Is it possible to use FTorch in ABAQUS Fortran subroutine?

[Fengyixin-research]

I want to utilize the neural network (Pytorch) in ABAQUS Fortran subroutine, is it possible? Do I need any extra operation?
Any advice or suggestion is welcome.

[jatkinson1000]

Hi @Fengyixin-research we are not ABAQUS users, so we'll need a bit more information to try and help you with this.

How do you compile ABAQUS code?
Can you you link to external libraries when building and running ABAQUS code?
What platform, compilers etc are you working on?
What have you tried so far?

FTorch can be linked to as a Fortran library similarly to any other external library (see details below).

A quick search suggests it is possible to link to shared libraries from ABAQUS, but you may also need to compile your code as a shared library: https://abaqus-docs.mit.edu/2017/English/SIMACAEEXCRefMap/simaexc-c-makeproc.htm

I also found this article detailing how to use the external LAPACK library in ABAQUS Fortran subroutines through linking: https://www.bibekanandadatta.com/blog/2024/lapack-Intel-Fortran-Abaqus/

Do you have anyone who has expertise in ABAQUS or can you raise with them how to link to external libraries when building?

---

Linking to FTorch as an external library:

FTorch can be linked to as a Fortran library similarly to any other external library.
Whilst we use CMake in the examples, the FTorch documentation explains how to do this in the traditional way with Make by linking:

FCFLAGS+=" -I<path/to/FTorch/install/location>/include/ftorch"
LDFLAGS+=" -L<path/to/FTorch/install/location>/lib -lftorch"

Details

To build a project with make or similar you need to include the FTorch's header (.h) and module (.mod) files and link the executable to the Ftorch library (e.g., .so, .dll, .dylib depending on your system) when compiling.

To compile with make use the following compiler flag for any files that use ftorch to include the module and header files:

-I<path/to/FTorch/install/location>/include/ftorch

This is often done by appending to an FCFLAGS compiler flags variable or similar:

FCFLAGS+=" -I<path/to/FTorch/install/location>/include/ftorch"

When compiling the final executable add the following linker flag:

-L<path/to/FTorch/install/location>/lib -lftorch

This is often done by appending to an LDFLAGS linker flags variable or similar:

LDFLAGS+=" -L<path/to/FTorch/install/location>/lib -lftorch"

If you have pkg-config installed, you can easily query the compiler and linker flags of FTorch rather than manually specifying them as was shown above. FTorch provides a standard pkg-config file in both the directory in which FTorch was built (e.g., </path/to/FTorch>/build) as well as the library directory in which it was installed (e.g., </path/to/FTorch/install/location>/lib/pkgconfig). For example, the following commands are equivalent to adding the manually specified flags:

FCFLAGS+=" $(pkg-config --cflags </path/to/FTorch/install/location>/lib/pkgconfig/ftorch.pc)"
LDFLAGS+=" $(pkg-config --libs </path/to/FTorch/install/location>/lib/pkgconfig/ftorch.pc)"

You can simplify these commands by adding FTorch to the PKG_CONFIG_PATH environment variable:

export PKG_CONFIG_PATH=${PKG_CONFIG_PATH}:</path/to/FTorch/install/location>/lib/pkgconfig/
FCFLAGS+=" $(pkg-config --cflags ftorch)"
LDFLAGS+=" $(pkg-config --libs ftorch)"

You may also need to add the location of the dynamic library .so files to your LD_LIBRARY_PATH environment variable unless installing in a default location:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path/to/FTorch/installation>/lib

[Fengyixin-research]

Thanks for your quick reply.
I think the first link could help me to use FTorch in ABAQUS Fortran subroutine.
The second link I have used before, it is mainly dealed with math computation. Quite simple, but not suitable for neural network.
The reason for posting this question is that I have seen many researchers manually recode the neural network in ABAQUS Fortran subroutine, it might be possible for shallow neural network. But it is really huge work for deep learning neural network. Therefore, I want to use FTorch for simplification.
I will try to do some test.

[jatkinson1000]

Great.

I shared the second link not suggesting you use LAPACK, but because it discusses linking an external library into ABAQUS Fortran subroutines.

Direct linking for complex architectures is definitely the sensible approach in our opinion.

Good luck.
Please let us know how you get on and share anything that might help future users.

[jatkinson1000]

Hi @Fengyixin-research I have converted this to a discussion as it pertains to use-cases and integration rather than an issue with FTorch itself.

Please do feed back any progress you have as there may well be others that want to integrate FTorch to ABAQUS and can build off your experiences.

Thanks.

[Fengyixin-research]

Thanks. I'm not familiar with compiling, therefore, it is not very easy for me. I might need some help.
First, ABAQUS could use intel visual fortran to compiled fortran code, and the corresponding win86_64.env could be modified as follows,

Details

``` # Location of the /include directory in the ABAQUS installation abaHomeInc = os.path.abspath(os.path.join(os.environ.get('ABA_HOME', ''), os.pardir))

compile_cpp=['cl', '/c', '/W0', '/MD', '/TP',
'/EHsc', '/DNDEBUG', '/DWIN32', '/DTP_IP', '/D_CONSOLE',
'/DNTI', '/DFLT_LIC', '/DOL_DOC', '/D__LIB__', '/DHKS_NT',
'/D_WINDOWS_SOURCE', '/DFAR=', '/D_WINDOWS', '/DABQ_WIN86_64', '%P',
# '/O1', # <-- Optimization
# '/Zi', # <-- Debug symbols
'/I%I', '/I'+abaHomeInc]

compile_fmu=['win64CmpWrp', '-m64', '-msvc9', 'cl', '/LD',
'/D_WINDOWS', '/TC', '/W0', '/I%I', '/I'+abaHomeInc]

Fortran compile command for User Subroutines

compile_fortran=['"C:\Program Files (x86)\Intel\oneAPI\compiler\latest\bin\ifort.exe"','/Qmkl:parallel',
'/c', '/fpp', '/extend-source',
'/DABQ_WIN86_64', '/DABQ_FORTRAN',
'/iface:cref', '/recursive',
'/Qauto', # <-- important for thread-safety of parallel user subroutines
'/align:array64byte',
'/Qpc64', # set FPU precision to 53 bit significand
'/Qprec-div', '/Qprec-sqrt', # improve precision of FP divides and sqrt
'/Qfma-', # disable floating point fused multiply-add
'/fp:precise', # floating point model: precise
'/Qimf-arch-consistency:true', # math library consistent results
'/Qfp-speculation:safe', # floating point speculations only when safe
'/Qprotect-parens', # honor parenthesis during expression evaluation
'/Qfp-stack-check', # enable stack overflow protection checks
'/reentrancy:threaded', # important for thread-safety
#'/Qinit=zero','/Qinit=arrays', # automatically initialize all arrays to zero
#'/Qinit=snan', '/Qinit=arrays', # automatically initialize all arrays to SNAN
'/QxSSE3', '/QaxAVX', '/Qopenmp','/libs:dll','/threads', # generate SSE3, SSE2, and SSE instructions
'/heap-arrays:1000000000000000',
# '/Od', '/Ob0', # <-- Disable Optimization for Debugging
# '/Zi', # <-- Debugging Information
'/include:%I', '/include:'+abaHomeInc,
'/I"C:\PROGRA~2\FTorch\include\ftorch"',
'%P']

link_sl=['LINK',
'/nologo', '/NOENTRY', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64',
'/NODEFAULTLIB:LIBC.LIB', '/NODEFAULTLIB:LIBCMT.LIB',
'/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB', '/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB',
'/DEFAULTLIB:kernel32.lib', '/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
'/FIXED:NO', '/dll',
#'/debug', # <-- Debugging
'/libpath:"C:\PROGRA2\FTorch\lib"',
'/libpath:"D:\09TRAN1\01UMAT1\Windows_FTorch\FTorch\.ftorch\Lib\site-packages\torch\lib"',
'/libpath:"D:\09TRAN1\01UMAT~1\Windows_FTorch\FTorch\.ftorch\Lib\site-packages"',
'/def:%E', '/out:%U', '%F', '%A', '%L', '%B',
'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib',
'ftorch.lib', 'torch_cpu.lib', 'c10.lib', 'torch.lib']

link_exe=['LINK',
'/nologo', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64', '/STACK:20000000',
'/NODEFAULTLIB:LIBC.LIB', '/NODEFAULTLIB:LIBCMT.LIB', '/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB',
'/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB', '/DEFAULTLIB:kernel32.lib',
'/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
'/FIXED:NO', '/LARGEADDRESSAWARE',
# '/debug', # <-- Debugging
'/libpath:"C:\PROGRA2\FTorch\lib"',
'/libpath:"D:\09TRAN1\01UMAT1\Windows_FTorch\FTorch\.ftorch\Lib\site-packages\torch\lib"',
'/libpath:"D:\09TRAN1\01UMAT~1\Windows_FTorch\FTorch\.ftorch\Lib\site-packages"',
'/out:%J', '%F', '%M', '%L', '%B', '%O',
'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib',
'ftorch.lib', 'torch_cpu.lib', 'c10.lib', 'torch.lib']

However, when I want to compile it with ABAQUS cmd, it raises errors.

D:\09 transformer\02 FTorch_test\01 lapack_abaqus\02 abaqus_make>abaqus make library=uel_final.for
Abaqus JOB uel_final.for
Begin Compiling Abaqus/Standard User Subroutines
1/5/2026 6:54:43 PM
Intel(R) Fortran Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.12.0 Build 20240222_000000
Copyright (C) 1985-2024 Intel Corporation. All rights reserved.

ifort: remark #10448: Intel(R) Fortran Compiler Classic (ifort) is now deprecated and will be discontinued late 2024. Intel recommends that customers transition now to using the LLVM-based Intel(R) Fortran Compiler (ifx) for continued Windows* and Linux* support, new language support, new language features, and optimizations. Use '/Qdiag-disable:10448' to disable this message.
End Compiling Abaqus/Standard User Subroutines
Begin Linking Abaqus/Standard User Subroutines
Creating library standardU.lib and object standardU.exp
uel_final.obj : error LNK2019: Unresolved external symbol ftorch_mp_torch_tensor_from_array_real32_1d_default_layout，The symbol is referenced in the function uel.
uel_final.obj : error LNK2019: Unresolved external symbol ftorch_mp_torch_model_load，The symbol is referenced in the function uel.
uel_final.obj : error LNK2019: Unresolved external symbol ftorch_mp_torch_model_forward，The symbol is referenced in the function uel.
uel_final.obj : error LNK2019: Unresolved external symbol ftorch_mp_torch_model_delete，The symbol is referenced in the function uel.
uel_final.obj : error LNK2019: Unresolved external symbol ftorch_mp_torch_tensor_array_delete，The symbol is referenced in the function uel.
uel_final.obj : error LNK2001: Unresolved external symbol ftorch_mp_torch_tensor_get_rank
uel_final.obj : error LNK2001: Unresolved external symbol ftorch_mp_torch_tensor_get_shape
...
standardU.dll : fatal error LNK1120: 14 unresolvable external commands
Abaqus Error: Problem during linking - Abaqus/Standard User Subroutines.
This error may be due to a mismatch in the Abaqus user subroutine arguments.
These arguments sometimes change from release to release, so user subroutines
used with a previous release of Abaqus may need to be adjusted.
The Abaqus Make execution procedure exited with errors

PS. I have successfully installed FTorch on Windows. I have tested the usage is passed:

call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
make

Makefile:

FC = "C:\Program Files (x86)\Intel\oneAPI\compiler\latest\bin\ifort.exe"
FCFLAGS = -O3 -I "C:\Program Files (x86)\FTorch\include\ftorch"
LDFLAGS = /link -LIBPATH:"C:\Program Files (x86)\FTorch\lib" ftorch.lib
PROGRAM = simplenet_infer_fortran
SRC = simplenet_infer_fortran.f90
OBJECTS = $(SRC:.f90=.obj)
all: $(PROGRAM)
$(PROGRAM): $(OBJECTS)
$(FC) $(FCFLAGS) -o $@ $^ $(LDFLAGS)
%.o: %.f90
$(FC) $(FCFLAGS) -c $<
clean:
del /Q *.o *.mod 2>NUL

Then

set PATH=D:\09 transformer\01 Umat_NN_test\Windows_FTorch\FTorch.ftorch\Lib\site-packages;%PATH%
set PATH=C:\Program Files (x86)\FTorch\bin;%PATH%
set PATH=D:\09 transformer\01 Umat_NN_test\Windows_FTorch\FTorch.ftorch\Lib\site-packages\torch\lib;%PATH%
.\simplenet_infer_fortran saved_simplenet_model_cpu.pt

<img width="2308" height="212" alt="image" src="https://github.com/user-attachments/assets/001c104e-9e38-4399-ab3d-f3dd52f5de25" />

**Any suggestion is welcome.**
<details>

[jatkinson1000]

Hey @Fengyixin-research the fact FTorch is working standalone is a good sign and starting point!!

Regarding ABAQUS, it looks like there is a linking error.
Initial thoughts are that it could be related to the order you pass the libraries to the linker command or the paths you provide.

Shortened paths may/may not be causing issues.
Try to replace '/I"C:\\PROGRA~2\\FTorch\\include\\ftorch"' with '/I"C:\\[FULL_PATH_HERE]\\FTorch\\include\\ftorch"',

         '/libpath:"C:\\PROGRA~2\\FTorch\\lib"',
         '/libpath:"D:\\09TRAN~1\\01UMAT~1\\Windows_FTorch\\FTorch\\.ftorch\\Lib\\site-packages\\torch\\lib"',
         '/libpath:"D:\\09TRAN~1\\01UMAT~1\\Windows_FTorch\\FTorch\\.ftorch\\Lib\\site-packages"',

with

         '/libpath:"C:\\[FULL_PATH_HERE]\\FTorch\\lib"',
         '/libpath:"D:\\[FULL_PATH_HERE]\\Windows_FTorch\\FTorch\\.ftorch\\Lib\\site-packages\\torch\\lib"',

Another thing is order.
Since FTorch depends on Torch could you try passing

'ftorch.lib', 'c10.lib', 'torch_cpu.lib', 'torch.lib'

instead of

'ftorch.lib', 'torch_cpu.lib', 'c10.lib', 'torch.lib']

though this may require some further iteration.

[jatkinson1000]

Hi @Fengyixin-research

1. FTorch is downloaded and built in the first directory, but when you run cmake --build . --install or similar it installs FTorch to the chosen install location. If you did not specify a install location this will be the default which on Windows is C:Program Files (x86)/FTorch.
2. I think this depends on how you built your executable. If building statically (all code bundled into one final executable) it is true you only need the final executable. However, if building dynamically (referring to external code at run time) then you need to provide the location of the libraries in the path so that they can be found to call out to.

As a note I'm not very well versed on Windows as almost all of our development and users to date are on unix systems, but I think the question will be the same on both.

[TomMelt]

@Fengyixin-research I agree with @jatkinson1000 . Because the library (on Windows) is a .dll you need to add the paths to the FTorch installation to your PATH when you run it. Also I would avoid the short name. It might work for DOS commands but it may not work for other parts of the tool chain. Again, we don't use Windows so this is more of a guess. But it will help narrow down the problem either way.

[Fengyixin-research]

@jatkinson1000 @TomMelt Thanks for your reply. However, I have tried to reinstall FTorch in shor folder directory. The same error still occurs.

Details

``` D:\ftorch_test>abaqus make library=uel_final.for Abaqus JOB uel_final.for Begin Compiling Abaqus/Standard User Subroutines 1/6/2026 12:01:33 PM Intel(R) Fortran Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.12.0 Build 20240222_000000 Copyright (C) 1985-2024 Intel Corporation. All rights reserved.

ifort: remark #10448: Intel(R) Fortran Compiler Classic (ifort) is now deprecated and will be discontinued late 2024. Intel recommends that customers transition now to using the LLVM-based Intel(R) Fortran Compiler (ifx) for continued Windows* and Linux* support, new language support, new language features, and optimizations. Use '/Qdiag-disable:10448' to disable this message.
End Compiling Abaqus/Standard User Subroutines
Begin Linking Abaqus/Standard User Subroutines
Creating library standardU.lib and object standardU.exp
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_tensor_from_array_real32_1d_default_layout referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_load referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_forward referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_delete referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_tensor_array_delete referenced in function uel
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_rank
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_shape
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_stride
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_dtype
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_device_type
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_device_index
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_requires_grad
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_zero
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_zero_grad
standardU.dll : fatal error LNK1120: 14 unresolved externals
Abaqus Error: Problem during linking - Abaqus/Standard User Subroutines.
This error may be due to a mismatch in the Abaqus user subroutine arguments.
These arguments sometimes change from release to release, so user subroutines
used with a previous release of Abaqus may need to be adjusted.
The Abaqus Make execution procedure exited with errors

The full win86_64.env:

#-- mode: python --

#############################################################################

Compile and Link commands for ABAQUS on the Windows 64 Platform

#############################################################################

import os # <-- Debugging

os.environ['GLOBAL_ENABLE_FPE'] = 'TRUE' # <-- Debugging

Location of the /include directory in the ABAQUS installation

abaHomeInc = os.path.abspath(os.path.join(os.environ.get('ABA_HOME', ''), os.pardir))

compile_cpp=['cl', '/c', '/W0', '/MD', '/TP',
'/EHsc', '/DNDEBUG', '/DWIN32', '/DTP_IP', '/D_CONSOLE',
'/DNTI', '/DFLT_LIC', '/DOL_DOC', '/D__LIB__', '/DHKS_NT',
'/D_WINDOWS_SOURCE', '/DFAR=', '/D_WINDOWS', '/DABQ_WIN86_64', '%P',
# '/O1', # <-- Optimization
# '/Zi', # <-- Debug symbols
'/I%I', '/I'+abaHomeInc]

compile_fmu=['win64CmpWrp', '-m64', '-msvc9', 'cl', '/LD',
'/D_WINDOWS', '/TC', '/W0', '/I%I', '/I'+abaHomeInc]

Fortran compile command for User Subroutines

compile_fortran=['ifort','/Qmkl:parallel',
'/c', '/fpp', '/extend-source',
'/DABQ_WIN86_64', '/DABQ_FORTRAN',
'/iface:cref', '/recursive',
'/Qauto', # <-- important for thread-safety of parallel user subroutines
'/align:array64byte',
'/Qpc64', # set FPU precision to 53 bit significand
'/Qprec-div', '/Qprec-sqrt', # improve precision of FP divides and sqrt
'/Qfma-', # disable floating point fused multiply-add
'/fp:precise', # floating point model: precise
'/Qimf-arch-consistency:true', # math library consistent results
'/Qfp-speculation:safe', # floating point speculations only when safe
'/Qprotect-parens', # honor parenthesis during expression evaluation
'/Qfp-stack-check', # enable stack overflow protection checks
'/reentrancy:threaded', # important for thread-safety
#'/Qinit=zero','/Qinit=arrays', # automatically initialize all arrays to zero
#'/Qinit=snan', '/Qinit=arrays', # automatically initialize all arrays to SNAN
'/QxSSE3', '/QaxAVX', '/Qopenmp','/libs:dll','/threads', # generate SSE3, SSE2, and SSE instructions
'/heap-arrays:1000000000000000',
# '/Od', '/Ob0', # <-- Disable Optimization for Debugging
# '/Zi', # <-- Debugging Information
'/include:%I', '/include:'+abaHomeInc,
'/I"D:\FTorch_install\include\ftorch"',
'%P']

link_sl=['LINK',
'/nologo', '/NOENTRY', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64',
'/NODEFAULTLIB:LIBC.LIB', '/NODEFAULTLIB:LIBCMT.LIB',
'/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB', '/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB',
'/DEFAULTLIB:kernel32.lib', '/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
'/FIXED:NO', '/dll',
#'/debug', # <-- Debugging
'/libpath:"D:\FTorch_install\lib"',
'/libpath:"D:\FTorch\.ftorch\Lib\site-packages\torch\lib"',
'/def:%E', '/out:%U', '%F', '%A', '%L', '%B',
'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib',
'torch.lib','torch_cpu.lib', 'c10.lib','ftorch.lib']

link_exe=['LINK',
'/nologo', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64', '/STACK:20000000',
'/NODEFAULTLIB:LIBC.LIB', '/NODEFAULTLIB:LIBCMT.LIB', '/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB',
'/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB', '/DEFAULTLIB:kernel32.lib',
'/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
'/FIXED:NO', '/LARGEADDRESSAWARE',
# '/debug', # <-- Debugging
'/libpath:"D:\FTorch_install\lib"',
'/libpath:"D:\FTorch\.ftorch\Lib\site-packages\torch\lib"',
'/out:%J', '%F', '%M', '%L', '%B', '%O',
'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib',
'torch.lib','torch_cpu.lib', 'c10.lib','ftorch.lib']

#usub_dir_lib="standardU"

Link command to be used for ABAQUS/MAKE on machines without the Fortran compiler.

Uncomment it to enable.

#link_exe=['LINK', '/nologo', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64', '/NODEFAULTLIB:LIBC.LIB', '/NODEFAULTLIB:LIBCMT.LIB',

'/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:MSVCRT.LIB', '/DEFAULTLIB:kernel32.lib', '/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',

'/FIXED:NO', '/LARGEADDRESSAWARE', '/out:%J', '%F', '%M', '%L', '%B', '%O',

'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',

'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib']

Remove the temporary names from the namespace

del abaHomeInc

I have also tested the standalone example2, it also passed. Therefore, the reason could be the connection problem between the external .lib/.dll and the ABAQUS compile environment.
<details>

[jatkinson1000]

Thanks for the update @Fengyixin-research
Since FTorch is working standalone on your machine it would seem that this is a problem with how ABAQUS is linking out rather than with FTorch, and I'm afraid we don't have any experience using this.

Do you have some software managers or system administrators with ABAQUS you can speak to to assist you in troubleshooting this?
If they are able to assist please do let us know the outcome.

[Fengyixin-research]

@jatkinson1000 Thanks for your reply. I have figured out a way to validate the link option:

Details

``` D:\ftorch_test>call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64 vs2022 :: initializing oneAPI environment... Initializing Visual Studio command-line environment... Visual Studio version 17.10.3 environment configured. "C:\Program Files\Microsoft Visual Studio\2022\Professional\" Visual Studio command-line environment initialized for: 'x64' : advisor -- latest : compiler -- latest : dal -- latest : debugger -- latest : dev-utilities -- latest : dnnl -- latest : dpcpp-ct -- latest : dpl -- latest : ipp -- latest : ippcp -- latest : mkl -- latest : mpi -- latest : ocloc -- latest : tbb -- latest : vtune -- latest :: oneAPI environment initialized ::

D:\ftorch_test>ifort /Qmkl:parallel /c /fpp /extend-source /DABQ_WIN86_64 /DABQ_FORTRAN /iface:cref /recursive /Qauto /align:array64byte /Qpc64 /Qprec-div /Qprec-sqrt /Qfma- /fp:precise /Qimf-arch-consistency:true /Qfp-speculation:safe /Qprotect-parens /Qfp-stack-check /reentrancy:threaded /QxSSE3 /QaxAVX /Qopenmp /libs:dll /threads /heap-arrays:1000000000000000 /I"D:\FTorch_install\include\ftorch" /I"C:\SIMULIA\EstProducts\2024\SMAUsubs\PublicInterfaces" uel_final.for
Intel(R) Fortran Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.12.0 Build 20240222_000000
Copyright (C) 1985-2024 Intel Corporation. All rights reserved.

ifort: remark #10448: Intel(R) Fortran Compiler Classic (ifort) is now deprecated and will be discontinued late 2024. Intel recommends that customers transition now to using the LLVM-based Intel(R) Fortran Compiler (ifx) for continued Windows* and Linux* support, new language support, new language features, and optimizations. Use '/Qdiag-disable:10448' to disable this message.

D:\ftorch_test>link /nologo /machine:AMD64 /dll /NOENTRY /INCREMENTAL:NO /subsystem:console /DEFAULTLIB:OLDNAMES.LIB /DEFAULTLIB:LIBIFCOREMD.LIB /DEFAULTLIB:LIBIFPORTMD.LIB /DEFAULTLIB:LIBMMD.LIB /DEFAULTLIB:msvcrt.lib /DEFAULTLIB:vcruntime.lib /DEFAULTLIB:ucrt.lib /DEFAULTLIB:libcmt.lib /DEFAULTLIB:kernel32.lib /DEFAULTLIB:user32.lib /DEFAULTLIB:advapi32.lib /libpath:"D:\FTorch_install\lib" /libpath:"D:\FTorch.ftorch\Lib\site-packages\torch\lib" /libpath:"C:\Program Files (x86)\Intel\oneAPI\mkl\2024.0\lib\intel64" /OUT:test_uel.dll uel_final.obj ftorch.lib torch.lib torch_cpu.lib c10.lib oldnames.lib user32.lib ws2_32.lib netapi32.lib advapi32.lib mkl_intel_lp64.lib mkl_core.lib mkl_sequential.lib vcruntime.lib ucrt.lib libcmt.lib
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_tensor_from_array_real32_1d_default_layout referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_load referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_forward referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_model_delete referenced in function uel
uel_final.obj : error LNK2019: unresolved external symbol ftorch_mp_torch_tensor_array_delete referenced in function uel
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_rank
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_shape
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_stride
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_dtype
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_device_type
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_get_device_index
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_requires_grad
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_zero
uel_final.obj : error LNK2001: unresolved external symbol ftorch_mp_torch_tensor_zero_grad
test_uel.dll : fatal error LNK1120: 14 unresolved externals

<img width="1384" height="118" alt="image" src="https://github.com/user-attachments/assets/a0cc1cdd-ac51-41f3-8fb6-1f6273c4a70d" />
We could see the .obj file could be generated well. However, the same link error occurs. 

**Therefore, the link option could be somewhere wrong even without the ABAQUS environment.**

PS. I have also checked the offical windows installation link(https://cambridge-iccs.github.io/FTorch/page/troubleshooting.html#windows). Only one thing needs to change:

Modify CMakeLists.txt
Change
target_link_libraries(${LIB_NAME}
PRIVATE ${TORCH_LIBRARIES} INTERFACE stdc++)
to
target_link_libraries(${LIB_NAME}
PRIVATE ${TORCH_LIBRARIES} INTERFACE msvcprt)

Also, I have checked the implement of example 2, the test is passed.
<img width="1374" height="322" alt="image" src="https://github.com/user-attachments/assets/be4593cf-8334-4407-93c6-911b2879deaa" />
**I am wondering why the Makefile does not need these extra options like the first cmd link options but works well.
How should I modify the above link options?**

`link /nologo /machine:AMD64 /dll /NOENTRY /INCREMENTAL:NO /subsystem:console /DEFAULTLIB:OLDNAMES.LIB /DEFAULTLIB:LIBIFCOREMD.LIB /DEFAULTLIB:LIBIFPORTMD.LIB /DEFAULTLIB:LIBMMD.LIB /DEFAULTLIB:msvcrt.lib /DEFAULTLIB:vcruntime.lib /DEFAULTLIB:ucrt.lib /DEFAULTLIB:libcmt.lib /DEFAULTLIB:kernel32.lib /DEFAULTLIB:user32.lib /DEFAULTLIB:advapi32.lib /libpath:"D:\FTorch_install\lib" /libpath:"D:\FTorch\.ftorch\Lib\site-packages\torch\lib" /libpath:"C:\Program Files (x86)\Intel\oneAPI\mkl\2024.0\lib\intel64" /OUT:test_uel.dll uel_final.obj ftorch.lib torch.lib torch_cpu.lib c10.lib oldnames.lib user32.lib ws2_32.lib netapi32.lib advapi32.lib mkl_intel_lp64.lib mkl_core.lib mkl_sequential.lib vcruntime.lib ucrt.lib libcmt.lib`

<img width="2234" height="1120" alt="image" src="https://github.com/user-attachments/assets/830c4f5b-0ec4-44c3-9fbc-8668c6f1f239" />

<details>

[Fengyixin-research]

Updated. @jatkinson1000 @TomMelt .
Fortunately, I have successfully solved the above problem with the help of Deepseek and Gemini.
The main question is that ABAQUS could not recognize the upperclass word in the function name, therfore, we need to manually add the /names:lowercase to the build option for FTorch.
For example, we could use dumpbin to extract the function name in FTorch.lib:
dumpbin /exports "D:\FTorch_install\lib\ftorch.lib" > ftorch_exports.txt
If we use the default build option for FTorch, the corresponding function names are:

However, if we add the the /names:lowercase to the build option for FTorch:

Obviously, the latter names could be found by ABAQUS.
Then, the code works well:

abaqus make library=uel_final.for
abaqus job=test.inp

And the corresponding result is that:

To sum up:
Two things need to do:

1. modify the build option for FTorch library:

FTorch with vs2022 and oneAPI fortran:
call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64 vs2022

git clone https://github.com/Cambridge-ICCS/FTorch.git
cd FTorch

python -m venv .ftorch
call .ftorch\Scripts\activate
pip install torch torchvision torchaudio
ECHO ON

cmake -Bbuild -G "NMake Makefiles" -DCMAKE_Fortran_FLAGS="/fpscomp:logicals /iface:cref  /names:lowercase /libs:dll /threads" ^
 -DCMAKE_PREFIX_PATH="D:\FTorch\.ftorch\Lib\site-packages" ^
-DCMAKE_BUILD_TYPE=Release ^
-DCMAKE_BUILD_TESTS=True ^
-DCMAKE_Fortran_COMPILER=ifort ^
-DCMAKE_C_COMPILER=cl ^
-DCMAKE_CXX_COMPILER=cl ^
-DMKL_LIBRARIES="C:\Program Files (x86)\Intel\oneAPI\mkl\latest\lib" ^
-DGPU_DEVICE=CUDA ^
-DCMAKE_INSTALL_PREFIX="D:\FTorch_install" ^
-DMULTI_GPU=OFF

Modify CMakeLists.txt
Change
target_link_libraries(${LIB_NAME}
  PRIVATE ${TORCH_LIBRARIES} INTERFACE stdc++)
to
target_link_libraries(${LIB_NAME}
  PRIVATE ${TORCH_LIBRARIES} INTERFACE msvcprt)
 
cmake --build build
cmake --install build

2. modify the win86_64.env file in ABAQUS site directory:

compile_cpp=['cl', '/c', '/W0', '/MD', '/TP',
             '/EHsc', '/DNDEBUG', '/DWIN32', '/DTP_IP', '/D_CONSOLE',
             '/DNTI', '/DFLT_LIC', '/DOL_DOC', '/D__LIB__', '/DHKS_NT',
             '/D_WINDOWS_SOURCE', '/DFAR=', '/D_WINDOWS', '/DABQ_WIN86_64', '%P',
             # '/O1', # <-- Optimization
             # '/Zi', # <-- Debug symbols
             '/I%I', '/I'+abaHomeInc,
			 '/include:%I',
			 r'/I"D:\FTorch_install\include\ftorch"',
			 ]
compile_fortran=['ifort','/Qmkl:parallel',
                 '/c', '/fpp', '/extend-source', 
                 '/DABQ_WIN86_64',  '/DABQ_FORTRAN',
                 '/iface:cref', '/recursive',
                 '/Qauto',  # <-- important for thread-safety of parallel user subroutines
                 '/align:array64byte',
                 '/Qpc64',                    # set FPU precision to 53 bit significand
                 '/Qprec-div', '/Qprec-sqrt', # improve precision of FP divides and sqrt
                 '/Qfma-',                    # disable floating point fused multiply-add
                 '/fp:precise',               # floating point model: precise 
                 '/Qimf-arch-consistency:true', # math library consistent results
                 '/Qfp-speculation:safe',     # floating point speculations only when safe
                 '/Qprotect-parens',          # honor parenthesis during expression evaluation
                 '/Qfp-stack-check',          # enable stack overflow protection checks
                 '/reentrancy:threaded',      # important for thread-safety
                 '/QxSSE3', '/QaxAVX', '/Qopenmp','/libs:dll','/threads',        # generate SSE3, SSE2, and SSE instructions
                 '/heap-arrays:1000000000000000', 
                 '/I%I', '/I'+abaHomeInc, 
                 r'/ID:\FTorch_install\include\ftorch',
                 '%P']
link_sl=['LINK',
         '/nologo', '/NOENTRY', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64',
         '/NODEFAULTLIB:LIBC.LIB', 
         '/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB', '/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB',
         '/DEFAULTLIB:kernel32.lib', '/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
         '/FIXED:NO', '/dll',
         r'/libpath:"D:\FTorch_install\lib"',
         r'/libpath:"D:\FTorch\.ftorch\Lib\site-packages\torch\lib"',
         '/def:%E', '/out:%U', 
         '%F', '%A', '%L', '%B', 
         r'D:\FTorch_install\lib\ftorch.lib',
         'torch.lib','torch_cpu.lib', 'c10.lib', 
         'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
         'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib']
link_exe=['LINK',
          '/nologo', '/INCREMENTAL:NO', '/subsystem:console', '/machine:AMD64', '/STACK:20000000',
          '/NODEFAULTLIB:LIBC.LIB', 
          '/DEFAULTLIB:OLDNAMES.LIB', '/DEFAULTLIB:LIBIFCOREMD.LIB',
          '/DEFAULTLIB:LIBIFPORTMD.LIB', '/DEFAULTLIB:LIBMMD.LIB', '/DEFAULTLIB:kernel32.lib',
          '/DEFAULTLIB:user32.lib', '/DEFAULTLIB:advapi32.lib',
          '/FIXED:NO', '/LARGEADDRESSAWARE',
          r'/libpath:"D:\FTorch_install\lib"',
          r'/libpath:"D:\FTorch\.ftorch\Lib\site-packages\torch\lib"',
          '/out:%J', '%F', '%M', '%L', '%B', '%O',
          'ftorch.lib', 
          'torch.lib','torch_cpu.lib', 'c10.lib', 
          'oldnames.lib', 'user32.lib', 'ws2_32.lib', 'netapi32.lib',
          'advapi32.lib', 'msvcrt.lib', 'vcruntime.lib', 'ucrt.lib']

Then, you could enjoy using the FTorch with ABAQUS Fortran subroutine.