02 Fortran 1x2
02 Fortran 1x2
02 Fortran 1x2
Why Fortran? - A lot of legacy software written in it (well, often in Fortran 77). - High-level abstraction (as compared to C); e.g. handling of arrays. - Large number of numerical libraries built for Fortran. Why not Fortran? - Compilers (particularly free ones) not so easy to find on different environments. - Some things easier to implement in C. Often a question of personal taste and the fact which language you learned first. Different generations of Fortran1: 1. FORTRAN66 2. FORTRAN77 3. Fortran 90 4. Fortran 95 5. Fortran 2003 6. Fortran 2008
These two must not be used anymore! These two are now in use. No large differences between F90 and F95. Most features implemented by most compilers. Some features implemented by many compilers.
- From now on I call the language simply Fortran or F90+ (meaning all other version except FORTRAN77) - If there is need to distinguish between 90, 95, 2003, and 2008 versions will tell it explicitly. - A good source of compiler support of the latest Fortran features can be found at http://www.fortranplus.co.uk/resources/fortran_2003_2008_compiler_support.pdf
1. For more info on history of programming languages see e.g. http://www.levenez.com/lang/ Tools for High Performance Computing 2013: 2. Introduction to Fortran 1
Introduction to Fortran
Material - Fortran 95/2003 Explained by Michael Metcalf, John Reid and Malcolm Cohen, Oxford University Press - A newer version Modern Fortran Explained by the same authors and publisher - CSC has published a free Fortran 95/2003 guide in Finnish (see http://www.csc.fi/csc/julkaisut/oppaat) - gfortran home page: http://gcc.gnu.org/wiki/GFortran - Small example programs presented in these lecture notes can be downloaded from http://www.physics.helsinki.fi/courses/s/stltk/progs/ - IBM has a nice reference manual for their compiler at http://www-01.ibm.com/software/awdtools/fortran/xlfortran/library/ (choose the XL Fortran Advanced Edition for Linux PDF files and there Language Reference...) - Note, however, that it is a language reference manual not a textbook for learning Fortran.
Introduction to Fortran
Structure of a Fortran program: program name_of_the_program ! This is a comment implicit none type and variable definitions . . . executable statements . . . stop contains ! Comment again local subroutine definitions . . . end program name_of_the_program - implicit none is the most important statement in F90 - In older Fortrans implicit type definitions were commonly used: variables beginning with i,j,k,l,m,n are integers others are real. - If you leave out implicit none the same convention is used in F90+! Then...you get what you deserve!
- Subroutines defined after contains statement are local in the sense that they are not visible to any other routines and that they see the local variables. - Comments begin with exclamation mark. The rest of the line is ignored (just as the //-comment in C++).
Urban legend (?): Fortran compilers used to ignore white space. This combined with implicit typing caused a typo go unnoticed: in a do loop start a period was written instead of a comma: do 100 i=1,10 do 100 i=1.10 . This was interpreted as an assignment: do100i=1.10 i.e. no loop at all!
Introduction to Fortran
Structure of a Fortran procedure definition subroutine sub(p1,p2) implicit none procedure parameter definitions local type and variable definitions . . . executable statements . . . return contains ! Comment again local subroutine definitions . . . end subroutine sub function func(p1,p2) implicit none definitions of the function and it parameters local type and variable definitions . . . executable statements . . . func=... ! assign a return value return contains ! Comment again local subroutine definitions . . . end function func
subroutine never returns a value (as a C function defined as void) - Refer to subroutine as call sub(p1,p2) function always returns a value - Refer to function simply by using it in an expression or as a procedure parameter x=func(p1,p2)
Introduction to Fortran
Built-in datatypes - Integer integer,parameter :: ik=selected_int_kind(9) integer (kind=ik) :: i - With the kind keyword and intrinsic function selected_int_kind() one can select the size of the datatype: selected_int_kind(p) returns the kind type that represents all integers n in the range 10 p - Example
program intkind implicit none integer,parameter :: ik=selected_int_kind(2) integer(kind=ik) :: i integer :: j i=1 do j=1,20 i=i*2 write(0,*) j,i end do stop end program intkind progs> gfortran intkind.f90 progs> ./a.out 1 2 2 4 3 8 4 16 5 32 6 64 7 -128 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0
10 p .
- Default integer (i.e. if you say only integer :: i ) is in most environments of size 32 bits (corresponding to p = 9 ).
Tools for High Performance Computing 2013: 2. Introduction to Fortran 5
Introduction to Fortran
- Real integer,parameter :: rk=selected_real_kind(6,15) real(kind=rk) :: x (or more simply real(rk) :: x) - Parameters of the function selected_real_kind(p,r) tell the minimum decimal precision (p) and range of the exponent (r) of the real type. - Example
progs> gfortran realkind.f90 program realkind progs> a.out implicit none 1 100.0000 integer,parameter :: rk=selected_real_kind(6,15) 2 10000.00 real(kind=rk) :: x 3 1000000. integer :: j 4 1.0000000E+08 5 1.0000000E+10 x=1.0 ... do j=1,64 18 1.0000001E+36 x=x*100.0 19 1.0000001E+38 write(0,*) j,x 20 +Infinity end do 21 +Infinity 22 +Infinity stop ... end program realkind
- Default real (using only real :: x) is in most systems 32 bit long (i.e. corresponds to float in C). - But not always! - This kind-qualifier is exactly the feature that should prevent problems with different word lengths. - By using it your code is easier to port to other systems. (Well, in practice, most machines today use IEEE floating point numbers)
Tools for High Performance Computing 2013: 2. Introduction to Fortran 6
Introduction to Fortran
- Complex integer,parameter :: rk=selected_real_kind(10,40) complex (kind=rk) :: z z=(1.0,1.0) - Boolean logical :: firsttime=.true. ... if (firsttime) then ... firsttime=.false. end if - Strings character (len=80) :: c c=No, joo! write(0,*) c - Substring: c(5:10), c(:4), c(11:) - Note that string handling in F90 is much easier than in C, methinks:
Assigning and concatenation character(len=80) :: c,d,e c=first d=last e=trim(c)// and //trim(d) write(0,*) e Comparison character(len=80) :: c,d c=first ... d=last if (c==d) then write(0,*) equal! endif
Function
Introduction to Fortran
- Own type definitions (structs) program typedeftest implicit none type person character(len=20) :: first,last integer :: age end type person type(person) :: student student%first=Matti student%last=Meiklinen student%age=75 ! Not sure whether outputting the whole ! struct is standard comforming. write(6,*) student stop end program typedeftest
75
Introduction to Fortran
Conditional execution If-then-else Case program casetest implicit none integer :: i character(len=10) :: c read(5,*) i select case(i) case(1) c=one case(2) c=two case(3) c=three case default c=many end select write(0,*) c stop end program casetest
if (logical expression) then executable statements else if (logical expression) then executable statements else executable statements end if
In case there is only one statement after the if construction then-endif pair can be left out; e.g.; if (i>10) a(i)=0.0
- Instead of integers you can also use logical and character expressions in the case construct. - Indivdual cases can also be lists of values and ranges, e.g. case(4:10,12,14) - No need to use break commands as in C.
Introduction to Fortran
Comparison operators - In FORTRAN77 the comparison operators had the dotted forms given below in the last column: Operator
equal not equal less greater less or equal greater or equal
New form
== /= < > <= >=
Old form
.eq. .ne. .lt. .gt. .le. .ge.
- Both forms can be used in F90+. Logical operators - These only have the dotted forms. Operator
Negation Logical and Logical or Logical equivalence Logical inequivalence
Examples: Fortran
.not. .and. .or. .eqv. .neqv.
10
Introduction to Fortran
Iteration (do loop) [name:] do iter_var=first,last[,step] executable statements end do [name] - Inside the do loop you can use the statement exit to stop the iteration (jump out of the loop) or cycle to stop the current iteration and continue the next one. - Examples: loop1: do i=1,n do j=i,m if (a(j)>b(i)) then a(j)=b(i) cycle loop1 end if end do end do loop1 Do loop without an iteration variable: do executable statements ... if (x>-1.0) exit ! To prevent an infinite loop end do
- In F77 real iteration variables were allowed. This is no more the case in F90+. - Naming the loops is helpful if you have many loops within each other and want e.g. jump out from the outermost one. - It also makes the code more readable.
11
Introduction to Fortran
Arrays - In F77 only statically allocated arrays were allowed. - F90 introduced allocatable arrays. Static allocation integer,dimension(10) :: k real,dimension(10,10) :: x real,dimension(100) :: x1 real(kind=rk),dimension(4,4,4) :: y Or more tersely integer :: k(10) real :: x(10,10),x1(100) real(kind=rk) :: y(4,4,4) Dynamic allocation real(kind=rk),dimension(:),allocatable :: x real(kind=rk),dimension(:,:),allocatable :: y integer :: n,err ... print *,Give array size. read *,n allocate(x(n),y(0:n,0:n),stat=err) if (err/=0) then print *,"Could not allocate." stop end if ... x=1.0 y=2.0 ... deallocate(x,y)
- Some restrictions of allocatable arrays: they cant be used in 1. formal parameters of subroutines (use modules instead) 2. function results 3. user datatypes (use pointers instead)
Tools for High Performance Computing 2013: 2. Introduction to Fortran 12
Introduction to Fortran
- Arrays or sections of arrays can be easily handled: real :: x(100),y(100),z(100) x=1.0 x(4:6)=2.0 y=2.0 z=x*y ! Multiply element by element - More about array handling later. Pointers (material from http://www.personal.psu.edu/faculty/j/h/jhm/f90/lectures/42.html) - Pointers in Fortran are more abstract than those in C. - Those variables that you want to be pointed by pointers must have the target qualifier: real, target :: a, b(1000), c(10,10) integer, target :: i, j(100), k(20,20) - Pointers itself have the pointer qualifier: real, pointer :: pa, aptr, pb(:), pc1(:), pc2(:,:) - To associate a pointer with a variable you use the => operator pa => a - Now you can use pa and a interchangibly: pa = 1.23456 print *, a = , a - With array pointer you can do tricks as pb => b pb => b(101:200) pc2 => c(3:5, 4:6) - Pointers in F90+ can be seen as a generalization of the equivalence1 statement in F77. - Personal opinion: Not much needed in HPC.
1.
equivalence(x,y)-->
code.
variables x and y can now be used interchangibly. In F77, however, this is a static definition. A good tool for making obfuscated 13
Introduction to Fortran
Reading and writing files - Open a file: unit number identifies an open file (cf. C file pointer) integer :: st character (len=40) :: fname real :: x ... fname=in.dat open(unit=10,file=fname,status=old,iostat=st) if (st/=0) then print *,Error in opening file. stop end if - Reading from a file
n=0 do read(unit=10,iostat=st,fmt=(a)) string if (st<0) exit ! end of file if (string(1:1)=="#") cycle ! comment line if (len_trim(string)==0) cycle ! ignore empty lines read(unit=string ,fmt=*,iostat=st) x,y,z if (st/=0) then print *,"line <",trim(string),"> in wrong format" stop end if n=n+1 atom(n)%x=x atom(n)%y=y atom(n)%z=z end do close(unit=10)
file should exist file should not exist if the files exists it is deleted and a new one is created
Specifier iostat=st set the status value of the io operation to integer variable st. st==0 indicates success st<0 means that EOF occured
14
Introduction to Fortran
- Writing to a file (formatted, i.e. ASCII) open(unit=10,file=fname,status=replace,iostat=ios) write(unit=10,*) x =,x close(unit=10) - It is usual in open, close, read, and write statements to leave out the keywords unit and fmt open(unit=10,file=fname) open(10,file=fname) close(unit=10) close(10) write(unit=10,fmt=*) write(10,*) read(unit=10,fmt=(a)) read(10,(a)) - The format string defined by keyword fmt determines in which form data is read in and printed out (cf. format string in sprintf in C; see next slide) - Probably the easiest way (in the beginning at least) is to use so called list-directed formatting; i.e. using * as the format string. - In this case the system decides what format to use based on what is being printed or read; e.g. write(6,*) Time step ,dt read(5,*) x - Predetermined units: 0 = stderr 5 = stdin 6 = stdout - These units need not be explicitly opened. - Units 5 and 6 can be replaced by * in reading and writing, respectively. - Output to stdout can also be done using the print statement: print format_specifier,io_list - E.g. print *,X is now ,x
Tools for High Performance Computing 2013: 2. Introduction to Fortran 15
Introduction to Fortran
- The most important format specifiers: Writing
Specifier iw fw.d ew.d gw.d aw a nx / Meaning write an integer to next w positions write a real with d decimals to next w positions write a real with d decimals in exponent form to next w positions write a real with d decimals in normal or exponent form to next w positions write a string to next w positions write a string starting from the next position skip n next positions line break
Reading
Specifier iw fw.d ew.d aw a nx / Meaning read an integer from next w positions read next w characters as a real with d decimals in case there is no period same as above read next w characters as a string read as many characters as is needed to fill the character variable skip n next positions line break
- You can use these for detailed formatting of io but the easiest way is to use the list-directed io (fmt=*)
Tools for High Performance Computing 2013: 2. Introduction to Fortran 16
Introduction to Fortran
- A simple example of reading from stdin and writing to stdout: program readtest implicit none real :: x integer :: i,s character (len=20) :: c do print *,Give x,i,c. read(*,*,iostat=s) x,i,c if (s/=0) exit print *,2*x,10*i,len_trim(c) end do print *,Bye! stop end program readtest len_trim(c) is an intrinsic
function returning the length of the string without the trailing spaces
17
Introduction to Fortran
Procedures - Fortran has two kinds of procedures functions that return a value subroutines that return nothing - Functions are called as they are in C: as a part of an expression x=sin(th) y=myfunc(i,j,z) - Definition of a function; example
function area(r) implicit none integer,parameter :: rk=selected_real_kind(10,40) real(kind=rk) :: area real(kind=rk),intent(in) :: r area=4.0*atan2(1.0,1.0)*r**2 return end function area
- With intent(in), intent(out), intent(inout) you can specify whether the function or subroutine parameters are only input, only output or both input and output parameters, respectively. For example:
program intenttest real :: x,y,f x=1.0 y=f(x) stop end program intenttest function f(x) real :: f real,intent(in) :: x x=2.0*x f=x return end function f
Compilation:
progs> gfortran intenttest.f90 In file intenttest.f90:11 x=2.0*x 1 Error: Can't assign to INTENT(IN) variable 'x' at (1)
18
Introduction to Fortran
- Subroutines are called by statement call: call mysub(a,b,i,j) - Definition of a subroutine; example
subroutine sub(x,y,n) implicit none integer,parameter :: rk=selected_real_kind(10,40) integer,intent(in) :: n real(kind=rk),intent(inout) :: x(n) real(kind=rk),intent(inout) :: y(0:n,0:n) x=2.0*x y=2.0*y return end subroutine sub
- And call
call sub(x,y,n)
- Just as in C it is advisable to use function prototypes to get rid of errors in parameters, in Fortran one can use interface blocks in those program units where the procedure is called:
interface interface_body_1 interface_body_2 ... end interface
- Each interface body consists of the procedure definition without the actual statements; see on the right.
interface function area(r) implicit none integer,parameter :: rk=selected_real_kind(10,40) real(kind=rk) :: area real(kind=rk),intent(in) :: r end function area subroutine sub(x,y,n) implicit none integer,parameter :: rk=selected_real_kind(10,40) integer,intent(in) :: n real(kind=rk),intent(inout) :: x(n) real(kind=rk),intent(inout) :: y(0:n,0:n) end subroutine sub end interface
19
Introduction to Fortran
- An array of a character string as a parameter of a procedure can be of assumed length or shape. - In this case the procedure interface has to be written out explicitly.
program assumedtest implicit none interface subroutine sub(x,y) implicit none real :: x(:),y(:,:) end subroutine sub end interface real :: x(3),y(3,3) x=1.0 y=2.0 call sub(x,y) print *,x print *,y stop end program assumedtest subroutine sub(x,y) implicit none real :: x(:),y(:,:) print *,size(x),size(y) x=2.0*x y=3.0*y return end subroutine sub progs> ifort assumedtest.f90 progs> a.out 3 9 2.000000 2.000000 6.000000 6.000000 6.000000 6.000000
6.000000 6.000000
6.000000
Personal opinion: The clearest way to pass arrays to subroutines is to pass also the sizes; for example:
call sub(x,y,3) ... subroutine sub(x,y,n) implicit none integer :: n real :: x(n),y(n,n) print *,size(x),size(y) x=2.0*x y=3.0*y return end subroutine sub
20
Introduction to Fortran
- If you want to save the value of a local procedure variable between procedure calls you can use the (you guessed it!) save specifier. - For example your subroutine needs to do some initialization only once: subroutine sub(x,y,n) ... logical,save :: firsttime=.true. ... if (firsttime) then ! Initialization ... firsttime=.false. endif ! Normal execution ...
- Same as static in C.
21
Introduction to Fortran
- Argument association may be positional (normal way) or by keywords:
program argassoc implicit none integer :: i,j i=1;j=2 call sub(i,j) ! Positional write(0,*) i,j i=1;j=2 call sub(arg1=i,arg2=j) ! By keyword write(0,*) i,j i=1;j=2 call sub(arg1=j,arg2=i) ! By keyword write(0,*) i,j stop end program argassoc subroutine sub(arg1,arg2) implicit none integer,intent(inout) :: arg1,arg2 arg1=arg1*arg2 arg2=-arg1 return end subroutine sub
22
Introduction to Fortran
Modules - Modules is one of the most handy features of F90+. - It replaces the old Fortran common areas but it is not restricted to that. - Modules can contain global variables, constants, commonly used procedures. - Example:
module stuff implicit none integer,parameter ::& & rk=selected_real_kind(p=15,r=100) real(kind=rk) :: pi=3.14159265358979 contains function area(r) real(kind=rk),intent(in) :: r real(kind=rk) :: area area=pi*r**2 end function area subroutine readr(r) real(kind=rk),intent(out) :: r print *,"Give r" read *,r end subroutine readr end module stuff program moduleexample use stuff implicit none real(kind=rk) :: r call readr(r) print *,"Radius : " , r print *,"Circumference : " ,2.0*pi*r print *,"Area : " , area(r) end program moduleexample
- You can only use a part of the module by giving the only qualifier: use stuff, only : rk - The name of the object in the module can also be changed: use stuff, only : rk => real_type
Tools for High Performance Computing 2013: 2. Introduction to Fortran 23
Introduction to Fortran
Main differences between Fortran and C Fortran Array indexing a() starts from 1 columnwise C a[] starts from 0 rowwise
address
value
- It is possible to use both Fortran and C in the same program but the Fortran-C interface is not (yet) standardized.
24
Introduction to Fortran
Intrinsic procedures - Subroutines and functions that are a part of the language standard - Math functions - Generic and specific names; e.g. sqrt(), dsqrt(), csqrt() - Generic names: type of result depends on the argument:
program genericfunctionname implicit none integer,parameter :: r1=selected_real_kind(5,10) integer,parameter :: r2=selected_real_kind(10,40) print print print print *,sqrt(2.0_r1) *,sqrt(2.0_r2) *,kind(sqrt(2.0_r1)) *,kind(sqrt(2.0_r2)) progs> ifort genericfunctionname.f90 progs> a.out 1.414214 1.41421356237310 4 8
- No reason to use specific names! - The most common math functions included.
25
Introduction to Fortran
- Whats going on here?
program pi_accuracy integer,parameter :: rk0=selected_real_kind(5,10) integer,parameter :: rk1=selected_real_kind(10,40) integer,parameter :: rk2=selected_real_kind(30,200) real(rk2) :: pi0 real(rk2) :: pi1 real(rk2) :: pi2 pi0=4.0*atan2(1.0_rk0,1.0_rk0) pi1=4.0*atan2(1.0_rk1,1.0_rk1) pi2=4.0*atan2(1.0_rk2,1.0_rk2) print *,pi0 print *,pi1 print *,pi2 stop end program pi_accuracy
26
Introduction to Fortran
- Below is a list of the standard intrinsic functions (from Intel Fortran compiler Language Reference Manual, /opt/intel_fc_80/doc/for_lang.pdf):
Function name ABS(a) ACHAR(i) ACOS(x) ADJUSTL(string) ADJUSTR(string) AIMAG(z) AINT(a[,kind]) ALL(mask[,dim]) ALLOCATED(array) ANINT(a[,kind]) ANY(mask[,dim]) ASIN(x) ASSOCIATED(pointer[,target]) ATAN(x) ATAN2(y,x) BIT_SIZE(i) BTEST(i,pos)
Tools for High Performance Computing 2013: 2. Introduction to Fortran
Classa E E E E E E E T I E T E I E E I E
Description The absolute value of an argument The character in the specified position of the ASCII character set The arccosine(in radians) of the argument The specified string with leading blanks removed and placed at the end of the string The specified string with trailing blanks removed and placed at the beginning of the string The imaginary part of a complex argument A real value truncated to a whole number .TRUE. if all elements of the masked array are true The allocation status of the argument array A real value rounded to a whole number .TRUE. if any elements of the masked array are true The arcsine(in radians) of the argument .TRUE. if the pointer argument is associated or the pointer is associated with the specified target The arctangent(in radians) of the argument The arctangent(in radians) of the arguments The number of bits(s) in the bit model .TRUE. if the specified position of argument I is one
27
Function name CEILING(a[,kind]) CHAR(i[,kind]) CONJG(z) COS(x) COSH(x) COUNT(mask[,dim][,kind]) CSHIFT(array,shift[,dim]) DBLE(a) DIGITS(x) DIM(x,y) DOT_PRODUCT(vector_a,vector_b) EOSHIFT(array,shift[,boundary][,dim]) EPSILON(x) EXP(x) EXPONENT(x) FLOOR(a[,kind]) FRACTION(x) HUGE(x) IACHAR(c) IAND(i,j) IBCLR(i,pos)
Tools for High Performance Computing 2013: 2. Introduction to Fortran
Classa E E E E E T T E I E T T I E E E E I E E E
Description The smallest integer greater than or equal to the argument value The character in the specified position of the processor character set The conjugate of a complex number The cosine of the argument, which is in radians The hyperbolic cosine of the argument The number of .TRUE. elements in the argument array An array that has the elements of the argument array circularly shifted The corresponding double precision value of the argument integer argument The number of significant digits in the model for the argument The positive difference between the two arguments The dot product of two rank-one arrays(also called a vector multiply function) An array that has the elements of the argument array end-off shifted The number that is almost negligible when compared to one The exponential ex for the argument x The value of the exponent part of a real argument The largest integer less than or equal to the argument value The fractional part of a real argument The largest number in the model for the argument The position of the specified character in the ASCII character set The logical AND of the two arguments The specified position of argument I cleared(set to zero)
28
Function name IBITS(i,pos,len) IBSET(i,pos) ICHAR(c) IEOR(i,j) INDEX(string,substring[,back][,kind]) INT(a[,kind]) IOR(i,j) ISHFT(i,shift) ISHFTC(i,shift[,size]) KIND(x) LBOUND(array[,dim][,kind]) LEN(string[,kind]) LEN_TRIM(string[,kind]) LGE(string_a,string_b) LGT(string_a,string_b LLE(string_a,string_b) LLT(string_a,string_b) LOG(x) LOG10(x) LOGICAL(l[,kind]) MATMUL(matrix_a,matrix_b) MAX(a1,a2[,a3,...])
Tools for High Performance Computing 2013: 2. Introduction to Fortran
Classa E E E E E E E E E I I I E E E E E E E E T E
Description The specified substring of bits of argument I The specified bit in argument I set to one The position of the specified character in the processor character set The logical exclusive OR of the corresponding bit arguments The position of the specified substring in a character expression The corresponding integer value(truncated) of the argument The logical inclusive OR of the corresponding bit arguments The logical end-off shift of the bits in argument I The logical circular shift of the bits in argument I The kind type parameter of the argument The lower bounds of an array(or one of its dimensions) The length(number of characters) of the argument character string The length of the specified string without trailing blanks A logical value determined by a > or = comparison of the arguments A logical value determined by a > comparison of the arguments A logical value determined by a < or = comparison of the arguments A logical value determined by a < comparison of the arguments The natural logarithm of the argument The common logarithm(base 10) of the argument The logical value of the argument converted to a logical of type KIND The result of matrix multiplication(also called a matrix multiply function) The maximum value in the set of arguments
29
Function name MAXEXPONENT(x) MAXLOC(array[,dim][,mask][,kind]) MAXVAL(array[,dim][,mask]) MERGE(tsource,fsource,mask) MIN(a1,a2[,a3,...]) MINEXPONENT(x) MINLOC(array[,dim][,mask][,kind]) MINVAL(array[,dim][,mask]) MOD(a,p) MODULO(a,p) NEAREST(x,s) NINT(a[,kind]) NOT(i) NULL([mold]) PACK(array,mask[,vector]) PRECISION(x) PRESENT(a) PRODUCT(array[,dim][,mask]) RADIX(x) RANGE(x)
Tools for High Performance Computing 2013: 2. Introduction to Fortran
Classa I T T E E I T T E E E E E T T I I T I I
Description The maximum exponent in the model for the argument The rank-one array that has the location of the maximum element in the argument array The maximum value of the elements in the argument array An array that is the combination of two conformable arrays(under a mask) The minimum value in the set of arguments The minimum exponent in the model for the argument The rank-one array that has the location of the minimum element in the argument array The minimum value of the elements in the argument array The remainder of the arguments(has the sign of the first argument) The modulo of the arguments(has the sign of the second argument) The nearest different machine-representable number in a given direction A real value rounded to the nearest integer The logical complement of the argument A disassociated pointer A packed array of rank one(under a mask) The decimal precision(real or complex) of the argument .TRUE. if an actual argument has been provided for an optional dummy argument The product of the elements of the argument array The base of the model for the argument The decimal exponent range of the model for the argument
30
Function name REAL(a[,kind]) REPEAT(string,ncopies) RESHAPE(source,shape[,pad][,order]) RRSPACING(x) SCALE(x,I) SCAN(string,SET[,back][,kind]) SELECTED_INT_KIND(r) SELECTED_REAL_KIND([p][,r]) SET_EXPONENT(x,i) SHAPE(SOURCE[,kind]) SIGN(a,b) SIN(x) SINH(x) SIZE(array[,dim][,kind]) SPACING(x) SPREAD(source,dim,ncopies) SQRT(x) SUM(array[,dim][,mask]) TAN(x)
Tools for High Performance Computing 2013: 2. Introduction to Fortran
Classa E T T E E E T T E I E E E I E T E T E
Description The corresponding real value of the argument The concatenation of zero or more copies of the specified string An array that has a different shape than the argument array, but the same elements The reciprocal of the relative spacing near the argument The value of the exponent part(of the model for the argument) changed by a specified value The position of the specified character(or set of characters) within a string The integer kind parameter of the argument The real kind parameter of the argument; one of the optional arguments must be specified The value of the exponent part(of the model for the argument) set to a specified value The shape(rank and extents) of an array or scalar A value with the sign transferred from its second argument The sine of the argument, which is in radians The hyperbolic sine of the argument The size (total number of elements) of the argument array(or one of its dimensions) The value of the absolute spacing of model numbers near the argument A replicated array that has an added dimension The square root of the argument The sum of the elements of the argument array The tangent of the argument, which is in radians
31
Function name TANH(x) TINY(x) TRANSFER(source,mold[,size]) TRANSPOSE(matrix) TRIM(string) UBOUND(array[,dim][,kind]) UNPACK(vector,mask,field) VERIFY(string,set[,back][,kind])
a. Class: E=elemental, T=transformational, I=inquiry
Classa E I T T T I T E
Description The hyperbolic tangent of the argument The smallest positive number in the model for the argument The bit pattern of SOURCE converted to the type and kind parameters of MOLD The matrix transpose for the rank-two argument array The argument with trailing blanks removed The upper bounds of an array(or one of its dimensions) An array(under a mask) unpacked from a rank-one array The position of the first character in a string that does not appear in the given set of characters
- Below is a list of the standard intrinsic subroutines (from Intel Fortran compiler Language Reference Manual, /opt/intel_fc_80/doc/for_lang.pdf):
Description The processor time in seconds Date and time information from the real-time clock A sequence of bits(bit field) is copied from one location to another A pseudorandom number taken from a sequence of pseudorandom numbers uniformly distributed within the range 0.0 to 1.0 Initializes or retrieves the pseudorandom number generator seed value Data from the processors real-time clock
32
Introduction to Fortran
- Inquiry functions include functions describing the floating point system used: program fpinquiryfunctions implicit none integer,parameter :: r1=selected_real_kind(5,10) integer,parameter :: r2=selected_real_kind(10,40) integer,parameter :: r3=selected_real_kind(30,200) real(kind=r1) :: x real(kind=r2) :: y real(kind=r3) :: z print *,digits(x),digits(y),digits(z) print *,epsilon(x),epsilon(y),epsilon(z) print *,huge(x),huge(y),huge(z) - Also inquiries about the integer system: program bitsizeexample integer,parameter i1=selected_int_kind(3) integer,parameter i2=selected_int_kind(6) integer,parameter i3=selected_int_kind(12) integer(kind=i1) :: i integer(kind=i2) :: j
Tools for High Performance Computing 2013: 2. Introduction to Fortran
progs> ifort fpinquiryfunctions.f90 progs> a.out 24 53 113 1.1920929E-07 2.220446049250313E016 1.925929944387235853055977942584927 E-0034 3.4028235E+38 1.797693134862316E+308
:: :: ::
64
33
Introduction to Fortran
- Functions for array handling (one of the greatest features of F90+):
program anyexample implicit none integer :: i,j,a(10) do i=1,10 a(i)=i end do print '(10(i0,x))',a if (any(a(5:10)>9)) print '(a)','yes' if (all(a(5:10)>9)) print '(a)','yes' print '(10l1)',a(5:10)>9 stop end program anyexample
program dotproductexample implicit none integer,parameter :: & & rk=selected_real_kind(10,40) real(kind=rk) :: x(5)=[1.0,1.0,2.0,4.0,7.0] real(kind=rk) :: y(5)=[2.0,3.0,1.0,0.0,5.0] print '(100f5.2)',x print '(100f5.2)',y print '(100f5.2)',dot_product(x,y) stop end program dotproductexample
progs> ifort dotproductexample.f90 progs> ./a.out 1.00 1.00 2.00 4.00 7.00 2.00 3.00 1.00 0.00 5.00 42.00
34
Introduction to Fortran
- More examples of array functions
program matmulexample implicit none integer,parameter ::& & rk=selected_real_kind(10,40) real(kind=rk) :: x(3,2),y(2,3),z(3,3) character(len=80) :: fm='(a,x,100f6.2)' x=reshape([1.0,2.0,4.0,5.0,9.0,1.0],[3,2]) y=reshape([7.0,4.0,4.0,6.0,1.0,0.0],[2,3]) print fm,'x',x print * print fm,'y',y z=matmul(x,y) print * print fm,'z',z print * print fm,'max',maxloc(z),maxval(z) stop end program matmulexample progs> ifort matmulexample.f90 progs> ./a.out x 1.00 2.00 4.00 5.00 9.00 y z 7.00 4.00 4.00 6.00 1.00
max
35
Introduction to Fortran
- Examples of string handling functions:
program stringhandling character(len=20) :: s1 read *,s1 print *,|,s1,| print *,|,trim(s1),| print *,len(s1),len_trim(s1) stop end program stringhandling progs> ifort stringhandling.f90 progs> a.out aaabbbccc |aaabbbccc | |aaabbbccc| 20 9
36
Introduction to Fortran
- When optimizing programs it is essential to measure the CPU time used:
program cputime implicit none integer,parameter ::& & rk=selected_real_kind(10,40),NMAX=100000000 real(kind=rk) :: x,t1,t2 integer :: n x=0.0 call cpu_time(t1) do n=NMAX,1,-1 x=x+1.0_rk/real(n,rk) end do call cpu_time(t2) print *,Result ,x print *,CPU time in seconds ,t2-t1 stop end program cputime
progs> ifort cputime.f90 progs> a.out Result 18.9978964138534 CPU time in seconds 1.63975200000000
37
Introduction to Fortran
F90+ has two constructs that helps the handling of arrays. - With the where construct you can modify arrays without using explicit do-loops:
program whereexample implicit none integer,parameter ::& & rk=selected_real_kind(10,40),i1=1,i2=10 real(kind=rk),dimension(i1:i2) :: x,y integer :: n do n=i1,i2 x(n)=n-i2/2 end do where (x<0.0) y=-10.0 elsewhere y=0.0 end where print (10f6.1),x print (10f6.1),y stop end program whereexample
mask expression
progs> ifort whereexample.f90 progs> ./a.out -4.0 -3.0 -2.0 -1.0 0.0 -10.0 -10.0 -10.0 -10.0 0.0
1.0 0.0
2.0 0.0
3.0 0.0
4.0 0.0
5.0 0.0
38
Introduction to Fortran
- In the where construct the mask expression can be any logical expression that has the same shape as the left hand sides of the assignments inside the construct:
program whereexample1 implicit none integer,parameter :: & & rk=selected_real_kind(10,40),i1=1,i2=10 real(kind=rk),dimension(i1:i2) :: x logical :: mask(i1:i2) integer :: n x=1.0 mask=.false. mask(i1:i2:2)=.true. where (mask) x=-1.0 end where print (50l6),mask print (50f6.1),x stop end program whereexample1 progs> ifort whereexample1.f90 progs> ./a.out T F T F T -1.0 1.0 -1.0 1.0 -1.0
F 1.0
T -1.0
F 1.0
T -1.0
F 1.0
39
Introduction to Fortran
- Fortran95 added the forall construct to the language. - In a way it is a do-loop or an array construct where the order of iteration does not matter; i.e. it can be parallellized:
program forallexample implicit none integer,parameter :: & & rk=selected_real_kind(10,40) integer :: a(10,10) integer :: i,j a=0 forall (i=1:10:2,j=1:10:2,i+j>10) a(i,j)=(i+j)/2 end forall print (10i4),a stop end program forallexample
progs> progs> 0 0 0 0 0 0 0 0 0 0
0 0 6 0 7 0 8 0 9 0
0 0 0 0 0 0 0 0 0 0
40