DLNetBench/Makefile.ALPS at main · HicrestLaboratory/DLNetBench · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
# -------------------------
# Makefile.ALPS
# Alps (CSCS) — nvcc, Cray MPI
# -------------------------

# IMPORTANT: UENV
# uenv start --view=modules prgenv-gnu-openmpi/25.12:v1
# ml gcc/14.3.0 cuda/12.9.1 nccl/2.28.9-1 openmpi/5.0.9 python/3.14.0 libfabric/2.3.1 aws-ofi-nccl/1.17.1

CXX := nvcc
CXXFLAGS := -arch=sm_90

# -- System paths (CUDA auto-detected from nvcc; NCCL_HOME from environment) --
# Override if auto-detection fails:
CUDA_HOME := /user-environment/linux-neoverse_v2/cuda-12.9.1-gbxd4g7cdadbxiev6ynbxgjlt5cs5h4s
NCCL_HOME := /user-environment/linux-neoverse_v2/nccl-2.28.9-1-c6miglsp7xn7ls6eohm5xfd7aiueydnl
MPI_HOME := /user-environment/linux-neoverse_v2/openmpi-5.0.9-emqmwdrzfeh265pelxc3h3leqozwhsn5

# -------------------------------------------------------
# Config: nccl
#   CUDA buffers + NCCL collectives
# -------------------------------------------------------
CONFIGS += nccl
BACKEND_nccl := nccl
EXTRA_CXXFLAGS_nccl := -arch=sm_90

# -------------------------------------------------------
# Config: mpi_gpu_cuda
#   CUDA buffers, MPI collectives only (no NCCL)
# -------------------------------------------------------
CONFIGS += mpi_gpu_cuda
BACKEND_mpi_gpu_cuda := mpi_gpu_cuda
EXTRA_CXXFLAGS_mpi_gpu_cuda := -arch=sm_90


include Makefile.flags.mk
include Makefile.common