Add nightly testing for ROCm 5.4 (#23463)

The script is pretty similar to the CUDA test scripts on EX.

Tested in jenkins manually.

[Reviewed by @stonea]

runtime/src/gpu/amd/Makefile.share

@@ -23,6 +23,9 @@ GPU_OBJS = $(addprefix $(GPU_OBJDIR)/,$(addsuffix .o,$(basename $(GPU_SRCS))))
 
 RUNTIME_CXXFLAGS += -x hip --offload-arch=$(CHPL_MAKE_GPU_ARCH)
 
+# Some ROCm headers have `foo()`, suppress warnings/errors from them
+RUNTIME_CFLAGS += -Wno-strict-prototypes
+
 $(RUNTIME_OBJ_DIR)/gpu-amd-reduce.o: gpu-amd-reduce.cc \
                                          $(RUNTIME_OBJ_DIR_STAMP)
 	PATH=$(PATH):$(CHPL_MAKE_ROCM_PATH)/llvm/bin $(CXX) -c -std=c++14 $(RUNTIME_CXXFLAGS) $(RUNTIME_INCLS) -o $@ $<

util/cron/test-gpu-ex-rocm-54.bash

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+#
+# GPU native testing on a Cray EX (using none for CHPL_COMM)
+
+CWD=$(cd $(dirname ${BASH_SOURCE[0]}) ; pwd)
+source $CWD/common-native-gpu.bash
+source $CWD/common-hpe-cray-ex.bash
+
+module load rocm/5.4.3  # pin to rocm 5.4.3
+
+export CHPL_COMM=none
+export CHPL_LLVM=bundled
+unset CHLP_LLVM_CONFIG  # we need this to avoid warnings
+export CHPL_LOCALE_MODEL=gpu
+export CHPL_LAUNCHER_PARTITION=bardpeak  # bardpeak is the default queue
+
+export CHPL_GPU=amd  # also detected by default
+export CHPL_GPU_ARCH=gfx90a
+
+export CHPL_NIGHTLY_TEST_DIRS="gpu/native"
+
+export CHPL_NIGHTLY_TEST_CONFIG_NAME="gpu-ex-rocm-54"
+$CWD/nightly -cron ${nightly_args}