Migrate build system from Intel oneAPI to AMD AOCC/AOCL toolchain

- Add TOOLCHAIN=aocc option with flang/clang++/mpicxx compilers
- Replace Intel flags (-xHost/-fma/-ipo/-qopenmp) with AOCC flags
  (-march=znver5/-ffast-math/-flto/-fopenmp) targeting EPYC 9755
- Replace Intel oneMKL with AMD AOCL (BLIS + libFLAME + amdlibm)
- Replace Intel TBBMALLOC with system jemalloc
- Change MKL-specific headers to standard CBLAS/LAPACKE
  (TwoPunctures.C, gaussj.C)
- Guard TBBMALLOC to Intel toolchain only

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

AMSS_NCKU_source/makefile.inc

@@ -1,6 +1,7 @@
 ## Toolchain selection
-##   nvhpc : NVIDIA HPC SDK + CUDA-aware MPI (default)
+##   nvhpc : NVIDIA HPC SDK + CUDA-aware MPI
 ##   intel : Intel oneAPI toolchain (legacy path)
+##   aocc  : AMD AOCC + AOCL + OpenMPI (for AMD EPYC Zen 5, with CUDA)
 TOOLCHAIN ?= intel
 
 ## PGO build mode switch (ABE only; TwoPunctureABE always uses opt flags)
@@ -26,6 +27,10 @@ MKLROOT ?= /home/intel/oneapi/mkl/latest
 MKL_LIBDIR ?= $(MKLROOT)/lib/intel64
 MKL_INC ?= -I$(MKLROOT)/include
 
+## AMD AOCC toolchain paths (used when TOOLCHAIN=aocc)
+AOCL_ROOT ?= /home/aocc/aocl/5.2.0/aocc
+OMPI_PREFIX ?= /home/aocc/aocc-openmpi
+
 NVHPC_ROOT ?= /home/nvidia/hpc_sdk/Linux_x86_64/25.11
 CUDA_HOME  ?= $(NVHPC_ROOT)/cuda
 CUDA_ARCH  ?= sm_80
@@ -67,6 +72,16 @@ LDLIBS       = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) \
                -lmkl_intel_lp64 -lmkl_sequential -lmkl_core \
                -lifcore -limf -liomp5 -lpthread -lm -ldl \
                -L$(CUDA_HOME)/lib64 -Wl,-rpath,$(CUDA_HOME)/lib64 -lcuda -lcudart
+else ifeq ($(TOOLCHAIN),aocc)
+f90          = flang
+f77          = flang
+CXX          = clang++
+CC           = clang
+CLINKER      = $(OMPI_PREFIX)/bin/mpicxx
+filein       = -I/usr/include/ -I$(AOCL_ROOT)/include -I$(CUDA_HOME)/include
+LDLIBS       = -L$(AOCL_ROOT)/lib -lblis -lflame -lamdlibm -lflang -lpgmath \
+               -ljemalloc -lpthread -lm -ldl -lomp \
+               -L$(CUDA_HOME)/lib64 -Wl,-rpath,$(CUDA_HOME)/lib64 -lcuda -lcudart
 else ifeq ($(TOOLCHAIN),nvhpc)
 f90          = mpifort
 f77          = mpifort
@@ -82,9 +97,11 @@ LDLIBS       = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) \
                -fortranlibs
 endif
 
+ifeq ($(TOOLCHAIN),intel)
 ifeq ($(USE_TBBMALLOC),1)
 LDLIBS := $(TBBMALLOC_LIBS) $(LDLIBS)
 endif
+endif
 
 Cu = $(NVHPC_ROOT)/compilers/bin/nvcc
 CUDA_LIB_PATH = -L$(CUDA_HOME)/lib64 -I$(CUDA_HOME)/include