Skip to content

Commit 933eab0

Browse files
committed
build updates
1 parent f662bb5 commit 933eab0

File tree

4 files changed

+66
-5
lines changed

4 files changed

+66
-5
lines changed

Cxx11/Makefile

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -283,7 +283,7 @@ endif
283283
$(NVCC) $(CUDAFLAGS) $(CPPFLAGS) -DPRK_USE_CUBLAS $< -lcublas -lcublasLt -o $@
284284

285285
%-nccl: %-nccl.cu prk_util.h prk_cuda.h prk_nccl.h transpose-kernel.h
286-
$(NVCC) $(CUDAFLAGS) $(CPPFLAGS) $(MPIINC) $< $(MPILIB) -lnccl -o $@
286+
$(NVCC) $(CUDAFLAGS) $(CPPFLAGS) $(MPIINC) $(NCCLINC) $< $(MPILIB) $(NCCLLIB) -o $@
287287

288288
%-nvshmem: %-nvshmem.cu prk_util.h prk_cuda.h prk_nvshmem.h transpose-kernel.h
289289
$(NVCC) $(NVSHMEMFLAGS) $(CUDAFLAGS) $(CPPFLAGS) $< -o $@

common/make.defs.cuda

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -196,6 +196,24 @@ CUDAFLAGS+=-D_X86INTRIN_H_INCLUDED
196196
#CUDAFLAGS+=-D_AVX512IFMAINTRIN_H_INCLUDED
197197
#CUDAFLAGS+=-D_AVX512IFMAVLINTRIN_H_INCLUDED
198198
#CUDAFLAGS+=-D_AVX512ERINTRIN_H_INCLUDED
199+
CUDAFLAGS += -I/usr/local/cuda-12.6/targets/x86_64-linux/include/cub/detail
200+
#
201+
# NCCL
202+
#
203+
NCCLDIR=../deps/nccl
204+
NCCLINC=-I${NCCLDIR}/include
205+
NCCLLIB=-L${NCCLDIR}/lib -lnccl
206+
#
207+
# NVSHMEM (Apt packages not reliable...)
208+
#
209+
NVSHMEMFLAGS=-rdc=true -diag-suppress 3012,3013
210+
#NVSHMEMFLAGS+=-I/usr/include/nvshmem_12
211+
NVSHMEM_DIR=${HOME}/NVSHMEM/nvshmem/build/src
212+
#NVSHMEM_DIR=/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem
213+
NVSHMEMFLAGS+=-I${NVSHMEM_DIR}/include
214+
NVSHMEMFLAGS+=-L${NVSHMEM_DIR}/lib
215+
NVSHMEMFLAGS+=-Wl,-rpath=${NVSHMEM_DIR}/lib
216+
NVSHMEMFLAGS+=-lnvshmem_device -lnvshmem_host
199217
#
200218
# CUDASTF
201219
#
@@ -221,6 +239,13 @@ MPILIB+=-lopen-pal -lopen-rte
221239
#MPIINC=-I/usr/include/mpich-3.2-$$(uname -m)
222240
#MPILIB=-L/usr/lib64/mpich-3.2/lib -lmpi
223241
#
242+
# OpenSHMEM
243+
#
244+
OSHDIR=/opt/mpich/oshmpi2-mpich
245+
OSHCC=${OSHDIR}/bin/oshcc
246+
OSHCXX=${OSHDIR}/bin/oshc++ -std=c++20
247+
OSHINC=-I${OSHDIR}/include
248+
#
224249
# Global Arrays
225250
#
226251
GADIR=../deps/ga
@@ -231,6 +256,7 @@ GAFLAG+=-L${GADIR}/../armci-mpi/lib -larmci # ARMCI-MPI
231256
GAFLAG+=${MPIINC} ${MPILIB}
232257
GAFLAG+=-lmpifort -lmpi
233258
GAFLAG+=-i8 # GA is compiled with -i8 on 64-bit systems
259+
#GAFLAG+=-fdefault-integer-8 # GA is compiled with 64b integers on 64-bit systems
234260
#
235261
# PETSc
236262
#

common/make.defs.gcc

Lines changed: 20 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -177,16 +177,32 @@ CUDAFLAGS+=-arch=native
177177
CUDAFLAGS+=--forward-unknown-to-host-compiler # MPI rpath
178178
# https://github.com/tensorflow/tensorflow/issues/1066#issuecomment-200574233
179179
#CUDAFLAGS+=-D_MWAITXINTRIN_H_INCLUDED
180+
CUDAFLAGS += -I/usr/local/cuda-12.6/targets/x86_64-linux/include/cub/detail
181+
#
182+
# NCCL
183+
#
184+
NCCLDIR=../deps/nccl
185+
NCCLINC=-I${NCCLDIR}/include
186+
NCCLLIB=-L${NCCLDIR}/lib -lnccl
180187
#
181188
# NVSHMEM (Apt packages not reliable...)
182189
#
190+
NVSHMEM_DIR=/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem
183191
NVSHMEMFLAGS=-rdc=true -diag-suppress 3012,3013
184192
#NVSHMEMFLAGS+=-I/usr/include/nvshmem_12
185-
NVSHMEMFLAGS+=-I/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem/include
186-
NVSHMEMFLAGS+=-L/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem/lib
187-
NVSHMEMFLAGS+=-Wl,-rpath=/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem/lib
193+
NVSHMEMFLAGS+=-I${NVSHMEM_DIR}/include
194+
NVSHMEMFLAGS+=-L${NVSHMEM_DIR}/lib
195+
NVSHMEMFLAGS+=-Wl,-rpath=${NVSHMEM_DIR}/lib
188196
NVSHMEMFLAGS+=-lnvshmem_device -lnvshmem_host
189197
#
198+
# CUDASTF
199+
#
200+
CUDASTF_DIR=../deps/stf/
201+
CUDASTF_CFLAGS=--expt-relaxed-constexpr
202+
CUDASTF_CFLAGS+=-I${CUDASTF_DIR}/cccl/cudax/include/
203+
CUDASTF_CFLAGS+=-I${CUDASTF_DIR}/cccl/libcudacxx/include
204+
CUDASTF_CFLAGS+=-lcuda
205+
#
190206
# Halide
191207
#
192208
HALIDECXX=c++
@@ -212,7 +228,7 @@ MPIINC=-I${MPIDIR}/include
212228
MPILIB=-L${MPIDIR}/lib -Wl,-rpath -Wl,${MPIDIR}/lib -lmpi_usempif08 -lmpi
213229
#MPILIB=-L${MPIDIR}/lib -lmpifort -lmpi
214230
#MPILIB=-L/usr/local/opt/libevent/lib -L${MPIDIR}/lib -lmpi
215-
#MPIINC=-I/usr/include/mpich-3.2-x86_64
231+
#MPIINC=-I/usr/include/mpich-3.2-$$(uname -m)
216232
#MPILIB=-L/usr/lib64/mpich-3.2/lib -lmpi
217233
#
218234
# OpenSHMEM

common/make.defs.nvhpc

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -112,6 +112,25 @@ CUDAFLAGS+=-D_X86INTRIN_H_INCLUDED
112112
#CUDAFLAGS+=-D_AVX512IFMAINTRIN_H_INCLUDED
113113
#CUDAFLAGS+=-D_AVX512IFMAVLINTRIN_H_INCLUDED
114114
#CUDAFLAGS+=-D_AVX512ERINTRIN_H_INCLUDED
115+
CUDAFLAGS += -I/usr/local/cuda-12.6/targets/x86_64-linux/include/cub/detail
116+
#
117+
#
118+
# NCCL
119+
#
120+
NCCLDIR=../deps/nccl
121+
NCCLINC=-I${NCCLDIR}/include
122+
NCCLLIB=-L${NCCLDIR}/lib -lnccl
123+
#
124+
# NVSHMEM (Apt packages not reliable...)
125+
#
126+
NVSHMEM_DIR=${HOME}/NVSHMEM/nvshmem/build/src
127+
NVSHMEMFLAGS=-rdc=true -diag-suppress 3012,3013
128+
#NVSHMEMFLAGS+=-I/usr/include/nvshmem_12
129+
#NVSHMEM_DIR=/opt/nvidia/hpc_sdk/Linux_x86_64/24.11/comm_libs/12.6/nvshmem
130+
NVSHMEMFLAGS+=-I${NVSHMEM_DIR}/include
131+
NVSHMEMFLAGS+=-L${NVSHMEM_DIR}/lib
132+
NVSHMEMFLAGS+=-Wl,-rpath=${NVSHMEM_DIR}/lib
133+
NVSHMEMFLAGS+=-lnvshmem_device -lnvshmem_host
115134
#
116135
# CUDASTF
117136
#

0 commit comments

Comments
 (0)