Compiling Magma for arm architecture

[Ashar Alam]

Hi, 

Is it possible to cross-compile and use MAGMA for ARM architecture and embedded processors; or is only x64 architecture supported?

Thanks

Best,

Ashar

[Stanimire Tomov]

Hi Ashar,

Yes, it is possible to cross compile and use MAGMA on ARM.

We have done it successfully on NVIDIA’s Tegra devices.

I assume you would still want to use a GPU where the host is powered by ARM.

Best regards,

Stan

--
You received this message because you are subscribed to the Google Groups "MAGMA User" group.
To unsubscribe from this group and stop receiving emails from it, send an email to magma-user+...@icl.utk.edu.
To view this discussion on the web visit https://groups.google.com/a/icl.utk.edu/d/msgid/magma-user/a3742636-1f9d-4ceb-a026-d2ccdb1295f0n%40icl.utk.edu.

[Ashar Alam]

Hi Stan,

Thank you so much for your reply. I think I want to use it for AGX Xavier. Is there a guideline or document on the compiling procedure?

Thanks.

Best,

Ashar

[Stanimire Tomov]

Ashar,

You may have to look at the AGX Xavier development kit documentation how to compile for it,

how to get compilers, etc.

After that compiling magma is as on any other system - see the instructions in README.

There are a number of make.inc examples in make.inc-examples but not for arm.

We have to add one - or if you manage to compile it and are willing to contribute it,

that would be great.

Last time I compiled for an arm machine I used a make.inc like this:

more make.inc

#//////////////////////////////////////////////////////////////////////////////

#   -- MAGMA (version 2.0) --

#      Univ. of Tennessee, Knoxville

#      Univ. of California, Berkeley

#      Univ. of Colorado, Denver

#      @date

#//////////////////////////////////////////////////////////////////////////////

# GPU_TARGET contains one or more of Fermi, Kepler, Maxwell, Pascal, Volta

# to specify for which GPUs you want to compile MAGMA:

#     Fermi   - NVIDIA compute capability 2.x cards

#     Kepler  - NVIDIA compute capability 3.x cards

#     Maxwell - NVIDIA compute capability 5.x cards

#     Pascal  - NVIDIA compute capability 6.x cards

#     Volta   - NVIDIA compute capability 7.x cards

# The default is "Kepler Maxwell Pascal".

# Note that NVIDIA no longer supports 1.x cards, as of CUDA 6.5.

# See http://developer.nvidia.com/cuda-gpus

#

#GPU_TARGET ?= Kepler Maxwell Pascal

GPU_TARGET = Volta

# --------------------

# programs

CC        = armclang++

CXX       = armclang++

NVCC      = nvcc

FORT      = armflang

ARCH      = ar

ARCHFLAGS = cr

RANLIB    = ranlib

# --------------------

# flags

# Use -fPIC to make shared (.so) and static (.a) library;

# can be commented out if making only static library.

FPIC      = -fPIC

CFLAGS    = -O3 $(FPIC) -DNDEBUG -DADD_ -Wall -fopenmp

FFLAGS    = -O3 $(FPIC) -DNDEBUG -DADD_ -Wall -Wno-unused-dummy-argument

F90FLAGS  = -O3 $(FPIC) -DNDEBUG -DADD_ -Wall -Wno-unused-dummy-argument -x f95-cpp-input

NVCCFLAGS = -O3         -DNDEBUG -DADD_       -Xcompiler "$(FPIC)" -std=c++11

LDFLAGS   =     $(FPIC)                       -fopenmp

# C++11 (gcc >= 4.7) is not required, but has benefits like atomic operations

CXXFLAGS := $(CFLAGS) -std=c++11

CFLAGS   += -std=c99

# --------------------

# libraries

BLASmp = /sw/wombat/ARM_Compiler/19.3/armpl-19.3.0_ThunderX2CN99_RHEL-7_arm-hpc-compiler_19.3_aarch64-linux/lib/libarmpl_lp64_mp.a

LAPACKmp = /sw/wombat/ARM_Compiler/19.3/armpl-19.3.0_ThunderX2CN99_RHEL-7_arm-hpc-compiler_19.3_aarch64-linux/lib/libarmpl_lp64_mp.a

# gcc with OpenBLAS (includes LAPACK)

LIB       = $(BLASmp) $(LAPACKmp) 

LIB      += -lcublas -lcusparse -lcudart -lcudadevrt -lflang-omp -lomp

# --------------------

# directories

# define library directories preferably in your environment, or here.

#OPENBLASDIR ?= /usr/local/openblas

CUDADIR ?= $(CUDA_DIR)

#-include make.check-openblas

#-include make.check-cuda

LIBDIR    = -L$(CUDADIR)/lib64

INC       = -I$(CUDADIR)/include

Note that you would need BLAS and LAPACK for arm. Usually this will be from openblas - you have

to get them for arm or compile them for arm. In the above example the vendor had them.

Hope this helps. Tell us how it goes and if there are problems (in which case we can try to find

a system like this and see how exactly to compile there).

Thanks,

Stan