toolchain: add option to compile DBCSR for NVIDIA's V100

Enables use in CP2K of support for NVIDIA's V100 introduced in DBCSR by PR 172. And remove dangerous manual definition of GNU compiler version.
cp2k · Jul 12, 2019 · 3e58b4d · 3e58b4d
1 parent 31b514a
commit 3e58b4d
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 4 deletions.
diff --git a/tools/toolchain/install_cp2k_toolchain.sh b/tools/toolchain/install_cp2k_toolchain.sh
@@ -94,7 +94,7 @@ OPTIONS:
                           switch --math-mode to the respective modes, BUT
                           --with-reflapack option do not have this effect.
 --gpu-ver                 Selects the GPU architecture for which to compile. Available
-                          options are: K20X, K40, K80, P100, no. Default: no.
+                          options are: K20X, K40, K80, P100, V100, no. Default: no.
                           The script will determine the correct corresponding value for
                           nvcc's '-arch' flag.
 --libint-lmax             Maximum supported angular momentum by libint.
@@ -445,13 +445,16 @@ while [ $# -ge 1 ] ; do
                 P100)
                     export GPUVER=P100
                     ;;
+                V100)
+                    export GPUVER=V100
+                    ;;
                 no)
                     export GPUVER=no
                     ;;
                 *)
                     export GPUVER=no
                     report_error ${LINENO} \
-                    "--gpu-ver currently only supports K20X, K40, K80, P100 and no as options"
+                    "--gpu-ver currently only supports K20X, K40, K80, P100, V100 and no as options"
                     exit 1
             esac
             ;;
@@ -889,12 +892,15 @@ case $GPUVER in
     P100)
         export ARCH_NUM=60
         ;;
+    V100)
+        export ARCH_NUM=70
+        ;;
     no)
         export ARCH_NUM=no
         ;;
     *)
         report_error ${LINENO} \
-        "--gpu-ver currently only supports K20X, K40, K80, P100 as options"
+        "--gpu-ver currently only supports K20X, K40, K80, P100, V100 as options"
 esac
 
 # write toolchain environment

diff --git a/tools/toolchain/scripts/generate_arch_files.sh b/tools/toolchain/scripts/generate_arch_files.sh
@@ -160,7 +160,7 @@ gen_arch_file() {
 CXX         = \${CC}
 CXXFLAGS    = \${CXXFLAGS} -I\\\${CUDA_PATH}/include -std=c++11 IF_OMP(-fopenmp|)
 GPUVER      = \${GPUVER}
-NVCC        = \${NVCC} -D__GNUC__=4 -D__GNUC_MINOR__=9
+NVCC        = \${NVCC}
 NVFLAGS     = \${NVFLAGS}
 EOF
     fi