T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add an initial wasi-nn implementation for Wasmtime (#2208)

Browse Source 
* Add an initial wasi-nn implementation for Wasmtime

This change adds a crate, `wasmtime-wasi-nn`, that uses `wiggle` to expose the current state of the wasi-nn API and `openvino` to implement the exposed functions. It includes an end-to-end test demonstrating how to do classification using wasi-nn:
 - `crates/wasi-nn/tests/classification-example` contains Rust code that is compiled to the `wasm32-wasi` target and run with a Wasmtime embedding that exposes the wasi-nn calls
 - the example uses Rust bindings for wasi-nn contained in `crates/wasi-nn/tests/wasi-nn-rust-bindings`; this crate contains code generated by `witx-bindgen` and eventually should be its own standalone crate

* Test wasi-nn as a CI step

This change adds:
 - a GitHub action for installing OpenVINO
 - a script, `ci/run-wasi-nn-example.sh`, to run the classification example
This commit is contained in:
Andrew Brown
2020-11-16 10:54:00 -08:00
committed by GitHub
parent 61a0bcbdc6
commit a61f068c64
33 changed files with 1554 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
.github/actions/install-openvino/README.md vendored Normal file
View File

@@ -0,0 +1,8 @@
# install-openvino
A GitHub action to install OpenVINO from a package repository. This is only necessary for `wasi-nn` support but there
are enough steps here to package the functionality separately and avoid cluttering the CI.
Future improvements:
- make this installer work for different OS/distributions (e.g. https://docs.openvinotoolkit.org/latest/openvino_docs_install_guides_installing_openvino_windows.html)
- it would be nice to output the install directory (i.e. `/opt/intel/openvino`)

14
.github/actions/install-openvino/action.yml vendored Normal file
View File

@@ -0,0 +1,14 @@
name: 'Install OpenVINO'
description: 'Install OpenVINO binaries from a package repository; this is significantly faster than building from source'
inputs:
version:
description: 'The release version of OpenVINO to install'
required: false
default: '2020.4.287'
runs:
using: composite
steps:
- run: ${{ github.action_path }}/install.sh ${{ inputs.version }}
shell: bash

16
.github/actions/install-openvino/install.sh vendored Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/bash
set -e
# Retrieve OpenVINO checksum.
wget https://apt.repos.intel.com/openvino/2020/GPG-PUB-KEY-INTEL-OPENVINO-2020
echo '5f5cff8a2d26ba7de91942bd0540fa4d GPG-PUB-KEY-INTEL-OPENVINO-2020' > CHECKSUM
md5sum --check CHECKSUM
# Add OpenVINO repository (deb).
sudo apt-key add GPG-PUB-KEY-INTEL-OPENVINO-2020
echo "deb https://apt.repos.intel.com/openvino/2020 all main" | sudo tee /etc/apt/sources.list.d/intel-openvino-2020.list
sudo apt update
# Install OpenVINO package.
sudo apt install -y intel-openvino-runtime-ubuntu18-$1

20
.github/workflows/main.yml vendored
View File

@@ -51,6 +51,7 @@ jobs:
runs-on: ubuntu-latest
env:
RUSTDOCFLAGS: -Dbroken_intra_doc_links
OPENVINO_SKIP_LINKING: 1
steps:
- uses: actions/checkout@v2
with:
@@ -252,6 +253,7 @@ jobs:
--all \
--exclude lightbeam \
--exclude wasmtime-lightbeam \
--exclude wasmtime-wasi-nn \
--exclude peepmatic \
--exclude peepmatic-automata \
--exclude peepmatic-fuzzing \
@@ -304,6 +306,23 @@ jobs:
CARGO_VERSION: "+nightly"
RUST_BACKTRACE: 1
# Build and test the wasi-nn module.
test_wasi_nn:
name: Test wasi-nn module
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v2
with:
submodules: true
- uses: ./.github/actions/install-rust
with:
toolchain: nightly-2020-08-25
- run: rustup target add wasm32-wasi
- uses: ./.github/actions/install-openvino
- run: ./ci/run-wasi-nn-example.sh
env:
RUST_BACKTRACE: 1
# Verify that cranelift's code generation is deterministic
meta_determinist_check:
name: Meta deterministic check
@@ -411,6 +430,7 @@ jobs:
--all \
--exclude lightbeam \
--exclude wasmtime-lightbeam \
--exclude wasmtime-wasi-nn \
--exclude peepmatic \
--exclude peepmatic-automata \
--exclude peepmatic-fuzzing \

3
.gitmodules vendored
View File

@@ -7,3 +7,6 @@
[submodule "WASI"]
path = crates/wasi-common/WASI
url = https://github.com/WebAssembly/WASI
[submodule "crates/wasi-nn/spec"]
path = crates/wasi-nn/spec
url = https://github.com/WebAssembly/wasi-nn

127
Cargo.lock generated
View File

@@ -128,6 +128,30 @@ dependencies = [
"serde",
]
[[package]]
name = "bindgen"
version = "0.55.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "75b13ce559e6433d360c26305643803cb52cfbabbc2b9c47ce04a58493dfb443"
dependencies = [
"bitflags",
"cexpr",
"cfg-if 0.1.10",
"clang-sys",
"clap",
"env_logger 0.7.1",
"lazy_static",
"lazycell",
"log",
"peeking_take_while",
"proc-macro2",
"quote",
"regex",
"rustc-hash",
"shlex",
"which",
]
[[package]]
name = "bit-set"
version = "0.5.2"
@@ -210,6 +234,15 @@ dependencies = [
"jobserver",
]
[[package]]
name = "cexpr"
version = "0.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f4aedb84272dbe89af497cf81375129abda4fc0a9e7c5d317498c15cc30c0d27"
dependencies = [
"nom",
]
[[package]]
name = "cfg-if"
version = "0.1.10"
@@ -235,6 +268,17 @@ dependencies = [
"winapi",
]
[[package]]
name = "clang-sys"
version = "1.0.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9da1484c6a890e374ca5086062d4847e0a2c1e5eba9afa5d48c09e8eb39b2519"
dependencies = [
"glob",
"libc",
"libloading",
]
[[package]]
name = "clap"
version = "2.33.3"
@@ -1048,6 +1092,12 @@ version = "1.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e2abad23fbc42b3700f2f279844dc832adb2b2eb069b2df918f455c4e18cc646"
[[package]]
name = "lazycell"
version = "1.2.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b294d6fa9ee409a054354afc4352b0b9ef7ca222c69b8812cbea9e7d2bf3783f"
[[package]]
name = "leb128"
version = "0.2.4"
@@ -1070,6 +1120,16 @@ dependencies = [
"cc",
]
[[package]]
name = "libloading"
version = "0.6.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2443d8f0478b16759158b2f66d525991a05491138bc05814ef52a250148ef4f9"
dependencies = [
"cfg-if 0.1.10",
"winapi",
]
[[package]]
name = "lightbeam"
version = "0.21.0"
@@ -1184,6 +1244,16 @@ version = "0.2.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0debeb9fcf88823ea64d64e4a815ab1643f33127d995978e099942ce38f25238"
[[package]]
name = "nom"
version = "5.1.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ffb4262d26ed83a1c0a33a38fe2bb15797329c85770da05e6b828ddb782627af"
dependencies = [
"memchr",
"version_check",
]
[[package]]
name = "num-integer"
version = "0.1.44"
@@ -1242,6 +1312,26 @@ version = "0.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "624a8340c38c1b80fd549087862da4ba43e08858af025b236e509b6649fc13d5"
[[package]]
name = "openvino"
version = "0.1.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "87a74f90f07f134153e3ad2ffa724a3ebda92cdc6e099f7fe7d9185cf960f028"
dependencies = [
"openvino-sys",
"thiserror",
]
[[package]]
name = "openvino-sys"
version = "0.1.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e72a2e5bd353bd3cf39b2663767e0ae0325a7588c47fd496cbf9a09237ef7ca8"
dependencies = [
"bindgen",
"cmake",
]
[[package]]
name = "os_pipe"
version = "0.9.2"
@@ -1252,6 +1342,12 @@ dependencies = [
"winapi",
]
[[package]]
name = "peeking_take_while"
version = "0.1.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "19b17cddbe7ec3f8bc800887bab5e717348c95ea2ca0b1bf0837fb964dc67099"
[[package]]
name = "peepmatic"
version = "0.68.0"
@@ -1828,6 +1924,12 @@ dependencies = [
"dirs",
]
[[package]]
name = "shlex"
version = "0.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7fdf1b9db47230893d76faad238fd6097fd6d6a9245cd7a4d90dbd639536bbd2"
[[package]]
name = "smallvec"
version = "1.4.2"
@@ -2344,6 +2446,7 @@ dependencies = [
"wasmtime-obj",
"wasmtime-runtime",
"wasmtime-wasi",
"wasmtime-wasi-nn",
"wasmtime-wast",
"wat",
]
@@ -2547,6 +2650,21 @@ dependencies = [
"wiggle",
]
[[package]]
name = "wasmtime-wasi-nn"
version = "0.21.0"
dependencies = [
"anyhow",
"log",
"openvino",
"thiserror",
"wasmtime",
"wasmtime-runtime",
"wasmtime-wasi",
"wasmtime-wiggle",
"wiggle",
]
[[package]]
name = "wasmtime-wast"
version = "0.21.0"
@@ -2604,6 +2722,15 @@ dependencies = [
"wast 27.0.0",
]
[[package]]
name = "which"
version = "3.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d011071ae14a2f6671d0b74080ae0cd8ebf3a6f8c9589a2cd45f23126fe29724"
dependencies = [
"libc",
]
[[package]]
name = "wig"
version = "0.21.0"

2
Cargo.toml
View File

@@ -30,6 +30,7 @@ wasmtime-jit = { path = "crates/jit", version = "0.21.0" }
wasmtime-obj = { path = "crates/obj", version = "0.21.0" }
wasmtime-wast = { path = "crates/wast", version = "0.21.0" }
wasmtime-wasi = { path = "crates/wasi", version = "0.21.0" }
wasmtime-wasi-nn = { path = "crates/wasi-nn", version = "0.21.0", optional = true }
wasi-common = { path = "crates/wasi-common", version = "0.21.0" }
structopt = { version = "0.3.5", features = ["color", "suggestions"] }
object = { version = "0.22.0", default-features = false, features = ["write"] }
@@ -80,6 +81,7 @@ default = ["jitdump", "wasmtime/wat", "wasmtime/parallel-compilation"]
lightbeam = ["wasmtime/lightbeam"]
jitdump = ["wasmtime/jitdump"]
vtune = ["wasmtime/vtune"]
wasi-nn = ["wasmtime-wasi-nn"]
# Try the experimental, work-in-progress new x86_64 backend. This is not stable
# as of June 2020.

1
ci/run-experimental-x64-ci.sh
View File

@@ -13,6 +13,7 @@ cargo $CARGO_VERSION \
--features experimental_x64 \
--all \
--exclude wasmtime-lightbeam \
--exclude wasmtime-wasi-nn \
--exclude peepmatic \
--exclude peepmatic-automata \
--exclude peepmatic-fuzzing \

35
ci/run-wasi-nn-example.sh Executable file
View File

@@ -0,0 +1,35 @@
#!/bin/bash
# The following script demonstrates how to execute a machine learning inference using the wasi-nn module optionally
# compiled into Wasmtime. Calling it will download the necessary model and tensor files stored separately in $FIXTURE
# into $TMP_DIR (optionally pass a directory with existing files as the first argument to re-try the script). Then,
# it will compile the example code in crates/wasi-nn/tests/example into a Wasm file that is subsequently
# executed with the Wasmtime CLI.
set -e
WASMTIME_DIR=$(dirname "$0" | xargs dirname)
FIXTURE=https://gist.github.com/abrown/c7847bf3701f9efbb2070da1878542c1/raw/07a9f163994b0ff8f0d7c5a5c9645ec3d8b24024
# Inform the environment of OpenVINO library locations. Then we use OPENVINO_INSTALL_DIR below to avoid building all of
# OpenVINO from source (quite slow).
source /opt/intel/openvino/bin/setupvars.sh
# Build Wasmtime with wasi-nn enabled; we attempt this first to avoid extra work if the build fails.
OPENVINO_INSTALL_DIR=/opt/intel/openvino cargo build -p wasmtime-cli --features wasi-nn
# Download all necessary test fixtures to the temporary directory.
TMP_DIR=${1:-$(mktemp -d -t ci-XXXXXXXXXX)}
wget --no-clobber --directory-prefix=$TMP_DIR $FIXTURE/frozen_inference_graph.bin
wget --no-clobber --directory-prefix=$TMP_DIR $FIXTURE/frozen_inference_graph.xml
wget --no-clobber --directory-prefix=$TMP_DIR $FIXTURE/tensor-1x3x300x300-f32.bgr
# Now build an example that uses the wasi-nn API.
pushd $WASMTIME_DIR/crates/wasi-nn/examples/classification-example
cargo build --release --target=wasm32-wasi
cp target/wasm32-wasi/release/wasi-nn-example.wasm $TMP_DIR
popd
# Run the example in Wasmtime (note that the example uses `fixture` as the expected location of the model/tensor files).
OPENVINO_INSTALL_DIR=/opt/intel/openvino cargo run --features wasi-nn -- run --mapdir fixture::$TMP_DIR $TMP_DIR/wasi-nn-example.wasm
# Clean up.
rm -rf $TMP_DIR

29
crates/wasi-nn/Cargo.toml Normal file
View File

@@ -0,0 +1,29 @@
[package]
name = "wasmtime-wasi-nn"
version = "0.21.0"
authors = ["The Wasmtime Project Developers"]
description = "Wasmtime implementation of the wasi-nn API"
documentation = "https://docs.rs/wasmtime-wasi-nn"
license = "Apache-2.0 WITH LLVM-exception"
categories = ["wasm", "computer-vision"]
keywords = ["webassembly", "wasm", "neural network"]
repository = "https://github.com/bytecodealliance/wasmtime"
readme = "README.md"
edition = "2018"
[dependencies]
# These dependencies are necessary for the witx-generation macros to work:
anyhow = "1.0"
log = { version = "0.4", default-features = false }
wasmtime = { path = "../wasmtime", version = "0.21.0", default-features = false }
wasmtime-runtime = { path = "../runtime", version = "0.21.0" }
wasmtime-wiggle = { path = "../wiggle/wasmtime", version = "0.21.0" }
wasmtime-wasi = { path = "../wasi", version = "0.21.0" }
wiggle = { path = "../wiggle", version = "0.21.0" }
# These dependencies are necessary for the wasi-nn implementation:
openvino = "0.1.5"
thiserror = "1.0"
[badges]
maintenance = { status = "experimental" }

220
crates/wasi-nn/LICENSE Normal file
View File

@@ -0,0 +1,220 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
--- LLVM Exceptions to the Apache 2.0 License ----
As an exception, if, as a result of your compiling your source code, portions
of this Software are embedded into an Object form of such source code, you
may redistribute such embedded portions in such Object form without complying
with the conditions of Sections 4(a), 4(b) and 4(d) of the License.
In addition, if you combine or link compiled forms of this Software with
software that is licensed under the GPLv2 ("Combined Software") and if a
court of competent jurisdiction determines that the patent provision (Section
3), the indemnity provision (Section 9) or other Section of the License
conflicts with the conditions of the GPLv2, you may retroactively and
prospectively choose to deem waived or otherwise exclude such Section(s) of
the License, but only in their entirety and only with respect to the Combined
Software.

38
crates/wasi-nn/README.md Normal file
View File

@@ -0,0 +1,38 @@
# wasmtime-wasi-nn
This crate enables support for the [wasi-nn] API in Wasmtime. Currently it contains an implementation of [wasi-nn] using
OpenVINO™ but in the future it could support multiple machine learning backends. Since the [wasi-nn] API is expected
to be an optional feature of WASI, this crate is currently separate from the [wasi-common] crate. This crate is
experimental and its API, functionality, and location could quickly change.
[examples]: examples
[openvino]: https://crates.io/crates/openvino
[wasi-nn]: https://github.com/WebAssembly/wasi-nn
[wasi-common]: ../wasi-common
### Use
Use the Wasmtime APIs to instantiate a Wasm module and link in the `WasiNn` implementation as follows:
```
let wasi_nn = WasiNn::new(&store, WasiNnCtx::new()?);
wasi_nn.add_to_linker(&mut linker)?;
```
### Build
This crate should build as usual (i.e. `cargo build`) but note that using an existing installation of OpenVINO™, rather
than building from source, will drastically improve the build times. See the [openvino] crate for more information
### Example
An end-to-end example demonstrating ML classification is included in [examples]:
- `tests/wasi-nn-rust-bindings` contains ergonomic bindings for writing Rust code against the [wasi-nn] APIs
- `tests/classification-example` contains a standalone Rust project that uses the [wasi-nn] APIs and is compiled to the
`wasm32-wasi` target using the `wasi-nn-rust-bindings`
Run the example from the Wasmtime project directory:
```
ci/run-wasi-nn-example.sh
```

10
crates/wasi-nn/build.rs Normal file
View File

@@ -0,0 +1,10 @@
//! This build script:
//! - has the configuration necessary for the wiggle and witx macros.
use std::path::PathBuf;
fn main() {
// This is necessary for Wiggle/Witx macros.
let wasi_root = PathBuf::from("./spec").canonicalize().unwrap();
println!("cargo:rustc-env=WASI_ROOT={}", wasi_root.display());
}

1
crates/wasi-nn/examples/README.md Normal file
View File

@@ -0,0 +1 @@
See `ci/run-wasi-nn-example.sh` for how the classification example is tested during CI.

13
crates/wasi-nn/examples/classification-example/Cargo.lock generated Normal file
View File

@@ -0,0 +1,13 @@
# This file is automatically @generated by Cargo.
# It is not intended for manual editing.
[[package]]
name = "wasi-nn"
version = "0.1.0"
[[package]]
name = "wasi-nn-example"
version = "0.19.0"
dependencies = [
"wasi-nn 0.1.0",
]

15
crates/wasi-nn/examples/classification-example/Cargo.toml Normal file
View File

@@ -0,0 +1,15 @@
[package]
name = "wasi-nn-example"
version = "0.19.0"
authors = ["The Wasmtime Project Developers"]
readme = "README.md"
edition = "2018"
publish = false
[dependencies]
wasi-nn = { path = "../wasi-nn-rust-bindings", version = "0.1.0" }
# This crate is built with the wasm32-wasi target, so it's separate
# from the main Wasmtime build, so use this directive to exclude it
# from the parent directory's workspace.
[workspace]

2
crates/wasi-nn/examples/classification-example/README.md Normal file
View File

@@ -0,0 +1,2 @@
This example project demonstrates using the `wasi-nn` API to perform ML inference. It consists of Rust code that is
built using the `wasm32-wasi` target. See `ci/run-wasi-nn-example.sh` for how this is used.

54
crates/wasi-nn/examples/classification-example/src/main.rs Normal file
View File

@@ -0,0 +1,54 @@
use std::convert::TryInto;
use std::fs;
use wasi_nn;
pub fn main() {
let xml = fs::read_to_string("fixture/frozen_inference_graph.xml").unwrap();
println!("First 50 characters of graph: {}", &xml[..50]);
let weights = fs::read("fixture/frozen_inference_graph.bin").unwrap();
println!("Size of weights: {}", weights.len());
let graph = unsafe {
wasi_nn::load(
&[&xml.into_bytes(), &weights],
wasi_nn::GRAPH_ENCODING_OPENVINO,
wasi_nn::EXECUTION_TARGET_CPU,
)
.unwrap()
};
println!("Graph handle ID: {}", graph);
let context = unsafe { wasi_nn::init_execution_context(graph).unwrap() };
println!("Execution context ID: {}", context);
// Load a tensor that precisely matches the graph input tensor (see
// `fixture/frozen_inference_graph.xml`).
let tensor_data = fs::read("fixture/tensor-1x3x300x300-f32.bgr").unwrap();
println!("Tensor bytes: {}", tensor_data.len());
let tensor = wasi_nn::Tensor {
dimensions: &[1, 3, 300, 300],
r#type: wasi_nn::TENSOR_TYPE_F32,
data: &tensor_data,
};
unsafe {
wasi_nn::set_input(context, 0, tensor).unwrap();
}
// Execute the inference.
unsafe {
wasi_nn::compute(context).unwrap();
}
// Retrieve the output (TODO output looks incorrect).
let mut output_buffer = vec![0f32; 1 << 20];
unsafe {
wasi_nn::get_output(
context,
0,
&mut output_buffer[..] as *mut [f32] as *mut u8,
(output_buffer.len() * 4).try_into().unwrap(),
);
}
println!("output tensor: {:?}", &output_buffer[..1000])
}

1
crates/wasi-nn/examples/wasi-nn-rust-bindings/.gitignore vendored Normal file
View File

@@ -0,0 +1 @@
/target

5
crates/wasi-nn/examples/wasi-nn-rust-bindings/Cargo.lock generated Normal file
View File

@@ -0,0 +1,5 @@
# This file is automatically @generated by Cargo.
# It is not intended for manual editing.
[[package]]
name = "wasi-nn"
version = "0.1.0"

14
crates/wasi-nn/examples/wasi-nn-rust-bindings/Cargo.toml Normal file
View File

@@ -0,0 +1,14 @@
[package]
name = "wasi-nn"
version = "0.1.0"
authors = ["The Wasmtime Project Developers"]
readme = "README.md"
edition = "2018"
publish = false
[dependencies]
# This crate is only used when building the example, so it's separate
# from the main Wasmtime build, so use this directive to exclude it
# from the parent directory's workspace.
[workspace]

201
crates/wasi-nn/examples/wasi-nn-rust-bindings/LICENSE Normal file
View File

@@ -0,0 +1,201 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

65
crates/wasi-nn/examples/wasi-nn-rust-bindings/README.md Normal file
View File

@@ -0,0 +1,65 @@
wasi-nn Rust Bindings
=====================
This crate contains API bindings for [wasi-nn] system calls in Rust. It is similar in purpose to the [wasi bindings] but
this crate provides access to the optional neural network functionality from WebAssembly.
[wasi-nn]: https://github.com/WebAssembly/wasi-nn
[wasi bindings]: https://github.com/bytecodealliance/wasi
> __NOTE__: These bindings are experimental (use at your own risk) and subject to upstream changes in the wasi-nn
> specification.
> __NOTE__: In the future this crate may be (should be) moved to its own repository, like the [wasi bindings].
### Use
Depend on this crate in your `Cargo.toml`:
```toml
[dependencies]
wasi-nn = "0.1.0"
```
Use the wasi-nn APIs in your application:
```rust
use wasi_nn;
unsafe {
wasi_nn::load(
&[&xml.into_bytes(), &weights],
wasi_nn::GRAPH_ENCODING_OPENVINO,
wasi_nn::EXECUTION_TARGET_CPU,
)
.unwrap()
}
```
Compile the application to WebAssembly:
```shell script
cargo build --target=wasm32-wasi
```
Run the generated Wasm in a runtime supporting wasi-nn. Currently Wasmtime has experimental support using the Wasmtime
APIs; see [main.rs](../main.rs) for an example of how this is accomplished.
### Generation
This crate contains code ([`src/generated.rs`](src/generated.rs)) generated by
[`witx-bindgen`](https://github.com/bytecodealliance/wasi/tree/main/crates/witx-bindgen).
To regenerate this code, run `witx-bindgen` against the [`wasi-nn` WITX file](https://github.com/WebAssembly/wasi-nn/blob/master/phases/ephemeral/witx/wasi_ephemeral_nn.witx):
```shell script
.../crates/witx-bindgen$ cargo run .../wasi-nn/phases/ephemeral/witx/wasi_ephemeral_nn.witx
```
### License
This project is licensed under the Apache 2.0 license. See [LICENSE](LICENSE) for more details.
### Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this project by you,
as defined in the Apache-2.0 license, shall be licensed as above, without any additional terms or conditions.

76
crates/wasi-nn/examples/wasi-nn-rust-bindings/src/error.rs Normal file
View File

@@ -0,0 +1,76 @@
use super::NnErrno;
use core::fmt;
use core::num::NonZeroU16;
/// A raw error returned by wasi-nn APIs, internally containing a 16-bit error
/// code.
#[derive(Copy, Clone, PartialEq, Eq, Ord, PartialOrd)]
pub struct Error {
code: NonZeroU16,
}
impl Error {
/// Constructs a new error from a raw error code, returning `None` if the
/// error code is zero (which means success).
pub fn from_raw_error(error: NnErrno) -> Option<Error> {
Some(Error {
code: NonZeroU16::new(error)?,
})
}
/// Returns the raw error code that this error represents.
pub fn raw_error(&self) -> u16 {
self.code.get()
}
}
impl fmt::Display for Error {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
write!(f, "{} (error {})", strerror(self.code.get()), self.code)?;
Ok(())
}
}
impl fmt::Debug for Error {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("Error")
.field("code", &self.code)
.field("message", &strerror(self.code.get()))
.finish()
}
}
/// This should be generated automatically by witx-bindgen but is not yet for enums other than
/// `Errno` (this API uses `NnErrno` to avoid naming conflicts). TODO: https://github.com/bytecodealliance/wasi/issues/52.
fn strerror(code: u16) -> &'static str {
match code {
super::NN_ERRNO_SUCCESS => "No error occurred.",
super::NN_ERRNO_INVALID_ARGUMENT => "Caller module passed an invalid argument.",
super::NN_ERRNO_MISSING_MEMORY => "Caller module is missing a memory export.",
super::NN_ERRNO_BUSY => "Device or resource busy.",
_ => "Unknown error.",
}
}
#[cfg(feature = "std")]
extern crate std;
#[cfg(feature = "std")]
impl std::error::Error for Error {}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn error_from_success_code() {
assert_eq!(None, Error::from_raw_error(0));
}
#[test]
fn error_from_invalid_argument_code() {
assert_eq!(
"Caller module passed an invalid argument. (error 1)",
Error::from_raw_error(1).unwrap().to_string()
);
}
}

199
crates/wasi-nn/examples/wasi-nn-rust-bindings/src/generated.rs Normal file
View File

@@ -0,0 +1,199 @@
// This file is automatically generated, DO NOT EDIT
//
// To regenerate this file run the `crates/witx-bindgen` command
use core::mem::MaybeUninit;
pub use crate::error::Error;
pub type Result<T, E = Error> = core::result::Result<T, E>;
pub type BufferSize = u32;
pub type NnErrno = u16;
/// No error occurred.
pub const NN_ERRNO_SUCCESS: NnErrno = 0;
/// Caller module passed an invalid argument.
pub const NN_ERRNO_INVALID_ARGUMENT: NnErrno = 1;
/// Caller module is missing a memory export.
pub const NN_ERRNO_MISSING_MEMORY: NnErrno = 2;
/// Device or resource busy.
pub const NN_ERRNO_BUSY: NnErrno = 3;
pub type TensorDimensions<'a> = &'a [u32];
pub type TensorType = u8;
pub const TENSOR_TYPE_F16: TensorType = 0;
pub const TENSOR_TYPE_F32: TensorType = 1;
pub const TENSOR_TYPE_U8: TensorType = 2;
pub const TENSOR_TYPE_I32: TensorType = 3;
pub type TensorData<'a> = &'a [u8];
#[repr(C)]
#[derive(Copy, Clone, Debug)]
pub struct Tensor<'a> {
/// Describe the size of the tensor (e.g. 2x2x2x2 -> [2, 2, 2, 2]). To represent a tensor containing a single value,
/// use `[1]` for the tensor dimensions.
pub dimensions: TensorDimensions<'a>,
pub r#type: TensorType,
/// Contains the tensor data.
pub data: TensorData<'a>,
}
pub type GraphBuilder<'a> = &'a [u8];
pub type GraphBuilderArray<'a> = &'a [GraphBuilder<'a>];
pub type Graph = u32;
pub type GraphEncoding = u8;
/// TODO document buffer order
pub const GRAPH_ENCODING_OPENVINO: GraphEncoding = 0;
pub type ExecutionTarget = u8;
pub const EXECUTION_TARGET_CPU: ExecutionTarget = 0;
pub const EXECUTION_TARGET_GPU: ExecutionTarget = 1;
pub const EXECUTION_TARGET_TPU: ExecutionTarget = 2;
pub type GraphExecutionContext = u32;
/// Load an opaque sequence of bytes to use for inference.
///
/// This allows runtime implementations to support multiple graph encoding formats. For unsupported graph encodings,
/// return `errno::inval`.
///
/// ## Parameters
///
/// * `builder` - The bytes necessary to build the graph.
/// * `encoding` - The encoding of the graph.
/// * `target` - Where to execute the graph.
pub unsafe fn load(
builder: GraphBuilderArray,
encoding: GraphEncoding,
target: ExecutionTarget,
) -> Result<Graph> {
let mut graph = MaybeUninit::uninit();
let rc = wasi_ephemeral_nn::load(
builder.as_ptr(),
builder.len(),
encoding,
target,
graph.as_mut_ptr(),
);
if let Some(err) = Error::from_raw_error(rc) {
Err(err)
} else {
Ok(graph.assume_init())
}
}
/// TODO Functions like `describe_graph_inputs` and `describe_graph_outputs` (returning
/// an array of `$tensor_description`s) might be useful for introspecting the graph but are not yet included here.
/// Create an execution instance of a loaded graph.
/// TODO this may need to accept flags that might affect the compilation or execution of the graph.
pub unsafe fn init_execution_context(graph: Graph) -> Result<GraphExecutionContext> {
let mut context = MaybeUninit::uninit();
let rc = wasi_ephemeral_nn::init_execution_context(graph, context.as_mut_ptr());
if let Some(err) = Error::from_raw_error(rc) {
Err(err)
} else {
Ok(context.assume_init())
}
}
/// Define the inputs to use for inference.
///
/// This should return an $nn_errno (TODO define) if the input tensor does not match the expected dimensions and type.
///
/// ## Parameters
///
/// * `index` - The index of the input to change.
/// * `tensor` - The tensor to set as the input.
pub unsafe fn set_input(context: GraphExecutionContext, index: u32, tensor: Tensor) -> Result<()> {
let rc = wasi_ephemeral_nn::set_input(context, index, &tensor as *const _ as *mut _);
if let Some(err) = Error::from_raw_error(rc) {
Err(err)
} else {
Ok(())
}
}
/// Extract the outputs after inference.
///
/// This should return an $nn_errno (TODO define) if the inference has not yet run.
///
/// ## Parameters
///
/// * `index` - The index of the output to retrieve.
/// * `out_buffer` - An out parameter to which to copy the tensor data. The caller is responsible for allocating enough memory for
/// the tensor data or an error will be returned. Currently there is no dynamic way to extract the additional
/// tensor metadata (i.e. dimension, element type) but this should be added at some point.
///
/// ## Return
///
/// * `bytes_written` - The number of bytes of tensor data written to the `$out_buffer`.
pub unsafe fn get_output(
context: GraphExecutionContext,
index: u32,
out_buffer: *mut u8,
out_buffer_max_size: BufferSize,
) -> Result<BufferSize> {
let mut bytes_written = MaybeUninit::uninit();
let rc = wasi_ephemeral_nn::get_output(
context,
index,
out_buffer,
out_buffer_max_size,
bytes_written.as_mut_ptr(),
);
if let Some(err) = Error::from_raw_error(rc) {
Err(err)
} else {
Ok(bytes_written.assume_init())
}
}
/// Compute the inference on the given inputs (see `set_input`).
///
/// This should return an $nn_errno (TODO define) if the inputs are not all defined.
pub unsafe fn compute(context: GraphExecutionContext) -> Result<()> {
let rc = wasi_ephemeral_nn::compute(context);
if let Some(err) = Error::from_raw_error(rc) {
Err(err)
} else {
Ok(())
}
}
pub mod wasi_ephemeral_nn {
use super::*;
#[link(wasm_import_module = "wasi_ephemeral_nn")]
extern "C" {
/// Load an opaque sequence of bytes to use for inference.
///
/// This allows runtime implementations to support multiple graph encoding formats. For unsupported graph encodings,
/// return `errno::inval`.
pub fn load(
builder_ptr: *const GraphBuilder,
builder_len: usize,
encoding: GraphEncoding,
target: ExecutionTarget,
graph: *mut Graph,
) -> NnErrno;
/// TODO Functions like `describe_graph_inputs` and `describe_graph_outputs` (returning
/// an array of `$tensor_description`s) might be useful for introspecting the graph but are not yet included here.
/// Create an execution instance of a loaded graph.
/// TODO this may need to accept flags that might affect the compilation or execution of the graph.
pub fn init_execution_context(graph: Graph, context: *mut GraphExecutionContext)
-> NnErrno;
/// Define the inputs to use for inference.
///
/// This should return an $nn_errno (TODO define) if the input tensor does not match the expected dimensions and type.
pub fn set_input(
context: GraphExecutionContext,
index: u32,
tensor: *mut Tensor,
) -> NnErrno;
/// Extract the outputs after inference.
///
/// This should return an $nn_errno (TODO define) if the inference has not yet run.
pub fn get_output(
context: GraphExecutionContext,
index: u32,
out_buffer: *mut u8,
out_buffer_max_size: BufferSize,
bytes_written: *mut BufferSize,
) -> NnErrno;
/// Compute the inference on the given inputs (see `set_input`).
///
/// This should return an $nn_errno (TODO define) if the inputs are not all defined.
pub fn compute(context: GraphExecutionContext) -> NnErrno;
}
}

3
crates/wasi-nn/examples/wasi-nn-rust-bindings/src/lib.rs Normal file
View File

@@ -0,0 +1,3 @@
mod error;
mod generated;
pub use generated::*;

1
crates/wasi-nn/spec Submodule

Submodule crates/wasi-nn/spec added at 68e73cf612

125
crates/wasi-nn/src/ctx.rs Normal file
View File

@@ -0,0 +1,125 @@
//! Implements the base structure (i.e. [WasiNnCtx]) that will provide the implementation of the
//! wasi-nn API.
use crate::r#impl::UsageError;
use crate::witx::types::{Graph, GraphExecutionContext};
use openvino::InferenceError;
use std::cell::RefCell;
use std::collections::HashMap;
use std::hash::Hash;
use thiserror::Error;
use wiggle::GuestError;
/// Possible errors for interacting with [WasiNnCtx].
#[derive(Debug, Error)]
pub enum WasiNnError {
#[error("guest error")]
GuestError(#[from] GuestError),
#[error("openvino error")]
OpenvinoError(#[from] InferenceError),
#[error("usage error")]
UsageError(#[from] UsageError),
}
pub(crate) type WasiNnResult<T> = std::result::Result<T, WasiNnError>;
pub struct Table<K, V> {
entries: HashMap<K, V>,
next_key: u32,
}
impl<K, V> Default for Table<K, V> {
fn default() -> Self {
Self {
entries: HashMap::new(),
next_key: 0,
}
}
}
impl<K, V> Table<K, V>
where
K: Eq + Hash + From<u32> + Copy,
{
pub fn insert(&mut self, value: V) -> K {
let key = self.use_next_key();
self.entries.insert(key, value);
key
}
pub fn remove(&mut self, key: K) -> Option<V> {
self.entries.remove(&key)
}
pub fn get(&self, key: K) -> Option<&V> {
self.entries.get(&key)
}
pub fn get_mut(&mut self, key: K) -> Option<&mut V> {
self.entries.get_mut(&key)
}
pub fn len(&self) -> usize {
self.entries.len()
}
fn use_next_key(&mut self) -> K {
let current = self.next_key;
self.next_key += 1;
K::from(current)
}
}
pub struct ExecutionContext {
pub(crate) graph: Graph,
pub(crate) request: openvino::InferRequest,
}
impl ExecutionContext {
pub(crate) fn new(graph: Graph, request: openvino::InferRequest) -> Self {
Self { graph, request }
}
}
/// Capture the state necessary for calling into `openvino`.
pub struct Ctx {
pub(crate) core: openvino::Core,
pub(crate) graphs: Table<Graph, (openvino::CNNNetwork, openvino::ExecutableNetwork)>,
pub(crate) executions: Table<GraphExecutionContext, ExecutionContext>,
}
impl Ctx {
/// Make a new `WasiNnCtx` with the default settings.
pub fn new() -> WasiNnResult<Self> {
Ok(Self {
core: openvino::Core::new(None)?,
graphs: Table::default(),
executions: Table::default(),
})
}
}
/// This structure provides the Rust-side context necessary for implementing the wasi-nn API. At the
/// moment, it is specialized for a single inference implementation (i.e. OpenVINO) but conceivably
/// this could support more than one backing implementation.
pub struct WasiNnCtx {
pub(crate) ctx: RefCell<Ctx>,
}
impl WasiNnCtx {
/// Make a new `WasiNnCtx` with the default settings.
pub fn new() -> WasiNnResult<Self> {
Ok(Self {
ctx: RefCell::new(Ctx::new()?),
})
}
}
#[cfg(test)]
mod test {
use super::*;
#[test]
fn instantiate() {
WasiNnCtx::new().unwrap();
}
}

176
crates/wasi-nn/src/impl.rs Normal file
View File

@@ -0,0 +1,176 @@
//! Implements the wasi-nn API.
use crate::ctx::{ExecutionContext, WasiNnResult as Result};
use crate::witx::types::{
ExecutionTarget, Graph, GraphBuilderArray, GraphEncoding, GraphExecutionContext, Tensor,
TensorType,
};
use crate::witx::wasi_ephemeral_nn::WasiEphemeralNn;
use crate::WasiNnCtx;
use openvino::{Layout, Precision, TensorDesc};
use thiserror::Error;
use wiggle::GuestPtr;
#[derive(Debug, Error)]
pub enum UsageError {
#[error("Only OpenVINO's IR is currently supported, passed encoding: {0}")]
InvalidEncoding(GraphEncoding),
#[error("OpenVINO expects only two buffers (i.e. [ir, weights]), passed: {0}")]
InvalidNumberOfBuilders(u32),
#[error("Invalid graph handle; has it been loaded?")]
InvalidGraphHandle,
#[error("Invalid execution context handle; has it been initialized?")]
InvalidExecutionContextHandle,
#[error("Not enough memory to copy tensor data of size: {0}")]
NotEnoughMemory(u32),
}
impl<'a> WasiEphemeralNn for WasiNnCtx {
fn load<'b>(
&self,
builders: &GraphBuilderArray<'_>,
encoding: GraphEncoding,
target: ExecutionTarget,
) -> Result<Graph> {
if encoding != GraphEncoding::Openvino {
return Err(UsageError::InvalidEncoding(encoding).into());
}
if builders.len() != 2 {
return Err(UsageError::InvalidNumberOfBuilders(builders.len()).into());
}
let builders = builders.as_ptr();
let xml = builders.read()?.as_slice()?;
let weights = builders.add(1)?.read()?.as_slice()?;
let graph = self
.ctx
.borrow_mut()
.core
.read_network_from_buffer(&xml, &weights)?;
let executable_graph = self
.ctx
.borrow_mut()
.core
.load_network(&graph, map_execution_target_to_string(target))?;
let id = self
.ctx
.borrow_mut()
.graphs
.insert((graph, executable_graph));
Ok(id)
}
fn init_execution_context(&self, graph: Graph) -> Result<GraphExecutionContext> {
let request =
if let Some((_, executable_graph)) = self.ctx.borrow_mut().graphs.get_mut(graph) {
executable_graph.create_infer_request()?
} else {
return Err(UsageError::InvalidGraphHandle.into());
};
let execution_context = ExecutionContext::new(graph, request);
let handle = self.ctx.borrow_mut().executions.insert(execution_context);
Ok(handle)
}
fn set_input<'b>(
&self,
context: GraphExecutionContext,
index: u32,
tensor: &Tensor<'b>,
) -> Result<()> {
let graph = if let Some(execution) = self.ctx.borrow_mut().executions.get_mut(context) {
execution.graph
} else {
return Err(UsageError::InvalidExecutionContextHandle.into());
};
let input_name = if let Some((graph, _)) = self.ctx.borrow().graphs.get(graph) {
graph.get_input_name(index as usize)?
} else {
unreachable!("It should be impossible to attempt to access an execution's graph and for that graph not to exist--this is a bug.")
};
// Construct the blob structure.
let dimensions = tensor
.dimensions
.as_slice()?
.iter()
.map(|d| *d as u64)
.collect::<Vec<_>>();
let precision = match tensor.type_ {
TensorType::F16 => Precision::FP16,
TensorType::F32 => Precision::FP32,
TensorType::U8 => Precision::U8,
TensorType::I32 => Precision::I32,
};
// TODO There must be some good way to discover the layout here; this should not have to default to NHWC.
let desc = TensorDesc::new(Layout::NHWC, &dimensions, precision);
let data = tensor.data.as_slice()?;
let blob = openvino::Blob::new(desc, &data)?;
// Actually assign the blob to the request (TODO avoid duplication with the borrow above).
if let Some(execution) = self.ctx.borrow_mut().executions.get_mut(context) {
execution.request.set_blob(&input_name, blob)?;
} else {
return Err(UsageError::InvalidExecutionContextHandle.into());
}
Ok(())
}
fn compute(&self, context: GraphExecutionContext) -> Result<()> {
if let Some(execution) = self.ctx.borrow_mut().executions.get_mut(context) {
Ok(execution.request.infer()?)
} else {
return Err(UsageError::InvalidExecutionContextHandle.into());
}
}
fn get_output<'b>(
&self,
context: GraphExecutionContext,
index: u32,
out_buffer: &GuestPtr<'_, u8>,
out_buffer_max_size: u32,
) -> Result<u32> {
let graph = if let Some(execution) = self.ctx.borrow_mut().executions.get_mut(context) {
execution.graph
} else {
return Err(UsageError::InvalidExecutionContextHandle.into());
};
let output_name = if let Some((graph, _)) = self.ctx.borrow().graphs.get(graph) {
graph.get_output_name(index as usize)?
} else {
unreachable!("It should be impossible to attempt to access an execution's graph and for that graph not to exist--this is a bug.")
};
// Retrieve the tensor data.
let (mut blob, blob_size) =
if let Some(execution) = self.ctx.borrow_mut().executions.get_mut(context) {
let mut blob = execution.request.get_blob(&output_name)?; // TODO shouldn't need to be mut
let blob_size = blob.byte_len()? as u32;
if blob_size > out_buffer_max_size {
return Err(UsageError::NotEnoughMemory(blob_size).into());
}
(blob, blob_size)
} else {
return Err(UsageError::InvalidExecutionContextHandle.into());
};
// Copy the tensor data over to the `out_buffer`.
let mut out_slice = out_buffer.as_array(out_buffer_max_size).as_slice()?;
(&mut out_slice[..blob_size as usize]).copy_from_slice(blob.buffer()?);
Ok(blob_size)
}
}
/// Return the execution target string expected by OpenVINO from the `ExecutionTarget` enum provided
/// by wasi-nn.
fn map_execution_target_to_string(target: ExecutionTarget) -> &'static str {
match target {
ExecutionTarget::Cpu => "CPU",
ExecutionTarget::Gpu => "GPU",
ExecutionTarget::Tpu => unimplemented!("OpenVINO does not support TPU execution targets"),
}
}

26
crates/wasi-nn/src/lib.rs Normal file
View File

@@ -0,0 +1,26 @@
mod ctx;
mod r#impl;
mod witx;
pub use ctx::WasiNnCtx;
// Defines a `struct WasiNn` with member fields and appropriate APIs for dealing with all the
// various WASI exports.
wasmtime_wiggle::wasmtime_integration!({
// The wiggle code to integrate with lives here:
target: witx,
// This must be the same witx document as used above:
witx: ["$WASI_ROOT/phases/ephemeral/witx/wasi_ephemeral_nn.witx"],
// This must be the same ctx type as used for the target:
ctx: WasiNnCtx,
// This macro will emit a struct to represent the instance, with this name and docs:
modules: {
wasi_ephemeral_nn => {
name: WasiNn,
docs: "An instantiated instance of the wasi-nn exports.",
function_override: {}
}
},
// Error to return when caller module is missing memory export:
missing_memory: { witx::types::Errno::MissingMemory },
});

40
crates/wasi-nn/src/witx.rs Normal file
View File

@@ -0,0 +1,40 @@
//! Contains the macro-generated implementation of wasi-nn from the its witx definition file.
use crate::ctx::WasiNnCtx;
use crate::ctx::WasiNnError;
// Generate the traits and types of wasi-nn in several Rust modules (e.g. `types`).
wiggle::from_witx!({
witx: ["$WASI_ROOT/phases/ephemeral/witx/wasi_ephemeral_nn.witx"],
ctx: WasiNnCtx,
errors: { errno => WasiNnError }
});
use types::Errno;
/// Wiggle generates code that performs some input validation on the arguments passed in by users of
/// wasi-nn. Here we convert the validation error into one (or more, eventually) of the error
/// variants defined in the witx.
impl types::GuestErrorConversion for WasiNnCtx {
fn into_errno(&self, e: wiggle::GuestError) -> Errno {
eprintln!("Guest error: {:?}", e);
Errno::InvalidArgument
}
}
impl<'a> types::UserErrorConversion for WasiNnCtx {
fn errno_from_wasi_nn_error(&self, e: WasiNnError) -> Errno {
eprintln!("Host error: {:?}", e);
match e {
WasiNnError::OpenvinoError(_) => unimplemented!(),
WasiNnError::GuestError(_) => unimplemented!(),
WasiNnError::UsageError(_) => unimplemented!(),
}
}
}
/// Additionally, we must let Wiggle know which of our error codes represents a successful operation.
impl wiggle::GuestErrorType for Errno {
fn success() -> Self {
Self::Success
}
}

6
scripts/publish.rs
View File

@@ -64,6 +64,7 @@ const CRATES_TO_PUBLISH: &[&str] = &[
"wasmtime",
"wasmtime-wiggle",
"wasmtime-wasi",
"wasmtime-wasi-nn",
"wasmtime-rust-macro",
"wasmtime-rust",
"wasmtime-wast",
@@ -308,7 +309,10 @@ fn verify(crates: &[Crate]) {
.arg("--manifest-path")
.arg(&krate.manifest)
.env("CARGO_TARGET_DIR", "./target");
if krate.name.contains("lightbeam") || krate.name == "witx" {
if krate.name.contains("lightbeam")
|| krate.name == "witx"
|| krate.name.contains("wasi-nn")
{
cmd.arg("--no-verify");
}
let status = cmd.status().unwrap();

9
src/commands/run.rs
View File

@@ -15,6 +15,9 @@ use wasi_common::{preopen_dir, WasiCtxBuilder};
use wasmtime::{Engine, Func, Linker, Module, Store, Trap, Val, ValType};
use wasmtime_wasi::Wasi;
#[cfg(feature = "wasi-nn")]
use wasmtime_wasi_nn::{WasiNn, WasiNnCtx};
fn parse_module(s: &OsStr) -> Result<PathBuf, OsString> {
// Do not accept wasmtime subcommand names as the module name
match s.to_str() {
@@ -353,6 +356,12 @@ fn populate_with_wasi(
let wasi = Wasi::new(linker.store(), cx);
wasi.add_to_linker(linker)?;
#[cfg(feature = "wasi-nn")]
{
let wasi_nn = WasiNn::new(linker.store(), WasiNnCtx::new()?);
wasi_nn.add_to_linker(linker)?;
}
// Repeat the above, but this time for snapshot 0.
let mut cx = wasi_common::old::snapshot_0::WasiCtxBuilder::new();
cx.inherit_stdio().args(argv).envs(vars);