portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 09:15:38 +01:00
Code Releases Activity

Hadrons: DiskVector Eigen specialisation with binary I/O and sha256 correctness check

Browse Source
This commit is contained in:
Antonin Portelli 2018-10-08 19:02:00 +01:00
parent 936eaac8e1
commit efc0c65056
2 changed files with 113 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

115
Hadrons/DiskVector.hpp
View File

@ -34,6 +34,12 @@ See the full license in the file "LICENSE" in the top level distribution directo
#include <ftw.h> #include <ftw.h>
#include <unistd.h> #include <unistd.h>
#ifdef DV_DEBUG
#define DV_DEBUG_MSG(dv, stream) LOG(Debug) << "diskvector " << (dv) << ": " << stream << std::endl
#else
#define DV_DEBUG_MSG(dv, stream)
#endif
BEGIN_HADRONS_NAMESPACE BEGIN_HADRONS_NAMESPACE
/****************************************************************************** /******************************************************************************
@ -56,9 +62,7 @@ public:
// write to disk and cache // write to disk and cache
T &operator=(const T &obj) const T &operator=(const T &obj) const
{ {
#ifdef DV_DEBUG DV_DEBUG_MSG(&master_, "writing to " << i_);
LOG(Debug) << "diskvector " << &master_ << ": writing to " << i_ << std::endl;
#endif
master_.cacheInsert(i_, obj); master_.cacheInsert(i_, obj);
master_.save(master_.filename(i_), obj); master_.save(master_.filename(i_), obj);
@ -82,6 +86,8 @@ public:
virtual ~DiskVectorBase(void); virtual ~DiskVectorBase(void);
const T & operator[](const unsigned int i) const; const T & operator[](const unsigned int i) const;
RwAccessHelper operator[](const unsigned int i); RwAccessHelper operator[](const unsigned int i);
double hitRatio(void) const;
void resetStat(void);
private: private:
virtual void load(T &obj, const std::string filename) const = 0; virtual void load(T &obj, const std::string filename) const = 0;
virtual void save(const std::string filename, const T &obj) const = 0; virtual void save(const std::string filename, const T &obj) const = 0;
@ -93,6 +99,7 @@ private:
private: private:
std::string dirname_; std::string dirname_;
unsigned int size_, cacheSize_; unsigned int size_, cacheSize_;
double access_{0.}, hit_{0.};
bool clean_; bool clean_;
// using pointers to allow modifications when class is const // using pointers to allow modifications when class is const
// semantic: const means data unmodified, but cache modification allowed // semantic: const means data unmodified, but cache modification allowed
@ -115,6 +122,7 @@ private:
read(reader, basename(filename), obj); read(reader, basename(filename), obj);
} }
virtual void save(const std::string filename, const T &obj) const virtual void save(const std::string filename, const T &obj) const
{ {
Writer writer(filename); Writer writer(filename);
@ -123,13 +131,70 @@ private:
} }
}; };
/******************************************************************************
* Specialisation for Eigen matrices *
******************************************************************************/
template <typename T>
using EigenDiskVectorMat = Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>;
template <typename T>
class EigenDiskVector: public DiskVectorBase<EigenDiskVectorMat<T>>
{
public:
using DiskVectorBase<EigenDiskVectorMat<T>>::DiskVectorBase;
typedef EigenDiskVectorMat<T> Matrix;
private:
virtual void load(EigenDiskVectorMat<T> &obj, const std::string filename) const
{
std::ifstream f(filename, std::ios::binary);
std::vector<unsigned char> hash(SHA256_DIGEST_LENGTH);
Eigen::Index nRow, nCol;
size_t matSize;
double t;
f.read(reinterpret_cast<char *>(hash.data()), hash.size()*sizeof(unsigned char));
f.read(reinterpret_cast<char *>(&nRow), sizeof(Eigen::Index));
f.read(reinterpret_cast<char *>(&nCol), sizeof(Eigen::Index));
obj.resize(nRow, nCol);
matSize = nRow*nCol*sizeof(T);
t = -usecond();
f.read(reinterpret_cast<char *>(obj.data()), matSize);
t += usecond();
DV_DEBUG_MSG(this, "Eigen read " << matSize/t*1.0e6/1024/1024 << " MB/s");
auto check = GridChecksum::sha256(obj.data(), matSize);
DV_DEBUG_MSG(this, "Eigen sha256 " << GridChecksum::sha256_string(check));
if (hash != check)
{
HADRONS_ERROR(Io, "checksum failed")
}
}
virtual void save(const std::string filename, const EigenDiskVectorMat<T> &obj) const
{
std::ofstream f(filename, std::ios::binary);
std::vector<unsigned char> hash(SHA256_DIGEST_LENGTH);
Eigen::Index nRow, nCol;
size_t matSize;
double t;
nRow = obj.rows();
nCol = obj.cols();
matSize = nRow*nCol*sizeof(T);
hash = GridChecksum::sha256(obj.data(), matSize);
DV_DEBUG_MSG(this, "Eigen sha256 " << GridChecksum::sha256_string(hash));
f.write(reinterpret_cast<char *>(hash.data()), hash.size()*sizeof(unsigned char));
f.write(reinterpret_cast<char *>(&nRow), sizeof(Eigen::Index));
f.write(reinterpret_cast<char *>(&nCol), sizeof(Eigen::Index));
t = -usecond();
f.write(reinterpret_cast<const char *>(obj.data()), matSize);
t += usecond();
DV_DEBUG_MSG(this, "Eigen write " << matSize/t*1.0e6/1024/1024 << " MB/s");
}
};
/****************************************************************************** /******************************************************************************
* DiskVectorBase implementation * * DiskVectorBase implementation *
******************************************************************************/ ******************************************************************************/
#ifdef DV_DEBUG
#define DV_DEBUG_MSG(stream) LOG(Debug) << "diskvector " << this << ": " << stream << std::endl
#endif
template <typename T> template <typename T>
DiskVectorBase<T>::DiskVectorBase(const std::string dirname, DiskVectorBase<T>::DiskVectorBase(const std::string dirname,
const unsigned int size, const unsigned int size,
@ -160,28 +225,29 @@ DiskVectorBase<T>::~DiskVectorBase(void)
template <typename T> template <typename T>
const T & DiskVectorBase<T>::operator[](const unsigned int i) const const T & DiskVectorBase<T>::operator[](const unsigned int i) const
{ {
auto &cache = *cachePtr_; auto &cache = *cachePtr_;
auto &loads = *loadsPtr_; auto &loads = *loadsPtr_;
DV_DEBUG_MSG("accessing " << i << " (RO)"); DV_DEBUG_MSG(this, "accessing " << i << " (RO)");
if (i >= size_) if (i >= size_)
{ {
HADRONS_ERROR(Size, "index out of range"); HADRONS_ERROR(Size, "index out of range");
} }
const_cast<double &>(access_)++;
if (cache.find(i) == cache.end()) if (cache.find(i) == cache.end())
{ {
// cache miss // cache miss
DV_DEBUG_MSG("cache miss"); DV_DEBUG_MSG(this, "cache miss");
fetch(i); fetch(i);
} }
else else
{ {
DV_DEBUG_MSG("cache hit"); DV_DEBUG_MSG(this, "cache hit");
auto pos = std::find(loads.begin(), loads.end(), i); auto pos = std::find(loads.begin(), loads.end(), i);
const_cast<double &>(hit_)++;
loads.erase(pos); loads.erase(pos);
loads.push_back(i); loads.push_back(i);
} }
@ -193,7 +259,7 @@ const T & DiskVectorBase<T>::operator[](const unsigned int i) const
{ {
msg += std::to_string(p) + " "; msg += std::to_string(p) + " ";
} }
DV_DEBUG_MSG("in cache: " << msg); DV_DEBUG_MSG(this, "in cache: " << msg);
#endif #endif
return cache.at(i); return cache.at(i);
@ -202,7 +268,7 @@ const T & DiskVectorBase<T>::operator[](const unsigned int i) const
template <typename T> template <typename T>
typename DiskVectorBase<T>::RwAccessHelper DiskVectorBase<T>::operator[](const unsigned int i) typename DiskVectorBase<T>::RwAccessHelper DiskVectorBase<T>::operator[](const unsigned int i)
{ {
DV_DEBUG_MSG("accessing " << i << " (RW)"); DV_DEBUG_MSG(this, "accessing " << i << " (RW)");
if (i >= size_) if (i >= size_)
{ {
@ -212,6 +278,19 @@ typename DiskVectorBase<T>::RwAccessHelper DiskVectorBase<T>::operator[](const u
return RwAccessHelper(*this, i); return RwAccessHelper(*this, i);
} }
template <typename T>
double DiskVectorBase<T>::hitRatio(void) const
{
return hit_/access_;
}
template <typename T>
void DiskVectorBase<T>::resetStat(void)
{
access_ = 0.;
hit_ = 0.;
}
template <typename T> template <typename T>
std::string DiskVectorBase<T>::filename(const unsigned int i) const std::string DiskVectorBase<T>::filename(const unsigned int i) const
{ {
@ -226,7 +305,7 @@ void DiskVectorBase<T>::evict(void) const
if (cache.size() >= cacheSize_) if (cache.size() >= cacheSize_)
{ {
DV_DEBUG_MSG("evicting " << loads.front()); DV_DEBUG_MSG(this, "evicting " << loads.front());
cache.erase(loads.front()); cache.erase(loads.front());
loads.pop_front(); loads.pop_front();
} }
@ -239,7 +318,7 @@ void DiskVectorBase<T>::fetch(const unsigned int i) const
auto &loads = *loadsPtr_; auto &loads = *loadsPtr_;
struct stat s; struct stat s;
DV_DEBUG_MSG("loading " << i << " from disk"); DV_DEBUG_MSG(this, "loading " << i << " from disk");
evict(); evict();
if(stat(filename(i).c_str(), &s) != 0) if(stat(filename(i).c_str(), &s) != 0)
@ -267,7 +346,7 @@ void DiskVectorBase<T>::cacheInsert(const unsigned int i, const T &obj) const
{ {
msg += std::to_string(p) + " "; msg += std::to_string(p) + " ";
} }
DV_DEBUG_MSG("in cache: " << msg); DV_DEBUG_MSG(this, "in cache: " << msg);
#endif #endif
} }

16
tests/hadrons/Test_diskvector.cc
View File

@ -91,6 +91,22 @@ int main(int argc, char *argv[])
v13r = v[13]; v13r = v[13];
LOG(Message) << "v[13] correct? " LOG(Message) << "v[13] correct? "
<< ((v13r == v13w) ? "yes" : "no" ) << std::endl; << ((v13r == v13w) ? "yes" : "no" ) << std::endl;
LOG(Message) << "hit ratio " << v.hitRatio() << std::endl;
EigenDiskVector<ComplexD> w("eigendiskvector_test", 1000, 4);
EigenDiskVector<ComplexD>::Matrix m,n;
w[2] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
m = w[2];
w[3] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
w[4] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
w[5] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
w[6] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
w[7] = EigenDiskVectorMat<ComplexD>::Random(2000, 2000);
n = w[2];
LOG(Message) << "w[2] correct? "
<< ((m == n) ? "yes" : "no" ) << std::endl;
LOG(Message) << "hit ratio " << w.hitRatio() << std::endl;
Grid_finalize(); Grid_finalize();