Procedures

Add an int64 variable which contains the FLOPs we use int64 since we might reach very large FLOP counts

this is the interface for the BLAS ASUM routines, it will call SASUM, DASUM, SCASUM, DZASUM Usage: result = blas_asum(x, incx) where x is the input vector, incx is the increment.

explicit interface for BLAS AXPY routines

explicit interface for BLAS COPY routines

explicit interface for BLAS DOT routines

explicit interface for BLAS GEMM routines

explicit interface for BLAS GEMV routines

explicit interface for BLAS IAMAX routines

explicit interface for BLAS SCAL routines

Close the log file, needs to be called at the end of the program

Get the current logger verbosity configuration

Configure the logger to be a certain verbosity level

Configure the logger to file to be a certain verbosity level

copy provides a blas-less implementation of xcopy where x is (i,s,d) icopy, scopy, dcopy if you built pic with BLAS use the copy interface provided there, I will not beat BLAS copy is implemented for (int32, int64, sp, dp) for 1 and 2d arrays of the same types

Log a message that will only be printed at the debug level of verbosity

this subroutine runs a random dgemm to create work so that timers and other testing utils work nicely

Log a message that will only be printed at the error of verbosity

fill provides a generic interface to assing a value alpha of types (int32, int64, sp, dp) as defined in pic_types.F90 The inteface supports filling 1d and 2d arrays of the specified variables

add flops to the flop rate type, this will add the flops to the flop recorder Usage: call my_flop_rate%add_flops(1000)

get the flop rate in GFLOP/s, this will calculate the flop rate based on the number of flops and the elapsed time

get the number of flops recorded in the flop rate type

get the elapsed time in seconds from the timer contained in the flop rate type

report the flop rate, this will print the flop rate in GFLOP/s

reset the flop rate type, this will reset the timer and the flop recorder

Calls the start method for the timer contained in the flop rate type

Calls the stop method for the timer contained in the flop rate type

get the first argument from the command line, this is expected to be a filename

Get the int64 number of FLOPs we currently have in the counter

Get the current precision for real numbers

Obtain the current precision being used to print variables to strings

get the current threading mode for the array routines Usage: mode = get_threading_mode()

get_threading_mode returns the current threading mode for the array routines Usage: mode = get_threading_mode()

Log a message that will only be printed at the info level of verbosity

The is equal interface is used to circumvent the fact that Fortran, rightfully, complains about comparing two reals without involving a tolerance, i.e. if(a == b) since there is no guarantee that the two reals are exactly equal due to floating point precision issues. The functions are defined as elemental

is_sorted provides a simple way to checking if a 1d array is sorted it is implemented for int32, int64, sp, and dp datatypes. The default is to check if an array is sorted in ascending fashion.

internal subroutines that processes the message and filters it according to the verbosity level set by the user or the default this is a private subroutine so it is not exposed to the user

function to pad a string with a certain number of characters for nice printing

adds a number X of spaces to the left of a “string” whcih is just a collection of characters. Mostly used for nice printing

Log a message that will only be printed at the performance of verbosity

general interface of the BLAS ASUM routines, will call SASUM, DASUM, SCASUM, DZASUM

general interface of the BLAS AXPY routines, will call SAXPY, DAXPY, CAXPY, ZAXPY

interface for single precision complex AXPY

interface for single precision complex copy

interface for single precision complex dot product

finalize the MPI library

initilalize the MPI library and get the size and rank variables

general interface of the BLAS COPY routines, will call SCOPY, DCOPY, CCOPY, ZCOPY

interface for single precision complex scaling

interface for double precision absolute sum

interface for double precision AXPY

interface for double precision copy

interface for double precision dot product

interface for single precision matrix multiplication

interface for double precision matrix-vector multiplication

general interface of the BLAS DOT routines, will call SDOT, DDOT, CDOTC, ZDOTC

interface for double precision scaling

interface for double precision complex absolute sum

general interface of the BLAS GEMM routines, will call SGEMM, DGEMM, CGEMM, ZGEMM

general interface of the BLAS GEMV routines, will call SGEMV, DGEMV, CGEMV, ZGEMV

general interface of the BLAS IAMAX routines, will call ISAMAX, IDAMAX, ICAMAX, IZAMAX

interface for single precision complex index of maximum absolute value

interface for double precision index of maximum absolute value

interface for single precision index of maximum absolute value

interface for double precision complex index of maximum absolute value

my cool banner, work in progress

interface for single precision absolute sum

interface for single precision AXPY

general interface of the BLAS SCAL routines, will call SSCAL, DSCAL, CSCAL, ZSCAL

interface for single precision complex absolute sum

interface for single precision copy

interface for single precision dot product

interface for single precision matrix multiplication

interface for single precision matrix-vector multiplication

interface for single precision scaling

pic_sum provides a threaded alternative to the sum(array) Fortran intrinsic which will be too slow for large sizes of vectors and matrices. Note that this provides the total sum. As opposed to the blas alternative XASUM which does the absolute sum

pic_transpose provides a blas-less, threaded alternative to the Fortran transpose intrinsic which will be slow for large matrix sizes. pic_transpose does not assume symmetric matrices

interface for double precision complex AXPY

interface for double precision complex copy

interface for double precision complex dot product

interface for single precision matrix multiplication

interface for double precision complex scaling

Print a 3D tensor of dp values

Print a 3D tensor of sp values

general interface for printing a one or two dimensional array will be deprecated

Generic interface for printing arrays of different types

general interface to print an array within certain bounds

prints a convenient row of asterisks of length n

print a matrix in a given format

print a matrix of dp values

private subroutine to print a matrix in format

print a matrix of int32 values

print a matrix of int64 values

Print a matrix of n_cols by n_rows

print a matrix of sp values

Print a packed lower triangular matrix of dp values

Print a packed lower triangular matrix of int32 values

Print a packed lower triangular matrix of int64 values

Print a packed lower triangular matrix of sp values

private subroutine that prints a plain matrix of n_cols by n_rows

private subroutine that prints a vector of n_elements

print a vector

print a vector of dp values

private subroutine that prints a vector in a format

print a vector of int32 values

print a vector of int64 values

print a vector from start up to n_elements

print a vector of sp values

Resets the flop counter to 0_int64

Set brackets based on output format type

Set the precision for real numbers

This routine overrides the default dp precision used for printing strings in the to_string function, the default is : integer(default_int), parameter :: default_dp_precision = 12

set the threading mode for the array routines, this will set the use_threaded variable to true or false depending on the input

set_threading sets the threading mode for the array routines this will set the use_threaded variable to true or false depending on the input Usage: call set_threading_mode(.true.) or call set_threading_mode(.false.)

Returns the elapsed time as a real(dp) variable

Prints the elapsed time at the time of calling

starts the timer. If OMP is enabled, it will use omp_get_wtime() if not, it will use Fortran’s system_clock

stop the timer. If OMP is enabled, it will use omp_get_wtime() if not, it will use Fortran’s system_clock

converts a variable of type (int32, int64, sp, dp, char, logical) to a “string” which is just a collecting of chars.

transform a character to a string

transform a real(dp) to a string

transform an int32 to a string

transform an int64 to a string

tranform a logical to a string either true or false

transform a real(sp) to a string

takes a character variable and transforms it to uppercase

Log a message that will only be printed at the verbose level of verbosity

Log a message that will only be printed at the warning level of verbosity

Internal subroutine that will write the message to the log no interface to the public