lars/geant4
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Import Geant4 10.4.0 source tree

Browse Source
This commit is contained in:
Gabriele Cosmo
2017-12-08 12:52:30 +01:00
parent 98e455a940
commit fc6af9e721
2166 changed files with 276760 additions and 100873 deletions
Download Patch File Download Diff File Expand all files Collapse all files
source/externals/clhep/src/MixMaxRng.cc
Vendored
+510 -72
View File
@@ -1,4 +1,4 @@
// $Id:$
//
// -*- C++ -*-
//
// -----------------------------------------------------------------------
@@ -7,22 +7,27 @@
// class implementation file
// -----------------------------------------------------------------------
//
// This file interfaces the PseudoRandom Number Generator
// proposed by N.Z. Akopov, G.K.Saviddy & N.G. Ter-Arutyunian
// "Matrix Generator of Pseudorandom Numbers"
// J. Compt. Phy. 97, 573 (1991)
// Preprint: EPI-867(18)-86, Yerevan June 1986.
// This file interfaces the MixMax PseudoRandom Number Generator
// proposed by:
//
// Implementation by Konstantin Savvidy - 2004-2015
// "The MIXMAX random number generator"
// Comp. Phys. Commun. (2015)
// http://dx.doi.org/10.1016/j.cpc.2015.06.003
// G.K.Savvidy and N.G.Ter-Arutyunian,
// On the Monte Carlo simulation of physical systems,
// J.Comput.Phys. 97, 566 (1991);
// Preprint EPI-865-16-86, Yerevan, Jan. 1986
// http://dx.doi.org/10.1016/0021-9991(91)90015-D
//
// K.Savvidy
// "The MIXMAX random number generator"
// Comp. Phys. Commun. (2015)
// http://dx.doi.org/10.1016/j.cpc.2015.06.003
//
// K.Savvidy and G.Savvidy
// "Spectrum and Entropy of C-systems. MIXMAX random number generator"
// Chaos, Solitons & Fractals, Volume 91, (2016) pp. 33-38
// http://dx.doi.org/10.1016/j.chaos.2016.05.003
//
// Release 0.99 and later: released under the LGPL license version 3.0
// =======================================================================
// CLHEP interface implemented by
// J. Apostolakis, G. Cosmo & K. Savvidy - Created: 6th July 2015
// CLHEP interface released under the LGPL license version 3.0
// Implementation by Konstantin Savvidy - Copyright 2004-2017
// =======================================================================
#include "CLHEP/Random/Random.h"
@@ -32,10 +37,6 @@
#include <string.h> // for strcmp
#include <cmath>
#include <cstdlib>
#include <stdint.h>
#include "CLHEP/Random/mixmax.h"
const unsigned long MASK32=0xffffffff;
@@ -48,20 +49,17 @@ namespace {
static const int MarkerLen = 64; // Enough room to hold a begin or end marker.
std::string MixMaxRng::name() const { return "MixMaxRng"; } // N=" + N
MixMaxRng::MixMaxRng()
: HepRandomEngine()
{
int numEngines = ++numberOfEngines;
fRngState= rng_alloc();
setSeed(static_cast<long>(numEngines));
}
MixMaxRng::MixMaxRng(long seed)
: HepRandomEngine()
{
fRngState= rng_alloc();
theSeed=seed;
setSeed(seed);
}
@@ -73,15 +71,14 @@ MixMaxRng::MixMaxRng(std::istream& is)
MixMaxRng::~MixMaxRng()
{
rng_free( fRngState );
}
MixMaxRng::MixMaxRng(const MixMaxRng& rng)
: HepRandomEngine(rng)
{
fRngState= rng_copy( rng.fRngState->V );
fRngState->sumtot= rng.fRngState->sumtot;
fRngState->counter= rng.fRngState->counter;
S.V = rng.S.V;
S.sumtot= rng.S.sumtot;
S.counter= rng.S.counter;
}
MixMaxRng& MixMaxRng::operator=(const MixMaxRng& rng)
@@ -94,12 +91,9 @@ MixMaxRng& MixMaxRng::operator=(const MixMaxRng& rng)
//
HepRandomEngine::operator=(rng);
// Copy data
//
rng_free( fRngState );
fRngState= rng_copy( rng.fRngState->V );
fRngState->sumtot= rng.fRngState->sumtot;
fRngState->counter= rng.fRngState->counter;
S.V = rng.S.V;
S.sumtot= rng.S.sumtot;
S.counter= rng.S.counter;
return *this;
}
@@ -110,40 +104,127 @@ void MixMaxRng::saveStatus( const char filename[] ) const
FILE *fh= fopen(filename, "w");
if( fh )
{
fRngState->fh= fh;
print_state(fRngState);
int j;
fprintf(fh, "mixmax state, file version 1.0\n" );
fprintf(fh, "N=%u; V[N]={", rng_get_N() );
for (j=0; (j< (rng_get_N()-1) ); j++) {
fprintf(fh, "%llu, ", S.V[j] );
}
fprintf(fh, "%llu", S.V[rng_get_N()-1] );
fprintf(fh, "}; " );
fprintf(fh, "counter=%u; ", S.counter );
fprintf(fh, "sumtot=%llu;\n", S.sumtot );
fclose(fh);
}
fRngState->fh= 0;
}
#define MIXMAX_ARRAY_INDEX_OUT_OF_BOUNDS 0xFF01
#define MIXMAX_SEED_WAS_ZERO 0xFF02
#define MIXMAX_ERROR_READING_STATE_FILE 0xFF03
#define MIXMAX_ERROR_READING_STATE_COUNTER 0xFF04
#define MIXMAX_ERROR_READING_STATE_CHECKSUM 0xFF05
void MixMaxRng::restoreStatus( const char filename[] )
{
read_state(fRngState, filename);
// a function for reading the state from a file
FILE* fin;
if( ( fin = fopen(filename, "r") ) )
{
char l=0;
while ( l != '{' ) { // 0x7B = "{"
l=fgetc(fin); // proceed until hitting opening bracket
}
ungetc(' ', fin);
}
else
{
fprintf(stderr, "mixmax -> read_state: error reading file %s\n", filename);
exit(MIXMAX_ERROR_READING_STATE_FILE);
}
myuint_t vecVal;
//printf("mixmax -> read_state: starting to read state from file\n");
if (!fscanf(fin, "%llu", &S.V[0]) )
{
fprintf(stderr, "mixmax -> read_state: error reading file %s\n", filename);
exit(MIXMAX_ERROR_READING_STATE_FILE);
}
int i;
for( i = 1; i < rng_get_N(); i++)
{
if (!fscanf(fin, ", %llu", &vecVal) )
{
fprintf(stderr, "mixmax -> read_state: error reading vector component i=%d from file %s\n", i, filename);
exit(MIXMAX_ERROR_READING_STATE_FILE);
}
if( vecVal <= MixMaxRng::M61 )
{
S.V[i] = vecVal;
}
else
{
fprintf(stderr, "mixmax -> read_state: Invalid state vector value= %llu"
" ( must be less than %llu ) "
" obtained from reading file %s\n"
, vecVal, MixMaxRng::M61, filename);
}
}
int counter;
if (!fscanf( fin, "}; counter=%i; ", &counter))
{
fprintf(stderr, "mixmax -> read_state: error reading counter from file %s\n", filename);
exit(MIXMAX_ERROR_READING_STATE_FILE);
}
if( counter <= rng_get_N() )
{
S.counter= counter;
}
else
{
fprintf(stderr, "mixmax -> read_state: Invalid counter = %d"
" Must be 0 <= counter < %u\n" , counter, rng_get_N());
print_state();
exit(MIXMAX_ERROR_READING_STATE_COUNTER);
}
precalc();
myuint_t sumtot;
if (!fscanf( fin, "sumtot=%llu\n", &sumtot))
{
fprintf(stderr, "mixmax -> read_state: error reading checksum from file %s\n", filename);
exit(MIXMAX_ERROR_READING_STATE_FILE);
}
if (S.sumtot != sumtot)
{
fprintf(stderr, "mixmax -> checksum error while reading state from file %s - corrupted?\n", filename);
exit(MIXMAX_ERROR_READING_STATE_CHECKSUM);
}
fclose(fin);
}
#undef MIXMAX_ARRAY_INDEX_OUT_OF_BOUNDS
#undef MIXMAX_SEED_WAS_ZERO
#undef MIXMAX_ERROR_READING_STATE_FILE
#undef MIXMAX_ERROR_READING_STATE_COUNTER
#undef MIXMAX_ERROR_READING_STATE_CHECKSUM
void MixMaxRng::showStatus() const
{
std::cout << std::endl;
std::cout << "------- MixMaxRng engine status -------" << std::endl;
std::cout << " Current state vector is:" << std::endl;
fRngState->fh=stdout;
print_state(fRngState);
print_state();
std::cout << "---------------------------------------" << std::endl;
}
void MixMaxRng::setSeed(long longSeed, int /* extraSeed */)
{
unsigned long seed0;
//seed_uniquestream(0,0,0,longSeed);
theSeed = longSeed;
if( sizeof(long) > 4) // C standard says long is at least 32-bits
seed0= static_cast<unsigned long>(longSeed) & MASK32 ;
else
seed0= longSeed;
seed_spbox(fRngState, seed0);
seed_spbox(longSeed);
}
// Preferred Seeding method
@@ -174,24 +255,93 @@ void MixMaxRng::setSeeds(const long* Seeds, int seedNum)
}
theSeed = Seeds[0];
theSeeds = Seeds;
seed_uniquestream(fRngState, seed3, seed2, seed1, seed0);
seed_uniquestream(seed3, seed2, seed1, seed0);
}
double MixMaxRng::flat()
std::string MixMaxRng::engineName()
{
return get_next_float(fRngState);
return "MixMaxRng";
}
constexpr int MixMaxRng::rng_get_N()
{
return N;
}
constexpr long long int MixMaxRng::rng_get_SPECIAL()
{
return SPECIAL;
}
constexpr int MixMaxRng::rng_get_SPECIALMUL()
{
return SPECIALMUL;
}
double MixMaxRng::generate(int i)
{
S.counter++;
#if defined(__clang__) || defined(__llvm__)
return INV_M61*static_cast<double>(S.V[i]);
#elif defined(__GNUC__) && (__GNUC__ < 7) && (!defined(__ICC)) && defined(__x86_64__) && defined(__SSE2_MATH__)
int64_t Z=S.V[i];
double F=0.0;
//#warning Using the inline assembler
/* using SSE inline assemly to zero the xmm register, just before int64 -> double conversion,
not necessary in GCC-5 or better, but huge penalty on earlier compilers
*/
__asm__ __volatile__( "pxor %0, %0;"
"cvtsi2sdq %1, %0;"
:"=x"(F)
:"r"(Z)
);
return F*INV_M61;
#else
//#warning other method
return convert1double(S.V[i]); //get_next_float_packbits();
#endif
}
double MixMaxRng::iterate()
{
myuint_t* Y=S.V.data();
myuint_t tempP, tempV;
Y[0] = ( tempV = S.sumtot);
myuint_t sumtot = Y[0], ovflow = 0; // will keep a running sum of all new elements
tempP = 0; // will keep a partial sum of all old elements
myuint_t tempPO;
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[1] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[1] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[2] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[2] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[3] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[3] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[4] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[4] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[5] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[5] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[6] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[6] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[7] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[7] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[8] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[8] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[9] ); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[9] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[10]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[10] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[11]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[11] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[12]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[12] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[13]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[13] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[14]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[14] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[15]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[15] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
tempPO = MULWU(tempP); tempP = modadd(tempP, Y[16]); tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); Y[16] = tempV; sumtot += tempV; if (sumtot < tempV) {ovflow++;};
S.sumtot = MIXMAX_MOD_MERSENNE(MIXMAX_MOD_MERSENNE(sumtot) + (ovflow <<3 ));
S.counter=2;
return double(S.V[1])*INV_M61;
}
void MixMaxRng::flatArray(const int size, double* vect )
{
// fill_array( fRngState, size, arrayDbl );
// fill_array( S, size, arrayDbl );
for (int i=0; i<size; ++i) { vect[i] = flat(); }
}
MixMaxRng::operator unsigned int()
{
return static_cast<unsigned int>(get_next(fRngState));
// get_next returns a 64-bit integer, of which the lower 61 bits
return static_cast<unsigned int>(get_next());
// clhep_get_next returns a 64-bit integer, of which the lower 61 bits
// are random and upper 3 bits are zero
}
@@ -202,12 +352,12 @@ std::ostream & MixMaxRng::put ( std::ostream& os ) const
int pr = os.precision(24);
os << beginMarker << " ";
os << theSeed << " ";
os << theSeed << "\n";
for (int i=0; i<rng_get_N(); ++i) {
os << fRngState->V[i] << "\n";
os << S.V[i] << "\n";
}
os << fRngState->counter << "\n";
os << fRngState->sumtot << "\n";
os << S.counter << "\n";
os << S.sumtot << "\n";
os << endMarker << "\n";
os.precision(pr);
return os;
@@ -217,15 +367,16 @@ std::vector<unsigned long> MixMaxRng::put () const
{
std::vector<unsigned long> v;
v.push_back (engineIDulong<MixMaxRng>());
for (int i=0; i<rng_get_N(); ++i) {
v.push_back(static_cast<unsigned long>(fRngState->V[i] & MASK32));
for (int i=0; i<rng_get_N(); ++i)
{
v.push_back(static_cast<unsigned long>(S.V[i] & MASK32));
// little-ended order on all platforms
v.push_back(static_cast<unsigned long>(fRngState->V[i] >> 32 ));
v.push_back(static_cast<unsigned long>(S.V[i] >> 32 ));
// pack uint64 into a data structure which is 32-bit on some platforms
}
v.push_back(static_cast<unsigned long>(fRngState->counter));
v.push_back(static_cast<unsigned long>(fRngState->sumtot & MASK32));
v.push_back(static_cast<unsigned long>(fRngState->sumtot >> 32));
v.push_back(static_cast<unsigned long>(S.counter));
v.push_back(static_cast<unsigned long>(S.sumtot & MASK32));
v.push_back(static_cast<unsigned long>(S.sumtot >> 32));
return v;
}
@@ -256,9 +407,9 @@ std::istream & MixMaxRng::getState ( std::istream& is )
{
char endMarker[MarkerLen];
is >> theSeed;
for (int i=0; i<rng_get_N(); ++i) is >> fRngState->V[i];
is >> fRngState->counter;
myuint checksum;
for (int i=0; i<rng_get_N(); ++i) is >> S.V[i];
is >> S.counter;
myuint_t checksum;
is >> checksum;
is >> std::ws;
is.width(MarkerLen);
@@ -269,14 +420,14 @@ std::istream & MixMaxRng::getState ( std::istream& is )
<< "\nInput stream is probably mispositioned now.\n";
return is;
}
if ( fRngState->counter < 0 || fRngState->counter > rng_get_N() ) {
if ( S.counter < 0 || S.counter > rng_get_N() ) {
std::cerr << "\nMixMaxRng::getState(): "
<< "vector read wrong value of counter from file!"
<< "\nInput stream is probably mispositioned now.\n";
return is;
}
precalc(fRngState);
if ( checksum != fRngState->sumtot) {
precalc();
if ( checksum != S.sumtot) {
std::cerr << "\nMixMaxRng::getState(): "
<< "checksum disagrees with value stored in file!"
<< "\nInput stream is probably mispositioned now.\n";
@@ -303,13 +454,13 @@ bool MixMaxRng::getState (const std::vector<unsigned long> & v)
return false;
}
for (int i=1; i<2*rng_get_N() ; i=i+2) {
fRngState->V[i/2]= ( (v[i] & MASK32) | ( (myuint)(v[i+1]) << 32 ) );
S.V[i/2]= ( (v[i] & MASK32) | ( (myuint_t)(v[i+1]) << 32 ) );
// unpack from a data structure which is 32-bit on some platforms
}
fRngState->counter = v[2*rng_get_N()+1];
precalc(fRngState);
S.counter = v[2*rng_get_N()+1];
precalc();
if ( ( (v[2*rng_get_N()+2] & MASK32)
| ( (myuint)(v[2*rng_get_N()+3]) << 32 ) ) != fRngState->sumtot) {
| ( (myuint_t)(v[2*rng_get_N()+3]) << 32 ) ) != S.sumtot) {
std::cerr << "\nMixMaxRng::getState(): vector has wrong checksum!"
<< "\nInput vector is probably mispositioned now.\n";
return false;
@@ -317,4 +468,291 @@ bool MixMaxRng::getState (const std::vector<unsigned long> & v)
return true;
}
myuint_t MixMaxRng ::MOD_MULSPEC(myuint_t k)
{
switch (N)
{
case 17:
return 0;
break;
case 8:
return 0;
break;
case 240:
return fmodmulM61( 0, SPECIAL , (k) );
break;
default:
std::cerr << "MIXMAX ERROR: " << "Disallowed value of parameter N\n";
std::terminate();
break;
}
}
myuint_t MixMaxRng::MULWU (myuint_t k)
{
return (( (k)<<(SPECIALMUL) & M61) ^ ( (k) >> (BITS-SPECIALMUL)) );
}
myuint_t MixMaxRng::iterate_raw_vec(myuint_t* Y, myuint_t sumtotOld)
{
// operates with a raw vector, uses known sum of elements of Y
int i;
myuint_t tempP, tempV;
Y[0] = ( tempV = sumtotOld);
myuint_t sumtot = Y[0], ovflow = 0; // will keep a running sum of all new elements
tempP = 0; // will keep a partial sum of all old elements
for (i=1; (i<N); i++)
{
myuint_t tempPO = MULWU(tempP);
tempP = modadd(tempP, Y[i]);
tempV = MIXMAX_MOD_MERSENNE(tempV+tempP+tempPO); // new Y[i] = old Y[i] + old partial * m
Y[i] = tempV;
sumtot += tempV; if (sumtot < tempV) {ovflow++;}
}
return MIXMAX_MOD_MERSENNE(MIXMAX_MOD_MERSENNE(sumtot) + (ovflow <<3 ));
}
myuint_t MixMaxRng::get_next()
{
int i;
i=S.counter;
if ((i<=(N-1)) )
{
S.counter++;
return S.V[i];
}
else
{
S.sumtot = iterate_raw_vec(S.V.data(), S.sumtot);
S.counter=2;
return S.V[1];
}
}
myuint_t MixMaxRng::precalc()
{
int i;
myuint_t temp;
temp = 0;
for (i=0; i < N; i++){
temp = MIXMAX_MOD_MERSENNE(temp + S.V[i]);
}
S.sumtot = temp;
return temp;
}
double MixMaxRng::get_next_float_packbits()
{
myuint_t Z=get_next();
return convert1double(Z);
}
void MixMaxRng::seed_vielbein(unsigned int index)
{
int i;
if (index<N)
{
for (i=0; i < N; i++){
S.V[i] = 0;
}
S.V[index] = 1;
}
else
{
std::terminate();
}
S.counter = N; // set the counter to N if iteration should happen right away
S.sumtot = 1;
}
#define MIXMAX_SEED_WAS_ZERO 0xFF02
void MixMaxRng::seed_spbox(myuint_t seed)
{
// a 64-bit LCG from Knuth line 26, in combination with a bit swap is used to seed
const myuint_t MULT64=6364136223846793005ULL;
int i;
myuint_t sumtot=0,ovflow=0;
if (seed == 0)
{
fprintf(stderr, " try seeding with nonzero seed next time!\n");
exit(MIXMAX_SEED_WAS_ZERO);
}
myuint_t l = seed;
for (i=0; i < N; i++){
l*=MULT64; l = (l << 32) ^ (l>>32);
S.V[i] = l & M61;
sumtot += S.V[(i)]; if (sumtot < S.V[(i)]) {ovflow++;}
}
S.counter = N; // set the counter to N if iteration should happen right away
S.sumtot = MIXMAX_MOD_MERSENNE(MIXMAX_MOD_MERSENNE(sumtot) + (ovflow <<3 ));
}
#undef MIXMAX_SEED_WAS_ZERO
void MixMaxRng::seed_uniquestream( myID_t clusterID, myID_t machineID, myID_t runID, myID_t streamID )
{
seed_vielbein(0);
S.sumtot = apply_bigskip(S.V.data(), S.V.data(), clusterID, machineID, runID, streamID );
S.counter = 1;
}
myuint_t MixMaxRng::apply_bigskip( myuint_t* Vout, myuint_t* Vin, myID_t clusterID, myID_t machineID, myID_t runID, myID_t streamID )
{
/*
makes a derived state vector, Vout, from the mother state vector Vin
by skipping a large number of steps, determined by the given seeding ID's
it is mathematically guaranteed that the substreams derived in this way from the SAME (!!!) Vin will not collide provided
1) at least one bit of ID is different
2) less than 10^100 numbers are drawn from the stream
(this is good enough : a single CPU will not exceed this in the lifetime of the universe, 10^19 sec,
even if it had a clock cycle of Planch time, 10^44 Hz )
Caution: never apply this to a derived vector, just choose some mother vector Vin, for example the unit vector by seed_vielbein(X,0),
and use it in all your runs, just change runID to get completely nonoverlapping streams of random numbers on a different day.
clusterID and machineID are provided for the benefit of large organizations who wish to ensure that a simulation
which is running in parallel on a large number of clusters and machines will have non-colliding source of random numbers.
did i repeat it enough times? the non-collision guarantee is absolute, not probabilistic
*/
const myuint_t skipMat17[128][17] =
#include "CLHEP/Random/mixmax_skip_N17.icc"
;
const myuint_t* skipMat[128];
for (int i=0; i<128; i++) { skipMat[i] = skipMat17[i];}
myID_t IDvec[4] = {streamID, runID, machineID, clusterID};
int r,i,j, IDindex;
myID_t id;
myuint_t Y[N], cum[N];
myuint_t coeff;
myuint_t* rowPtr;
myuint_t sumtot=0;
for (i=0; i<N; i++) { Y[i] = Vin[i]; sumtot = modadd( sumtot, Vin[i]); } ;
for (IDindex=0; IDindex<4; IDindex++)
{ // go from lower order to higher order ID
id=IDvec[IDindex];
//printf("now doing ID at level %d, with ID = %d\n", IDindex, id);
r = 0;
while (id)
{
if (id & 1)
{
rowPtr = (myuint_t*)skipMat[r + IDindex*8*sizeof(myID_t)];
for (i=0; i<N; i++){ cum[i] = 0; }
for (j=0; j<N; j++)
{ // j is lag, enumerates terms of the poly
// for zero lag Y is already given
coeff = rowPtr[j]; // same coeff for all i
for (i =0; i<N; i++){
cum[i] = fmodmulM61( cum[i], coeff , Y[i] ) ;
}
sumtot = iterate_raw_vec(Y, sumtot);
}
sumtot=0;
for (i=0; i<N; i++){ Y[i] = cum[i]; sumtot = modadd( sumtot, cum[i]); } ;
}
id = (id >> 1); r++; // bring up the r-th bit in the ID
}
}
sumtot=0;
for (i=0; i<N; i++){ Vout[i] = Y[i]; sumtot = modadd( sumtot, Y[i]); }
// returns sumtot, and copy the vector over to Vout
return (sumtot) ;
}
#if defined(__x86_64__)
myuint_t MixMaxRng::mod128(__uint128_t s)
{
myuint_t s1;
s1 = ( ( ((myuint_t)s)&M61 ) + ( ((myuint_t)(s>>64)) * 8 ) + ( ((myuint_t)s) >>BITS) );
return MIXMAX_MOD_MERSENNE(s1);
}
myuint_t MixMaxRng::fmodmulM61(myuint_t cum, myuint_t a, myuint_t b)
{
__uint128_t temp;
temp = (__uint128_t)a*(__uint128_t)b + cum;
return mod128(temp);
}
#else // on all other platforms, including 32-bit linux, PPC and PPC64, ARM and all Windows
myuint_t MixMaxRng::fmodmulM61(myuint_t cum, myuint_t s, myuint_t a)
{
const myuint_t MASK32=0xFFFFFFFFULL;
myuint_t o,ph,pl,ah,al;
o=(s)*a;
ph = ((s)>>32);
pl = (s) & MASK32;
ah = a>>32;
al = a & MASK32;
o = (o & M61) + ((ph*ah)<<3) + ((ah*pl+al*ph + ((al*pl)>>32))>>29) ;
o += cum;
o = (o & M61) + ((o>>61));
return o;
}
#endif
myuint_t MixMaxRng::modadd(myuint_t foo, myuint_t bar)
{
#if (defined(__x86_64__) || defined(__i386__)) && defined(__GNUC__) && (!defined(__ICC))
//#warning Using assembler routine in modadd
myuint_t out;
/* Assembler trick suggested by Andrzej Görlich */
__asm__ ("addq %2, %0; "
"btrq $61, %0; "
"adcq $0, %0; "
:"=r"(out)
:"0"(foo), "r"(bar)
);
return out;
#else
return MIXMAX_MOD_MERSENNE(foo+bar);
#endif
}
void MixMaxRng::print_state() const
{
int j;
std::cout << "mixmax state, file version 1.0\n";
std::cout << "N=" << rng_get_N() << "; V[N]={";
for (j=0; (j< (rng_get_N()-1) ); j++) {
std::cout << S.V[j] << ", ";
}
std::cout << S.V[rng_get_N()-1];
std::cout << "}; ";
std::cout << "counter= " << S.counter;
std::cout << "sumtot= " << S.sumtot << "\n";
}
MixMaxRng MixMaxRng::Branch()
{
S.sumtot = iterate_raw_vec(S.V.data(), S.sumtot); S.counter = 1;
MixMaxRng tmp=*this;
tmp.BranchInplace(0); // daughter id
return tmp;
}
void MixMaxRng::BranchInplace(int id)
{
// Dont forget to iterate the mother, when branching the daughter, or else will have collisions!
// a 64-bit LCG from Knuth line 26, is used to mangle a vector component
constexpr myuint_t MULT64=6364136223846793005ULL;
myuint_t tmp=S.V[id];
S.V[1] *= MULT64; S.V[id] &= M61;
S.sumtot = MIXMAX_MOD_MERSENNE( S.sumtot + S.V[id] - tmp + M61);
S.sumtot = iterate_raw_vec(S.V.data(), S.sumtot);// printf("iterating!\n");
S.counter = 1;
}
} // namespace CLHEP
source/externals/clhep/src/Random.cc
Vendored
+2 -2
View File
@@ -19,7 +19,7 @@
// =======================================================================
#include <assert.h>
#include "CLHEP/Random/JamesRandom.h"
#include "CLHEP/Random/MixMaxRng.h"
#include "CLHEP/Random/Random.h"
#include "CLHEP/Random/StaticRandomStates.h"
#include "CLHEP/Utility/memory.h"
@@ -65,7 +65,7 @@ namespace CLHEP {
private:
HepRandom theDefaultGenerator;
HepJamesRandom theDefaultEngine;
MixMaxRng theDefaultEngine;
public:
source/externals/clhep/src/mixmax.cc
Vendored
-452
View File
@@ -1,452 +0,0 @@
// $Id:$
// -*- C++ -*-
//
// -----------------------------------------------------------------------
// MixMax Matrix PseudoRandom Number Generator
// --- MixMax ---
// class header file
// -----------------------------------------------------------------------
//
//
// Created by Konstantin Savvidy on Sun Feb 22 2004.
// As of version 0.99 and later, the code is being released under
// GNU Lesser General Public License v3
//
// Generator described in
// N.Z.Akopov, G.K.Savvidy and N.G.Ter-Arutyunian, Matrix Generator of Pseudorandom Numbers,
// J.Comput.Phys. 97, 573 (1991);
// Preprint EPI-867(18)-86, Yerevan Jun.1986;
//
// and
//
// K.Savvidy
// The MIXMAX random number generator
// Comp. Phys. Commun. (2015)
// http://dx.doi.org/10.1016/j.cpc.2015.06.003
//
// -----------------------------------------------------------------------
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#define __MIXMAX_C // do NOT define it in your own program, just include mixmax.h
#include "CLHEP/Random/mixmax.h"
namespace CLHEP {
int iterate(rng_state_t* X){
X->sumtot = iterate_raw_vec(X->V, X->sumtot);
return 0;
}
#if (SPECIALMUL!=0)
inline uint64_t MULWU (uint64_t k){ return (( (k)<<(SPECIALMUL) & M61) | ( (k) >> (BITS-SPECIALMUL)) ) ;}
#elif (SPECIALMUL==0)
inline uint64_t MULWU (uint64_t){ return 0;}
#else
#error SPECIALMUL not undefined
#endif
myuint iterate_raw_vec(myuint* Y, myuint sumtotOld){
// operates with a raw vector, uses known sum of elements of Y
int i;
#ifdef SPECIAL
myuint temp2 = Y[1];
#endif
myuint tempP, tempV;
Y[0] = ( tempV = sumtotOld);
myuint sumtot = Y[0], ovflow = 0; // will keep a running sum of all new elements (except Y[0])
tempP = 0; // will keep a partial sum of all old elements (except Y[0])
for (i=1; i<N; i++){
#if (SPECIALMUL!=0)
myuint tempPO = MULWU(tempP);
tempP = modadd(tempP,Y[i]);
tempV = MOD_MERSENNE(tempV + tempP + tempPO); // edge cases ?
#else
tempP = modadd(tempP , Y[i]);
tempV = modadd(tempV , tempP);
#endif
Y[i] = tempV;
sumtot += tempV; if (sumtot < tempV) {ovflow++;}
}
#ifdef SPECIAL
temp2 = MOD_MULSPEC(temp2);
Y[2] = modadd( Y[2] , temp2 );
sumtot += temp2; if (sumtot < temp2) {ovflow++;}
#endif
return MOD_MERSENNE(MOD_MERSENNE(sumtot) + (ovflow <<3 ));
}
myuint get_next(rng_state_t* X) {
return GET_BY_MACRO(X);
}
double get_next_float(rng_state_t* X){
return get_next_float_BY_MACRO(X);
}
void fill_array(rng_state_t* X, unsigned int n, double *array)
{
// Return an array of n random numbers uniformly distributed in (0,1]
unsigned int i,j;
const int M=N-1;
for (i=0; i<(n/M); i++){
iterate_and_fill_array(X, array+i*M);
}
unsigned int rem=(n % M);
if (rem) {
iterate(X);
for (j=0; j< (rem); j++){
array[M*i+j] = (int64_t)X->V[j] * (double)(INV_MERSBASE);
}
X->counter = j; // needed to continue with single fetches from the exact spot, but if you only use fill_array to get numbers then it is not necessary
}else{
X->counter = N;
}
}
void iterate_and_fill_array(rng_state_t* X, double *array){
myuint* Y=X->V;
int i;
myuint tempP, tempV;
#if (SPECIAL != 0)
myuint temp2 = Y[1];
#endif
Y[0] = (tempV = modadd(Y[0] , X->sumtot));
//array[0] = (double)tempV * (double)(INV_MERSBASE);
myuint sumtot = 0, ovflow = 0; // will keep a running sum of all new elements (except Y[0])
tempP = 0; // will keep a partial sum of all old elements (except Y[0])
for (i=1; i<N; i++){
tempP = modadd(tempP,Y[i]);
Y[i] = ( tempV = modadd(tempV,tempP) );
sumtot += tempV; if (sumtot < tempV) {ovflow++;}
array[i-1] = (int64_t)tempV * (double)(INV_MERSBASE);
}
#if (SPECIAL != 0)
temp2 = MOD_MULSPEC(temp2);
Y[2] = modadd( Y[2] , temp2 );
sumtot += temp2; if (sumtot < temp2) {ovflow++;}
#endif
X->sumtot = MOD_MERSENNE(MOD_MERSENNE(sumtot) + (ovflow <<3 ));
}
myuint modadd(myuint foo, myuint bar){
#if defined(__x86_64__) && defined(USE_INLINE_ASM)
myuint out;
/* Assembler trick suggested by Andrzej Görlich */
__asm__ ("addq %2, %0; "
"btrq $61, %0; "
"adcq $0, %0; "
:"=r"(out)
:"0"(foo), "r"(bar)
);
return out;
#else
return MOD_MERSENNE(foo+bar);
#endif
}
rng_state_t* rng_alloc()
{
/* allocate the state */
rng_state_t *p = (rng_state_t*)malloc(sizeof(rng_state_t));
p->fh=NULL; // by default, set the output file handle to stdout
return p;
}
int rng_free(rng_state_t* X) /* free the memory occupied by the state */
{
free(X);
return 0;
}
rng_state_t* rng_copy(myuint *Y)
{
/* copy the vector stored at Y, and return pointer to the newly allocated and initialized state.
It is the user's responsibility to make sure that Y is properly allocated with rng_alloc,
then pass Y->V or it can also be an array -- such as myuint Y[N+1] and Y[1]...Y[N] have been set to legal values [0 .. MERSBASE-1]
Partial sums on this new state are recalculated, and counter set to zero, so that when get_next is called,
it will output the initial vector before any new numbers are produced, call iterate(X) if you want to advance right away */
rng_state_t* X = rng_alloc();
myuint sumtot=0,ovflow=0;
X->counter = 2;
int i;
for ( i=0; i < N; i++){
X->V[i] = Y[i];
sumtot += X->V[(i)]; if (sumtot < X->V[(i)]) {ovflow++;}
}
X->sumtot = MOD_MERSENNE(MOD_MERSENNE(sumtot) + (ovflow <<3 ));
return X;
}
void seed_vielbein(rng_state_t* X, unsigned int index)
{
int i;
if (index<N){
for (i=0; i < N; i++){
X->V[i] = 0;
}
X->V[index] = 1;
}else{
fprintf(stderr, "Out of bounds index, is not ( 0 <= index < N )\n"); exit(ARRAY_INDEX_OUT_OF_BOUNDS);
}
X->counter = N; // set the counter to N if iteration should happen right away
//precalc(X);
X->sumtot = 1; //(index ? 1:0);
if (X->fh==NULL){X->fh=stdout;}
}
void seed_spbox(rng_state_t* X, myuint seed)
{ // a 64-bit LCG from Knuth line 26, in combination with a bit swap is used to seed
const myuint MULT64=6364136223846793005ULL;
int i;
myuint sumtot=0,ovflow=0;
if (seed == 0){
fprintf(stderr, " try seeding with nonzero seed next time!\n");
exit(SEED_WAS_ZERO);
}
myuint l = seed;
//X->V[0] = l & MERSBASE;
if (X->fh==NULL){X->fh=stdout;} // if the filehandle is not yet set, make it stdout
for (i=0; i < N; i++){
l*=MULT64; l = (l << 32) ^ (l>>32);
X->V[i] = l & MERSBASE;
sumtot += X->V[(i)]; if (sumtot < X->V[(i)]) {ovflow++;}
}
X->counter = N; // set the counter to N if iteration should happen right away
X->sumtot = MOD_MERSENNE(MOD_MERSENNE(sumtot) + (ovflow <<3 ));
}
myuint precalc(rng_state_t* X){
int i;
myuint temp;
temp = 0;
for (i=0; i < N; i++){
temp = MOD_MERSENNE(temp + X->V[i]);
}
X->sumtot = temp;
return temp;
}
int rng_get_N(void){return N;}
#if defined(__x86_64__)
inline myuint mod128(__uint128_t s){
myuint s1;
s1 = ( ( ((myuint)s)&MERSBASE ) + ( ((myuint)(s>>64)) * 8 ) + ( ((myuint)s) >>BITS) );
return MOD_MERSENNE(s1);
}
inline myuint fmodmulM61(myuint cum, myuint a, myuint b){
__uint128_t temp;
temp = (__uint128_t)a*(__uint128_t)b + cum;
return mod128(temp);
}
#else // on all other platforms, including 32-bit linux, PPC and PPC64 and all Windows
#define MASK32 0xFFFFFFFFULL
inline myuint fmodmulM61(myuint cum, myuint s, myuint a)
{
myuint o,ph,pl,ah,al;
o=(s)*a;
ph = ((s)>>32);
pl = (s) & MASK32;
ah = a>>32;
al = a & MASK32;
o = (o & M61) + ((ph*ah)<<3) + ((ah*pl+al*ph + ((al*pl)>>32))>>29) ;
o += cum;
o = (o & M61) + ((o>>61));
return o;
}
#endif
void print_state(rng_state_t* X){
int j;
fprintf(X->fh, "mixmax state, file version 1.0\n" );
fprintf(X->fh, "N=%u; V[N]={", rng_get_N() );
for (j=0; (j< (rng_get_N()-1) ); j++) {
fprintf(X->fh, "%llu, ", X->V[j] );
}
fprintf(X->fh, "%llu", X->V[rng_get_N()-1] );
fprintf(X->fh, "}; " );
fprintf(X->fh, "counter=%u; ", X->counter );
fprintf(X->fh, "sumtot=%llu;\n", X->sumtot );
}
void read_state(rng_state_t* X, const char filename[] ){
// a function for reading the state from a file, after J. Apostolakis
FILE* fin;
if( ( fin = fopen(filename, "r") ) ){
char l=0;
while ( l != '{' ) { // 0x7B = "{"
l=fgetc(fin); // proceed until hitting opening bracket
}
ungetc(' ', fin);
}else{
fprintf(stderr, "mixmax -> read_state: error reading file %s\n", filename);
exit(ERROR_READING_STATE_FILE);
}
myuint vecVal;
//printf("mixmax -> read_state: starting to read state from file\n");
if (!fscanf(fin, "%llu", &X->V[0]) ) {fprintf(stderr, "mixmax -> read_state: error reading file %s\n", filename); exit(ERROR_READING_STATE_FILE);}
//printf("V[%d] = %llu\n",0, X->V[0]);
int i;
for( i = 1; i < rng_get_N(); i++){
if (!fscanf(fin, ", %llu", &vecVal) ) {fprintf(stderr, "mixmax -> read_state: error reading vector component i=%d from file %s\n", i, filename); exit(ERROR_READING_STATE_FILE);}
//printf("V[%d] = %llu\n",i, vecVal);
if( vecVal <= MERSBASE ){
X->V[i] = vecVal;
}else{
fprintf(stderr, "mixmax -> read_state: Invalid state vector value= %llu"
" ( must be less than %llu ) "
" obtained from reading file %s\n"
, vecVal, MERSBASE, filename);
}
}
unsigned int counter;
if (!fscanf( fin, "}; counter=%u; ", &counter)){fprintf(stderr, "mixmax -> read_state: error reading counter from file %s\n", filename); exit(ERROR_READING_STATE_FILE);}
if( counter <= N ) {
X->counter= counter;
}else{
fprintf(stderr, "mixmax -> read_state: Invalid counter = %d"
" Must be 0 <= counter < %u\n" , counter, N);
print_state(X);
exit(ERROR_READING_STATE_COUNTER);
}
precalc(X);
myuint sumtot;
if (!fscanf( fin, "sumtot=%llu\n", &sumtot)){fprintf(stderr, "mixmax -> read_state: error reading checksum from file %s\n", filename); exit(ERROR_READING_STATE_FILE);}
if (X->sumtot != sumtot) {
fprintf(stderr, "mixmax -> checksum error while reading state from file %s - corrupted?\n", filename);
exit(ERROR_READING_STATE_CHECKSUM);
}
// else{fprintf(stderr, "mixmax -> read_state: checksum ok: %llu == %llu\n",X->sumtot, sumtot);}
fclose(fin);
}
#define FUSEDMODMULVEC \
{ for (i =0; i<N; i++){ \
cum[i] = fmodmulM61( cum[i], coeff , Y[i] ) ; \
} }
#define SKIPISON 1
#if (BITS==61 && SKIPISON!=0)
void seed_uniquestream( rng_state_t* Xin, myID_t clusterID, myID_t machineID, myID_t runID, myID_t streamID ){
seed_vielbein(Xin,0);
Xin->sumtot = apply_bigskip(Xin->V, Xin->V, clusterID, machineID, runID, streamID );
if (Xin->fh==NULL){Xin->fh=stdout;} // if the filehandle is not yet set, make it stdout
}
void branch_inplace( rng_state_t* Xin, myID_t* IDvec ){
Xin->sumtot = apply_bigskip(Xin->V, Xin->V, IDvec[3], IDvec[2], IDvec[1], IDvec[0] );
}
myuint apply_bigskip(myuint* Vout, myuint* Vin, myID_t clusterID, myID_t machineID, myID_t runID, myID_t streamID ){
/*
makes a derived state vector, Vout, from the mother state vector Vin
by skipping a large number of steps, determined by the given seeding ID's
it is mathematically guaranteed that the substreams derived in this way from the SAME (!!!) Vin will not collide provided
1) at least one bit of ID is different
2) less than 10^100 numbers are drawn from the stream
(this is good enough : a single CPU will not exceed this in the lifetime of the universe, 10^19 sec,
even if it had a clock cycle of Planch time, 10^44 Hz )
Caution: never apply this to a derived vector, just choose some mother vector Vin, for example the unit vector by seed_vielbein(X,0),
and use it in all your runs, just change runID to get completely nonoverlapping streams of random numbers on a different day.
clusterID and machineID are provided for the benefit of large organizations who wish to ensure that a simulation
which is running in parallel on a large number of clusters and machines will have non-colliding source of random numbers.
did i repeat it enough times? the non-collision guarantee is absolute, not probabilistic
*/
const myuint skipMat[128][N] =
//#if (N==8)
//#include "CLHEP/Random/mixmax_skip_N8.icc"
//#elif (N==17)
#include "CLHEP/Random/mixmax_skip_N17.icc"
//#elif (N==88)
//#include "CLHEP/Random/mixmax_skip_N88.icc" // to make this file, delete all except some chosen 128 rows of the coefficients table
//#elif (N==256)
//#include "CLHEP/Random/mixmax_skip_N256.icc"
//#elif (N==1000)
//#include "CLHEP/Random/mixmax_skip_N1000.icc"
//#elif (N==3150)
//#include "CLHEP/Random/mixmax_skip_N3150.icc"
//#endif
;
myID_t IDvec[4] = {streamID, runID, machineID, clusterID};
int r,i,j, IDindex;
myID_t id;
myuint Y[N], cum[N];
myuint coeff;
myuint* rowPtr;
myuint sumtot=0;
for (i=0; i<N; i++) { Y[i] = Vin[i]; sumtot = modadd( sumtot, Vin[i]); } ;
for (IDindex=0; IDindex<4; IDindex++) { // go from lower order to higher order ID
id=IDvec[IDindex];
//printf("now doing ID at level %d, with ID = %d\n", IDindex, id);
r = 0;
while (id){
if (id & 1) {
rowPtr = (myuint*)skipMat[r + IDindex*8*sizeof(myID_t)];
//printf("free coeff for row %d is %llu\n", r, rowPtr[0]);
for (i=0; i<N; i++){ cum[i] = 0; }
for (j=0; j<N; j++){ // j is lag, enumerates terms of the poly
// for zero lag Y is already given
coeff = rowPtr[j]; // same coeff for all i
FUSEDMODMULVEC;
sumtot = iterate_raw_vec(Y, sumtot);
}
sumtot=0;
for (i=0; i<N; i++){ Y[i] = cum[i]; sumtot = modadd( sumtot, cum[i]); } ;
}
id = (id >> 1); r++; // bring up the r-th bit in the ID
}
}
sumtot=0;
for (i=0; i<N; i++){ Vout[i] = Y[i]; sumtot = modadd( sumtot, Y[i]); } ; // returns sumtot, and copy the vector over to Vout
return (sumtot) ;
}
#else
#warning For this N, we dont have the skipping coefficients yet, using alternative method to seed
void seed_uniquestream( rng_state_t* Xin, myID_t clusterID, myID_t machineID, myID_t runID, myID_t streamID ){
Xin->V[0] = (myuint)clusterID;
Xin->V[1] = (myuint)machineID;
Xin->V[2] = (myuint)runID;
Xin->V[3] = (myuint)streamID;
Xin->V[4] = (myuint)clusterID << 5;
Xin->V[5] = (myuint)machineID << 7;
Xin->V[6] = (myuint)runID << 11;
Xin->V[7] = (myuint)streamID << 13;
precalc(Xin);
Xin->sumtot = iterate_raw_vec(Xin->V, Xin->sumtot);
Xin->sumtot = iterate_raw_vec(Xin->V, Xin->sumtot);
}
#endif // SKIPISON
} // namespace CLHEP