Release Notes for QUDA v0.4.1 (PRE-RELEASE)                 XX April 2012
-----------------------------

Overview:

QUDA is a library for performing calculations in lattice QCD on
graphics processing units (GPUs) using NVIDIA's "C for CUDA" API.
This release includes optimized kernels for applying a variety of
Dirac operators (Wilson, clover-improved Wilson, twisted mass,
improved staggered, and domain wall), kernels for performing various
BLAS-like operations, and full inverters built on these kernels.
Mixed-precision implementations of both CG and BiCGstab are provided,
with support for double, single, and half (16-bit fixed-point)
precision.  The staggered implementation additionally includes support
for asqtad link fattening, force terms for the asqtad fermion action
and one-loop improved Symanzik gauge action, and a multi-shift CG
solver.  Use of multiple GPUs in parallel is supported for all actions
except domain wall.


Software Compatibility:

The library has been tested under Linux (CentOS 5.7 and Ubuntu 10.04)
using release 4.1 of the CUDA toolkit.  CUDA 3.x and earlier are no
longer supported.  The library also seems to work under Mac OS X
10.6.8 ("Snow Leopard") and 10.7.3 ("Lion") on recent 64-bit
Intel-based Macs.

See also "Known Issues" below.


Hardware Compatibility:

For a list of supported devices, see

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

Before building the library, you should determine the "compute
capability" of your card, either from NVIDIA's documentation or by
running the deviceQuery example in the CUDA SDK, and pass the
appropriate value to QUDA's configure script.  For example, the Tesla
C1060 is listed on the above website as having compute capability 1.3,
and so to configure the library for this card, you'd run "configure
--enable-gpu-arch=sm_13 [other options]" before typing "make".

As of QUDA 0.4.0, only devices of compute capability 1.1 or greater
are supported.


Installation:

Installing the library involves running "configure" followed by
"make".  See "./configure --help" for a list of configure options.
At a minimum, you'll probably want to set the GPU architecture; see
"Hardware Compatibility" above.

Enabling multi-GPU support requires passing the --enable-multi-gpu
flag to configure, as well as --with-mpi=<PATH> and optionally
--with-qmp=<PATH>.  If the latter is given, QUDA will use QMP for
communications; otherwise, MPI will be called directly.  By default,
it is assumed that the MPI compiler wrappers are <MPI_PATH>/bin/mpicc
and <MPI_PATH>/bin/mpicxx for C and C++, respectively.  These choices
may be overriden by setting the CC and CXX variables on the command
line as follows:

./configure --enable-multi-gpu --with-mpi=<MPI_PATH> \
[--with-qmp=<QMP_PATH>] [OTHER_OPTIONS] CC=my_mpicc CXX=my_mpicxx

Finally, with some MPI implementations, executables compiled against
MPI will not run without "mpirun".  This has the side effect of
causing the configure script to believe that the compiler is failing
to produce a valid executable.  To skip these checks, one can trick
configure into thinking that it's cross-compiling by setting the
--build=none and --host=<HOST> flags.  For the latter,
"--host=x86_64-linux-gnu" should work on a 64-bit linux system.

Throughout the library, auto-tuning is used to select optimal launch
parameters for most performance-critical kernels.  This tuning
process takes some time and will generally slow things down the first
time a given kernel is called during a run.  To avoid this one-time
overhead in subsequent runs (using the same action, solver, lattice
volume, etc.), the optimal parameters are cached to disk.  For this
to work, the QUDA_RESOURCE_PATH environment variable must be set,
pointing to a writeable directory.  Note that since the tuned parameters
are hardware-specific, this "resource directory" should not be shared
between jobs running on different systems (e.g., two clusters
with different GPUs installed).  Attempting to use parameters tuned
for one card on a different card may lead to unexpected errors.


Using the Library:

Include the header file include/quda.h in your application, link
against lib/libquda.a, and study tests/wilson_invert_test.c (or the
corresponding tests for staggered, domain wall, and twisted mass) for
an example of the interface.  The various inverter options are
enumerated in include/enum_quda.h.


Known Issues:

* For compatibility with CUDA, on 32-bit platforms the library is
  compiled with the GCC option -malign-double.  This differs from the
  GCC default and may affect the alignment of various structures,
  notably those of type QudaGaugeParam and QudaInvertParam, defined in
  quda.h.  Therefore, any code to be linked against QUDA should also
  be compiled with this option.

* With current drivers, Fermi-based GeForce cards suffer from
  occasional hangs when reading from double-precision textures,
  recoverable only with a soft reset.  Such texture reads are thus
  disabled by default at the expense of some performance.  To
  re-enable double-precision texture reads on Fermi-based Tesla and
  Quadro cards, pass the --enable-fermi-double-tex flag to configure.
  This flag has no effect when the library is build for GPU
  architectures other than Fermi (sm_20).

* The auto-tuner reports "0 Gflop/s" and "0 GB/s" for several of the
  Dslash kernels (visible if the verbosity is set to at least
  QUDA_SUMMARIZE), rather than the correct values.  This does not
  affect the tuning process or actual performance.

* At present, using MPI directly for communications (as opposed to
  QMP) requires calling some initialization routines that are not
  exposed in quda.h.  This will be corrected in the next release.


Getting Help:

Please visit http://lattice.github.com/quda for contact information.
Bug reports are especially welcome.


Acknowledging QUDA:

If you find this software useful in your work, please cite:

M. A. Clark, R. Babich, K. Barros, R. Brower, and C. Rebbi, "Solving
Lattice QCD systems of equations using mixed precision solvers on GPUs,"
Comput. Phys. Commun. 181, 1517 (2010) [arXiv:0911.3191 [hep-lat]].

When taking advantage of multi-GPU support, please also cite:

R. Babich, M. A. Clark, B. Joo, G. Shi, R. C. Brower, and S. Gottlieb,
"Scaling lattice QCD beyond 100 GPUs," International Conference for
High Performance Computing, Networking, Storage and Analysis (SC),
2011 [arXiv:1109.2935 [hep-lat]].

Several other papers that might be of interest are listed at
http://lattice.github.com/quda .


Authors:

Ronald Babich (NVIDIA)
Kipton Barros (Los Alamos National Laboratory)
Richard Brower (Boston University)
Michael Clark (NVIDIA)
Justin Foley (University of Utah)
Joel Giedt (Rensselaer Polytechnic Institute)
Steven Gottlieb (Indiana University)
Balint Joo (Jefferson Laboratory)
Claudio Rebbi (Boston University)
Guochun Shi (NCSA)
Alexei Strelchenko (Cyprus Institute)

Portions of this software were developed at the Innovative Systems Lab,
National Center for Supercomputing Applications
http://www.ncsa.uiuc.edu/AboutUs/Directorates/ISL.html

Development was supported in part by the U.S. Department of Energy
under grants DE-FC02-06ER41440, DE-FC02-06ER41449, and
DE-AC05-06OR23177, as well as by the National Science Foundation under
grants DGE-0221680, PHY-0427646, PHY-0835713, OCI-0946441, and
OCI-1060067.  Any opinions, findings, and conclusions or
recommendations expressed in this material are those of the authors
and do not necessarily reflect the views of the Department of Energy
or the National Science Foundation.