Merge pull request #13 from TensoriumCore/CUDA_GPU_Support

  ## Summary

  This update merges the `moving-puncture-regrid` work into the current CUDA GPU support branch.

  ## Highlights

  - Added the new Z4c-oriented moving puncture workflow, replacing the older BSSN test layout with `z4c`
  test targets and runners.
  - Added moving-puncture regridding support, including nested and BBH split fixed mesh refinement
  layouts.
  - Added FMR memory pooling utilities to reduce allocation pressure during moving puncture evolutions.
  - Added optional HDF5 export support for simulation and diagnostic output.
  - Improved moving puncture boundary handling, including Sommerfeld/radiative boundary configuration and
  sponge support.
  - Preserved and integrated the CUDA backend path for Z4c/BSSN RK stepping, CUDA RHS kernels, and CUDA-
  related regression tests.
  - Added/updated Python bindings and Python-side Z4c helpers.
  - Updated CMake targets and test organization around the new `z4c` naming.

  ## Validation

  Built successfully with:

  - `cmake --build build-codex-cuda -- -j8`
  - `cmake --build build-codex-check -- -j8`

  Both CUDA and non-CUDA builds compile successfully.

Author: at0m741

.gitignore

@@ -3,7 +3,7 @@
 dist/
 __pycache__/
 .venv/
-
+*.h5
 # Build / CMake
 build/
 build-*/

CMakeLists.txt

@@ -8,11 +8,12 @@ option(AVX2 "Enable AVX2+FMA codegen" OFF)
 option(AVX512 "Enable AVX-512" OFF)
 option(VERBOSE "Verbose runtime logs" OFF)
 option(DEBUG "Build with debug symbols" OFF)
-option(BUILD_PLUGINS "Build Clang/LLVM plugins" ON)
+option(BUILD_PLUGINS "Build Clang/LLVM plugins" OFF)
 option(BUILD_TESTS "Build test executables" ON)
 option(BUILD_PYBIND "Build Python bindings" OFF)
 option(USE_CUDA "Enable CUDA kernels (requires nvcc)" OFF)
 option(USE_CBLAS "Enable BLAS" ON)
+option(USE_HDF5 "Enable HDF5 export support when HDF5 is available" ON)
 option(TENSORIUM_BSSN_PROFILE_KERNELS "Enable BSSN RHS timing instrumentation" OFF)
 option(TENSORIUM_BSSN_VALIDATE_TILDE_GAMMA_SYMBOLS "Validate on-the-fly Γ̃ symbols" OFF)
 option(TENSORIUM_ENABLE_COMPILER_OPT_REPORTS
@@ -104,24 +105,42 @@ if (USE_CUDA)
 
   add_compile_definitions(TENSORIUM_CUDA)
 
+  if (CUDAToolkit_NVCC_EXECUTABLE AND NOT CMAKE_CUDA_COMPILER)
+    set(CMAKE_CUDA_COMPILER "${CUDAToolkit_NVCC_EXECUTABLE}" CACHE FILEPATH "CUDA compiler" FORCE)
+  endif()
+
   enable_language(CUDA)
-  set(CMAKE_CUDA_COMPILER "${CUDAToolkit_NVCC_EXECUTABLE}")
 
-  set(CMAKE_CUDA_ARCHITECTURES 61 70 75 80 86 90)
-  add_compile_definitions(TENSORIUM_CUDA_ARCH=${CMAKE_CUDA_ARCHITECTURES})
+  set(CMAKE_CUDA_ARCHITECTURES 60 61 70 75 80 86 90)
+  string(JOIN "_" TENSORIUM_CUDA_ARCH_LIST ${CMAKE_CUDA_ARCHITECTURES})
+  add_compile_definitions(TENSORIUM_CUDA_ARCH_LIST="${TENSORIUM_CUDA_ARCH_LIST}")
 
   set(CMAKE_CUDA_STANDARD 17)
   set(CMAKE_CUDA_STANDARD_REQUIRED ON)
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3 -Xcompiler=-fPIC --use_fast_math -lineinfo -Wno-deprecated-gpu-targets")
 
   include_directories(${CUDAToolkit_INCLUDE_DIRS})
-  link_directories(${CUDAToolkit_LIBRARY_DIR})
 else()
   add_compile_definitions(TENSORIUM_NO_CUDA)
 endif()
 
 include_directories(${CMAKE_SOURCE_DIR}/includes)
 
+set(TENSORIUM_HDF5_FOUND FALSE)
+set(TENSORIUM_HDF5_LIBRARIES "")
+if(USE_HDF5)
+  find_package(HDF5 QUIET COMPONENTS C)
+  if(HDF5_FOUND)
+    set(TENSORIUM_HDF5_FOUND TRUE)
+    set(TENSORIUM_HDF5_LIBRARIES ${HDF5_LIBRARIES})
+    include_directories(${HDF5_INCLUDE_DIRS})
+    add_compile_definitions(TENSORIUM_USE_HDF5)
+    message(STATUS "HDF5 export support enabled")
+  else()
+    message(STATUS "HDF5 not found; export support disabled")
+  endif()
+endif()
+
 if(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64" OR CMAKE_SYSTEM_PROCESSOR MATCHES "amd64")
   if(USE_KNL)
     set(BASE_FLAGS "-O3 -march=knl -mtune=knl -Wno-ignored-attributes")

Plugins/TensoriumLLVM_IRCheck.cpp

@@ -4,7 +4,7 @@
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/Passes/PassBuilder.h"
-#include "llvm/Passes/PassPlugin.h"
+#include "llvm/Plugins/PassPlugin.h"
 #include "llvm/Support/raw_ostream.h"
 #include <iostream>
 

Plugins/TensoriumLLVM_ML_IR_GPU.cpp

@@ -3,7 +3,7 @@
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/Passes/PassBuilder.h"
-#include "llvm/Passes/PassPlugin.h"
+#include "llvm/Plugins/PassPlugin.h"
 
 using namespace llvm;
 

Pybind/README.md

@@ -73,25 +73,25 @@ PY
   - `tns.compute_christoffel`
   - `tns.compute_riemann_tensor`
   - Ricci contraction/scalar helpers
-- BSSN grid initial-data bindings:
-  - `bssn.BSSNGrid`
-  - `bssn.minkowski`
-  - `bssn.schwarzschild_isotropic`
-  - `bssn.kerr_schild_single`
-  - `bssn.binary_bowen_york_puncture_init`
-  - `bssn.binary_bowen_york_puncture_interpolated_init`
-  - `bssn.binary_bowen_york_puncture_twopunctures_c_init`
+- Z4c grid initial-data bindings:
+  - `z4c.Z4cGrid`
+  - `z4c.minkowski`
+  - `z4c.schwarzschild_isotropic`
+  - `z4c.kerr_schild_single`
+  - `z4c.binary_bowen_york_puncture_init`
+  - `z4c.binary_bowen_york_puncture_interpolated_init`
+  - `z4c.binary_bowen_york_puncture_twopunctures_c_init`
 - NumPy conversion helpers are available on bound classes via
   - `Class.from_numpy(...)`
   - `instance.to_numpy()`
 
-## BSSN quick start
+## Z4c quick start
 
 ```bash
 python3 - <<'PY'
 import tensorium
-g = tensorium.bssn.BSSNGrid(24, 24, 24, 4, 0.25, 0.25, 0.25)
-tensorium.bssn.minkowski(g)
+g = tensorium.z4c.Z4cGrid(24, 24, 24, 4, 0.25, 0.25, 0.25)
+tensorium.z4c.minkowski(g)
 alpha = g.alpha()
 print(alpha.shape, alpha.min(), alpha.max())
 PY
@@ -109,7 +109,7 @@ Runtime check:
 ```bash
 python3 - <<'PY'
 import tensorium
-print("two_punctures_c:", tensorium.bssn.has_twopunctures_c())
+print("two_punctures_c:", tensorium.z4c.has_twopunctures_c())
 PY
 ```
 
@@ -118,8 +118,8 @@ Explicit TwoPuncturesC init:
 ```bash
 python3 - <<'PY'
 import tensorium
-g = tensorium.bssn.BSSNGrid(64, 64, 64, 4, 0.125, 0.125, 0.125)
-tensorium.bssn.binary_bowen_york_puncture_twopunctures_c_init(
+g = tensorium.z4c.Z4cGrid(64, 64, 64, 4, 0.125, 0.125, 0.125)
+tensorium.z4c.binary_bowen_york_puncture_twopunctures_c_init(
     g,
     0.5, -1.0, 0.0, 0.0, [0.1, 0.0, 0.0], [0.0, 0.0, 0.0],
     0.5,  1.0, 0.0, 0.0, [-0.1, 0.0, 0.0], [0.0, 0.0, 0.0],
@@ -152,6 +152,6 @@ PY
 
 ## Not Yet Exposed
 
-- `BSSN_Grid` steppers and gauge runtime controls
+- `Z4c` steppers and gauge runtime controls
 - MPI-distributed paths and CUDA backends
 - High-level Python orchestration APIs for full NR runs