Fix HIP on CUDA build

cp2k · Apr 28, 2022 · 57db85b · 57db85b
1 parent 9eaad38
commit 57db85b
Show file tree

Hide file tree

Showing 4 changed files with 11 additions and 11 deletions.
diff --git a/exts/Makefile.inc b/exts/Makefile.inc
@@ -20,7 +20,7 @@ DBCSR_ACC      := $(OFFLOAD_CC)
 DBCSR_ACCFLAGS := $(OFFLOAD_FLAGS) -Xcompiler='-Wno-error'
 else ifeq (hip,$(DBCSR_USE_ACCEL))
 DBCSR_ACC      := $(CXX)
-DBCSR_ACCFLAGS := $(CXXFLAGS)
+DBCSR_ACCFLAGS := $(CXXFLAGS) -Wno-error=unused-parameter
 else ifeq (opencl,$(DBCSR_USE_ACCEL))
 DBCSR_ACC      := $(CC)
 DBCSR_ACCFLAGS := $(CFLAGS)

diff --git a/src/grid/hip/grid_hip_context.cu b/src/grid/hip/grid_hip_context.cu
@@ -89,7 +89,6 @@ extern "C" void grid_hip_create_task_list(
   memset(ctx->number_of_tasks_per_level_.data(), 0, sizeof(int) * nlevels);
 
   std::vector<rocm_backend::task_info> tasks_host(ntasks);
-  memset(tasks_host.data(), 0, sizeof(rocm_backend::task_info) * ntasks);
 
   size_t coef_size = 0;
 
@@ -112,6 +111,7 @@ extern "C" void grid_hip_create_task_list(
     const grid_basis_set *ibasis = basis_sets[ikind];
     const grid_basis_set *jbasis = basis_sets[jkind];
 
+    tasks_host[i] = {};
     tasks_host[i].level = level;
     tasks_host[i].iatom = iatom;
     tasks_host[i].jatom = jatom;

diff --git a/src/grid/hip/grid_hip_context.h b/src/grid/hip/grid_hip_context.h
@@ -198,7 +198,7 @@ template <typename T> class grid_info {
         ((fabs(dh_[0] * dh_[3] + dh_[1] * dh_[4] + dh_[2] * dh_[5]) * norm1 *
           norm2) < 1e-12);
 
-    orthorhombic_ = orthogonal[0] * orthogonal[1] * orthogonal[2];
+    orthorhombic_ = orthogonal[0] && orthogonal[1] && orthogonal[2];
   }
 
   inline void copy_to_host(double *data__, offloadStream_t &stream) {

diff --git a/tools/toolchain/scripts/generate_arch_files.sh b/tools/toolchain/scripts/generate_arch_files.sh
@@ -122,7 +122,7 @@ if [ "${ENABLE_CUDA}" = __TRUE__ ] && [ "${GPUVER}" != no ]; then
   CUDA_DFLAGS="-D__OFFLOAD_CUDA -D__DBCSR_ACC IF_DEBUG(-D__OFFLOAD_PROFILING|)"
   LIBS="${LIBS} IF_CUDA(${CUDA_LIBS}|)"
   DFLAGS="IF_CUDA(${CUDA_DFLAGS}|) ${DFLAGS}"
-  NVFLAGS="-g -arch sm_${ARCH_NUM} -O3 -allow-unsupported-compiler -Xcompiler='-fopenmp' --std=c++11 \$(DFLAGS)"
+  NVFLAGS="-g -arch sm_${ARCH_NUM} -O3 -allow-unsupported-compiler -Xcompiler='-fopenmp -Wall -Wextra -Werror' --std=c++11 \$(DFLAGS)"
   check_command nvcc "cuda"
   check_lib -lcudart "cuda"
   check_lib -lnvrtc "cuda"
@@ -171,10 +171,10 @@ if [ "${ENABLE_HIP}" = __TRUE__ ] && [ "${GPUVER}" != no ]; then
       add_lib_from_paths HIP_LDFLAGS "libroctx64.*" $LIB_PATHS
       check_lib -lroctracer64 "hip"
       add_lib_from_paths HIP_LDFLAGS "libroctracer64.*" $LIB_PATHS
-      HIP_FLAGS+="-fPIE -D__HIP_PLATFORM_AMD__ -g --offload-arch=gfx906 -O3 --std=c++11 \$(DFLAGS)"
+      HIP_FLAGS+="-fPIE -D__HIP_PLATFORM_AMD__ -g --offload-arch=gfx906 -O3 --std=c++11 -Wall -Wextra -Werror \$(DFLAGS)"
       LIBS+=" IF_HIP(-lamdhip64 -lhipfft -lhipblas -lrocblas IF_DEBUG(-lroctx64 -lroctracer64|)|)"
       DFLAGS+=" IF_HIP(-D__HIP_PLATFORM_AMD__ -D__OFFLOAD_HIP IF_DEBUG(-D__OFFLOAD_PROFILING|)|)  -D__DBCSR_ACC"
-      CXXFLAGS+=" -fopenmp -std=c++11"
+      CXXFLAGS+=" -fopenmp -std=c++11 -Wall -Wextra -Werror"
       ;;
     Mi100)
       check_lib -lamdhip64 "hip"
@@ -187,10 +187,10 @@ if [ "${ENABLE_HIP}" = __TRUE__ ] && [ "${GPUVER}" != no ]; then
       add_lib_from_paths HIP_LDFLAGS "libroctx64.*" $LIB_PATHS
       check_lib -lroctracer64 "hip"
       add_lib_from_paths HIP_LDFLAGS "libroctracer64.*" $LIB_PATHS
-      HIP_FLAGS+="-fPIE -D__HIP_PLATFORM_AMD__ -g --offload-arch=gfx908 -O3 --std=c++11 \$(DFLAGS)"
+      HIP_FLAGS+="-fPIE -D__HIP_PLATFORM_AMD__ -g --offload-arch=gfx908 -O3 --std=c++11 -Wall -Wextra -Werror \$(DFLAGS)"
       LIBS+=" IF_HIP(-lamdhip64 -lhipfft -lhipblas -lrocblas IF_DEBUG(-lroctx64 -lroctracer64|)|)"
       DFLAGS+=" IF_HIP(-D__HIP_PLATFORM_AMD__ -D__OFFLOAD_HIP IF_DEBUG(-D__OFFLOAD_PROFILING|)|) -D__DBCSR_ACC"
-      CXXFLAGS+=" -fopenmp -std=c++11"
+      CXXFLAGS+=" -fopenmp -std=c++11 -Wall -Wextra -Werror"
       ;;
     *)
       check_command nvcc "cuda"
@@ -200,7 +200,7 @@ if [ "${ENABLE_HIP}" = __TRUE__ ] && [ "${GPUVER}" != no ]; then
       check_lib -lcufft "cuda"
       check_lib -lcublas "cuda"
       DFLAGS+=" IF_HIP(-D__HIP_PLATFORM_NVIDIA__ -D__HIP_PLATFORM_NVCC__ -D__OFFLOAD_HIP |) -D__DBCSR_ACC"
-      HIP_FLAGS=" -g -arch sm_${ARCH_NUM} -O3 -Xcompiler='-fopenmp' --std=c++11 \$(DFLAGS)"
+      HIP_FLAGS=" -g -arch sm_${ARCH_NUM} -O3 -Xcompiler='-fopenmp -Wall -Wextra -Werror' --std=c++11 \$(DFLAGS)"
       add_include_from_paths CUDA_CFLAGS "cuda.h" $INCLUDE_PATHS
       HIP_INCLUDES+=" -I${CUDA_PATH:-${CUDA_HOME:-/CUDA_HOME-notset}}/include"
       # GCC issues lots of warnings for hip/nvidia_detail/hip_runtime_api.h
@@ -277,7 +277,7 @@ gen_arch_file() {
 #
 GPUVER        = \${GPUVER}
 OFFLOAD_CC    = \${NVCC}
-OFFLOAD_FLAGS = \${NVFLAGS} -Xcompiler='-Wall -Wextra -Werror'
+OFFLOAD_FLAGS = \${NVFLAGS}
 OFFLOAD_TARGET = cuda
 EOF
   fi
@@ -287,7 +287,7 @@ EOF
 #
 GPUVER        = \${GPUVER}
 OFFLOAD_CC    = \${ROCM_PATH}/hip/bin/hipcc
-OFFLOAD_FLAGS = \${HIP_FLAGS} \${HIP_INCLUDES} -Wall -Wextra -Werror
+OFFLOAD_FLAGS = \${HIP_FLAGS} \${HIP_INCLUDES}
 OFFLOAD_TARGET = hip
 EOF
   fi