commit | e06b0b9225585020cb7fb84dfc142ed5fa6a9f7d | [log] [tgz] |
---|---|---|
author | Kaustav Goswami <kggoswami@ucdavis.edu> | Sat Oct 30 00:27:15 2021 -0700 |
committer | Kaustav Goswami <kggoswami@ucdavis.edu> | Mon Nov 15 01:46:07 2021 +0000 |
tree | 1145bd80c9c5c4d5897df4fedc0c30f3d6872195 | |
parent | 153455c998c81daffa33ed873dab03fabee29795 [diff] |
resources: Fixes re-building SPEC CPU2017 benchmarks. This change fixes the issue with running SPEC CPU2017 benchmarks via gem5. It provides a valid path to /usr/bin in the disk-image during execution and avoids re-building the same benchmark programs. Change-Id: Ice2ec9bbf52416558b8fdb5f3f37d501274cf141 Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5-resources/+/52243 Reviewed-by: Jason Lowe-Power <power.jg@gmail.com> Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu> Maintainer: Jason Lowe-Power <power.jg@gmail.com> Maintainer: Bobby R. Bruce <bbruce@ucdavis.edu> Tested-by: Jason Lowe-Power <power.jg@gmail.com> Tested-by: Bobby R. Bruce <bbruce@ucdavis.edu>
This repository contains the sources needed to compile the gem5 resources. The compiled resources are found in the gem5 resources bucket, http://dist.gem5.org/dist. Though these resources are not needed to compile or run gem5, they may be required to execute some gem5 tests or may be useful when carrying out specific simulations.
The following sections outline our versioning policy, how to make changes to this repository, and describe each resource and how they may be built.
We ensure that for each version of the gem5 source there is a corresponding version of the gem5-resources, with the assumption that version X of the gem5 source will be used with version X of the gem5-resources. The gem5-resources repository contains two branches, develop and stable. The stable branch's HEAD points towards the latest gem5 resources release, which will be the same version id as the that of the latest gem5 source. E.g., if the latest release of gem5 is v20.2.0.0, then the latest release of gem5-resources will be v20.2.0.0, with the HEAD of its stable branch tagged as v20.2.0.0. Previous versions will be tagged within the stable branch. Past versions gem5-resources can thereby be checked out with git checkout <VERSION>
. A complete list of versions can be found with git tag
. The develop branch contains code under development and will be merged into the stable branch, then tagged, as part of the next release of gem5. More information on gem5 release procedures can be found here. Any release procedures related to the gem5 source can be assumed to be applicable to gem5-resources.
The compiled resources for gem5 can be found under http://dist.gem5.org/dist/{VERSION}. E.g. compiled resources for gem5 v20.2 are under http://dist.gem5.org/dist/develop and are compiled from gem5-resources v20.2. http://dist.gem5.org/dist/develop is kept in sync with the develop branch, and therefore should not be depended upon for stable, regular usage.
Note: Resource files for gem5 v19.0.0.0, our legacy release, can be found under http://dist.gem5.org/dist/current.
Changes to this repository are made to the develop branch via our Gerrit code review system. Therefore, to make changes, first clone the repository checkout the develop branch:
git clone https://gem5.googlesource.com/public/gem5-resources git checkout --track origin/develop
Then make changes and commit. When ready, push to Gerrit with:
git push origin HEAD:refs/for/develop
The change will then be reviewed via our Gerrit code review system. Once fully accepted and merged into the gem5-resources repository, please contact Bobby R. Bruce bbruce@ucdavis.edu to have the compiled sources uploaded to the gem5 resources bucket.
The RISCV Tests source can be found in the src/riscv-tests
directory. More information about these tests can be found in src/riscv-tests/README.md
.
The RISCV Tests in this repository were obtained from https://github.com/riscv-software-src/riscv-tests.git, revision e65ecdf941a5484af27f9be223fb655ebcb0398b.
To compile the RISCV Tests the RISCV GNU Compiler must be installed.
Then, to compile:
cd src/riscv-tests autoconf ./configure --prefix=/opt/riscv/target make -C src/riscv-tests
This RISCV binaries can then be found within the src/riscv-tests/benchmarks
directory.
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/dhrystone.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/median.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/mm.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/mt-matmul.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/mt-vvadd.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/multiply.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/pmp.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/qsort.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/rsort.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/spmv.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/towers.riscv
http://dist.gem5.org/dist/v21-1/test-progs/riscv-tests/vvadd.riscv
The simple resources are small binaries, often used to run quick tests and checks in gem5. They are baremetal.
Simple single source file per executable userland or baremetal examples.
The toplevel executables under src/simple
can be built for any ISA that we have a cross compiler for. The current cross compilers supported are :
x86_64
(as installed via APT with sudo apt install build-essential
)aarch64-linux-gnu-gcc/arch64-linux-gnu-g++
arm-linux-gnueabihf-gcc/arm-linux-gnueabihf-g++
riscv64-linux-gnu-gcc/riscv64-linux-gnu-g++
Examples that build only for some ISAs specific ones are present under src/simple/<ISA>
subdirs, e.g. src/simple/aarch64/
,
The ISA names are meant to match uname -m
, e.g.:
aarch64
arm
riscv
x86_64
sparc64
You have to specify the path to the gem5 source code with GEM5_ROOT
variable so that m5ops can be used from there. For example for a native build:
cd src/simple make -j`nproc` GEM5_ROOT=../../../
The default of that variable is such that if you place this repository and the gem5 repository in the same directory:
./gem5/ ./gem5-resources/
you can omit that variable and build just with:
make
After the building, the generated files are located under:
./out/<ISA>/
For example, some of the userland executables built on x86 are:
./out/x86_64/user/hello.out ./out/x86_64/user/x86_64/mwait.out
Or if you build for a different ISA:
make ISA=aarch64
some of the executables would be:
./out/aarch64/user/hello.out ./out/aarch64/user/aarch64/futex_ldxr_stxr.out
By default, only userland executables are built. You can build just the baremetal ones instead with:
make ISA=aarch64 bare
or both userland and baremetal with:
make ISA=aarch64 all
A sample baremetal executable generated by this is:
out/aarch64/bare/m5_exit.out
Only ISAs that have a corresponding src/simple/bootloader/
file can build for baremetal, e.g. src/simple/bootloader/aarch64.S
.
Note that a some C source files can produce both a baremetal and an userland. For example m5_exit.c
produces both:
out/aarch64/bare/m5_exit.out out/aarch64/user/m5_exit.out
However, since the regular userland toolchain is used rather than a more specialized baremetal toolchain, the C standard library is not available. Therefore, only very few C examples can build for baremetal, notably the ones that use m5ops.
There are also examples that can only build for baremetal, e.g. aarch64/semihost_exit
only builds for baremetal, as semihosting is not available on userland.
The simple
directory is also able to generate squashfs images containing only a single userland executable at /sbin/init
for any of the userland executables. This can be done with a command of type:
make ISA=aarch64 out/aarch64/squashfs/m5_exit.squashfs
Squashfs is a filesystem type that the Linux kernel understands natively, exactly like ext4, except that it is a bit more convenient to create, and write-only.
You can therefore give those squashfs images to gem5 exactly as you would give a normal ext4 raw image, by pointing to it for example with fs.py --disk-image=m5_exit.squashfs
as shown at: https://www.gem5.org/documentation/general_docs/fullsystem/building_arm_kernel Linux will then run the given userland executable after Linux boots as the init program.
The initial motivation for this was to generate simple test images for Linux boot.
Since this is a less common use case, squashfs images are not currently generated by any single phony target all at once.
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_pthread_create_seq
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_pthread_create_para
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_pthread_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_atomic
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_pthread_cond
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_std_thread
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_std_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/x86/test_std_condition_variable
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_pthread_create_seq
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_pthread_create_para
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_pthread_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_atomic
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_pthread_cond
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_std_thread
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_std_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch32/test_std_condition_variable
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_pthread_create_seq
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_pthread_create_para
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_pthread_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_atomic
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_pthread_cond
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_std_thread
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_std_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/aarch64/test_std_condition_variable
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_pthread_create_seq
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_pthread_create_para
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_pthread_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_atomic
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_pthread_cond
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_std_thread
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_std_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/riscv64/test_std_condition_variable
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_pthread_create_seq
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_pthread_create_para
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_pthread_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_atomic
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_pthread_cond
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_std_thread
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_std_mutex
http://dist.gem5.org/dist/v21-1/test-progs/pthreads/sparc64/test_std_condition_variable
To compile:
cd src/gpu/square docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make gfx8-apu
The compiled binary can be found in src/gpu/square/bin
http://dist.gem5.org/dist/v21-1/test-progs/square/square.o
Based off of the Square resource in this repository, this resource serves as an example for using an HSA Agent Packet to send commands to the GPU command processor included in the GCN_X86 build of gem5.
The example command extracts the kernel‘s completion signal from the domain of the command processor and the GPU’s dispatcher. Initially this was a workaround for the hipDeviceSynchronize bug, now fixed. The method of waiting on a signal can be applied to other agent packet commands though.
Custom commands can be added to the command processor in gem5 to control the GPU in novel ways.
To compile:
cd src/gpu/hsa-agent-pkt docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make gfx8-apu
The compiled binary can be found in src/gpu/hsa-agent-pkt/bin
The HIP sample apps contain applications that introduce various GPU programming concepts that are usable in HIP.
The samples cover topics such as using and accessing different parts of GPU memory, running multiple GPU streams, and optimization techniques for GPU code.
Certain apps aren't included due to complexities with either ROCm or Docker (hipEvent, profiler), or due to lack of feature support in gem5 (peer2peer)
cd src/gpu/hip-samples docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make
Individual programs can be made by specifying the name of the program
By default, this code builds for gfx801, a GCN3-based APU. This can be overridden by specifying -e HCC_AMDGPU_TARGET=<target>
in the build command.
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/2dshfl
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/dynamic_shared
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/inline_asm
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/MatrixTranspose
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/sharedMemory
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/shfl
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/stream
http://dist.gem5.org/dist/v21-1/test-progs/hip-samples/unroll
Heterosync is a benchmark suite used to test the performance of various types of fine-grained synchronization on tightly-coupled GPUs. The version in gem5-resources contains only the HIP code.
The README in the heterosync folder details the various synchronization primitives and the other command-line arguments for use with heterosync.
cd src/gpu/heterosync docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make release-gfx8
The release-gfx8 target builds for gfx801, a GCN3-based APU, and gfx803, a GCN3-based dGPU. There are other targets (release) that build for GPU types that are currently unsupported in gem5.
http://dist.gem5.org/dist/v21-1/test-progs/heterosync/gcn3/allSyncPrims-1kernel
lulesh is a DOE proxy application that is used as an example of hydrodynamics modeling. The version provided is for use with the gpu-compute model of gem5.
cd src/gpu/lulesh docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make
By default, the Makefile builds for gfx801, and is placed in the src/gpu/lulesh/bin
folder.
lulesh is a GPU application, which requires that gem5 is built with the GCN3_X86 architecture. To build GCN3_X86:
# Working directory is your gem5 directory docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu scons -sQ -j$(nproc) build/GCN3_X86/gem5.opt
The following command shows how to run lulesh
Note: lulesh has two optional command-line arguments, to specify the stop time and number of iterations. To set the arguments, add --options="<stop_time> <num_iters>
to the run command. The default arguments are equivalent to --options="1.0e-2 10"
# Assuming gem5 and gem5-resources are in your working directory docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu gem5/build/GCN3_X86/gem5.opt gem5/configs/example/apu_se.py -n3 --mem-size=8GB --benchmark-root=gem5-resources/src/gpu/lulesh/bin -clulesh
http://dist.gem5.org/dist/v21-1/test-progs/lulesh/lulesh
HACC is a DoE application designed to simulate the evolution of the universe by simulating the formation of structure in collisionless fluids under the influence of gravity. The halo-finder code can be GPU accelerated by using the code in RCBForceTree.cxx
src/gpu/halo-finder/src
contains the code required to build and run ForceTreeTest from src/halo_finder
in the main HACC codebase. src/gpu/halo-finder/src/dfft
contains the dfft code from src/dfft
in the main HACC codebase.
halo-finder requires that certain libraries that aren't installed by default in the GCN3 docker container provided by gem5, and that the environment is configured properly in order to build. We provide a Dockerfile that installs those libraries and sets the environment.
In order to test the GPU code in halo-finder, we compile and run ForceTreeTest.
To build the Docker image and the benchmark:
cd src/gpu/halo-finder docker build -t <image_name> . docker run --rm -v ${PWD}:${PWD} -w ${PWD}/src -u $UID:$GID <image_name> make hip/ForceTreeTest
The binary is built for gfx801 by default and is placed at src/gpu/halo-finder/src/hip/ForceTreeTest
ForceTreeTest is a GPU application, which requires that gem5 is built with the GCN3_X86 architecture. To build GCN3_X86:
# Working directory is your gem5 directory docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID <image_name> scons -sQ -j$(nproc) build/GCN3_X86/gem5.opt
To run ForceTreeTest:
# Assuming gem5 and gem5-resources are in the working directory docker run --rm -v $PWD:$PWD -w $PWD -u $UID:$GID <image_name> gem5/build/GCN3_X86/gem5.opt gem5/configs/example/apu_se.py -n3 --benchmark-root=gem5-resources/src/gpu/halo-finder/src/hip -cForceTreeTest --options="0.5 0.1 64 0.1 1 N 12 rcb"
http://dist.gem5.org/dist/v21-1/test-progs/halo-finder/ForceTreeTest
DNNMark is a benchmark framework used to characterize the performance of deep neural network (DNN) primitive workloads.
To build DNNMark: NOTE: Due to DNNMark building a library, it‘s important to mount gem5-resources to the same directory within the docker container when building and running, as otherwise the benchmarks won’t be able to link against the library. The example commands do this by using -v ${PWD}:${PWD}
in the docker run commands
cd src/gpu/DNNMark docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu ./setup.sh HIP docker run --rm -v ${PWD}:${PWD} -w ${PWD}/build -u $UID:$GID gcr.io/gem5-test/gcn-gpu make
DNNMark uses MIOpen kernels, which are unable to be compiled on-the-fly in gem5. We have provided a python script to generate these kernels for a subset of the benchmarks for a gfx801 GPU with 4 CUs by default
To generate the MIOpen kernels:
cd src/gpu/DNNMark docker run --rm -v ${PWD}:${PWD} -v${PWD}/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD} gcr.io/gem5-test/gcn-gpu python3 generate_cachefiles.py cachefiles.csv [--gfx-version={gfx801,gfx803}] [--num-cus=N]
Due to the large amounts of memory that need to be set up for DNNMark, we have added in the ability to MMAP a file to reduce setup time, as well as added a program that can generate a 2GB file of floats.
To make the MMAP file:
cd src/gpu/DNNMark g++ -std=c++0x generate_rand_data.cpp -o generate_rand_data ./generate_rand_data
DNNMark is a GPU application, which requires that gem5 is built with the GCN3_X86 architecture. To build GCN3_X86:
# Working directory is your gem5 directory docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu scons -sQ -j$(nproc) build/GCN3_X86/gem5.opt
To run one of the benchmarks (fwd softmax) in gem5:
# Assuming gem5 and gem5-resources are sub-directories of the current directory docker run --rm -v ${PWD}:${PWD} -v ${PWD}/gem5-resources/src/gpu/DNNMark/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD} gcr.io/gem5-test/gcn-gpu gem5/build/GCN3_X86/gem5.opt gem5/configs/example/apu_se.py -n3 --benchmark-root=gem5-resources/src/gpu/DNNMark/build/benchmarks/test_fwd_softmax -cdnnmark_test_fwd_softmax --options="-config gem5-resources/src/gpu/DNNMark/config_example/softmax_config.dnnmark -mmap gem5-resources/src/gpu/DNNMark/mmap.bin"
pennant is an unstructured mesh physics mini-app designed for advanced architecture research. It contains mesh data structures and a few physics algorithms adapted from the LANL rad-hydro code FLAG, and gives a sample of the typical memory access patterns of FLAG.
cd src/gpu/pennant docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu make
By default, the binary is built for gfx801 and is placed in src/gpu/pennant/build
pennant is a GPU application, which requires that gem5 is built with the GCN3_X86 architecture.
pennant has sample input files located at src/gpu/pennant/test
. The following command shows how to run the sample noh
# Assuming gem5 and gem5-resources are in your working directory docker run --rm -v ${PWD}:${PWD} -w ${PWD} -u $UID:$GID gcr.io/gem5-test/gcn-gpu gem5/build/GCN3_X86/gem5.opt gem5/configs/example/apu_se.py -n3 --benchmark-root=gem5-resources/src/gpu/pennant/build -cpennant --options="gem5-resources/src/gpu/pennant/test/noh/noh.pnt"
The output gets placed in src/gpu/pennant/test/noh/
, and the file noh.xy
against the noh.xy.std
file. Note: Only some tests have .xy.std
files to compare against, and there may be slight differences due to floating-point rounding
http://dist.gem5.org/dist/v21-1/test-progs/pennant/pennant
The Standard Performance Evaluation Corporation (SPEC) CPU 2006 benchmarks are designed to provide performance measurements that can be used to compare compute-intensive workloads on different computer systems. SPEC CPU 2006 contains 12 different benchmark tests.
src/spec-2006
provides resources on creating a SPEC 2006 disk image, and necessary scripts to run the SPEC 2006 benchmarks within X86 gem5 simulations. Please consult the src/spec-2006/README.md
for more information.
Please note, due to licensing issues, the SPEC 2006 iso cannot be provided as part of this repository.
The Standard Performance Evaluation Corporation (SPEC) CPU 2017 benchmarks are designed to provide performance measurements that can be used to compare compute-intensive workloads on different computer systems. SPEC CPU 2017 contains 43 benchmarks organized into four suites: SPECspeed 2017 Integer, SPECspeed 2017 Floating Point, SPECrate 2017 Integer, and SPECrate 2017 Floating Point.
src/spec-2017
provides resources on creating a SPEC 2017 disk image, and necessary scripts to run the SPEC 2017 benchmarks within X86 gem5 simulations. Please consult the src/spec-2017/README.md
for more information.
Please note, due to licensing issues, the SPEC 2017 iso cannot be provided as part of this repository.
GAPBS is a graph processing benchmark suite and it contains 6 kernels: Breadth-First Search, PageRank, Connected Components, Betweenness Centrality, Single-Source Shortest Paths, and Triangle Counting.
We obtained the GAPBS benchmark suite from http://gap.cs.berkeley.edu/benchmark.html
src/gapbs
contains resources to build a GAPBS disk image which may be used to run the benchmark on gem5 X86 simulations. src/gapbs/README.md
contains build and usage instructions.
http://dist.gem5.org/dist/v21-1/images/x86/ubuntu-18-04/gapbs.img.gz.
The Princeton Application Repository for Shared-Memory Computers (PARSEC) is a benchmark suite composed of multithreaded programs.
We used PARSEC 3.0, available from https://parsec.cs.princeton.edu.
In src/parsec
we provide the source to build a disk image which may be used, alongside configuration files, to run the PARSEC Benchmark Suite on gem5 architectural simulations. Please consult src/parsec/README.md
for build and execution information.
http://dist.gem5.org/dist/v21-1/images/x86/ubuntu-18-04/parsec.img.gz.
The NAS Parallel Benchmarks (NPB) are a small set of programs designed to help evaluate the performance of parallel supercomputers. The set consists of five Linux Kernels and three pseudo-applications. gem5 resources provides a disk image, and scripts allowing for the NPB image to be run within gem5 X86 simulations.
We use NPB 3.4.1, available from https://www.nas.nasa.gov/publications/npb.html.
The npb resources can be found in src/npb
. It consists of:
The instructions to build the npb disk image, a Linux kernel binary, and how to use gem5 run scripts to run npb are available in the README file.
http://dist.gem5.org/dist/v21-1/images/x86/ubuntu-18-04/npb.img.gz
The Linux boot tests refer to the tests performed with different gem5 configurations to check its ability to boot a Linux kernel. More information on Linux boot tests can be found here.
The boot-tests resources consist of three main components:
The instructions to build the x86-ubuntu disk image, the Linux binaries, and how to use gem5 run scripts to run boot-tests are available in this README file.
The RISCV Full System resource includes a RISCV boot loader (berkeley bootloader (bbl)
) to boot the Linux 5.10 kernel on a RISCV system, and an image which includes the BusyBox software suite. The resource also contains simple gem5 run/config scripts to run Linux full system simulations in which a user may telnet into.
Further information on building a riscv disk image, a riscv boot loader, and how to use gem5 scripts to run riscv Linux full system simulations, is available in the README file.
http://dist.gem5.org/dist/develop/images/riscv/busybox/riscv-disk.img.gz
http://dist.gem5.org/dist/develop/kernels/riscv/static/bootloader-vmlinux-5.10
The RISCV Full System resource includes a RISCV bootloader (berkeley bootloader (bbl)
) to boot the Linux 5.10 kernel on a RISCV system. The workload and the Linux utils (provided by BusyBox) are also included in the bootloader. The resource also contains simple gem5 run/config scripts to run Linux full system simulations in which a user may telnet into.
More details on building such a RISCV bootloader and hwo does it work are available in the README.md file.
http://dist.gem5.org/dist/develop/misc/riscv/bbl-busybox-boot-exit
The Insttests test SPARC instructions.
Creating the SPARC Insttest binary requires a SPARC cross compile. Instructions on creating a cross compiler can be found here.
To compile:
cd src/insttest make
We provide a docker image with a pre-loaded SPARC cross compiler. To use:
cd src/insttest docker run --volume $(pwd):$(pwd) -w $(pwd) --rm gcr.io/gem5-test/sparc64-gnu-cross:latest make
The compiled binary can be found in src/insttest/bin
.
http://dist.gem5.org/dist/v21-1/test-progs/insttest/bin/sparc/linux/insttest
Contains scripts to create a Linux kernel binary.
Instructions on how to use the scripts can be found here src/linux-kernel/README.md
.
http://dist.gem5.org/dist/v21-1/kernels/x86/static/vmlinux-4.4.186 http://dist.gem5.org/dist/v21-1/kernels/x86/static/vmlinux-4.9.186 http://dist.gem5.org/dist/v21-1/kernels/x86/static/vmlinux-4.14.134 http://dist.gem5.org/dist/v21-1/kernels/x86/static/vmlinux-4.19.83
There is no universal license encompassing all this repository's contents. The licences covering the individual gem5 resources are therefore highlighted below.
src/asmtest/LICENSE
.src/riscv-tests/LICENSE
.src/gpu/square
.src/gpu/hsa-agent-pkt/square.cpp
is licensed under the same licence as ‘src/gpu/square/square.cpp’. src/gpu/hsa-agent-pkt/HSA_Interface.[h|.cpp]
are licensed under a BSD Lisense (A University of Maryland copyright).src/gpu/heterosync/LICENSE.txt
src/gpu/lulesh/src/gpu/lulesh.hip.cc
src/gpu/DNNMark/LICENSE
src/gpu/pennant/LICENSE
src/gpu/square.src/spec-2006
.src/spec-2017
.src/gapbs
.src/parsec/disk-image/parsec/parsec-benchmark/LICENSE
). For the remaining files, please consult copyright notices in individual source files.src/npb
. The NAS Parallel Benchmarks utilize a permissive BSD-style license.src/x86-ubuntu
.src/insttest
.src/linux-kernel
.src/hack-back
.src/simple
.