JDFTx  1.7.0
Shared utilities

Files

file  Blip.h
 
file  EnergyComponents.h
 Represent components of the (free) energy.
 
file  GpuKernelUtils.h
 Common utility functions/macros for the gpu kernels and launchers in the .cu files.
 
file  GpuUtil.h
 
file  LoopMacros.h
 
file  MPIUtil.h
 Helper classes for MPI parallelization.
 
file  SphericalHarmonics.h
 
file  Spline.h
 Spline interpolation routines.
 
file  string.h
 STL strings and streams with case insensitive comparison.
 
file  Thread.h
 Utilities for threading (wrappers around std::thread)
 
file  Units.h
 Commonly used measurement units in terms of atomic units.
 
file  Util.h
 Miscellaneous utilities.
 
file  symbols.h
 
file  Euler.h
 Various Euler angle related utilities.
 

Classes

class  BlipConverter
 
class  BlipResampler
 
class  EnergyComponents
 
struct  GpuLaunchConfig
 Base-class for launch configuration for gpu kernels. More...
 
struct  GpuLaunchConfig1D
 1D launch configuration More...
 
struct  GpuLaunchConfig3D
 3D launch configuration More...
 
struct  GpuLaunchConfigHalf3D
 3D launch configuration for symmetry-reduced G-space loops (z dimension folded for real data sets) More...
 
class  ManagedMemory< T >
 Base class for managed memory of a specified data type. More...
 
class  MPIUtil
 MPI wrapper class. More...
 
class  TaskDivision
 Helper for optimally dividing a specified number of (equal) tasks over MPI. More...
 
struct  YlmProdTerm
 Term in real spherical harmonic expansion of a product of two real spherical harmonics. More...
 
struct  ichar_traits
 Case insensitive character trait. More...
 
struct  ifstream
 
struct  ofstream
 
struct  istringstream
 
struct  ostringstream
 
struct  InitParams
 Parameters used for common initialization functions. More...
 
class  StopWatch
 
class  EnumStringMap< Enum >
 A template to ease option parsing (maps enums <--> strings) More...
 

Macros

#define kernelIndex(dir)   (blockIdx.dir * blockDim.dir + threadIdx.dir)
 
#define kernelIndex1D()   ((blockIdx.y*gridDim.x+blockIdx.x) * blockDim.x + threadIdx.x)
 
#define THREAD_rLoop(code)
 One thread of a loop over real space (see applyFunc_r_sub for example) More...
 
#define COMPUTE_rIndices
 The GPU equivalent of THREAD_rLoop. More...
 
#define THREAD_fullGspaceLoop(code)
 One thread of a loop over G-space (see L_sub in PHLO.cpp for example) More...
 
#define COMPUTE_fullGindices
 
#define THREAD_halfGspaceLoop(code)
 One thread of a loop over symmetry-reduced G-space (see applyFuncGsq_sub in Operators.h for example) More...
 
#define COMPUTE_halfGindices
 
#define IS_NYQUIST   ( (!(2*iG[0]-S[0])) | (!(2*iG[1]-S[1])) | (!(2*iG[2]-S[2])) )
 
#define SwitchTemplate_lm(l, m, fTemplate, argList)
 Switch a function templated over l,m for all supported l,m with parenthesis enclosed argument list argList.
 
#define DECLARE_YlmPrime(l, m)   template<> __hostanddev__ vector3<> YlmPrime<l*(l+1)+m>(const vector3<>& qHat) { return YlmPrime<l,m>(qHat); }
 
#define TIME(title, fp, code)
 Elapsed time in seconds (from start of program) More...
 
#define assert(expr)    (void)((expr) ? 0 : assertStackTraceExit(#expr, __func__, __FILE__, __LINE__))
 A custom assertion with stack trace (NOTE: enabled in release modes as well)
 
#define logPrintf(...)   fprintf(globalLog, __VA_ARGS__)
 printf() for log files
 
#define logFlush()   fflush(globalLog)
 fflush() for log files
 
#define die(...)
 Quit with an error message (formatted using printf()). Must be called from all processes. More...
 
#define die_alone(...)
 Version of die that should only be used when it is impossible to guarantee synchronized calls from all processes. More...
 
#define PRIdPTR   "zd"
 

Typedefs

typedef std::basic_string< char, ichar_traitsstring
 Case-insensitive string.
 

Enumerations

enum class  AtomicSymbol : int {
  H = 1 , He = 2 , Li = 3 , Be = 4 ,
  B = 5 , C = 6 , N = 7 , O = 8 ,
  F = 9 , Ne = 10 , Na = 11 , Mg = 12 ,
  Al = 13 , Si = 14 , P = 15 , S = 16 ,
  Cl = 17 , Ar = 18 , K = 19 , Ca = 20 ,
  Sc = 21 , Ti = 22 , V = 23 , Cr = 24 ,
  Mn = 25 , Fe = 26 , Co = 27 , Ni = 28 ,
  Cu = 29 , Zn = 30 , Ga = 31 , Ge = 32 ,
  As = 33 , Se = 34 , Br = 35 , Kr = 36 ,
  Rb = 37 , Sr = 38 , Y = 39 , Zr = 40 ,
  Nb = 41 , Mo = 42 , Tc = 43 , Ru = 44 ,
  Rh = 45 , Pd = 46 , Ag = 47 , Cd = 48 ,
  In = 49 , Sn = 50 , Sb = 51 , Te = 52 ,
  I = 53 , Xe = 54 , Cs = 55 , Ba = 56 ,
  La = 57 , Ce = 58 , Pr = 59 , Nd = 60 ,
  Pm = 61 , Sm = 62 , Eu = 63 , Gd = 64 ,
  Tb = 65 , Dy = 66 , Ho = 67 , Er = 68 ,
  Tm = 69 , Yb = 70 , Lu = 71 , Hf = 72 ,
  Ta = 73 , W = 74 , Re = 75 , Os = 76 ,
  Ir = 77 , Pt = 78 , Au = 79 , Hg = 80 ,
  Tl = 81 , Pb = 82 , Bi = 83 , Po = 84 ,
  At = 85 , Rn = 86 , Fr = 87 , Ra = 88 ,
  Ac = 89 , Th = 90 , Pa = 91 , U = 92 ,
  Np = 93 , Pu = 94 , Am = 95 , Cm = 96 ,
  Bk = 97 , Cf = 98 , Es = 99 , Fm = 100 ,
  Md = 101 , No = 102 , Lr = 103 , Rf = 104 ,
  Db = 105 , Sg = 106 , Bh = 107 , Hs = 108 ,
  Mt = 109 , Ds = 110 , Rg = 111 , Cn = 112 ,
  Uut = 113 , Uuq = 114 , Uup = 115 , Uuh = 116 ,
  Uus = 117 , Uuo = 118
}
 Atomic symbol to atomic number.
 

Functions

double Tblip (const complexScalarField &phi, double *tMax=0, int *i0max=0, int *i1max=0, int *i2max=0)
 Compute the kinetic energy for a blip orbital phi (and set max local KE and location in unit cell)
 
double Vblip (const complexScalarField &phi, const ScalarField &V)
 Compute the local potential energy for blip orbital phi in blip potential V.
 
void gpuErrorCheck ()
 Check for gpu errors and print a useful message (implemented in GpuUtils.cpp) More...
 
bool gpuInit (FILE *fpLog=stdout, const class MPIUtil *mpiHostGpu=0, double *nGPUs=0)
 
bool isGpuMine ()
 
bool isGpuEnabled ()
 A utility function to eliminate ugly #ifdef's when not required.
 
double bessel_jl (int l, double x)
 Spherical bessel function.
 
template<int lm>
__hostanddev__ double Ylm (const vector3<> &qhat)
 Index by combined lm := l*(l+1)+m index (useful when static-looping over all l,m)
 
template<int l, int m>
__hostanddev__ double Ylm (const vector3<> &qhat)
 Index by l and m separately.
 
template<int l, int m>
void set_Ylm (const vector3<> qHat, double &result)
 Use above macro to provide a non-templated version of the function.
 
double Ylm (int l, int m, const vector3<> &qHat)
 
std::vector< YlmProdTermexpandYlmProd (int lm1, int lm2)
 
std::vector< YlmProdTermexpandYlmProd (int l1, int m1, int l2, int m2)
 Wrapper function expandYlmProd with individual indices.
 
template<int l, int m>
__hostanddev__ vector3 YlmPrime (const vector3<> &qHat)
 Derivative of spherical Harmonic with repect to iDir'th component of qHat with separate l,m indices. More...
 
template<int lm>
__hostanddev__ vector3 YlmPrime (const vector3<> &qHat)
 Derivative of spherical Harmonic with repect to iDir'th component of qHat with combined lm index. More...
 
 DECLARE_YlmPrime (0, 0) DECLARE_YlmPrime(1
 
 DECLARE_YlmPrime (1, 0) DECLARE_YlmPrime(1
 
 DECLARE_YlmPrime (2,-2) DECLARE_YlmPrime(2
 
 DECLARE_YlmPrime (2, 0) DECLARE_YlmPrime(2
 
 DECLARE_YlmPrime (2, 2) DECLARE_YlmPrime(3
 
 DECLARE_YlmPrime (3,-2) DECLARE_YlmPrime(3
 
 DECLARE_YlmPrime (3, 0) DECLARE_YlmPrime(3
 
 DECLARE_YlmPrime (3, 2) DECLARE_YlmPrime(3
 
 DECLARE_YlmPrime (4,-4) DECLARE_YlmPrime(4
 
 DECLARE_YlmPrime (4, 0) DECLARE_YlmPrime(4
 
 DECLARE_YlmPrime (4, 2) DECLARE_YlmPrime(4
 
 DECLARE_YlmPrime (4, 4) DECLARE_YlmPrime(5
 
 DECLARE_YlmPrime (5,-4) DECLARE_YlmPrime(5
 
 DECLARE_YlmPrime (5, 0) DECLARE_YlmPrime(5
 
 DECLARE_YlmPrime (5, 2) DECLARE_YlmPrime(5
 
 DECLARE_YlmPrime (5, 4) DECLARE_YlmPrime(5
 
 DECLARE_YlmPrime (6,-6) DECLARE_YlmPrime(6
 
 DECLARE_YlmPrime (6, 0) DECLARE_YlmPrime(6
 
 DECLARE_YlmPrime (6, 2) DECLARE_YlmPrime(6
 
 DECLARE_YlmPrime (6, 4) DECLARE_YlmPrime(6
 
 DECLARE_YlmPrime (6, 6) template< int l
 Non-templated version of YlmPrime (for debugging)
 
int m void set_YlmPrime (const vector3<> qHat, vector3<> &result)
 
vector3 YlmPrime (int l, int m, const vector3<> &qHat)
 
void trim (string &s)
 Remove leading and trailing spaces from a string.
 
istream & operator>> (istream &is, string &str)
 
ostream & operator<< (ostream &os, const string &str)
 
istream & getline (istream &is, string &str, char delim='\n')
 
bool shouldThreadOperators ()
 
void suspendOperatorThreading ()
 call from multi-threaded top-level code to disable threading within operators called from a parallel section
 
void resumeOperatorThreading ()
 call after a parallel section in top-level code to resume threading within subsequent operator calls
 
template<typename Callable , typename ... Args>
void threadLaunch (int nThreads, Callable *func, size_t nJobs, Args... args)
 A simple utility for running muliple threads. More...
 
template<typename Callable , typename ... Args>
void threadLaunch (Callable *func, size_t nJobs, Args... args)
 
template<typename Callable , typename ... Args>
void threadedLoop (Callable *func, size_t nIter, Args... args)
 A parallelized loop. More...
 
template<typename Callable , typename ... Args>
double threadedAccumulate (Callable *func, size_t nIter, Args... args)
 A parallelized loop with an accumulated return value. More...
 
void printVersionBanner (const InitParams *ip=0)
 Print package name, version, revision etc. to log.
 
void initSystem (int argc, char **argv, const InitParams *ip=0)
 Init MPI (if not already done), print banner, set up threads (play nice with job schedulers), GPU and signal handlers.
 
void initSystemCmdline (int argc, char **argv, InitParams &ip)
 initSystem along with commandline options
 
void finalizeSystem (bool successful=true)
 Clean-up corresponding to initSystem(), final messages (depending on successful) and clean-up MPI.
 
double clock_us ()
 
double clock_sec ()
 Elapsed time in microseconds (from start of program)
 
void printStack (bool detailedStackScript=false)
 Print a minimal stack trace and optionally write a script that, when run, will print a more detailed stacktrace.
 
void stackTraceExit (int code)
 Exit on error with stack trace.
 
int assertStackTraceExit (const char *expr, const char *function, const char *file, long line)
 
void logSuspend ()
 temporarily disable all log output (until logResume())
 
void logResume ()
 re-enable logging after a logSuspend() call
 
off_t fileSize (const char *filename)
 Get the size of a file.
 
void convertToLE (void *ptr, size_t size, size_t nmemb)
 Convert data from operating endianness to little-endian.
 
void convertFromLE (void *ptr, size_t size, size_t nmemb)
 Convert data from little-endian to operating endianness.
 
size_t freadLE (void *ptr, size_t size, size_t nmemb, FILE *fp)
 Read from a little-endian binary file, regardless of operating endianness.
 
size_t fwriteLE (const void *ptr, size_t size, size_t nmemb, FILE *fp)
 Write to a little-endian binary file, regardless of operating endianness.
 
uint16_t positiveRemainder (int16_t x, uint16_t y)
 For any x and y>0, compute z = x % y such that 0 <= z < y.
 
bool fftSuitable (int N)
 Check if an integer is suitable for Fast Fourier Transforms (small prime factors only)
 
double atomicMass (AtomicSymbol)
 retrieve atomic mass for each element
 
vector3 polarUnitVector (double alpha, double beta)
 Get the position of the new Z-axis, given alpha and beta (does not depend on gamma)
 
void getEulerAxis (const vector3<> &newZ, vector3<> &euler)
 Calculates euler[0]=alpha and euler[1]=beta given newZ, the direction of the Z-axis in the rotated frame.
 
matrix3 matrixFromEuler (const vector3<> &euler)
 Calculate rotation matrices from euler angles.
 
vector3 eulerFromMatrix (const matrix3<> &mat)
 Calculate euler angles from rotation matrices.
 
double wigner_d (const int j, const int m1, const int m2, const double beta)
 Wigner d-function d^j_{m1 m2}(beta) for ZYZ Euler angles.
 

Variables

cudaDeviceProp cudaDevProps
 cached properties of currently running device (defined in GpuUtil.cpp)
 
cublasHandle_t cublasHandle
 global handle to cublas (defined in GpuUtil.cpp)
 
cublasHandle_t cublasHandle
 global handle to cublas (defined in GpuUtil.cpp)
 
int nProcsAvailable
 number of available processors (initialized to number of online processors, can be overriden)
 
const double eV = 1/27.21138505
 eV in Hartrees
 
const double Ryd = 0.5
 Rydberg in Hartrees.
 
const double Joule = 1/4.35974434e-18
 Joule in Hartrees.
 
const double KJoule = 1000*Joule
 KJoule in Hartrees.
 
const double Kcal = KJoule * 4.184
 Kcal in Hartrees.
 
const double Kelvin = 1.3806488e-23*Joule
 Kelvin in Hartrees.
 
const double invcm = 1./219474.6313705
 Inverse cm in Hartrees.
 
const double Angstrom = 1/0.5291772
 Angstrom in bohrs.
 
const double meter = 1e10*Angstrom
 meter in bohrs
 
const double liter = 1e-3*pow(meter,3)
 liter in cubic bohrs
 
const double amu = 1822.88839
 atomic mass unit in electron masses
 
const double kg = 1./9.10938291e-31
 kilogram in electron masses
 
const double mol = 6.0221367e23
 mole in number (i.e. Avogadro number)
 
const double Newton = Joule/meter
 Newton in Hartree/bohr.
 
const double Pascal = Newton/(meter*meter)
 Pascal in Hartree/bohr^3.
 
const double KPascal = 1000*Pascal
 KPa in Hartree/bohr^3.
 
const double Bar = 100*KPascal
 bar in Hartree/bohr^3
 
const double mmHg = 133.322387415*Pascal
 mm Hg in Hartree/bohr^3
 
const double sec = sqrt((kg*meter)/Newton)
 second in inverse Hartrees
 
const double invSec = 1./sec
 inverse second in Hartrees
 
const double fs = sec*1.0e-15
 femtosecond in inverse Hartrees
 
const double Coul = Joule/eV
 Coulomb in electrons.
 
const double Volt = Joule/Coul
 Volt in Hartrees.
 
const double Ampere = Coul/sec
 Ampere in electrons/inverse Hartree.
 
const double Ohm = Volt/Ampere
 Ohm in inverse conductance quanta.
 
const double Tesla = Volt*sec/(meter*meter)
 Tesla in atomic units.
 
const double bohrMagneton = 0.5
 
const double gElectron = 2.0023193043617
 electron gyromagnetic ratio
 
string inputBasename
 Basename of input file or "stdin" that can be used as a default run-name.
 
bool killFlag
 Flag set by signal handlers - all compute loops should quit cleanly when this is set.
 
MPIUtilmpiWorld
 MPI across all processes.
 
MPIUtilmpiGroup
 MPI within current group of processes.
 
MPIUtilmpiGroupHead
 MPI across equal ranks in each group.
 
bool mpiDebugLog
 If true, all processes output to seperate debug log files, otherwise only head process outputs (set before calling initSystem())
 
size_t mempoolSize
 If non-zero, size of memory pool managed internally by JDFTx.
 
FILE * globalLog
 
FILE * nullLog
 pointer to /dev/null
 
EnumStringMap< AtomicSymbolatomicSymbolMap
 Relate chemical symbols and atomic numbers.
 

Detailed Description

Macro Definition Documentation

◆ COMPUTE_fullGindices

#define COMPUTE_fullGindices
Value:
zBlock * blockDim.z + threadIdx.z, \
kernelIndex(y), \
kernelIndex(x) ); \
if(iG[0]>=S[0] || iG[1]>=S[1] || iG[2]>=S[2]) return; \
size_t i = iG[2] + S[2]*size_t(iG[1] + S[1]*iG[0]); \
for(int j=0; j<3; j++) if(2*iG[j]>S[j]) iG[j]-=S[j];

The GPU equivalent of THREAD_fullGspaceLoop NOTE: x and z are swapped in kernel indices since x is contiguous on GPU

◆ COMPUTE_halfGindices

#define COMPUTE_halfGindices
Value:
int size2 = S[2]/2+1; \
vector3<int> iG( \
threadIdx.z + zBlock * blockDim.z, \
kernelIndex(y), \
kernelIndex(x) ); \
if(iG[0]>=S[0] || iG[1]>=S[1] || iG[2]>=size2) return; \
size_t i = iG[2] + size2*size_t(iG[1] + S[1]*iG[0]); \
for(int j=0; j<3; j++) if(2*iG[j]>S[j]) iG[j]-=S[j];

The GPU equivalent of THREAD_halfGspaceLoop NOTE: x and z are swapped in kernel indices since x is contiguous on GPU

◆ COMPUTE_rIndices

#define COMPUTE_rIndices
Value:
threadIdx.z + zBlock * blockDim.z, \
kernelIndex(y), \
kernelIndex(x) ); \
if(iv[0]>=S[0] || iv[1]>=S[1] || iv[2]>=S[2]) return; \
size_t i = iv[2] + S[2]*size_t(iv[1] + S[1]*iv[0]);

The GPU equivalent of THREAD_rLoop.

◆ die

#define die (   ...)
Value:
{ fprintf(globalLog, __VA_ARGS__); \
if(mpiWorld->isHead() && globalLog != stdout) \
fprintf(stderr, __VA_ARGS__); \
finalizeSystem(false); \
exit(1); \
}
bool isHead() const
whether this is the root process (makes code more readable)
Definition: MPIUtil.h:51
MPIUtil * mpiWorld
MPI across all processes.

Quit with an error message (formatted using printf()). Must be called from all processes.

◆ die_alone

#define die_alone (   ...)
Value:
{ fprintf(globalLog, __VA_ARGS__); \
fflush(globalLog); \
if(mpiWorld->isHead() && globalLog != stdout) \
fprintf(stderr, __VA_ARGS__); \
mpiWorld->exit(1); \
}

Version of die that should only be used when it is impossible to guarantee synchronized calls from all processes.

◆ IS_NYQUIST

#define IS_NYQUIST   ( (!(2*iG[0]-S[0])) | (!(2*iG[1]-S[1])) | (!(2*iG[2]-S[2])) )

Determine if this is a nyquist component NOTE: no component is nyquist for odd sizes in this convention

◆ THREAD_fullGspaceLoop

#define THREAD_fullGspaceLoop (   code)
Value:
size_t i=iStart; \
vector3<int> iG( i / (S[2]*S[1]), (i/S[2]) % S[1], i % S[2] ); \
for(int j=0; j<3; j++) if(2*iG[j]>S[j]) iG[j]-=S[j]; \
while(true) \
{ \
code \
\
i++; if(i==iStop) break; \
iG[2]++; \
if(2*iG[2]>S[2]) iG[2]-=S[2]; \
if(iG[2]==0) \
{ iG[1]++; \
if(2*iG[1]>S[1]) iG[1]-=S[1]; \
if(iG[1]==0) \
{ iG[0]++; \
if(2*iG[0]>S[0]) iG[0]-=S[0]; \
} \
} \
}

One thread of a loop over G-space (see L_sub in PHLO.cpp for example)

◆ THREAD_halfGspaceLoop

#define THREAD_halfGspaceLoop (   code)
Value:
int size2 = S[2]/2+1; \
size_t i=iStart; \
vector3<int> iG( i / (size2*S[1]), (i/size2) % S[1], i % size2 ); \
for(int j=0; j<3; j++) if(2*iG[j]>S[j]) iG[j]-=S[j]; \
while(i<iStop) \
{ \
code \
\
i++; if(i==iStop) break; \
iG[2]++; \
if(iG[2]==size2) \
{ iG[2]=0; \
iG[1]++; \
if(2*iG[1]>S[1]) iG[1]-=S[1]; \
if(iG[1]==0) \
{ iG[0]++; \
if(2*iG[0]>S[0]) iG[0]-=S[0]; \
} \
} \
}

One thread of a loop over symmetry-reduced G-space (see applyFuncGsq_sub in Operators.h for example)

◆ THREAD_rLoop

#define THREAD_rLoop (   code)
Value:
size_t i=iStart; \
vector3<int> iv( \
i / (S[2]*S[1]), \
(i/S[2]) % S[1], \
i % S[2] ); \
while(i<iStop) \
{ \
code \
\
i++; if(i==iStop) break; \
iv[2]++; \
if(iv[2]==S[2]) \
{ iv[2]=0; \
iv[1]++; \
if(iv[1]==S[1]) \
{ iv[1] = 0; \
iv[0]++; \
} \
} \
}

One thread of a loop over real space (see applyFunc_r_sub for example)

◆ TIME

#define TIME (   title,
  fp,
  code 
)
Value:
{ double runTime = clock_us(); \
{ code } \
runTime = clock_us() - runTime; \
fprintf(fp, "%s took %.2le s.\n", title, runTime*1e-6); \
}

Elapsed time in seconds (from start of program)

Time a code section and print the timing (with title).
Code must be self-contained as a scope, as it will be surrounded by { }, and should not define "runTime"

Parameters
titleName to use when printing the timing info for the code block
fpStream to output the timing info to
codeThe code block to time (must be a self-contained scope)

Function Documentation

◆ assertStackTraceExit()

int assertStackTraceExit ( const char *  expr,
const char *  function,
const char *  file,
long  line 
)

stack trace on failed assertions

◆ expandYlmProd()

std::vector<YlmProdTerm> expandYlmProd ( int  lm1,
int  lm2 
)
inline

Real spherical harmonic expansion of product of two real spherical harmonics (effectively returns list of non-zero Clebsch-Gordon coefficients in the modified real Ylm basis)

◆ gpuErrorCheck()

void gpuErrorCheck ( )

Check for gpu errors and print a useful message (implemented in GpuUtils.cpp)

Check for gpu errors, and if any, abort with a useful messsage.

◆ gpuInit()

bool gpuInit ( FILE *  fpLog = stdout,
const class MPIUtil mpiHostGpu = 0,
double *  nGPUs = 0 
)

Must be called before any GPU use (preferably from main(), see isGpuMine) If mpiHostGpu (group of GPU processes on the same node) is specified, divide compatible GPUs amongst processes on same node, else select one with max memory nGPUs returns the number of physical GPUs used (fraction if a GPU is shared with other processes) Returns false on failure to find a suitable GPU

◆ isGpuMine()

bool isGpuMine ( )

Only the thread that called gpuInit() is allowed to use GPU resources This function will return true only on the one thread that rules the gpu.

◆ shouldThreadOperators()

bool shouldThreadOperators ( )

Operators should run multithreaded if this returns true, and should run in a single thread if this returns false. Note that all the thread launching functions in this file automatically prevent nested threading, so operator codes using those functions need not explicitly check this.

This only affects CPU threading, GPU operators should only be called from a single thread anyway.

◆ threadedAccumulate()

template<typename Callable , typename ... Args>
double threadedAccumulate ( Callable *  func,
size_t  nIter,
Args...  args 
)

A parallelized loop with an accumulated return value.

Same as threadedLoop, but func() returns double, which is summed over and returned

Returns
Accumulated return value of all calls to func()

◆ threadedLoop()

template<typename Callable , typename ... Args>
void threadedLoop ( Callable *  func,
size_t  nIter,
Args...  args 
)

A parallelized loop.

Given a callable object func and an argument list args, this routine calls func(i, args) for each i in [0:nIter-1], and accumulates the return values.

Note that the calls are made from multiple threads, so func must be thread safe. (Hint: pass mutexes as a part of args if synchronization is required).

As many threads as online processors are launched and the nIter iterations are evenly split between all the threads. Threaded loops will become single threaded if suspendOperatorThreading().

Parameters
funcThe function / object with operator() to be looped over
nIterThe number of loop 'iterations'
argsArguments to pass to func

◆ threadLaunch() [1/2]

template<typename Callable , typename ... Args>
void threadLaunch ( Callable *  func,
size_t  nJobs,
Args...  args 
)

Same as threadLaunch(int nThreads, Callable* func, size_t nJobs, Args... args) with nThreads = number of online processors.

◆ threadLaunch() [2/2]

template<typename Callable , typename ... Args>
void threadLaunch ( int  nThreads,
Callable *  func,
size_t  nJobs,
Args...  args 
)

A simple utility for running muliple threads.

Given a callable object func and an argument list args, this routine calls nThreads threads of func invoked as func(iMin, iMax, args). The nJobs jobs are evenly split between all the threads; each instance of func should handle job index i satisfying iMin <= i < iMax.

If nJobs <= 0, the behaviour changes: the function is invoked as func(iThread, nThreads, args) instead, where 0 <= iThread < nThreads. This mode allows for more flexible threading than the evenly split job management indicated above. This could be used as a convenient interface for launching threads for any parallel routine requiring as many threads as processors.

Parameters
nThreadsNumber of threads to launch (if <=0, as many as processors on system)
funcThe function / object with operator() to invoke in a multithreaded fashion
nJobsThe number of jobs to be split between the various func threads
argsArguments to pass to func

◆ YlmPrime() [1/2]

template<int l, int m>
__hostanddev__ vector3 YlmPrime ( const vector3<> &  qHat)

Derivative of spherical Harmonic with repect to iDir'th component of qHat with separate l,m indices.

Derivative of spherical Harmonic with repect to iDir'th component of qHat with combined lm index.

◆ YlmPrime() [2/2]

template<int lm>
__hostanddev__ vector3 YlmPrime ( const vector3<> &  qHat)

Derivative of spherical Harmonic with repect to iDir'th component of qHat with combined lm index.

Derivative of spherical Harmonic with repect to iDir'th component of qHat with combined lm index.