fixing bugs in destructor of LUstruct_v100

xiaoyeli · Oct 21, 2023 · b19bce4 · b19bce4
1 parent eef70c7
commit b19bce4
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 5 deletions.
diff --git a/SRC/TRF3dV100/batch_factorize.cu b/SRC/TRF3dV100/batch_factorize.cu
@@ -804,8 +804,9 @@ void freeBatchFactorizeWorkspace(BatchFactorizeWorkspace* ws)
     gpuErrchk( cudaFree(d_localLU.Unzval_br_new_dat) );
     gpuErrchk( cudaFree(d_localLU.Unzval_br_new_offset) );
     gpuErrchk( cudaFree(d_localLU.Unzval_br_new_ptr) );
-
+#ifdef HAVE_MAGMA
     magma_queue_destroy(ws->magma_queue);
+#endif
     cublasDestroy( ws->cuhandle );
     gpuErrchk( cudaStreamDestroy(ws->stream) );
 }

diff --git a/SRC/TRF3dV100/lupanels.hpp b/SRC/TRF3dV100/lupanels.hpp
@@ -437,13 +437,16 @@ struct LUstruct_v100
 
         for (int_t i = 0; i < CEILING(nsupers, Pc); ++i)
             if (i * Pc + mycol < nsupers && isNodeInMyGrid[i * Pc + mycol] == 1){
-                SUPERLU_FREE(lPanelVec[i].index);
+                if(lPanelVec[i].index)
+                    SUPERLU_FREE(lPanelVec[i].index);
                 // SUPERLU_FREE(lPanelVec[i].val);
             }
 
         for (int_t i = 0; i < CEILING(nsupers, Pr); ++i)
             if (i * Pr + myrow < nsupers && isNodeInMyGrid[i * Pr + myrow] == 1){
+                if(uPanelVec[i].index)
                 SUPERLU_FREE(uPanelVec[i].index);
+                if(uPanelVec[i].val)
                 SUPERLU_FREE(uPanelVec[i].val);
             }
 

diff --git a/example_scripts/run_cmake_build_perlmutter_gcc_nvshmem.sh b/example_scripts/run_cmake_build_perlmutter_gcc_nvshmem.sh
@@ -42,9 +42,9 @@ export LD_LIBRARY_PATH=${LD_LIBRARY_PATH//\/usr\/local\/cuda-11.7\/compat:/}
 NVSHMEM_HOME=/global/cfs/cdirs/m3894/lib/PrgEnv-gnu/nvshmem_src_2.8.0-3/build/
 #NVSHMEM_HOME=${CRAY_NVIDIA_PREFIX}/comm_libs/nvshmem/
 cmake .. \
-  -DCMAKE_C_FLAGS="-O2 -DGPU_SOLVE -std=c11 -DPRNTlevel=0 -DPROFlevel=0 -DDEBUGlevel=0 -DAdd_" \
-  -DCMAKE_CXX_FLAGS="-O2" \
-  -DCMAKE_Fortran_FLAGS="-O2" \
+  -DCMAKE_C_FLAGS="-DGPU_SOLVE -std=c11 -DPRNTlevel=0 -DPROFlevel=0 -DDEBUGlevel=0 -DAdd_" \
+  -DCMAKE_CXX_FLAGS="" \
+  -DCMAKE_Fortran_FLAGS="" \
   -DCMAKE_CXX_COMPILER=CC \
   -DCMAKE_C_COMPILER=cc \
   -DCMAKE_Fortran_COMPILER=ftn \