From 62eee6bcad8413195d37da11b9aed0a99bed53d0 Mon Sep 17 00:00:00 2001
From: Adrian Pope <apope@anl.gov>
Date: Thu, 27 Jun 2024 19:58:07 -0500
Subject: [PATCH] fixed llvm gpu arch cross compile; made compiler flags easier
 to modify; matched llvm run params to nvhpc

---
 polaris/make.llvm.sh  | 11 +++++++++--
 polaris/make.nvhpc.sh | 10 +++++++++-
 polaris/run.llvm.sh   |  4 ++--
 3 files changed, 20 insertions(+), 5 deletions(-)

diff --git a/polaris/make.llvm.sh b/polaris/make.llvm.sh
index 9ba0f88..6692083 100755
--- a/polaris/make.llvm.sh
+++ b/polaris/make.llvm.sh
@@ -3,7 +3,14 @@
 module restore
 module use /soft/modulefiles
 module load llvm
-module load mpiwrappers/cray-mpich-llvm
 module load cudatoolkit-standalone
+module load mpiwrappers/cray-mpich-llvm
+
+#dbg="-g"
+dbg=""
+
+opt="-O3"
+
+flags="${dbg} -fopenmp ${opt} -fopenmp-offload-mandatory -fopenmp-targets=nvptx64 --offload-arch=sm_80"
 
-mpicxx -cxx=clang++ -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_80 flops.cpp -o flops.llvm.exe
+mpicxx -cxx=clang++ ${flags} flops.cpp -o flops.llvm.exe
diff --git a/polaris/make.nvhpc.sh b/polaris/make.nvhpc.sh
index dd0e1dd..9710f62 100755
--- a/polaris/make.nvhpc.sh
+++ b/polaris/make.nvhpc.sh
@@ -2,4 +2,12 @@
 
 module restore
 
-CC -mp=gpu flops.cpp -o flops.nvhpc.exe
+#dbg="-g"
+dbg=""
+
+opt="-O3"
+
+#flags="${dbg} ${opt} -mp=gpu -gpu=cc80,cuda11.0"
+flags="${dbg} ${opt} -mp=gpu -gpu=cc80"
+
+CC ${flags} flops.cpp -o flops.nvhpc.exe
diff --git a/polaris/run.llvm.sh b/polaris/run.llvm.sh
index 3361e32..bec107a 100755
--- a/polaris/run.llvm.sh
+++ b/polaris/run.llvm.sh
@@ -7,8 +7,8 @@ module load mpiwrappers/cray-mpich-llvm
 module load cudatoolkit-standalone
 
 EXE=./flops.llvm.exe
-OUTER_ITER=10
-INNER_ITER=50
+OUTER_ITER=5
+INNER_ITER=200
 
 export OMP_NUM_THREADS=1
 
-- 
GitLab