
auto [it,inserted]=map.emplace(key,uint32_t(reps.size()));
|
||||
if(inserted)reps.push_back(makePosition(x,z));
|
||||
pc.push_back(it->second);
|
||||
}
|
||||
}
|
||||
if(!positions.empty())CUDA(cudaMemcpy(d.positions,positions.data(),positions.size()*sizeof(Position),cudaMemcpyHostToDevice));
|
||||
Run r{};r.classes=mode>=2?reps.size():positions.size();r.buildMs=ms(start);
|
||||
const auto rootStart=Clock::now();
|
||||
if(mode>=2){
|
||||
CUDA(cudaMemcpy(extra.reps,reps.data(),reps.size()*sizeof(Position),cudaMemcpyHostToDevice));
|
||||
CUDA(cudaMemset(d.total,0,sizeof(*d.total)));CUDA(cudaMemset(d.counts,0,reps.size()*sizeof(uint32_t)));
|
||||
rootKernel<0><<<unsigned(((uint64_t(reps.size())<<18)+255)/256),256>>>(extra.reps,reps.size(),nullptr,d.counts,d.jobs,d.total,true);
|
||||
}else d.roots(positions.size(),0,true);
|
||||
uint64_t n=d.totalCount();if(n>JOB_CAPACITY)throw std::runtime_error("root capacity exceeded");
|
||||
r.rootMs=ms(rootStart);
|
||||
const auto expandStart=Clock::now();
|
||||
std::vector<uint32_t> originalIndices;
|
||||
if(mode>=2){
|
||||
std::vector<Job> cj(n);if(n)CUDA(cudaMemcpy(cj.data(),d.jobs,n*sizeof(Job),cudaMemcpyDeviceToHost));
|
||||
std::vector<std::vector<uint32_t>> lists(reps.size());
|
||||
for(auto j:cj)lists[j.position].push_back(j.root);
|
||||
if(mode==3){
|
||||
std::vector<uint32_t> compactClasses;
|
||||
for(uint32_t i=0;i<coords.size();++i)if(!lists[pc[i]].empty()){
|
||||
auto [x,z]=coords[i];positions.push_back(makePosition(x,z));
|
||||
compactClasses.push_back(pc[i]);originalIndices.push_back(i);
|
||||
}
|
||||
pc=std::move(compactClasses);
|
||||
if(!positions.empty())CUDA(cudaMemcpy(d.positions,positions.data(),positions.size()*sizeof(Position),cudaMemcpyHostToDevice));
|
||||
}
|
||||
std::vector<uint32_t> roots,co{0},offsets{0};
|
||||
for(auto& list:lists){std::sort(list.begin(),list.end());roots.insert(roots.end(),list.begin(),list.end());co.push_back(roots.size());}
|
||||
uint64_t expanded=0;
|
||||
for(auto c:pc){expanded+=lists[c].size();if(expanded>JOB_CAPACITY)throw std::runtime_error("expanded capacity exceeded; use smaller batch");offsets.push_back(expanded);}
|
||||
n=expanded;
|
||||
CUDA(cudaMemcpy(extra.positionClass,pc.data(),pc.size()*sizeof(uint32_t),cudaMemcpyHostToDevice));
|
||||
CUDA(cudaMemcpy(extra.offsets,offsets.data(),offsets.size()*sizeof(uint32_t),cudaMemcpyHostToDevice));
|
||||
CUDA(cudaMemcpy(extra.classOffsets,co.data(),co.size()*sizeof(uint32_t),cudaMemcpyHostToDevice));
|
||||
if(!roots.empty())CUDA(cudaMemcpy(extra.roots,roots.data(),roots.size()*sizeof(uint32_t),cudaMemcpyHostToDevice));
|
||||
if(!positions.empty())expandJobs<<<unsigned((positions.size()+255)/256),256>>>(extra.positionClass,extra.offsets,extra.classOffsets,extra.roots,positions.size(),d.jobs);
|
||||
CUDA(cudaGetLastError());CUDA(cudaDeviceSynchronize());
|
||||
}
|
||||
r.expandMs=ms(expandStart);r.jobs=n;r.activePositions=positions.size();
|
||||
const auto searchStart=Clock::now();
|
||||
d.search(n,mode==0?3:7);r.matches=d.results();
|
||||
if(mode==3)for(auto& match:r.matches)match.position=originalIndices[match.position];
|
||||
r.searchMs=ms(searchStart);r.hostMs=ms(start);
|
||||
// Validation-only transfers/sorting are outside reported pipeline time.
|
||||
r.rootList.resize(n);if(n)CUDA(cudaMemcpy(r.rootList.data(),d.jobs,n*sizeof(Job),cudaMemcpyDeviceToHost));
|
||||
if(mode==3)for(auto& job:r.rootList)job.position=originalIndices[job.position];
|
||||
std::sort(r.rootList.begin(),r.rootList.end(),[](Job a,Job b){return std::tie(a.position,a.root)<std::tie(b.position,b.root);});
|
||||
return r;
|
||||
}
|
||||
int main(int argc,char**argv){try{
|
||||
if(argc<2)throw std::runtime_error("usage: pipeline_bench positions.txt [repeats=5]");
|
||||
const auto process=Clock::now();
|
||||
std::ifstream in(argv[1]);if(!in)throw std::runtime_error("cannot open input");
|
||||
std::vector<std::pair<int32_t,int32_t>> coords;int32_t x,z;while(in>>x>>z)coords.emplace_back(x,z);
|
||||
if(coords.empty()||coords.size()>65536)throw std::runtime_error("expected 1..65536 coordinates");
|
||||
int repeats=argc>2?std::stoi(argv[2]):5;if(repeats<1)throw std::runtime_error("invalid repeats");
|
||||
Device d(std::vector<Position>(coords.size()));Extra e(coords.size());
|
||||
std::cout<<std::setprecision(10);
|
||||
std::cout<<"{\"type\":\"setup\",\"host_ms\":"<<std::chrono::duration<double,std::milli>(Clock::now()-process).count()<<"}"<<std::endl;
|
||||
Run ref;
|
||||
for(int round=-1;round<repeats;++round)for(int k=0;k<4;++k){
|
||||
int mode=round<0?k:(round+k)%4;
|
||||
Run r=run(coords,d,e,mode);
|
||||
if(round==-1&&mode==0)ref=r;
|
||||
else {
|
||||
if(r.matches!=ref.matches||r.rootList.size()!=ref.rootList.size())throw std::runtime_error("pipeline mismatch");
|
||||
for(size_t i=0;i<r.rootList.size();++i)if(r.rootList[i].position!=ref.rootList[i].position||r.rootList[i].root!=ref.rootList[i].root)throw std::runtime_error("complete root list mismatch");
|
||||
}
|
||||
if(round>=0)std::cout<<"{\"type\":\"pipeline\",\"mode\":"<<mode<<",\"round\":"<<round<<",\"positions\":"<<coords.size()<<",\"classes\":"<<r.classes<<",\"active_positions\":"<<r.activePositions<<",\"jobs\":"<<r.jobs<<",\"matches\":"<<r.matches.size()<<",\"host_ms\":"<<r.hostMs<<",\"build_ms\":"<<r.buildMs<<",\"root_ms\":"<<r.rootMs<<",\"expand_ms\":"<<r.expandMs<<",\"search_ms\":"<<r.searchMs<<"}"<<std::endl;
|
||||
}
|
||||
std::cout<<"{\"type\":\"validation\",\"complete_roots_equal\":true,\"complete_matches_equal\":true,\"matches\":[";
|
||||
for(size_t i=0;i<ref.matches.size();++i){if(i)std::cout<<",";auto m=ref.matches[i];std::cout<<"["<<m.seed<<","<<coords[m.position].first<<","<<coords[m.position].second<<"]";}
|
||||
std::cout<<"]}"<<std::endl;
|
||||
return 0;
|
||||
}catch(const std::exception& e){std::cerr<<e.what()<<"\n";return 1;}}
|
||||
File diff suppressed because it is too large
Load diff
32768
research/experiments/normal_search/pipeline_positions/mixed512.txt
Normal file
32768
research/experiments/normal_search/pipeline_positions/mixed512.txt
Normal file
File diff suppressed because it is too large
Load diff
32768
research/experiments/normal_search/pipeline_positions/shuffled_mixed.txt
Normal file
32768
research/experiments/normal_search/pipeline_positions/shuffled_mixed.txt
Normal file
File diff suppressed because it is too large
Load diff
1653
research/experiments/normal_search/pipeline_positions/zero_lifts.txt
Normal file
1653
research/experiments/normal_search/pipeline_positions/zero_lifts.txt
Normal file
File diff suppressed because it is too large
Load diff
|
|
@ -0,0 +1,339 @@
|
|||
{
|
||||
"zero_lifts": {
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"modes": [
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 0,
|
||||
"round": 0,
|
||||
"positions": 1653,
|
||||
"classes": 1653,
|
||||
"active_positions": 1653,
|
||||
"jobs": 0,
|
||||
"matches": 0,
|
||||
"host_ms": 5.571692,
|
||||
"build_ms": 0.090188,
|
||||
"root_ms": 5.379603,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 0.101716
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 1,
|
||||
"round": 0,
|
||||
"positions": 1653,
|
||||
"classes": 1653,
|
||||
"active_positions": 1653,
|
||||
"jobs": 0,
|
||||
"matches": 0,
|
||||
"host_ms": 5.729661,
|
||||
"build_ms": 0.1019,
|
||||
"root_ms": 5.497826,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 0.12975
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 2,
|
||||
"round": 0,
|
||||
"positions": 1653,
|
||||
"classes": 1,
|
||||
"active_positions": 1653,
|
||||
"jobs": 0,
|
||||
"matches": 0,
|
||||
"host_ms": 0.56059,
|
||||
"build_ms": 0.10771,
|
||||
"root_ms": 0.251016,
|
||||
"expand_ms": 0.131594,
|
||||
"search_ms": 0.070177
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 3,
|
||||
"round": 0,
|
||||
"positions": 1653,
|
||||
"classes": 1,
|
||||
"active_positions": 0,
|
||||
"jobs": 0,
|
||||
"matches": 0,
|
||||
"host_ms": 0.230821,
|
||||
"build_ms": 0.006363,
|
||||
"root_ms": 0.083272,
|
||||
"expand_ms": 0.071838,
|
||||
"search_ms": 0.069348
|
||||
}
|
||||
]
|
||||
},
|
||||
"stress": {
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": [
|
||||
[
|
||||
10358701894766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
15959335054446,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
18914276748398,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
24514909908078,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
27469851602030,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
33070484761710,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
36025426455662,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
41626059615342,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
44581001309294,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
50181634468974,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
53136576162926,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
58737209322606,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
61692151016558,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
67292784176238,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
70247725870190,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
75848359029870,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
78803300723822,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
84403933883502,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
87358875577454,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
92959508737134,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
95914450431086,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
101515083590766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
104470025284718,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
110070658444398,
|
||||
51069,
|
||||
419124
|
||||
]
|
||||
]
|
||||
},
|
||||
"modes": [
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 0,
|
||||
"round": 0,
|
||||
"positions": 21,
|
||||
"classes": 21,
|
||||
"active_positions": 21,
|
||||
"jobs": 474,
|
||||
"matches": 24,
|
||||
"host_ms": 1.476311,
|
||||
"build_ms": 0.003135,
|
||||
"root_ms": 0.17549,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 1.297501
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 1,
|
||||
"round": 0,
|
||||
"positions": 21,
|
||||
"classes": 21,
|
||||
"active_positions": 21,
|
||||
"jobs": 474,
|
||||
"matches": 24,
|
||||
"host_ms": 0.505168,
|
||||
"build_ms": 0.004611,
|
||||
"root_ms": 0.175213,
|
||||
"expand_ms": 9.3e-05,
|
||||
"search_ms": 0.325159
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 2,
|
||||
"round": 0,
|
||||
"positions": 21,
|
||||
"classes": 20,
|
||||
"active_positions": 21,
|
||||
"jobs": 474,
|
||||
"matches": 24,
|
||||
"host_ms": 0.834753,
|
||||
"build_ms": 0.011158,
|
||||
"root_ms": 0.242071,
|
||||
"expand_ms": 0.210348,
|
||||
"search_ms": 0.370991
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 3,
|
||||
"round": 0,
|
||||
"positions": 21,
|
||||
"classes": 20,
|
||||
"active_positions": 13,
|
||||
"jobs": 474,
|
||||
"matches": 24,
|
||||
"host_ms": 0.782097,
|
||||
"build_ms": 0.004335,
|
||||
"root_ms": 0.1967,
|
||||
"expand_ms": 0.261529,
|
||||
"search_ms": 0.319533
|
||||
}
|
||||
]
|
||||
},
|
||||
"shuffled_mixed": {
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"modes": [
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 0,
|
||||
"round": 0,
|
||||
"positions": 32768,
|
||||
"classes": 32768,
|
||||
"active_positions": 32768,
|
||||
"jobs": 15424,
|
||||
"matches": 0,
|
||||
"host_ms": 152.132767,
|
||||
"build_ms": 4.096027,
|
||||
"root_ms": 113.387564,
|
||||
"expand_ms": 0.000184,
|
||||
"search_ms": 34.648623
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 1,
|
||||
"round": 0,
|
||||
"positions": 32768,
|
||||
"classes": 32768,
|
||||
"active_positions": 32768,
|
||||
"jobs": 15424,
|
||||
"matches": 0,
|
||||
"host_ms": 124.087013,
|
||||
"build_ms": 4.152926,
|
||||
"root_ms": 113.998689,
|
||||
"expand_ms": 0.000184,
|
||||
"search_ms": 5.934844
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 2,
|
||||
"round": 0,
|
||||
"positions": 32768,
|
||||
"classes": 512,
|
||||
"active_positions": 32768,
|
||||
"jobs": 15424,
|
||||
"matches": 0,
|
||||
"host_ms": 13.605316,
|
||||
"build_ms": 5.613835,
|
||||
"root_ms": 1.992268,
|
||||
"expand_ms": 0.604117,
|
||||
"search_ms": 5.394911
|
||||
},
|
||||
{
|
||||
"type": "pipeline",
|
||||
"mode": 3,
|
||||
"round": 0,
|
||||
"positions": 32768,
|
||||
"classes": 512,
|
||||
"active_positions": 1216,
|
||||
"jobs": 15424,
|
||||
"matches": 0,
|
||||
"host_ms": 9.609161,
|
||||
"build_ms": 1.342779,
|
||||
"root_ms": 2.238581,
|
||||
"expand_ms": 0.603472,
|
||||
"search_ms": 5.424052
|
||||
}
|
||||
]
|
||||
},
|
||||
"source_sha256": {
|
||||
"pipeline_bench.cu": "31da815aa0ee8904c702e55d2d685282fc64bfada24dd208e151faac5c55c37c",
|
||||
"cuda_bench.cu": "3e1fd6fe3bdb87ac68fe853dc4e6d2275d6aa00a692c7d630391fd12c0105c20",
|
||||
"grouped.hpp": "5d48c4a45e7f7b07a362c69b5163f48d44bd45d18e7f0f02b00bda0edaadeb23",
|
||||
"verify_pipeline_edges.py": "97a2b4642e871149519a79c73b7b617dd7a0cc47cd0f0dfe59cc7c8afdf0507c"
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,301 @@
|
|||
{
|
||||
"metadata": {
|
||||
"date": "2026-10-11T10:49:17.203764+08:00",
|
||||
"gpu": "name, driver_version\nNVIDIA GeForce RTX 3060 Laptop GPU, 596.49\n"
|
||||
},
|
||||
"known_lifts": {
|
||||
"0": {
|
||||
"host_ms": 359.494613,
|
||||
"build_ms": 0.100726,
|
||||
"root_ms": 8.470989,
|
||||
"expand_ms": 9.6e-05,
|
||||
"search_ms": 350.821402,
|
||||
"jobs": 138852,
|
||||
"classes": 1653,
|
||||
"positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 61.726379,
|
||||
"build_ms": 0.127304,
|
||||
"root_ms": 8.947655,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 52.659413,
|
||||
"jobs": 138852,
|
||||
"classes": 1653,
|
||||
"positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 53.625264,
|
||||
"build_ms": 0.166015,
|
||||
"root_ms": 0.356104,
|
||||
"expand_ms": 0.319126,
|
||||
"search_ms": 52.783635,
|
||||
"jobs": 138852,
|
||||
"classes": 1,
|
||||
"positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": [
|
||||
[
|
||||
10358701894766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
15959335054446,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
18914276748398,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
24514909908078,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
27469851602030,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
33070484761710,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
36025426455662,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
41626059615342,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
44581001309294,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
50181634468974,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
53136576162926,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
58737209322606,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
61692151016558,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
67292784176238,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
70247725870190,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
75848359029870,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
78803300723822,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
84403933883502,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
87358875577454,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
92959508737134,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
95914450431086,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
101515083590766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
104470025284718,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
110070658444398,
|
||||
51069,
|
||||
419124
|
||||
]
|
||||
]
|
||||
},
|
||||
"setup_ms": 269.594069
|
||||
},
|
||||
"nonempty32": {
|
||||
"0": {
|
||||
"host_ms": 4621.165908,
|
||||
"build_ms": 7.240293,
|
||||
"root_ms": 257.582953,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 4356.341998,
|
||||
"jobs": 1693228,
|
||||
"classes": 52213,
|
||||
"positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 926.774796,
|
||||
"build_ms": 7.307786,
|
||||
"root_ms": 251.583683,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 668.081633,
|
||||
"jobs": 1693228,
|
||||
"classes": 52213,
|
||||
"positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 664.598058,
|
||||
"build_ms": 8.052305,
|
||||
"root_ms": 0.602059,
|
||||
"expand_ms": 1.15488,
|
||||
"search_ms": 654.776953,
|
||||
"jobs": 1693228,
|
||||
"classes": 32,
|
||||
"positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 281.326323
|
||||
},
|
||||
"mixed512": {
|
||||
"0": {
|
||||
"host_ms": 170.81455,
|
||||
"build_ms": 4.3475,
|
||||
"root_ms": 125.510086,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 41.628402,
|
||||
"jobs": 15424,
|
||||
"classes": 32768,
|
||||
"positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 138.433459,
|
||||
"build_ms": 4.575848,
|
||||
"root_ms": 128.008956,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 5.843008,
|
||||
"jobs": 15424,
|
||||
"classes": 32768,
|
||||
"positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 14.418511,
|
||||
"build_ms": 5.463323,
|
||||
"root_ms": 2.456199,
|
||||
"expand_ms": 0.614328,
|
||||
"search_ms": 5.820053,
|
||||
"jobs": 15424,
|
||||
"classes": 512,
|
||||
"positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 259.425362
|
||||
},
|
||||
"random4096": {
|
||||
"0": {
|
||||
"host_ms": 22.41199,
|
||||
"build_ms": 0.318132,
|
||||
"root_ms": 15.081804,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 6.895335,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 16.864992,
|
||||
"build_ms": 0.311931,
|
||||
"root_ms": 15.251468,
|
||||
"expand_ms": 0,
|
||||
"search_ms": 1.256979,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 19.192994,
|
||||
"build_ms": 1.265494,
|
||||
"root_ms": 15.992881,
|
||||
"expand_ms": 0.767331,
|
||||
"search_ms": 1.202645,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 248.445216
|
||||
},
|
||||
"source_sha256": {
|
||||
"pipeline_bench.cu": "249b6119832788d37643591407bc7eeebc086af452fe22cf0c756004a28596b0",
|
||||
"cuda_bench.cu": "f762ea050d7c93c00903928f972e41106e917d37a4caceb79630bacd0f9316e1",
|
||||
"grouped.hpp": "5d48c4a45e7f7b07a362c69b5163f48d44bd45d18e7f0f02b00bda0edaadeb23",
|
||||
"run_pipeline.py": "a1aaea904d5a3a0d341ed1956e15f20af93c42c76205a2b8df108878f445ac59"
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,14 @@
|
|||
{"type":"setup","host_ms":220.229718}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":351.890247,"build_ms":0.123965,"root_ms":8.648427,"expand_ms":0.000189,"search_ms":343.117004}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":60.282088,"build_ms":0.106201,"root_ms":8.500842,"expand_ms":0.000189,"search_ms":51.674478}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":54.114753,"build_ms":0.118106,"root_ms":0.32588,"expand_ms":0.297156,"search_ms":53.373222}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":56.8665,"build_ms":0.0101,"root_ms":0.2025,"expand_ms":0.6349,"search_ms":56.0187}
|
||||
{"type":"pipeline","mode":1,"round":1,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":63.2884,"build_ms":0.113,"root_ms":9.1249,"expand_ms":0.0003,"search_ms":54.0495}
|
||||
{"type":"pipeline","mode":2,"round":1,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":55.3725,"build_ms":0.1507,"root_ms":0.4095,"expand_ms":0.2649,"search_ms":54.5471}
|
||||
{"type":"pipeline","mode":3,"round":1,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":55.7387,"build_ms":0.0109,"root_ms":0.2574,"expand_ms":0.7803,"search_ms":54.6898}
|
||||
{"type":"pipeline","mode":0,"round":1,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":374.3942,"build_ms":0.114,"root_ms":8.904,"expand_ms":0.0001,"search_ms":365.3759}
|
||||
{"type":"pipeline","mode":2,"round":2,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":56.1846,"build_ms":0.133,"root_ms":0.3246,"expand_ms":0.2974,"search_ms":55.4289}
|
||||
{"type":"pipeline","mode":3,"round":2,"positions":1653,"classes":1,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":57.1526,"build_ms":0.0088,"root_ms":0.1615,"expand_ms":0.6781,"search_ms":56.3039}
|
||||
{"type":"pipeline","mode":0,"round":2,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":364.022951,"build_ms":0.1147,"root_ms":8.7547,"expand_ms":0.0001,"search_ms":355.153251}
|
||||
{"type":"pipeline","mode":1,"round":2,"positions":1653,"classes":1653,"active_positions":1653,"jobs":138852,"matches":24,"host_ms":62.015757,"build_ms":0.113887,"root_ms":8.742087,"expand_ms":0.000191,"search_ms":53.159211}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[[10358701894766,51069,419124],[15959335054446,51069,419124],[18914276748398,51069,419124],[24514909908078,51069,419124],[27469851602030,51069,419124],[33070484761710,51069,419124],[36025426455662,51069,419124],[41626059615342,51069,419124],[44581001309294,51069,419124],[50181634468974,51069,419124],[53136576162926,51069,419124],[58737209322606,51069,419124],[61692151016558,51069,419124],[67292784176238,51069,419124],[70247725870190,51069,419124],[75848359029870,51069,419124],[78803300723822,51069,419124],[84403933883502,51069,419124],[87358875577454,51069,419124],[92959508737134,51069,419124],[95914450431086,51069,419124],[101515083590766,51069,419124],[104470025284718,51069,419124],[110070658444398,51069,419124]]}
|
||||
|
|
@ -0,0 +1,14 @@
|
|||
{"type":"setup","host_ms":242.015308}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":172.222139,"build_ms":5.020376,"root_ms":125.961985,"expand_ms":0.000192,"search_ms":41.239203}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":141.138417,"build_ms":4.616912,"root_ms":130.393185,"expand_ms":0.000287,"search_ms":6.127841}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":32768,"classes":512,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":14.093525,"build_ms":5.009729,"root_ms":2.227974,"expand_ms":0.638962,"search_ms":6.216476}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":32768,"classes":512,"active_positions":1216,"jobs":15424,"matches":0,"host_ms":9.527453,"build_ms":0.399052,"root_ms":2.04888,"expand_ms":0.821317,"search_ms":6.257916}
|
||||
{"type":"pipeline","mode":1,"round":1,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":137.613046,"build_ms":4.711591,"root_ms":126.754716,"expand_ms":9.6e-05,"search_ms":6.145971}
|
||||
{"type":"pipeline","mode":2,"round":1,"positions":32768,"classes":512,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":14.235207,"build_ms":5.358898,"root_ms":2.203034,"expand_ms":0.604141,"search_ms":6.068846}
|
||||
{"type":"pipeline","mode":3,"round":1,"positions":32768,"classes":512,"active_positions":1216,"jobs":15424,"matches":0,"host_ms":9.585776,"build_ms":0.416318,"root_ms":1.860291,"expand_ms":0.534307,"search_ms":6.774668}
|
||||
{"type":"pipeline","mode":0,"round":1,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":171.867118,"build_ms":4.452687,"root_ms":126.737258,"expand_ms":0.000288,"search_ms":40.676693}
|
||||
{"type":"pipeline","mode":2,"round":2,"positions":32768,"classes":512,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":15.318882,"build_ms":5.280144,"root_ms":2.701464,"expand_ms":0.670138,"search_ms":6.666944}
|
||||
{"type":"pipeline","mode":3,"round":2,"positions":32768,"classes":512,"active_positions":1216,"jobs":15424,"matches":0,"host_ms":10.647196,"build_ms":0.441931,"root_ms":2.114877,"expand_ms":0.690187,"search_ms":7.400009}
|
||||
{"type":"pipeline","mode":0,"round":2,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":174.051349,"build_ms":4.55245,"root_ms":127.973647,"expand_ms":0.000192,"search_ms":41.524773}
|
||||
{"type":"pipeline","mode":1,"round":2,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":144.526135,"build_ms":4.499307,"root_ms":133.867236,"expand_ms":0.000192,"search_ms":6.159304}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[]}
|
||||
|
|
@ -0,0 +1,14 @@
|
|||
{"type":"setup","host_ms":219.927973}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":4362.48845,"build_ms":7.490479,"root_ms":252.85323,"expand_ms":0.000286,"search_ms":4102.144074}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":909.690741,"build_ms":7.012591,"root_ms":242.747725,"expand_ms":0.000183,"search_ms":659.929967}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":636.817316,"build_ms":7.196841,"root_ms":0.554217,"expand_ms":1.022633,"search_ms":628.04335}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":633.461125,"build_ms":0.338158,"root_ms":0.504258,"expand_ms":8.844275,"search_ms":623.77425}
|
||||
{"type":"pipeline","mode":1,"round":1,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":915.445942,"build_ms":7.53445,"root_ms":245.545208,"expand_ms":9.2e-05,"search_ms":662.365642}
|
||||
{"type":"pipeline","mode":2,"round":1,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":656.710025,"build_ms":9.4644,"root_ms":0.576767,"expand_ms":1.500033,"search_ms":645.168458}
|
||||
{"type":"pipeline","mode":3,"round":1,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":676.742292,"build_ms":0.3428,"root_ms":0.671,"expand_ms":9.1207,"search_ms":666.607392}
|
||||
{"type":"pipeline","mode":0,"round":1,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":4691.310731,"build_ms":7.216464,"root_ms":255.097691,"expand_ms":0.000286,"search_ms":4428.995719}
|
||||
{"type":"pipeline","mode":2,"round":2,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":685.115982,"build_ms":7.652735,"root_ms":0.58709,"expand_ms":1.326998,"search_ms":675.548969}
|
||||
{"type":"pipeline","mode":3,"round":2,"positions":52213,"classes":32,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":681.412155,"build_ms":0.344485,"root_ms":0.473976,"expand_ms":8.571836,"search_ms":672.021573}
|
||||
{"type":"pipeline","mode":0,"round":2,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":4816.532778,"build_ms":9.669654,"root_ms":263.964214,"expand_ms":0.000191,"search_ms":4542.898147}
|
||||
{"type":"pipeline","mode":1,"round":2,"positions":52213,"classes":52213,"active_positions":52213,"jobs":1693228,"matches":0,"host_ms":980.983116,"build_ms":7.505432,"root_ms":259.835073,"expand_ms":0.000191,"search_ms":713.641749}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[]}
|
||||
|
|
@ -0,0 +1,14 @@
|
|||
{"type":"setup","host_ms":216.555239}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":22.564546,"build_ms":0.29315,"root_ms":15.171827,"expand_ms":0.000192,"search_ms":7.099377}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":17.388386,"build_ms":0.316459,"root_ms":15.843212,"expand_ms":0.000192,"search_ms":1.228427}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":19.225847,"build_ms":1.497691,"root_ms":15.909497,"expand_ms":0.624573,"search_ms":1.193798}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":4096,"classes":4096,"active_positions":118,"jobs":2960,"matches":0,"host_ms":17.880678,"build_ms":1.123676,"root_ms":14.852298,"expand_ms":0.623134,"search_ms":1.281474}
|
||||
{"type":"pipeline","mode":1,"round":1,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":17.902261,"build_ms":0.291998,"root_ms":16.411382,"expand_ms":0.000287,"search_ms":1.19821}
|
||||
{"type":"pipeline","mode":2,"round":1,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":19.238988,"build_ms":1.595823,"root_ms":15.836689,"expand_ms":0.631288,"search_ms":1.174708}
|
||||
{"type":"pipeline","mode":3,"round":1,"positions":4096,"classes":4096,"active_positions":118,"jobs":2960,"matches":0,"host_ms":21.575071,"build_ms":1.05058,"root_ms":17.458508,"expand_ms":0.716758,"search_ms":2.349129}
|
||||
{"type":"pipeline","mode":0,"round":1,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":33.127423,"build_ms":0.307347,"root_ms":24.924514,"expand_ms":0.000192,"search_ms":7.895082}
|
||||
{"type":"pipeline","mode":2,"round":2,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":20.589721,"build_ms":1.355912,"root_ms":17.315675,"expand_ms":0.630137,"search_ms":1.287517}
|
||||
{"type":"pipeline","mode":3,"round":2,"positions":4096,"classes":4096,"active_positions":118,"jobs":2960,"matches":0,"host_ms":19.501921,"build_ms":1.103531,"root_ms":16.471047,"expand_ms":0.668987,"search_ms":1.257972}
|
||||
{"type":"pipeline","mode":0,"round":2,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":24.022332,"build_ms":0.288545,"root_ms":16.179721,"expand_ms":0.000191,"search_ms":7.553683}
|
||||
{"type":"pipeline","mode":1,"round":2,"positions":4096,"classes":4096,"active_positions":4096,"jobs":2960,"matches":0,"host_ms":18.798496,"build_ms":0.350033,"root_ms":16.509706,"expand_ms":0.000191,"search_ms":1.938279}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[]}
|
||||
|
|
@ -0,0 +1,6 @@
|
|||
{"type":"setup","host_ms":200.876289}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":152.132767,"build_ms":4.096027,"root_ms":113.387564,"expand_ms":0.000184,"search_ms":34.648623}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":32768,"classes":32768,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":124.087013,"build_ms":4.152926,"root_ms":113.998689,"expand_ms":0.000184,"search_ms":5.934844}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":32768,"classes":512,"active_positions":32768,"jobs":15424,"matches":0,"host_ms":13.605316,"build_ms":5.613835,"root_ms":1.992268,"expand_ms":0.604117,"search_ms":5.394911}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":32768,"classes":512,"active_positions":1216,"jobs":15424,"matches":0,"host_ms":9.609161,"build_ms":1.342779,"root_ms":2.238581,"expand_ms":0.603472,"search_ms":5.424052}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[]}
|
||||
|
|
@ -0,0 +1,6 @@
|
|||
{"type":"setup","host_ms":175.300766}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":21,"classes":21,"active_positions":21,"jobs":474,"matches":24,"host_ms":1.476311,"build_ms":0.003135,"root_ms":0.17549,"expand_ms":0,"search_ms":1.297501}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":21,"classes":21,"active_positions":21,"jobs":474,"matches":24,"host_ms":0.505168,"build_ms":0.004611,"root_ms":0.175213,"expand_ms":9.3e-05,"search_ms":0.325159}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":21,"classes":20,"active_positions":21,"jobs":474,"matches":24,"host_ms":0.834753,"build_ms":0.011158,"root_ms":0.242071,"expand_ms":0.210348,"search_ms":0.370991}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":21,"classes":20,"active_positions":13,"jobs":474,"matches":24,"host_ms":0.782097,"build_ms":0.004335,"root_ms":0.1967,"expand_ms":0.261529,"search_ms":0.319533}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[[10358701894766,51069,419124],[15959335054446,51069,419124],[18914276748398,51069,419124],[24514909908078,51069,419124],[27469851602030,51069,419124],[33070484761710,51069,419124],[36025426455662,51069,419124],[41626059615342,51069,419124],[44581001309294,51069,419124],[50181634468974,51069,419124],[53136576162926,51069,419124],[58737209322606,51069,419124],[61692151016558,51069,419124],[67292784176238,51069,419124],[70247725870190,51069,419124],[75848359029870,51069,419124],[78803300723822,51069,419124],[84403933883502,51069,419124],[87358875577454,51069,419124],[92959508737134,51069,419124],[95914450431086,51069,419124],[101515083590766,51069,419124],[104470025284718,51069,419124],[110070658444398,51069,419124]]}
|
||||
361
research/experiments/normal_search/pipeline_results/summary.json
Normal file
361
research/experiments/normal_search/pipeline_results/summary.json
Normal file
|
|
@ -0,0 +1,361 @@
|
|||
{
|
||||
"metadata": {
|
||||
"date": "2026-10-11T10:51:13.411084+08:00",
|
||||
"gpu": "name, driver_version\nNVIDIA GeForce RTX 3060 Laptop GPU, 596.49\n"
|
||||
},
|
||||
"known_lifts": {
|
||||
"0": {
|
||||
"host_ms": 364.022951,
|
||||
"build_ms": 0.1147,
|
||||
"root_ms": 8.7547,
|
||||
"expand_ms": 0.0001,
|
||||
"search_ms": 355.153251,
|
||||
"jobs": 138852,
|
||||
"classes": 1653,
|
||||
"positions": 1653,
|
||||
"active_positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 62.015757,
|
||||
"build_ms": 0.113,
|
||||
"root_ms": 8.742087,
|
||||
"expand_ms": 0.000191,
|
||||
"search_ms": 53.159211,
|
||||
"jobs": 138852,
|
||||
"classes": 1653,
|
||||
"positions": 1653,
|
||||
"active_positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 55.3725,
|
||||
"build_ms": 0.133,
|
||||
"root_ms": 0.32588,
|
||||
"expand_ms": 0.297156,
|
||||
"search_ms": 54.5471,
|
||||
"jobs": 138852,
|
||||
"classes": 1,
|
||||
"positions": 1653,
|
||||
"active_positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"3": {
|
||||
"host_ms": 56.8665,
|
||||
"build_ms": 0.0101,
|
||||
"root_ms": 0.2025,
|
||||
"expand_ms": 0.6781,
|
||||
"search_ms": 56.0187,
|
||||
"jobs": 138852,
|
||||
"classes": 1,
|
||||
"positions": 1653,
|
||||
"active_positions": 1653,
|
||||
"matches": 24
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": [
|
||||
[
|
||||
10358701894766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
15959335054446,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
18914276748398,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
24514909908078,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
27469851602030,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
33070484761710,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
36025426455662,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
41626059615342,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
44581001309294,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
50181634468974,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
53136576162926,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
58737209322606,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
61692151016558,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
67292784176238,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
70247725870190,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
75848359029870,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
78803300723822,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
84403933883502,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
87358875577454,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
92959508737134,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
95914450431086,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
101515083590766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
104470025284718,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
110070658444398,
|
||||
51069,
|
||||
419124
|
||||
]
|
||||
]
|
||||
},
|
||||
"setup_ms": 220.229718
|
||||
},
|
||||
"nonempty32": {
|
||||
"0": {
|
||||
"host_ms": 4691.310731,
|
||||
"build_ms": 7.490479,
|
||||
"root_ms": 255.097691,
|
||||
"expand_ms": 0.000286,
|
||||
"search_ms": 4428.995719,
|
||||
"jobs": 1693228,
|
||||
"classes": 52213,
|
||||
"positions": 52213,
|
||||
"active_positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 915.445942,
|
||||
"build_ms": 7.505432,
|
||||
"root_ms": 245.545208,
|
||||
"expand_ms": 0.000183,
|
||||
"search_ms": 662.365642,
|
||||
"jobs": 1693228,
|
||||
"classes": 52213,
|
||||
"positions": 52213,
|
||||
"active_positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 656.710025,
|
||||
"build_ms": 7.652735,
|
||||
"root_ms": 0.576767,
|
||||
"expand_ms": 1.326998,
|
||||
"search_ms": 645.168458,
|
||||
"jobs": 1693228,
|
||||
"classes": 32,
|
||||
"positions": 52213,
|
||||
"active_positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"3": {
|
||||
"host_ms": 676.742292,
|
||||
"build_ms": 0.3428,
|
||||
"root_ms": 0.504258,
|
||||
"expand_ms": 8.844275,
|
||||
"search_ms": 666.607392,
|
||||
"jobs": 1693228,
|
||||
"classes": 32,
|
||||
"positions": 52213,
|
||||
"active_positions": 52213,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 219.927973
|
||||
},
|
||||
"mixed512": {
|
||||
"0": {
|
||||
"host_ms": 172.222139,
|
||||
"build_ms": 4.55245,
|
||||
"root_ms": 126.737258,
|
||||
"expand_ms": 0.000192,
|
||||
"search_ms": 41.239203,
|
||||
"jobs": 15424,
|
||||
"classes": 32768,
|
||||
"positions": 32768,
|
||||
"active_positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 141.138417,
|
||||
"build_ms": 4.616912,
|
||||
"root_ms": 130.393185,
|
||||
"expand_ms": 0.000192,
|
||||
"search_ms": 6.145971,
|
||||
"jobs": 15424,
|
||||
"classes": 32768,
|
||||
"positions": 32768,
|
||||
"active_positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 14.235207,
|
||||
"build_ms": 5.280144,
|
||||
"root_ms": 2.227974,
|
||||
"expand_ms": 0.638962,
|
||||
"search_ms": 6.216476,
|
||||
"jobs": 15424,
|
||||
"classes": 512,
|
||||
"positions": 32768,
|
||||
"active_positions": 32768,
|
||||
"matches": 0
|
||||
},
|
||||
"3": {
|
||||
"host_ms": 9.585776,
|
||||
"build_ms": 0.416318,
|
||||
"root_ms": 2.04888,
|
||||
"expand_ms": 0.690187,
|
||||
"search_ms": 6.774668,
|
||||
"jobs": 15424,
|
||||
"classes": 512,
|
||||
"positions": 32768,
|
||||
"active_positions": 1216,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 242.015308
|
||||
},
|
||||
"random4096": {
|
||||
"0": {
|
||||
"host_ms": 24.022332,
|
||||
"build_ms": 0.29315,
|
||||
"root_ms": 16.179721,
|
||||
"expand_ms": 0.000192,
|
||||
"search_ms": 7.553683,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"active_positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"1": {
|
||||
"host_ms": 17.902261,
|
||||
"build_ms": 0.316459,
|
||||
"root_ms": 16.411382,
|
||||
"expand_ms": 0.000192,
|
||||
"search_ms": 1.228427,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"active_positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"2": {
|
||||
"host_ms": 19.238988,
|
||||
"build_ms": 1.497691,
|
||||
"root_ms": 15.909497,
|
||||
"expand_ms": 0.630137,
|
||||
"search_ms": 1.193798,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"active_positions": 4096,
|
||||
"matches": 0
|
||||
},
|
||||
"3": {
|
||||
"host_ms": 19.501921,
|
||||
"build_ms": 1.103531,
|
||||
"root_ms": 16.471047,
|
||||
"expand_ms": 0.668987,
|
||||
"search_ms": 1.281474,
|
||||
"jobs": 2960,
|
||||
"classes": 4096,
|
||||
"positions": 4096,
|
||||
"active_positions": 118,
|
||||
"matches": 0
|
||||
},
|
||||
"validation": {
|
||||
"type": "validation",
|
||||
"complete_roots_equal": true,
|
||||
"complete_matches_equal": true,
|
||||
"matches": []
|
||||
},
|
||||
"setup_ms": 216.555239
|
||||
},
|
||||
"source_sha256": {
|
||||
"pipeline_bench.cu": "31da815aa0ee8904c702e55d2d685282fc64bfada24dd208e151faac5c55c37c",
|
||||
"cuda_bench.cu": "3e1fd6fe3bdb87ac68fe853dc4e6d2275d6aa00a692c7d630391fd12c0105c20",
|
||||
"grouped.hpp": "5d48c4a45e7f7b07a362c69b5163f48d44bd45d18e7f0f02b00bda0edaadeb23",
|
||||
"run_pipeline.py": "ce3f9e01fad8978541c581233bfeaa4023acade24bbc736f479be6ac14e88591"
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,6 @@
|
|||
{"type":"setup","host_ms":197.591869}
|
||||
{"type":"pipeline","mode":0,"round":0,"positions":1653,"classes":1653,"active_positions":1653,"jobs":0,"matches":0,"host_ms":5.571692,"build_ms":0.090188,"root_ms":5.379603,"expand_ms":0,"search_ms":0.101716}
|
||||
{"type":"pipeline","mode":1,"round":0,"positions":1653,"classes":1653,"active_positions":1653,"jobs":0,"matches":0,"host_ms":5.729661,"build_ms":0.1019,"root_ms":5.497826,"expand_ms":0,"search_ms":0.12975}
|
||||
{"type":"pipeline","mode":2,"round":0,"positions":1653,"classes":1,"active_positions":1653,"jobs":0,"matches":0,"host_ms":0.56059,"build_ms":0.10771,"root_ms":0.251016,"expand_ms":0.131594,"search_ms":0.070177}
|
||||
{"type":"pipeline","mode":3,"round":0,"positions":1653,"classes":1,"active_positions":0,"jobs":0,"matches":0,"host_ms":0.230821,"build_ms":0.006363,"root_ms":0.083272,"expand_ms":0.071838,"search_ms":0.069348}
|
||||
{"type":"validation","complete_roots_equal":true,"complete_matches_equal":true,"matches":[]}
|
||||
|
|
@ -0,0 +1 @@
|
|||
{"q_checks":5000000,"group_masks":888128,"partial_filter_survivors":159120,"known_filter_survivors":36,"passed":true}
|
||||
40
research/experiments/normal_search/results/fresh65536.jsonl
Normal file
40
research/experiments/normal_search/results/fresh65536.jsonl
Normal file
File diff suppressed because one or more lines are too long
40
research/experiments/normal_search/results/known.jsonl
Normal file
40
research/experiments/normal_search/results/known.jsonl
Normal file
|
|
@ -0,0 +1,40 @@
|
|||
{"type":"device","name":"NVIDIA GeForce RTX 3060 Laptop GPU","positions":1}
|
||||
{"type":"search_timing","mode":3,"round":0,"kernel_ms":0.3235839903,"scan_ms":0.46062,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":0,"kernel_ms":0.2693119943,"scan_ms":0.393283,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":0,"kernel_ms":0.2897920012,"scan_ms":0.44485,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":0,"kernel_ms":0.2846719921,"scan_ms":0.431015,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":0,"kernel_ms":0.1566720009,"scan_ms":0.292181,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":1,"kernel_ms":0.2682879865,"scan_ms":0.380609,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":1,"kernel_ms":0.2959359884,"scan_ms":0.421824,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":1,"kernel_ms":0.2887679935,"scan_ms":0.400152,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":1,"kernel_ms":0.1556479931,"scan_ms":0.275249,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":1,"kernel_ms":0.3235839903,"scan_ms":0.449784,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":2,"kernel_ms":0.2938880026,"scan_ms":0.40615,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":2,"kernel_ms":0.2897920012,"scan_ms":0.40199,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":2,"kernel_ms":0.1566720009,"scan_ms":0.269735,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":2,"kernel_ms":0.3225600123,"scan_ms":0.437013,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":2,"kernel_ms":0.2693119943,"scan_ms":0.3779,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":3,"kernel_ms":0.2897920012,"scan_ms":0.411375,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":3,"kernel_ms":0.1556479931,"scan_ms":0.269929,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":3,"kernel_ms":0.3194879889,"scan_ms":0.427629,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":3,"kernel_ms":0.2682879865,"scan_ms":0.375288,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":3,"kernel_ms":0.2969599962,"scan_ms":0.414665,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":4,"kernel_ms":0.1566720009,"scan_ms":0.268187,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":4,"kernel_ms":0.3235839903,"scan_ms":0.426952,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":4,"kernel_ms":0.2682879865,"scan_ms":0.425307,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":4,"kernel_ms":0.2949120104,"scan_ms":0.415051,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":4,"kernel_ms":0.2897920012,"scan_ms":0.407022,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":5,"kernel_ms":0.3235839903,"scan_ms":0.427339,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":5,"kernel_ms":0.2631680071,"scan_ms":0.37432,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":5,"kernel_ms":0.2949120104,"scan_ms":0.403248,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":5,"kernel_ms":0.2846719921,"scan_ms":0.436142,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":5,"kernel_ms":0.1505279988,"scan_ms":0.388832,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":6,"kernel_ms":0.4679679871,"scan_ms":0.896375,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":6,"kernel_ms":0.5027840137,"scan_ms":0.714295,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":6,"kernel_ms":0.4864000082,"scan_ms":0.691172,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":6,"kernel_ms":0.1587200016,"scan_ms":0.454428,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":6,"kernel_ms":0.3205119967,"scan_ms":0.426854,"matches":24}
|
||||
{"type":"search_info","root_ms":1.090559959,"jobs":84,"lower35":11010048,"matches":[[10358701894766,51069,419124],[15959335054446,51069,419124],[18914276748398,51069,419124],[24514909908078,51069,419124],[27469851602030,51069,419124],[33070484761710,51069,419124],[36025426455662,51069,419124],[41626059615342,51069,419124],[44581001309294,51069,419124],[50181634468974,51069,419124],[53136576162926,51069,419124],[58737209322606,51069,419124],[61692151016558,51069,419124],[67292784176238,51069,419124],[70247725870190,51069,419124],[75848359029870,51069,419124],[78803300723822,51069,419124],[84403933883502,51069,419124],[87358875577454,51069,419124],[92959508737134,51069,419124],[95914450431086,51069,419124],[101515083590766,51069,419124],[104470025284718,51069,419124],[110070658444398,51069,419124]]}
|
||||
{"type":"periodic_validation","classes":1,"positions":1,"equal":true}
|
||||
{"type":"root_counts","counts":[84]}
|
||||
{"type":"process","ms":234.661262}
|
||||
40
research/experiments/normal_search/results/random4096.jsonl
Normal file
40
research/experiments/normal_search/results/random4096.jsonl
Normal file
File diff suppressed because one or more lines are too long
40
research/experiments/normal_search/results/random65536.jsonl
Normal file
40
research/experiments/normal_search/results/random65536.jsonl
Normal file
File diff suppressed because one or more lines are too long
40
research/experiments/normal_search/results/stress.jsonl
Normal file
40
research/experiments/normal_search/results/stress.jsonl
Normal file
|
|
@ -0,0 +1,40 @@
|
|||
{"type":"device","name":"NVIDIA GeForce RTX 3060 Laptop GPU","positions":21}
|
||||
{"type":"search_timing","mode":3,"round":0,"kernel_ms":1.288192034,"scan_ms":1.497474,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":0,"kernel_ms":0.9256960154,"scan_ms":1.163402,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":0,"kernel_ms":1.083392024,"scan_ms":1.280467,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":0,"kernel_ms":1.010687947,"scan_ms":1.163015,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":0,"kernel_ms":0.224255994,"scan_ms":0.397249,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":1,"kernel_ms":0.922623992,"scan_ms":1.157113,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":1,"kernel_ms":2.477056026,"scan_ms":2.639978,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":1,"kernel_ms":1.007616043,"scan_ms":1.288401,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":1,"kernel_ms":0.2191359997,"scan_ms":0.471262,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":1,"kernel_ms":1.288192034,"scan_ms":1.444263,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":2,"kernel_ms":1.078271985,"scan_ms":1.256764,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":2,"kernel_ms":1.006592035,"scan_ms":1.21168,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":2,"kernel_ms":0.2232320011,"scan_ms":0.403248,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":2,"kernel_ms":1.288192034,"scan_ms":1.439328,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":2,"kernel_ms":0.9277439713,"scan_ms":1.074297,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":3,"kernel_ms":1.010687947,"scan_ms":1.17298,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":3,"kernel_ms":0.2191359997,"scan_ms":0.391735,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":3,"kernel_ms":1.28512001,"scan_ms":1.408369,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":3,"kernel_ms":0.922623992,"scan_ms":1.240607,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":3,"kernel_ms":1.338368058,"scan_ms":1.72406,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":4,"kernel_ms":0.2273280025,"scan_ms":0.686431,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":4,"kernel_ms":1.284096003,"scan_ms":1.587644,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":4,"kernel_ms":0.9267200232,"scan_ms":1.158467,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":4,"kernel_ms":1.08544004,"scan_ms":1.29943,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":4,"kernel_ms":1.010687947,"scan_ms":1.240704,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":5,"kernel_ms":1.289216042,"scan_ms":1.494572,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":5,"kernel_ms":0.9267200232,"scan_ms":1.150921,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":5,"kernel_ms":1.084416032,"scan_ms":1.289078,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":5,"kernel_ms":1.010687947,"scan_ms":1.221837,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":5,"kernel_ms":0.2201599926,"scan_ms":0.470488,"matches":24}
|
||||
{"type":"search_timing","mode":4,"round":6,"kernel_ms":0.9277439713,"scan_ms":1.136892,"matches":24}
|
||||
{"type":"search_timing","mode":5,"round":6,"kernel_ms":1.083392024,"scan_ms":1.321102,"matches":24}
|
||||
{"type":"search_timing","mode":6,"round":6,"kernel_ms":1.781759977,"scan_ms":1.983442,"matches":24}
|
||||
{"type":"search_timing","mode":7,"round":6,"kernel_ms":0.2252800018,"scan_ms":0.355744,"matches":24}
|
||||
{"type":"search_timing","mode":3,"round":6,"kernel_ms":1.93740797,"scan_ms":2.027463,"matches":24}
|
||||
{"type":"search_info","root_ms":1.161216021,"jobs":474,"lower35":62128128,"matches":[[10358701894766,51069,419124],[15959335054446,51069,419124],[18914276748398,51069,419124],[24514909908078,51069,419124],[27469851602030,51069,419124],[33070484761710,51069,419124],[36025426455662,51069,419124],[41626059615342,51069,419124],[44581001309294,51069,419124],[50181634468974,51069,419124],[53136576162926,51069,419124],[58737209322606,51069,419124],[61692151016558,51069,419124],[67292784176238,51069,419124],[70247725870190,51069,419124],[75848359029870,51069,419124],[78803300723822,51069,419124],[84403933883502,51069,419124],[87358875577454,51069,419124],[92959508737134,51069,419124],[95914450431086,51069,419124],[101515083590766,51069,419124],[104470025284718,51069,419124],[110070658444398,51069,419124]]}
|
||||
{"type":"periodic_validation","classes":20,"positions":21,"equal":true}
|
||||
{"type":"root_counts","counts":[84,152,16,0,2,20,61,22,0,1,35,30,20,29,0,2,0,0,0,0,0]}
|
||||
{"type":"process","ms":276.458389}
|
||||
331
research/experiments/normal_search/results/summary.json
Normal file
331
research/experiments/normal_search/results/summary.json
Normal file
|
|
@ -0,0 +1,331 @@
|
|||
{
|
||||
"metadata": {
|
||||
"date": "2026-10-11T09:46:12.120634+08:00",
|
||||
"cpu_validation": {
|
||||
"q_checks": 5000000,
|
||||
"group_masks": 888128,
|
||||
"partial_filter_survivors": 159120,
|
||||
"known_filter_survivors": 36,
|
||||
"passed": true
|
||||
},
|
||||
"gpu": "name, driver_version\nNVIDIA GeForce RTX 3060 Laptop GPU, 596.49\n"
|
||||
},
|
||||
"known": {
|
||||
"3": 0.3235839903,
|
||||
"4": 0.2682879865,
|
||||
"5": 0.2949120104,
|
||||
"6": 0.2897920012,
|
||||
"7": 0.1566720009,
|
||||
"info": {
|
||||
"type": "search_info",
|
||||
"root_ms": 1.090559959,
|
||||
"jobs": 84,
|
||||
"lower35": 11010048,
|
||||
"matches": [
|
||||
[
|
||||
10358701894766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
15959335054446,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
18914276748398,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
24514909908078,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
27469851602030,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
33070484761710,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
36025426455662,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
41626059615342,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
44581001309294,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
50181634468974,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
53136576162926,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
58737209322606,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
61692151016558,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
67292784176238,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
70247725870190,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
75848359029870,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
78803300723822,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
84403933883502,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
87358875577454,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
92959508737134,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
95914450431086,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
101515083590766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
104470025284718,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
110070658444398,
|
||||
51069,
|
||||
419124
|
||||
]
|
||||
]
|
||||
}
|
||||
},
|
||||
"stress": {
|
||||
"3": 1.288192034,
|
||||
"4": 0.9267200232,
|
||||
"5": 1.084416032,
|
||||
"6": 1.010687947,
|
||||
"7": 0.2232320011,
|
||||
"info": {
|
||||
"type": "search_info",
|
||||
"root_ms": 1.161216021,
|
||||
"jobs": 474,
|
||||
"lower35": 62128128,
|
||||
"matches": [
|
||||
[
|
||||
10358701894766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
15959335054446,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
18914276748398,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
24514909908078,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
27469851602030,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
33070484761710,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
36025426455662,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
41626059615342,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
44581001309294,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
50181634468974,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
53136576162926,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
58737209322606,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
61692151016558,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
67292784176238,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
70247725870190,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
75848359029870,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
78803300723822,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
84403933883502,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
87358875577454,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
92959508737134,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
95914450431086,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
101515083590766,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
104470025284718,
|
||||
51069,
|
||||
419124
|
||||
],
|
||||
[
|
||||
110070658444398,
|
||||
51069,
|
||||
419124
|
||||
]
|
||||
]
|
||||
}
|
||||
},
|
||||
"random4096": {
|
||||
"3": 8.857600212,
|
||||
"4": 5.859327793,
|
||||
"5": 6.7450881,
|
||||
"6": 5.53574419,
|
||||
"7": 1.141759992,
|
||||
"info": {
|
||||
"type": "search_info",
|
||||
"root_ms": 19.01977539,
|
||||
"jobs": 2960,
|
||||
"lower35": 387973120,
|
||||
"matches": []
|
||||
}
|
||||
},
|
||||
"random65536": {
|
||||
"3": 130.1360626,
|
||||
"4": 98.65420532,
|
||||
"5": 114.5487366,
|
||||
"6": 108.295166,
|
||||
"7": 19.98233604,
|
||||
"info": {
|
||||
"type": "search_info",
|
||||
"root_ms": 267.9224243,
|
||||
"jobs": 45356,
|
||||
"lower35": 5944901632,
|
||||
"matches": []
|
||||
}
|
||||
},
|
||||
"fresh65536": {
|
||||
"3": 134.184967,
|
||||
"4": 102.5822754,
|
||||
"5": 118.9703674,
|
||||
"6": 112.1454086,
|
||||
"7": 20.44006348,
|
||||
"info": {
|
||||
"type": "search_info",
|
||||
"root_ms": 277.3555298,
|
||||
"jobs": 46148,
|
||||
"lower35": 6048710656,
|
||||
"matches": []
|
||||
}
|
||||
},
|
||||
"source_sha256": {
|
||||
"cuda_bench.cu": "f762ea050d7c93c00903928f972e41106e917d37a4caceb79630bacd0f9316e1",
|
||||
"grouped.hpp": "5d48c4a45e7f7b07a362c69b5163f48d44bd45d18e7f0f02b00bda0edaadeb23",
|
||||
"cpu_verify.cpp": "2e2b6799bd624b3e0a3022c2144c4a6f665c08de530d8f2846e6bc748ba6412f",
|
||||
"run.py": "daca9d7de3463fd501a99b10d921974b73c206acc6e1438a246ae5729043c15e"
|
||||
}
|
||||
}
|
||||
28
research/experiments/normal_search/run.py
Normal file
28
research/experiments/normal_search/run.py
Normal file
|
|
@ -0,0 +1,28 @@
|
|||
#!/usr/bin/env python3
|
||||
import subprocess,json,statistics,hashlib,random,datetime
|
||||
from pathlib import Path
|
||||
HERE=Path(__file__).resolve().parent
|
||||
POS=HERE.parent/'full_search_feasibility'/'positions'
|
||||
OUT=HERE/'results'
|
||||
OUT.mkdir(exist_ok=True)
|
||||
subprocess.run(['nvcc','-O3','-std=c++17','-arch=sm_86',str(HERE/'cuda_bench.cu'),'-o','/tmp/slime_normal_cuda'],check=True)
|
||||
subprocess.run(['g++','-O3','-std=c++17','-Wall','-Wextra',str(HERE/'cpu_verify.cpp'),'-o','/tmp/slime_normal_verify'],check=True)
|
||||
validation=subprocess.check_output(['/tmp/slime_normal_verify'],text=True)
|
||||
(OUT/'cpu_verify.json').write_text(validation)
|
||||
rng=random.Random(2026101104)
|
||||
fresh=HERE/'fresh65536.txt'
|
||||
fresh.write_text(''.join(f'{rng.randrange(-1875000,1875000)} {rng.randrange(-1875000,1875000)}\n' for _ in range(65536)))
|
||||
summary={'metadata':{'date':datetime.datetime.now().astimezone().isoformat(),'cpu_validation':json.loads(validation),'gpu':subprocess.check_output(['nvidia-smi','--query-gpu=name,driver_version','--format=csv'],text=True)}}
|
||||
for name in ['known','stress','random4096','random65536','fresh65536']:
|
||||
cmd=['/tmp/slime_normal_cuda',str(fresh if name=='fresh65536' else POS/(name+'.txt')),'normal','7']
|
||||
r=subprocess.run(cmd,capture_output=True,text=True,timeout=300)
|
||||
(OUT/(name+'.jsonl')).write_text(r.stdout)
|
||||
(OUT/(name+'.stderr')).write_text(r.stderr)
|
||||
if r.returncode: raise RuntimeError(r.stderr)
|
||||
rows=[json.loads(s) for s in r.stdout.splitlines()]
|
||||
summary[name]={str(m):statistics.median(v['kernel_ms'] for v in rows if v['type']=='search_timing' and v['mode']==m) for m in [3,4,5,6,7]}
|
||||
summary[name]['info']=next(v for v in rows if v['type']=='search_info')
|
||||
print(name,json.dumps(summary[name]),flush=True)
|
||||
summary['source_sha256']={name:hashlib.sha256((HERE/name).read_bytes()).hexdigest() for name in ['cuda_bench.cu','grouped.hpp','cpu_verify.cpp','run.py']}
|
||||
summary['shared_headers_sha256']={name:hashlib.sha256((HERE.parents[2]/'include'/'slime'/name).read_bytes()).hexdigest() for name in ['solver_core.hpp','upper_interval.hpp']}
|
||||
(OUT/'summary.json').write_text(json.dumps(summary,indent=2))
|
||||
44
research/experiments/normal_search/run_pipeline.py
Normal file
44
research/experiments/normal_search/run_pipeline.py
Normal file
|
|
@ -0,0 +1,44 @@
|
|||
#!/usr/bin/env python3
|
||||
import subprocess,json,random,statistics,hashlib,datetime
|
||||
from pathlib import Path
|
||||
HERE=Path(__file__).resolve().parent
|
||||
OLD=HERE.parent/'full_search_feasibility'
|
||||
OUT=HERE/'pipeline_results'
|
||||
POS=HERE/'pipeline_positions'
|
||||
OUT.mkdir(exist_ok=True);POS.mkdir(exist_ok=True)
|
||||
def lifts(x,z):
|
||||
return [(xx,zz) for xx in range(-1875000+(x+1875000)%65536,1875000,65536)
|
||||
for zz in range(-1875000+(z+1875000)%131072,1875000,131072)]
|
||||
def save(name,coords):
|
||||
p=POS/(name+'.txt');p.write_text(''.join(f'{x} {z}\n' for x,z in coords));return p
|
||||
known=save('known_lifts',lifts(51069,419124))
|
||||
rng=random.Random(2026101105)
|
||||
coords=[]
|
||||
classes=set()
|
||||
while len(classes)<512:
|
||||
c=(rng.randrange(65536),rng.randrange(131072))
|
||||
if c in classes:continue
|
||||
classes.add(c)
|
||||
coords.extend(rng.sample(lifts(*c),64))
|
||||
mixed=save('mixed512',coords)
|
||||
cases={'known_lifts':known,'nonempty32':OLD/'positions/lifts.txt','mixed512':mixed,'random4096':OLD/'positions/random4096.txt'}
|
||||
subprocess.run(['nvcc','-O3','-std=c++17','-arch=sm_86',str(HERE/'pipeline_bench.cu'),'-o','/tmp/slime_pipeline_cuda'],check=True)
|
||||
summary={'metadata':{'date':datetime.datetime.now().astimezone().isoformat(),'gpu':subprocess.check_output(['nvidia-smi','--query-gpu=name,driver_version','--format=csv'],text=True)}}
|
||||
for name,path in cases.items():
|
||||
cmd=['/tmp/slime_pipeline_cuda',str(path),'3']
|
||||
p=subprocess.run(cmd,capture_output=True,text=True,timeout=300)
|
||||
(OUT/(name+'.jsonl')).write_text(p.stdout);(OUT/(name+'.stderr')).write_text(p.stderr)
|
||||
if p.returncode:raise RuntimeError(p.stderr)
|
||||
rows=[json.loads(l) for l in p.stdout.splitlines()]
|
||||
d={}
|
||||
for m in range(4):
|
||||
a=[r for r in rows if r['type']=='pipeline' and r['mode']==m]
|
||||
d[str(m)]={k:statistics.median(r[k] for r in a) for k in ['host_ms','build_ms','root_ms','expand_ms','search_ms','jobs','classes','positions','active_positions','matches']}
|
||||
d['validation']=next(r for r in rows if r['type']=='validation')
|
||||
d['setup_ms']=next(r['host_ms'] for r in rows if r['type']=='setup')
|
||||
summary[name]=d
|
||||
(OUT/'summary.json').write_text(json.dumps(summary,indent=2))
|
||||
print(name,json.dumps({k:v for k,v in d.items() if k!='validation'}),flush=True)
|
||||
summary['source_sha256']={n:hashlib.sha256((HERE/n).read_bytes()).hexdigest() for n in ['pipeline_bench.cu','cuda_bench.cu','grouped.hpp','run_pipeline.py']}
|
||||
summary['shared_headers_sha256']={name:hashlib.sha256((HERE.parents[2]/'include'/'slime'/name).read_bytes()).hexdigest() for name in ['solver_core.hpp','upper_interval.hpp']}
|
||||
(OUT/'summary.json').write_text(json.dumps(summary,indent=2))
|
||||
28
research/experiments/normal_search/verify_pipeline_edges.py
Normal file
28
research/experiments/normal_search/verify_pipeline_edges.py
Normal file
|
|
@ -0,0 +1,28 @@
|
|||
#!/usr/bin/env python3
|
||||
import json,subprocess,random,hashlib
|
||||
from pathlib import Path
|
||||
H=Path(__file__).resolve().parent
|
||||
OLD=H.parent/'full_search_feasibility'
|
||||
POS=H/'pipeline_positions';OUT=H/'pipeline_results'
|
||||
rows=[json.loads(l) for l in (OLD/'results/random4096.jsonl').read_text().splitlines()]
|
||||
counts=next(r['counts'] for r in rows if r['type']=='root_counts')
|
||||
points=[tuple(map(int,l.split())) for l in (OLD/'positions/random4096.txt').read_text().splitlines()]
|
||||
x,z=points[counts.index(0)]
|
||||
zero=[(xx,zz) for xx in range(-1875000+(x+1875000)%65536,1875000,65536) for zz in range(-1875000+(z+1875000)%131072,1875000,131072)]
|
||||
p=POS/'zero_lifts.txt';p.write_text(''.join(f'{x} {z}\n' for x,z in zero))
|
||||
shuffled=(POS/'mixed512.txt').read_text().splitlines();random.Random(2026101106).shuffle(shuffled)
|
||||
q=POS/'shuffled_mixed.txt';q.write_text('\n'.join(shuffled)+'\n')
|
||||
subprocess.run(['nvcc','-O3','-std=c++17','-arch=sm_86',str(H/'pipeline_bench.cu'),'-o','/tmp/slime_pipeline_cuda'],check=True)
|
||||
checks={}
|
||||
for name,path in [('zero_lifts',p),('stress',OLD/'positions/stress.txt'),('shuffled_mixed',q)]:
|
||||
r=subprocess.run(['/tmp/slime_pipeline_cuda',str(path),'1'],capture_output=True,text=True,timeout=120)
|
||||
(OUT/(name+'.jsonl')).write_text(r.stdout);(OUT/(name+'.stderr')).write_text(r.stderr)
|
||||
if r.returncode:raise RuntimeError(r.stderr)
|
||||
rows=[json.loads(l) for l in r.stdout.splitlines()]
|
||||
checks[name]={'validation':next(r for r in rows if r['type']=='validation'),'modes':[r for r in rows if r['type']=='pipeline']}
|
||||
if name=='zero_lifts':assert all(r['jobs']==0 and r['matches']==0 for r in checks[name]['modes'])
|
||||
if name=='stress':assert len(checks[name]['validation']['matches'])==24
|
||||
print(name,'passed',flush=True)
|
||||
checks['source_sha256']={n:hashlib.sha256((H/n).read_bytes()).hexdigest() for n in ['pipeline_bench.cu','cuda_bench.cu','grouped.hpp','verify_pipeline_edges.py']}
|
||||
checks['shared_headers_sha256']={name:hashlib.sha256((H.parents[2]/'include'/'slime'/name).read_bytes()).hexdigest() for name in ['solver_core.hpp','upper_interval.hpp']}
|
||||
(OUT/'edge_validation.json').write_text(json.dumps(checks,indent=2))
|
||||
Loading…
Add table
Add a link
Reference in a new issue