Toolchain: Make Dockerfile.cuda_mkl cache-friendly

tools/toolchain/Dockerfile.cuda_mkl

@@ -32,7 +32,17 @@ ENV MKLROOT=/opt/intel/compilers_and_libraries/linux/mkl
 
 # Build toolchain.
 WORKDIR /opt/cp2k-toolchain
-COPY ./scripts ./scripts
+RUN mkdir scripts
+COPY ./scripts/VERSION \
+     ./scripts/parse_if.py \
+     ./scripts/tool_kit.sh \
+     ./scripts/common_vars.sh \
+     ./scripts/signal_trap.sh \
+     ./scripts/get_openblas_arch.sh \
+     ./scripts/
+
+# Dry-run leaves behind config files for the followup install scripts.
+# This breaks up the lengthy installation into smaller docker build steps.
 COPY ./install_cp2k_toolchain.sh .
 RUN ./install_cp2k_toolchain.sh  \
     --mpi-mode=mpich             \
@@ -40,6 +50,38 @@ RUN ./install_cp2k_toolchain.sh  \
     --with-scalapack=no          \
     --libint-lmax=${LIBINT_LMAX} \
     --gpu-ver=${GPU_VERSION}     \
-    && rm -rf ./build
+    --dry-run
+
+COPY ./scripts/stage0/ ./scripts/stage0/
+RUN  ./scripts/stage0/install_stage0.sh && rm -rf ./build
+
+COPY ./scripts/stage1/ ./scripts/stage1/
+RUN  ./scripts/stage1/install_stage1.sh && rm -rf ./build
+
+COPY ./scripts/stage2/ ./scripts/stage2/
+RUN  ./scripts/stage2/install_stage2.sh && rm -rf ./build
+
+COPY ./scripts/stage3/ ./scripts/stage3/
+RUN  ./scripts/stage3/install_stage3.sh && rm -rf ./build
+
+COPY ./scripts/stage4/ ./scripts/stage4/
+RUN  ./scripts/stage4/install_stage4.sh && rm -rf ./build
+
+COPY ./scripts/stage5/ ./scripts/stage5/
+RUN  ./scripts/stage5/install_stage5.sh && rm -rf ./build
+
+COPY ./scripts/stage6/ ./scripts/stage6/
+RUN  ./scripts/stage6/install_stage6.sh && rm -rf ./build
+
+COPY ./scripts/stage7/ ./scripts/stage7/
+RUN  ./scripts/stage7/install_stage7.sh && rm -rf ./build
+
+COPY ./scripts/stage8/ ./scripts/stage8/
+RUN  ./scripts/stage8/install_stage8.sh && rm -rf ./build
+
+COPY ./scripts/arch_base.tmpl \
+     ./scripts/generate_arch_files.sh \
+     ./scripts/
+RUN ./scripts/generate_arch_files.sh && rm -rf ./build
 
 #EOF