多PE scan：消息里到底应该传什么

// Original CPU teaching model; not a device simulator or benchmark.
#include <iostream>
#include <vector>
#include <stdexcept>

void require(bool ok) { if (!ok) throw std::runtime_error("model check failed"); }

int main() {
 const std::vector<std::vector<int>> parts{{3,-1},{4},{2,2}};
 std::vector<int> out,flat;
 int offset=0;
 for(const auto& part:parts) {
  int local=0;
  for(int value:part){local+=value;out.push_back(offset+local);flat.push_back(value);}
  offset+=local;
 }
 std::vector<int> oracle;int sum=0;
 for(int value:flat){sum+=value;oracle.push_back(sum);}
 require(out==oracle && out==std::vector<int>({3,2,6,8,10}));
 std::cout<<"scan=";
 for(std::size_t i=0;i<out.size();++i) std::cout<<(i?" ":"")<<out[i];
 std::cout<<'\n';
}

// Original CPU teaching model; not a device simulator or benchmark.
#include <iostream>
#include <vector>
#include <stdexcept>

void require(bool ok) { if (!ok) throw std::runtime_error("model check failed"); }

std::vector<int> prefix(const std::vector<int>& a) {
 std::vector<int> out;int sum=0;for(int x:a){sum+=x;out.push_back(sum);}return out;
}
int main() {
 const auto a=prefix({2,4,4}),b=prefix({4,2,4});
 require(a.back()==b.back() && a!=b);
 std::cout<<"same_total="<<(a.back()==b.back())<<" same_scan="<<(a==b)<<'\n';
}