Skip to content

Commit 6405c76

Browse files
gilbertlee-amdgilbertlee-amd
authored
Fixing install script hip_compiler bug and improving logging on fallback (ROCm#156)
* Fixing install script hip_compiler bug and improving logging on fallback
1 parent 33cc4df commit 6405c76

File tree

1 file changed

+66
-61
lines changed

1 file changed

+66
-61
lines changed

‎install.sh‎

Lines changed: 66 additions & 61 deletions
Original file line numberDiff line numberDiff line change
@@ -37,49 +37,49 @@ gpu_targets=""
3737
# check if we have a modern version of getopt that can handle whitespace and long parameters
3838
getopt -T
3939
if [[ $? -eq 4 ]]; then
40-
GETOPT_PARSE=$(getopt --name "${0}" --longoptions help,mpi,test,rocm_home:,rccl_home:,mpi_home:,hip_compiler:,gpu_targets: --options hmt -- "$@")
40+
GETOPT_PARSE=$(getopt --name "${0}" --longoptions help,mpi,test,rocm_home:,rccl_home:,mpi_home:,hip_compiler:,gpu_targets: --options hmt -- "$@")
4141
else
42-
echo "Need a new version of getopt"
43-
exit 1
42+
echo "Need a new version of getopt"
43+
exit 1
4444
fi
4545

4646
if [[ $? -ne 0 ]]; then
47-
echo "getopt invocation failed; could not parse the command line";
48-
exit 1
47+
echo "getopt invocation failed; could not parse the command line";
48+
exit 1
4949
fi
5050

5151
eval set -- "${GETOPT_PARSE}"
5252

5353
while true; do
54-
case "${1}" in
55-
-h|--help)
56-
display_help
57-
exit 0 ;;
58-
-m|--mpi)
59-
mpi_enabled=true
60-
shift ;;
61-
-t|--test)
62-
run_tests=true
63-
shift ;;
64-
--rocm_home)
65-
rocm_dir=${2}
66-
shift 2 ;;
67-
--rccl_home)
68-
rccl_dir=${2}
69-
shift 2 ;;
70-
--mpi_home)
71-
mpi_dir=${2}
72-
shift 2 ;;
73-
--hip_compiler)
74-
hip_compiler=${2}
75-
shift 2 ;;
76-
--gpu_targets)
77-
gpu_targets=${2}
78-
shift 2 ;;
79-
--) shift ; break ;;
80-
*) echo "Unexpected command line parameter received; aborting";
81-
exit 1 ;;
82-
esac
54+
case "${1}" in
55+
-h|--help)
56+
display_help
57+
exit 0 ;;
58+
-m|--mpi)
59+
mpi_enabled=true
60+
shift ;;
61+
-t|--test)
62+
run_tests=true
63+
shift ;;
64+
--rocm_home)
65+
rocm_dir=${2}
66+
shift 2 ;;
67+
--rccl_home)
68+
rccl_dir=${2}
69+
shift 2 ;;
70+
--mpi_home)
71+
mpi_dir=${2}
72+
shift 2 ;;
73+
--hip_compiler)
74+
hip_compiler=${2}
75+
shift 2 ;;
76+
--gpu_targets)
77+
gpu_targets=${2}
78+
shift 2 ;;
79+
--) shift ; break ;;
80+
*) echo "Unexpected command line parameter received; aborting";
81+
exit 1 ;;
82+
esac
8383
done
8484

8585
# throw error code after running a command in the install script
@@ -101,49 +101,54 @@ build_dir=./build
101101
rm -rf ${build_dir}
102102

103103
if [[ -z ${rocm_dir} ]]; then
104-
echo "ROCM_PATH does not exist at ${rocm_dir}. Defaulting to /opt/rocm"
105-
rocm_dir=/opt/rocm
104+
echo "[WARN] ROCM_PATH does not exist at ${rocm_dir}. Defaulting to /opt/rocm"
105+
rocm_dir=/opt/rocm
106106
fi
107107

108108
if ! command -v ${hip_compiler} 2>&1 >/dev/null ; then
109-
echo "HIP Compiler does not exist at ${hip_compiler}. Please check the path."
110-
echo "Defaulting to /opt/rocm/bin/amdclang++"
111-
hip_compiler=${rocm_dir}/bin/amdclang++
109+
echo "[WARN] HIP Compiler does not exist at ${hip_compiler}. Please check the path."
110+
echo "[WARN] - Falling back to ${rocm_dir}/bin/amdclang++"
111+
hip_compiler=${rocm_dir}/bin/amdclang++
112+
113+
if ! command -v ${hip_compiler} 2>&1 >/dev/null ; then
114+
echo "[WARN] ${hip_compiler} does not exist. Please be advised."
115+
echo "[WARN] - Falling back to ${rocm_dir}/bin/hipcc"
116+
hip_compiler=${rocm_dir}/bin/hipcc
112117

113118
if ! command -v ${hip_compiler} 2>&1 >/dev/null ; then
114-
echo "${hip_compiler} does not exist. Please be advised."
115-
echo "Defaulting to /opt/rocm/bin/hipcc"
116-
hip_compiler=${rocm_dir}/bin/hipcc
117-
118-
if ! command -v ${hip_compiler} 2>&1 >/dev/null ; then
119-
echo "${hip_compiler} does not exist!. Please check your ROCm installation."
120-
echo "Cannot proceed with building rccl-tests!"
121-
exit 1
122-
fi
119+
echo "[ERROR] ${hip_compiler} does not exist!. Please check your ROCm installation." >&2
120+
echo "[ERROR] Cannot proceed with building rccl-tests!" >&2
121+
exit 1
123122
fi
123+
fi
124124
fi
125+
echo "[INFO] Compiling with ${hip_compiler}"
125126

126127
if [[ -n ${gpu_targets} ]]; then
127-
GPU_TARGETS="GPU_TARGETS=${gpu_targets}"
128+
GPU_TARGETS="GPU_TARGETS=${gpu_targets}"
128129
fi
129130

130131
if ($mpi_enabled); then
131-
if [[ ${mpi_dir} == "" ]]; then
132-
echo "MPI flag enabled but path to MPI installation not specified. See --mpi_home command line argument."
133-
exit 1
134-
else
135-
make NCCL_HOME=${rccl_dir} CUSTOM_RCCL_LIB=${rccl_dir}/lib/librccl.so MPI=1 MPI_HOME=${mpi_dir} HIPCC=${hip_compiler} ${GPU_TARGETS} -j$(nproc)
136-
fi
132+
if [[ ${mpi_dir} == "" ]]; then
133+
echo "[ERROR] MPI flag enabled but path to MPI installation not specified. See --mpi_home command line argument." >&2
134+
exit 1
135+
else
136+
echo "[INFO] Compiling with MPI support (Using MPI from ${mpi_dir})"
137+
echo
138+
make NCCL_HOME=${rccl_dir} CUSTOM_RCCL_LIB=${rccl_dir}/lib/librccl.so MPI=1 MPI_HOME=${mpi_dir} HIPCC=${hip_compiler} ${GPU_TARGETS} -j$(nproc)
139+
fi
137140
else
138-
make NCCL_HOME=${rccl_dir} CUSTOM_RCCL_LIB=${rccl_dir}/lib/librccl.so HIP_COMPILER=${hip_compiler} ${GPU_TARGETS} -j$(nproc)
141+
echo "[INFO] Compiling without MPI support (MPI support requires -m and --mpi_home)"
142+
echo
143+
make NCCL_HOME=${rccl_dir} CUSTOM_RCCL_LIB=${rccl_dir}/lib/librccl.so HIPCC=${hip_compiler} ${GPU_TARGETS} -j$(nproc)
139144
fi
140145
check_exit_code "$?"
141146

142147
# Optionally, run tests if they're enabled.
143148
if ($run_tests); then
144-
if ($mpi_enabled); then
145-
cd test; LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${rccl_dir}/lib:${mpi_dir}/lib PATH=$PATH:${mpi_dir}/bin python3 -m pytest
146-
else
147-
cd test; LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${rccl_dir}/lib python3 -m pytest -k "not MPI"
148-
fi
149+
if ($mpi_enabled); then
150+
cd test; LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${rccl_dir}/lib:${mpi_dir}/lib PATH=$PATH:${mpi_dir}/bin python3 -m pytest
151+
else
152+
cd test; LD_LIBRARY_PATH=$LD_LIBRARY_PATH:${rccl_dir}/lib python3 -m pytest -k "not MPI"
153+
fi
149154
fi

0 commit comments

Comments
 (0)