ELPA updates

.github/workflows/github_actions.yml

@@ -190,11 +190,15 @@ jobs:
           echo F77 compiler version `"$F77" -v`
       - name: patch source code if needed
         run: |
-          if [[ "$CC" == "icx" ]]; then
+          if [ $(uname -s) == "Linux" ]; then
           sudo apt-get install -q -y patch
+          fi
+          if [[ "$CC" == "icx" ]]; then
           echo patching g2test to prevent icx failures
           patch -p0  < ./travis/g2t.patch
           fi
+          echo patching sprs_perf.c
+          patch -p0  < ./travis/sprs_perf.patch
       - name: before_install
         if: steps.ga-cache-install.outputs.cache-hit != 'true'
         run: |

global/src/scalapack.F

@@ -3729,8 +3729,6 @@ subroutine ga_evp_real(g_a, g_b, eval, nb8, mout)
 c     
       call SLinit2(n)
       oactive=iam.lt.maxproc
-      mpA = numroc(dimA1, nb, myrow2, zero4, nprow2)
-      nqA = numroc(dimA2, nb, mycol2, zero4, npcol2)
       if(g_a.ne.g_b) then
          if(.not.ga_deallocate(g_b))
      E        call ga_error(' ga_evp_real: ga_deallocate failed',0)
@@ -3903,10 +3901,10 @@ subroutine ga_evp_real(g_a, g_b, eval, nb8, mout)
 #endif
 #if HAVE_ELPA_NVIDIA
 C checks for the following env. variable setting
-c ELPA_DEFAULT_real_kernel=ELPA_2STAGE_REAL_NVIDIA_GPU
+c     ELPA_DEFAULT_nvidia-gpu=1
            call get_environment_variable(
-     A          "ELPA_DEFAULT_real_kernel", value=env_out, status=err8)
-           if(trim(env_out).eq."ELPA_2STAGE_REAL_NVIDIA_GPU") then
+     A          "ELPA_DEFAULT_nvidia_gpu", value=env_out, status=err8)
+           if(trim(env_out).eq."1") then
               call e%set("nvidia-gpu", int(1,c_int), elpa_err)
               if(elpa_err.ne.ELPA_OK) call ga_error(
      A             'ga-evp_real: e%set nvidia-gpu failed ',elpa_err)

global/testing/CMakeLists.txt

@@ -83,8 +83,8 @@ ga_add_parallel_test (perf2         "perf2.c"            )
 ga_add_parallel_test (ga-mpi        "ga-mpi.c util.c"    )
 ga_add_parallel_test (lock          "lock.c util.c"      )
 ga_add_parallel_test (comm_init     "comm_init.c util.c" )
-ga_add_parallel_test (comm_init     "sprs_test.c util.c" )
-ga_add_parallel_test (comm_init     "sprs_perf.c util.c" )
+ga_add_parallel_test (sprs_test     "sprs_test.c util.c" )
+ga_add_parallel_test (sprs_perf     "sprs_perf.c util.c" )
 
 ga_add_parallel_test (patch_enumc         "patch_enumc.c util.c"        )
 ga_add_parallel_test (simple_groups_commc "simple_groups_commc.c util.c")

travis/sprs_perf.patch

@@ -0,0 +1,17 @@
+diff --git global/testing/sprs_perf.c global/testing/sprs_perf.c
+index 311222a6..97dc9808 100644
+--- global/testing/sprs_perf.c
++++ global/testing/sprs_perf.c
+@@ -9,8 +9,10 @@
+
+ #define WRITE_VTK
+ #define CG_SOLVE 1
+-#define NDIM_SPRS 1048576
+-#define NDIM_DNS 16384
++#define NDIM_SPRS_ 1048576
++#define NDIM_DNS_ 16384
++#define NDIM_SPRS 262144
++#define NDIM_DNS 8192
+
+ /**
+  *  Solve Laplace's equation on a cubic domain using the sparse matrix