Closed shreya-dupa closed 5 years ago
darioizzo
commented
5 years ago Not sure this is the right place to seek for help ...
I did compile snopt7.5 on mingw64. In my case the only DLL needed where the mingw ones. And only a few (like 3-4).
A note on dependency walker ... many system libraries are often False negatives as the software is rather old ... my guess is that most of yours are.
https://stackoverflow.com/questions/36240215/dependency-walker-missing-dlls
shreya-dupa
commented
5 years ago @darioizzo That's good to know...
Do you have any sample code that I can use to test the SNOPT pass through?
darioizzo
commented
5 years ago You mean from pagmo plugins?
Sure, I am on the mobile now, will post tomorrow...
shreya-dupa
commented
5 years ago Yes, thanks!
darioizzo
commented
5 years ago Here is a test I normally use:
Python 3.7.2 (default, Jan 10 2019, 23:51:51)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.2.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: import pygmo as pg
In [2]: import pygmo_plugins_nonfree as pnf
In [3]: uda = pnf.snopt7(library = r'/usr/local/lib/libsnopt72_c.so', screen_output=True, minor_version=2)
In [4]: algo = pg.algorithm(uda)
In [5]: prob = pg.problem(pg.ackley(10))
In [6]: pop = pg.population(prob, 1)
In [7]: pop = algo.evolve(pop)
==============================
SNOPT C interface 2.0.0
==============================
S N O P T 7.2-4 (Jun 2006)
==============================
SNMEMA EXIT 100 -- finished successfully
SNMEMA INFO 104 -- memory requirements estimated
Nonlinear constraints 0 Linear constraints 1
Nonlinear variables 10 Linear variables 0
Jacobian variables 0 Objective variables 10
Total constraints 1 Total variables 10
The user has defined 0 out of 10 first derivatives
Major Minors Step nObj Feasible Optimal Objective nS
0 10 1 3.2E-01 2.1009715E+01 8 r
1 1 3.8E-01 2 4.9E-01 2.0453627E+01 8 n rl
2 1 1.9E-01 3 1.5E-01 1.9435252E+01 8 s
3 1 1.0E+00 4 3.1E-01 1.9350316E+01 8
4 2 1.0E+00 5 1.7E-01 1.9329938E+01 9
5 2 1.0E+00 6 4.6E-02 1.9319270E+01 10
6 2 1.0E+00 7 3.1E-02 1.9318807E+01 9
7 2 1.0E+00 8 6.0E-03 1.9318126E+01 10
8 1 1.0E+00 9 1.3E-03 1.9318121E+01 10
9 1 1.0E+00 10 1.8E-04 1.9318117E+01 10
Major Minors Step nObj Feasible Optimal Objective nS
10 1 4.8E-01 11 5.3E-05 1.9318117E+01 10
11 1 1.0E+00 12 7.5E-06 1.9318117E+01 10 c
11 2 1.0E+00 12 4.7E-05 1.9318117E+01 10 c
12 1 4.8E-01 13 1.9E-05 1.9318117E+01 10 c
13 1 1.0E+00 14 3.3E-06 1.9318117E+01 10 c
14 1 1.0E+00 15 (9.1E-09) 1.9318117E+01 10 c
SNOPTA EXIT 0 -- finished successfully
SNOPTA INFO 1 -- optimality conditions satisfied
Problem name Ackley F
No. of iterations 29 Objective value 1.9318117416E+01
No. of major iterations 14 Linear objective 0.0000000000E+00
Penalty parameter 0.000E+00 Nonlinear objective 1.9318117416E+01
No. of calls to funobj 232 No. of calls to funcon 232
Calls with modes 1,2 (known g) 15 Calls with modes 1,2 (known g) 15
Calls for forward differencing 120 Calls for forward differencing 120
Calls for central differencing 80 Calls for central differencing 80
No. of superbasics 10 No. of basic nonlinears 0
No. of degenerate steps 0 Percentage 0.00
Max x 9 2.8E+01 Max pi 1 1.0E+00
Max Primal infeas 0 0.0E+00 Max Dual infeas 10 2.3E-08
Solution not printed
Time for MPS input 0.00 seconds
Time for solving problem 0.00 seconds
Time for solution output 0.00 seconds
Time for constraint functions 0.00 seconds
Time for objective function 0.00 seconds
shreya-udupa
commented
5 years ago ==============================
SNOPT C interface 2.0.0 At line 94 of file src/sn03prnt.f (unit = 219)
Fortran runtime error: Unit number in I/O statement too large@darioizzo I'm running into this error. Have you encountered it before?
darioizzo
commented
5 years ago No but you can see in your Fortran file why is there. Pagmo plugins is not using most of smooth I/o machinery .... so that if you just suppress the error it could work.
Alternatively you can post here the lines of that file and I can see what happens.
darioizzo
commented
5 years ago I can send you the patched file maybe...
shreya-udupa
commented
5 years ago It's in the snPRNT subroutine... I've put asterisks around the line where the code is erroring out.
Does this error indicate that the libraries have been linked correctly?
subroutine snPRNT
& ( mode, string, iw, leniw )
implicit
& none
character*(*)
& string
integer
& mode, leniw, iw(leniw)
* ==================================================================
* snPRNT prints a trimmed form of "string" on various files.
* If mode = 0, nothing is output.
* If mode = 1, string is output to iPrint.
* If mode = 2, string is output to iSumm.
* If mode = 3 or 4, string is output to iPrint and iSumm.
* If mode = 4, string is output to iStdo (standard output)
* if iPrint and iSumm are both zero. This mode
* is intended for error messages.
* If mode = 5, string is output to iStdo (standard output)
* This mode is to be used when the elements of
* the integer work array iw cannot be trusted.
*
* mode 11-15 are the same as mode 1-5 with blank line before output.
*
* If mode > 15 then nothing is printed unless lvlSys > 0.
* mode 21-25 are the same as mode 1-5
* mode 31-35 are the same as mode 11-15
*
* 25 Sep 2002: First version of snPRNT.
* 31 Jul 2003: mode 11-14 added. form introduced.
* 27 Dec 2003: mode 5 added to allow printing before iw is set.
* 12 Mar 2004: s1trim called to trim the string.
* 22 Jun 2004: System printing option added.
* 22 Jun 2004: Current version of snPRNT.
* ==================================================================
external
& s1outpt
integer
& iPrint, iSumm, iStdo, length, lvlSys, m, s1outpt
character
& form*4, form1*4, form2*4
* ------------------------------------------------------------------
parameter (form1 = '( a)')
parameter (form2 = '(/a)')
* ------------------------------------------------------------------
lvlSys = iw( 71) ! > 0 => print system info
m = 0
if (mode .le. 0) then
! Relax
else if (mode .lt. 10) then
m = mode
form = form1
else if (mode .lt. 20) then ! Blank line first
m = mode - 10
form = form2
else if (lvlSys .gt. 0) then ! Print system Info
if (mode .lt. 30) then
m = mode - 20
form = form1
else
m = mode - 30
form = form2
end if
end if
if (m .gt. 0) then
! length = len_trim(string) ! An F90 intrinsic
call s1trim( string, length ) ! The F77 equivalent
if (m .eq. 5) then
iStdo = s1outpt()
if (iStdo .gt. 0) write(iStdo, form) string(1:length)
else
iStdo = iw( 10) ! Standard output
iPrint = iw( 12) ! Print file
iSumm = iw( 13) ! Summary file
if (m .eq. 1 .or. m .ge. 3) then
if (iPrint .gt. 0) write(iPrint, form) string(1:length)
end if
if (m .eq. 2 .or. m .ge. 3) then
*** if (iSumm .gt. 0) write(iSumm , form) string(1:length) *** # This is where the error occurs
end if
if (m .eq. 4) then
if (iPrint .le. 0 .and. iSumm .le. 0) then
if (iStdo .gt. 0) write(iStdo, form) string(1:length)
end if
end if
end if
end if
end ! subroutine snPRNT@darioizzo Still struggling with this error. Were you going to send me a patched file? Not sure how to suppress the error...
darioizzo
commented
5 years ago I have the same identical file in my fortran files. Just checked. Not sure whats happening here ... but it must be some problem of the dll you assembled. Pagmo is only dlopening them and calling the snoptA interface.
darioizzo
commented
5 years ago Will close this. In case some new info arises reopen.
I decided to start a new issue for this even though it's more of an SNOPT problem than an interface problems. I was able to compile both the Fortran library and the C-interface ( 2. (snopt/snopt-interface@76b166e)) for SNOPT 7.6.
While the build executes without errors, there are a lot of dependencies missing during runtime. I checked this using the Dependency Walker software. I don't know where these missing libraries are located. Are they Windows DLLs? or MSYS64/MinGW64 DLLs? The screenshot below from the dependency walker shows some but not all of the missing dependencies.
Appreciate the help!