Development environment setup for RMCProfile on Windows

This blog is for noting down the details about setting up the development environment for RMCProfile on Windows machine. The notes here are still based on the conventional way of compiling using make files. There is some work happening in the background now to transfer the compiling to using cmake. Before the whole pipeline of cmake is ready, we will still rely on the route as in the current blog. In fact, even the current route is depracated, some of the notes will still apply when setting up the development environment.

Nowadays, we are pretty much using the intel Fortran compiler which now is free, through the oneAPI toolkit. To install the compiler, we need to first install Microsoft Visual Studio (the community version is fine, and it is free). We may need to remove whatever existing old versions of Microsoft Visual Studio (e.g., the 2019 version may need to be removed, if the we intend to use the 2022 version). I am not sure whether this is the cause for my issue, but in my case, I have been struggling a lot with the oneAPI setup in the down stream. Anyhow, having a clean version of 2022 turns out to be working fine for me.
Then we want to install the OneAPI toolkit.

Here is the official website for download and instructions (as of writing), oneAPI Base Toolkit download and install

See the Run Sample Code to Verify Installation section on the website for the command to run for terminal (e.g., powershell) initialization.
To use make and also for the purpose of back-compatibility for some plotting routines, we need to install Cygwin and set up the pgplot library inside the Cygwin environment (see the section at the very bottom for details about setting up pgplot in Cygwin). When installing Cygwin, the following selections are necessary: make, x11 and xorg. Later on when installing the pgplot library, if some of the X11 stuff are missing, we can Google what we need to include for X11 installation according to the error message we got.
We will need to install nvcc for compiling CUDA codes involved in the RMCProfile project. As of writing, the offial website for download the installation file is this. In case it is changed, we can just google ‘CUDA nvcc install’ or something relevant to search for information.
In preparation for the compiling, we may need to delete unnecessary entries in the PATH environment variable. Otherwise, the nvcc command will fail when it is trying to set up the environment on-the-fly. It is something relevant to the running of vcvars64.bat script which will be running on-the-fly when running nvcc, and the fundamental reason for the potential failure is the value of the environment variable PATH is too long. On Windows, we can use Start to search for ‘environment variable’ to bring up the GUI window for changing the PATH variable.
On powershell, we need to run
```
 cmd.exe "/K" '"C:\Program Files (x86)\Intel\oneAPI\setvars.bat" && powershell'
```
to initialize the terminal environment.

This command needs to be run each time when launching a new powershell terminal.

When we see error like this,
```
 LINK : fatal error LNK1104: cannot open file 'ifconsol.lib'
```
it infers the oneAPI environment initializer was not run, in which case we just need to run the command above to initialize the environment and try again the compiling.
As a side note, with the latest intel oneAPI compiler (ifort), the argument that goes into c_loc (the function used in the source codes for fetching the address of a pointer) should be defined as a pointer. The original definition of some of the variables (used as the argument for c_loc) as allocatable arrays would not work.
When encountering the issue with unresolved CUDA symbols, we may need to change the cuda library files to the version that is compatible with the version of nvcc being used for the compiling. Specifically, two CUDA library files cublas.lib and cudart.lib need to be linked while making the rmcprofile.exe executable. For specific command for linking, etc., refer to those Makefile files in the RMCProfile repository – for the moment, the source code is not open source yet. To get access, get in touch with me here. We need to copy the proper version of these two files to somewhere (and the path to the two library files will be specified while linking to make rmcprofile.exe).

In my case, I was using the 12.6 version of nvcc and my library files could be found here, C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\lib\x64.

=================I AM A SEPARATOR=================

Here follows are the detaild instructions about setting up pgplot in Cygwin.

Grab the pgplot5.2.tar.gz file which has already been included in the main directory of the RMCProfile repository.
Decompress and extract the contents of the distribution file in a source directory. In this sense, I always place pgplot5.2.tar.gz under /usr/local/src. Then we can do,
```
  cd /usr/local/src
  mv [SOMEWHERE]/pgplot5.2.tar.gz .
  tar zxvf pgplot5.2.tar.gz
```
This will create /usr/local/src/pgplot and subdirectories.
Create the directory where PGPLOT will be actually installed,
```
  mkdir /usr/local/pgplot
  cd /usr/local/pgplot
```
Copy the file drivers.list from the source directory to the installation directory,
```
  cp /usr/local/src/pgplot/drivers.list .
```
Edit that file and remove the exclamation mark (first column of each row) in front of the following graphic devices,
```
  /PS, /VPS, /CPS, /VCPS,  /Xserve and /XWindow
```
Create the makefile. From the installation directory /usr/local/pgplot, execute,
```
  /usr/local/src/pgplot/makemake /usr/local/src/pgplot linux g77_gcc_aout
```
Edit the file makefile and change the line,
```
  FCOMPL=g77
```
to
```
  FCOMPL=gfortran
```
Compile the source files,
```
  make
  make cpg
  make clean
```

=================I AM A SEPARATOR=================

Here are some random notes on compiling with cmake.

In the main CMakeLists.txt file, there is one line project(RMCProfile LANGUAGES CXX C Fortran) specifying what languages are included in the project. cmake will perform the necessary configuration checking and testing at the configuration stage for all the included languages. With CUDA, I found that on my Windows machine, including CUDA in the language list would never pass the initial configuration stage. However, leaving out CUDA from the list, thus skipping the CUDA checking and testing, is still OK – the compiling could still be finished without problems with CUDA codes compiled successfully. Not sure why, but suspecting whether it is the following line that puts down CUDA specification explicitly actually did the magic,
```
  find_package(CUDA 11.4 REQUIRED)

  set(CMAKE_CUDA_STANDARD 11)
  set(CMAKE_CUDA_STANDARD_REQUIRED ON)

  find_package(CUDAToolkit REQUIRED)

  if(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 12)
      set(CMAKE_CUDA_ARCHITECTURES "61;70;75;80;86;89")
  else()
      set(CMAKE_CUDA_ARCHITECTURES "61;70;75;80;86")
  endif()

  message(STATUS "Using Nvidia GPU architectures ${CMAKE_CUDA_ARCHITECTURES}")

  if(CMAKE_BUILD_TYPE STREQUAL "Debug")
      set(CMAKE_CUDA_FLAGS ${CMAKE_CUDA_FLAGS} "-g -G")  # enable cuda-gdb
  endif()
```