blob: e29f0e4d15fa3ba2f06173913f228b6686c1c302 [file] [log] [blame]
gemmlowp: a small self-contained low-precision GEMM library
This is not a full linear algebra library, only a GEMM library: it only does
general matrix multiplication ("GEMM").
The meaning of "low precision" is detailed in this document:
Some of the general design is explained in
This is not an official Google product (experimental or otherwise), it is just
code that happens to be owned by Google.
Mailing list
gemmlowp-related discussion, about either development or usage, is welcome
on this Google Group (mailing list / forum):!forum/gemmlowp
Portability, target platforms/architectures
Should be portable to any platform with some C++11 and POSIX support,
while we have optional optimized code paths for specific architectures.
C++11 (a small conservative subset of it)
Required for some features:
* Some POSIX interfaces:
* pthreads (for multi-threaded operation and for profiling).
* sysconf (for multi-threaded operation to detect number of cores;
may be bypassed).
Architecture-specific code paths use intrinsics or inline assembly.
See "Architecture-specific optimized code paths" below.
Architecture-specific optimized code paths
We have some optimized code paths for specific instruction sets.
Some are written in inline assembly, some are written in C++ using
intrinsics. Both GCC and Clang are supported.
At the moment, we have a full set of optimized code paths (kernels,
packing and unpacking paths) only for ARM NEON, supporting both
ARMv7 (32bit) and ARMv8 (64bit).
We also have a partial set of optimized code paths (only kernels
at the moment) for Intel SSE. It supports both x86 and x86-64 but
only targets SSE4. The lack of packing/unpacking code paths means
that performance isn't optimal yet.
Details of what it takes to make an efficient port of gemmlowp, namely
writing a suitable GEMM kernel and accompanying packing code, are
explained in this file:
Public interfaces
1. gemmlowp public interface
gemmlowp's main public interface is in the public/ subdirectory. The
header to include is
This is a headers-only library, so there is nothing to link to.
2. EightBitIntGemm standard interface
Additionally, the eight_bit_int_gemm/ subdirectory provides an
implementation of the standard EightBitIntGemm interface. The header
to include is
This is *NOT* a headers-only library, users need to link to
The API is similar to the standard BLAS GEMM interface, and implements
C = A * B. If the transpose flags for a matrix argument are false, its memory
order is treated as column major, and row major if its true.
Building by manually invoking your compiler
Because gemmlowp is so simple, working with it involves only
single-command-line compiler invokations. Therefore we expect that
most people working with gemmlowp will either manually invoke their
compiler, or write their own rules for their own preferred build
Keep in mind (previous section) that gemmlowp itself is a pure-headers-only
library so there is nothing to build, and the eight_bit_int_gemm library
consists of a single file to build.
For a Android gemmlowp development workflow, the scripts/ directory
contains a script to build and run a program on an Android device:
Building using Bazel
That being said, we also maintain a Bazel BUILD system as part of
gemmlowp. Its usage is not mandatory at all and is only one
possible way that gemmlowp libraries and tests may be built. If
you are interested, Bazel's home page is
And you can get started with using Bazel to build gemmlowp targets
by first creating an empty WORKSPACE file in a parent directory,
for instance:
$ cd gemmlowp/.. # change to parent directory containing gemmlowp/
$ touch WORKSPACE # declare that to be our workspace root
$ bazel build gemmlowp:all
Testing by manually building and running tests
The test/ directory contains unit tests. The primary unit test is
Since it covers also the EightBitIntGemm interface, it needs to be
linked against
It also uses realistic data captured from a neural network run in
Thus you'll want to pass the following list of source files to your
The scripts/ directory contains a script to build and run a program
on an Android device:
It expects the CXX environment variable to point to an Android toolchain's
C++ compiler, and expects source files (and optionally, cflags) as
command-line parameters. To build and run the above-mentioned main unit test,
first set CXX e.g.:
$ export CXX=/some/toolchains/arm-linux-androideabi-4.8/bin/arm-linux-androideabi-g++
Then run:
$ ./scripts/ \
test/ \
eight_bit_int_gemm/ \
Testing using Bazel
Alternatively, you can use Bazel to build and run tests. See the Bazel
instruction in the above section on building. Once your Bazel workspace
is set up, you can for instance do:
$ bazel test gemmlowp:all
Troubleshooting Compilation
If you're having trouble finding the compiler, follow these instructions to
build a standalone toolchain:
Here's an example of setting up Clang 3.5:
$ export INSTALL_DIR=~/toolchains/clang-21-stl-gnu
$ $NDK/build/tools/ \
--toolchain=arm-linux-androideabi-clang3.5 --platform=android-21 \
$ export CXX="$INSTALL_DIR/bin/arm-linux-androideabi-g++ \
Some compilers (e.g. the default clang++ in the same bin directory) don't
support NEON assembly. The benchmark build process will issue a warning if
support isn't detected, and you should make sure you're using a compiler like
arm-linux-androideabi-g++ that does include NEON.
The main benchmark is
It doesn't need to be linked to any
other source file. We recommend building with assertions disabled (-DNDEBUG).
For example, the benchmark can be built and run on an Android device by doing:
$ ./scripts/ test/ -DNDEBUG
If GEMMLOWP_TEST_PROFILE is defined then the benchmark will be built with
profiling instrumentation (which makes it slower) and will dump profiles.
See next section on profiling.
The profiling/ subdirectory offers a very simple non-interrupting sampling
profiler that only requires pthreads (no signals).
It relies on source code being instrumented with pseudo-stack labels.
See profiling/instrumentation.h.
A full example of using this profiler is given in profiling/profiler.h.
Contribution-related discussion is always welcome on the gemmlowp
mailing list (see above).
We try to keep a current list of TODO items in the todo/ directory.
Prospective contributors are welcome to pick one to work on, and
communicate about it on the gemmlowp mailing list.
Details of the contributing process, including legalese, are in CONTRIBUTING.
Performance goals
Our performance goals differ from typical GEMM performance goals in the
following ways:
1. We care not only about speed, but also about minimizing power usage.
We specifically care about charge usage in mobile/embedded devices.
This implies that we care doubly about minimizing memory bandwidth usage:
we care about it, like any GEMM, because of the impact on speed, and we
also care about it because it is a key factor of power usage.
2. Most GEMMs are optimized primarily for large dense matrix sizes (>= 1000).
We do care about large sizes, but we also care specifically about the
typically smaller matrix sizes encountered in various mobile applications.
This means that we have to optimize for all sizes, not just for large enough