blob: ac737c281151462d339ff5a2930cb2abe40a0019 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
|
# WASI-NN
## How to use
Enable WASI-NN in the WAMR by spefiying it in the cmake building configuration as follows,
```
set (WAMR_BUILD_WASI_NN 1)
```
The definition of the functions provided by WASI-NN is in the header file `core/iwasm/libraries/wasi-nn/wasi_nn.h`.
By only including this file in your WASM application you will bind WASI-NN into your module.
## Tests
To run the tests we assume that the current directory is the root of the repository.
### Build the runtime
Build the runtime image for your execution target type.
`EXECUTION_TYPE` can be:
* `cpu`
* `nvidia-gpu`
* `vx-delegate`
```
EXECUTION_TYPE=cpu
docker build -t wasi-nn-${EXECUTION_TYPE} -f core/iwasm/libraries/wasi-nn/test/Dockerfile.${EXECUTION_TYPE} .
```
### Build wasm app
```
docker build -t wasi-nn-compile -f core/iwasm/libraries/wasi-nn/test/Dockerfile.compile .
```
```
docker run -v $PWD/core/iwasm/libraries/wasi-nn:/wasi-nn wasi-nn-compile
```
### Run wasm app
If all the tests have run properly you will the the following message in the terminal,
```
Tests: passed!
```
* CPU
```
docker run \
-v $PWD/core/iwasm/libraries/wasi-nn/test:/assets wasi-nn-cpu \
--dir=/assets \
--env="TARGET=cpu" \
/assets/test_tensorflow.wasm
```
* (NVIDIA) GPU
```
docker run \
--runtime=nvidia \
-v $PWD/core/iwasm/libraries/wasi-nn/test:/assets wasi-nn-nvidia-gpu \
--dir=/assets \
--env="TARGET=gpu" \
/assets/test_tensorflow.wasm
```
* vx-delegate for NPU (x86 simulater)
```
docker run \
-v $PWD/core/iwasm/libraries/wasi-nn/test:/assets wasi-nn-vx-delegate \
--dir=/assets \
--env="TARGET=gpu" \
/assets/test_tensorflow.wasm
```
Requirements:
* [NVIDIA docker](https://github.com/NVIDIA/nvidia-docker).
## What is missing
Supported:
* Graph encoding: `tensorflowlite`.
* Execution target: `cpu` and `gpu`.
* Tensor type: `fp32`.
|
