Merge pull request #95 from UoB-HPC/rust

Rust implementation
This commit is contained in:
Tom Deakin 2021-12-09 16:45:28 +00:00 committed by GitHub
commit 9ec3018b7f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
17 changed files with 2357 additions and 18 deletions

View File

@ -11,6 +11,22 @@ on:
jobs: jobs:
test-rust:
runs-on: ubuntu-18.04
defaults:
run:
working-directory: ./src/rust/rust-stream
steps:
- uses: actions/checkout@v2
- name: Setup project
run: rustup install nightly
- name: Compile project
run: cargo +nightly build --release
- name: Test project
run: cargo +nightly test --release
- name: Test run project
run: ./target/release/rust-stream --arraysize 2048
test-java: test-java:
runs-on: ubuntu-18.04 runs-on: ubuntu-18.04
defaults: defaults:

View File

@ -17,8 +17,8 @@ This code was previously called GPU-STREAM.
- [Programming Models](#programming-models) - [Programming Models](#programming-models)
- [How is this different to STREAM?](#how-is-this-different-to-stream) - [How is this different to STREAM?](#how-is-this-different-to-stream)
- [Building](#building) - [Building](#building)
- [CMake](#cmake) - [CMake](#cmake)
- [GNU Make (removed)](#gnu-make) - [GNU Make (removed)](#gnu-make)
- [Results](#results) - [Results](#results)
- [Contributing](#contributing) - [Contributing](#contributing)
- [Citing](#citing) - [Citing](#citing)
@ -29,23 +29,23 @@ This code was previously called GPU-STREAM.
BabelStream is currently implemented in the following parallel programming models, listed in no particular order: BabelStream is currently implemented in the following parallel programming models, listed in no particular order:
- OpenCL - OpenCL
- CUDA - CUDA
- HIP - HIP
- OpenACC - OpenACC
- OpenMP 3 and 4.5 - OpenMP 3 and 4.5
- C++ Parallel STL - C++ Parallel STL
- Kokkos - Kokkos
- RAJA - RAJA
- SYCL and SYCL 2020 - SYCL and SYCL 2020
- TBB - TBB
- Thrust (via CUDA or HIP) - Thrust (via CUDA or HIP)
This project also contains implementations in alternative languages with different build systems: This project also contains implementations in alternative languages with different build systems:
* Julia - [JuliaStream.jl](./src/julia/JuliaStream.jl) * Julia - [JuliaStream.jl](./src/julia/JuliaStream.jl)
* Java - [java-stream](./src/java/java-stream) * Java - [java-stream](./src/java/java-stream)
* Scala - [scala-stream](./src/scala/scala-stream) * Scala - [scala-stream](./src/scala/scala-stream)
* Rust - [rust-stream](./src/rust/rust-stream)
## How is this different to STREAM? ## How is this different to STREAM?
@ -136,7 +136,7 @@ For example:
Alternatively, refer to the [CI script](./src/ci-test-compile.sh), which test-compiles most of the models, and see which flags are used there. Alternatively, refer to the [CI script](./src/ci-test-compile.sh), which test-compiles most of the models, and see which flags are used there.
*It is recommended that you delete the `build` directory when you change any of the build flags.* *It is recommended that you delete the `build` directory when you change any of the build flags.*
### GNU Make ### GNU Make
@ -170,15 +170,15 @@ Deakin T, Price J, Martineau M, McIntosh-Smith S. GPU-STREAM v2.0: Benchmarking
* Deakin T, Price J, Martineau M, McIntosh-Smith S. Evaluating attainable memory bandwidth of parallel programming models via BabelStream. International Journal of Computational Science and Engineering. Special issue. Vol. 17, No. 3, pp. 247262. 2018.DOI: 10.1504/IJCSE.2018.095847 * Deakin T, Price J, Martineau M, McIntosh-Smith S. Evaluating attainable memory bandwidth of parallel programming models via BabelStream. International Journal of Computational Science and Engineering. Special issue. Vol. 17, No. 3, pp. 247262. 2018.DOI: 10.1504/IJCSE.2018.095847
* Deakin T, McIntosh-Smith S. GPU-STREAM: Benchmarking the achievable memory bandwidth of Graphics Processing Units. 2015. Poster session presented at IEEE/ACM SuperComputing, Austin, United States. * Deakin T, McIntosh-Smith S. GPU-STREAM: Benchmarking the achievable memory bandwidth of Graphics Processing Units. 2015. Poster session presented at IEEE/ACM SuperComputing, Austin, United States.
You can view the [Poster and Extended Abstract](http://sc15.supercomputing.org/sites/all/themes/SC15images/tech_poster/tech_poster_pages/post150.html). You can view the [Poster and Extended Abstract](http://sc15.supercomputing.org/sites/all/themes/SC15images/tech_poster/tech_poster_pages/post150.html).
* Deakin T, Price J, Martineau M, McIntosh-Smith S. GPU-STREAM: Now in 2D!. 2016. Poster session presented at IEEE/ACM SuperComputing, Salt Lake City, United States. * Deakin T, Price J, Martineau M, McIntosh-Smith S. GPU-STREAM: Now in 2D!. 2016. Poster session presented at IEEE/ACM SuperComputing, Salt Lake City, United States.
You can view the [Poster and Extended Abstract](http://sc16.supercomputing.org/sc-archive/tech_poster/tech_poster_pages/post139.html). You can view the [Poster and Extended Abstract](http://sc16.supercomputing.org/sc-archive/tech_poster/tech_poster_pages/post139.html).
* Raman K, Deakin T, Price J, McIntosh-Smith S. Improving achieved memory bandwidth from C++ codes on Intel Xeon Phi Processor (Knights Landing). IXPUG Spring Meeting, Cambridge, UK, 2017. * Raman K, Deakin T, Price J, McIntosh-Smith S. Improving achieved memory bandwidth from C++ codes on Intel Xeon Phi Processor (Knights Landing). IXPUG Spring Meeting, Cambridge, UK, 2017.
* Deakin T, Price J, McIntosh-Smith S. Portable methods for measuring cache hierarchy performance. 2017. Poster sessions presented at IEEE/ACM SuperComputing, Denver, United States. * Deakin T, Price J, McIntosh-Smith S. Portable methods for measuring cache hierarchy performance. 2017. Poster sessions presented at IEEE/ACM SuperComputing, Denver, United States.
You can view the [Poster and Extended Abstract](http://sc17.supercomputing.org/SC17%20Archive/tech_poster/tech_poster_pages/post155.html) You can view the [Poster and Extended Abstract](http://sc17.supercomputing.org/SC17%20Archive/tech_poster/tech_poster_pages/post155.html)
[1]: McCalpin, John D., 1995: "Memory Bandwidth and Machine Balance in Current High Performance Computers", IEEE Computer Society Technical Committee on Computer Architecture (TCCA) Newsletter, December 1995. [1]: McCalpin, John D., 1995: "Memory Bandwidth and Machine Balance in Current High Performance Computers", IEEE Computer Society Technical Committee on Computer Architecture (TCCA) Newsletter, December 1995.

View File

@ -0,0 +1,2 @@
[build]
rustflags = ["-C", "target-cpu=native"]

2
src/rust/rust-stream/.gitignore vendored Normal file
View File

@ -0,0 +1,2 @@
/target
.idea

636
src/rust/rust-stream/Cargo.lock generated Normal file
View File

@ -0,0 +1,636 @@
# This file is automatically @generated by Cargo.
# It is not intended for manual editing.
version = 3
[[package]]
name = "ansi_term"
version = "0.11.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ee49baf6cb617b853aa8d93bf420db2383fab46d314482ca2803b40d5fde979b"
dependencies = [
"winapi 0.3.9",
]
[[package]]
name = "atty"
version = "0.2.14"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d9b39be18770d11421cdb1b9947a45dd3f37e93092cbf377614828a319d5fee8"
dependencies = [
"hermit-abi",
"libc",
"winapi 0.3.9",
]
[[package]]
name = "autocfg"
version = "1.0.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "cdb031dd78e28731d87d56cc8ffef4a8f36ca26c38fe2de700543e627f8a464a"
[[package]]
name = "bitflags"
version = "1.3.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bef38d45163c2f1dde094a7dfd33ccf595c92905c8f8f4fdc18d06fb1037718a"
[[package]]
name = "cfg-if"
version = "1.0.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
[[package]]
name = "clap"
version = "2.33.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "37e58ac78573c40708d45522f0d80fa2f01cc4f9b4e2bf749807255454312002"
dependencies = [
"ansi_term",
"atty",
"bitflags",
"strsim",
"textwrap",
"unicode-width",
"vec_map",
]
[[package]]
name = "colour"
version = "0.6.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a27e4532f26f510c24bb8477d963c0c3ef27e293c3b2c507cccb0536d493201a"
dependencies = [
"crossterm",
]
[[package]]
name = "core_affinity"
version = "0.5.10"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7f8a03115cc34fb0d7c321dd154a3914b3ca082ccc5c11d91bf7117dbbe7171f"
dependencies = [
"kernel32-sys",
"libc",
"num_cpus",
"winapi 0.2.8",
]
[[package]]
name = "crossbeam"
version = "0.8.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4ae5588f6b3c3cb05239e90bd110f257254aecd01e4635400391aeae07497845"
dependencies = [
"cfg-if",
"crossbeam-channel",
"crossbeam-deque",
"crossbeam-epoch",
"crossbeam-queue",
"crossbeam-utils",
]
[[package]]
name = "crossbeam-channel"
version = "0.5.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "06ed27e177f16d65f0f0c22a213e17c696ace5dd64b14258b52f9417ccb52db4"
dependencies = [
"cfg-if",
"crossbeam-utils",
]
[[package]]
name = "crossbeam-deque"
version = "0.8.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6455c0ca19f0d2fbf751b908d5c55c1f5cbc65e03c4225427254b46890bdde1e"
dependencies = [
"cfg-if",
"crossbeam-epoch",
"crossbeam-utils",
]
[[package]]
name = "crossbeam-epoch"
version = "0.9.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4ec02e091aa634e2c3ada4a392989e7c3116673ef0ac5b72232439094d73b7fd"
dependencies = [
"cfg-if",
"crossbeam-utils",
"lazy_static",
"memoffset",
"scopeguard",
]
[[package]]
name = "crossbeam-queue"
version = "0.3.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9b10ddc024425c88c2ad148c1b0fd53f4c6d38db9697c9f1588381212fa657c9"
dependencies = [
"cfg-if",
"crossbeam-utils",
]
[[package]]
name = "crossbeam-utils"
version = "0.8.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d82cfc11ce7f2c3faef78d8a684447b40d503d9681acebed6cb728d45940c4db"
dependencies = [
"cfg-if",
"lazy_static",
]
[[package]]
name = "crossterm"
version = "0.19.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7c36c10130df424b2f3552fcc2ddcd9b28a27b1e54b358b45874f88d1ca6888c"
dependencies = [
"bitflags",
"crossterm_winapi",
"lazy_static",
"libc",
"mio",
"parking_lot",
"signal-hook",
"winapi 0.3.9",
]
[[package]]
name = "crossterm_winapi"
version = "0.7.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0da8964ace4d3e4a044fd027919b2237000b24315a37c916f61809f1ff2140b9"
dependencies = [
"winapi 0.3.9",
]
[[package]]
name = "either"
version = "1.6.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e78d4f1cc4ae33bbfc157ed5d5a5ef3bc29227303d595861deb238fcec4e9457"
[[package]]
name = "heck"
version = "0.3.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6d621efb26863f0e9924c6ac577e8275e5e6b77455db64ffa6c65c904e9e132c"
dependencies = [
"unicode-segmentation",
]
[[package]]
name = "hermit-abi"
version = "0.1.19"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "62b467343b94ba476dcb2500d242dadbb39557df889310ac77c5d99100aaac33"
dependencies = [
"libc",
]
[[package]]
name = "instant"
version = "0.1.12"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7a5bbe824c507c5da5956355e86a746d82e0e1464f65d862cc5e71da70e94b2c"
dependencies = [
"cfg-if",
]
[[package]]
name = "kernel32-sys"
version = "0.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7507624b29483431c0ba2d82aece8ca6cdba9382bff4ddd0f7490560c056098d"
dependencies = [
"winapi 0.2.8",
"winapi-build",
]
[[package]]
name = "lazy_static"
version = "1.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e2abad23fbc42b3700f2f279844dc832adb2b2eb069b2df918f455c4e18cc646"
[[package]]
name = "libc"
version = "0.2.108"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8521a1b57e76b1ec69af7599e75e38e7b7fad6610f037db8c79b127201b5d119"
[[package]]
name = "lock_api"
version = "0.4.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "712a4d093c9976e24e7dbca41db895dabcbac38eb5f4045393d17a95bdfb1109"
dependencies = [
"scopeguard",
]
[[package]]
name = "log"
version = "0.4.14"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "51b9bbe6c47d51fc3e1a9b945965946b4c44142ab8792c50835a980d362c2710"
dependencies = [
"cfg-if",
]
[[package]]
name = "memoffset"
version = "0.6.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "59accc507f1338036a0477ef61afdae33cde60840f4dfe481319ce3ad116ddf9"
dependencies = [
"autocfg",
]
[[package]]
name = "mio"
version = "0.7.14"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8067b404fe97c70829f082dec8bcf4f71225d7eaea1d8645349cb76fa06205cc"
dependencies = [
"libc",
"log",
"miow",
"ntapi",
"winapi 0.3.9",
]
[[package]]
name = "miow"
version = "0.3.7"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b9f1c5b025cda876f66ef43a113f91ebc9f4ccef34843000e0adf6ebbab84e21"
dependencies = [
"winapi 0.3.9",
]
[[package]]
name = "ntapi"
version = "0.3.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3f6bb902e437b6d86e03cce10a7e2af662292c5dfef23b65899ea3ac9354ad44"
dependencies = [
"winapi 0.3.9",
]
[[package]]
name = "num-traits"
version = "0.2.14"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9a64b1ec5cda2586e284722486d802acf1f7dbdc623e2bfc57e65ca1cd099290"
dependencies = [
"autocfg",
]
[[package]]
name = "num_cpus"
version = "1.13.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "05499f3756671c15885fee9034446956fff3f243d6077b91e5767df161f766b3"
dependencies = [
"hermit-abi",
"libc",
]
[[package]]
name = "parking_lot"
version = "0.11.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7d17b78036a60663b797adeaee46f5c9dfebb86948d1255007a1d6be0271ff99"
dependencies = [
"instant",
"lock_api",
"parking_lot_core",
]
[[package]]
name = "parking_lot_core"
version = "0.8.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d76e8e1493bcac0d2766c42737f34458f1c8c50c0d23bcb24ea953affb273216"
dependencies = [
"cfg-if",
"instant",
"libc",
"redox_syscall",
"smallvec",
"winapi 0.3.9",
]
[[package]]
name = "pest"
version = "2.1.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "10f4872ae94d7b90ae48754df22fd42ad52ce740b8f370b03da4835417403e53"
dependencies = [
"ucd-trie",
]
[[package]]
name = "proc-macro-error"
version = "1.0.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "da25490ff9892aab3fcf7c36f08cfb902dd3e71ca0f9f9517bea02a73a5ce38c"
dependencies = [
"proc-macro-error-attr",
"proc-macro2",
"quote",
"syn",
"version_check",
]
[[package]]
name = "proc-macro-error-attr"
version = "1.0.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a1be40180e52ecc98ad80b184934baf3d0d29f979574e439af5a55274b35f869"
dependencies = [
"proc-macro2",
"quote",
"version_check",
]
[[package]]
name = "proc-macro2"
version = "1.0.32"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ba508cc11742c0dc5c1659771673afbab7a0efab23aa17e854cbab0837ed0b43"
dependencies = [
"unicode-xid",
]
[[package]]
name = "quote"
version = "1.0.10"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "38bc8cc6a5f2e3655e0899c1b848643b2562f853f114bfec7be120678e3ace05"
dependencies = [
"proc-macro2",
]
[[package]]
name = "rayon"
version = "1.5.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c06aca804d41dbc8ba42dfd964f0d01334eceb64314b9ecf7c5fad5188a06d90"
dependencies = [
"autocfg",
"crossbeam-deque",
"either",
"rayon-core",
]
[[package]]
name = "rayon-core"
version = "1.9.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d78120e2c850279833f1dd3582f730c4ab53ed95aeaaaa862a2a5c71b1656d8e"
dependencies = [
"crossbeam-channel",
"crossbeam-deque",
"crossbeam-utils",
"lazy_static",
"num_cpus",
]
[[package]]
name = "redox_syscall"
version = "0.2.10"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8383f39639269cde97d255a32bdb68c047337295414940c68bdd30c2e13203ff"
dependencies = [
"bitflags",
]
[[package]]
name = "rstest"
version = "0.10.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "041bb0202c14f6a158bbbf086afb03d0c6e975c2dec7d4912f8061ed44f290af"
dependencies = [
"cfg-if",
"proc-macro2",
"quote",
"rustc_version",
"syn",
]
[[package]]
name = "rust-stream"
version = "3.4.0"
dependencies = [
"colour",
"core_affinity",
"crossbeam",
"libc",
"num-traits",
"num_cpus",
"rayon",
"rstest",
"rustversion",
"structopt",
"tabular",
]
[[package]]
name = "rustc_version"
version = "0.3.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f0dfe2087c51c460008730de8b57e6a320782fbfb312e1f4d520e6c6fae155ee"
dependencies = [
"semver",
]
[[package]]
name = "rustversion"
version = "1.0.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "61b3909d758bb75c79f23d4736fac9433868679d3ad2ea7a61e3c25cfda9a088"
[[package]]
name = "scopeguard"
version = "1.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d29ab0c6d3fc0ee92fe66e2d99f700eab17a8d57d1c1d3b748380fb20baa78cd"
[[package]]
name = "semver"
version = "0.11.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f301af10236f6df4160f7c3f04eec6dbc70ace82d23326abad5edee88801c6b6"
dependencies = [
"semver-parser",
]
[[package]]
name = "semver-parser"
version = "0.10.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "00b0bef5b7f9e0df16536d3961cfb6e84331c065b4066afb39768d0e319411f7"
dependencies = [
"pest",
]
[[package]]
name = "signal-hook"
version = "0.1.17"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7e31d442c16f047a671b5a71e2161d6e68814012b7f5379d269ebd915fac2729"
dependencies = [
"libc",
"mio",
"signal-hook-registry",
]
[[package]]
name = "signal-hook-registry"
version = "1.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e51e73328dc4ac0c7ccbda3a494dfa03df1de2f46018127f60c693f2648455b0"
dependencies = [
"libc",
]
[[package]]
name = "smallvec"
version = "1.7.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1ecab6c735a6bb4139c0caafd0cc3635748bbb3acf4550e8138122099251f309"
[[package]]
name = "strsim"
version = "0.8.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8ea5119cdb4c55b55d432abb513a0429384878c15dde60cc77b1c99de1a95a6a"
[[package]]
name = "structopt"
version = "0.3.25"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "40b9788f4202aa75c240ecc9c15c65185e6a39ccdeb0fd5d008b98825464c87c"
dependencies = [
"clap",
"lazy_static",
"structopt-derive",
]
[[package]]
name = "structopt-derive"
version = "0.4.18"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "dcb5ae327f9cc13b68763b5749770cb9e048a99bd9dfdfa58d0cf05d5f64afe0"
dependencies = [
"heck",
"proc-macro-error",
"proc-macro2",
"quote",
"syn",
]
[[package]]
name = "syn"
version = "1.0.82"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8daf5dd0bb60cbd4137b1b587d2fc0ae729bc07cf01cd70b36a1ed5ade3b9d59"
dependencies = [
"proc-macro2",
"quote",
"unicode-xid",
]
[[package]]
name = "tabular"
version = "0.1.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e7e35bee02dcefe64a74065b6b869d241eab1a02fea0d65e6074ce4e51894c3b"
dependencies = [
"unicode-width",
]
[[package]]
name = "textwrap"
version = "0.11.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d326610f408c7a4eb6f51c37c330e496b08506c9457c9d34287ecc38809fb060"
dependencies = [
"unicode-width",
]
[[package]]
name = "ucd-trie"
version = "0.1.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "56dee185309b50d1f11bfedef0fe6d036842e3fb77413abef29f8f8d1c5d4c1c"
[[package]]
name = "unicode-segmentation"
version = "1.8.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8895849a949e7845e06bd6dc1aa51731a103c42707010a5b591c0038fb73385b"
[[package]]
name = "unicode-width"
version = "0.1.9"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3ed742d4ea2bd1176e236172c8429aaf54486e7ac098db29ffe6529e0ce50973"
[[package]]
name = "unicode-xid"
version = "0.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8ccb82d61f80a663efe1f787a51b16b5a51e3314d6ac365b08639f52387b33f3"
[[package]]
name = "vec_map"
version = "0.8.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f1bddf1187be692e79c5ffeab891132dfb0f236ed36a43c7ed39f1165ee20191"
[[package]]
name = "version_check"
version = "0.9.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5fecdca9a5291cc2b8dcf7dc02453fee791a280f3743cb0905f8822ae463b3fe"
[[package]]
name = "winapi"
version = "0.2.8"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "167dc9d6949a9b857f3451275e911c3f44255842c1f7a76f33c55103a909087a"
[[package]]
name = "winapi"
version = "0.3.9"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5c839a674fcd7a98952e593242ea400abe93992746761e38641405d28b00f419"
dependencies = [
"winapi-i686-pc-windows-gnu",
"winapi-x86_64-pc-windows-gnu",
]
[[package]]
name = "winapi-build"
version = "0.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2d315eee3b34aca4797b2da6b13ed88266e6d612562a0c46390af8299fc699bc"
[[package]]
name = "winapi-i686-pc-windows-gnu"
version = "0.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ac3b87c63620426dd9b991e5ce0329eff545bccbbb34f3be09ff6fb6ab51b7b6"
[[package]]
name = "winapi-x86_64-pc-windows-gnu"
version = "0.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f"

View File

@ -0,0 +1,34 @@
[package]
name = "rust-stream"
version = "3.4.0"
authors = ["Wei-Chen Lin <wl14928@bristol.ac.uk>"]
edition = "2018"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
num-traits = "0.2.14"
structopt = "0.3.13"
tabular = "0.1.4"
rayon = "1.5.1"
crossbeam = "0.8.1"
num_cpus = "1.13.0"
rustversion = "1.0"
libc = "0.2.97"
core_affinity = "0.5.10"
colour = "0.6.0"
[dev-dependencies]
rstest = "0.10.0"
[build-dependencies]
rustversion = "1.0"
[profile.dev]
opt-level = 2
overflow-checks = true
[profile.release]
opt-level = 3
lto = "thin" # fully enabling this (i.e true) negatively affects performance as tested on both AMD and Intel

View File

@ -0,0 +1,78 @@
rust-stream
===========
This is an implementation of BabelStream in Rust.
Currently, we support three CPU threading API as devices:
* Plain - basic single-threaded `for` version, see [plain_stream.rs](src/plain_stream.rs)
* [Rayon](https://github.com/rayon-rs/rayon) - Parallel with high level API,
see [rayon_stream.rs](src/rayon_stream.rs)
* [Crossbeam](https://github.com/crossbeam-rs/crossbeam) - Parallel with partitions per thread,
see [crossbeam_stream.rs](src/crossbeam_stream.rs)
* Arc - Parallel with `Vec` per thread (static partitions) wrapped in `Mutex` contained in `Arc`s,
see [crossbeam_stream.rs](src/arc_stream.rs)
* Unsafe - Parallel with unsafe pointer per thread (static partitions) to `Vec`,
see [crossbeam_stream.rs](src/unsafe_stream.rs)
In addition, this implementation also supports the following extra flags:
****
```
--init Initialise each benchmark array at allocation time on the main thread
--malloc Use libc malloc instead of the Rust's allocator for benchmark array allocation
--pin Pin threads to distinct cores, this has NO effect in Rayon devices
```
Max thread count is controlled by the environment variable `BABELSTREAM_NUM_THREADS` which is compatible for all devices (avoid setting `RAYON_NUM_THREADS`, the implementation will issue a warning if this happened).
There is an ongoing investigation on potential performance issues under NUMA situations. As part of
the experiment, this implementation made use of the
provisional [Allocator traits](https://github.com/rust-lang/rust/issues/32838) which requires rust
unstable. We hope a NUMA aware allocator will be available once the allocator API reaches rust
stable.
### Build & Run
Prerequisites:
* [Rust toolchain](https://www.rust-lang.org/tools/install)
Once the toolchain is installed, enable the nightly channel:
```shell
> rustup install nightly
> rustup default nightly # optional, this sets `+nightly` automatically for cargo calls later
```
With `cargo` on path, compile and run the benchmark with:
```shell
> cd rust-stream/
> cargo +nightly build --release # or simply `cargo build --release` if nightly channel is the default
> ./target/release/rust-stream --help
rust-stream 3.4.0
USAGE:
rust-stream [FLAGS] [OPTIONS]
FLAGS:
--csv Output as csv table
--float Use floats (rather than doubles)
-h, --help Prints help information
--init Initialise each benchmark array at allocation time on the main thread
--list List available devices
--malloc Use libc malloc instead of the Rust's allocator for benchmark array allocation
--mibibytes Use MiB=2^20 for bandwidth calculation (default MB=10^6)
--nstream-only Only run nstream
--pin Pin threads to distinct cores, this has NO effect in Rayon devices
--triad-only Only run triad
-V, --version Prints version information
OPTIONS:
-s, --arraysize <arraysize> Use <arraysize> elements in the array [default: 33554432]
--device <device> Select device at <device> [default: 0]
-n, --numtimes <numtimes> Run the test <numtimes> times (NUM >= 2) [default: 100]
```

View File

@ -0,0 +1,68 @@
max_width = 100
hard_tabs = false
tab_spaces = 2
newline_style = "Auto"
use_small_heuristics = "Max"
indent_style = "Block"
wrap_comments = false
format_code_in_doc_comments = false
comment_width = 80
normalize_comments = false
normalize_doc_attributes = false
license_template_path = ""
format_strings = false
format_macro_matchers = false
format_macro_bodies = true
empty_item_single_line = true
struct_lit_single_line = true
fn_single_line = true
where_single_line = true
imports_indent = "Block"
imports_layout = "Mixed"
imports_granularity = "Preserve"
group_imports = "Preserve"
reorder_imports = true
reorder_modules = true
reorder_impl_items = false
type_punctuation_density = "Wide"
space_before_colon = false
space_after_colon = true
spaces_around_ranges = false
binop_separator = "Front"
remove_nested_parens = true
combine_control_expr = true
overflow_delimited_expr = false
struct_field_align_threshold = 0
enum_discrim_align_threshold = 0
match_arm_blocks = true
match_arm_leading_pipes = "Never"
force_multiline_blocks = false
fn_args_layout = "Compressed"
brace_style = "PreferSameLine"
control_brace_style = "AlwaysSameLine"
trailing_semicolon = true
trailing_comma = "Vertical"
match_block_trailing_comma = false
blank_lines_upper_bound = 1
blank_lines_lower_bound = 0
edition = "2015"
version = "One"
inline_attribute_width = 0
merge_derives = true
use_try_shorthand = false
use_field_init_shorthand = false
force_explicit_abi = true
condense_wildcard_suffixes = false
color = "Auto"
required_version = "1.4.38"
unstable_features = false
disable_all_formatting = false
skip_children = false
hide_parse_errors = false
error_on_line_overflow = false
error_on_unformatted = false
report_todo = "Never"
report_fixme = "Never"
ignore = []
emit_mode = "Files"
make_backup = false

View File

@ -0,0 +1,254 @@
use std::iter::Sum;
use std::sync::{Arc, Mutex};
use self::core_affinity::CoreId;
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
struct ArcHeapData<T: ArrayType, A: AllocatorType> {
a_chunks: Vec<Arc<Mutex<Vec<T, A>>>>,
b_chunks: Vec<Arc<Mutex<Vec<T, A>>>>,
c_chunks: Vec<Arc<Mutex<Vec<T, A>>>>,
}
pub struct ArcDevice<T: ArrayType, A: AllocatorType> {
pub(crate) ncore: usize,
pub(crate) pin: bool,
pub(crate) core_ids: Vec<CoreId>,
data: ArcHeapData<T, A>,
}
impl<T: ArrayType, A: AllocatorType> ArcDevice<T, A> {
pub fn new(ncore: usize, pin: bool, alloc: A) -> Self {
let mut core_ids = match core_affinity::get_core_ids() {
Some(xs) => xs,
None => {
colour::e_red_ln!("Cannot enumerate cores, pinning will not work if enabled");
(0..ncore).map(|i| CoreId { id: i }).collect()
}
};
core_ids.resize(ncore, core_ids[0]);
let lift =
|| (0..ncore).map(|_| return Arc::new(Mutex::new(Vec::new_in(alloc)))).collect::<Vec<_>>();
let data = ArcHeapData { a_chunks: lift(), b_chunks: lift(), c_chunks: lift() };
ArcDevice { ncore, pin, core_ids, data }
}
pub fn ref_a(&self, t: usize) -> Arc<Mutex<Vec<T, A>>> { self.data.a_chunks[t].clone() }
pub fn ref_b(&self, t: usize) -> Arc<Mutex<Vec<T, A>>> { self.data.b_chunks[t].clone() }
pub fn ref_c(&self, t: usize) -> Arc<Mutex<Vec<T, A>>> { self.data.c_chunks[t].clone() }
// divide the length by the number of cores, the last core gets less work if it does not divide
fn chunk_size(&self, len: usize, t: usize) -> usize {
assert!(t < self.ncore);
let chunk = (len as f64 / self.ncore as f64).ceil() as usize;
if t == self.ncore - 1 {
len - (t * chunk)
} else {
chunk
}
}
}
extern crate core_affinity;
// Arc+Mutex threaded version, it should be semantically equal to the single threaded version
impl<T: 'static + ArrayType + Sync + Send + Sum, A: AllocatorType + Sync + Send + 'static>
RustStream<T> for StreamData<T, ArcDevice<T, A>, A>
{
fn init_arrays(&mut self) {
let init = self.init;
let pin = self.device.pin;
(0..self.device.ncore)
.map(&|t| {
let ref_a = self.device.ref_a(t);
let ref_b = self.device.ref_b(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
ref_a.lock().unwrap().resize(n, init.0);
ref_b.lock().unwrap().resize(n, init.1);
ref_c.lock().unwrap().resize(n, init.2);
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn read_arrays(&mut self) {
let range = self.size;
let unlift = |drain: &mut Vec<T, A>, source: &Vec<Arc<Mutex<Vec<T, A>>>>| {
let xs =
source.into_iter().flat_map(|x| x.lock().unwrap().clone().into_iter()).collect::<Vec<_>>();
for i in 0..range {
drain[i] = xs[i];
}
};
unlift(&mut self.a, &self.device.data.a_chunks);
unlift(&mut self.b, &self.device.data.b_chunks);
unlift(&mut self.c, &self.device.data.c_chunks);
}
fn copy(&mut self) {
let pin = self.device.pin;
(0..self.device.ncore)
.map(move |t| {
let ref_a = self.device.ref_a(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = ref_a.lock().unwrap();
let mut c = ref_c.lock().unwrap();
for i in 0..n {
c[i] = a[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn mul(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
(0..self.device.ncore)
.map(move |t| {
let ref_b = self.device.ref_b(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let mut b = ref_b.lock().unwrap();
let c = ref_c.lock().unwrap();
for i in 0..n {
b[i] = scalar * c[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn add(&mut self) {
let pin = self.device.pin;
(0..self.device.ncore)
.map(&|t| {
let ref_a = self.device.ref_a(t);
let ref_b = self.device.ref_b(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = ref_a.lock().unwrap();
let b = ref_b.lock().unwrap();
let mut c = ref_c.lock().unwrap();
for i in 0..n {
c[i] = a[i] + b[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn triad(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
(0..self.device.ncore)
.map(&|t| {
let ref_a = self.device.ref_a(t);
let ref_b = self.device.ref_b(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let mut a = ref_a.lock().unwrap();
let b = ref_b.lock().unwrap();
let c = ref_c.lock().unwrap();
for i in 0..n {
a[i] = b[i] + scalar * c[i]
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn nstream(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
(0..self.device.ncore)
.map(&|t| {
let ref_a = self.device.ref_a(t);
let ref_b = self.device.ref_b(t);
let ref_c = self.device.ref_c(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let mut a = ref_a.lock().unwrap();
let b = ref_b.lock().unwrap();
let c = ref_c.lock().unwrap();
for i in 0..n {
a[i] += b[i] + scalar * c[i]
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn dot(&mut self) -> T {
let pin = self.device.pin;
(0..self.device.ncore)
.map(&|t| {
let ref_a = self.device.ref_a(t);
let ref_b = self.device.ref_b(t);
let core = self.device.core_ids[t];
let n = self.device.chunk_size(self.size, t);
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = ref_a.lock().unwrap();
let b = ref_b.lock().unwrap();
let mut p = T::default();
for i in 0..n {
p += a[i] * b[i];
}
p
})
})
.collect::<Vec<_>>()
.into_iter()
.map(|t| t.join().unwrap())
.sum()
}
}

View File

@ -0,0 +1,221 @@
use std::iter::Sum;
use std::slice::{Chunks, ChunksMut};
use crossbeam::thread;
use self::core_affinity::CoreId;
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
pub struct CrossbeamDevice {
pub(crate) ncore: usize,
pub(crate) pin: bool,
pub(crate) core_ids: Vec<CoreId>,
}
impl CrossbeamDevice {
pub fn new(ncore: usize, pin: bool) -> Self {
let mut core_ids = match core_affinity::get_core_ids() {
Some(xs) => xs,
None => {
colour::e_red_ln!("Cannot enumerate cores, pinning will not work if enabled");
(0..ncore).map(|i| CoreId { id: i }).collect()
}
};
core_ids.resize(ncore, core_ids[0]);
CrossbeamDevice { ncore, pin, core_ids }
}
}
impl CrossbeamDevice {
// divide the length by the number of cores, the last core gets less work if it does not divide
fn chunk_size(&self, len: usize) -> usize { (len as f64 / self.ncore as f64).ceil() as usize }
// make a mutable chunk from the vec
fn mk_mut_chunks<'a, T, A: AllocatorType>(&self, xs: &'a mut Vec<T, A>) -> ChunksMut<'a, T> {
let len = xs.len();
xs.chunks_mut(self.chunk_size(len))
}
// make a immutable chunk from the vec
fn mk_chunks<'a, T, A: AllocatorType>(&self, xs: &'a mut Vec<T, A>) -> Chunks<'a, T> {
xs.chunks(self.chunk_size(xs.len()))
}
}
extern crate core_affinity;
// Crossbeam threaded version, it should be semantically equal to the single threaded version
impl<T: ArrayType + Sync + Send + Sum, A: AllocatorType + Sync + Send> RustStream<T>
for StreamData<T, CrossbeamDevice, A>
{
fn init_arrays(&mut self) {
thread::scope(|s| {
let init = self.init;
let pin = self.device.pin;
for (t, ((a, b), c)) in self.device.core_ids.iter().zip(
self
.device
.mk_mut_chunks(&mut self.a)
.zip(self.device.mk_mut_chunks(&mut self.b))
.zip(self.device.mk_mut_chunks(&mut self.c)),
) {
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for x in a.into_iter() {
*x = init.0;
}
for x in b.into_iter() {
*x = init.1;
}
for x in c.into_iter() {
*x = init.2;
}
});
}
})
.unwrap()
}
fn copy(&mut self) {
thread::scope(|s| {
let pin = self.device.pin;
for (t, (c, a)) in self
.device
.core_ids
.iter()
.zip(self.device.mk_mut_chunks(&mut self.c).zip(self.device.mk_chunks(&mut self.a)))
{
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for i in 0..c.len() {
c[i] = a[i];
}
});
}
})
.unwrap()
}
fn mul(&mut self) {
thread::scope(|s| {
let pin = self.device.pin;
let scalar = self.scalar;
for (t, (b, c)) in self
.device
.core_ids
.iter()
.zip(self.device.mk_mut_chunks(&mut self.b).zip(self.device.mk_chunks(&mut self.c)))
{
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for i in 0..b.len() {
b[i] = scalar * c[i];
}
});
}
})
.unwrap()
}
fn add(&mut self) {
thread::scope(|s| {
let pin = self.device.pin;
for (t, (c, (a, b))) in (&mut self.device.core_ids.iter()).zip(
self
.device
.mk_mut_chunks(&mut self.c)
.zip(self.device.mk_chunks(&mut self.a).zip(self.device.mk_chunks(&mut self.b))),
) {
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for i in 0..c.len() {
c[i] = a[i] + b[i];
}
});
}
})
.unwrap()
}
fn triad(&mut self) {
thread::scope(|s| {
let pin = self.device.pin;
let scalar = self.scalar;
for (t, (a, (b, c))) in self.device.core_ids.iter().zip(
self
.device
.mk_mut_chunks(&mut self.a)
.zip(self.device.mk_chunks(&mut self.b).zip(self.device.mk_chunks(&mut self.c))),
) {
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for i in 0..a.len() {
a[i] = b[i] + scalar * c[i]
}
});
}
})
.unwrap()
}
fn nstream(&mut self) {
thread::scope(|s| {
let pin = self.device.pin;
let scalar = self.scalar;
for (t, (a, (b, c))) in self.device.core_ids.iter().zip(
self
.device
.mk_mut_chunks(&mut self.a)
.zip(self.device.mk_chunks(&mut self.b).zip(self.device.mk_chunks(&mut self.c))),
) {
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
for i in 0..a.len() {
a[i] += b[i] + scalar * c[i]
}
});
}
})
.unwrap()
}
fn dot(&mut self) -> T {
let mut partial_sum = vec![T::zero(); self.device.ncore];
thread::scope(|s| {
let pin = self.device.pin;
let a = &self.a;
let b = &self.b;
let chunk_indices = |i: usize| {
let chunk_size = self.device.chunk_size(self.size);
let start = i * chunk_size;
start..((start + chunk_size).min(self.size))
};
for (t, (n, acc)) in self.device.core_ids.iter().zip(partial_sum.iter_mut().enumerate()) {
s.spawn(move |_| {
if pin {
core_affinity::set_for_current(*t);
}
let mut p = T::zero();
for i in chunk_indices(n) {
p += a[i] * b[i];
}
*acc = p;
});
}
})
.unwrap();
partial_sum.into_iter().sum()
}
}

View File

@ -0,0 +1,435 @@
#![feature(allocator_api)]
#![feature(vec_into_raw_parts)]
use std::alloc::System;
use std::env;
use std::fmt::{Debug, Display};
use std::iter::Sum;
use std::mem::size_of;
use std::time::Duration;
use num_traits::abs;
use structopt::StructOpt;
use tabular::{Row, Table};
use crate::arc_stream::ArcDevice;
use crate::crossbeam_stream::CrossbeamDevice;
use crate::plain_stream::SerialDevice;
use crate::rayon_stream::RayonDevice;
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
use crate::unsafe_stream::UnsafeDevice;
mod arc_stream;
mod crossbeam_stream;
mod plain_stream;
mod rayon_stream;
mod stream;
mod unsafe_stream;
#[derive(Debug, StructOpt)]
struct Options {
/// List available devices
#[structopt(long)]
list: bool,
/// Select device at <device>
#[structopt(long, default_value = "0")]
device: usize,
/// Run the test <numtimes> times (NUM >= 2)
#[structopt(long, short = "n", default_value = "100")]
numtimes: usize,
/// Use <arraysize> elements in the array
#[structopt(long, short = "s", default_value = "33554432")]
arraysize: usize,
/// Use floats (rather than doubles)
#[structopt(long)]
float: bool,
/// Only run triad
#[structopt(long)]
triad_only: bool,
/// Only run nstream
#[structopt(long)]
nstream_only: bool,
/// Output as csv table
#[structopt(long)]
csv: bool,
/// Use MiB=2^20 for bandwidth calculation (default MB=10^6)
#[structopt(long)]
mibibytes: bool,
/// Use libc malloc instead of the Rust's allocator for benchmark array allocation
#[structopt(name = "malloc", long)]
malloc: bool,
/// Initialise each benchmark array at allocation time on the main thread
#[structopt(name = "init", long)]
init: bool,
/// Pin threads to distinct cores, this has NO effect in Rayon devices
#[structopt(long)]
pin: bool,
}
#[derive(PartialEq)]
enum Benchmark {
All,
Triad,
NStream,
}
fn check_solution<T: ArrayType + Display + Sum + Into<f64>, D, A: AllocatorType>(
benchmark: Benchmark, numtimes: usize, vec: &StreamData<T, D, A>, dot_sum: Option<T>,
) -> bool {
let (mut gold_a, mut gold_b, mut gold_c) = vec.init;
for _ in 0..numtimes {
match benchmark {
Benchmark::All => {
gold_c = gold_a;
gold_b = vec.scalar * gold_c;
gold_c = gold_a + gold_b;
gold_a = gold_b + vec.scalar * gold_c;
}
Benchmark::Triad => {
gold_a = gold_b + vec.scalar * gold_c;
}
Benchmark::NStream => {
gold_a += gold_b + vec.scalar * gold_c;
}
};
}
let tolerance = T::epsilon().into() * 100.0f64;
let validate_xs = |name: &str, xs: &Vec<T, A>, from: T| {
let error = (xs.iter().map(|x| abs(*x - from)).sum::<T>()).into() / xs.len() as f64;
let fail = error > tolerance;
if fail {
eprintln!("Validation failed on {}[]. Average error {} ", name, error);
}
!fail
};
let a_ok = validate_xs("a", &vec.a, gold_a);
let b_ok = validate_xs("b", &vec.b, gold_b);
let c_ok = validate_xs("c", &vec.c, gold_c);
let dot_ok = dot_sum.map_or(true, |sum| {
let gold_sum = (gold_a * gold_b).into() * vec.size as f64;
let error = abs((sum.into() - gold_sum) / gold_sum);
let fail = error > 1.0e-8;
if fail {
eprintln!(
"Validation failed on sum. Error {} \nSum was {} but should be {}",
error, sum, gold_sum
);
}
!fail
});
a_ok && b_ok && c_ok && dot_ok
}
fn run_cpu<T: ArrayType + Sync + Send + Sum + Into<f64> + Display, D, A: AllocatorType>(
option: &Options, mut stream: StreamData<T, D, A>,
) -> bool
where StreamData<T, D, A>: RustStream<T> {
let benchmark = match (option.nstream_only, option.triad_only) {
(true, false) => Benchmark::NStream,
(false, true) => Benchmark::Triad,
(false, false) => Benchmark::All,
(true, true) => {
panic!("Both triad and nstream are enabled, pick one or omit both to run all benchmarks")
}
};
let array_bytes = option.arraysize * size_of::<T>();
let total_bytes = array_bytes * 3;
let (mega_scale, mega_suffix, giga_scale, giga_suffix) = if !option.mibibytes {
(1.0e-6, "MB", 1.0e-9, "GB")
} else {
(2f64.powi(-20), "MiB", 2f64.powi(-30), "GiB")
};
if !option.csv {
println!(
"Running {} {} times",
match benchmark {
Benchmark::All => "kernels",
Benchmark::Triad => "triad",
Benchmark::NStream => "nstream",
},
option.numtimes
);
if benchmark == Benchmark::Triad {
println!("Number of elements: {}", option.arraysize);
}
println!("Precision: {}", if option.float { "float" } else { "double" });
println!(
"Array size: {:.1} {} (={:.1} {})",
mega_scale * array_bytes as f64,
mega_suffix,
giga_scale * array_bytes as f64,
giga_suffix
);
println!(
"Total size: {:.1} {} (={:.1} {})",
mega_scale * total_bytes as f64,
mega_suffix,
giga_scale * total_bytes as f64,
giga_suffix
);
}
stream.init_arrays();
let tabulate = |xs: &Vec<Duration>, name: &str, t_size: usize| -> Vec<(&str, String)> {
let tail = &xs[1..]; // tail only
// do stats
let max = tail.iter().max().map(|d| d.as_secs_f64());
let min = tail.iter().min().map(|d| d.as_secs_f64());
match (min, max) {
(Some(min), Some(max)) => {
let avg: f64 = tail.iter().map(|d| d.as_secs_f64()).sum::<f64>() / tail.len() as f64;
let mbps = mega_scale * (t_size as f64) / min;
if option.csv {
vec![
("function", name.to_string()),
("num_times", option.numtimes.to_string()),
("n_elements", option.arraysize.to_string()),
("sizeof", t_size.to_string()),
(
if option.mibibytes { "max_mibytes_per_sec" } else { "max_mbytes_per_sec" },
mbps.to_string(),
),
("min_runtime", min.to_string()),
("max_runtime", max.to_string()),
("avg_runtime", avg.to_string()),
]
} else {
vec![
("Function", name.to_string()),
(if option.mibibytes { "MiBytes/sec" } else { "MBytes/sec" }, format!("{:.3}", mbps)),
("Min (sec)", format!("{:.5}", min)),
("Max", format!("{:.5}", max)),
("Average", format!("{:.5}", avg)),
]
}
}
(_, _) => panic!("No min/max element for {}(size={})", name, t_size),
}
};
let tabulate_all = |xs: Vec<Vec<(&str, String)>>| {
match xs.as_slice() {
[head, ..] => {
if option.csv {
println!("{}", head.iter().map(|(col, _)| *col).collect::<Vec<_>>().join(","));
for kvs in xs {
println!("{}", kvs.iter().map(|(_, val)| val.clone()).collect::<Vec<_>>().join(","));
}
} else {
let mut table = Table::new(&vec!["{:<}"; head.len()].join(" "));
table.add_row(head.iter().fold(Row::new(), |row, (col, _)| row.with_cell(col)));
for kvs in xs {
table.add_row(kvs.iter().fold(Row::new(), |row, (_, val)| row.with_cell(val)));
}
print!("{}", table);
}
}
_ => panic!("Empty tabulation"),
};
};
let solutions_correct = match benchmark {
Benchmark::All => {
let (results, sum) = stream.run_all(option.numtimes);
stream.read_arrays();
let correct = check_solution(benchmark, option.numtimes, &stream, Some(sum));
tabulate_all(vec![
tabulate(&results.copy, "Copy", 2 * array_bytes),
tabulate(&results.mul, "Mul", 2 * array_bytes),
tabulate(&results.add, "Add", 3 * array_bytes),
tabulate(&results.triad, "Triad", 3 * array_bytes),
tabulate(&results.dot, "Dot", 2 * array_bytes),
]);
correct
}
Benchmark::NStream => {
let results = stream.run_nstream(option.numtimes);
stream.read_arrays();
let correct = check_solution(benchmark, option.numtimes, &stream, None);
tabulate_all(vec![tabulate(&results, "Nstream", 4 * array_bytes)]);
correct
}
Benchmark::Triad => {
let results = stream.run_triad(option.numtimes);
stream.read_arrays();
let correct = check_solution(benchmark, option.numtimes, &stream, None);
let total_bytes = 3 * array_bytes * option.numtimes;
let bandwidth = giga_scale * (total_bytes as f64 / results.as_secs_f64());
println!("Runtime (seconds): {:.5}", results.as_secs_f64());
println!("Bandwidth ({}/s): {:.3} ", giga_suffix, bandwidth);
correct
}
};
stream.clean_up();
solutions_correct
}
const VERSION: Option<&'static str> = option_env!("CARGO_PKG_VERSION");
static START_A: f32 = 0.1;
static START_B: f32 = 0.2;
static START_C: f32 = 0.0;
static START_SCALAR: f32 = 0.4;
static FLOAT_INIT_SCALAR: f32 = START_SCALAR;
static FLOAT_INIT: (f32, f32, f32) = (START_A, START_B, START_C);
static DOUBLE_INIT_SCALAR: f64 = START_SCALAR as f64;
static DOUBLE_INIT: (f64, f64, f64) = (START_A as f64, START_B as f64, START_C as f64);
pub fn run(args: &Vec<String>) -> bool {
let opt: Options = Options::from_iter(args);
if opt.numtimes < 2 {
panic!("numtimes must be >= 2")
}
let alloc = System;
let alloc_name = if opt.malloc { "libc-malloc" } else { "rust-system" };
fn mk_data<T: ArrayType, D, A: AllocatorType>(
opt: &Options, init: (T, T, T), scalar: T, dev: D, alloc: A,
) -> StreamData<T, D, A> {
StreamData::new_in(opt.arraysize, scalar, init, dev, alloc, opt.malloc, opt.init)
}
let num_thread_key = "BABELSTREAM_NUM_THREADS";
let max_ncores = num_cpus::get();
let ncores = match env::var(num_thread_key) {
Ok(v) => match v.parse::<i64>() {
Err(bad) => {
colour::e_yellow_ln!(
"Cannot parse {} (reason: {}), defaulting to {}",
bad,
num_thread_key,
max_ncores
);
max_ncores
}
Ok(n) if n <= 0 || n > max_ncores as i64 => {
println!("{} out of bound ({}), defaulting to {}", num_thread_key, n, max_ncores);
max_ncores
}
Ok(n) => n as usize,
},
Err(_) => {
println!("{} not set, defaulting to max ({})", num_thread_key, max_ncores);
max_ncores
}
};
let rayon_device = &|| {
let rayon_num_thread_key = "RAYON_NUM_THREADS";
if env::var(rayon_num_thread_key).is_ok() {
colour::e_yellow_ln!("{} is ignored, set {} instead", rayon_num_thread_key, num_thread_key)
}
let dev = RayonDevice {
pool: rayon::ThreadPoolBuilder::default().num_threads(ncores).build().unwrap(),
};
if !opt.csv {
println!("Using {} thread(s), alloc={}", dev.pool.current_num_threads(), alloc_name);
if opt.pin {
colour::e_yellow_ln!("Pinning threads have no effect on Rayon!")
}
}
if opt.float {
run_cpu(&opt, mk_data(&opt, FLOAT_INIT, FLOAT_INIT_SCALAR, dev, alloc))
} else {
run_cpu(&opt, mk_data(&opt, DOUBLE_INIT, DOUBLE_INIT_SCALAR, dev, alloc))
}
};
let arc_device = &|| {
if !opt.csv {
println!("Using {} thread, pin={}, alloc={}", ncores, opt.pin, alloc_name);
}
if opt.float {
let dev = ArcDevice::<f32, _>::new(ncores, opt.pin, alloc);
run_cpu(&opt, mk_data(&opt, FLOAT_INIT, FLOAT_INIT_SCALAR, dev, alloc))
} else {
let dev = ArcDevice::<f64, _>::new(ncores, opt.pin, alloc);
run_cpu(&opt, mk_data(&opt, DOUBLE_INIT, DOUBLE_INIT_SCALAR, dev, alloc))
}
};
let unsafe_device = &|| {
if !opt.csv {
println!("Using {} thread, pin={}, alloc={}", ncores, opt.pin, alloc_name);
}
if opt.float {
let dev = UnsafeDevice::<f32>::new(ncores, opt.pin);
run_cpu(&opt, mk_data(&opt, FLOAT_INIT, FLOAT_INIT_SCALAR, dev, alloc))
} else {
let dev = UnsafeDevice::<f64>::new(ncores, opt.pin);
run_cpu(&opt, mk_data(&opt, DOUBLE_INIT, DOUBLE_INIT_SCALAR, dev, alloc))
}
};
let crossbeam_device = &|| {
let dev = CrossbeamDevice::new(ncores, opt.pin);
if !opt.csv {
println!("Using {} thread(s), pin={}, alloc={}", ncores, opt.pin, alloc_name)
}
if opt.float {
run_cpu(&opt, mk_data(&opt, FLOAT_INIT, FLOAT_INIT_SCALAR, dev, alloc))
} else {
run_cpu(&opt, mk_data(&opt, DOUBLE_INIT, DOUBLE_INIT_SCALAR, dev, alloc))
}
};
let st_device = &|| {
let dev = SerialDevice { pin: opt.pin };
if !opt.csv {
println!("Using 1 thread, pin={}, alloc={}", opt.pin, alloc_name);
}
if opt.float {
run_cpu(&opt, mk_data(&opt, FLOAT_INIT, FLOAT_INIT_SCALAR, dev, alloc))
} else {
run_cpu(&opt, mk_data(&opt, DOUBLE_INIT, DOUBLE_INIT_SCALAR, dev, alloc))
}
};
let devices: Vec<(String, &'_ dyn Fn() -> bool)> = vec![
("CPU (Single threaded)".to_string(), st_device),
("CPU (Rayon)".to_string(), rayon_device),
(format!("CPU (Arc, pinning={})", opt.pin), arc_device),
(format!("CPU (Unsafe, pinning={})", opt.pin), unsafe_device),
(format!("CPU (Crossbeam, pinning={})", opt.pin), crossbeam_device),
];
if opt.list {
devices.iter().enumerate().for_each(|(i, (name, _))| {
println!("[{}] {}", i, name);
});
true
} else {
match devices.get(opt.device) {
Some((name, run)) => {
if !&opt.csv {
println!(
"BabelStream\n\
Version: {}\n\
Implementation: Rust; {}",
VERSION.unwrap_or("unknown"),
name
);
if opt.init {
println!("Initialising arrays on main thread");
}
}
run()
}
None => {
eprintln!("Device index {} not available", opt.device);
false
}
}
}
}

View File

@ -0,0 +1,5 @@
fn main() {
if !rust_stream::run(&std::env::args().collect::<Vec<_>>()) {
std::process::exit(1);
}
}

View File

@ -0,0 +1,61 @@
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
use core_affinity::CoreId;
pub struct SerialDevice {
pub(crate) pin: bool,
}
// single threaded version
impl<T: ArrayType, A: AllocatorType> RustStream<T> for StreamData<T, SerialDevice, A> {
fn init_arrays(&mut self) {
if self.device.pin {
core_affinity::set_for_current(
match core_affinity::get_core_ids().as_ref().map(|x| x.first()) {
Some(Some(x)) => *x,
_ => CoreId { id: 0 },
},
);
}
self.a.fill(self.init.0);
self.b.fill(self.init.1);
self.c.fill(self.init.2);
}
fn copy(&mut self) {
for i in 0..self.size {
self.c[i] = self.a[i];
}
}
fn mul(&mut self) {
for i in 0..self.size {
self.b[i] = self.scalar * self.c[i];
}
}
fn add(&mut self) {
for i in 0..self.size {
self.c[i] = self.a[i] + self.b[i];
}
}
fn triad(&mut self) {
for i in 0..self.size {
self.a[i] = self.b[i] + self.scalar * self.c[i];
}
}
fn nstream(&mut self) {
for i in 0..self.size {
self.a[i] += self.b[i] + self.scalar * self.c[i];
}
}
fn dot(&mut self) -> T {
let mut sum = T::default();
for i in 0..self.size {
sum += self.a[i] * self.b[i];
}
sum
}
}

View File

@ -0,0 +1,77 @@
use std::iter::Sum;
use rayon::prelude::*;
use rayon::ThreadPool;
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
pub struct RayonDevice {
pub(crate) pool: ThreadPool,
}
// Rayon version, it should be semantically equal to the single threaded version
impl<T: ArrayType + Sync + Send + Sum, A: AllocatorType + Sync + Send> RustStream<T>
for StreamData<T, RayonDevice, A>
{
fn init_arrays(&mut self) {
let init = self.init;
self.a.par_iter_mut().for_each(|v| *v = init.0);
self.b.par_iter_mut().for_each(|v| *v = init.1);
self.c.par_iter_mut().for_each(|v| *v = init.2);
}
fn copy(&mut self) {
let a = &self.a;
let c = &mut self.c;
self.device.pool.install(|| {
(*c).par_iter_mut().enumerate().for_each(|(i, c)| *c = a[i]);
});
}
fn mul(&mut self) {
let scalar = self.scalar;
let c = &self.c;
let b = &mut self.b;
self
.device
.pool
.install(|| (*b).par_iter_mut().enumerate().for_each(|(i, b)| *b = scalar * c[i]));
}
fn add(&mut self) {
let a = &self.a;
let b = &self.b;
let c = &mut self.c;
self.device.pool.install(|| (*c).par_iter_mut().enumerate().for_each(|(i, c)| *c = a[i] + b[i]))
}
fn triad(&mut self) {
let scalar = self.scalar;
let a = &mut self.a;
let b = &self.b;
let c = &self.c;
self
.device
.pool
.install(|| (*a).par_iter_mut().enumerate().for_each(|(i, a)| *a = b[i] + scalar * c[i]))
}
fn nstream(&mut self) {
let scalar = self.scalar;
let a = &mut self.a;
let b = &self.b;
let c = &self.c;
self
.device
.pool
.install(|| (*a).par_iter_mut().enumerate().for_each(|(i, a)| *a += b[i] + scalar * c[i]))
}
fn dot(&mut self) -> T {
let a = &self.a;
let b = &self.b;
self.device.pool.install(|| {
(0..self.size).into_par_iter().fold(|| T::default(), |acc, i| acc + a[i] * b[i]).sum::<T>()
})
}
}

View File

@ -0,0 +1,167 @@
use num_traits::real::Real;
use num_traits::{NumAssign, Signed};
use std::alloc::Allocator;
use std::fmt::Debug;
use std::time::{Duration, Instant};
pub trait AllocatorType: Allocator + Copy + Clone + Default + Debug {}
impl<T: Allocator + Copy + Clone + Default + Debug> AllocatorType for T {}
pub struct StreamData<T, D, A: AllocatorType> {
pub device: D,
pub size: usize,
pub scalar: T,
pub init: (T, T, T),
pub a: Vec<T, A>,
pub b: Vec<T, A>,
pub c: Vec<T, A>,
pub needs_dealloc: bool,
}
#[inline(always)]
fn timed<F: FnOnce()>(f: F) -> Duration {
let start = Instant::now();
f();
start.elapsed()
}
#[inline(always)]
fn timed_mut<T, F: FnMut() -> T>(f: &mut F) -> (Duration, T) {
let start = Instant::now();
let x = f();
(start.elapsed(), x)
}
pub struct AllTiming<T> {
pub copy: T,
pub mul: T,
pub add: T,
pub triad: T,
pub dot: T,
}
pub trait ArrayType: Real + NumAssign + Signed + Default + Debug {}
impl<T: Real + NumAssign + Signed + Default + Debug> ArrayType for T {}
impl<T: Default + Clone, D, A: AllocatorType> StreamData<T, D, A> {
pub fn new_in(
size: usize,
scalar: T,
init: (T, T, T),
device: D,
allocator: A,
malloc: bool, //
initialise: bool, //
) -> StreamData<T, D, A> {
let mk_vec = || {
if malloc {
extern crate libc;
use std::mem;
unsafe {
// we do the typical C malloc with a NULL check here
let bytes = mem::size_of::<T>() * size;
let ptr = libc::malloc(bytes as libc::size_t) as *mut T;
if ptr.is_null() {
panic!(
"Cannot allocate {} bytes in `sizeof(T) * size` (T = {}, size = {})",
bytes,
mem::size_of::<T>(),
size
);
}
let mut xs = Vec::from_raw_parts_in(ptr, size, size, allocator);
if initialise {
xs.fill(T::default());
}
xs
}
} else {
if initialise {
let mut xs = Vec::new_in(allocator);
xs.resize(size, T::default());
xs
} else {
// try not to touch the vec after allocation
let mut xs = Vec::with_capacity_in(size, allocator);
unsafe {
xs.set_len(size);
}
xs
}
}
};
StreamData {
device,
size,
scalar,
init,
a: mk_vec(),
b: mk_vec(),
c: mk_vec(),
needs_dealloc: malloc,
}
}
pub fn clean_up(self) {
if self.needs_dealloc {
unsafe {
extern crate libc;
let free_ts = move |xs: Vec<T, A>| {
// make sure we don't call dealloc for vec anymore
// XXX it's important we don't free xs.as_mut_ptr() here and use xs.into_raw_parts_with_alloc()
// as that function handles drops semantic for us
// if we free the the raw ptr directly, the compiler will still drop the vec and then segfault
let (ptr, _, _, _) = xs.into_raw_parts_with_alloc();
libc::free(ptr as *mut libc::c_void);
};
free_ts(self.a);
free_ts(self.b);
free_ts(self.c);
}
}
}
}
pub trait RustStream<T: Default> {
fn init_arrays(&mut self);
fn read_arrays(&mut self) {} // default to no-op as most impl. doesn't need this
fn copy(&mut self);
fn mul(&mut self);
fn add(&mut self);
fn triad(&mut self);
fn nstream(&mut self);
fn dot(&mut self) -> T;
fn run_all(&mut self, n: usize) -> (AllTiming<Vec<Duration>>, T) {
let mut timings: AllTiming<Vec<Duration>> = AllTiming {
copy: vec![Duration::default(); n],
mul: vec![Duration::default(); n],
add: vec![Duration::default(); n],
triad: vec![Duration::default(); n],
dot: vec![Duration::default(); n],
};
let mut last_sum = T::default();
for i in 0..n {
timings.copy[i] = timed(|| self.copy());
timings.mul[i] = timed(|| self.mul());
timings.add[i] = timed(|| self.add());
timings.triad[i] = timed(|| self.triad());
let (dot, sum) = timed_mut(&mut || self.dot());
timings.dot[i] = dot;
last_sum = sum;
}
(timings, last_sum)
}
fn run_triad(&mut self, n: usize) -> Duration {
timed(|| {
for _ in 0..n {
self.triad();
}
})
}
fn run_nstream(&mut self, n: usize) -> Vec<Duration> {
(0..n).map(|_| timed(|| self.nstream())).collect::<Vec<_>>()
}
}

View File

@ -0,0 +1,266 @@
extern crate core_affinity;
use std::alloc::Allocator;
use std::iter::Sum;
use std::ops::Range;
use crate::stream::{AllocatorType, ArrayType, RustStream, StreamData};
use self::core_affinity::CoreId;
#[derive(Debug, Copy, Clone)]
struct UnsafeData<T>(*mut T, usize);
impl<T: ArrayType> UnsafeData<T> {
fn empty() -> UnsafeData<T> { UnsafeData(([] as [T; 0]).as_mut_ptr(), 0) }
fn new<A: Allocator>(xs: &mut Vec<T, A>) -> UnsafeData<T> {
UnsafeData(xs.as_mut_ptr(), xs.len())
}
fn get_slice(&self) -> &mut [T] { unsafe { std::slice::from_raw_parts_mut(self.0, self.1) } }
}
unsafe impl<T> Send for UnsafeData<T> {}
unsafe impl<T> Sync for UnsafeData<T> {}
#[derive(Debug, Copy, Clone)]
struct UnsafeRefs<T> {
a: UnsafeData<T>,
b: UnsafeData<T>,
c: UnsafeData<T>,
}
unsafe impl<T> Send for UnsafeRefs<T> {}
unsafe impl<T> Sync for UnsafeRefs<T> {}
pub struct UnsafeDevice<T: ArrayType> {
pub(crate) ncore: usize,
pub(crate) pin: bool,
pub(crate) core_ids: Vec<CoreId>,
data: UnsafeRefs<T>,
}
impl<T: ArrayType> UnsafeDevice<T> {
pub fn new(ncore: usize, pin: bool) -> Self {
let mut core_ids = match core_affinity::get_core_ids() {
Some(xs) => xs,
None => {
colour::e_red_ln!("Cannot enumerate cores, pinning will not work if enabled");
(0..ncore).map(|i| CoreId { id: i }).collect()
}
};
core_ids.resize(ncore, core_ids[0]);
UnsafeDevice {
ncore,
pin,
core_ids,
data: UnsafeRefs { a: UnsafeData::empty(), b: UnsafeData::empty(), c: UnsafeData::empty() },
}
}
fn thread_ranges(&self, len: usize) -> Vec<(usize, Range<usize>)> {
let chunk = (len as f64 / self.ncore as f64).ceil() as usize;
(0..self.ncore)
.map(|t| {
(t, if t == self.ncore - 1 { (t * chunk)..len } else { (t * chunk)..((t + 1) * chunk) })
})
.collect::<Vec<_>>()
}
}
// Unsafe threaded version, it should be semantically equal to the single threaded version
impl<T: 'static + ArrayType + Sync + Send + Sum, A: AllocatorType + Sync + Send> RustStream<T>
for StreamData<T, UnsafeDevice<T>, A>
{
fn init_arrays(&mut self) {
self.device.data.a = UnsafeData::new(&mut self.a);
self.device.data.b = UnsafeData::new(&mut self.b);
self.device.data.c = UnsafeData::new(&mut self.c);
let init = self.init;
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let b = data.b.get_slice();
let c = data.c.get_slice();
for i in r {
a[i] = init.0;
b[i] = init.1;
c[i] = init.2;
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn copy(&mut self) {
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let c = data.c.get_slice();
for i in r {
c[i] = a[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn mul(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let b = data.b.get_slice();
let c = data.c.get_slice();
for i in r {
b[i] = scalar * c[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn add(&mut self) {
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let b = data.b.get_slice();
let c = data.c.get_slice();
for i in r {
c[i] = a[i] + b[i];
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn triad(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let b = data.b.get_slice();
let c = data.c.get_slice();
for i in r {
a[i] = b[i] + scalar * c[i]
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn nstream(&mut self) {
let scalar = self.scalar;
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let b = data.b.get_slice();
let c = data.c.get_slice();
for i in r {
a[i] += b[i] + scalar * c[i]
}
})
})
.collect::<Vec<_>>()
.into_iter()
.for_each(|t| t.join().unwrap());
}
fn dot(&mut self) -> T {
let pin = self.device.pin;
let data = self.device.data;
self
.device
.thread_ranges(self.size)
.into_iter()
.map(|(t, r)| {
let core = self.device.core_ids[t];
std::thread::spawn(move || {
if pin {
core_affinity::set_for_current(core);
}
let a = data.a.get_slice();
let b = data.b.get_slice();
let mut p = T::default();
for i in r {
p += a[i] * b[i];
}
p
})
})
.collect::<Vec<_>>()
.into_iter()
.map(|t| t.join().unwrap())
.sum()
}
}

View File

@ -0,0 +1,17 @@
use rstest::rstest;
#[rstest]
fn test_main(
#[values(0, 1, 2, 3, 4)] device: usize, //
#[values("", "--pin")] pin: &str, //
#[values("", "--malloc")] malloc: &str, //
#[values("", "--init")] init: &str, //
#[values("", "--triad-only", "--nstream-only")] option: &str, //
) {
let line = format!(
"rust-stream --arraysize 2048 --device {} {} {} {} {}",
device, pin, malloc, init, option
);
let args = line.split_whitespace().map(|s| s.to_string()).collect::<Vec<_>>();
assert!(rust_stream::run(&args));
}