Merge pull request #63 from mkstoyanov/fixing_compiler_warnings

Fixes numerous warnings across several compilers.

include/heffte_backend_oneapi.h

@@ -508,15 +508,15 @@ class onemkl_executor : public executor_base{
         }else if (size2 == 0){
             plan.set_value(oneapi::mkl::dft::config_param::NUMBER_OF_TRANSFORMS, (MKL_LONG) howmanyffts);
             plan.set_value(oneapi::mkl::dft::config_param::PLACEMENT, DFTI_INPLACE);
-            plan.set_value(oneapi::mkl::dft::config_param::INPUT_STRIDES, embed.data());
-            plan.set_value(oneapi::mkl::dft::config_param::OUTPUT_STRIDES, embed.data());
+            plan.set_value(oneapi::mkl::dft::config_param::FWD_STRIDES, embed.data());
+            plan.set_value(oneapi::mkl::dft::config_param::BWD_STRIDES, embed.data());
             plan.set_value(oneapi::mkl::dft::config_param::FWD_DISTANCE, (MKL_LONG) dist);
             plan.set_value(oneapi::mkl::dft::config_param::BWD_DISTANCE, (MKL_LONG) dist);
         }else{
             plan.set_value(oneapi::mkl::dft::config_param::NUMBER_OF_TRANSFORMS, (MKL_LONG) howmanyffts);
             plan.set_value(oneapi::mkl::dft::config_param::PLACEMENT, DFTI_INPLACE);
-            plan.set_value(oneapi::mkl::dft::config_param::INPUT_STRIDES, embed.data());
-            plan.set_value(oneapi::mkl::dft::config_param::OUTPUT_STRIDES, embed.data());
+            plan.set_value(oneapi::mkl::dft::config_param::FWD_STRIDES, embed.data());
+            plan.set_value(oneapi::mkl::dft::config_param::BWD_STRIDES, embed.data());
             plan.set_value(oneapi::mkl::dft::config_param::FWD_DISTANCE, (MKL_LONG) dist);
             plan.set_value(oneapi::mkl::dft::config_param::BWD_DISTANCE, (MKL_LONG) dist);
         }
@@ -620,8 +620,8 @@ class onemkl_executor_r2c : public executor_base{
         plan.set_value(oneapi::mkl::dft::config_param::PLACEMENT, DFTI_NOT_INPLACE);
         plan.set_value(oneapi::mkl::dft::config_param::CONJUGATE_EVEN_STORAGE, DFTI_COMPLEX_COMPLEX);
         MKL_LONG slstride[] = {0, static_cast<MKL_LONG>(stride)};
-        plan.set_value(oneapi::mkl::dft::config_param::INPUT_STRIDES, slstride);
-        plan.set_value(oneapi::mkl::dft::config_param::OUTPUT_STRIDES, slstride);
+        plan.set_value(oneapi::mkl::dft::config_param::FWD_STRIDES, slstride);
+        plan.set_value(oneapi::mkl::dft::config_param::BWD_STRIDES, slstride);
         plan.set_value(oneapi::mkl::dft::config_param::FWD_DISTANCE, (MKL_LONG) rdist);
         plan.set_value(oneapi::mkl::dft::config_param::BWD_DISTANCE, (MKL_LONG) cdist);
         plan.commit(q);

include/heffte_r2r_executor.h

@@ -143,7 +143,7 @@ struct cpu_cos1_pre_pos_processor{
     static void pre_forward(void*, int length, precision const input[], precision fft_signal[]){
         for (int i = 0; i < length-1; i++){
             fft_signal[2*i] = input[i];
-            fft_signal[2*i+1] = 0.0; 
+            fft_signal[2*i+1] = 0.0;
         }
         for (int i = 1; i < length; i++){
             fft_signal[4*(length-1)-2*i] = input[i];
@@ -190,6 +190,11 @@ struct cpu_sin1_pre_pos_processor{};
  */
 template<typename fft_backend_tag, typename prepost_processor>
 struct real2real_executor : public executor_base{
+    //! \brief Will not be used, suppresses compiler warnings
+    using executor_base::forward;
+    //! \brief Will not be used, suppresses compiler warnings
+    using executor_base::backward;
+
     //! \brief Construct a plan for batch 1D transforms.
     template<typename index>
     real2real_executor(typename backend::device_instance<typename backend::buffer_traits<fft_backend_tag>::location>::stream_type cstream, box3d<index> const box, int dimension) :

test/post_install_test.cmake.sh

@@ -21,6 +21,7 @@ fi
     -DCMAKE_CXX_FLAGS="@CMAKE_CXX_FLAGS@" \
     -DHeffte_DIR=@CMAKE_INSTALL_FULL_LIBDIR@/cmake/Heffte \
     -DMPI_CXX_COMPILER="@MPI_CXX_COMPILER@" \
+    -DMPIEXEC_EXECUTABLE="@MPIEXEC_EXECUTABLE@" \
     $heffte_mpic_compiler \
     $heffte_mpif_compiler \
     -DMPIEXEC_NUMPROC_FLAG="@MPIEXEC_NUMPROC_FLAG@" \

test/test_common.h

@@ -202,11 +202,15 @@ using gpu_backend = heffte::backend::rocfft;
 
 hipStream_t make_stream(backend::rocfft){
     hipStream_t result;
-    hipStreamCreateWithFlags(&result, hipStreamNonBlocking);
+    rocm::check_error( hipStreamCreateWithFlags(&result, hipStreamNonBlocking), "hipStreamCreateWithFlags()");
     return result;
 }
-void sync_stream(hipStream_t){ hipDeviceSynchronize(); }
-void free_stream(hipStream_t stream){ hipStreamDestroy(stream); }
+void sync_stream(hipStream_t){
+    rocm::check_error( hipDeviceSynchronize(), "hipDeviceSynchronize()");
+}
+void free_stream(hipStream_t stream){
+    rocm::check_error( hipStreamDestroy(stream), "hipStreamDestroy()");
+}
 #endif
 #ifdef Heffte_ENABLE_ONEAPI
 using gpu_backend = heffte::backend::onemkl;

test/test_units_nompi.cpp

@@ -597,15 +597,21 @@ void test_in_node_transpose(){
 
     auto active_intput = test_traits<backend_tag>::load(input);
     vcontainer result(24);
-    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), nullptr);
+
+    // when doing out-of-place transpose we do not need a workspace and can use a nullptr instead,
+    // but the in/out-place is checked at runtime  and one of the branches that is never used
+    // will still be compiled with the hard-coded nullptr, to suppress the warning we are using a dummy-null dnull
+    scalar_type *dnull = result.data();
+
+    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), dnull);
 
     sassert(match(result, reference));
 
     // test 2, transpose the data to order (2, 1, 0)
     box3d<> destination2(std::array<int, 3>{0, 0, 0}, std::array<int, 3>{1, 2, 3}, std::array<int, 3>{2, 1, 0});
     plans.clear();
     heffte::compute_overlap_map_transpose_pack(0, 1, destination2, {inbox}, proc, offset, sizes, plans);
-    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), nullptr);
+    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), dnull);
 
     reference = {1.0,  7.0, 13.0, 19.0,  3.0,  9.0, 15.0, 21.0,  5.0, 11.0, 17.0, 23.0,
                  2.0,  8.0, 14.0, 20.0,  4.0, 10.0, 16.0, 22.0,  6.0, 12.0, 18.0, 24.0};
@@ -615,14 +621,14 @@ void test_in_node_transpose(){
     plans.clear();
     heffte::compute_overlap_map_transpose_pack(0, 1, inbox, {destination2}, proc, offset, sizes, plans);
     auto active_reference = test_traits<backend_tag>::load(reference);
-    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_reference.data(), result.data(), nullptr);
+    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_reference.data(), result.data(), dnull);
     sassert(match(result, input));
 
     // test 3, transpose the data to order (0, 2, 1)
     box3d<> destination3(std::array<int, 3>{0, 0, 0}, std::array<int, 3>{1, 2, 3}, std::array<int, 3>{0, 2, 1});
     plans.clear();
     heffte::compute_overlap_map_transpose_pack(0, 1, destination3, {inbox}, proc, offset, sizes, plans);
-    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), nullptr);
+    heffte::reshape3d_transpose<ltag, int>(device.stream(), plans[0]).apply(1, active_intput.data(), result.data(), dnull);
 
     reference = {1.0, 2.0,  7.0,  8.0, 13.0, 14.0, 19.0, 20.0,
                  3.0, 4.0,  9.0, 10.0, 15.0, 16.0, 21.0, 22.0,