From: Frank Duan (f_duan_at_yahoo.com)
Date: Wed Jun 16 2021 - 09:06:18 CDT
Hi All,
I tried to build CUDA-11 enabled NAMD 2.14 on HPE Cray EX (formerly known as Cray Shasta) system which has NVIDIA V100 GPU. I used gcc 9.03 with following Make.config:
----------Make.config----------CHARMBASE = .rootdir/charm-6.10.2include .rootdir/arch/Linux-x86_64-g++.archCHARMARCH = multicore-linux-x86_64-gccCHARM = $(CHARMBASE)/$(CHARMARCH)NAMD_PLATFORM = $(NAMD_ARCH)-multicore-CUDAinclude .rootdir/arch/$(NAMD_ARCH).baseinclude .rootdir/arch/$(NAMD_ARCH).tclinclude .rootdir/arch/$(NAMD_ARCH).fftwTCLDIR = .rootdir/tcl-threadedFFTDIR = .rootdir/fftwinclude .rootdir/arch/$(NAMD_ARCH).cudainclude .rootdir/arch/$(NAMD_ARCH).cuda11CUDADIR = /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/cuda/11.0CUDASODIR = /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/cuda/11.0/lib64LIBCUDARTSO =LIBCUFFTSO =CUDAGENCODE = -gencode arch=compute_70,code=sm_70CUDADLINKOPTS = -gencode arch=compute_70,code=sm_70CXXOPTS = -I/opt/nvidia/hpc_sdk/Linux_x86_64/21.3/math_libs/11.0/targets/x86_64-linux/include/ -L/opt/nvidia/hpc_sdk/Linux_x86_64/21.3/math_libs/11.0/targets/x86_64-linux/lib-----------------------------------------------------------------------
When the compiler linked namd2, it gave following errors for libcufft_static.a:
-----------Errors----------------/usr/bin/ld: /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/math_libs/11.0/targets/x86_64-linux/lib/libcufft_static.a(cbdouble_32bit_prime_callback_RT_SM35_plus.o): in function `__sti____cudaRegisterAll()':cbdouble_32bit_prime_callback_RT_SM35_plus.compute_80.cudafe1.cpp:(.text.startup+0x1d): undefined reference to `__cudaRegisterLinkedBinary_61_cbdouble_32bit_prime_callback_RT_SM35_plus_compute_80_cpp1_ii_dc5d5345'/usr/bin/ld: /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/math_libs/11.0/targets/x86_64-linux/lib/libcufft_static.a(cbdouble_32bit_regular_bluestein_callback_RT_SM35_plus.o): in function `__sti____cudaRegisterAll()':cbdouble_32bit_regular_bluestein_callback_RT_SM35_plus.compute_80.cudafe1.cpp:(.text.startup+0x1d): undefined reference to `__cudaRegisterLinkedBinary_73_cbdouble_32bit_regular_bluestein_callback_RT_SM35_plus_compute_80_cpp1_ii_6a12427e'/usr/bin/ld: /opt/nvidia/hpc_sdk/Linux_x86_64/21.3/math_libs/11.0/targets/x86_64-linux/lib/libcufft_static.a(cbdouble_32bit_regular_callback_RT_SM35_plus.o): in function `__sti____cudaRegisterAll()':........................-------------------------
Please suggest and advise!
Thanks,
Frank
This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:11 CST