Re: Build for CUDA support

From: Mahmood Naderan (mahmood.nt_at_gmail.com)
Date: Sat Mar 03 2018 - 10:35:06 CST

So, I ran with the cuda switch. However, It seems that the cuda home is not
set correctly. Assume that the NAMD_ARCH is Linux_x86_64 in my case. Mine
is cuda-9.0 but I see cuda-6.5 in the arch file. See below

mahmood_at_orca:NAMD_2.12_Source$ ./config Linux-x86_64-g++ --charm-arch
multicore-linux64 --with-cuda --cuda-prefix /usr/local/cuda-9.0

Selected arch file arch/Linux-x86_64-g++.arch contains:

NAMD_ARCH = Linux-x86_64
CHARMARCH = multicore-linux64

CXX = g++ -m64 -std=c++0x -O3
CXXOPTS = -fexpensive-optimizations -ffast-math
CC = gcc -m64 -O3
COPTS = -fexpensive-optimizations -ffast-math

Creating directory: Linux-x86_64-g++
Creating link: .. to .rootdir
Using FFTW build found in main build directory
Writing build options to Linux-x86_64-g++/Make.config
Using Charm++ 6.7.1 build found in main build directory
Using Tcl build found in main build directory
glob: No match.
glob: No match.
Linking Makefile
Linking Make.depends
Linking src directory
Linking plugins directory
Linking psfgen directory

Generated Linux-x86_64-g++/Make.config contains the following:

CHARMBASE = .rootdir/charm-6.7.1
include .rootdir/arch/Linux-x86_64-g++.arch
CHARMARCH = multicore-linux64
CHARM = $(CHARMBASE)/$(CHARMARCH)
NAMD_PLATFORM = $(NAMD_ARCH)-multicore-CUDA
include .rootdir/arch/$(NAMD_ARCH).base
include .rootdir/arch/$(NAMD_ARCH).tcl
include .rootdir/arch/$(NAMD_ARCH).fftw
include .rootdir/arch/$(NAMD_ARCH).cuda
TCLDIR = .rootdir/tcl-threaded
FFTDIR = .rootdir/fftw
CUDADIR = /usr/local/cuda-9.0
CUDASODIR = /usr/local/cuda-9.0/lib64
LIBCUDARTSO = libcudart.so.9.0
LIBCUFFTSO = libcufft.so.9.0

You are ready to run make in directory Linux-x86_64-g++ now.
mahmood_at_orca:NAMD_2.12_Source$ grep CUDADIR
Linux-x86_64-g++/.rootdir/arch/Linux-x86_64.cuda

CUDADIR=/usr/local/encap/cuda-6.5
CUDAINCL=-I$(CUDADIR)/include
CUDALIB=-L$(CUDADIR)/lib64 -lcufft_static -lculibos -lcudart_static -lrt
CUDASODIR=$(CUDADIR)/lib64
CUDACC=$(CUDADIR)/bin/nvcc -O3 --maxrregcount 48 $(CUDAGENCODE) -Xcompiler
"-m64" $(CUDA)
mahmood_at_orca:NAMD_2.12_Source$

Regards,
Mahmood

On Sat, Mar 3, 2018 at 7:40 PM, Giacomo Fiorin <giacomo.fiorin_at_gmail.com>
wrote:

> Run the NAMD ./config script without argument to see the syntax for CUDA
> builds; check notes.txt as well.
>
>

This archive was generated by hypermail 2.1.6 : Sun Sep 15 2019 - 23:19:28 CDT