Use cuda/12.0 and fix cuda/hip compile issue

OP-DSL · Sep 20, 2024 · 85c0228 · 85c0228
1 parent 368eba0
commit 85c0228
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 5 deletions.
diff --git a/opp_lib/Makefile b/opp_lib/Makefile
@@ -24,7 +24,7 @@ HIPCC      = $(HIP_COMPILER)
 SYCLCC     = icpx
 
 CPPFLAGS   = -std=c++17 -Wall -fPIC -march=native $(CPPFLAGS_ADD)
-NVCCFLAGS  = -std=c++17 --extended-lambda -m64 -Xptxas -dlcm=ca -Xptxas=-v $(NVCCFLAGS_ADD)
+NVCCFLAGS  = -std=c++17 --extended-lambda -m64 -Xptxas -dlcm=ca $(NVCCFLAGS_ADD)
 HIPCCFLAGS = $(HIPCCFLAGS_ADD) -std=c++17 -isystem -D__HIP_PLATFORM_AMD__ -munsafe-fp-atomics -march=native 
 SYCLFLAGS  = -w -fsycl -std=c++17 $(SYCLFLAGS_ADD) -DUSE_SYCL
 
@@ -64,6 +64,7 @@ ifeq ($(EXT_HALOS), 0)
 	CPPFLAGS   += -DNO_EXTENDED_HALOS
 	NVCCFLAGS  += -DNO_EXTENDED_HALOS
 	HIPCCFLAGS += -DNO_EXTENDED_HALOS
+	SYCLFLAGS  += -DNO_EXTENDED_HALOS
 endif
 
 PETSC ?= 0
@@ -318,6 +319,7 @@ omp_mpi: clean mklib
 clean: 
 	     rm -rf *.o *.d *.a
 	     rm -rf $(OBJ)/*
-	     rm -rf $(LIB)/*
+
+#  rm -rf $(LIB)/*
 
 # ------------------------------------------------------------------------------------------
diff --git a/opp_lib/include/opp_cuda.h b/opp_lib/include/opp_cuda.h
@@ -58,6 +58,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define cutilCheckMsg(msg) __cutilCheckMsg(msg, __FILE__, __LINE__)
 
 #define OPP_GPU_THREADS_PER_BLOCK 32
+constexpr bool debugger = false;
 
 #define OPP_PARTICLE_MOVE_DONE { m.move_status = OPP_MOVE_DONE; }
 #define OPP_PARTICLE_NEED_MOVE { m.move_status = OPP_NEED_MOVE; }

diff --git a/opp_lib/include/opp_hip.h b/opp_lib/include/opp_hip.h
@@ -57,6 +57,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define cutilCheckMsg(msg) __cutilCheckMsg(msg, __FILE__, __LINE__)
 
 #define OPP_GPU_THREADS_PER_BLOCK 32
+constexpr bool debugger = false;
 
 #define OPP_PARTICLE_MOVE_DONE { m.move_status = OPP_MOVE_DONE; }
 #define OPP_PARTICLE_NEED_MOVE { m.move_status = OPP_NEED_MOVE; }
@@ -490,7 +491,7 @@ class opp_mem {
     // Copy data from host to device, create new device arrays if requested
     template <typename T>
     inline static void copy_host_to_dev(T*& data_d, const T *data_h, size_t copy_count, 
-                                       bool create_new = false, size_t alloc_count = 0) {
+                            bool no_wait = false, bool create_new = false, size_t alloc_count = 0) {
         if (create_new) {
             if (data_d != nullptr)  
                 opp_mem::dev_free<T>(data_d);

diff --git a/scripts/source/demos_sycl b/scripts/source/demos_sycl
@@ -19,9 +19,12 @@ export SYCLFLAGS_ADD="-fp-model precise -fsycl-targets=nvptx64-nvidia-cuda -Xsyc
 # export SYCLFLAGS_ADD=" -fsycl-targets=spir64_x86_64 "
 # export SYCLFLAGS_ADD="-fp-model precise -fsycl-targets=nvptx64-nvidia-cuda -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_90 -Xcuda-ptxas --maxrregcount=128"
 
-module load cuda/12.5
+# module load cuda/12.5
+# export CUDA_INSTALL_PATH="/opt/cuda/12.5"
+module load cuda/12.0
+export CUDA_INSTALL_PATH="/opt/nvidia/hpc_sdk/Linux_x86_64/23.1/cuda/12.0"
+
 export NVCCFLAGS_ADD='-gencode arch=compute_90,code=sm_90'
-export CUDA_INSTALL_PATH="/opt/cuda/12.5"
 export CUDA_ARCH=90
 
 . /home/zl/spack_install/spack_demos/opt/spack/linux-debian12-zen3/gcc-12.2.0/intel-oneapi-compilers-2024.2.1-jpypt464ei5wbrl2bly2y3hbngnhp3j4/setvars.sh --include-intel-llvm
@@ -31,6 +34,13 @@ export PETSC_INSTALL_PATH=/home/zl/lib_install/petsc-3.20.5-oneapi-2024.2.1_demo
 export HDF5_INSTALL_PATH=$(spack location -i hdf5@1.14.3)
 export LD_LIBRARY_PATH=$HDF5_INSTALL_PATH/lib:$LD_LIBRARY_PATH
 
+# install cuda plugin!
+# sh oneapi-for-nvidia-gpus-2024.2.1-cuda-12.0-linux.sh --install-dir /home/zl/spack_install/spack_demos/opt/spack/linux-debian12-zen3/gcc-12.2.0/intel-oneapi-compilers-2024.2.1-jpypt464ei5wbrl2bly2y3hbngnhp3j4
+
+# dpct conversions
+# dpct --cuda-include-path=/opt/nvidia/hpc_sdk/Linux_x86_64/23.7/cuda/include --process-all --in-root=. --out-root=/home/zl/phd/test/OP-PIC/dpct_lib
+# dpct --extra-arg="-DUSE_THRUST" --extra-arg="-DUSE_MPI" --extra-arg="-I/home/zl/phd/test/OP-PIC/opp_lib/include" --cuda-include-path=/opt/nvidia/hpc_sdk/Linux_x86_64/23.7/cuda/include --process-all --in-root=. --out-root=/home/zl/phd/test/OP-PIC/dpct_lib
+# dpct --extra-arg="-DUSE_THRUST" --extra-arg="-DUSE_MPI" --extra-arg="-I/home/zl/phd/test/OP-PIC/opp_lib/include" --cuda-include-path=/opt/nvidia/hpc_sdk/Linux_x86_64/23.7/cuda/include --process-all --in-root=. --out-root=/home/zl/phd/test/OP-PIC/dpct_cabanapic_cg
 
 
 # module load compiler-rt/latest