Legion is a parallel programming model
for distributed, heterogeneous machines.
The Legion team uses this repository for active development, so please make sure you're
using the right branch for your needs:
- stable - This is the default branch if you clone the
repository. It is generally about a month behind the master branch, allowing us to get some
mileage on larger changes before foisting them on everybody. Most users of Legion should use this
branch, although you should be prepared to try the master branch if you run into issues.
Updates are moved to the stable branch roughly monthly, although important bug fixes may be
applied directly when needed. Each batch of updates is given a "version" number, and
CHANGES.txt lists the
- master - This is the "mainline" used by the Legion team,
and contains changes and bug fixes that may not have made it into the stable branch yet. If you
are a user of "bleeding-edge" Legion functionality, you will probably need to be using this branch.
- lots of other feature branches - These exist as necessary for larger changes, and users will
generally want to steer clear of them. :)
Legion is a programming model and runtime system designed for decoupling the specification
of parallel algorithms from their mapping onto distributed heterogeneous architectures. Since
running on the target class of machines requires distributing not just computation but data
as well, Legion presents the abstraction of logical regions for describing the structure of
program data in a machine independent way. Programmers specify the partitioning of logical
regions into subregions, which provides a mechanism for communicating both the independence
and locality of program data to the programming system. Since the programming system
has knowledge of both the structure of tasks and data within the program, it can aid the
programmer in host of problems that are commonly the burden of the programmer:
- Discovering/verifying correctness of parallel execution: determining when two tasks
can be run in parallel without a data race is often difficult. Legion provides mechanisms
for creating both implicit and explicit parallel task launches. For implicit constructs
Legion will automatically discover parallelism. For explicit constructs, Legion will
notify the programmer if there are potential data races between tasks intended to be
run in parallel.
- Managing communication: when Legion determines that there are data dependencies between
two tasks run in different locations, Legion will automatically insert the necessary
copies and apply the necessary constraints so the second task will not run until
its data is available. We describe how tasks and data are placed in the next paragraph
on mapping Legion programs.
The Legion programming model is designed to abstract computations in a way that makes
them portable across many different potential architectures. The challenge then is to make
it easy to map the abstracted computation of the program onto actual architectures. At
a high level, mapping a Legion program entails making two kinds of decisions:
- For each task: select a processor on which to run the task.
- For each logical region a task needs: select a memory in which to create
a physical instance of the logical region for the task to use.
To facilitate this process Legion introduces a novel runtime 'mapping' interface. One of the
NON-goals of the Legion project was to design a programming system that was magically capable
of making intelligent mapping decisions. Instead the mapping interface provides a declarative
mechanism for the programmer to communicate mapping decisions to the runtime system
without having to actually write any code to perform the mapping (e.g. actually writing
the code to perform a copy or synchronization). Furthermore, by making the mapping interface
dynamic, it allows the programmer to make mapping decisions based on information that
may only be available at runtime. This includes decisions based on:
- Program data: some computations are dependent on data (e.g. is our irregular graph
sparse or dense in the number of edges).
- System data: which processors or nodes are currently up or down, or which are running
fast or slow to conserve power.
- Execution data: profiling data that is fed back to the mapper about how a certain
mapping performed previously. Alternatively which processors are currently over-
or under- loaded.
All of this information is made available to the mapper via various mapper calls, some
of which query the mapping interface while others simply are communicating information
to the mapper.
One very important property of the mapping interface is that no mapping decisions are
capable of impacting the correctness of the program. Consequently, all mapping decisions
made are only performance decisions. Programmers can then easily tune a Legion application
by modifying the mapping interface implementation without needing to be concerned
with how their decisions impact correctness. Ultimately, this makes it possible in Legion
to explore whole spaces of mapping choices (which tasks run on CPUs or GPUs, or where data
gets placed in the memory hierarchy) simply by enumerating all the possible mapping
decisions and trying them.
To make it easy to get a working program, Legion provides a default mapper implementation
that uses heuristics to make mapping decisions. In general these decision are good, but
they are certain to be sub-optimal across all applications and architectures. All calls
in the mapping interface are C++ virtual functions that can be overridden, so programmers
can extend the default mapper and only override the mapping calls that are impacting performance.
Alternatively a program can implement the mapping interface entirely from scratch.
For more details on the Legion programming model and its current implementation
we refer to you to our Supercomputing paper.
This repository includes the following contents:
tutorial: Source code for the tutorials.
examples: Larger examples for advanced programming techniques.
apps: Several complete Legion applications.
language: The Regent programming language compiler and examples.
runtime: The core runtime components:
legion: The Legion runtime itself (see
realm: The Realm low-level runtime (see
mappers: Several mappers, including the default mapper (see
tools: Miscellaneous tools:
To get started with Legion, you'll need:
- Linux, macOS, or another Unix
- A C++ 98 (or newer) compiler (GCC, Clang, Intel, or PGI) and GNU Make
- Optional: Python 2.7 (used for profiling/debugging tools)
- Optional: CUDA 5.0 or newer (for NVIDIA GPUs)
- Optional: GASNet (for networking, see
- Optional: LLVM 3.5 (for dynamic code generation)
- Optional: HDF5 (for file I/O)
Legion is currently compiled with each application. To try a Legion
application, just call
make in the directory in question. The
LG_RT_DIR variable is used to locate the Legion
directory. For example:
git clone https://github.com/StanfordLegion/legion.git export LG_RT_DIR="$PWD/legion/runtime" cd legion/examples/full_circuit make ./ckt_sim
The Legion Makefile includes several variables which influence the
build. These may either be set in the environment (e.g.
make) or at the top of each application's Makefile.
DEBUG=<0,1>: controls optimization level and enables various
dynamic checks which are too expensive for release builds.
OUTPUT_LEVEL=<level_name>: controls the compile-time logging
USE_CUDA=<0,1>: enables CUDA support.
USE_GASNET=<0,1>: enables GASNet support (see installation instructions).
USE_LLVM=<0,1>: enables LLVM support.
USE_HDF=<0,1>: enables HDF5 support.
In addition to Makefile variables, compilation is influenced by a
number of build flags. These flags may be added to the environment
CC_FLAGS (or again set inside the Makefile).
CC_FLAGS=-DLEGION_SPY: enables Legion Spy.
CC_FLAGS=-DPRIVILEGE_CHECKS: enables extra privilege checks.
CC_FLAGS=-DBOUNDS_CHECKS: enables dynamic bounds checks.
Legion and Realm accept command-line arguments for various runtime
parameters. Below are some of the more commonly used flags:
sets logging level for
directs logging output to
-ll:cpu <int>: CPU processors to create per process
-ll:gpu <int>: GPU processors to create per process
-ll:cpu <int>: utility processors to create per process
-ll:csize <int>: size of CPU DRAM memory per process (in MB)
-ll:gsize <int>: size of GASNET global memory available per process (in MB)
-ll:rsize <int>: size of GASNET registered RDMA memory available per process (in MB)
-ll:fsize <int>: size of framebuffer memory for each GPU (in MB)
-ll:zsize <int>: size of zero-copy memory for each GPU (in MB)
-hl:window <int>: maximum number of tasks that can be created in a parent task window
-hl:sched <int>: minimum number of tasks to try to schedule for each invocation of the scheduler
The default mapper also has several flags for controlling the default mapping.
default_mapper.cc for more details.
To start a new Legion application, make a new directory and copy
apps/Makefile.template into your directory under the name
Makefile. Fill in the appropriate fields at the top of the Makefile
with the filenames needed for your application.
Most Legion APIs are described in
legion.h; a smaller number are
described in the various header files in the
directory. The default mapper is available in
Legion has a number of tools to aid in debugging programs.
Extended Correctness Checks
DEBUG=1 CC_FLAGS="-DPRIVILEGE_CHECKS -DBOUNDS_CHECKS"
make and rerun the application. This enables dynamic checks for
privilege and out-of-bounds errors in the application. (These checks
are not enabled by default because they are relatively expensive.) If
the application runs without terminating with an error, then continue
on to Legion Spy.
Legion provides a task-level visualization tool called Legion
Spy. This captures the logical and physical dependence graphs. These
may help, for example, as a sanity check to ensure that the correct
sequence of tasks is being launched (and the tasks have the correct
dependencies). Legion Spy also has a self-checking mode which can
validate the correctness of the runtime's logical and physical
To capture a trace, invoke the application with
spy_%.log. (No special compile-time flags are required.) This will
produce a log file per node. Call the post-processing script to render
PDF files of the dependence graphs:
./app -hl:spy -logfile spy_%.log $LG_RT_DIR/../tools/legion_spy.py -dez spy_*.log
To run Legion Spy's self-checking mode, Legion must be built with the
-DLEGION_SPY. Following this, the application can be run again,
and the script used to validate (or render) the trace.
DEBUG=1 CC_FLAGS="-DLEGION_SPY" make ./app -hl:spy -logfile spy_%.log $LG_RT_DIR/../tools/legion_spy.py -lpa spy_*.log $LG_RT_DIR/../tools/legion_spy.py -dez spy_*.log
Legion contains a task-level profiler. No special compile-time flags
are required. However, it is recommended to build with
to avoid any undesired performance issues.
Run the application with
-hl:prof <N> -logfile prof_%.log where
is the number of nodes to be profiled. (
N can be less than the total
node count---this profiles a subset of the nodes.) This will produce a
log file per node. The profiler itself runs offline and produces an
HTML file which can be loaded in a web browser.
DEBUG=0 make ./app -hl:prof 1 -logfile prof_%.log $LG_RT_DIR/../tools/legion_prof.py prof_*.log
Inorder Execution: Users can force the high-level runtime to execute
all tasks in program order by passing
-hl:inorderflag on the
Dynamic Independence Tests: Users can request the high-level runtime
perform dynamic independence tests between regions and partitions by
-hl:dynamicflag on the command-line.