MagneX ML code PR (#89)

* docs

* docs update

* docs update

* docs update

* use forward euler to simple build works

* Update README.md

* Update README.md

* Update README.md

* whitespace

* enable DMI and anis in scaling test

* skyrmion diagnostic post processing utility

* renname file

* restart bugfix

* sundials 7.1.1

* inputs file for paper demag plot

* license and copyright

* patches to get regression tests running again
some benchmarks will be changing

* fix sundials settings

* cleanup per Weiqun

* logic for different physics wasn't right for single-rate case

* add SUNDIALS_HOME

* sundials update

* comments in inputs file

* initial commit

* updated ML part of the MagneX by adding: 1. USE_ML flag for during compile. 2. Added 'ml_enable' flag in the input file to control inference with ML or not

* clean up makefile

* fix trailing whitespace

* no longer need fftw includes, these are contained in USE_FFT=TRUE

* non ml compile

* put in ml_demag protections when not implemented

* minor

* fixed MagneX.H header for compiling issue

* cleanup include statements

* pytorch install instructions

---------

Co-authored-by: Andy Nonaka <AJNonaka@lbl.gov>

Author: tyh0123

Exec/GNUmakefile

@@ -9,10 +9,51 @@ USE_HIP      = FALSE
 COMP         = gnu
 DIM          = 3
 USE_FFT      = TRUE
-
+USE_ML       = FALSE
+TINY_PROFILE = FALSE
+PROFILE      = FALSE
 USE_SUNDIALS = FALSE
 SUNDIALS_HOME ?= ../../sundials/instdir
 
+ifeq ($(USE_ML),TRUE)
+
+  CPPFLAGS += -DAMREX_USE_ML
+
+  # Define a macro for the C++ preprocessor
+  DEFINES += -DML_ENABLE -D_GLIBCXX_USE_CXX11_ABI=1
+
+  # Pytorch root directory selection
+  ifeq ($(USE_CUDA),TRUE)
+    PYTORCH_ROOT := ../../libtorch_cuda
+  else
+    PYTORCH_ROOT := ../../libtorch_cpu
+  endif
+
+  TORCH_LIBPATH = $(PYTORCH_ROOT)/lib
+
+  # Library definitions
+  ifeq ($(USE_CUDA),TRUE)
+    # Note: Modern LibTorch often requires both torch_cuda and torch_cpu
+    TORCH_LIBS = -ltorch -ltorch_cuda -ltorch_cpu -lc10 -lc10_cuda -lcuda 
+  else
+    TORCH_LIBS = -ltorch -ltorch_cpu -lc10
+  endif
+
+  # Header search paths
+  INCLUDE_LOCATIONS += $(PYTORCH_ROOT)/include \
+                       $(PYTORCH_ROOT)/include/torch/csrc/api/include
+
+  # Library search paths
+  LIBRARY_LOCATIONS += $(TORCH_LIBPATH)
+
+  # Linker flags (rpath ensures the .so files are found at runtime)
+  ifeq ($(USE_CUDA),TRUE)
+    LDFLAGS += -Xlinker "--no-as-needed,-rpath,$(TORCH_LIBPATH)" $(TORCH_LIBS)
+  else
+    LDFLAGS += -Wl,--no-as-needed,-rpath=$(TORCH_LIBPATH) $(TORCH_LIBS)
+  endif
+endif
+
 include $(AMREX_HOME)/Tools/GNUMake/Make.defs
 
 include ../Source/Make.package

Exec/README

@@ -0,0 +1,36 @@
+# PyTorch (libtorch) Download and Setup
+
+This guide downloads the libtorch CUDA 11.8 C++ distribution and unzips it in the same directory that contains `MagneX`, then renames the extracted folder to `libtorch_cuda`.
+
+## Steps
+
+1. Change to the parent directory of `MagneX`:
+
+```bash
+cd <MAGNEX_PARENT_DIR>
+```
+
+2. Download the libtorch archive with `wget`:
+
+```bash
+wget https://download.pytorch.org/libtorch/cu118/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu118.zip
+```
+
+3. Unzip the archive in the same directory:
+
+```bash
+unzip libtorch-cxx11-abi-shared-with-deps-2.7.1+cu118.zip
+```
+
+4. Rename the extracted folder to `libtorch_cuda`:
+
+```bash
+mv libtorch libtorch_cuda
+```
+
+## Result
+
+After the steps above, you should have:
+
+- `<MAGNEX_PARENT_DIR>/MagneX`
+- `<MAGNEX_PARENT_DIR>/pytorch_cuda`

Exec/README_md.pytorch

@@ -0,0 +1,6 @@
+# SUNDIALS Setup
+
+Refer to https://amrex-codes.github.io/amrex/docs_html/TimeIntegration_Chapter.html
+for SUNDIALS installation, build, and usage instructions.
+
+Make sure that `SUNDIALS_HOME` in the `GNUmakefile` points to the installation directory.